Loading...
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 */
20struct 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#else /* !CONFIG_SLOB */
34
35struct memcg_cache_array {
36 struct rcu_head rcu;
37 struct kmem_cache *entries[0];
38};
39
40/*
41 * This is the main placeholder for memcg-related information in kmem caches.
42 * Both the root cache and the child caches will have it. For the root cache,
43 * this will hold a dynamically allocated array large enough to hold
44 * information about the currently limited memcgs in the system. To allow the
45 * array to be accessed without taking any locks, on relocation we free the old
46 * version only after a grace period.
47 *
48 * Root and child caches hold different metadata.
49 *
50 * @root_cache: Common to root and child caches. NULL for root, pointer to
51 * the root cache for children.
52 *
53 * The following fields are specific to root caches.
54 *
55 * @memcg_caches: kmemcg ID indexed table of child caches. This table is
56 * used to index child cachces during allocation and cleared
57 * early during shutdown.
58 *
59 * @root_caches_node: List node for slab_root_caches list.
60 *
61 * @children: List of all child caches. While the child caches are also
62 * reachable through @memcg_caches, a child cache remains on
63 * this list until it is actually destroyed.
64 *
65 * The following fields are specific to child caches.
66 *
67 * @memcg: Pointer to the memcg this cache belongs to.
68 *
69 * @children_node: List node for @root_cache->children list.
70 *
71 * @kmem_caches_node: List node for @memcg->kmem_caches list.
72 */
73struct memcg_cache_params {
74 struct kmem_cache *root_cache;
75 union {
76 struct {
77 struct memcg_cache_array __rcu *memcg_caches;
78 struct list_head __root_caches_node;
79 struct list_head children;
80 bool dying;
81 };
82 struct {
83 struct mem_cgroup *memcg;
84 struct list_head children_node;
85 struct list_head kmem_caches_node;
86 struct percpu_ref refcnt;
87
88 void (*work_fn)(struct kmem_cache *);
89 union {
90 struct rcu_head rcu_head;
91 struct work_struct work;
92 };
93 };
94 };
95};
96#endif /* CONFIG_SLOB */
97
98#ifdef CONFIG_SLAB
99#include <linux/slab_def.h>
100#endif
101
102#ifdef CONFIG_SLUB
103#include <linux/slub_def.h>
104#endif
105
106#include <linux/memcontrol.h>
107#include <linux/fault-inject.h>
108#include <linux/kasan.h>
109#include <linux/kmemleak.h>
110#include <linux/random.h>
111#include <linux/sched/mm.h>
112
113/*
114 * State of the slab allocator.
115 *
116 * This is used to describe the states of the allocator during bootup.
117 * Allocators use this to gradually bootstrap themselves. Most allocators
118 * have the problem that the structures used for managing slab caches are
119 * allocated from slab caches themselves.
120 */
121enum slab_state {
122 DOWN, /* No slab functionality yet */
123 PARTIAL, /* SLUB: kmem_cache_node available */
124 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
125 UP, /* Slab caches usable but not all extras yet */
126 FULL /* Everything is working */
127};
128
129extern enum slab_state slab_state;
130
131/* The slab cache mutex protects the management structures during changes */
132extern struct mutex slab_mutex;
133
134/* The list of all slab caches on the system */
135extern struct list_head slab_caches;
136
137/* The slab cache that manages slab cache information */
138extern struct kmem_cache *kmem_cache;
139
140/* A table of kmalloc cache names and sizes */
141extern const struct kmalloc_info_struct {
142 const char *name;
143 unsigned int size;
144} kmalloc_info[];
145
146#ifndef CONFIG_SLOB
147/* Kmalloc array related functions */
148void setup_kmalloc_cache_index_table(void);
149void create_kmalloc_caches(slab_flags_t);
150
151/* Find the kmalloc slab corresponding for a certain size */
152struct kmem_cache *kmalloc_slab(size_t, gfp_t);
153#endif
154
155
156/* Functions provided by the slab allocators */
157int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
158
159struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
160 slab_flags_t flags, unsigned int useroffset,
161 unsigned int usersize);
162extern void create_boot_cache(struct kmem_cache *, const char *name,
163 unsigned int size, slab_flags_t flags,
164 unsigned int useroffset, unsigned int usersize);
165
166int slab_unmergeable(struct kmem_cache *s);
167struct kmem_cache *find_mergeable(unsigned size, unsigned align,
168 slab_flags_t flags, const char *name, void (*ctor)(void *));
169#ifndef CONFIG_SLOB
170struct kmem_cache *
171__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
172 slab_flags_t flags, void (*ctor)(void *));
173
174slab_flags_t kmem_cache_flags(unsigned int object_size,
175 slab_flags_t flags, const char *name,
176 void (*ctor)(void *));
177#else
178static inline struct kmem_cache *
179__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
180 slab_flags_t flags, void (*ctor)(void *))
181{ return NULL; }
182
183static inline slab_flags_t kmem_cache_flags(unsigned int object_size,
184 slab_flags_t flags, const char *name,
185 void (*ctor)(void *))
186{
187 return flags;
188}
189#endif
190
191
192/* Legal flag mask for kmem_cache_create(), for various configurations */
193#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | \
194 SLAB_CACHE_DMA32 | SLAB_PANIC | \
195 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
196
197#if defined(CONFIG_DEBUG_SLAB)
198#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
199#elif defined(CONFIG_SLUB_DEBUG)
200#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
201 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
202#else
203#define SLAB_DEBUG_FLAGS (0)
204#endif
205
206#if defined(CONFIG_SLAB)
207#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
208 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
209 SLAB_ACCOUNT)
210#elif defined(CONFIG_SLUB)
211#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
212 SLAB_TEMPORARY | SLAB_ACCOUNT)
213#else
214#define SLAB_CACHE_FLAGS (0)
215#endif
216
217/* Common flags available with current configuration */
218#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
219
220/* Common flags permitted for kmem_cache_create */
221#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
222 SLAB_RED_ZONE | \
223 SLAB_POISON | \
224 SLAB_STORE_USER | \
225 SLAB_TRACE | \
226 SLAB_CONSISTENCY_CHECKS | \
227 SLAB_MEM_SPREAD | \
228 SLAB_NOLEAKTRACE | \
229 SLAB_RECLAIM_ACCOUNT | \
230 SLAB_TEMPORARY | \
231 SLAB_ACCOUNT)
232
233bool __kmem_cache_empty(struct kmem_cache *);
234int __kmem_cache_shutdown(struct kmem_cache *);
235void __kmem_cache_release(struct kmem_cache *);
236int __kmem_cache_shrink(struct kmem_cache *);
237void __kmemcg_cache_deactivate(struct kmem_cache *s);
238void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s);
239void slab_kmem_cache_release(struct kmem_cache *);
240void kmem_cache_shrink_all(struct kmem_cache *s);
241
242struct seq_file;
243struct file;
244
245struct slabinfo {
246 unsigned long active_objs;
247 unsigned long num_objs;
248 unsigned long active_slabs;
249 unsigned long num_slabs;
250 unsigned long shared_avail;
251 unsigned int limit;
252 unsigned int batchcount;
253 unsigned int shared;
254 unsigned int objects_per_slab;
255 unsigned int cache_order;
256};
257
258void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
259void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
260ssize_t slabinfo_write(struct file *file, const char __user *buffer,
261 size_t count, loff_t *ppos);
262
263/*
264 * Generic implementation of bulk operations
265 * These are useful for situations in which the allocator cannot
266 * perform optimizations. In that case segments of the object listed
267 * may be allocated or freed using these operations.
268 */
269void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
270int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
271
272static inline int cache_vmstat_idx(struct kmem_cache *s)
273{
274 return (s->flags & SLAB_RECLAIM_ACCOUNT) ?
275 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE;
276}
277
278#ifdef CONFIG_MEMCG_KMEM
279
280/* List of all root caches. */
281extern struct list_head slab_root_caches;
282#define root_caches_node memcg_params.__root_caches_node
283
284/*
285 * Iterate over all memcg caches of the given root cache. The caller must hold
286 * slab_mutex.
287 */
288#define for_each_memcg_cache(iter, root) \
289 list_for_each_entry(iter, &(root)->memcg_params.children, \
290 memcg_params.children_node)
291
292static inline bool is_root_cache(struct kmem_cache *s)
293{
294 return !s->memcg_params.root_cache;
295}
296
297static inline bool slab_equal_or_root(struct kmem_cache *s,
298 struct kmem_cache *p)
299{
300 return p == s || p == s->memcg_params.root_cache;
301}
302
303/*
304 * We use suffixes to the name in memcg because we can't have caches
305 * created in the system with the same name. But when we print them
306 * locally, better refer to them with the base name
307 */
308static inline const char *cache_name(struct kmem_cache *s)
309{
310 if (!is_root_cache(s))
311 s = s->memcg_params.root_cache;
312 return s->name;
313}
314
315static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
316{
317 if (is_root_cache(s))
318 return s;
319 return s->memcg_params.root_cache;
320}
321
322/*
323 * Expects a pointer to a slab page. Please note, that PageSlab() check
324 * isn't sufficient, as it returns true also for tail compound slab pages,
325 * which do not have slab_cache pointer set.
326 * So this function assumes that the page can pass PageSlab() && !PageTail()
327 * check.
328 *
329 * The kmem_cache can be reparented asynchronously. The caller must ensure
330 * the memcg lifetime, e.g. by taking rcu_read_lock() or cgroup_mutex.
331 */
332static inline struct mem_cgroup *memcg_from_slab_page(struct page *page)
333{
334 struct kmem_cache *s;
335
336 s = READ_ONCE(page->slab_cache);
337 if (s && !is_root_cache(s))
338 return READ_ONCE(s->memcg_params.memcg);
339
340 return NULL;
341}
342
343/*
344 * Charge the slab page belonging to the non-root kmem_cache.
345 * Can be called for non-root kmem_caches only.
346 */
347static __always_inline int memcg_charge_slab(struct page *page,
348 gfp_t gfp, int order,
349 struct kmem_cache *s)
350{
351 struct mem_cgroup *memcg;
352 struct lruvec *lruvec;
353 int ret;
354
355 rcu_read_lock();
356 memcg = READ_ONCE(s->memcg_params.memcg);
357 while (memcg && !css_tryget_online(&memcg->css))
358 memcg = parent_mem_cgroup(memcg);
359 rcu_read_unlock();
360
361 if (unlikely(!memcg || mem_cgroup_is_root(memcg))) {
362 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
363 (1 << order));
364 percpu_ref_get_many(&s->memcg_params.refcnt, 1 << order);
365 return 0;
366 }
367
368 ret = memcg_kmem_charge_memcg(page, gfp, order, memcg);
369 if (ret)
370 goto out;
371
372 lruvec = mem_cgroup_lruvec(page_pgdat(page), memcg);
373 mod_lruvec_state(lruvec, cache_vmstat_idx(s), 1 << order);
374
375 /* transer try_charge() page references to kmem_cache */
376 percpu_ref_get_many(&s->memcg_params.refcnt, 1 << order);
377 css_put_many(&memcg->css, 1 << order);
378out:
379 css_put(&memcg->css);
380 return ret;
381}
382
383/*
384 * Uncharge a slab page belonging to a non-root kmem_cache.
385 * Can be called for non-root kmem_caches only.
386 */
387static __always_inline void memcg_uncharge_slab(struct page *page, int order,
388 struct kmem_cache *s)
389{
390 struct mem_cgroup *memcg;
391 struct lruvec *lruvec;
392
393 rcu_read_lock();
394 memcg = READ_ONCE(s->memcg_params.memcg);
395 if (likely(!mem_cgroup_is_root(memcg))) {
396 lruvec = mem_cgroup_lruvec(page_pgdat(page), memcg);
397 mod_lruvec_state(lruvec, cache_vmstat_idx(s), -(1 << order));
398 memcg_kmem_uncharge_memcg(page, order, memcg);
399 } else {
400 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
401 -(1 << order));
402 }
403 rcu_read_unlock();
404
405 percpu_ref_put_many(&s->memcg_params.refcnt, 1 << order);
406}
407
408extern void slab_init_memcg_params(struct kmem_cache *);
409extern void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg);
410
411#else /* CONFIG_MEMCG_KMEM */
412
413/* If !memcg, all caches are root. */
414#define slab_root_caches slab_caches
415#define root_caches_node list
416
417#define for_each_memcg_cache(iter, root) \
418 for ((void)(iter), (void)(root); 0; )
419
420static inline bool is_root_cache(struct kmem_cache *s)
421{
422 return true;
423}
424
425static inline bool slab_equal_or_root(struct kmem_cache *s,
426 struct kmem_cache *p)
427{
428 return s == p;
429}
430
431static inline const char *cache_name(struct kmem_cache *s)
432{
433 return s->name;
434}
435
436static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
437{
438 return s;
439}
440
441static inline struct mem_cgroup *memcg_from_slab_page(struct page *page)
442{
443 return NULL;
444}
445
446static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
447 struct kmem_cache *s)
448{
449 return 0;
450}
451
452static inline void memcg_uncharge_slab(struct page *page, int order,
453 struct kmem_cache *s)
454{
455}
456
457static inline void slab_init_memcg_params(struct kmem_cache *s)
458{
459}
460
461static inline void memcg_link_cache(struct kmem_cache *s,
462 struct mem_cgroup *memcg)
463{
464}
465
466#endif /* CONFIG_MEMCG_KMEM */
467
468static inline struct kmem_cache *virt_to_cache(const void *obj)
469{
470 struct page *page;
471
472 page = virt_to_head_page(obj);
473 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
474 __func__))
475 return NULL;
476 return page->slab_cache;
477}
478
479static __always_inline int charge_slab_page(struct page *page,
480 gfp_t gfp, int order,
481 struct kmem_cache *s)
482{
483 if (is_root_cache(s)) {
484 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
485 1 << order);
486 return 0;
487 }
488
489 return memcg_charge_slab(page, gfp, order, s);
490}
491
492static __always_inline void uncharge_slab_page(struct page *page, int order,
493 struct kmem_cache *s)
494{
495 if (is_root_cache(s)) {
496 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
497 -(1 << order));
498 return;
499 }
500
501 memcg_uncharge_slab(page, order, s);
502}
503
504static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
505{
506 struct kmem_cache *cachep;
507
508 /*
509 * When kmemcg is not being used, both assignments should return the
510 * same value. but we don't want to pay the assignment price in that
511 * case. If it is not compiled in, the compiler should be smart enough
512 * to not do even the assignment. In that case, slab_equal_or_root
513 * will also be a constant.
514 */
515 if (!memcg_kmem_enabled() &&
516 !IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
517 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
518 return s;
519
520 cachep = virt_to_cache(x);
521 WARN_ONCE(cachep && !slab_equal_or_root(cachep, s),
522 "%s: Wrong slab cache. %s but object is from %s\n",
523 __func__, s->name, cachep->name);
524 return cachep;
525}
526
527static inline size_t slab_ksize(const struct kmem_cache *s)
528{
529#ifndef CONFIG_SLUB
530 return s->object_size;
531
532#else /* CONFIG_SLUB */
533# ifdef CONFIG_SLUB_DEBUG
534 /*
535 * Debugging requires use of the padding between object
536 * and whatever may come after it.
537 */
538 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
539 return s->object_size;
540# endif
541 if (s->flags & SLAB_KASAN)
542 return s->object_size;
543 /*
544 * If we have the need to store the freelist pointer
545 * back there or track user information then we can
546 * only use the space before that information.
547 */
548 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
549 return s->inuse;
550 /*
551 * Else we can use all the padding etc for the allocation
552 */
553 return s->size;
554#endif
555}
556
557static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
558 gfp_t flags)
559{
560 flags &= gfp_allowed_mask;
561
562 fs_reclaim_acquire(flags);
563 fs_reclaim_release(flags);
564
565 might_sleep_if(gfpflags_allow_blocking(flags));
566
567 if (should_failslab(s, flags))
568 return NULL;
569
570 if (memcg_kmem_enabled() &&
571 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
572 return memcg_kmem_get_cache(s);
573
574 return s;
575}
576
577static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
578 size_t size, void **p)
579{
580 size_t i;
581
582 flags &= gfp_allowed_mask;
583 for (i = 0; i < size; i++) {
584 p[i] = kasan_slab_alloc(s, p[i], flags);
585 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
586 kmemleak_alloc_recursive(p[i], s->object_size, 1,
587 s->flags, flags);
588 }
589
590 if (memcg_kmem_enabled())
591 memcg_kmem_put_cache(s);
592}
593
594#ifndef CONFIG_SLOB
595/*
596 * The slab lists for all objects.
597 */
598struct kmem_cache_node {
599 spinlock_t list_lock;
600
601#ifdef CONFIG_SLAB
602 struct list_head slabs_partial; /* partial list first, better asm code */
603 struct list_head slabs_full;
604 struct list_head slabs_free;
605 unsigned long total_slabs; /* length of all slab lists */
606 unsigned long free_slabs; /* length of free slab list only */
607 unsigned long free_objects;
608 unsigned int free_limit;
609 unsigned int colour_next; /* Per-node cache coloring */
610 struct array_cache *shared; /* shared per node */
611 struct alien_cache **alien; /* on other nodes */
612 unsigned long next_reap; /* updated without locking */
613 int free_touched; /* updated without locking */
614#endif
615
616#ifdef CONFIG_SLUB
617 unsigned long nr_partial;
618 struct list_head partial;
619#ifdef CONFIG_SLUB_DEBUG
620 atomic_long_t nr_slabs;
621 atomic_long_t total_objects;
622 struct list_head full;
623#endif
624#endif
625
626};
627
628static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
629{
630 return s->node[node];
631}
632
633/*
634 * Iterator over all nodes. The body will be executed for each node that has
635 * a kmem_cache_node structure allocated (which is true for all online nodes)
636 */
637#define for_each_kmem_cache_node(__s, __node, __n) \
638 for (__node = 0; __node < nr_node_ids; __node++) \
639 if ((__n = get_node(__s, __node)))
640
641#endif
642
643void *slab_start(struct seq_file *m, loff_t *pos);
644void *slab_next(struct seq_file *m, void *p, loff_t *pos);
645void slab_stop(struct seq_file *m, void *p);
646void *memcg_slab_start(struct seq_file *m, loff_t *pos);
647void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
648void memcg_slab_stop(struct seq_file *m, void *p);
649int memcg_slab_show(struct seq_file *m, void *p);
650
651#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
652void dump_unreclaimable_slab(void);
653#else
654static inline void dump_unreclaimable_slab(void)
655{
656}
657#endif
658
659void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
660
661#ifdef CONFIG_SLAB_FREELIST_RANDOM
662int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
663 gfp_t gfp);
664void cache_random_seq_destroy(struct kmem_cache *cachep);
665#else
666static inline int cache_random_seq_create(struct kmem_cache *cachep,
667 unsigned int count, gfp_t gfp)
668{
669 return 0;
670}
671static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
672#endif /* CONFIG_SLAB_FREELIST_RANDOM */
673
674static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
675{
676 if (static_branch_unlikely(&init_on_alloc)) {
677 if (c->ctor)
678 return false;
679 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
680 return flags & __GFP_ZERO;
681 return true;
682 }
683 return flags & __GFP_ZERO;
684}
685
686static inline bool slab_want_init_on_free(struct kmem_cache *c)
687{
688 if (static_branch_unlikely(&init_on_free))
689 return !(c->ctor ||
690 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
691 return false;
692}
693
694#endif /* MM_SLAB_H */
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 */
20struct 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 */
58enum 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
66extern enum slab_state slab_state;
67
68/* The slab cache mutex protects the management structures during changes */
69extern struct mutex slab_mutex;
70
71/* The list of all slab caches on the system */
72extern struct list_head slab_caches;
73
74/* The slab cache that manages slab cache information */
75extern struct kmem_cache *kmem_cache;
76
77/* A table of kmalloc cache names and sizes */
78extern 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 */
85void setup_kmalloc_cache_index_table(void);
86void create_kmalloc_caches(slab_flags_t);
87
88/* Find the kmalloc slab corresponding for a certain size */
89struct kmem_cache *kmalloc_slab(size_t, gfp_t);
90#endif
91
92gfp_t kmalloc_fix_flags(gfp_t flags);
93
94/* Functions provided by the slab allocators */
95int __kmem_cache_create(struct kmem_cache *, slab_flags_t flags);
96
97struct kmem_cache *create_kmalloc_cache(const char *name, unsigned int size,
98 slab_flags_t flags, unsigned int useroffset,
99 unsigned int usersize);
100extern 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
104int slab_unmergeable(struct kmem_cache *s);
105struct kmem_cache *find_mergeable(unsigned size, unsigned align,
106 slab_flags_t flags, const char *name, void (*ctor)(void *));
107#ifndef CONFIG_SLOB
108struct kmem_cache *
109__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
110 slab_flags_t flags, void (*ctor)(void *));
111
112slab_flags_t kmem_cache_flags(unsigned int object_size,
113 slab_flags_t flags, const char *name);
114#else
115static 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
120static 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
169bool __kmem_cache_empty(struct kmem_cache *);
170int __kmem_cache_shutdown(struct kmem_cache *);
171void __kmem_cache_release(struct kmem_cache *);
172int __kmem_cache_shrink(struct kmem_cache *);
173void slab_kmem_cache_release(struct kmem_cache *);
174
175struct seq_file;
176struct file;
177
178struct 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
191void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
192void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
193ssize_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 */
202void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
203int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
204
205static 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
213DECLARE_STATIC_KEY_TRUE(slub_debug_enabled);
214#else
215DECLARE_STATIC_KEY_FALSE(slub_debug_enabled);
216#endif
217extern void print_tracking(struct kmem_cache *s, void *object);
218long validate_slab_cache(struct kmem_cache *s);
219static inline bool __slub_debug_enabled(void)
220{
221 return static_branch_unlikely(&slub_debug_enabled);
222}
223#else
224static inline void print_tracking(struct kmem_cache *s, void *object)
225{
226}
227static inline bool __slub_debug_enabled(void)
228{
229 return false;
230}
231#endif
232
233/*
234 * Returns true if any of the specified slub_debug flags is enabled for the
235 * cache. Use only for flags parsed by setup_slub_debug() as it also enables
236 * the static key.
237 */
238static inline bool kmem_cache_debug_flags(struct kmem_cache *s, slab_flags_t flags)
239{
240 if (IS_ENABLED(CONFIG_SLUB_DEBUG))
241 VM_WARN_ON_ONCE(!(flags & SLAB_DEBUG_FLAGS));
242 if (__slub_debug_enabled())
243 return s->flags & flags;
244 return false;
245}
246
247#ifdef CONFIG_MEMCG_KMEM
248int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s,
249 gfp_t gfp, bool new_page);
250void mod_objcg_state(struct obj_cgroup *objcg, struct pglist_data *pgdat,
251 enum node_stat_item idx, int nr);
252
253static inline void memcg_free_page_obj_cgroups(struct page *page)
254{
255 kfree(page_objcgs(page));
256 page->memcg_data = 0;
257}
258
259static inline size_t obj_full_size(struct kmem_cache *s)
260{
261 /*
262 * For each accounted object there is an extra space which is used
263 * to store obj_cgroup membership. Charge it too.
264 */
265 return s->size + sizeof(struct obj_cgroup *);
266}
267
268/*
269 * Returns false if the allocation should fail.
270 */
271static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
272 struct obj_cgroup **objcgp,
273 size_t objects, gfp_t flags)
274{
275 struct obj_cgroup *objcg;
276
277 if (!memcg_kmem_enabled())
278 return true;
279
280 if (!(flags & __GFP_ACCOUNT) && !(s->flags & SLAB_ACCOUNT))
281 return true;
282
283 objcg = get_obj_cgroup_from_current();
284 if (!objcg)
285 return true;
286
287 if (obj_cgroup_charge(objcg, flags, objects * obj_full_size(s))) {
288 obj_cgroup_put(objcg);
289 return false;
290 }
291
292 *objcgp = objcg;
293 return true;
294}
295
296static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
297 struct obj_cgroup *objcg,
298 gfp_t flags, size_t size,
299 void **p)
300{
301 struct page *page;
302 unsigned long off;
303 size_t i;
304
305 if (!memcg_kmem_enabled() || !objcg)
306 return;
307
308 for (i = 0; i < size; i++) {
309 if (likely(p[i])) {
310 page = virt_to_head_page(p[i]);
311
312 if (!page_objcgs(page) &&
313 memcg_alloc_page_obj_cgroups(page, s, flags,
314 false)) {
315 obj_cgroup_uncharge(objcg, obj_full_size(s));
316 continue;
317 }
318
319 off = obj_to_index(s, page, p[i]);
320 obj_cgroup_get(objcg);
321 page_objcgs(page)[off] = objcg;
322 mod_objcg_state(objcg, page_pgdat(page),
323 cache_vmstat_idx(s), obj_full_size(s));
324 } else {
325 obj_cgroup_uncharge(objcg, obj_full_size(s));
326 }
327 }
328 obj_cgroup_put(objcg);
329}
330
331static inline void memcg_slab_free_hook(struct kmem_cache *s_orig,
332 void **p, int objects)
333{
334 struct kmem_cache *s;
335 struct obj_cgroup **objcgs;
336 struct obj_cgroup *objcg;
337 struct page *page;
338 unsigned int off;
339 int i;
340
341 if (!memcg_kmem_enabled())
342 return;
343
344 for (i = 0; i < objects; i++) {
345 if (unlikely(!p[i]))
346 continue;
347
348 page = virt_to_head_page(p[i]);
349 objcgs = page_objcgs_check(page);
350 if (!objcgs)
351 continue;
352
353 if (!s_orig)
354 s = page->slab_cache;
355 else
356 s = s_orig;
357
358 off = obj_to_index(s, page, p[i]);
359 objcg = objcgs[off];
360 if (!objcg)
361 continue;
362
363 objcgs[off] = NULL;
364 obj_cgroup_uncharge(objcg, obj_full_size(s));
365 mod_objcg_state(objcg, page_pgdat(page), cache_vmstat_idx(s),
366 -obj_full_size(s));
367 obj_cgroup_put(objcg);
368 }
369}
370
371#else /* CONFIG_MEMCG_KMEM */
372static inline struct mem_cgroup *memcg_from_slab_obj(void *ptr)
373{
374 return NULL;
375}
376
377static inline int memcg_alloc_page_obj_cgroups(struct page *page,
378 struct kmem_cache *s, gfp_t gfp,
379 bool new_page)
380{
381 return 0;
382}
383
384static inline void memcg_free_page_obj_cgroups(struct page *page)
385{
386}
387
388static inline bool memcg_slab_pre_alloc_hook(struct kmem_cache *s,
389 struct obj_cgroup **objcgp,
390 size_t objects, gfp_t flags)
391{
392 return true;
393}
394
395static inline void memcg_slab_post_alloc_hook(struct kmem_cache *s,
396 struct obj_cgroup *objcg,
397 gfp_t flags, size_t size,
398 void **p)
399{
400}
401
402static inline void memcg_slab_free_hook(struct kmem_cache *s,
403 void **p, int objects)
404{
405}
406#endif /* CONFIG_MEMCG_KMEM */
407
408static inline struct kmem_cache *virt_to_cache(const void *obj)
409{
410 struct page *page;
411
412 page = virt_to_head_page(obj);
413 if (WARN_ONCE(!PageSlab(page), "%s: Object is not a Slab page!\n",
414 __func__))
415 return NULL;
416 return page->slab_cache;
417}
418
419static __always_inline void account_slab_page(struct page *page, int order,
420 struct kmem_cache *s,
421 gfp_t gfp)
422{
423 if (memcg_kmem_enabled() && (s->flags & SLAB_ACCOUNT))
424 memcg_alloc_page_obj_cgroups(page, s, gfp, true);
425
426 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
427 PAGE_SIZE << order);
428}
429
430static __always_inline void unaccount_slab_page(struct page *page, int order,
431 struct kmem_cache *s)
432{
433 if (memcg_kmem_enabled())
434 memcg_free_page_obj_cgroups(page);
435
436 mod_node_page_state(page_pgdat(page), cache_vmstat_idx(s),
437 -(PAGE_SIZE << order));
438}
439
440static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
441{
442 struct kmem_cache *cachep;
443
444 if (!IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED) &&
445 !kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS))
446 return s;
447
448 cachep = virt_to_cache(x);
449 if (WARN(cachep && cachep != s,
450 "%s: Wrong slab cache. %s but object is from %s\n",
451 __func__, s->name, cachep->name))
452 print_tracking(cachep, x);
453 return cachep;
454}
455
456static inline size_t slab_ksize(const struct kmem_cache *s)
457{
458#ifndef CONFIG_SLUB
459 return s->object_size;
460
461#else /* CONFIG_SLUB */
462# ifdef CONFIG_SLUB_DEBUG
463 /*
464 * Debugging requires use of the padding between object
465 * and whatever may come after it.
466 */
467 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
468 return s->object_size;
469# endif
470 if (s->flags & SLAB_KASAN)
471 return s->object_size;
472 /*
473 * If we have the need to store the freelist pointer
474 * back there or track user information then we can
475 * only use the space before that information.
476 */
477 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
478 return s->inuse;
479 /*
480 * Else we can use all the padding etc for the allocation
481 */
482 return s->size;
483#endif
484}
485
486static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
487 struct obj_cgroup **objcgp,
488 size_t size, gfp_t flags)
489{
490 flags &= gfp_allowed_mask;
491
492 might_alloc(flags);
493
494 if (should_failslab(s, flags))
495 return NULL;
496
497 if (!memcg_slab_pre_alloc_hook(s, objcgp, size, flags))
498 return NULL;
499
500 return s;
501}
502
503static inline void slab_post_alloc_hook(struct kmem_cache *s,
504 struct obj_cgroup *objcg, gfp_t flags,
505 size_t size, void **p, bool init)
506{
507 size_t i;
508
509 flags &= gfp_allowed_mask;
510
511 /*
512 * As memory initialization might be integrated into KASAN,
513 * kasan_slab_alloc and initialization memset must be
514 * kept together to avoid discrepancies in behavior.
515 *
516 * As p[i] might get tagged, memset and kmemleak hook come after KASAN.
517 */
518 for (i = 0; i < size; i++) {
519 p[i] = kasan_slab_alloc(s, p[i], flags, init);
520 if (p[i] && init && !kasan_has_integrated_init())
521 memset(p[i], 0, s->object_size);
522 kmemleak_alloc_recursive(p[i], s->object_size, 1,
523 s->flags, flags);
524 }
525
526 memcg_slab_post_alloc_hook(s, objcg, flags, size, p);
527}
528
529#ifndef CONFIG_SLOB
530/*
531 * The slab lists for all objects.
532 */
533struct kmem_cache_node {
534 spinlock_t list_lock;
535
536#ifdef CONFIG_SLAB
537 struct list_head slabs_partial; /* partial list first, better asm code */
538 struct list_head slabs_full;
539 struct list_head slabs_free;
540 unsigned long total_slabs; /* length of all slab lists */
541 unsigned long free_slabs; /* length of free slab list only */
542 unsigned long free_objects;
543 unsigned int free_limit;
544 unsigned int colour_next; /* Per-node cache coloring */
545 struct array_cache *shared; /* shared per node */
546 struct alien_cache **alien; /* on other nodes */
547 unsigned long next_reap; /* updated without locking */
548 int free_touched; /* updated without locking */
549#endif
550
551#ifdef CONFIG_SLUB
552 unsigned long nr_partial;
553 struct list_head partial;
554#ifdef CONFIG_SLUB_DEBUG
555 atomic_long_t nr_slabs;
556 atomic_long_t total_objects;
557 struct list_head full;
558#endif
559#endif
560
561};
562
563static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
564{
565 return s->node[node];
566}
567
568/*
569 * Iterator over all nodes. The body will be executed for each node that has
570 * a kmem_cache_node structure allocated (which is true for all online nodes)
571 */
572#define for_each_kmem_cache_node(__s, __node, __n) \
573 for (__node = 0; __node < nr_node_ids; __node++) \
574 if ((__n = get_node(__s, __node)))
575
576#endif
577
578void *slab_start(struct seq_file *m, loff_t *pos);
579void *slab_next(struct seq_file *m, void *p, loff_t *pos);
580void slab_stop(struct seq_file *m, void *p);
581int memcg_slab_show(struct seq_file *m, void *p);
582
583#if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
584void dump_unreclaimable_slab(void);
585#else
586static inline void dump_unreclaimable_slab(void)
587{
588}
589#endif
590
591void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
592
593#ifdef CONFIG_SLAB_FREELIST_RANDOM
594int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
595 gfp_t gfp);
596void cache_random_seq_destroy(struct kmem_cache *cachep);
597#else
598static inline int cache_random_seq_create(struct kmem_cache *cachep,
599 unsigned int count, gfp_t gfp)
600{
601 return 0;
602}
603static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
604#endif /* CONFIG_SLAB_FREELIST_RANDOM */
605
606static inline bool slab_want_init_on_alloc(gfp_t flags, struct kmem_cache *c)
607{
608 if (static_branch_maybe(CONFIG_INIT_ON_ALLOC_DEFAULT_ON,
609 &init_on_alloc)) {
610 if (c->ctor)
611 return false;
612 if (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON))
613 return flags & __GFP_ZERO;
614 return true;
615 }
616 return flags & __GFP_ZERO;
617}
618
619static inline bool slab_want_init_on_free(struct kmem_cache *c)
620{
621 if (static_branch_maybe(CONFIG_INIT_ON_FREE_DEFAULT_ON,
622 &init_on_free))
623 return !(c->ctor ||
624 (c->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)));
625 return false;
626}
627
628#if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG)
629void debugfs_slab_release(struct kmem_cache *);
630#else
631static inline void debugfs_slab_release(struct kmem_cache *s) { }
632#endif
633
634#ifdef CONFIG_PRINTK
635#define KS_ADDRS_COUNT 16
636struct kmem_obj_info {
637 void *kp_ptr;
638 struct page *kp_page;
639 void *kp_objp;
640 unsigned long kp_data_offset;
641 struct kmem_cache *kp_slab_cache;
642 void *kp_ret;
643 void *kp_stack[KS_ADDRS_COUNT];
644 void *kp_free_stack[KS_ADDRS_COUNT];
645};
646void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page);
647#endif
648
649#endif /* MM_SLAB_H */