1 /* SPDX-License-Identifier: GPL-2.0 */
5 * Internal slab definitions
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.
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.
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 */
33 #else /* !CONFIG_SLOB */
35 struct memcg_cache_array
{
37 struct kmem_cache
*entries
[0];
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.
48 * Root and child caches hold different metadata.
50 * @root_cache: Common to root and child caches. NULL for root, pointer to
51 * the root cache for children.
53 * The following fields are specific to root caches.
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.
59 * @root_caches_node: List node for slab_root_caches list.
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.
65 * The following fields are specific to child caches.
67 * @memcg: Pointer to the memcg this cache belongs to.
69 * @children_node: List node for @root_cache->children list.
71 * @kmem_caches_node: List node for @memcg->kmem_caches list.
73 struct memcg_cache_params
{
74 struct kmem_cache
*root_cache
;
77 struct memcg_cache_array __rcu
*memcg_caches
;
78 struct list_head __root_caches_node
;
79 struct list_head children
;
83 struct mem_cgroup
*memcg
;
84 struct list_head children_node
;
85 struct list_head kmem_caches_node
;
86 struct percpu_ref refcnt
;
88 void (*work_fn
)(struct kmem_cache
*);
90 struct rcu_head rcu_head
;
91 struct work_struct work
;
96 #endif /* CONFIG_SLOB */
99 #include <linux/slab_def.h>
103 #include <linux/slub_def.h>
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>
114 * State of the slab allocator.
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.
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 */
129 extern enum slab_state slab_state
;
131 /* The slab cache mutex protects the management structures during changes */
132 extern struct mutex slab_mutex
;
134 /* The list of all slab caches on the system */
135 extern struct list_head slab_caches
;
137 /* The slab cache that manages slab cache information */
138 extern struct kmem_cache
*kmem_cache
;
140 /* A table of kmalloc cache names and sizes */
141 extern const struct kmalloc_info_struct
{
142 const char *name
[NR_KMALLOC_TYPES
];
147 /* Kmalloc array related functions */
148 void setup_kmalloc_cache_index_table(void);
149 void create_kmalloc_caches(slab_flags_t
);
151 /* Find the kmalloc slab corresponding for a certain size */
152 struct kmem_cache
*kmalloc_slab(size_t, gfp_t
);
156 /* Functions provided by the slab allocators */
157 int __kmem_cache_create(struct kmem_cache
*, slab_flags_t flags
);
159 struct kmem_cache
*create_kmalloc_cache(const char *name
, unsigned int size
,
160 slab_flags_t flags
, unsigned int useroffset
,
161 unsigned int usersize
);
162 extern 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
);
166 int slab_unmergeable(struct kmem_cache
*s
);
167 struct kmem_cache
*find_mergeable(unsigned size
, unsigned align
,
168 slab_flags_t flags
, const char *name
, void (*ctor
)(void *));
171 __kmem_cache_alias(const char *name
, unsigned int size
, unsigned int align
,
172 slab_flags_t flags
, void (*ctor
)(void *));
174 slab_flags_t
kmem_cache_flags(unsigned int object_size
,
175 slab_flags_t flags
, const char *name
,
176 void (*ctor
)(void *));
178 static 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 *))
183 static inline slab_flags_t
kmem_cache_flags(unsigned int object_size
,
184 slab_flags_t flags
, const char *name
,
185 void (*ctor
)(void *))
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 )
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)
203 #define SLAB_DEBUG_FLAGS (0)
206 #if defined(CONFIG_SLAB)
207 #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
208 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
210 #elif defined(CONFIG_SLUB)
211 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
212 SLAB_TEMPORARY | SLAB_ACCOUNT)
214 #define SLAB_CACHE_FLAGS (0)
217 /* Common flags available with current configuration */
218 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
220 /* Common flags permitted for kmem_cache_create */
221 #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
226 SLAB_CONSISTENCY_CHECKS | \
229 SLAB_RECLAIM_ACCOUNT | \
233 bool __kmem_cache_empty(struct kmem_cache
*);
234 int __kmem_cache_shutdown(struct kmem_cache
*);
235 void __kmem_cache_release(struct kmem_cache
*);
236 int __kmem_cache_shrink(struct kmem_cache
*);
237 void __kmemcg_cache_deactivate(struct kmem_cache
*s
);
238 void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache
*s
);
239 void slab_kmem_cache_release(struct kmem_cache
*);
240 void kmem_cache_shrink_all(struct kmem_cache
*s
);
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
;
252 unsigned int batchcount
;
254 unsigned int objects_per_slab
;
255 unsigned int cache_order
;
258 void get_slabinfo(struct kmem_cache
*s
, struct slabinfo
*sinfo
);
259 void slabinfo_show_stats(struct seq_file
*m
, struct kmem_cache
*s
);
260 ssize_t
slabinfo_write(struct file
*file
, const char __user
*buffer
,
261 size_t count
, loff_t
*ppos
);
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.
269 void __kmem_cache_free_bulk(struct kmem_cache
*, size_t, void **);
270 int __kmem_cache_alloc_bulk(struct kmem_cache
*, gfp_t
, size_t, void **);
272 static inline int cache_vmstat_idx(struct kmem_cache
*s
)
274 return (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
275 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
;
278 #ifdef CONFIG_MEMCG_KMEM
280 /* List of all root caches. */
281 extern struct list_head slab_root_caches
;
282 #define root_caches_node memcg_params.__root_caches_node
285 * Iterate over all memcg caches of the given root cache. The caller must hold
288 #define for_each_memcg_cache(iter, root) \
289 list_for_each_entry(iter, &(root)->memcg_params.children, \
290 memcg_params.children_node)
292 static inline bool is_root_cache(struct kmem_cache
*s
)
294 return !s
->memcg_params
.root_cache
;
297 static inline bool slab_equal_or_root(struct kmem_cache
*s
,
298 struct kmem_cache
*p
)
300 return p
== s
|| p
== s
->memcg_params
.root_cache
;
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
308 static inline const char *cache_name(struct kmem_cache
*s
)
310 if (!is_root_cache(s
))
311 s
= s
->memcg_params
.root_cache
;
315 static inline struct kmem_cache
*memcg_root_cache(struct kmem_cache
*s
)
317 if (is_root_cache(s
))
319 return s
->memcg_params
.root_cache
;
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()
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.
332 static inline struct mem_cgroup
*memcg_from_slab_page(struct page
*page
)
334 struct kmem_cache
*s
;
336 s
= READ_ONCE(page
->slab_cache
);
337 if (s
&& !is_root_cache(s
))
338 return READ_ONCE(s
->memcg_params
.memcg
);
344 * Charge the slab page belonging to the non-root kmem_cache.
345 * Can be called for non-root kmem_caches only.
347 static __always_inline
int memcg_charge_slab(struct page
*page
,
348 gfp_t gfp
, int order
,
349 struct kmem_cache
*s
)
351 int nr_pages
= 1 << order
;
352 struct mem_cgroup
*memcg
;
353 struct lruvec
*lruvec
;
357 memcg
= READ_ONCE(s
->memcg_params
.memcg
);
358 while (memcg
&& !css_tryget_online(&memcg
->css
))
359 memcg
= parent_mem_cgroup(memcg
);
362 if (unlikely(!memcg
|| mem_cgroup_is_root(memcg
))) {
363 mod_node_page_state(page_pgdat(page
), cache_vmstat_idx(s
),
365 percpu_ref_get_many(&s
->memcg_params
.refcnt
, nr_pages
);
369 ret
= memcg_kmem_charge(memcg
, gfp
, nr_pages
);
373 lruvec
= mem_cgroup_lruvec(memcg
, page_pgdat(page
));
374 mod_lruvec_state(lruvec
, cache_vmstat_idx(s
), nr_pages
);
376 /* transer try_charge() page references to kmem_cache */
377 percpu_ref_get_many(&s
->memcg_params
.refcnt
, nr_pages
);
378 css_put_many(&memcg
->css
, nr_pages
);
380 css_put(&memcg
->css
);
385 * Uncharge a slab page belonging to a non-root kmem_cache.
386 * Can be called for non-root kmem_caches only.
388 static __always_inline
void memcg_uncharge_slab(struct page
*page
, int order
,
389 struct kmem_cache
*s
)
391 int nr_pages
= 1 << order
;
392 struct mem_cgroup
*memcg
;
393 struct lruvec
*lruvec
;
396 memcg
= READ_ONCE(s
->memcg_params
.memcg
);
397 if (likely(!mem_cgroup_is_root(memcg
))) {
398 lruvec
= mem_cgroup_lruvec(memcg
, page_pgdat(page
));
399 mod_lruvec_state(lruvec
, cache_vmstat_idx(s
), -nr_pages
);
400 memcg_kmem_uncharge(memcg
, nr_pages
);
402 mod_node_page_state(page_pgdat(page
), cache_vmstat_idx(s
),
407 percpu_ref_put_many(&s
->memcg_params
.refcnt
, nr_pages
);
410 extern void slab_init_memcg_params(struct kmem_cache
*);
411 extern void memcg_link_cache(struct kmem_cache
*s
, struct mem_cgroup
*memcg
);
413 #else /* CONFIG_MEMCG_KMEM */
415 /* If !memcg, all caches are root. */
416 #define slab_root_caches slab_caches
417 #define root_caches_node list
419 #define for_each_memcg_cache(iter, root) \
420 for ((void)(iter), (void)(root); 0; )
422 static inline bool is_root_cache(struct kmem_cache
*s
)
427 static inline bool slab_equal_or_root(struct kmem_cache
*s
,
428 struct kmem_cache
*p
)
433 static inline const char *cache_name(struct kmem_cache
*s
)
438 static inline struct kmem_cache
*memcg_root_cache(struct kmem_cache
*s
)
443 static inline struct mem_cgroup
*memcg_from_slab_page(struct page
*page
)
448 static inline int memcg_charge_slab(struct page
*page
, gfp_t gfp
, int order
,
449 struct kmem_cache
*s
)
454 static inline void memcg_uncharge_slab(struct page
*page
, int order
,
455 struct kmem_cache
*s
)
459 static inline void slab_init_memcg_params(struct kmem_cache
*s
)
463 static inline void memcg_link_cache(struct kmem_cache
*s
,
464 struct mem_cgroup
*memcg
)
468 #endif /* CONFIG_MEMCG_KMEM */
470 static inline struct kmem_cache
*virt_to_cache(const void *obj
)
474 page
= virt_to_head_page(obj
);
475 if (WARN_ONCE(!PageSlab(page
), "%s: Object is not a Slab page!\n",
478 return page
->slab_cache
;
481 static __always_inline
int charge_slab_page(struct page
*page
,
482 gfp_t gfp
, int order
,
483 struct kmem_cache
*s
)
485 if (is_root_cache(s
)) {
486 mod_node_page_state(page_pgdat(page
), cache_vmstat_idx(s
),
491 return memcg_charge_slab(page
, gfp
, order
, s
);
494 static __always_inline
void uncharge_slab_page(struct page
*page
, int order
,
495 struct kmem_cache
*s
)
497 if (is_root_cache(s
)) {
498 mod_node_page_state(page_pgdat(page
), cache_vmstat_idx(s
),
503 memcg_uncharge_slab(page
, order
, s
);
506 static inline struct kmem_cache
*cache_from_obj(struct kmem_cache
*s
, void *x
)
508 struct kmem_cache
*cachep
;
511 * When kmemcg is not being used, both assignments should return the
512 * same value. but we don't want to pay the assignment price in that
513 * case. If it is not compiled in, the compiler should be smart enough
514 * to not do even the assignment. In that case, slab_equal_or_root
515 * will also be a constant.
517 if (!memcg_kmem_enabled() &&
518 !IS_ENABLED(CONFIG_SLAB_FREELIST_HARDENED
) &&
519 !unlikely(s
->flags
& SLAB_CONSISTENCY_CHECKS
))
522 cachep
= virt_to_cache(x
);
523 WARN_ONCE(cachep
&& !slab_equal_or_root(cachep
, s
),
524 "%s: Wrong slab cache. %s but object is from %s\n",
525 __func__
, s
->name
, cachep
->name
);
529 static inline size_t slab_ksize(const struct kmem_cache
*s
)
532 return s
->object_size
;
534 #else /* CONFIG_SLUB */
535 # ifdef CONFIG_SLUB_DEBUG
537 * Debugging requires use of the padding between object
538 * and whatever may come after it.
540 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
541 return s
->object_size
;
543 if (s
->flags
& SLAB_KASAN
)
544 return s
->object_size
;
546 * If we have the need to store the freelist pointer
547 * back there or track user information then we can
548 * only use the space before that information.
550 if (s
->flags
& (SLAB_TYPESAFE_BY_RCU
| SLAB_STORE_USER
))
553 * Else we can use all the padding etc for the allocation
559 static inline struct kmem_cache
*slab_pre_alloc_hook(struct kmem_cache
*s
,
562 flags
&= gfp_allowed_mask
;
564 fs_reclaim_acquire(flags
);
565 fs_reclaim_release(flags
);
567 might_sleep_if(gfpflags_allow_blocking(flags
));
569 if (should_failslab(s
, flags
))
572 if (memcg_kmem_enabled() &&
573 ((flags
& __GFP_ACCOUNT
) || (s
->flags
& SLAB_ACCOUNT
)))
574 return memcg_kmem_get_cache(s
);
579 static inline void slab_post_alloc_hook(struct kmem_cache
*s
, gfp_t flags
,
580 size_t size
, void **p
)
584 flags
&= gfp_allowed_mask
;
585 for (i
= 0; i
< size
; i
++) {
586 p
[i
] = kasan_slab_alloc(s
, p
[i
], flags
);
587 /* As p[i] might get tagged, call kmemleak hook after KASAN. */
588 kmemleak_alloc_recursive(p
[i
], s
->object_size
, 1,
592 if (memcg_kmem_enabled())
593 memcg_kmem_put_cache(s
);
598 * The slab lists for all objects.
600 struct kmem_cache_node
{
601 spinlock_t list_lock
;
604 struct list_head slabs_partial
; /* partial list first, better asm code */
605 struct list_head slabs_full
;
606 struct list_head slabs_free
;
607 unsigned long total_slabs
; /* length of all slab lists */
608 unsigned long free_slabs
; /* length of free slab list only */
609 unsigned long free_objects
;
610 unsigned int free_limit
;
611 unsigned int colour_next
; /* Per-node cache coloring */
612 struct array_cache
*shared
; /* shared per node */
613 struct alien_cache
**alien
; /* on other nodes */
614 unsigned long next_reap
; /* updated without locking */
615 int free_touched
; /* updated without locking */
619 unsigned long nr_partial
;
620 struct list_head partial
;
621 #ifdef CONFIG_SLUB_DEBUG
622 atomic_long_t nr_slabs
;
623 atomic_long_t total_objects
;
624 struct list_head full
;
630 static inline struct kmem_cache_node
*get_node(struct kmem_cache
*s
, int node
)
632 return s
->node
[node
];
636 * Iterator over all nodes. The body will be executed for each node that has
637 * a kmem_cache_node structure allocated (which is true for all online nodes)
639 #define for_each_kmem_cache_node(__s, __node, __n) \
640 for (__node = 0; __node < nr_node_ids; __node++) \
641 if ((__n = get_node(__s, __node)))
645 void *slab_start(struct seq_file
*m
, loff_t
*pos
);
646 void *slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
);
647 void slab_stop(struct seq_file
*m
, void *p
);
648 void *memcg_slab_start(struct seq_file
*m
, loff_t
*pos
);
649 void *memcg_slab_next(struct seq_file
*m
, void *p
, loff_t
*pos
);
650 void memcg_slab_stop(struct seq_file
*m
, void *p
);
651 int memcg_slab_show(struct seq_file
*m
, void *p
);
653 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
654 void dump_unreclaimable_slab(void);
656 static inline void dump_unreclaimable_slab(void)
661 void ___cache_free(struct kmem_cache
*cache
, void *x
, unsigned long addr
);
663 #ifdef CONFIG_SLAB_FREELIST_RANDOM
664 int cache_random_seq_create(struct kmem_cache
*cachep
, unsigned int count
,
666 void cache_random_seq_destroy(struct kmem_cache
*cachep
);
668 static inline int cache_random_seq_create(struct kmem_cache
*cachep
,
669 unsigned int count
, gfp_t gfp
)
673 static inline void cache_random_seq_destroy(struct kmem_cache
*cachep
) { }
674 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
676 static inline bool slab_want_init_on_alloc(gfp_t flags
, struct kmem_cache
*c
)
678 if (static_branch_unlikely(&init_on_alloc
)) {
681 if (c
->flags
& (SLAB_TYPESAFE_BY_RCU
| SLAB_POISON
))
682 return flags
& __GFP_ZERO
;
685 return flags
& __GFP_ZERO
;
688 static inline bool slab_want_init_on_free(struct kmem_cache
*c
)
690 if (static_branch_unlikely(&init_on_free
))
692 (c
->flags
& (SLAB_TYPESAFE_BY_RCU
| SLAB_POISON
)));
696 #endif /* MM_SLAB_H */