]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - mm/slab.h
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
[mirror_ubuntu-artful-kernel.git] / mm / slab.h
CommitLineData
97d06609
CL
1#ifndef MM_SLAB_H
2#define MM_SLAB_H
3/*
4 * Internal slab definitions
5 */
6
07f361b2
JK
7#ifdef CONFIG_SLOB
8/*
9 * Common fields provided in kmem_cache by all slab allocators
10 * This struct is either used directly by the allocator (SLOB)
11 * or the allocator must include definitions for all fields
12 * provided in kmem_cache_common in their definition of kmem_cache.
13 *
14 * Once we can do anonymous structs (C11 standard) we could put a
15 * anonymous struct definition in these allocators so that the
16 * separate allocations in the kmem_cache structure of SLAB and
17 * SLUB is no longer needed.
18 */
19struct kmem_cache {
20 unsigned int object_size;/* The original size of the object */
21 unsigned int size; /* The aligned/padded/added on size */
22 unsigned int align; /* Alignment as calculated */
23 unsigned long flags; /* Active flags on the slab */
24 const char *name; /* Slab name for sysfs */
25 int refcount; /* Use counter */
26 void (*ctor)(void *); /* Called on object slot creation */
27 struct list_head list; /* List of all slab caches on the system */
28};
29
30#endif /* CONFIG_SLOB */
31
32#ifdef CONFIG_SLAB
33#include <linux/slab_def.h>
34#endif
35
36#ifdef CONFIG_SLUB
37#include <linux/slub_def.h>
38#endif
39
40#include <linux/memcontrol.h>
11c7aec2
JDB
41#include <linux/fault-inject.h>
42#include <linux/kmemcheck.h>
43#include <linux/kasan.h>
44#include <linux/kmemleak.h>
7c00fce9 45#include <linux/random.h>
07f361b2 46
97d06609
CL
47/*
48 * State of the slab allocator.
49 *
50 * This is used to describe the states of the allocator during bootup.
51 * Allocators use this to gradually bootstrap themselves. Most allocators
52 * have the problem that the structures used for managing slab caches are
53 * allocated from slab caches themselves.
54 */
55enum slab_state {
56 DOWN, /* No slab functionality yet */
57 PARTIAL, /* SLUB: kmem_cache_node available */
ce8eb6c4 58 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
97d06609
CL
59 UP, /* Slab caches usable but not all extras yet */
60 FULL /* Everything is working */
61};
62
63extern enum slab_state slab_state;
64
18004c5d
CL
65/* The slab cache mutex protects the management structures during changes */
66extern struct mutex slab_mutex;
9b030cb8
CL
67
68/* The list of all slab caches on the system */
18004c5d
CL
69extern struct list_head slab_caches;
70
9b030cb8
CL
71/* The slab cache that manages slab cache information */
72extern struct kmem_cache *kmem_cache;
73
af3b5f87
VB
74/* A table of kmalloc cache names and sizes */
75extern const struct kmalloc_info_struct {
76 const char *name;
77 unsigned long size;
78} kmalloc_info[];
79
45906855
CL
80unsigned long calculate_alignment(unsigned long flags,
81 unsigned long align, unsigned long size);
82
f97d5f63
CL
83#ifndef CONFIG_SLOB
84/* Kmalloc array related functions */
34cc6990 85void setup_kmalloc_cache_index_table(void);
f97d5f63 86void create_kmalloc_caches(unsigned long);
2c59dd65
CL
87
88/* Find the kmalloc slab corresponding for a certain size */
89struct kmem_cache *kmalloc_slab(size_t, gfp_t);
f97d5f63
CL
90#endif
91
92
9b030cb8 93/* Functions provided by the slab allocators */
8a13a4cc 94extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags);
97d06609 95
45530c44
CL
96extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
97 unsigned long flags);
98extern void create_boot_cache(struct kmem_cache *, const char *name,
99 size_t size, unsigned long flags);
100
423c929c
JK
101int slab_unmergeable(struct kmem_cache *s);
102struct kmem_cache *find_mergeable(size_t size, size_t align,
103 unsigned long flags, const char *name, void (*ctor)(void *));
12220dea 104#ifndef CONFIG_SLOB
2633d7a0 105struct kmem_cache *
a44cb944
VD
106__kmem_cache_alias(const char *name, size_t size, size_t align,
107 unsigned long flags, void (*ctor)(void *));
423c929c
JK
108
109unsigned long kmem_cache_flags(unsigned long object_size,
110 unsigned long flags, const char *name,
111 void (*ctor)(void *));
cbb79694 112#else
2633d7a0 113static inline struct kmem_cache *
a44cb944
VD
114__kmem_cache_alias(const char *name, size_t size, size_t align,
115 unsigned long flags, void (*ctor)(void *))
cbb79694 116{ return NULL; }
423c929c
JK
117
118static inline unsigned long kmem_cache_flags(unsigned long object_size,
119 unsigned long flags, const char *name,
120 void (*ctor)(void *))
121{
122 return flags;
123}
cbb79694
CL
124#endif
125
126
d8843922
GC
127/* Legal flag mask for kmem_cache_create(), for various configurations */
128#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
5f0d5a3a 129 SLAB_TYPESAFE_BY_RCU | SLAB_DEBUG_OBJECTS )
d8843922
GC
130
131#if defined(CONFIG_DEBUG_SLAB)
132#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
133#elif defined(CONFIG_SLUB_DEBUG)
134#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
becfda68 135 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
d8843922
GC
136#else
137#define SLAB_DEBUG_FLAGS (0)
138#endif
139
140#if defined(CONFIG_SLAB)
141#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
230e9fc2
VD
142 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
143 SLAB_NOTRACK | SLAB_ACCOUNT)
d8843922
GC
144#elif defined(CONFIG_SLUB)
145#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
230e9fc2 146 SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT)
d8843922
GC
147#else
148#define SLAB_CACHE_FLAGS (0)
149#endif
150
e70954fd 151/* Common flags available with current configuration */
d8843922
GC
152#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
153
e70954fd
TG
154/* Common flags permitted for kmem_cache_create */
155#define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \
156 SLAB_RED_ZONE | \
157 SLAB_POISON | \
158 SLAB_STORE_USER | \
159 SLAB_TRACE | \
160 SLAB_CONSISTENCY_CHECKS | \
161 SLAB_MEM_SPREAD | \
162 SLAB_NOLEAKTRACE | \
163 SLAB_RECLAIM_ACCOUNT | \
164 SLAB_TEMPORARY | \
165 SLAB_NOTRACK | \
166 SLAB_ACCOUNT)
167
945cf2b6 168int __kmem_cache_shutdown(struct kmem_cache *);
52b4b950 169void __kmem_cache_release(struct kmem_cache *);
c9fc5864
TH
170int __kmem_cache_shrink(struct kmem_cache *);
171void __kmemcg_cache_deactivate(struct kmem_cache *s);
41a21285 172void slab_kmem_cache_release(struct kmem_cache *);
945cf2b6 173
b7454ad3
GC
174struct seq_file;
175struct file;
b7454ad3 176
0d7561c6
GC
177struct slabinfo {
178 unsigned long active_objs;
179 unsigned long num_objs;
180 unsigned long active_slabs;
181 unsigned long num_slabs;
182 unsigned long shared_avail;
183 unsigned int limit;
184 unsigned int batchcount;
185 unsigned int shared;
186 unsigned int objects_per_slab;
187 unsigned int cache_order;
188};
189
190void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
191void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
b7454ad3
GC
192ssize_t slabinfo_write(struct file *file, const char __user *buffer,
193 size_t count, loff_t *ppos);
ba6c496e 194
484748f0
CL
195/*
196 * Generic implementation of bulk operations
197 * These are useful for situations in which the allocator cannot
9f706d68 198 * perform optimizations. In that case segments of the object listed
484748f0
CL
199 * may be allocated or freed using these operations.
200 */
201void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
865762a8 202int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
484748f0 203
127424c8 204#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
510ded33
TH
205
206/* List of all root caches. */
207extern struct list_head slab_root_caches;
208#define root_caches_node memcg_params.__root_caches_node
209
426589f5
VD
210/*
211 * Iterate over all memcg caches of the given root cache. The caller must hold
212 * slab_mutex.
213 */
214#define for_each_memcg_cache(iter, root) \
9eeadc8b
TH
215 list_for_each_entry(iter, &(root)->memcg_params.children, \
216 memcg_params.children_node)
426589f5 217
ba6c496e
GC
218static inline bool is_root_cache(struct kmem_cache *s)
219{
9eeadc8b 220 return !s->memcg_params.root_cache;
ba6c496e 221}
2633d7a0 222
b9ce5ef4 223static inline bool slab_equal_or_root(struct kmem_cache *s,
f7ce3190 224 struct kmem_cache *p)
b9ce5ef4 225{
f7ce3190 226 return p == s || p == s->memcg_params.root_cache;
b9ce5ef4 227}
749c5415
GC
228
229/*
230 * We use suffixes to the name in memcg because we can't have caches
231 * created in the system with the same name. But when we print them
232 * locally, better refer to them with the base name
233 */
234static inline const char *cache_name(struct kmem_cache *s)
235{
236 if (!is_root_cache(s))
f7ce3190 237 s = s->memcg_params.root_cache;
749c5415
GC
238 return s->name;
239}
240
f8570263
VD
241/*
242 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
f7ce3190
VD
243 * That said the caller must assure the memcg's cache won't go away by either
244 * taking a css reference to the owner cgroup, or holding the slab_mutex.
f8570263 245 */
2ade4de8
QH
246static inline struct kmem_cache *
247cache_from_memcg_idx(struct kmem_cache *s, int idx)
749c5415 248{
959c8963 249 struct kmem_cache *cachep;
f7ce3190 250 struct memcg_cache_array *arr;
f8570263
VD
251
252 rcu_read_lock();
f7ce3190 253 arr = rcu_dereference(s->memcg_params.memcg_caches);
959c8963
VD
254
255 /*
256 * Make sure we will access the up-to-date value. The code updating
257 * memcg_caches issues a write barrier to match this (see
f7ce3190 258 * memcg_create_kmem_cache()).
959c8963 259 */
f7ce3190 260 cachep = lockless_dereference(arr->entries[idx]);
8df0c2dc
PK
261 rcu_read_unlock();
262
959c8963 263 return cachep;
749c5415 264}
943a451a
GC
265
266static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
267{
268 if (is_root_cache(s))
269 return s;
f7ce3190 270 return s->memcg_params.root_cache;
943a451a 271}
5dfb4175 272
f3ccb2c4
VD
273static __always_inline int memcg_charge_slab(struct page *page,
274 gfp_t gfp, int order,
275 struct kmem_cache *s)
5dfb4175
VD
276{
277 if (!memcg_kmem_enabled())
278 return 0;
279 if (is_root_cache(s))
280 return 0;
7779f212 281 return memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
27ee57c9
VD
282}
283
284static __always_inline void memcg_uncharge_slab(struct page *page, int order,
285 struct kmem_cache *s)
286{
45264778
VD
287 if (!memcg_kmem_enabled())
288 return;
27ee57c9 289 memcg_kmem_uncharge(page, order);
5dfb4175 290}
f7ce3190
VD
291
292extern void slab_init_memcg_params(struct kmem_cache *);
510ded33 293extern void memcg_link_cache(struct kmem_cache *s);
01fb58bc
TH
294extern void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s,
295 void (*deact_fn)(struct kmem_cache *));
f7ce3190 296
127424c8 297#else /* CONFIG_MEMCG && !CONFIG_SLOB */
f7ce3190 298
510ded33
TH
299/* If !memcg, all caches are root. */
300#define slab_root_caches slab_caches
301#define root_caches_node list
302
426589f5
VD
303#define for_each_memcg_cache(iter, root) \
304 for ((void)(iter), (void)(root); 0; )
426589f5 305
ba6c496e
GC
306static inline bool is_root_cache(struct kmem_cache *s)
307{
308 return true;
309}
310
b9ce5ef4
GC
311static inline bool slab_equal_or_root(struct kmem_cache *s,
312 struct kmem_cache *p)
313{
314 return true;
315}
749c5415
GC
316
317static inline const char *cache_name(struct kmem_cache *s)
318{
319 return s->name;
320}
321
2ade4de8
QH
322static inline struct kmem_cache *
323cache_from_memcg_idx(struct kmem_cache *s, int idx)
749c5415
GC
324{
325 return NULL;
326}
943a451a
GC
327
328static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
329{
330 return s;
331}
5dfb4175 332
f3ccb2c4
VD
333static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
334 struct kmem_cache *s)
5dfb4175
VD
335{
336 return 0;
337}
338
27ee57c9
VD
339static inline void memcg_uncharge_slab(struct page *page, int order,
340 struct kmem_cache *s)
341{
342}
343
f7ce3190
VD
344static inline void slab_init_memcg_params(struct kmem_cache *s)
345{
346}
510ded33
TH
347
348static inline void memcg_link_cache(struct kmem_cache *s)
349{
350}
351
127424c8 352#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
b9ce5ef4
GC
353
354static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
355{
356 struct kmem_cache *cachep;
357 struct page *page;
358
359 /*
360 * When kmemcg is not being used, both assignments should return the
361 * same value. but we don't want to pay the assignment price in that
362 * case. If it is not compiled in, the compiler should be smart enough
363 * to not do even the assignment. In that case, slab_equal_or_root
364 * will also be a constant.
365 */
becfda68
LA
366 if (!memcg_kmem_enabled() &&
367 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
b9ce5ef4
GC
368 return s;
369
370 page = virt_to_head_page(x);
371 cachep = page->slab_cache;
372 if (slab_equal_or_root(cachep, s))
373 return cachep;
374
375 pr_err("%s: Wrong slab cache. %s but object is from %s\n",
2d16e0fd 376 __func__, s->name, cachep->name);
b9ce5ef4
GC
377 WARN_ON_ONCE(1);
378 return s;
379}
ca34956b 380
11c7aec2
JDB
381static inline size_t slab_ksize(const struct kmem_cache *s)
382{
383#ifndef CONFIG_SLUB
384 return s->object_size;
385
386#else /* CONFIG_SLUB */
387# ifdef CONFIG_SLUB_DEBUG
388 /*
389 * Debugging requires use of the padding between object
390 * and whatever may come after it.
391 */
392 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
393 return s->object_size;
394# endif
80a9201a
AP
395 if (s->flags & SLAB_KASAN)
396 return s->object_size;
11c7aec2
JDB
397 /*
398 * If we have the need to store the freelist pointer
399 * back there or track user information then we can
400 * only use the space before that information.
401 */
5f0d5a3a 402 if (s->flags & (SLAB_TYPESAFE_BY_RCU | SLAB_STORE_USER))
11c7aec2
JDB
403 return s->inuse;
404 /*
405 * Else we can use all the padding etc for the allocation
406 */
407 return s->size;
408#endif
409}
410
411static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
412 gfp_t flags)
413{
414 flags &= gfp_allowed_mask;
415 lockdep_trace_alloc(flags);
416 might_sleep_if(gfpflags_allow_blocking(flags));
417
fab9963a 418 if (should_failslab(s, flags))
11c7aec2
JDB
419 return NULL;
420
45264778
VD
421 if (memcg_kmem_enabled() &&
422 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
423 return memcg_kmem_get_cache(s);
424
425 return s;
11c7aec2
JDB
426}
427
428static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
429 size_t size, void **p)
430{
431 size_t i;
432
433 flags &= gfp_allowed_mask;
434 for (i = 0; i < size; i++) {
435 void *object = p[i];
436
437 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
438 kmemleak_alloc_recursive(object, s->object_size, 1,
439 s->flags, flags);
505f5dcb 440 kasan_slab_alloc(s, object, flags);
11c7aec2 441 }
45264778
VD
442
443 if (memcg_kmem_enabled())
444 memcg_kmem_put_cache(s);
11c7aec2
JDB
445}
446
44c5356f 447#ifndef CONFIG_SLOB
ca34956b
CL
448/*
449 * The slab lists for all objects.
450 */
451struct kmem_cache_node {
452 spinlock_t list_lock;
453
454#ifdef CONFIG_SLAB
455 struct list_head slabs_partial; /* partial list first, better asm code */
456 struct list_head slabs_full;
457 struct list_head slabs_free;
bf00bd34
DR
458 unsigned long total_slabs; /* length of all slab lists */
459 unsigned long free_slabs; /* length of free slab list only */
ca34956b
CL
460 unsigned long free_objects;
461 unsigned int free_limit;
462 unsigned int colour_next; /* Per-node cache coloring */
463 struct array_cache *shared; /* shared per node */
c8522a3a 464 struct alien_cache **alien; /* on other nodes */
ca34956b
CL
465 unsigned long next_reap; /* updated without locking */
466 int free_touched; /* updated without locking */
467#endif
468
469#ifdef CONFIG_SLUB
470 unsigned long nr_partial;
471 struct list_head partial;
472#ifdef CONFIG_SLUB_DEBUG
473 atomic_long_t nr_slabs;
474 atomic_long_t total_objects;
475 struct list_head full;
476#endif
477#endif
478
479};
e25839f6 480
44c5356f
CL
481static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
482{
483 return s->node[node];
484}
485
486/*
487 * Iterator over all nodes. The body will be executed for each node that has
488 * a kmem_cache_node structure allocated (which is true for all online nodes)
489 */
490#define for_each_kmem_cache_node(__s, __node, __n) \
9163582c
MP
491 for (__node = 0; __node < nr_node_ids; __node++) \
492 if ((__n = get_node(__s, __node)))
44c5356f
CL
493
494#endif
495
1df3b26f 496void *slab_start(struct seq_file *m, loff_t *pos);
276a2439
WL
497void *slab_next(struct seq_file *m, void *p, loff_t *pos);
498void slab_stop(struct seq_file *m, void *p);
bc2791f8
TH
499void *memcg_slab_start(struct seq_file *m, loff_t *pos);
500void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
501void memcg_slab_stop(struct seq_file *m, void *p);
b047501c 502int memcg_slab_show(struct seq_file *m, void *p);
5240ab40 503
55834c59
AP
504void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
505
7c00fce9
TG
506#ifdef CONFIG_SLAB_FREELIST_RANDOM
507int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
508 gfp_t gfp);
509void cache_random_seq_destroy(struct kmem_cache *cachep);
510#else
511static inline int cache_random_seq_create(struct kmem_cache *cachep,
512 unsigned int count, gfp_t gfp)
513{
514 return 0;
515}
516static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
517#endif /* CONFIG_SLAB_FREELIST_RANDOM */
518
5240ab40 519#endif /* MM_SLAB_H */