]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - mm/slab.h
slab: implement slab_root_caches list
[mirror_ubuntu-artful-kernel.git] / mm / slab.h
1 #ifndef MM_SLAB_H
2 #define MM_SLAB_H
3 /*
4 * Internal slab definitions
5 */
6
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 */
19 struct 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>
41 #include <linux/fault-inject.h>
42 #include <linux/kmemcheck.h>
43 #include <linux/kasan.h>
44 #include <linux/kmemleak.h>
45 #include <linux/random.h>
46
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 */
55 enum slab_state {
56 DOWN, /* No slab functionality yet */
57 PARTIAL, /* SLUB: kmem_cache_node available */
58 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
59 UP, /* Slab caches usable but not all extras yet */
60 FULL /* Everything is working */
61 };
62
63 extern enum slab_state slab_state;
64
65 /* The slab cache mutex protects the management structures during changes */
66 extern struct mutex slab_mutex;
67
68 /* The list of all slab caches on the system */
69 extern struct list_head slab_caches;
70
71 /* The slab cache that manages slab cache information */
72 extern struct kmem_cache *kmem_cache;
73
74 /* A table of kmalloc cache names and sizes */
75 extern const struct kmalloc_info_struct {
76 const char *name;
77 unsigned long size;
78 } kmalloc_info[];
79
80 unsigned long calculate_alignment(unsigned long flags,
81 unsigned long align, unsigned long size);
82
83 #ifndef CONFIG_SLOB
84 /* Kmalloc array related functions */
85 void setup_kmalloc_cache_index_table(void);
86 void create_kmalloc_caches(unsigned long);
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
93 /* Functions provided by the slab allocators */
94 extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags);
95
96 extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
97 unsigned long flags);
98 extern void create_boot_cache(struct kmem_cache *, const char *name,
99 size_t size, unsigned long flags);
100
101 int slab_unmergeable(struct kmem_cache *s);
102 struct kmem_cache *find_mergeable(size_t size, size_t align,
103 unsigned long flags, const char *name, void (*ctor)(void *));
104 #ifndef CONFIG_SLOB
105 struct kmem_cache *
106 __kmem_cache_alias(const char *name, size_t size, size_t align,
107 unsigned long flags, void (*ctor)(void *));
108
109 unsigned long kmem_cache_flags(unsigned long object_size,
110 unsigned long flags, const char *name,
111 void (*ctor)(void *));
112 #else
113 static inline struct kmem_cache *
114 __kmem_cache_alias(const char *name, size_t size, size_t align,
115 unsigned long flags, void (*ctor)(void *))
116 { return NULL; }
117
118 static 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 }
124 #endif
125
126
127 /* Legal flag mask for kmem_cache_create(), for various configurations */
128 #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
129 SLAB_DESTROY_BY_RCU | SLAB_DEBUG_OBJECTS )
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 | \
135 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
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 | \
142 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
143 SLAB_NOTRACK | SLAB_ACCOUNT)
144 #elif defined(CONFIG_SLUB)
145 #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
146 SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT)
147 #else
148 #define SLAB_CACHE_FLAGS (0)
149 #endif
150
151 /* Common flags available with current configuration */
152 #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
153
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
168 int __kmem_cache_shutdown(struct kmem_cache *);
169 void __kmem_cache_release(struct kmem_cache *);
170 int __kmem_cache_shrink(struct kmem_cache *, bool);
171 void slab_kmem_cache_release(struct kmem_cache *);
172
173 struct seq_file;
174 struct file;
175
176 struct slabinfo {
177 unsigned long active_objs;
178 unsigned long num_objs;
179 unsigned long active_slabs;
180 unsigned long num_slabs;
181 unsigned long shared_avail;
182 unsigned int limit;
183 unsigned int batchcount;
184 unsigned int shared;
185 unsigned int objects_per_slab;
186 unsigned int cache_order;
187 };
188
189 void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
190 void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
191 ssize_t slabinfo_write(struct file *file, const char __user *buffer,
192 size_t count, loff_t *ppos);
193
194 /*
195 * Generic implementation of bulk operations
196 * These are useful for situations in which the allocator cannot
197 * perform optimizations. In that case segments of the object listed
198 * may be allocated or freed using these operations.
199 */
200 void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
201 int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
202
203 #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
204
205 /* List of all root caches. */
206 extern struct list_head slab_root_caches;
207 #define root_caches_node memcg_params.__root_caches_node
208
209 /*
210 * Iterate over all memcg caches of the given root cache. The caller must hold
211 * slab_mutex.
212 */
213 #define for_each_memcg_cache(iter, root) \
214 list_for_each_entry(iter, &(root)->memcg_params.children, \
215 memcg_params.children_node)
216
217 static inline bool is_root_cache(struct kmem_cache *s)
218 {
219 return !s->memcg_params.root_cache;
220 }
221
222 static inline bool slab_equal_or_root(struct kmem_cache *s,
223 struct kmem_cache *p)
224 {
225 return p == s || p == s->memcg_params.root_cache;
226 }
227
228 /*
229 * We use suffixes to the name in memcg because we can't have caches
230 * created in the system with the same name. But when we print them
231 * locally, better refer to them with the base name
232 */
233 static inline const char *cache_name(struct kmem_cache *s)
234 {
235 if (!is_root_cache(s))
236 s = s->memcg_params.root_cache;
237 return s->name;
238 }
239
240 /*
241 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
242 * That said the caller must assure the memcg's cache won't go away by either
243 * taking a css reference to the owner cgroup, or holding the slab_mutex.
244 */
245 static inline struct kmem_cache *
246 cache_from_memcg_idx(struct kmem_cache *s, int idx)
247 {
248 struct kmem_cache *cachep;
249 struct memcg_cache_array *arr;
250
251 rcu_read_lock();
252 arr = rcu_dereference(s->memcg_params.memcg_caches);
253
254 /*
255 * Make sure we will access the up-to-date value. The code updating
256 * memcg_caches issues a write barrier to match this (see
257 * memcg_create_kmem_cache()).
258 */
259 cachep = lockless_dereference(arr->entries[idx]);
260 rcu_read_unlock();
261
262 return cachep;
263 }
264
265 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
266 {
267 if (is_root_cache(s))
268 return s;
269 return s->memcg_params.root_cache;
270 }
271
272 static __always_inline int memcg_charge_slab(struct page *page,
273 gfp_t gfp, int order,
274 struct kmem_cache *s)
275 {
276 int ret;
277
278 if (!memcg_kmem_enabled())
279 return 0;
280 if (is_root_cache(s))
281 return 0;
282
283 ret = memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg);
284 if (ret)
285 return ret;
286
287 memcg_kmem_update_page_stat(page,
288 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
289 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
290 1 << order);
291 return 0;
292 }
293
294 static __always_inline void memcg_uncharge_slab(struct page *page, int order,
295 struct kmem_cache *s)
296 {
297 if (!memcg_kmem_enabled())
298 return;
299
300 memcg_kmem_update_page_stat(page,
301 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
302 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
303 -(1 << order));
304 memcg_kmem_uncharge(page, order);
305 }
306
307 extern void slab_init_memcg_params(struct kmem_cache *);
308 extern void memcg_link_cache(struct kmem_cache *s);
309
310 #else /* CONFIG_MEMCG && !CONFIG_SLOB */
311
312 /* If !memcg, all caches are root. */
313 #define slab_root_caches slab_caches
314 #define root_caches_node list
315
316 #define for_each_memcg_cache(iter, root) \
317 for ((void)(iter), (void)(root); 0; )
318
319 static inline bool is_root_cache(struct kmem_cache *s)
320 {
321 return true;
322 }
323
324 static inline bool slab_equal_or_root(struct kmem_cache *s,
325 struct kmem_cache *p)
326 {
327 return true;
328 }
329
330 static inline const char *cache_name(struct kmem_cache *s)
331 {
332 return s->name;
333 }
334
335 static inline struct kmem_cache *
336 cache_from_memcg_idx(struct kmem_cache *s, int idx)
337 {
338 return NULL;
339 }
340
341 static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
342 {
343 return s;
344 }
345
346 static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
347 struct kmem_cache *s)
348 {
349 return 0;
350 }
351
352 static inline void memcg_uncharge_slab(struct page *page, int order,
353 struct kmem_cache *s)
354 {
355 }
356
357 static inline void slab_init_memcg_params(struct kmem_cache *s)
358 {
359 }
360
361 static inline void memcg_link_cache(struct kmem_cache *s)
362 {
363 }
364
365 #endif /* CONFIG_MEMCG && !CONFIG_SLOB */
366
367 static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
368 {
369 struct kmem_cache *cachep;
370 struct page *page;
371
372 /*
373 * When kmemcg is not being used, both assignments should return the
374 * same value. but we don't want to pay the assignment price in that
375 * case. If it is not compiled in, the compiler should be smart enough
376 * to not do even the assignment. In that case, slab_equal_or_root
377 * will also be a constant.
378 */
379 if (!memcg_kmem_enabled() &&
380 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
381 return s;
382
383 page = virt_to_head_page(x);
384 cachep = page->slab_cache;
385 if (slab_equal_or_root(cachep, s))
386 return cachep;
387
388 pr_err("%s: Wrong slab cache. %s but object is from %s\n",
389 __func__, s->name, cachep->name);
390 WARN_ON_ONCE(1);
391 return s;
392 }
393
394 static inline size_t slab_ksize(const struct kmem_cache *s)
395 {
396 #ifndef CONFIG_SLUB
397 return s->object_size;
398
399 #else /* CONFIG_SLUB */
400 # ifdef CONFIG_SLUB_DEBUG
401 /*
402 * Debugging requires use of the padding between object
403 * and whatever may come after it.
404 */
405 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
406 return s->object_size;
407 # endif
408 if (s->flags & SLAB_KASAN)
409 return s->object_size;
410 /*
411 * If we have the need to store the freelist pointer
412 * back there or track user information then we can
413 * only use the space before that information.
414 */
415 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
416 return s->inuse;
417 /*
418 * Else we can use all the padding etc for the allocation
419 */
420 return s->size;
421 #endif
422 }
423
424 static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
425 gfp_t flags)
426 {
427 flags &= gfp_allowed_mask;
428 lockdep_trace_alloc(flags);
429 might_sleep_if(gfpflags_allow_blocking(flags));
430
431 if (should_failslab(s, flags))
432 return NULL;
433
434 if (memcg_kmem_enabled() &&
435 ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT)))
436 return memcg_kmem_get_cache(s);
437
438 return s;
439 }
440
441 static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
442 size_t size, void **p)
443 {
444 size_t i;
445
446 flags &= gfp_allowed_mask;
447 for (i = 0; i < size; i++) {
448 void *object = p[i];
449
450 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
451 kmemleak_alloc_recursive(object, s->object_size, 1,
452 s->flags, flags);
453 kasan_slab_alloc(s, object, flags);
454 }
455
456 if (memcg_kmem_enabled())
457 memcg_kmem_put_cache(s);
458 }
459
460 #ifndef CONFIG_SLOB
461 /*
462 * The slab lists for all objects.
463 */
464 struct kmem_cache_node {
465 spinlock_t list_lock;
466
467 #ifdef CONFIG_SLAB
468 struct list_head slabs_partial; /* partial list first, better asm code */
469 struct list_head slabs_full;
470 struct list_head slabs_free;
471 unsigned long total_slabs; /* length of all slab lists */
472 unsigned long free_slabs; /* length of free slab list only */
473 unsigned long free_objects;
474 unsigned int free_limit;
475 unsigned int colour_next; /* Per-node cache coloring */
476 struct array_cache *shared; /* shared per node */
477 struct alien_cache **alien; /* on other nodes */
478 unsigned long next_reap; /* updated without locking */
479 int free_touched; /* updated without locking */
480 #endif
481
482 #ifdef CONFIG_SLUB
483 unsigned long nr_partial;
484 struct list_head partial;
485 #ifdef CONFIG_SLUB_DEBUG
486 atomic_long_t nr_slabs;
487 atomic_long_t total_objects;
488 struct list_head full;
489 #endif
490 #endif
491
492 };
493
494 static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
495 {
496 return s->node[node];
497 }
498
499 /*
500 * Iterator over all nodes. The body will be executed for each node that has
501 * a kmem_cache_node structure allocated (which is true for all online nodes)
502 */
503 #define for_each_kmem_cache_node(__s, __node, __n) \
504 for (__node = 0; __node < nr_node_ids; __node++) \
505 if ((__n = get_node(__s, __node)))
506
507 #endif
508
509 void *slab_start(struct seq_file *m, loff_t *pos);
510 void *slab_next(struct seq_file *m, void *p, loff_t *pos);
511 void slab_stop(struct seq_file *m, void *p);
512 void *memcg_slab_start(struct seq_file *m, loff_t *pos);
513 void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos);
514 void memcg_slab_stop(struct seq_file *m, void *p);
515 int memcg_slab_show(struct seq_file *m, void *p);
516
517 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
518
519 #ifdef CONFIG_SLAB_FREELIST_RANDOM
520 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
521 gfp_t gfp);
522 void cache_random_seq_destroy(struct kmem_cache *cachep);
523 #else
524 static inline int cache_random_seq_create(struct kmem_cache *cachep,
525 unsigned int count, gfp_t gfp)
526 {
527 return 0;
528 }
529 static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { }
530 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
531
532 #endif /* MM_SLAB_H */