]>
Commit | Line | Data |
---|---|---|
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 | unsigned long calculate_alignment(unsigned long flags, | |
75 | unsigned long align, unsigned long size); | |
76 | ||
77 | #ifndef CONFIG_SLOB | |
78 | /* Kmalloc array related functions */ | |
79 | void setup_kmalloc_cache_index_table(void); | |
80 | void create_kmalloc_caches(unsigned long); | |
81 | ||
82 | /* Find the kmalloc slab corresponding for a certain size */ | |
83 | struct kmem_cache *kmalloc_slab(size_t, gfp_t); | |
84 | #endif | |
85 | ||
86 | ||
87 | /* Functions provided by the slab allocators */ | |
88 | extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags); | |
89 | ||
90 | extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size, | |
91 | unsigned long flags); | |
92 | extern void create_boot_cache(struct kmem_cache *, const char *name, | |
93 | size_t size, unsigned long flags); | |
94 | ||
95 | int slab_unmergeable(struct kmem_cache *s); | |
96 | struct kmem_cache *find_mergeable(size_t size, size_t align, | |
97 | unsigned long flags, const char *name, void (*ctor)(void *)); | |
98 | #ifndef CONFIG_SLOB | |
99 | struct kmem_cache * | |
100 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
101 | unsigned long flags, void (*ctor)(void *)); | |
102 | ||
103 | unsigned long kmem_cache_flags(unsigned long object_size, | |
104 | unsigned long flags, const char *name, | |
105 | void (*ctor)(void *)); | |
106 | #else | |
107 | static inline struct kmem_cache * | |
108 | __kmem_cache_alias(const char *name, size_t size, size_t align, | |
109 | unsigned long flags, void (*ctor)(void *)) | |
110 | { return NULL; } | |
111 | ||
112 | static inline unsigned long kmem_cache_flags(unsigned long object_size, | |
113 | unsigned long flags, const char *name, | |
114 | void (*ctor)(void *)) | |
115 | { | |
116 | return flags; | |
117 | } | |
118 | #endif | |
119 | ||
120 | ||
121 | /* Legal flag mask for kmem_cache_create(), for various configurations */ | |
122 | #define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \ | |
123 | SLAB_DESTROY_BY_RCU | SLAB_DEBUG_OBJECTS ) | |
124 | ||
125 | #if defined(CONFIG_DEBUG_SLAB) | |
126 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) | |
127 | #elif defined(CONFIG_SLUB_DEBUG) | |
128 | #define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
129 | SLAB_TRACE | SLAB_CONSISTENCY_CHECKS) | |
130 | #else | |
131 | #define SLAB_DEBUG_FLAGS (0) | |
132 | #endif | |
133 | ||
134 | #if defined(CONFIG_SLAB) | |
135 | #define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \ | |
136 | SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \ | |
137 | SLAB_NOTRACK | SLAB_ACCOUNT) | |
138 | #elif defined(CONFIG_SLUB) | |
139 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \ | |
140 | SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT) | |
141 | #else | |
142 | #define SLAB_CACHE_FLAGS (0) | |
143 | #endif | |
144 | ||
145 | /* Common flags available with current configuration */ | |
146 | #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS) | |
147 | ||
148 | /* Common flags permitted for kmem_cache_create */ | |
149 | #define SLAB_FLAGS_PERMITTED (SLAB_CORE_FLAGS | \ | |
150 | SLAB_RED_ZONE | \ | |
151 | SLAB_POISON | \ | |
152 | SLAB_STORE_USER | \ | |
153 | SLAB_TRACE | \ | |
154 | SLAB_CONSISTENCY_CHECKS | \ | |
155 | SLAB_MEM_SPREAD | \ | |
156 | SLAB_NOLEAKTRACE | \ | |
157 | SLAB_RECLAIM_ACCOUNT | \ | |
158 | SLAB_TEMPORARY | \ | |
159 | SLAB_NOTRACK | \ | |
160 | SLAB_ACCOUNT) | |
161 | ||
162 | int __kmem_cache_shutdown(struct kmem_cache *); | |
163 | void __kmem_cache_release(struct kmem_cache *); | |
164 | int __kmem_cache_shrink(struct kmem_cache *); | |
165 | void slab_kmem_cache_release(struct kmem_cache *); | |
166 | ||
167 | struct seq_file; | |
168 | struct file; | |
169 | ||
170 | struct slabinfo { | |
171 | unsigned long active_objs; | |
172 | unsigned long num_objs; | |
173 | unsigned long active_slabs; | |
174 | unsigned long num_slabs; | |
175 | unsigned long shared_avail; | |
176 | unsigned int limit; | |
177 | unsigned int batchcount; | |
178 | unsigned int shared; | |
179 | unsigned int objects_per_slab; | |
180 | unsigned int cache_order; | |
181 | }; | |
182 | ||
183 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo); | |
184 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s); | |
185 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, | |
186 | size_t count, loff_t *ppos); | |
187 | ||
188 | /* | |
189 | * Generic implementation of bulk operations | |
190 | * These are useful for situations in which the allocator cannot | |
191 | * perform optimizations. In that case segments of the object listed | |
192 | * may be allocated or freed using these operations. | |
193 | */ | |
194 | void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **); | |
195 | int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **); | |
196 | ||
197 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) | |
198 | /* | |
199 | * Iterate over all memcg caches of the given root cache. The caller must hold | |
200 | * slab_mutex. | |
201 | */ | |
202 | #define for_each_memcg_cache(iter, root) \ | |
203 | list_for_each_entry(iter, &(root)->memcg_params.list, \ | |
204 | memcg_params.list) | |
205 | ||
206 | static inline bool is_root_cache(struct kmem_cache *s) | |
207 | { | |
208 | return s->memcg_params.is_root_cache; | |
209 | } | |
210 | ||
211 | static inline bool slab_equal_or_root(struct kmem_cache *s, | |
212 | struct kmem_cache *p) | |
213 | { | |
214 | return p == s || p == s->memcg_params.root_cache; | |
215 | } | |
216 | ||
217 | /* | |
218 | * We use suffixes to the name in memcg because we can't have caches | |
219 | * created in the system with the same name. But when we print them | |
220 | * locally, better refer to them with the base name | |
221 | */ | |
222 | static inline const char *cache_name(struct kmem_cache *s) | |
223 | { | |
224 | if (!is_root_cache(s)) | |
225 | s = s->memcg_params.root_cache; | |
226 | return s->name; | |
227 | } | |
228 | ||
229 | /* | |
230 | * Note, we protect with RCU only the memcg_caches array, not per-memcg caches. | |
231 | * That said the caller must assure the memcg's cache won't go away by either | |
232 | * taking a css reference to the owner cgroup, or holding the slab_mutex. | |
233 | */ | |
234 | static inline struct kmem_cache * | |
235 | cache_from_memcg_idx(struct kmem_cache *s, int idx) | |
236 | { | |
237 | struct kmem_cache *cachep; | |
238 | struct memcg_cache_array *arr; | |
239 | ||
240 | rcu_read_lock(); | |
241 | arr = rcu_dereference(s->memcg_params.memcg_caches); | |
242 | ||
243 | /* | |
244 | * Make sure we will access the up-to-date value. The code updating | |
245 | * memcg_caches issues a write barrier to match this (see | |
246 | * memcg_create_kmem_cache()). | |
247 | */ | |
248 | cachep = lockless_dereference(arr->entries[idx]); | |
249 | rcu_read_unlock(); | |
250 | ||
251 | return cachep; | |
252 | } | |
253 | ||
254 | static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s) | |
255 | { | |
256 | if (is_root_cache(s)) | |
257 | return s; | |
258 | return s->memcg_params.root_cache; | |
259 | } | |
260 | ||
261 | static __always_inline int memcg_charge_slab(struct page *page, | |
262 | gfp_t gfp, int order, | |
263 | struct kmem_cache *s) | |
264 | { | |
265 | int ret; | |
266 | ||
267 | if (!memcg_kmem_enabled()) | |
268 | return 0; | |
269 | if (is_root_cache(s)) | |
270 | return 0; | |
271 | ||
272 | ret = memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg); | |
273 | if (ret) | |
274 | return ret; | |
275 | ||
276 | memcg_kmem_update_page_stat(page, | |
277 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
278 | MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE, | |
279 | 1 << order); | |
280 | return 0; | |
281 | } | |
282 | ||
283 | static __always_inline void memcg_uncharge_slab(struct page *page, int order, | |
284 | struct kmem_cache *s) | |
285 | { | |
286 | if (!memcg_kmem_enabled()) | |
287 | return; | |
288 | ||
289 | memcg_kmem_update_page_stat(page, | |
290 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
291 | MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE, | |
292 | -(1 << order)); | |
293 | memcg_kmem_uncharge(page, order); | |
294 | } | |
295 | ||
296 | extern void slab_init_memcg_params(struct kmem_cache *); | |
297 | ||
298 | #else /* CONFIG_MEMCG && !CONFIG_SLOB */ | |
299 | ||
300 | #define for_each_memcg_cache(iter, root) \ | |
301 | for ((void)(iter), (void)(root); 0; ) | |
302 | ||
303 | static inline bool is_root_cache(struct kmem_cache *s) | |
304 | { | |
305 | return true; | |
306 | } | |
307 | ||
308 | static inline bool slab_equal_or_root(struct kmem_cache *s, | |
309 | struct kmem_cache *p) | |
310 | { | |
311 | return true; | |
312 | } | |
313 | ||
314 | static inline const char *cache_name(struct kmem_cache *s) | |
315 | { | |
316 | return s->name; | |
317 | } | |
318 | ||
319 | static inline struct kmem_cache * | |
320 | cache_from_memcg_idx(struct kmem_cache *s, int idx) | |
321 | { | |
322 | return NULL; | |
323 | } | |
324 | ||
325 | static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s) | |
326 | { | |
327 | return s; | |
328 | } | |
329 | ||
330 | static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order, | |
331 | struct kmem_cache *s) | |
332 | { | |
333 | return 0; | |
334 | } | |
335 | ||
336 | static inline void memcg_uncharge_slab(struct page *page, int order, | |
337 | struct kmem_cache *s) | |
338 | { | |
339 | } | |
340 | ||
341 | static inline void slab_init_memcg_params(struct kmem_cache *s) | |
342 | { | |
343 | } | |
344 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ | |
345 | ||
346 | static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x) | |
347 | { | |
348 | struct kmem_cache *cachep; | |
349 | struct page *page; | |
350 | ||
351 | /* | |
352 | * When kmemcg is not being used, both assignments should return the | |
353 | * same value. but we don't want to pay the assignment price in that | |
354 | * case. If it is not compiled in, the compiler should be smart enough | |
355 | * to not do even the assignment. In that case, slab_equal_or_root | |
356 | * will also be a constant. | |
357 | */ | |
358 | if (!memcg_kmem_enabled() && | |
359 | !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS)) | |
360 | return s; | |
361 | ||
362 | page = virt_to_head_page(x); | |
363 | cachep = page->slab_cache; | |
364 | if (slab_equal_or_root(cachep, s)) | |
365 | return cachep; | |
366 | ||
367 | pr_err("%s: Wrong slab cache. %s but object is from %s\n", | |
368 | __func__, s->name, cachep->name); | |
369 | WARN_ON_ONCE(1); | |
370 | return s; | |
371 | } | |
372 | ||
373 | static inline size_t slab_ksize(const struct kmem_cache *s) | |
374 | { | |
375 | #ifndef CONFIG_SLUB | |
376 | return s->object_size; | |
377 | ||
378 | #else /* CONFIG_SLUB */ | |
379 | # ifdef CONFIG_SLUB_DEBUG | |
380 | /* | |
381 | * Debugging requires use of the padding between object | |
382 | * and whatever may come after it. | |
383 | */ | |
384 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
385 | return s->object_size; | |
386 | # endif | |
387 | if (s->flags & SLAB_KASAN) | |
388 | return s->object_size; | |
389 | /* | |
390 | * If we have the need to store the freelist pointer | |
391 | * back there or track user information then we can | |
392 | * only use the space before that information. | |
393 | */ | |
394 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
395 | return s->inuse; | |
396 | /* | |
397 | * Else we can use all the padding etc for the allocation | |
398 | */ | |
399 | return s->size; | |
400 | #endif | |
401 | } | |
402 | ||
403 | static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, | |
404 | gfp_t flags) | |
405 | { | |
406 | flags &= gfp_allowed_mask; | |
407 | lockdep_trace_alloc(flags); | |
408 | might_sleep_if(gfpflags_allow_blocking(flags)); | |
409 | ||
410 | if (should_failslab(s, flags)) | |
411 | return NULL; | |
412 | ||
413 | if (memcg_kmem_enabled() && | |
414 | ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT))) | |
415 | return memcg_kmem_get_cache(s); | |
416 | ||
417 | return s; | |
418 | } | |
419 | ||
420 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
421 | size_t size, void **p) | |
422 | { | |
423 | size_t i; | |
424 | ||
425 | flags &= gfp_allowed_mask; | |
426 | for (i = 0; i < size; i++) { | |
427 | void *object = p[i]; | |
428 | ||
429 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); | |
430 | kmemleak_alloc_recursive(object, s->object_size, 1, | |
431 | s->flags, flags); | |
432 | kasan_slab_alloc(s, object, flags); | |
433 | } | |
434 | ||
435 | if (memcg_kmem_enabled()) | |
436 | memcg_kmem_put_cache(s); | |
437 | } | |
438 | ||
439 | #ifndef CONFIG_SLOB | |
440 | /* | |
441 | * The slab lists for all objects. | |
442 | */ | |
443 | struct kmem_cache_node { | |
444 | spinlock_t list_lock; | |
445 | ||
446 | #ifdef CONFIG_SLAB | |
447 | struct list_head slabs_partial; /* partial list first, better asm code */ | |
448 | struct list_head slabs_full; | |
449 | struct list_head slabs_free; | |
450 | unsigned long total_slabs; /* length of all slab lists */ | |
451 | unsigned long free_slabs; /* length of free slab list only */ | |
452 | unsigned long free_objects; | |
453 | unsigned int free_limit; | |
454 | unsigned int colour_next; /* Per-node cache coloring */ | |
455 | struct array_cache *shared; /* shared per node */ | |
456 | struct alien_cache **alien; /* on other nodes */ | |
457 | unsigned long next_reap; /* updated without locking */ | |
458 | int free_touched; /* updated without locking */ | |
459 | #endif | |
460 | ||
461 | #ifdef CONFIG_SLUB | |
462 | unsigned long nr_partial; | |
463 | struct list_head partial; | |
464 | #ifdef CONFIG_SLUB_DEBUG | |
465 | atomic_long_t nr_slabs; | |
466 | atomic_long_t total_objects; | |
467 | struct list_head full; | |
468 | #endif | |
469 | #endif | |
470 | ||
471 | }; | |
472 | ||
473 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) | |
474 | { | |
475 | return s->node[node]; | |
476 | } | |
477 | ||
478 | /* | |
479 | * Iterator over all nodes. The body will be executed for each node that has | |
480 | * a kmem_cache_node structure allocated (which is true for all online nodes) | |
481 | */ | |
482 | #define for_each_kmem_cache_node(__s, __node, __n) \ | |
483 | for (__node = 0; __node < nr_node_ids; __node++) \ | |
484 | if ((__n = get_node(__s, __node))) | |
485 | ||
486 | #endif | |
487 | ||
488 | void *slab_start(struct seq_file *m, loff_t *pos); | |
489 | void *slab_next(struct seq_file *m, void *p, loff_t *pos); | |
490 | void slab_stop(struct seq_file *m, void *p); | |
491 | int memcg_slab_show(struct seq_file *m, void *p); | |
492 | ||
493 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr); | |
494 | ||
495 | #ifdef CONFIG_SLAB_FREELIST_RANDOM | |
496 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
497 | gfp_t gfp); | |
498 | void cache_random_seq_destroy(struct kmem_cache *cachep); | |
499 | #else | |
500 | static inline int cache_random_seq_create(struct kmem_cache *cachep, | |
501 | unsigned int count, gfp_t gfp) | |
502 | { | |
503 | return 0; | |
504 | } | |
505 | static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { } | |
506 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
507 | ||
508 | #endif /* MM_SLAB_H */ |