]>
Commit | Line | Data |
---|---|---|
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 | */ | |
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> | |
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 | */ | |
55 | enum 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 | ||
63 | extern enum slab_state slab_state; | |
64 | ||
18004c5d CL |
65 | /* The slab cache mutex protects the management structures during changes */ |
66 | extern struct mutex slab_mutex; | |
9b030cb8 CL |
67 | |
68 | /* The list of all slab caches on the system */ | |
18004c5d CL |
69 | extern struct list_head slab_caches; |
70 | ||
9b030cb8 CL |
71 | /* The slab cache that manages slab cache information */ |
72 | extern struct kmem_cache *kmem_cache; | |
73 | ||
45906855 CL |
74 | unsigned long calculate_alignment(unsigned long flags, |
75 | unsigned long align, unsigned long size); | |
76 | ||
f97d5f63 CL |
77 | #ifndef CONFIG_SLOB |
78 | /* Kmalloc array related functions */ | |
34cc6990 | 79 | void setup_kmalloc_cache_index_table(void); |
f97d5f63 | 80 | void create_kmalloc_caches(unsigned long); |
2c59dd65 CL |
81 | |
82 | /* Find the kmalloc slab corresponding for a certain size */ | |
83 | struct kmem_cache *kmalloc_slab(size_t, gfp_t); | |
f97d5f63 CL |
84 | #endif |
85 | ||
86 | ||
9b030cb8 | 87 | /* Functions provided by the slab allocators */ |
8a13a4cc | 88 | extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags); |
97d06609 | 89 | |
45530c44 CL |
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 | ||
423c929c JK |
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 *)); | |
12220dea | 98 | #ifndef CONFIG_SLOB |
2633d7a0 | 99 | struct kmem_cache * |
a44cb944 VD |
100 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
101 | unsigned long flags, void (*ctor)(void *)); | |
423c929c JK |
102 | |
103 | unsigned long kmem_cache_flags(unsigned long object_size, | |
104 | unsigned long flags, const char *name, | |
105 | void (*ctor)(void *)); | |
cbb79694 | 106 | #else |
2633d7a0 | 107 | static inline struct kmem_cache * |
a44cb944 VD |
108 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
109 | unsigned long flags, void (*ctor)(void *)) | |
cbb79694 | 110 | { return NULL; } |
423c929c JK |
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 | } | |
cbb79694 CL |
118 | #endif |
119 | ||
120 | ||
d8843922 GC |
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 | \ | |
becfda68 | 129 | SLAB_TRACE | SLAB_CONSISTENCY_CHECKS) |
d8843922 GC |
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 | \ | |
230e9fc2 VD |
136 | SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \ |
137 | SLAB_NOTRACK | SLAB_ACCOUNT) | |
d8843922 GC |
138 | #elif defined(CONFIG_SLUB) |
139 | #define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \ | |
230e9fc2 | 140 | SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT) |
d8843922 GC |
141 | #else |
142 | #define SLAB_CACHE_FLAGS (0) | |
143 | #endif | |
144 | ||
145 | #define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS) | |
146 | ||
945cf2b6 | 147 | int __kmem_cache_shutdown(struct kmem_cache *); |
52b4b950 | 148 | void __kmem_cache_release(struct kmem_cache *); |
d6e0b7fa | 149 | int __kmem_cache_shrink(struct kmem_cache *, bool); |
41a21285 | 150 | void slab_kmem_cache_release(struct kmem_cache *); |
945cf2b6 | 151 | |
b7454ad3 GC |
152 | struct seq_file; |
153 | struct file; | |
b7454ad3 | 154 | |
0d7561c6 GC |
155 | struct slabinfo { |
156 | unsigned long active_objs; | |
157 | unsigned long num_objs; | |
158 | unsigned long active_slabs; | |
159 | unsigned long num_slabs; | |
160 | unsigned long shared_avail; | |
161 | unsigned int limit; | |
162 | unsigned int batchcount; | |
163 | unsigned int shared; | |
164 | unsigned int objects_per_slab; | |
165 | unsigned int cache_order; | |
166 | }; | |
167 | ||
168 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo); | |
169 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s); | |
b7454ad3 GC |
170 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
171 | size_t count, loff_t *ppos); | |
ba6c496e | 172 | |
484748f0 CL |
173 | /* |
174 | * Generic implementation of bulk operations | |
175 | * These are useful for situations in which the allocator cannot | |
9f706d68 | 176 | * perform optimizations. In that case segments of the object listed |
484748f0 CL |
177 | * may be allocated or freed using these operations. |
178 | */ | |
179 | void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **); | |
865762a8 | 180 | int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **); |
484748f0 | 181 | |
127424c8 | 182 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
426589f5 VD |
183 | /* |
184 | * Iterate over all memcg caches of the given root cache. The caller must hold | |
185 | * slab_mutex. | |
186 | */ | |
187 | #define for_each_memcg_cache(iter, root) \ | |
188 | list_for_each_entry(iter, &(root)->memcg_params.list, \ | |
189 | memcg_params.list) | |
190 | ||
ba6c496e GC |
191 | static inline bool is_root_cache(struct kmem_cache *s) |
192 | { | |
f7ce3190 | 193 | return s->memcg_params.is_root_cache; |
ba6c496e | 194 | } |
2633d7a0 | 195 | |
b9ce5ef4 | 196 | static inline bool slab_equal_or_root(struct kmem_cache *s, |
f7ce3190 | 197 | struct kmem_cache *p) |
b9ce5ef4 | 198 | { |
f7ce3190 | 199 | return p == s || p == s->memcg_params.root_cache; |
b9ce5ef4 | 200 | } |
749c5415 GC |
201 | |
202 | /* | |
203 | * We use suffixes to the name in memcg because we can't have caches | |
204 | * created in the system with the same name. But when we print them | |
205 | * locally, better refer to them with the base name | |
206 | */ | |
207 | static inline const char *cache_name(struct kmem_cache *s) | |
208 | { | |
209 | if (!is_root_cache(s)) | |
f7ce3190 | 210 | s = s->memcg_params.root_cache; |
749c5415 GC |
211 | return s->name; |
212 | } | |
213 | ||
f8570263 VD |
214 | /* |
215 | * Note, we protect with RCU only the memcg_caches array, not per-memcg caches. | |
f7ce3190 VD |
216 | * That said the caller must assure the memcg's cache won't go away by either |
217 | * taking a css reference to the owner cgroup, or holding the slab_mutex. | |
f8570263 | 218 | */ |
2ade4de8 QH |
219 | static inline struct kmem_cache * |
220 | cache_from_memcg_idx(struct kmem_cache *s, int idx) | |
749c5415 | 221 | { |
959c8963 | 222 | struct kmem_cache *cachep; |
f7ce3190 | 223 | struct memcg_cache_array *arr; |
f8570263 VD |
224 | |
225 | rcu_read_lock(); | |
f7ce3190 | 226 | arr = rcu_dereference(s->memcg_params.memcg_caches); |
959c8963 VD |
227 | |
228 | /* | |
229 | * Make sure we will access the up-to-date value. The code updating | |
230 | * memcg_caches issues a write barrier to match this (see | |
f7ce3190 | 231 | * memcg_create_kmem_cache()). |
959c8963 | 232 | */ |
f7ce3190 | 233 | cachep = lockless_dereference(arr->entries[idx]); |
8df0c2dc PK |
234 | rcu_read_unlock(); |
235 | ||
959c8963 | 236 | return cachep; |
749c5415 | 237 | } |
943a451a GC |
238 | |
239 | static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s) | |
240 | { | |
241 | if (is_root_cache(s)) | |
242 | return s; | |
f7ce3190 | 243 | return s->memcg_params.root_cache; |
943a451a | 244 | } |
5dfb4175 | 245 | |
f3ccb2c4 VD |
246 | static __always_inline int memcg_charge_slab(struct page *page, |
247 | gfp_t gfp, int order, | |
248 | struct kmem_cache *s) | |
5dfb4175 | 249 | { |
27ee57c9 VD |
250 | int ret; |
251 | ||
5dfb4175 VD |
252 | if (!memcg_kmem_enabled()) |
253 | return 0; | |
254 | if (is_root_cache(s)) | |
255 | return 0; | |
27ee57c9 | 256 | |
45264778 | 257 | ret = memcg_kmem_charge_memcg(page, gfp, order, s->memcg_params.memcg); |
27ee57c9 VD |
258 | if (ret) |
259 | return ret; | |
260 | ||
261 | memcg_kmem_update_page_stat(page, | |
262 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
263 | MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE, | |
264 | 1 << order); | |
265 | return 0; | |
266 | } | |
267 | ||
268 | static __always_inline void memcg_uncharge_slab(struct page *page, int order, | |
269 | struct kmem_cache *s) | |
270 | { | |
45264778 VD |
271 | if (!memcg_kmem_enabled()) |
272 | return; | |
273 | ||
27ee57c9 VD |
274 | memcg_kmem_update_page_stat(page, |
275 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
276 | MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE, | |
277 | -(1 << order)); | |
278 | memcg_kmem_uncharge(page, order); | |
5dfb4175 | 279 | } |
f7ce3190 VD |
280 | |
281 | extern void slab_init_memcg_params(struct kmem_cache *); | |
282 | ||
127424c8 | 283 | #else /* CONFIG_MEMCG && !CONFIG_SLOB */ |
f7ce3190 | 284 | |
426589f5 VD |
285 | #define for_each_memcg_cache(iter, root) \ |
286 | for ((void)(iter), (void)(root); 0; ) | |
426589f5 | 287 | |
ba6c496e GC |
288 | static inline bool is_root_cache(struct kmem_cache *s) |
289 | { | |
290 | return true; | |
291 | } | |
292 | ||
b9ce5ef4 GC |
293 | static inline bool slab_equal_or_root(struct kmem_cache *s, |
294 | struct kmem_cache *p) | |
295 | { | |
296 | return true; | |
297 | } | |
749c5415 GC |
298 | |
299 | static inline const char *cache_name(struct kmem_cache *s) | |
300 | { | |
301 | return s->name; | |
302 | } | |
303 | ||
2ade4de8 QH |
304 | static inline struct kmem_cache * |
305 | cache_from_memcg_idx(struct kmem_cache *s, int idx) | |
749c5415 GC |
306 | { |
307 | return NULL; | |
308 | } | |
943a451a GC |
309 | |
310 | static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s) | |
311 | { | |
312 | return s; | |
313 | } | |
5dfb4175 | 314 | |
f3ccb2c4 VD |
315 | static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order, |
316 | struct kmem_cache *s) | |
5dfb4175 VD |
317 | { |
318 | return 0; | |
319 | } | |
320 | ||
27ee57c9 VD |
321 | static inline void memcg_uncharge_slab(struct page *page, int order, |
322 | struct kmem_cache *s) | |
323 | { | |
324 | } | |
325 | ||
f7ce3190 VD |
326 | static inline void slab_init_memcg_params(struct kmem_cache *s) |
327 | { | |
328 | } | |
127424c8 | 329 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
b9ce5ef4 GC |
330 | |
331 | static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x) | |
332 | { | |
333 | struct kmem_cache *cachep; | |
334 | struct page *page; | |
335 | ||
336 | /* | |
337 | * When kmemcg is not being used, both assignments should return the | |
338 | * same value. but we don't want to pay the assignment price in that | |
339 | * case. If it is not compiled in, the compiler should be smart enough | |
340 | * to not do even the assignment. In that case, slab_equal_or_root | |
341 | * will also be a constant. | |
342 | */ | |
becfda68 LA |
343 | if (!memcg_kmem_enabled() && |
344 | !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS)) | |
b9ce5ef4 GC |
345 | return s; |
346 | ||
347 | page = virt_to_head_page(x); | |
348 | cachep = page->slab_cache; | |
349 | if (slab_equal_or_root(cachep, s)) | |
350 | return cachep; | |
351 | ||
352 | pr_err("%s: Wrong slab cache. %s but object is from %s\n", | |
2d16e0fd | 353 | __func__, s->name, cachep->name); |
b9ce5ef4 GC |
354 | WARN_ON_ONCE(1); |
355 | return s; | |
356 | } | |
ca34956b | 357 | |
11c7aec2 JDB |
358 | static inline size_t slab_ksize(const struct kmem_cache *s) |
359 | { | |
360 | #ifndef CONFIG_SLUB | |
361 | return s->object_size; | |
362 | ||
363 | #else /* CONFIG_SLUB */ | |
364 | # ifdef CONFIG_SLUB_DEBUG | |
365 | /* | |
366 | * Debugging requires use of the padding between object | |
367 | * and whatever may come after it. | |
368 | */ | |
369 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
370 | return s->object_size; | |
371 | # endif | |
80a9201a AP |
372 | if (s->flags & SLAB_KASAN) |
373 | return s->object_size; | |
11c7aec2 JDB |
374 | /* |
375 | * If we have the need to store the freelist pointer | |
376 | * back there or track user information then we can | |
377 | * only use the space before that information. | |
378 | */ | |
379 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
380 | return s->inuse; | |
381 | /* | |
382 | * Else we can use all the padding etc for the allocation | |
383 | */ | |
384 | return s->size; | |
385 | #endif | |
386 | } | |
387 | ||
388 | static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, | |
389 | gfp_t flags) | |
390 | { | |
391 | flags &= gfp_allowed_mask; | |
392 | lockdep_trace_alloc(flags); | |
393 | might_sleep_if(gfpflags_allow_blocking(flags)); | |
394 | ||
fab9963a | 395 | if (should_failslab(s, flags)) |
11c7aec2 JDB |
396 | return NULL; |
397 | ||
45264778 VD |
398 | if (memcg_kmem_enabled() && |
399 | ((flags & __GFP_ACCOUNT) || (s->flags & SLAB_ACCOUNT))) | |
400 | return memcg_kmem_get_cache(s); | |
401 | ||
402 | return s; | |
11c7aec2 JDB |
403 | } |
404 | ||
405 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
406 | size_t size, void **p) | |
407 | { | |
408 | size_t i; | |
409 | ||
410 | flags &= gfp_allowed_mask; | |
411 | for (i = 0; i < size; i++) { | |
412 | void *object = p[i]; | |
413 | ||
414 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); | |
415 | kmemleak_alloc_recursive(object, s->object_size, 1, | |
416 | s->flags, flags); | |
505f5dcb | 417 | kasan_slab_alloc(s, object, flags); |
11c7aec2 | 418 | } |
45264778 VD |
419 | |
420 | if (memcg_kmem_enabled()) | |
421 | memcg_kmem_put_cache(s); | |
11c7aec2 JDB |
422 | } |
423 | ||
44c5356f | 424 | #ifndef CONFIG_SLOB |
ca34956b CL |
425 | /* |
426 | * The slab lists for all objects. | |
427 | */ | |
428 | struct kmem_cache_node { | |
429 | spinlock_t list_lock; | |
430 | ||
431 | #ifdef CONFIG_SLAB | |
432 | struct list_head slabs_partial; /* partial list first, better asm code */ | |
433 | struct list_head slabs_full; | |
434 | struct list_head slabs_free; | |
435 | unsigned long free_objects; | |
436 | unsigned int free_limit; | |
437 | unsigned int colour_next; /* Per-node cache coloring */ | |
438 | struct array_cache *shared; /* shared per node */ | |
c8522a3a | 439 | struct alien_cache **alien; /* on other nodes */ |
ca34956b CL |
440 | unsigned long next_reap; /* updated without locking */ |
441 | int free_touched; /* updated without locking */ | |
442 | #endif | |
443 | ||
444 | #ifdef CONFIG_SLUB | |
445 | unsigned long nr_partial; | |
446 | struct list_head partial; | |
447 | #ifdef CONFIG_SLUB_DEBUG | |
448 | atomic_long_t nr_slabs; | |
449 | atomic_long_t total_objects; | |
450 | struct list_head full; | |
451 | #endif | |
452 | #endif | |
453 | ||
454 | }; | |
e25839f6 | 455 | |
44c5356f CL |
456 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
457 | { | |
458 | return s->node[node]; | |
459 | } | |
460 | ||
461 | /* | |
462 | * Iterator over all nodes. The body will be executed for each node that has | |
463 | * a kmem_cache_node structure allocated (which is true for all online nodes) | |
464 | */ | |
465 | #define for_each_kmem_cache_node(__s, __node, __n) \ | |
9163582c MP |
466 | for (__node = 0; __node < nr_node_ids; __node++) \ |
467 | if ((__n = get_node(__s, __node))) | |
44c5356f CL |
468 | |
469 | #endif | |
470 | ||
1df3b26f | 471 | void *slab_start(struct seq_file *m, loff_t *pos); |
276a2439 WL |
472 | void *slab_next(struct seq_file *m, void *p, loff_t *pos); |
473 | void slab_stop(struct seq_file *m, void *p); | |
b047501c | 474 | int memcg_slab_show(struct seq_file *m, void *p); |
5240ab40 | 475 | |
55834c59 AP |
476 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr); |
477 | ||
7c00fce9 TG |
478 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
479 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
480 | gfp_t gfp); | |
481 | void cache_random_seq_destroy(struct kmem_cache *cachep); | |
482 | #else | |
483 | static inline int cache_random_seq_create(struct kmem_cache *cachep, | |
484 | unsigned int count, gfp_t gfp) | |
485 | { | |
486 | return 0; | |
487 | } | |
488 | static inline void cache_random_seq_destroy(struct kmem_cache *cachep) { } | |
489 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
490 | ||
5240ab40 | 491 | #endif /* MM_SLAB_H */ |