]>
Commit | Line | Data |
---|---|---|
039363f3 CL |
1 | /* |
2 | * Slab allocator functions that are independent of the allocator strategy | |
3 | * | |
4 | * (C) 2012 Christoph Lameter <cl@linux.com> | |
5 | */ | |
6 | #include <linux/slab.h> | |
7 | ||
8 | #include <linux/mm.h> | |
9 | #include <linux/poison.h> | |
10 | #include <linux/interrupt.h> | |
11 | #include <linux/memory.h> | |
12 | #include <linux/compiler.h> | |
13 | #include <linux/module.h> | |
20cea968 CL |
14 | #include <linux/cpu.h> |
15 | #include <linux/uaccess.h> | |
b7454ad3 GC |
16 | #include <linux/seq_file.h> |
17 | #include <linux/proc_fs.h> | |
039363f3 CL |
18 | #include <asm/cacheflush.h> |
19 | #include <asm/tlbflush.h> | |
20 | #include <asm/page.h> | |
2633d7a0 | 21 | #include <linux/memcontrol.h> |
928cec9c AR |
22 | |
23 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 24 | #include <trace/events/kmem.h> |
039363f3 | 25 | |
97d06609 CL |
26 | #include "slab.h" |
27 | ||
28 | enum slab_state slab_state; | |
18004c5d CL |
29 | LIST_HEAD(slab_caches); |
30 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 31 | struct kmem_cache *kmem_cache; |
97d06609 | 32 | |
657dc2f9 TH |
33 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
34 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
35 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
36 | slab_caches_to_rcu_destroy_workfn); | |
37 | ||
423c929c JK |
38 | /* |
39 | * Set of flags that will prevent slab merging | |
40 | */ | |
41 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
42 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ | |
7ed2f9e6 | 43 | SLAB_FAILSLAB | SLAB_KASAN) |
423c929c | 44 | |
230e9fc2 VD |
45 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
46 | SLAB_NOTRACK | SLAB_ACCOUNT) | |
423c929c JK |
47 | |
48 | /* | |
49 | * Merge control. If this is set then no merging of slab caches will occur. | |
50 | * (Could be removed. This was introduced to pacify the merge skeptics.) | |
51 | */ | |
52 | static int slab_nomerge; | |
53 | ||
54 | static int __init setup_slab_nomerge(char *str) | |
55 | { | |
56 | slab_nomerge = 1; | |
57 | return 1; | |
58 | } | |
59 | ||
60 | #ifdef CONFIG_SLUB | |
61 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
62 | #endif | |
63 | ||
64 | __setup("slab_nomerge", setup_slab_nomerge); | |
65 | ||
07f361b2 JK |
66 | /* |
67 | * Determine the size of a slab object | |
68 | */ | |
69 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
70 | { | |
71 | return s->object_size; | |
72 | } | |
73 | EXPORT_SYMBOL(kmem_cache_size); | |
74 | ||
77be4b13 | 75 | #ifdef CONFIG_DEBUG_VM |
794b1248 | 76 | static int kmem_cache_sanity_check(const char *name, size_t size) |
039363f3 CL |
77 | { |
78 | struct kmem_cache *s = NULL; | |
79 | ||
039363f3 CL |
80 | if (!name || in_interrupt() || size < sizeof(void *) || |
81 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
82 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
83 | return -EINVAL; | |
039363f3 | 84 | } |
b920536a | 85 | |
20cea968 CL |
86 | list_for_each_entry(s, &slab_caches, list) { |
87 | char tmp; | |
88 | int res; | |
89 | ||
90 | /* | |
91 | * This happens when the module gets unloaded and doesn't | |
92 | * destroy its slab cache and no-one else reuses the vmalloc | |
93 | * area of the module. Print a warning. | |
94 | */ | |
95 | res = probe_kernel_address(s->name, tmp); | |
96 | if (res) { | |
77be4b13 | 97 | pr_err("Slab cache with size %d has lost its name\n", |
20cea968 CL |
98 | s->object_size); |
99 | continue; | |
100 | } | |
20cea968 CL |
101 | } |
102 | ||
103 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
77be4b13 SK |
104 | return 0; |
105 | } | |
106 | #else | |
794b1248 | 107 | static inline int kmem_cache_sanity_check(const char *name, size_t size) |
77be4b13 SK |
108 | { |
109 | return 0; | |
110 | } | |
20cea968 CL |
111 | #endif |
112 | ||
484748f0 CL |
113 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
114 | { | |
115 | size_t i; | |
116 | ||
ca257195 JDB |
117 | for (i = 0; i < nr; i++) { |
118 | if (s) | |
119 | kmem_cache_free(s, p[i]); | |
120 | else | |
121 | kfree(p[i]); | |
122 | } | |
484748f0 CL |
123 | } |
124 | ||
865762a8 | 125 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
126 | void **p) |
127 | { | |
128 | size_t i; | |
129 | ||
130 | for (i = 0; i < nr; i++) { | |
131 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
132 | if (!x) { | |
133 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 134 | return 0; |
484748f0 CL |
135 | } |
136 | } | |
865762a8 | 137 | return i; |
484748f0 CL |
138 | } |
139 | ||
127424c8 | 140 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
f7ce3190 | 141 | void slab_init_memcg_params(struct kmem_cache *s) |
33a690c4 | 142 | { |
f7ce3190 | 143 | s->memcg_params.is_root_cache = true; |
426589f5 | 144 | INIT_LIST_HEAD(&s->memcg_params.list); |
f7ce3190 VD |
145 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
146 | } | |
147 | ||
148 | static int init_memcg_params(struct kmem_cache *s, | |
149 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
150 | { | |
151 | struct memcg_cache_array *arr; | |
33a690c4 | 152 | |
f7ce3190 VD |
153 | if (memcg) { |
154 | s->memcg_params.is_root_cache = false; | |
155 | s->memcg_params.memcg = memcg; | |
156 | s->memcg_params.root_cache = root_cache; | |
33a690c4 | 157 | return 0; |
f7ce3190 | 158 | } |
33a690c4 | 159 | |
f7ce3190 | 160 | slab_init_memcg_params(s); |
33a690c4 | 161 | |
f7ce3190 VD |
162 | if (!memcg_nr_cache_ids) |
163 | return 0; | |
33a690c4 | 164 | |
f7ce3190 VD |
165 | arr = kzalloc(sizeof(struct memcg_cache_array) + |
166 | memcg_nr_cache_ids * sizeof(void *), | |
167 | GFP_KERNEL); | |
168 | if (!arr) | |
169 | return -ENOMEM; | |
33a690c4 | 170 | |
f7ce3190 | 171 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); |
33a690c4 VD |
172 | return 0; |
173 | } | |
174 | ||
f7ce3190 | 175 | static void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 | 176 | { |
f7ce3190 VD |
177 | if (is_root_cache(s)) |
178 | kfree(rcu_access_pointer(s->memcg_params.memcg_caches)); | |
33a690c4 VD |
179 | } |
180 | ||
f7ce3190 | 181 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
6f817f4c | 182 | { |
f7ce3190 | 183 | struct memcg_cache_array *old, *new; |
6f817f4c | 184 | |
f7ce3190 VD |
185 | if (!is_root_cache(s)) |
186 | return 0; | |
6f817f4c | 187 | |
f7ce3190 VD |
188 | new = kzalloc(sizeof(struct memcg_cache_array) + |
189 | new_array_size * sizeof(void *), GFP_KERNEL); | |
190 | if (!new) | |
6f817f4c VD |
191 | return -ENOMEM; |
192 | ||
f7ce3190 VD |
193 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
194 | lockdep_is_held(&slab_mutex)); | |
195 | if (old) | |
196 | memcpy(new->entries, old->entries, | |
197 | memcg_nr_cache_ids * sizeof(void *)); | |
6f817f4c | 198 | |
f7ce3190 VD |
199 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
200 | if (old) | |
201 | kfree_rcu(old, rcu); | |
6f817f4c VD |
202 | return 0; |
203 | } | |
204 | ||
55007d84 GC |
205 | int memcg_update_all_caches(int num_memcgs) |
206 | { | |
207 | struct kmem_cache *s; | |
208 | int ret = 0; | |
55007d84 | 209 | |
05257a1a | 210 | mutex_lock(&slab_mutex); |
55007d84 | 211 | list_for_each_entry(s, &slab_caches, list) { |
f7ce3190 | 212 | ret = update_memcg_params(s, num_memcgs); |
55007d84 | 213 | /* |
55007d84 GC |
214 | * Instead of freeing the memory, we'll just leave the caches |
215 | * up to this point in an updated state. | |
216 | */ | |
217 | if (ret) | |
05257a1a | 218 | break; |
55007d84 | 219 | } |
55007d84 GC |
220 | mutex_unlock(&slab_mutex); |
221 | return ret; | |
222 | } | |
657dc2f9 TH |
223 | |
224 | static void unlink_memcg_cache(struct kmem_cache *s) | |
225 | { | |
226 | list_del(&s->memcg_params.list); | |
227 | } | |
33a690c4 | 228 | #else |
f7ce3190 VD |
229 | static inline int init_memcg_params(struct kmem_cache *s, |
230 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
33a690c4 VD |
231 | { |
232 | return 0; | |
233 | } | |
234 | ||
f7ce3190 | 235 | static inline void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 VD |
236 | { |
237 | } | |
657dc2f9 TH |
238 | |
239 | static inline void unlink_memcg_cache(struct kmem_cache *s) | |
240 | { | |
241 | } | |
127424c8 | 242 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
55007d84 | 243 | |
423c929c JK |
244 | /* |
245 | * Find a mergeable slab cache | |
246 | */ | |
247 | int slab_unmergeable(struct kmem_cache *s) | |
248 | { | |
249 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
250 | return 1; | |
251 | ||
252 | if (!is_root_cache(s)) | |
253 | return 1; | |
254 | ||
255 | if (s->ctor) | |
256 | return 1; | |
257 | ||
258 | /* | |
259 | * We may have set a slab to be unmergeable during bootstrap. | |
260 | */ | |
261 | if (s->refcount < 0) | |
262 | return 1; | |
263 | ||
264 | return 0; | |
265 | } | |
266 | ||
267 | struct kmem_cache *find_mergeable(size_t size, size_t align, | |
268 | unsigned long flags, const char *name, void (*ctor)(void *)) | |
269 | { | |
270 | struct kmem_cache *s; | |
271 | ||
c6e28895 | 272 | if (slab_nomerge) |
423c929c JK |
273 | return NULL; |
274 | ||
275 | if (ctor) | |
276 | return NULL; | |
277 | ||
278 | size = ALIGN(size, sizeof(void *)); | |
279 | align = calculate_alignment(flags, align, size); | |
280 | size = ALIGN(size, align); | |
281 | flags = kmem_cache_flags(size, flags, name, NULL); | |
282 | ||
c6e28895 GM |
283 | if (flags & SLAB_NEVER_MERGE) |
284 | return NULL; | |
285 | ||
54362057 | 286 | list_for_each_entry_reverse(s, &slab_caches, list) { |
423c929c JK |
287 | if (slab_unmergeable(s)) |
288 | continue; | |
289 | ||
290 | if (size > s->size) | |
291 | continue; | |
292 | ||
293 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
294 | continue; | |
295 | /* | |
296 | * Check if alignment is compatible. | |
297 | * Courtesy of Adrian Drzewiecki | |
298 | */ | |
299 | if ((s->size & ~(align - 1)) != s->size) | |
300 | continue; | |
301 | ||
302 | if (s->size - size >= sizeof(void *)) | |
303 | continue; | |
304 | ||
95069ac8 JK |
305 | if (IS_ENABLED(CONFIG_SLAB) && align && |
306 | (align > s->align || s->align % align)) | |
307 | continue; | |
308 | ||
423c929c JK |
309 | return s; |
310 | } | |
311 | return NULL; | |
312 | } | |
313 | ||
45906855 CL |
314 | /* |
315 | * Figure out what the alignment of the objects will be given a set of | |
316 | * flags, a user specified alignment and the size of the objects. | |
317 | */ | |
318 | unsigned long calculate_alignment(unsigned long flags, | |
319 | unsigned long align, unsigned long size) | |
320 | { | |
321 | /* | |
322 | * If the user wants hardware cache aligned objects then follow that | |
323 | * suggestion if the object is sufficiently large. | |
324 | * | |
325 | * The hardware cache alignment cannot override the specified | |
326 | * alignment though. If that is greater then use it. | |
327 | */ | |
328 | if (flags & SLAB_HWCACHE_ALIGN) { | |
329 | unsigned long ralign = cache_line_size(); | |
330 | while (size <= ralign / 2) | |
331 | ralign /= 2; | |
332 | align = max(align, ralign); | |
333 | } | |
334 | ||
335 | if (align < ARCH_SLAB_MINALIGN) | |
336 | align = ARCH_SLAB_MINALIGN; | |
337 | ||
338 | return ALIGN(align, sizeof(void *)); | |
339 | } | |
340 | ||
c9a77a79 VD |
341 | static struct kmem_cache *create_cache(const char *name, |
342 | size_t object_size, size_t size, size_t align, | |
343 | unsigned long flags, void (*ctor)(void *), | |
344 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
794b1248 VD |
345 | { |
346 | struct kmem_cache *s; | |
347 | int err; | |
348 | ||
349 | err = -ENOMEM; | |
350 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
351 | if (!s) | |
352 | goto out; | |
353 | ||
354 | s->name = name; | |
355 | s->object_size = object_size; | |
356 | s->size = size; | |
357 | s->align = align; | |
358 | s->ctor = ctor; | |
359 | ||
f7ce3190 | 360 | err = init_memcg_params(s, memcg, root_cache); |
794b1248 VD |
361 | if (err) |
362 | goto out_free_cache; | |
363 | ||
364 | err = __kmem_cache_create(s, flags); | |
365 | if (err) | |
366 | goto out_free_cache; | |
367 | ||
368 | s->refcount = 1; | |
369 | list_add(&s->list, &slab_caches); | |
794b1248 VD |
370 | out: |
371 | if (err) | |
372 | return ERR_PTR(err); | |
373 | return s; | |
374 | ||
375 | out_free_cache: | |
f7ce3190 | 376 | destroy_memcg_params(s); |
7c4da061 | 377 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
378 | goto out; |
379 | } | |
45906855 | 380 | |
77be4b13 SK |
381 | /* |
382 | * kmem_cache_create - Create a cache. | |
383 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
384 | * @size: The size of objects to be created in this cache. | |
385 | * @align: The required alignment for the objects. | |
386 | * @flags: SLAB flags | |
387 | * @ctor: A constructor for the objects. | |
388 | * | |
389 | * Returns a ptr to the cache on success, NULL on failure. | |
390 | * Cannot be called within a interrupt, but can be interrupted. | |
391 | * The @ctor is run when new pages are allocated by the cache. | |
392 | * | |
393 | * The flags are | |
394 | * | |
395 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
396 | * to catch references to uninitialised memory. | |
397 | * | |
398 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
399 | * for buffer overruns. | |
400 | * | |
401 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
402 | * cacheline. This can be beneficial if you're counting cycles as closely | |
403 | * as davem. | |
404 | */ | |
2633d7a0 | 405 | struct kmem_cache * |
794b1248 VD |
406 | kmem_cache_create(const char *name, size_t size, size_t align, |
407 | unsigned long flags, void (*ctor)(void *)) | |
77be4b13 | 408 | { |
40911a79 | 409 | struct kmem_cache *s = NULL; |
3dec16ea | 410 | const char *cache_name; |
3965fc36 | 411 | int err; |
039363f3 | 412 | |
77be4b13 | 413 | get_online_cpus(); |
03afc0e2 | 414 | get_online_mems(); |
05257a1a | 415 | memcg_get_cache_ids(); |
03afc0e2 | 416 | |
77be4b13 | 417 | mutex_lock(&slab_mutex); |
686d550d | 418 | |
794b1248 | 419 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 420 | if (err) { |
3965fc36 | 421 | goto out_unlock; |
3aa24f51 | 422 | } |
686d550d | 423 | |
e70954fd TG |
424 | /* Refuse requests with allocator specific flags */ |
425 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
426 | err = -EINVAL; | |
427 | goto out_unlock; | |
428 | } | |
429 | ||
d8843922 GC |
430 | /* |
431 | * Some allocators will constraint the set of valid flags to a subset | |
432 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
433 | * case, and we'll just provide them with a sanitized version of the | |
434 | * passed flags. | |
435 | */ | |
436 | flags &= CACHE_CREATE_MASK; | |
686d550d | 437 | |
794b1248 VD |
438 | s = __kmem_cache_alias(name, size, align, flags, ctor); |
439 | if (s) | |
3965fc36 | 440 | goto out_unlock; |
2633d7a0 | 441 | |
3dec16ea | 442 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
443 | if (!cache_name) { |
444 | err = -ENOMEM; | |
445 | goto out_unlock; | |
446 | } | |
7c9adf5a | 447 | |
c9a77a79 VD |
448 | s = create_cache(cache_name, size, size, |
449 | calculate_alignment(flags, align, size), | |
450 | flags, ctor, NULL, NULL); | |
794b1248 VD |
451 | if (IS_ERR(s)) { |
452 | err = PTR_ERR(s); | |
3dec16ea | 453 | kfree_const(cache_name); |
794b1248 | 454 | } |
3965fc36 VD |
455 | |
456 | out_unlock: | |
20cea968 | 457 | mutex_unlock(&slab_mutex); |
03afc0e2 | 458 | |
05257a1a | 459 | memcg_put_cache_ids(); |
03afc0e2 | 460 | put_online_mems(); |
20cea968 CL |
461 | put_online_cpus(); |
462 | ||
ba3253c7 | 463 | if (err) { |
686d550d CL |
464 | if (flags & SLAB_PANIC) |
465 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
466 | name, err); | |
467 | else { | |
1170532b | 468 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
469 | name, err); |
470 | dump_stack(); | |
471 | } | |
686d550d CL |
472 | return NULL; |
473 | } | |
039363f3 CL |
474 | return s; |
475 | } | |
794b1248 | 476 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 477 | |
657dc2f9 | 478 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 479 | { |
657dc2f9 TH |
480 | LIST_HEAD(to_destroy); |
481 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 482 | |
657dc2f9 TH |
483 | /* |
484 | * On destruction, SLAB_DESTROY_BY_RCU kmem_caches are put on the | |
485 | * @slab_caches_to_rcu_destroy list. The slab pages are freed | |
486 | * through RCU and and the associated kmem_cache are dereferenced | |
487 | * while freeing the pages, so the kmem_caches should be freed only | |
488 | * after the pending RCU operations are finished. As rcu_barrier() | |
489 | * is a pretty slow operation, we batch all pending destructions | |
490 | * asynchronously. | |
491 | */ | |
492 | mutex_lock(&slab_mutex); | |
493 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
494 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 495 | |
657dc2f9 TH |
496 | if (list_empty(&to_destroy)) |
497 | return; | |
498 | ||
499 | rcu_barrier(); | |
500 | ||
501 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
502 | #ifdef SLAB_SUPPORTS_SYSFS | |
503 | sysfs_slab_release(s); | |
504 | #else | |
505 | slab_kmem_cache_release(s); | |
506 | #endif | |
507 | } | |
d5b3cf71 VD |
508 | } |
509 | ||
657dc2f9 | 510 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 511 | { |
657dc2f9 TH |
512 | if (__kmem_cache_shutdown(s) != 0) |
513 | return -EBUSY; | |
d5b3cf71 | 514 | |
657dc2f9 TH |
515 | list_del(&s->list); |
516 | if (!is_root_cache(s)) | |
517 | unlink_memcg_cache(s); | |
d5b3cf71 | 518 | |
657dc2f9 TH |
519 | if (s->flags & SLAB_DESTROY_BY_RCU) { |
520 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); | |
521 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
522 | } else { | |
d5b3cf71 | 523 | #ifdef SLAB_SUPPORTS_SYSFS |
bf5eb3de | 524 | sysfs_slab_release(s); |
d5b3cf71 VD |
525 | #else |
526 | slab_kmem_cache_release(s); | |
527 | #endif | |
528 | } | |
657dc2f9 TH |
529 | |
530 | return 0; | |
d5b3cf71 VD |
531 | } |
532 | ||
127424c8 | 533 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
794b1248 | 534 | /* |
776ed0f0 | 535 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
536 | * @memcg: The memory cgroup the new cache is for. |
537 | * @root_cache: The parent of the new cache. | |
538 | * | |
539 | * This function attempts to create a kmem cache that will serve allocation | |
540 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
541 | * from its parent. | |
542 | */ | |
d5b3cf71 VD |
543 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
544 | struct kmem_cache *root_cache) | |
2633d7a0 | 545 | { |
3e0350a3 | 546 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
33398cf2 | 547 | struct cgroup_subsys_state *css = &memcg->css; |
f7ce3190 | 548 | struct memcg_cache_array *arr; |
bd673145 | 549 | struct kmem_cache *s = NULL; |
794b1248 | 550 | char *cache_name; |
f7ce3190 | 551 | int idx; |
794b1248 VD |
552 | |
553 | get_online_cpus(); | |
03afc0e2 VD |
554 | get_online_mems(); |
555 | ||
794b1248 VD |
556 | mutex_lock(&slab_mutex); |
557 | ||
2a4db7eb | 558 | /* |
567e9ab2 | 559 | * The memory cgroup could have been offlined while the cache |
2a4db7eb VD |
560 | * creation work was pending. |
561 | */ | |
b6ecd2de | 562 | if (memcg->kmem_state != KMEM_ONLINE) |
2a4db7eb VD |
563 | goto out_unlock; |
564 | ||
f7ce3190 VD |
565 | idx = memcg_cache_id(memcg); |
566 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
567 | lockdep_is_held(&slab_mutex)); | |
568 | ||
d5b3cf71 VD |
569 | /* |
570 | * Since per-memcg caches are created asynchronously on first | |
571 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
572 | * create the same cache, but only one of them may succeed. | |
573 | */ | |
f7ce3190 | 574 | if (arr->entries[idx]) |
d5b3cf71 VD |
575 | goto out_unlock; |
576 | ||
f1008365 | 577 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
73f576c0 JW |
578 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, |
579 | css->serial_nr, memcg_name_buf); | |
794b1248 VD |
580 | if (!cache_name) |
581 | goto out_unlock; | |
582 | ||
c9a77a79 VD |
583 | s = create_cache(cache_name, root_cache->object_size, |
584 | root_cache->size, root_cache->align, | |
f773e36d GT |
585 | root_cache->flags & CACHE_CREATE_MASK, |
586 | root_cache->ctor, memcg, root_cache); | |
d5b3cf71 VD |
587 | /* |
588 | * If we could not create a memcg cache, do not complain, because | |
589 | * that's not critical at all as we can always proceed with the root | |
590 | * cache. | |
591 | */ | |
bd673145 | 592 | if (IS_ERR(s)) { |
794b1248 | 593 | kfree(cache_name); |
d5b3cf71 | 594 | goto out_unlock; |
bd673145 | 595 | } |
794b1248 | 596 | |
426589f5 VD |
597 | list_add(&s->memcg_params.list, &root_cache->memcg_params.list); |
598 | ||
d5b3cf71 VD |
599 | /* |
600 | * Since readers won't lock (see cache_from_memcg_idx()), we need a | |
601 | * barrier here to ensure nobody will see the kmem_cache partially | |
602 | * initialized. | |
603 | */ | |
604 | smp_wmb(); | |
f7ce3190 | 605 | arr->entries[idx] = s; |
d5b3cf71 | 606 | |
794b1248 VD |
607 | out_unlock: |
608 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
609 | |
610 | put_online_mems(); | |
794b1248 | 611 | put_online_cpus(); |
2633d7a0 | 612 | } |
b8529907 | 613 | |
2a4db7eb VD |
614 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg) |
615 | { | |
616 | int idx; | |
617 | struct memcg_cache_array *arr; | |
d6e0b7fa | 618 | struct kmem_cache *s, *c; |
2a4db7eb VD |
619 | |
620 | idx = memcg_cache_id(memcg); | |
621 | ||
d6e0b7fa VD |
622 | get_online_cpus(); |
623 | get_online_mems(); | |
624 | ||
2a4db7eb VD |
625 | mutex_lock(&slab_mutex); |
626 | list_for_each_entry(s, &slab_caches, list) { | |
627 | if (!is_root_cache(s)) | |
628 | continue; | |
629 | ||
630 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
631 | lockdep_is_held(&slab_mutex)); | |
d6e0b7fa VD |
632 | c = arr->entries[idx]; |
633 | if (!c) | |
634 | continue; | |
635 | ||
290b6a58 | 636 | __kmem_cache_shrink(c, true); |
2a4db7eb VD |
637 | arr->entries[idx] = NULL; |
638 | } | |
639 | mutex_unlock(&slab_mutex); | |
d6e0b7fa VD |
640 | |
641 | put_online_mems(); | |
642 | put_online_cpus(); | |
2a4db7eb VD |
643 | } |
644 | ||
d5b3cf71 | 645 | void memcg_destroy_kmem_caches(struct mem_cgroup *memcg) |
b8529907 | 646 | { |
d5b3cf71 | 647 | struct kmem_cache *s, *s2; |
b8529907 | 648 | |
d5b3cf71 VD |
649 | get_online_cpus(); |
650 | get_online_mems(); | |
b8529907 | 651 | |
b8529907 | 652 | mutex_lock(&slab_mutex); |
d5b3cf71 | 653 | list_for_each_entry_safe(s, s2, &slab_caches, list) { |
f7ce3190 | 654 | if (is_root_cache(s) || s->memcg_params.memcg != memcg) |
d5b3cf71 VD |
655 | continue; |
656 | /* | |
657 | * The cgroup is about to be freed and therefore has no charges | |
658 | * left. Hence, all its caches must be empty by now. | |
659 | */ | |
657dc2f9 | 660 | BUG_ON(shutdown_cache(s)); |
d5b3cf71 VD |
661 | } |
662 | mutex_unlock(&slab_mutex); | |
b8529907 | 663 | |
d5b3cf71 VD |
664 | put_online_mems(); |
665 | put_online_cpus(); | |
b8529907 | 666 | } |
d60fdcc9 | 667 | |
657dc2f9 | 668 | static int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
669 | { |
670 | struct memcg_cache_array *arr; | |
671 | struct kmem_cache *c, *c2; | |
672 | LIST_HEAD(busy); | |
673 | int i; | |
674 | ||
675 | BUG_ON(!is_root_cache(s)); | |
676 | ||
677 | /* | |
678 | * First, shutdown active caches, i.e. caches that belong to online | |
679 | * memory cgroups. | |
680 | */ | |
681 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
682 | lockdep_is_held(&slab_mutex)); | |
683 | for_each_memcg_cache_index(i) { | |
684 | c = arr->entries[i]; | |
685 | if (!c) | |
686 | continue; | |
657dc2f9 | 687 | if (shutdown_cache(c)) |
d60fdcc9 VD |
688 | /* |
689 | * The cache still has objects. Move it to a temporary | |
690 | * list so as not to try to destroy it for a second | |
691 | * time while iterating over inactive caches below. | |
692 | */ | |
693 | list_move(&c->memcg_params.list, &busy); | |
694 | else | |
695 | /* | |
696 | * The cache is empty and will be destroyed soon. Clear | |
697 | * the pointer to it in the memcg_caches array so that | |
698 | * it will never be accessed even if the root cache | |
699 | * stays alive. | |
700 | */ | |
701 | arr->entries[i] = NULL; | |
702 | } | |
703 | ||
704 | /* | |
705 | * Second, shutdown all caches left from memory cgroups that are now | |
706 | * offline. | |
707 | */ | |
708 | list_for_each_entry_safe(c, c2, &s->memcg_params.list, | |
709 | memcg_params.list) | |
657dc2f9 | 710 | shutdown_cache(c); |
d60fdcc9 VD |
711 | |
712 | list_splice(&busy, &s->memcg_params.list); | |
713 | ||
714 | /* | |
715 | * A cache being destroyed must be empty. In particular, this means | |
716 | * that all per memcg caches attached to it must be empty too. | |
717 | */ | |
718 | if (!list_empty(&s->memcg_params.list)) | |
719 | return -EBUSY; | |
720 | return 0; | |
721 | } | |
722 | #else | |
657dc2f9 | 723 | static inline int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
724 | { |
725 | return 0; | |
726 | } | |
127424c8 | 727 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
97d06609 | 728 | |
41a21285 CL |
729 | void slab_kmem_cache_release(struct kmem_cache *s) |
730 | { | |
52b4b950 | 731 | __kmem_cache_release(s); |
f7ce3190 | 732 | destroy_memcg_params(s); |
3dec16ea | 733 | kfree_const(s->name); |
41a21285 CL |
734 | kmem_cache_free(kmem_cache, s); |
735 | } | |
736 | ||
945cf2b6 CL |
737 | void kmem_cache_destroy(struct kmem_cache *s) |
738 | { | |
d60fdcc9 | 739 | int err; |
d5b3cf71 | 740 | |
3942d299 SS |
741 | if (unlikely(!s)) |
742 | return; | |
743 | ||
945cf2b6 | 744 | get_online_cpus(); |
03afc0e2 VD |
745 | get_online_mems(); |
746 | ||
55834c59 | 747 | kasan_cache_destroy(s); |
945cf2b6 | 748 | mutex_lock(&slab_mutex); |
b8529907 | 749 | |
945cf2b6 | 750 | s->refcount--; |
b8529907 VD |
751 | if (s->refcount) |
752 | goto out_unlock; | |
753 | ||
657dc2f9 | 754 | err = shutdown_memcg_caches(s); |
d60fdcc9 | 755 | if (!err) |
657dc2f9 | 756 | err = shutdown_cache(s); |
b8529907 | 757 | |
cd918c55 | 758 | if (err) { |
756a025f JP |
759 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
760 | s->name); | |
cd918c55 VD |
761 | dump_stack(); |
762 | } | |
b8529907 VD |
763 | out_unlock: |
764 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 765 | |
03afc0e2 | 766 | put_online_mems(); |
945cf2b6 CL |
767 | put_online_cpus(); |
768 | } | |
769 | EXPORT_SYMBOL(kmem_cache_destroy); | |
770 | ||
03afc0e2 VD |
771 | /** |
772 | * kmem_cache_shrink - Shrink a cache. | |
773 | * @cachep: The cache to shrink. | |
774 | * | |
775 | * Releases as many slabs as possible for a cache. | |
776 | * To help debugging, a zero exit status indicates all slabs were released. | |
777 | */ | |
778 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
779 | { | |
780 | int ret; | |
781 | ||
782 | get_online_cpus(); | |
783 | get_online_mems(); | |
55834c59 | 784 | kasan_cache_shrink(cachep); |
290b6a58 | 785 | ret = __kmem_cache_shrink(cachep, false); |
03afc0e2 VD |
786 | put_online_mems(); |
787 | put_online_cpus(); | |
788 | return ret; | |
789 | } | |
790 | EXPORT_SYMBOL(kmem_cache_shrink); | |
791 | ||
fda90124 | 792 | bool slab_is_available(void) |
97d06609 CL |
793 | { |
794 | return slab_state >= UP; | |
795 | } | |
b7454ad3 | 796 | |
45530c44 CL |
797 | #ifndef CONFIG_SLOB |
798 | /* Create a cache during boot when no slab services are available yet */ | |
799 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
800 | unsigned long flags) | |
801 | { | |
802 | int err; | |
803 | ||
804 | s->name = name; | |
805 | s->size = s->object_size = size; | |
45906855 | 806 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
f7ce3190 VD |
807 | |
808 | slab_init_memcg_params(s); | |
809 | ||
45530c44 CL |
810 | err = __kmem_cache_create(s, flags); |
811 | ||
812 | if (err) | |
31ba7346 | 813 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
45530c44 CL |
814 | name, size, err); |
815 | ||
816 | s->refcount = -1; /* Exempt from merging for now */ | |
817 | } | |
818 | ||
819 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
820 | unsigned long flags) | |
821 | { | |
822 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
823 | ||
824 | if (!s) | |
825 | panic("Out of memory when creating slab %s\n", name); | |
826 | ||
827 | create_boot_cache(s, name, size, flags); | |
828 | list_add(&s->list, &slab_caches); | |
829 | s->refcount = 1; | |
830 | return s; | |
831 | } | |
832 | ||
9425c58e CL |
833 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
834 | EXPORT_SYMBOL(kmalloc_caches); | |
835 | ||
836 | #ifdef CONFIG_ZONE_DMA | |
837 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
838 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
839 | #endif | |
840 | ||
2c59dd65 CL |
841 | /* |
842 | * Conversion table for small slabs sizes / 8 to the index in the | |
843 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
844 | * of two cache sizes there. The size of larger slabs can be determined using | |
845 | * fls. | |
846 | */ | |
847 | static s8 size_index[24] = { | |
848 | 3, /* 8 */ | |
849 | 4, /* 16 */ | |
850 | 5, /* 24 */ | |
851 | 5, /* 32 */ | |
852 | 6, /* 40 */ | |
853 | 6, /* 48 */ | |
854 | 6, /* 56 */ | |
855 | 6, /* 64 */ | |
856 | 1, /* 72 */ | |
857 | 1, /* 80 */ | |
858 | 1, /* 88 */ | |
859 | 1, /* 96 */ | |
860 | 7, /* 104 */ | |
861 | 7, /* 112 */ | |
862 | 7, /* 120 */ | |
863 | 7, /* 128 */ | |
864 | 2, /* 136 */ | |
865 | 2, /* 144 */ | |
866 | 2, /* 152 */ | |
867 | 2, /* 160 */ | |
868 | 2, /* 168 */ | |
869 | 2, /* 176 */ | |
870 | 2, /* 184 */ | |
871 | 2 /* 192 */ | |
872 | }; | |
873 | ||
874 | static inline int size_index_elem(size_t bytes) | |
875 | { | |
876 | return (bytes - 1) / 8; | |
877 | } | |
878 | ||
879 | /* | |
880 | * Find the kmem_cache structure that serves a given size of | |
881 | * allocation | |
882 | */ | |
883 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
884 | { | |
885 | int index; | |
886 | ||
9de1bc87 | 887 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
907985f4 | 888 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
6286ae97 | 889 | return NULL; |
907985f4 | 890 | } |
6286ae97 | 891 | |
2c59dd65 CL |
892 | if (size <= 192) { |
893 | if (!size) | |
894 | return ZERO_SIZE_PTR; | |
895 | ||
896 | index = size_index[size_index_elem(size)]; | |
897 | } else | |
898 | index = fls(size - 1); | |
899 | ||
900 | #ifdef CONFIG_ZONE_DMA | |
b1e05416 | 901 | if (unlikely((flags & GFP_DMA))) |
2c59dd65 CL |
902 | return kmalloc_dma_caches[index]; |
903 | ||
904 | #endif | |
905 | return kmalloc_caches[index]; | |
906 | } | |
907 | ||
4066c33d GG |
908 | /* |
909 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
910 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
911 | * kmalloc-67108864. | |
912 | */ | |
af3b5f87 | 913 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
4066c33d GG |
914 | {NULL, 0}, {"kmalloc-96", 96}, |
915 | {"kmalloc-192", 192}, {"kmalloc-8", 8}, | |
916 | {"kmalloc-16", 16}, {"kmalloc-32", 32}, | |
917 | {"kmalloc-64", 64}, {"kmalloc-128", 128}, | |
918 | {"kmalloc-256", 256}, {"kmalloc-512", 512}, | |
919 | {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048}, | |
920 | {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192}, | |
921 | {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768}, | |
922 | {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072}, | |
923 | {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288}, | |
924 | {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152}, | |
925 | {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608}, | |
926 | {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432}, | |
927 | {"kmalloc-67108864", 67108864} | |
928 | }; | |
929 | ||
f97d5f63 | 930 | /* |
34cc6990 DS |
931 | * Patch up the size_index table if we have strange large alignment |
932 | * requirements for the kmalloc array. This is only the case for | |
933 | * MIPS it seems. The standard arches will not generate any code here. | |
934 | * | |
935 | * Largest permitted alignment is 256 bytes due to the way we | |
936 | * handle the index determination for the smaller caches. | |
937 | * | |
938 | * Make sure that nothing crazy happens if someone starts tinkering | |
939 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 940 | */ |
34cc6990 | 941 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 CL |
942 | { |
943 | int i; | |
944 | ||
2c59dd65 CL |
945 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
946 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
947 | ||
948 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
949 | int elem = size_index_elem(i); | |
950 | ||
951 | if (elem >= ARRAY_SIZE(size_index)) | |
952 | break; | |
953 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
954 | } | |
955 | ||
956 | if (KMALLOC_MIN_SIZE >= 64) { | |
957 | /* | |
958 | * The 96 byte size cache is not used if the alignment | |
959 | * is 64 byte. | |
960 | */ | |
961 | for (i = 64 + 8; i <= 96; i += 8) | |
962 | size_index[size_index_elem(i)] = 7; | |
963 | ||
964 | } | |
965 | ||
966 | if (KMALLOC_MIN_SIZE >= 128) { | |
967 | /* | |
968 | * The 192 byte sized cache is not used if the alignment | |
969 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
970 | * instead. | |
971 | */ | |
972 | for (i = 128 + 8; i <= 192; i += 8) | |
973 | size_index[size_index_elem(i)] = 8; | |
974 | } | |
34cc6990 DS |
975 | } |
976 | ||
ae6f2462 | 977 | static void __init new_kmalloc_cache(int idx, unsigned long flags) |
a9730fca CL |
978 | { |
979 | kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name, | |
980 | kmalloc_info[idx].size, flags); | |
981 | } | |
982 | ||
34cc6990 DS |
983 | /* |
984 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
985 | * may already have been created because they were needed to | |
986 | * enable allocations for slab creation. | |
987 | */ | |
988 | void __init create_kmalloc_caches(unsigned long flags) | |
989 | { | |
990 | int i; | |
991 | ||
a9730fca CL |
992 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
993 | if (!kmalloc_caches[i]) | |
994 | new_kmalloc_cache(i, flags); | |
f97d5f63 | 995 | |
956e46ef | 996 | /* |
a9730fca CL |
997 | * Caches that are not of the two-to-the-power-of size. |
998 | * These have to be created immediately after the | |
999 | * earlier power of two caches | |
956e46ef | 1000 | */ |
a9730fca CL |
1001 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) |
1002 | new_kmalloc_cache(1, flags); | |
1003 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) | |
1004 | new_kmalloc_cache(2, flags); | |
8a965b3b CL |
1005 | } |
1006 | ||
f97d5f63 CL |
1007 | /* Kmalloc array is now usable */ |
1008 | slab_state = UP; | |
1009 | ||
f97d5f63 CL |
1010 | #ifdef CONFIG_ZONE_DMA |
1011 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1012 | struct kmem_cache *s = kmalloc_caches[i]; | |
1013 | ||
1014 | if (s) { | |
1015 | int size = kmalloc_size(i); | |
1016 | char *n = kasprintf(GFP_NOWAIT, | |
1017 | "dma-kmalloc-%d", size); | |
1018 | ||
1019 | BUG_ON(!n); | |
1020 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
1021 | size, SLAB_CACHE_DMA | flags); | |
1022 | } | |
1023 | } | |
1024 | #endif | |
1025 | } | |
45530c44 CL |
1026 | #endif /* !CONFIG_SLOB */ |
1027 | ||
cea371f4 VD |
1028 | /* |
1029 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1030 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1031 | * know the allocation order to free the pages properly in kfree. | |
1032 | */ | |
52383431 VD |
1033 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
1034 | { | |
1035 | void *ret; | |
1036 | struct page *page; | |
1037 | ||
1038 | flags |= __GFP_COMP; | |
4949148a | 1039 | page = alloc_pages(flags, order); |
52383431 VD |
1040 | ret = page ? page_address(page) : NULL; |
1041 | kmemleak_alloc(ret, size, 1, flags); | |
505f5dcb | 1042 | kasan_kmalloc_large(ret, size, flags); |
52383431 VD |
1043 | return ret; |
1044 | } | |
1045 | EXPORT_SYMBOL(kmalloc_order); | |
1046 | ||
f1b6eb6e CL |
1047 | #ifdef CONFIG_TRACING |
1048 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1049 | { | |
1050 | void *ret = kmalloc_order(size, flags, order); | |
1051 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1052 | return ret; | |
1053 | } | |
1054 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1055 | #endif | |
45530c44 | 1056 | |
7c00fce9 TG |
1057 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1058 | /* Randomize a generic freelist */ | |
1059 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
1060 | size_t count) | |
1061 | { | |
1062 | size_t i; | |
1063 | unsigned int rand; | |
1064 | ||
1065 | for (i = 0; i < count; i++) | |
1066 | list[i] = i; | |
1067 | ||
1068 | /* Fisher-Yates shuffle */ | |
1069 | for (i = count - 1; i > 0; i--) { | |
1070 | rand = prandom_u32_state(state); | |
1071 | rand %= (i + 1); | |
1072 | swap(list[i], list[rand]); | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | /* Create a random sequence per cache */ | |
1077 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1078 | gfp_t gfp) | |
1079 | { | |
1080 | struct rnd_state state; | |
1081 | ||
1082 | if (count < 2 || cachep->random_seq) | |
1083 | return 0; | |
1084 | ||
1085 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1086 | if (!cachep->random_seq) | |
1087 | return -ENOMEM; | |
1088 | ||
1089 | /* Get best entropy at this stage of boot */ | |
1090 | prandom_seed_state(&state, get_random_long()); | |
1091 | ||
1092 | freelist_randomize(&state, cachep->random_seq, count); | |
1093 | return 0; | |
1094 | } | |
1095 | ||
1096 | /* Destroy the per-cache random freelist sequence */ | |
1097 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1098 | { | |
1099 | kfree(cachep->random_seq); | |
1100 | cachep->random_seq = NULL; | |
1101 | } | |
1102 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1103 | ||
b7454ad3 | 1104 | #ifdef CONFIG_SLABINFO |
e9b4db2b WL |
1105 | |
1106 | #ifdef CONFIG_SLAB | |
1107 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
1108 | #else | |
1109 | #define SLABINFO_RIGHTS S_IRUSR | |
1110 | #endif | |
1111 | ||
b047501c | 1112 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1113 | { |
1114 | /* | |
1115 | * Output format version, so at least we can change it | |
1116 | * without _too_ many complaints. | |
1117 | */ | |
1118 | #ifdef CONFIG_DEBUG_SLAB | |
1119 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1120 | #else | |
1121 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1122 | #endif | |
756a025f | 1123 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1124 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1125 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1126 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1127 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1128 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1129 | #endif | |
1130 | seq_putc(m, '\n'); | |
1131 | } | |
1132 | ||
1df3b26f | 1133 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1134 | { |
b7454ad3 | 1135 | mutex_lock(&slab_mutex); |
b7454ad3 GC |
1136 | return seq_list_start(&slab_caches, *pos); |
1137 | } | |
1138 | ||
276a2439 | 1139 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 GC |
1140 | { |
1141 | return seq_list_next(p, &slab_caches, pos); | |
1142 | } | |
1143 | ||
276a2439 | 1144 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1145 | { |
1146 | mutex_unlock(&slab_mutex); | |
1147 | } | |
1148 | ||
749c5415 GC |
1149 | static void |
1150 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1151 | { | |
1152 | struct kmem_cache *c; | |
1153 | struct slabinfo sinfo; | |
749c5415 GC |
1154 | |
1155 | if (!is_root_cache(s)) | |
1156 | return; | |
1157 | ||
426589f5 | 1158 | for_each_memcg_cache(c, s) { |
749c5415 GC |
1159 | memset(&sinfo, 0, sizeof(sinfo)); |
1160 | get_slabinfo(c, &sinfo); | |
1161 | ||
1162 | info->active_slabs += sinfo.active_slabs; | |
1163 | info->num_slabs += sinfo.num_slabs; | |
1164 | info->shared_avail += sinfo.shared_avail; | |
1165 | info->active_objs += sinfo.active_objs; | |
1166 | info->num_objs += sinfo.num_objs; | |
1167 | } | |
1168 | } | |
1169 | ||
b047501c | 1170 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1171 | { |
0d7561c6 GC |
1172 | struct slabinfo sinfo; |
1173 | ||
1174 | memset(&sinfo, 0, sizeof(sinfo)); | |
1175 | get_slabinfo(s, &sinfo); | |
1176 | ||
749c5415 GC |
1177 | memcg_accumulate_slabinfo(s, &sinfo); |
1178 | ||
0d7561c6 | 1179 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 1180 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1181 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1182 | ||
1183 | seq_printf(m, " : tunables %4u %4u %4u", | |
1184 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1185 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1186 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1187 | slabinfo_show_stats(m, s); | |
1188 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1189 | } |
1190 | ||
1df3b26f | 1191 | static int slab_show(struct seq_file *m, void *p) |
749c5415 GC |
1192 | { |
1193 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
1194 | ||
1df3b26f VD |
1195 | if (p == slab_caches.next) |
1196 | print_slabinfo_header(m); | |
b047501c VD |
1197 | if (is_root_cache(s)) |
1198 | cache_show(s, m); | |
1199 | return 0; | |
1200 | } | |
1201 | ||
127424c8 | 1202 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
b047501c VD |
1203 | int memcg_slab_show(struct seq_file *m, void *p) |
1204 | { | |
1205 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
1206 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1207 | ||
1208 | if (p == slab_caches.next) | |
1209 | print_slabinfo_header(m); | |
f7ce3190 | 1210 | if (!is_root_cache(s) && s->memcg_params.memcg == memcg) |
b047501c VD |
1211 | cache_show(s, m); |
1212 | return 0; | |
749c5415 | 1213 | } |
b047501c | 1214 | #endif |
749c5415 | 1215 | |
b7454ad3 GC |
1216 | /* |
1217 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1218 | * | |
1219 | * Output layout: | |
1220 | * cache-name | |
1221 | * num-active-objs | |
1222 | * total-objs | |
1223 | * object size | |
1224 | * num-active-slabs | |
1225 | * total-slabs | |
1226 | * num-pages-per-slab | |
1227 | * + further values on SMP and with statistics enabled | |
1228 | */ | |
1229 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1230 | .start = slab_start, |
276a2439 WL |
1231 | .next = slab_next, |
1232 | .stop = slab_stop, | |
1df3b26f | 1233 | .show = slab_show, |
b7454ad3 GC |
1234 | }; |
1235 | ||
1236 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1237 | { | |
1238 | return seq_open(file, &slabinfo_op); | |
1239 | } | |
1240 | ||
1241 | static const struct file_operations proc_slabinfo_operations = { | |
1242 | .open = slabinfo_open, | |
1243 | .read = seq_read, | |
1244 | .write = slabinfo_write, | |
1245 | .llseek = seq_lseek, | |
1246 | .release = seq_release, | |
1247 | }; | |
1248 | ||
1249 | static int __init slab_proc_init(void) | |
1250 | { | |
e9b4db2b WL |
1251 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
1252 | &proc_slabinfo_operations); | |
b7454ad3 GC |
1253 | return 0; |
1254 | } | |
1255 | module_init(slab_proc_init); | |
1256 | #endif /* CONFIG_SLABINFO */ | |
928cec9c AR |
1257 | |
1258 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1259 | gfp_t flags) | |
1260 | { | |
1261 | void *ret; | |
1262 | size_t ks = 0; | |
1263 | ||
1264 | if (p) | |
1265 | ks = ksize(p); | |
1266 | ||
0316bec2 | 1267 | if (ks >= new_size) { |
505f5dcb | 1268 | kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1269 | return (void *)p; |
0316bec2 | 1270 | } |
928cec9c AR |
1271 | |
1272 | ret = kmalloc_track_caller(new_size, flags); | |
1273 | if (ret && p) | |
1274 | memcpy(ret, p, ks); | |
1275 | ||
1276 | return ret; | |
1277 | } | |
1278 | ||
1279 | /** | |
1280 | * __krealloc - like krealloc() but don't free @p. | |
1281 | * @p: object to reallocate memory for. | |
1282 | * @new_size: how many bytes of memory are required. | |
1283 | * @flags: the type of memory to allocate. | |
1284 | * | |
1285 | * This function is like krealloc() except it never frees the originally | |
1286 | * allocated buffer. Use this if you don't want to free the buffer immediately | |
1287 | * like, for example, with RCU. | |
1288 | */ | |
1289 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) | |
1290 | { | |
1291 | if (unlikely(!new_size)) | |
1292 | return ZERO_SIZE_PTR; | |
1293 | ||
1294 | return __do_krealloc(p, new_size, flags); | |
1295 | ||
1296 | } | |
1297 | EXPORT_SYMBOL(__krealloc); | |
1298 | ||
1299 | /** | |
1300 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1301 | * @p: object to reallocate memory for. | |
1302 | * @new_size: how many bytes of memory are required. | |
1303 | * @flags: the type of memory to allocate. | |
1304 | * | |
1305 | * The contents of the object pointed to are preserved up to the | |
1306 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1307 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1308 | * %NULL pointer, the object pointed to is freed. | |
1309 | */ | |
1310 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1311 | { | |
1312 | void *ret; | |
1313 | ||
1314 | if (unlikely(!new_size)) { | |
1315 | kfree(p); | |
1316 | return ZERO_SIZE_PTR; | |
1317 | } | |
1318 | ||
1319 | ret = __do_krealloc(p, new_size, flags); | |
1320 | if (ret && p != ret) | |
1321 | kfree(p); | |
1322 | ||
1323 | return ret; | |
1324 | } | |
1325 | EXPORT_SYMBOL(krealloc); | |
1326 | ||
1327 | /** | |
1328 | * kzfree - like kfree but zero memory | |
1329 | * @p: object to free memory of | |
1330 | * | |
1331 | * The memory of the object @p points to is zeroed before freed. | |
1332 | * If @p is %NULL, kzfree() does nothing. | |
1333 | * | |
1334 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1335 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1336 | * careful when using this function in performance sensitive code. | |
1337 | */ | |
1338 | void kzfree(const void *p) | |
1339 | { | |
1340 | size_t ks; | |
1341 | void *mem = (void *)p; | |
1342 | ||
1343 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1344 | return; | |
1345 | ks = ksize(mem); | |
1346 | memset(mem, 0, ks); | |
1347 | kfree(mem); | |
1348 | } | |
1349 | EXPORT_SYMBOL(kzfree); | |
1350 | ||
1351 | /* Tracepoints definitions. */ | |
1352 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1353 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1354 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1355 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1356 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1357 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |