]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
3 | * Slab allocator functions that are independent of the allocator strategy | |
4 | * | |
5 | * (C) 2012 Christoph Lameter <cl@linux.com> | |
6 | */ | |
7 | #include <linux/slab.h> | |
8 | ||
9 | #include <linux/mm.h> | |
10 | #include <linux/poison.h> | |
11 | #include <linux/interrupt.h> | |
12 | #include <linux/memory.h> | |
13 | #include <linux/compiler.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/cpu.h> | |
16 | #include <linux/uaccess.h> | |
17 | #include <linux/seq_file.h> | |
18 | #include <linux/proc_fs.h> | |
19 | #include <asm/cacheflush.h> | |
20 | #include <asm/tlbflush.h> | |
21 | #include <asm/page.h> | |
22 | #include <linux/memcontrol.h> | |
23 | ||
24 | #define CREATE_TRACE_POINTS | |
25 | #include <trace/events/kmem.h> | |
26 | ||
27 | #include "slab.h" | |
28 | ||
29 | enum slab_state slab_state; | |
30 | LIST_HEAD(slab_caches); | |
31 | DEFINE_MUTEX(slab_mutex); | |
32 | struct kmem_cache *kmem_cache; | |
33 | ||
34 | static LIST_HEAD(slab_caches_to_rcu_destroy); | |
35 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
36 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
37 | slab_caches_to_rcu_destroy_workfn); | |
38 | ||
39 | /* | |
40 | * Set of flags that will prevent slab merging | |
41 | */ | |
42 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
43 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ | |
44 | SLAB_FAILSLAB | SLAB_KASAN) | |
45 | ||
46 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ | |
47 | SLAB_ACCOUNT) | |
48 | ||
49 | /* | |
50 | * Merge control. If this is set then no merging of slab caches will occur. | |
51 | */ | |
52 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); | |
53 | ||
54 | static int __init setup_slab_nomerge(char *str) | |
55 | { | |
56 | slab_nomerge = true; | |
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 | ||
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 | ||
75 | #ifdef CONFIG_DEBUG_VM | |
76 | static int kmem_cache_sanity_check(const char *name, size_t size) | |
77 | { | |
78 | struct kmem_cache *s = NULL; | |
79 | ||
80 | if (!name || in_interrupt() || size < sizeof(void *) || | |
81 | size > KMALLOC_MAX_SIZE) { | |
82 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); | |
83 | return -EINVAL; | |
84 | } | |
85 | ||
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) { | |
97 | pr_err("Slab cache with size %d has lost its name\n", | |
98 | s->object_size); | |
99 | continue; | |
100 | } | |
101 | } | |
102 | ||
103 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
104 | return 0; | |
105 | } | |
106 | #else | |
107 | static inline int kmem_cache_sanity_check(const char *name, size_t size) | |
108 | { | |
109 | return 0; | |
110 | } | |
111 | #endif | |
112 | ||
113 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) | |
114 | { | |
115 | size_t i; | |
116 | ||
117 | for (i = 0; i < nr; i++) { | |
118 | if (s) | |
119 | kmem_cache_free(s, p[i]); | |
120 | else | |
121 | kfree(p[i]); | |
122 | } | |
123 | } | |
124 | ||
125 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, | |
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); | |
134 | return 0; | |
135 | } | |
136 | } | |
137 | return i; | |
138 | } | |
139 | ||
140 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) | |
141 | ||
142 | LIST_HEAD(slab_root_caches); | |
143 | ||
144 | void slab_init_memcg_params(struct kmem_cache *s) | |
145 | { | |
146 | s->memcg_params.root_cache = NULL; | |
147 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); | |
148 | INIT_LIST_HEAD(&s->memcg_params.children); | |
149 | } | |
150 | ||
151 | static int init_memcg_params(struct kmem_cache *s, | |
152 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
153 | { | |
154 | struct memcg_cache_array *arr; | |
155 | ||
156 | if (root_cache) { | |
157 | s->memcg_params.root_cache = root_cache; | |
158 | s->memcg_params.memcg = memcg; | |
159 | INIT_LIST_HEAD(&s->memcg_params.children_node); | |
160 | INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node); | |
161 | return 0; | |
162 | } | |
163 | ||
164 | slab_init_memcg_params(s); | |
165 | ||
166 | if (!memcg_nr_cache_ids) | |
167 | return 0; | |
168 | ||
169 | arr = kvzalloc(sizeof(struct memcg_cache_array) + | |
170 | memcg_nr_cache_ids * sizeof(void *), | |
171 | GFP_KERNEL); | |
172 | if (!arr) | |
173 | return -ENOMEM; | |
174 | ||
175 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); | |
176 | return 0; | |
177 | } | |
178 | ||
179 | static void destroy_memcg_params(struct kmem_cache *s) | |
180 | { | |
181 | if (is_root_cache(s)) | |
182 | kvfree(rcu_access_pointer(s->memcg_params.memcg_caches)); | |
183 | } | |
184 | ||
185 | static void free_memcg_params(struct rcu_head *rcu) | |
186 | { | |
187 | struct memcg_cache_array *old; | |
188 | ||
189 | old = container_of(rcu, struct memcg_cache_array, rcu); | |
190 | kvfree(old); | |
191 | } | |
192 | ||
193 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) | |
194 | { | |
195 | struct memcg_cache_array *old, *new; | |
196 | ||
197 | new = kvzalloc(sizeof(struct memcg_cache_array) + | |
198 | new_array_size * sizeof(void *), GFP_KERNEL); | |
199 | if (!new) | |
200 | return -ENOMEM; | |
201 | ||
202 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
203 | lockdep_is_held(&slab_mutex)); | |
204 | if (old) | |
205 | memcpy(new->entries, old->entries, | |
206 | memcg_nr_cache_ids * sizeof(void *)); | |
207 | ||
208 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); | |
209 | if (old) | |
210 | call_rcu(&old->rcu, free_memcg_params); | |
211 | return 0; | |
212 | } | |
213 | ||
214 | int memcg_update_all_caches(int num_memcgs) | |
215 | { | |
216 | struct kmem_cache *s; | |
217 | int ret = 0; | |
218 | ||
219 | mutex_lock(&slab_mutex); | |
220 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { | |
221 | ret = update_memcg_params(s, num_memcgs); | |
222 | /* | |
223 | * Instead of freeing the memory, we'll just leave the caches | |
224 | * up to this point in an updated state. | |
225 | */ | |
226 | if (ret) | |
227 | break; | |
228 | } | |
229 | mutex_unlock(&slab_mutex); | |
230 | return ret; | |
231 | } | |
232 | ||
233 | void memcg_link_cache(struct kmem_cache *s) | |
234 | { | |
235 | if (is_root_cache(s)) { | |
236 | list_add(&s->root_caches_node, &slab_root_caches); | |
237 | } else { | |
238 | list_add(&s->memcg_params.children_node, | |
239 | &s->memcg_params.root_cache->memcg_params.children); | |
240 | list_add(&s->memcg_params.kmem_caches_node, | |
241 | &s->memcg_params.memcg->kmem_caches); | |
242 | } | |
243 | } | |
244 | ||
245 | static void memcg_unlink_cache(struct kmem_cache *s) | |
246 | { | |
247 | if (is_root_cache(s)) { | |
248 | list_del(&s->root_caches_node); | |
249 | } else { | |
250 | list_del(&s->memcg_params.children_node); | |
251 | list_del(&s->memcg_params.kmem_caches_node); | |
252 | } | |
253 | } | |
254 | #else | |
255 | static inline int init_memcg_params(struct kmem_cache *s, | |
256 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
257 | { | |
258 | return 0; | |
259 | } | |
260 | ||
261 | static inline void destroy_memcg_params(struct kmem_cache *s) | |
262 | { | |
263 | } | |
264 | ||
265 | static inline void memcg_unlink_cache(struct kmem_cache *s) | |
266 | { | |
267 | } | |
268 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ | |
269 | ||
270 | /* | |
271 | * Find a mergeable slab cache | |
272 | */ | |
273 | int slab_unmergeable(struct kmem_cache *s) | |
274 | { | |
275 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
276 | return 1; | |
277 | ||
278 | if (!is_root_cache(s)) | |
279 | return 1; | |
280 | ||
281 | if (s->ctor) | |
282 | return 1; | |
283 | ||
284 | /* | |
285 | * We may have set a slab to be unmergeable during bootstrap. | |
286 | */ | |
287 | if (s->refcount < 0) | |
288 | return 1; | |
289 | ||
290 | return 0; | |
291 | } | |
292 | ||
293 | struct kmem_cache *find_mergeable(size_t size, size_t align, | |
294 | slab_flags_t flags, const char *name, void (*ctor)(void *)) | |
295 | { | |
296 | struct kmem_cache *s; | |
297 | ||
298 | if (slab_nomerge) | |
299 | return NULL; | |
300 | ||
301 | if (ctor) | |
302 | return NULL; | |
303 | ||
304 | size = ALIGN(size, sizeof(void *)); | |
305 | align = calculate_alignment(flags, align, size); | |
306 | size = ALIGN(size, align); | |
307 | flags = kmem_cache_flags(size, flags, name, NULL); | |
308 | ||
309 | if (flags & SLAB_NEVER_MERGE) | |
310 | return NULL; | |
311 | ||
312 | list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) { | |
313 | if (slab_unmergeable(s)) | |
314 | continue; | |
315 | ||
316 | if (size > s->size) | |
317 | continue; | |
318 | ||
319 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
320 | continue; | |
321 | /* | |
322 | * Check if alignment is compatible. | |
323 | * Courtesy of Adrian Drzewiecki | |
324 | */ | |
325 | if ((s->size & ~(align - 1)) != s->size) | |
326 | continue; | |
327 | ||
328 | if (s->size - size >= sizeof(void *)) | |
329 | continue; | |
330 | ||
331 | if (IS_ENABLED(CONFIG_SLAB) && align && | |
332 | (align > s->align || s->align % align)) | |
333 | continue; | |
334 | ||
335 | return s; | |
336 | } | |
337 | return NULL; | |
338 | } | |
339 | ||
340 | /* | |
341 | * Figure out what the alignment of the objects will be given a set of | |
342 | * flags, a user specified alignment and the size of the objects. | |
343 | */ | |
344 | unsigned long calculate_alignment(slab_flags_t flags, | |
345 | unsigned long align, unsigned long size) | |
346 | { | |
347 | /* | |
348 | * If the user wants hardware cache aligned objects then follow that | |
349 | * suggestion if the object is sufficiently large. | |
350 | * | |
351 | * The hardware cache alignment cannot override the specified | |
352 | * alignment though. If that is greater then use it. | |
353 | */ | |
354 | if (flags & SLAB_HWCACHE_ALIGN) { | |
355 | unsigned long ralign = cache_line_size(); | |
356 | while (size <= ralign / 2) | |
357 | ralign /= 2; | |
358 | align = max(align, ralign); | |
359 | } | |
360 | ||
361 | if (align < ARCH_SLAB_MINALIGN) | |
362 | align = ARCH_SLAB_MINALIGN; | |
363 | ||
364 | return ALIGN(align, sizeof(void *)); | |
365 | } | |
366 | ||
367 | static struct kmem_cache *create_cache(const char *name, | |
368 | size_t object_size, size_t size, size_t align, | |
369 | slab_flags_t flags, void (*ctor)(void *), | |
370 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
371 | { | |
372 | struct kmem_cache *s; | |
373 | int err; | |
374 | ||
375 | err = -ENOMEM; | |
376 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
377 | if (!s) | |
378 | goto out; | |
379 | ||
380 | s->name = name; | |
381 | s->object_size = object_size; | |
382 | s->size = size; | |
383 | s->align = align; | |
384 | s->ctor = ctor; | |
385 | ||
386 | err = init_memcg_params(s, memcg, root_cache); | |
387 | if (err) | |
388 | goto out_free_cache; | |
389 | ||
390 | err = __kmem_cache_create(s, flags); | |
391 | if (err) | |
392 | goto out_free_cache; | |
393 | ||
394 | s->refcount = 1; | |
395 | list_add(&s->list, &slab_caches); | |
396 | memcg_link_cache(s); | |
397 | out: | |
398 | if (err) | |
399 | return ERR_PTR(err); | |
400 | return s; | |
401 | ||
402 | out_free_cache: | |
403 | destroy_memcg_params(s); | |
404 | kmem_cache_free(kmem_cache, s); | |
405 | goto out; | |
406 | } | |
407 | ||
408 | /* | |
409 | * kmem_cache_create - Create a cache. | |
410 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
411 | * @size: The size of objects to be created in this cache. | |
412 | * @align: The required alignment for the objects. | |
413 | * @flags: SLAB flags | |
414 | * @ctor: A constructor for the objects. | |
415 | * | |
416 | * Returns a ptr to the cache on success, NULL on failure. | |
417 | * Cannot be called within a interrupt, but can be interrupted. | |
418 | * The @ctor is run when new pages are allocated by the cache. | |
419 | * | |
420 | * The flags are | |
421 | * | |
422 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
423 | * to catch references to uninitialised memory. | |
424 | * | |
425 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
426 | * for buffer overruns. | |
427 | * | |
428 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
429 | * cacheline. This can be beneficial if you're counting cycles as closely | |
430 | * as davem. | |
431 | */ | |
432 | struct kmem_cache * | |
433 | kmem_cache_create(const char *name, size_t size, size_t align, | |
434 | slab_flags_t flags, void (*ctor)(void *)) | |
435 | { | |
436 | struct kmem_cache *s = NULL; | |
437 | const char *cache_name; | |
438 | int err; | |
439 | ||
440 | get_online_cpus(); | |
441 | get_online_mems(); | |
442 | memcg_get_cache_ids(); | |
443 | ||
444 | mutex_lock(&slab_mutex); | |
445 | ||
446 | err = kmem_cache_sanity_check(name, size); | |
447 | if (err) { | |
448 | goto out_unlock; | |
449 | } | |
450 | ||
451 | /* Refuse requests with allocator specific flags */ | |
452 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
453 | err = -EINVAL; | |
454 | goto out_unlock; | |
455 | } | |
456 | ||
457 | /* | |
458 | * Some allocators will constraint the set of valid flags to a subset | |
459 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
460 | * case, and we'll just provide them with a sanitized version of the | |
461 | * passed flags. | |
462 | */ | |
463 | flags &= CACHE_CREATE_MASK; | |
464 | ||
465 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
466 | if (s) | |
467 | goto out_unlock; | |
468 | ||
469 | cache_name = kstrdup_const(name, GFP_KERNEL); | |
470 | if (!cache_name) { | |
471 | err = -ENOMEM; | |
472 | goto out_unlock; | |
473 | } | |
474 | ||
475 | s = create_cache(cache_name, size, size, | |
476 | calculate_alignment(flags, align, size), | |
477 | flags, ctor, NULL, NULL); | |
478 | if (IS_ERR(s)) { | |
479 | err = PTR_ERR(s); | |
480 | kfree_const(cache_name); | |
481 | } | |
482 | ||
483 | out_unlock: | |
484 | mutex_unlock(&slab_mutex); | |
485 | ||
486 | memcg_put_cache_ids(); | |
487 | put_online_mems(); | |
488 | put_online_cpus(); | |
489 | ||
490 | if (err) { | |
491 | if (flags & SLAB_PANIC) | |
492 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
493 | name, err); | |
494 | else { | |
495 | pr_warn("kmem_cache_create(%s) failed with error %d\n", | |
496 | name, err); | |
497 | dump_stack(); | |
498 | } | |
499 | return NULL; | |
500 | } | |
501 | return s; | |
502 | } | |
503 | EXPORT_SYMBOL(kmem_cache_create); | |
504 | ||
505 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) | |
506 | { | |
507 | LIST_HEAD(to_destroy); | |
508 | struct kmem_cache *s, *s2; | |
509 | ||
510 | /* | |
511 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the | |
512 | * @slab_caches_to_rcu_destroy list. The slab pages are freed | |
513 | * through RCU and and the associated kmem_cache are dereferenced | |
514 | * while freeing the pages, so the kmem_caches should be freed only | |
515 | * after the pending RCU operations are finished. As rcu_barrier() | |
516 | * is a pretty slow operation, we batch all pending destructions | |
517 | * asynchronously. | |
518 | */ | |
519 | mutex_lock(&slab_mutex); | |
520 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
521 | mutex_unlock(&slab_mutex); | |
522 | ||
523 | if (list_empty(&to_destroy)) | |
524 | return; | |
525 | ||
526 | rcu_barrier(); | |
527 | ||
528 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
529 | #ifdef SLAB_SUPPORTS_SYSFS | |
530 | sysfs_slab_release(s); | |
531 | #else | |
532 | slab_kmem_cache_release(s); | |
533 | #endif | |
534 | } | |
535 | } | |
536 | ||
537 | static int shutdown_cache(struct kmem_cache *s) | |
538 | { | |
539 | /* free asan quarantined objects */ | |
540 | kasan_cache_shutdown(s); | |
541 | ||
542 | if (__kmem_cache_shutdown(s) != 0) | |
543 | return -EBUSY; | |
544 | ||
545 | memcg_unlink_cache(s); | |
546 | list_del(&s->list); | |
547 | ||
548 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { | |
549 | #ifdef SLAB_SUPPORTS_SYSFS | |
550 | sysfs_slab_unlink(s); | |
551 | #endif | |
552 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); | |
553 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
554 | } else { | |
555 | #ifdef SLAB_SUPPORTS_SYSFS | |
556 | sysfs_slab_unlink(s); | |
557 | sysfs_slab_release(s); | |
558 | #else | |
559 | slab_kmem_cache_release(s); | |
560 | #endif | |
561 | } | |
562 | ||
563 | return 0; | |
564 | } | |
565 | ||
566 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) | |
567 | /* | |
568 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. | |
569 | * @memcg: The memory cgroup the new cache is for. | |
570 | * @root_cache: The parent of the new cache. | |
571 | * | |
572 | * This function attempts to create a kmem cache that will serve allocation | |
573 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
574 | * from its parent. | |
575 | */ | |
576 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, | |
577 | struct kmem_cache *root_cache) | |
578 | { | |
579 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ | |
580 | struct cgroup_subsys_state *css = &memcg->css; | |
581 | struct memcg_cache_array *arr; | |
582 | struct kmem_cache *s = NULL; | |
583 | char *cache_name; | |
584 | int idx; | |
585 | ||
586 | get_online_cpus(); | |
587 | get_online_mems(); | |
588 | ||
589 | mutex_lock(&slab_mutex); | |
590 | ||
591 | /* | |
592 | * The memory cgroup could have been offlined while the cache | |
593 | * creation work was pending. | |
594 | */ | |
595 | if (memcg->kmem_state != KMEM_ONLINE) | |
596 | goto out_unlock; | |
597 | ||
598 | idx = memcg_cache_id(memcg); | |
599 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
600 | lockdep_is_held(&slab_mutex)); | |
601 | ||
602 | /* | |
603 | * Since per-memcg caches are created asynchronously on first | |
604 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
605 | * create the same cache, but only one of them may succeed. | |
606 | */ | |
607 | if (arr->entries[idx]) | |
608 | goto out_unlock; | |
609 | ||
610 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); | |
611 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, | |
612 | css->serial_nr, memcg_name_buf); | |
613 | if (!cache_name) | |
614 | goto out_unlock; | |
615 | ||
616 | s = create_cache(cache_name, root_cache->object_size, | |
617 | root_cache->size, root_cache->align, | |
618 | root_cache->flags & CACHE_CREATE_MASK, | |
619 | root_cache->ctor, memcg, root_cache); | |
620 | /* | |
621 | * If we could not create a memcg cache, do not complain, because | |
622 | * that's not critical at all as we can always proceed with the root | |
623 | * cache. | |
624 | */ | |
625 | if (IS_ERR(s)) { | |
626 | kfree(cache_name); | |
627 | goto out_unlock; | |
628 | } | |
629 | ||
630 | /* | |
631 | * Since readers won't lock (see cache_from_memcg_idx()), we need a | |
632 | * barrier here to ensure nobody will see the kmem_cache partially | |
633 | * initialized. | |
634 | */ | |
635 | smp_wmb(); | |
636 | arr->entries[idx] = s; | |
637 | ||
638 | out_unlock: | |
639 | mutex_unlock(&slab_mutex); | |
640 | ||
641 | put_online_mems(); | |
642 | put_online_cpus(); | |
643 | } | |
644 | ||
645 | static void kmemcg_deactivate_workfn(struct work_struct *work) | |
646 | { | |
647 | struct kmem_cache *s = container_of(work, struct kmem_cache, | |
648 | memcg_params.deact_work); | |
649 | ||
650 | get_online_cpus(); | |
651 | get_online_mems(); | |
652 | ||
653 | mutex_lock(&slab_mutex); | |
654 | ||
655 | s->memcg_params.deact_fn(s); | |
656 | ||
657 | mutex_unlock(&slab_mutex); | |
658 | ||
659 | put_online_mems(); | |
660 | put_online_cpus(); | |
661 | ||
662 | /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */ | |
663 | css_put(&s->memcg_params.memcg->css); | |
664 | } | |
665 | ||
666 | static void kmemcg_deactivate_rcufn(struct rcu_head *head) | |
667 | { | |
668 | struct kmem_cache *s = container_of(head, struct kmem_cache, | |
669 | memcg_params.deact_rcu_head); | |
670 | ||
671 | /* | |
672 | * We need to grab blocking locks. Bounce to ->deact_work. The | |
673 | * work item shares the space with the RCU head and can't be | |
674 | * initialized eariler. | |
675 | */ | |
676 | INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn); | |
677 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work); | |
678 | } | |
679 | ||
680 | /** | |
681 | * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a | |
682 | * sched RCU grace period | |
683 | * @s: target kmem_cache | |
684 | * @deact_fn: deactivation function to call | |
685 | * | |
686 | * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex | |
687 | * held after a sched RCU grace period. The slab is guaranteed to stay | |
688 | * alive until @deact_fn is finished. This is to be used from | |
689 | * __kmemcg_cache_deactivate(). | |
690 | */ | |
691 | void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s, | |
692 | void (*deact_fn)(struct kmem_cache *)) | |
693 | { | |
694 | if (WARN_ON_ONCE(is_root_cache(s)) || | |
695 | WARN_ON_ONCE(s->memcg_params.deact_fn)) | |
696 | return; | |
697 | ||
698 | /* pin memcg so that @s doesn't get destroyed in the middle */ | |
699 | css_get(&s->memcg_params.memcg->css); | |
700 | ||
701 | s->memcg_params.deact_fn = deact_fn; | |
702 | call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn); | |
703 | } | |
704 | ||
705 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg) | |
706 | { | |
707 | int idx; | |
708 | struct memcg_cache_array *arr; | |
709 | struct kmem_cache *s, *c; | |
710 | ||
711 | idx = memcg_cache_id(memcg); | |
712 | ||
713 | get_online_cpus(); | |
714 | get_online_mems(); | |
715 | ||
716 | mutex_lock(&slab_mutex); | |
717 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { | |
718 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
719 | lockdep_is_held(&slab_mutex)); | |
720 | c = arr->entries[idx]; | |
721 | if (!c) | |
722 | continue; | |
723 | ||
724 | __kmemcg_cache_deactivate(c); | |
725 | arr->entries[idx] = NULL; | |
726 | } | |
727 | mutex_unlock(&slab_mutex); | |
728 | ||
729 | put_online_mems(); | |
730 | put_online_cpus(); | |
731 | } | |
732 | ||
733 | void memcg_destroy_kmem_caches(struct mem_cgroup *memcg) | |
734 | { | |
735 | struct kmem_cache *s, *s2; | |
736 | ||
737 | get_online_cpus(); | |
738 | get_online_mems(); | |
739 | ||
740 | mutex_lock(&slab_mutex); | |
741 | list_for_each_entry_safe(s, s2, &memcg->kmem_caches, | |
742 | memcg_params.kmem_caches_node) { | |
743 | /* | |
744 | * The cgroup is about to be freed and therefore has no charges | |
745 | * left. Hence, all its caches must be empty by now. | |
746 | */ | |
747 | BUG_ON(shutdown_cache(s)); | |
748 | } | |
749 | mutex_unlock(&slab_mutex); | |
750 | ||
751 | put_online_mems(); | |
752 | put_online_cpus(); | |
753 | } | |
754 | ||
755 | static int shutdown_memcg_caches(struct kmem_cache *s) | |
756 | { | |
757 | struct memcg_cache_array *arr; | |
758 | struct kmem_cache *c, *c2; | |
759 | LIST_HEAD(busy); | |
760 | int i; | |
761 | ||
762 | BUG_ON(!is_root_cache(s)); | |
763 | ||
764 | /* | |
765 | * First, shutdown active caches, i.e. caches that belong to online | |
766 | * memory cgroups. | |
767 | */ | |
768 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
769 | lockdep_is_held(&slab_mutex)); | |
770 | for_each_memcg_cache_index(i) { | |
771 | c = arr->entries[i]; | |
772 | if (!c) | |
773 | continue; | |
774 | if (shutdown_cache(c)) | |
775 | /* | |
776 | * The cache still has objects. Move it to a temporary | |
777 | * list so as not to try to destroy it for a second | |
778 | * time while iterating over inactive caches below. | |
779 | */ | |
780 | list_move(&c->memcg_params.children_node, &busy); | |
781 | else | |
782 | /* | |
783 | * The cache is empty and will be destroyed soon. Clear | |
784 | * the pointer to it in the memcg_caches array so that | |
785 | * it will never be accessed even if the root cache | |
786 | * stays alive. | |
787 | */ | |
788 | arr->entries[i] = NULL; | |
789 | } | |
790 | ||
791 | /* | |
792 | * Second, shutdown all caches left from memory cgroups that are now | |
793 | * offline. | |
794 | */ | |
795 | list_for_each_entry_safe(c, c2, &s->memcg_params.children, | |
796 | memcg_params.children_node) | |
797 | shutdown_cache(c); | |
798 | ||
799 | list_splice(&busy, &s->memcg_params.children); | |
800 | ||
801 | /* | |
802 | * A cache being destroyed must be empty. In particular, this means | |
803 | * that all per memcg caches attached to it must be empty too. | |
804 | */ | |
805 | if (!list_empty(&s->memcg_params.children)) | |
806 | return -EBUSY; | |
807 | return 0; | |
808 | } | |
809 | #else | |
810 | static inline int shutdown_memcg_caches(struct kmem_cache *s) | |
811 | { | |
812 | return 0; | |
813 | } | |
814 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ | |
815 | ||
816 | void slab_kmem_cache_release(struct kmem_cache *s) | |
817 | { | |
818 | __kmem_cache_release(s); | |
819 | destroy_memcg_params(s); | |
820 | kfree_const(s->name); | |
821 | kmem_cache_free(kmem_cache, s); | |
822 | } | |
823 | ||
824 | void kmem_cache_destroy(struct kmem_cache *s) | |
825 | { | |
826 | int err; | |
827 | ||
828 | if (unlikely(!s)) | |
829 | return; | |
830 | ||
831 | get_online_cpus(); | |
832 | get_online_mems(); | |
833 | ||
834 | mutex_lock(&slab_mutex); | |
835 | ||
836 | s->refcount--; | |
837 | if (s->refcount) | |
838 | goto out_unlock; | |
839 | ||
840 | err = shutdown_memcg_caches(s); | |
841 | if (!err) | |
842 | err = shutdown_cache(s); | |
843 | ||
844 | if (err) { | |
845 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", | |
846 | s->name); | |
847 | dump_stack(); | |
848 | } | |
849 | out_unlock: | |
850 | mutex_unlock(&slab_mutex); | |
851 | ||
852 | put_online_mems(); | |
853 | put_online_cpus(); | |
854 | } | |
855 | EXPORT_SYMBOL(kmem_cache_destroy); | |
856 | ||
857 | /** | |
858 | * kmem_cache_shrink - Shrink a cache. | |
859 | * @cachep: The cache to shrink. | |
860 | * | |
861 | * Releases as many slabs as possible for a cache. | |
862 | * To help debugging, a zero exit status indicates all slabs were released. | |
863 | */ | |
864 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
865 | { | |
866 | int ret; | |
867 | ||
868 | get_online_cpus(); | |
869 | get_online_mems(); | |
870 | kasan_cache_shrink(cachep); | |
871 | ret = __kmem_cache_shrink(cachep); | |
872 | put_online_mems(); | |
873 | put_online_cpus(); | |
874 | return ret; | |
875 | } | |
876 | EXPORT_SYMBOL(kmem_cache_shrink); | |
877 | ||
878 | bool slab_is_available(void) | |
879 | { | |
880 | return slab_state >= UP; | |
881 | } | |
882 | ||
883 | #ifndef CONFIG_SLOB | |
884 | /* Create a cache during boot when no slab services are available yet */ | |
885 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
886 | slab_flags_t flags) | |
887 | { | |
888 | int err; | |
889 | ||
890 | s->name = name; | |
891 | s->size = s->object_size = size; | |
892 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); | |
893 | ||
894 | slab_init_memcg_params(s); | |
895 | ||
896 | err = __kmem_cache_create(s, flags); | |
897 | ||
898 | if (err) | |
899 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", | |
900 | name, size, err); | |
901 | ||
902 | s->refcount = -1; /* Exempt from merging for now */ | |
903 | } | |
904 | ||
905 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
906 | slab_flags_t flags) | |
907 | { | |
908 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
909 | ||
910 | if (!s) | |
911 | panic("Out of memory when creating slab %s\n", name); | |
912 | ||
913 | create_boot_cache(s, name, size, flags); | |
914 | list_add(&s->list, &slab_caches); | |
915 | memcg_link_cache(s); | |
916 | s->refcount = 1; | |
917 | return s; | |
918 | } | |
919 | ||
920 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; | |
921 | EXPORT_SYMBOL(kmalloc_caches); | |
922 | ||
923 | #ifdef CONFIG_ZONE_DMA | |
924 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
925 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
926 | #endif | |
927 | ||
928 | /* | |
929 | * Conversion table for small slabs sizes / 8 to the index in the | |
930 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
931 | * of two cache sizes there. The size of larger slabs can be determined using | |
932 | * fls. | |
933 | */ | |
934 | static s8 size_index[24] = { | |
935 | 3, /* 8 */ | |
936 | 4, /* 16 */ | |
937 | 5, /* 24 */ | |
938 | 5, /* 32 */ | |
939 | 6, /* 40 */ | |
940 | 6, /* 48 */ | |
941 | 6, /* 56 */ | |
942 | 6, /* 64 */ | |
943 | 1, /* 72 */ | |
944 | 1, /* 80 */ | |
945 | 1, /* 88 */ | |
946 | 1, /* 96 */ | |
947 | 7, /* 104 */ | |
948 | 7, /* 112 */ | |
949 | 7, /* 120 */ | |
950 | 7, /* 128 */ | |
951 | 2, /* 136 */ | |
952 | 2, /* 144 */ | |
953 | 2, /* 152 */ | |
954 | 2, /* 160 */ | |
955 | 2, /* 168 */ | |
956 | 2, /* 176 */ | |
957 | 2, /* 184 */ | |
958 | 2 /* 192 */ | |
959 | }; | |
960 | ||
961 | static inline int size_index_elem(size_t bytes) | |
962 | { | |
963 | return (bytes - 1) / 8; | |
964 | } | |
965 | ||
966 | /* | |
967 | * Find the kmem_cache structure that serves a given size of | |
968 | * allocation | |
969 | */ | |
970 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
971 | { | |
972 | int index; | |
973 | ||
974 | if (unlikely(size > KMALLOC_MAX_SIZE)) { | |
975 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); | |
976 | return NULL; | |
977 | } | |
978 | ||
979 | if (size <= 192) { | |
980 | if (!size) | |
981 | return ZERO_SIZE_PTR; | |
982 | ||
983 | index = size_index[size_index_elem(size)]; | |
984 | } else | |
985 | index = fls(size - 1); | |
986 | ||
987 | #ifdef CONFIG_ZONE_DMA | |
988 | if (unlikely((flags & GFP_DMA))) | |
989 | return kmalloc_dma_caches[index]; | |
990 | ||
991 | #endif | |
992 | return kmalloc_caches[index]; | |
993 | } | |
994 | ||
995 | /* | |
996 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
997 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
998 | * kmalloc-67108864. | |
999 | */ | |
1000 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { | |
1001 | {NULL, 0}, {"kmalloc-96", 96}, | |
1002 | {"kmalloc-192", 192}, {"kmalloc-8", 8}, | |
1003 | {"kmalloc-16", 16}, {"kmalloc-32", 32}, | |
1004 | {"kmalloc-64", 64}, {"kmalloc-128", 128}, | |
1005 | {"kmalloc-256", 256}, {"kmalloc-512", 512}, | |
1006 | {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048}, | |
1007 | {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192}, | |
1008 | {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768}, | |
1009 | {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072}, | |
1010 | {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288}, | |
1011 | {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152}, | |
1012 | {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608}, | |
1013 | {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432}, | |
1014 | {"kmalloc-67108864", 67108864} | |
1015 | }; | |
1016 | ||
1017 | /* | |
1018 | * Patch up the size_index table if we have strange large alignment | |
1019 | * requirements for the kmalloc array. This is only the case for | |
1020 | * MIPS it seems. The standard arches will not generate any code here. | |
1021 | * | |
1022 | * Largest permitted alignment is 256 bytes due to the way we | |
1023 | * handle the index determination for the smaller caches. | |
1024 | * | |
1025 | * Make sure that nothing crazy happens if someone starts tinkering | |
1026 | * around with ARCH_KMALLOC_MINALIGN | |
1027 | */ | |
1028 | void __init setup_kmalloc_cache_index_table(void) | |
1029 | { | |
1030 | int i; | |
1031 | ||
1032 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || | |
1033 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
1034 | ||
1035 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
1036 | int elem = size_index_elem(i); | |
1037 | ||
1038 | if (elem >= ARRAY_SIZE(size_index)) | |
1039 | break; | |
1040 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
1041 | } | |
1042 | ||
1043 | if (KMALLOC_MIN_SIZE >= 64) { | |
1044 | /* | |
1045 | * The 96 byte size cache is not used if the alignment | |
1046 | * is 64 byte. | |
1047 | */ | |
1048 | for (i = 64 + 8; i <= 96; i += 8) | |
1049 | size_index[size_index_elem(i)] = 7; | |
1050 | ||
1051 | } | |
1052 | ||
1053 | if (KMALLOC_MIN_SIZE >= 128) { | |
1054 | /* | |
1055 | * The 192 byte sized cache is not used if the alignment | |
1056 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
1057 | * instead. | |
1058 | */ | |
1059 | for (i = 128 + 8; i <= 192; i += 8) | |
1060 | size_index[size_index_elem(i)] = 8; | |
1061 | } | |
1062 | } | |
1063 | ||
1064 | static void __init new_kmalloc_cache(int idx, slab_flags_t flags) | |
1065 | { | |
1066 | kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name, | |
1067 | kmalloc_info[idx].size, flags); | |
1068 | } | |
1069 | ||
1070 | /* | |
1071 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
1072 | * may already have been created because they were needed to | |
1073 | * enable allocations for slab creation. | |
1074 | */ | |
1075 | void __init create_kmalloc_caches(slab_flags_t flags) | |
1076 | { | |
1077 | int i; | |
1078 | ||
1079 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1080 | if (!kmalloc_caches[i]) | |
1081 | new_kmalloc_cache(i, flags); | |
1082 | ||
1083 | /* | |
1084 | * Caches that are not of the two-to-the-power-of size. | |
1085 | * These have to be created immediately after the | |
1086 | * earlier power of two caches | |
1087 | */ | |
1088 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) | |
1089 | new_kmalloc_cache(1, flags); | |
1090 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) | |
1091 | new_kmalloc_cache(2, flags); | |
1092 | } | |
1093 | ||
1094 | /* Kmalloc array is now usable */ | |
1095 | slab_state = UP; | |
1096 | ||
1097 | #ifdef CONFIG_ZONE_DMA | |
1098 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1099 | struct kmem_cache *s = kmalloc_caches[i]; | |
1100 | ||
1101 | if (s) { | |
1102 | int size = kmalloc_size(i); | |
1103 | char *n = kasprintf(GFP_NOWAIT, | |
1104 | "dma-kmalloc-%d", size); | |
1105 | ||
1106 | BUG_ON(!n); | |
1107 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
1108 | size, SLAB_CACHE_DMA | flags); | |
1109 | } | |
1110 | } | |
1111 | #endif | |
1112 | } | |
1113 | #endif /* !CONFIG_SLOB */ | |
1114 | ||
1115 | /* | |
1116 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1117 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1118 | * know the allocation order to free the pages properly in kfree. | |
1119 | */ | |
1120 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) | |
1121 | { | |
1122 | void *ret; | |
1123 | struct page *page; | |
1124 | ||
1125 | flags |= __GFP_COMP; | |
1126 | page = alloc_pages(flags, order); | |
1127 | ret = page ? page_address(page) : NULL; | |
1128 | kmemleak_alloc(ret, size, 1, flags); | |
1129 | kasan_kmalloc_large(ret, size, flags); | |
1130 | return ret; | |
1131 | } | |
1132 | EXPORT_SYMBOL(kmalloc_order); | |
1133 | ||
1134 | #ifdef CONFIG_TRACING | |
1135 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1136 | { | |
1137 | void *ret = kmalloc_order(size, flags, order); | |
1138 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1139 | return ret; | |
1140 | } | |
1141 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1142 | #endif | |
1143 | ||
1144 | #ifdef CONFIG_SLAB_FREELIST_RANDOM | |
1145 | /* Randomize a generic freelist */ | |
1146 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
1147 | size_t count) | |
1148 | { | |
1149 | size_t i; | |
1150 | unsigned int rand; | |
1151 | ||
1152 | for (i = 0; i < count; i++) | |
1153 | list[i] = i; | |
1154 | ||
1155 | /* Fisher-Yates shuffle */ | |
1156 | for (i = count - 1; i > 0; i--) { | |
1157 | rand = prandom_u32_state(state); | |
1158 | rand %= (i + 1); | |
1159 | swap(list[i], list[rand]); | |
1160 | } | |
1161 | } | |
1162 | ||
1163 | /* Create a random sequence per cache */ | |
1164 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1165 | gfp_t gfp) | |
1166 | { | |
1167 | struct rnd_state state; | |
1168 | ||
1169 | if (count < 2 || cachep->random_seq) | |
1170 | return 0; | |
1171 | ||
1172 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1173 | if (!cachep->random_seq) | |
1174 | return -ENOMEM; | |
1175 | ||
1176 | /* Get best entropy at this stage of boot */ | |
1177 | prandom_seed_state(&state, get_random_long()); | |
1178 | ||
1179 | freelist_randomize(&state, cachep->random_seq, count); | |
1180 | return 0; | |
1181 | } | |
1182 | ||
1183 | /* Destroy the per-cache random freelist sequence */ | |
1184 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1185 | { | |
1186 | kfree(cachep->random_seq); | |
1187 | cachep->random_seq = NULL; | |
1188 | } | |
1189 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1190 | ||
1191 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) | |
1192 | #ifdef CONFIG_SLAB | |
1193 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
1194 | #else | |
1195 | #define SLABINFO_RIGHTS S_IRUSR | |
1196 | #endif | |
1197 | ||
1198 | static void print_slabinfo_header(struct seq_file *m) | |
1199 | { | |
1200 | /* | |
1201 | * Output format version, so at least we can change it | |
1202 | * without _too_ many complaints. | |
1203 | */ | |
1204 | #ifdef CONFIG_DEBUG_SLAB | |
1205 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1206 | #else | |
1207 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1208 | #endif | |
1209 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); | |
1210 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
1211 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1212 | #ifdef CONFIG_DEBUG_SLAB | |
1213 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); | |
1214 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); | |
1215 | #endif | |
1216 | seq_putc(m, '\n'); | |
1217 | } | |
1218 | ||
1219 | void *slab_start(struct seq_file *m, loff_t *pos) | |
1220 | { | |
1221 | mutex_lock(&slab_mutex); | |
1222 | return seq_list_start(&slab_root_caches, *pos); | |
1223 | } | |
1224 | ||
1225 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1226 | { | |
1227 | return seq_list_next(p, &slab_root_caches, pos); | |
1228 | } | |
1229 | ||
1230 | void slab_stop(struct seq_file *m, void *p) | |
1231 | { | |
1232 | mutex_unlock(&slab_mutex); | |
1233 | } | |
1234 | ||
1235 | static void | |
1236 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1237 | { | |
1238 | struct kmem_cache *c; | |
1239 | struct slabinfo sinfo; | |
1240 | ||
1241 | if (!is_root_cache(s)) | |
1242 | return; | |
1243 | ||
1244 | for_each_memcg_cache(c, s) { | |
1245 | memset(&sinfo, 0, sizeof(sinfo)); | |
1246 | get_slabinfo(c, &sinfo); | |
1247 | ||
1248 | info->active_slabs += sinfo.active_slabs; | |
1249 | info->num_slabs += sinfo.num_slabs; | |
1250 | info->shared_avail += sinfo.shared_avail; | |
1251 | info->active_objs += sinfo.active_objs; | |
1252 | info->num_objs += sinfo.num_objs; | |
1253 | } | |
1254 | } | |
1255 | ||
1256 | static void cache_show(struct kmem_cache *s, struct seq_file *m) | |
1257 | { | |
1258 | struct slabinfo sinfo; | |
1259 | ||
1260 | memset(&sinfo, 0, sizeof(sinfo)); | |
1261 | get_slabinfo(s, &sinfo); | |
1262 | ||
1263 | memcg_accumulate_slabinfo(s, &sinfo); | |
1264 | ||
1265 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", | |
1266 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, | |
1267 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); | |
1268 | ||
1269 | seq_printf(m, " : tunables %4u %4u %4u", | |
1270 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1271 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1272 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1273 | slabinfo_show_stats(m, s); | |
1274 | seq_putc(m, '\n'); | |
1275 | } | |
1276 | ||
1277 | static int slab_show(struct seq_file *m, void *p) | |
1278 | { | |
1279 | struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node); | |
1280 | ||
1281 | if (p == slab_root_caches.next) | |
1282 | print_slabinfo_header(m); | |
1283 | cache_show(s, m); | |
1284 | return 0; | |
1285 | } | |
1286 | ||
1287 | void dump_unreclaimable_slab(void) | |
1288 | { | |
1289 | struct kmem_cache *s, *s2; | |
1290 | struct slabinfo sinfo; | |
1291 | ||
1292 | /* | |
1293 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1294 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1295 | * risk of crash. | |
1296 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1297 | * without acquiring the mutex. | |
1298 | */ | |
1299 | if (!mutex_trylock(&slab_mutex)) { | |
1300 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1301 | return; | |
1302 | } | |
1303 | ||
1304 | pr_info("Unreclaimable slab info:\n"); | |
1305 | pr_info("Name Used Total\n"); | |
1306 | ||
1307 | list_for_each_entry_safe(s, s2, &slab_caches, list) { | |
1308 | if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT)) | |
1309 | continue; | |
1310 | ||
1311 | get_slabinfo(s, &sinfo); | |
1312 | ||
1313 | if (sinfo.num_objs > 0) | |
1314 | pr_info("%-17s %10luKB %10luKB\n", cache_name(s), | |
1315 | (sinfo.active_objs * s->size) / 1024, | |
1316 | (sinfo.num_objs * s->size) / 1024); | |
1317 | } | |
1318 | mutex_unlock(&slab_mutex); | |
1319 | } | |
1320 | ||
1321 | #if defined(CONFIG_MEMCG) | |
1322 | void *memcg_slab_start(struct seq_file *m, loff_t *pos) | |
1323 | { | |
1324 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1325 | ||
1326 | mutex_lock(&slab_mutex); | |
1327 | return seq_list_start(&memcg->kmem_caches, *pos); | |
1328 | } | |
1329 | ||
1330 | void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1331 | { | |
1332 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1333 | ||
1334 | return seq_list_next(p, &memcg->kmem_caches, pos); | |
1335 | } | |
1336 | ||
1337 | void memcg_slab_stop(struct seq_file *m, void *p) | |
1338 | { | |
1339 | mutex_unlock(&slab_mutex); | |
1340 | } | |
1341 | ||
1342 | int memcg_slab_show(struct seq_file *m, void *p) | |
1343 | { | |
1344 | struct kmem_cache *s = list_entry(p, struct kmem_cache, | |
1345 | memcg_params.kmem_caches_node); | |
1346 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1347 | ||
1348 | if (p == memcg->kmem_caches.next) | |
1349 | print_slabinfo_header(m); | |
1350 | cache_show(s, m); | |
1351 | return 0; | |
1352 | } | |
1353 | #endif | |
1354 | ||
1355 | /* | |
1356 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1357 | * | |
1358 | * Output layout: | |
1359 | * cache-name | |
1360 | * num-active-objs | |
1361 | * total-objs | |
1362 | * object size | |
1363 | * num-active-slabs | |
1364 | * total-slabs | |
1365 | * num-pages-per-slab | |
1366 | * + further values on SMP and with statistics enabled | |
1367 | */ | |
1368 | static const struct seq_operations slabinfo_op = { | |
1369 | .start = slab_start, | |
1370 | .next = slab_next, | |
1371 | .stop = slab_stop, | |
1372 | .show = slab_show, | |
1373 | }; | |
1374 | ||
1375 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1376 | { | |
1377 | return seq_open(file, &slabinfo_op); | |
1378 | } | |
1379 | ||
1380 | static const struct file_operations proc_slabinfo_operations = { | |
1381 | .open = slabinfo_open, | |
1382 | .read = seq_read, | |
1383 | .write = slabinfo_write, | |
1384 | .llseek = seq_lseek, | |
1385 | .release = seq_release, | |
1386 | }; | |
1387 | ||
1388 | static int __init slab_proc_init(void) | |
1389 | { | |
1390 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, | |
1391 | &proc_slabinfo_operations); | |
1392 | return 0; | |
1393 | } | |
1394 | module_init(slab_proc_init); | |
1395 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ | |
1396 | ||
1397 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1398 | gfp_t flags) | |
1399 | { | |
1400 | void *ret; | |
1401 | size_t ks = 0; | |
1402 | ||
1403 | if (p) | |
1404 | ks = ksize(p); | |
1405 | ||
1406 | if (ks >= new_size) { | |
1407 | kasan_krealloc((void *)p, new_size, flags); | |
1408 | return (void *)p; | |
1409 | } | |
1410 | ||
1411 | ret = kmalloc_track_caller(new_size, flags); | |
1412 | if (ret && p) | |
1413 | memcpy(ret, p, ks); | |
1414 | ||
1415 | return ret; | |
1416 | } | |
1417 | ||
1418 | /** | |
1419 | * __krealloc - like krealloc() but don't free @p. | |
1420 | * @p: object to reallocate memory for. | |
1421 | * @new_size: how many bytes of memory are required. | |
1422 | * @flags: the type of memory to allocate. | |
1423 | * | |
1424 | * This function is like krealloc() except it never frees the originally | |
1425 | * allocated buffer. Use this if you don't want to free the buffer immediately | |
1426 | * like, for example, with RCU. | |
1427 | */ | |
1428 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) | |
1429 | { | |
1430 | if (unlikely(!new_size)) | |
1431 | return ZERO_SIZE_PTR; | |
1432 | ||
1433 | return __do_krealloc(p, new_size, flags); | |
1434 | ||
1435 | } | |
1436 | EXPORT_SYMBOL(__krealloc); | |
1437 | ||
1438 | /** | |
1439 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1440 | * @p: object to reallocate memory for. | |
1441 | * @new_size: how many bytes of memory are required. | |
1442 | * @flags: the type of memory to allocate. | |
1443 | * | |
1444 | * The contents of the object pointed to are preserved up to the | |
1445 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1446 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1447 | * %NULL pointer, the object pointed to is freed. | |
1448 | */ | |
1449 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1450 | { | |
1451 | void *ret; | |
1452 | ||
1453 | if (unlikely(!new_size)) { | |
1454 | kfree(p); | |
1455 | return ZERO_SIZE_PTR; | |
1456 | } | |
1457 | ||
1458 | ret = __do_krealloc(p, new_size, flags); | |
1459 | if (ret && p != ret) | |
1460 | kfree(p); | |
1461 | ||
1462 | return ret; | |
1463 | } | |
1464 | EXPORT_SYMBOL(krealloc); | |
1465 | ||
1466 | /** | |
1467 | * kzfree - like kfree but zero memory | |
1468 | * @p: object to free memory of | |
1469 | * | |
1470 | * The memory of the object @p points to is zeroed before freed. | |
1471 | * If @p is %NULL, kzfree() does nothing. | |
1472 | * | |
1473 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1474 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1475 | * careful when using this function in performance sensitive code. | |
1476 | */ | |
1477 | void kzfree(const void *p) | |
1478 | { | |
1479 | size_t ks; | |
1480 | void *mem = (void *)p; | |
1481 | ||
1482 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1483 | return; | |
1484 | ks = ksize(mem); | |
1485 | memset(mem, 0, ks); | |
1486 | kfree(mem); | |
1487 | } | |
1488 | EXPORT_SYMBOL(kzfree); | |
1489 | ||
1490 | /* Tracepoints definitions. */ | |
1491 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1492 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1493 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1494 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1495 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1496 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |