Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
039363f3 CL |
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> | |
20cea968 CL |
15 | #include <linux/cpu.h> |
16 | #include <linux/uaccess.h> | |
b7454ad3 GC |
17 | #include <linux/seq_file.h> |
18 | #include <linux/proc_fs.h> | |
039363f3 CL |
19 | #include <asm/cacheflush.h> |
20 | #include <asm/tlbflush.h> | |
21 | #include <asm/page.h> | |
2633d7a0 | 22 | #include <linux/memcontrol.h> |
928cec9c AR |
23 | |
24 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 25 | #include <trace/events/kmem.h> |
039363f3 | 26 | |
97d06609 CL |
27 | #include "slab.h" |
28 | ||
29 | enum slab_state slab_state; | |
18004c5d CL |
30 | LIST_HEAD(slab_caches); |
31 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 32 | struct kmem_cache *kmem_cache; |
97d06609 | 33 | |
2d891fbc KC |
34 | #ifdef CONFIG_HARDENED_USERCOPY |
35 | bool usercopy_fallback __ro_after_init = | |
36 | IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK); | |
37 | module_param(usercopy_fallback, bool, 0400); | |
38 | MODULE_PARM_DESC(usercopy_fallback, | |
39 | "WARN instead of reject usercopy whitelist violations"); | |
40 | #endif | |
41 | ||
657dc2f9 TH |
42 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
43 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
44 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
45 | slab_caches_to_rcu_destroy_workfn); | |
46 | ||
423c929c JK |
47 | /* |
48 | * Set of flags that will prevent slab merging | |
49 | */ | |
50 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
5f0d5a3a | 51 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ |
7ed2f9e6 | 52 | SLAB_FAILSLAB | SLAB_KASAN) |
423c929c | 53 | |
230e9fc2 | 54 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
75f296d9 | 55 | SLAB_ACCOUNT) |
423c929c JK |
56 | |
57 | /* | |
58 | * Merge control. If this is set then no merging of slab caches will occur. | |
423c929c | 59 | */ |
7660a6fd | 60 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); |
423c929c JK |
61 | |
62 | static int __init setup_slab_nomerge(char *str) | |
63 | { | |
7660a6fd | 64 | slab_nomerge = true; |
423c929c JK |
65 | return 1; |
66 | } | |
67 | ||
68 | #ifdef CONFIG_SLUB | |
69 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
70 | #endif | |
71 | ||
72 | __setup("slab_nomerge", setup_slab_nomerge); | |
73 | ||
07f361b2 JK |
74 | /* |
75 | * Determine the size of a slab object | |
76 | */ | |
77 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
78 | { | |
79 | return s->object_size; | |
80 | } | |
81 | EXPORT_SYMBOL(kmem_cache_size); | |
82 | ||
77be4b13 | 83 | #ifdef CONFIG_DEBUG_VM |
794b1248 | 84 | static int kmem_cache_sanity_check(const char *name, size_t size) |
039363f3 CL |
85 | { |
86 | struct kmem_cache *s = NULL; | |
87 | ||
039363f3 CL |
88 | if (!name || in_interrupt() || size < sizeof(void *) || |
89 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
90 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
91 | return -EINVAL; | |
039363f3 | 92 | } |
b920536a | 93 | |
20cea968 CL |
94 | list_for_each_entry(s, &slab_caches, list) { |
95 | char tmp; | |
96 | int res; | |
97 | ||
98 | /* | |
99 | * This happens when the module gets unloaded and doesn't | |
100 | * destroy its slab cache and no-one else reuses the vmalloc | |
101 | * area of the module. Print a warning. | |
102 | */ | |
103 | res = probe_kernel_address(s->name, tmp); | |
104 | if (res) { | |
77be4b13 | 105 | pr_err("Slab cache with size %d has lost its name\n", |
20cea968 CL |
106 | s->object_size); |
107 | continue; | |
108 | } | |
20cea968 CL |
109 | } |
110 | ||
111 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
77be4b13 SK |
112 | return 0; |
113 | } | |
114 | #else | |
794b1248 | 115 | static inline int kmem_cache_sanity_check(const char *name, size_t size) |
77be4b13 SK |
116 | { |
117 | return 0; | |
118 | } | |
20cea968 CL |
119 | #endif |
120 | ||
484748f0 CL |
121 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
122 | { | |
123 | size_t i; | |
124 | ||
ca257195 JDB |
125 | for (i = 0; i < nr; i++) { |
126 | if (s) | |
127 | kmem_cache_free(s, p[i]); | |
128 | else | |
129 | kfree(p[i]); | |
130 | } | |
484748f0 CL |
131 | } |
132 | ||
865762a8 | 133 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
134 | void **p) |
135 | { | |
136 | size_t i; | |
137 | ||
138 | for (i = 0; i < nr; i++) { | |
139 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
140 | if (!x) { | |
141 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 142 | return 0; |
484748f0 CL |
143 | } |
144 | } | |
865762a8 | 145 | return i; |
484748f0 CL |
146 | } |
147 | ||
127424c8 | 148 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
510ded33 TH |
149 | |
150 | LIST_HEAD(slab_root_caches); | |
151 | ||
f7ce3190 | 152 | void slab_init_memcg_params(struct kmem_cache *s) |
33a690c4 | 153 | { |
9eeadc8b | 154 | s->memcg_params.root_cache = NULL; |
f7ce3190 | 155 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
9eeadc8b | 156 | INIT_LIST_HEAD(&s->memcg_params.children); |
f7ce3190 VD |
157 | } |
158 | ||
159 | static int init_memcg_params(struct kmem_cache *s, | |
160 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
161 | { | |
162 | struct memcg_cache_array *arr; | |
33a690c4 | 163 | |
9eeadc8b | 164 | if (root_cache) { |
f7ce3190 | 165 | s->memcg_params.root_cache = root_cache; |
9eeadc8b TH |
166 | s->memcg_params.memcg = memcg; |
167 | INIT_LIST_HEAD(&s->memcg_params.children_node); | |
bc2791f8 | 168 | INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node); |
33a690c4 | 169 | return 0; |
f7ce3190 | 170 | } |
33a690c4 | 171 | |
f7ce3190 | 172 | slab_init_memcg_params(s); |
33a690c4 | 173 | |
f7ce3190 VD |
174 | if (!memcg_nr_cache_ids) |
175 | return 0; | |
33a690c4 | 176 | |
f80c7dab JW |
177 | arr = kvzalloc(sizeof(struct memcg_cache_array) + |
178 | memcg_nr_cache_ids * sizeof(void *), | |
179 | GFP_KERNEL); | |
f7ce3190 VD |
180 | if (!arr) |
181 | return -ENOMEM; | |
33a690c4 | 182 | |
f7ce3190 | 183 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); |
33a690c4 VD |
184 | return 0; |
185 | } | |
186 | ||
f7ce3190 | 187 | static void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 | 188 | { |
f7ce3190 | 189 | if (is_root_cache(s)) |
f80c7dab JW |
190 | kvfree(rcu_access_pointer(s->memcg_params.memcg_caches)); |
191 | } | |
192 | ||
193 | static void free_memcg_params(struct rcu_head *rcu) | |
194 | { | |
195 | struct memcg_cache_array *old; | |
196 | ||
197 | old = container_of(rcu, struct memcg_cache_array, rcu); | |
198 | kvfree(old); | |
33a690c4 VD |
199 | } |
200 | ||
f7ce3190 | 201 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
6f817f4c | 202 | { |
f7ce3190 | 203 | struct memcg_cache_array *old, *new; |
6f817f4c | 204 | |
f80c7dab JW |
205 | new = kvzalloc(sizeof(struct memcg_cache_array) + |
206 | new_array_size * sizeof(void *), GFP_KERNEL); | |
f7ce3190 | 207 | if (!new) |
6f817f4c VD |
208 | return -ENOMEM; |
209 | ||
f7ce3190 VD |
210 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
211 | lockdep_is_held(&slab_mutex)); | |
212 | if (old) | |
213 | memcpy(new->entries, old->entries, | |
214 | memcg_nr_cache_ids * sizeof(void *)); | |
6f817f4c | 215 | |
f7ce3190 VD |
216 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
217 | if (old) | |
f80c7dab | 218 | call_rcu(&old->rcu, free_memcg_params); |
6f817f4c VD |
219 | return 0; |
220 | } | |
221 | ||
55007d84 GC |
222 | int memcg_update_all_caches(int num_memcgs) |
223 | { | |
224 | struct kmem_cache *s; | |
225 | int ret = 0; | |
55007d84 | 226 | |
05257a1a | 227 | mutex_lock(&slab_mutex); |
510ded33 | 228 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
f7ce3190 | 229 | ret = update_memcg_params(s, num_memcgs); |
55007d84 | 230 | /* |
55007d84 GC |
231 | * Instead of freeing the memory, we'll just leave the caches |
232 | * up to this point in an updated state. | |
233 | */ | |
234 | if (ret) | |
05257a1a | 235 | break; |
55007d84 | 236 | } |
55007d84 GC |
237 | mutex_unlock(&slab_mutex); |
238 | return ret; | |
239 | } | |
657dc2f9 | 240 | |
510ded33 | 241 | void memcg_link_cache(struct kmem_cache *s) |
657dc2f9 | 242 | { |
510ded33 TH |
243 | if (is_root_cache(s)) { |
244 | list_add(&s->root_caches_node, &slab_root_caches); | |
245 | } else { | |
246 | list_add(&s->memcg_params.children_node, | |
247 | &s->memcg_params.root_cache->memcg_params.children); | |
248 | list_add(&s->memcg_params.kmem_caches_node, | |
249 | &s->memcg_params.memcg->kmem_caches); | |
250 | } | |
251 | } | |
252 | ||
253 | static void memcg_unlink_cache(struct kmem_cache *s) | |
254 | { | |
255 | if (is_root_cache(s)) { | |
256 | list_del(&s->root_caches_node); | |
257 | } else { | |
258 | list_del(&s->memcg_params.children_node); | |
259 | list_del(&s->memcg_params.kmem_caches_node); | |
260 | } | |
657dc2f9 | 261 | } |
33a690c4 | 262 | #else |
f7ce3190 VD |
263 | static inline int init_memcg_params(struct kmem_cache *s, |
264 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) | |
33a690c4 VD |
265 | { |
266 | return 0; | |
267 | } | |
268 | ||
f7ce3190 | 269 | static inline void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 VD |
270 | { |
271 | } | |
657dc2f9 | 272 | |
510ded33 | 273 | static inline void memcg_unlink_cache(struct kmem_cache *s) |
657dc2f9 TH |
274 | { |
275 | } | |
127424c8 | 276 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
55007d84 | 277 | |
692ae74a BL |
278 | /* |
279 | * Figure out what the alignment of the objects will be given a set of | |
280 | * flags, a user specified alignment and the size of the objects. | |
281 | */ | |
282 | static unsigned long calculate_alignment(unsigned long flags, | |
283 | unsigned long align, unsigned long size) | |
284 | { | |
285 | /* | |
286 | * If the user wants hardware cache aligned objects then follow that | |
287 | * suggestion if the object is sufficiently large. | |
288 | * | |
289 | * The hardware cache alignment cannot override the specified | |
290 | * alignment though. If that is greater then use it. | |
291 | */ | |
292 | if (flags & SLAB_HWCACHE_ALIGN) { | |
293 | unsigned long ralign; | |
294 | ||
295 | ralign = cache_line_size(); | |
296 | while (size <= ralign / 2) | |
297 | ralign /= 2; | |
298 | align = max(align, ralign); | |
299 | } | |
300 | ||
301 | if (align < ARCH_SLAB_MINALIGN) | |
302 | align = ARCH_SLAB_MINALIGN; | |
303 | ||
304 | return ALIGN(align, sizeof(void *)); | |
305 | } | |
306 | ||
423c929c JK |
307 | /* |
308 | * Find a mergeable slab cache | |
309 | */ | |
310 | int slab_unmergeable(struct kmem_cache *s) | |
311 | { | |
312 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
313 | return 1; | |
314 | ||
315 | if (!is_root_cache(s)) | |
316 | return 1; | |
317 | ||
318 | if (s->ctor) | |
319 | return 1; | |
320 | ||
8eb8284b DW |
321 | if (s->usersize) |
322 | return 1; | |
323 | ||
423c929c JK |
324 | /* |
325 | * We may have set a slab to be unmergeable during bootstrap. | |
326 | */ | |
327 | if (s->refcount < 0) | |
328 | return 1; | |
329 | ||
330 | return 0; | |
331 | } | |
332 | ||
333 | struct kmem_cache *find_mergeable(size_t size, size_t align, | |
d50112ed | 334 | slab_flags_t flags, const char *name, void (*ctor)(void *)) |
423c929c JK |
335 | { |
336 | struct kmem_cache *s; | |
337 | ||
c6e28895 | 338 | if (slab_nomerge) |
423c929c JK |
339 | return NULL; |
340 | ||
341 | if (ctor) | |
342 | return NULL; | |
343 | ||
344 | size = ALIGN(size, sizeof(void *)); | |
345 | align = calculate_alignment(flags, align, size); | |
346 | size = ALIGN(size, align); | |
347 | flags = kmem_cache_flags(size, flags, name, NULL); | |
348 | ||
c6e28895 GM |
349 | if (flags & SLAB_NEVER_MERGE) |
350 | return NULL; | |
351 | ||
510ded33 | 352 | list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) { |
423c929c JK |
353 | if (slab_unmergeable(s)) |
354 | continue; | |
355 | ||
356 | if (size > s->size) | |
357 | continue; | |
358 | ||
359 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
360 | continue; | |
361 | /* | |
362 | * Check if alignment is compatible. | |
363 | * Courtesy of Adrian Drzewiecki | |
364 | */ | |
365 | if ((s->size & ~(align - 1)) != s->size) | |
366 | continue; | |
367 | ||
368 | if (s->size - size >= sizeof(void *)) | |
369 | continue; | |
370 | ||
95069ac8 JK |
371 | if (IS_ENABLED(CONFIG_SLAB) && align && |
372 | (align > s->align || s->align % align)) | |
373 | continue; | |
374 | ||
423c929c JK |
375 | return s; |
376 | } | |
377 | return NULL; | |
378 | } | |
379 | ||
c9a77a79 VD |
380 | static struct kmem_cache *create_cache(const char *name, |
381 | size_t object_size, size_t size, size_t align, | |
8eb8284b DW |
382 | slab_flags_t flags, size_t useroffset, |
383 | size_t usersize, void (*ctor)(void *), | |
c9a77a79 | 384 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
794b1248 VD |
385 | { |
386 | struct kmem_cache *s; | |
387 | int err; | |
388 | ||
8eb8284b DW |
389 | if (WARN_ON(useroffset + usersize > object_size)) |
390 | useroffset = usersize = 0; | |
391 | ||
794b1248 VD |
392 | err = -ENOMEM; |
393 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
394 | if (!s) | |
395 | goto out; | |
396 | ||
397 | s->name = name; | |
398 | s->object_size = object_size; | |
399 | s->size = size; | |
400 | s->align = align; | |
401 | s->ctor = ctor; | |
8eb8284b DW |
402 | s->useroffset = useroffset; |
403 | s->usersize = usersize; | |
794b1248 | 404 | |
f7ce3190 | 405 | err = init_memcg_params(s, memcg, root_cache); |
794b1248 VD |
406 | if (err) |
407 | goto out_free_cache; | |
408 | ||
409 | err = __kmem_cache_create(s, flags); | |
410 | if (err) | |
411 | goto out_free_cache; | |
412 | ||
413 | s->refcount = 1; | |
414 | list_add(&s->list, &slab_caches); | |
510ded33 | 415 | memcg_link_cache(s); |
794b1248 VD |
416 | out: |
417 | if (err) | |
418 | return ERR_PTR(err); | |
419 | return s; | |
420 | ||
421 | out_free_cache: | |
f7ce3190 | 422 | destroy_memcg_params(s); |
7c4da061 | 423 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
424 | goto out; |
425 | } | |
45906855 | 426 | |
77be4b13 | 427 | /* |
8eb8284b | 428 | * kmem_cache_create_usercopy - Create a cache. |
77be4b13 SK |
429 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
430 | * @size: The size of objects to be created in this cache. | |
431 | * @align: The required alignment for the objects. | |
432 | * @flags: SLAB flags | |
8eb8284b DW |
433 | * @useroffset: Usercopy region offset |
434 | * @usersize: Usercopy region size | |
77be4b13 SK |
435 | * @ctor: A constructor for the objects. |
436 | * | |
437 | * Returns a ptr to the cache on success, NULL on failure. | |
438 | * Cannot be called within a interrupt, but can be interrupted. | |
439 | * The @ctor is run when new pages are allocated by the cache. | |
440 | * | |
441 | * The flags are | |
442 | * | |
443 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
444 | * to catch references to uninitialised memory. | |
445 | * | |
446 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
447 | * for buffer overruns. | |
448 | * | |
449 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
450 | * cacheline. This can be beneficial if you're counting cycles as closely | |
451 | * as davem. | |
452 | */ | |
2633d7a0 | 453 | struct kmem_cache * |
8eb8284b DW |
454 | kmem_cache_create_usercopy(const char *name, size_t size, size_t align, |
455 | slab_flags_t flags, size_t useroffset, size_t usersize, | |
456 | void (*ctor)(void *)) | |
77be4b13 | 457 | { |
40911a79 | 458 | struct kmem_cache *s = NULL; |
3dec16ea | 459 | const char *cache_name; |
3965fc36 | 460 | int err; |
039363f3 | 461 | |
77be4b13 | 462 | get_online_cpus(); |
03afc0e2 | 463 | get_online_mems(); |
05257a1a | 464 | memcg_get_cache_ids(); |
03afc0e2 | 465 | |
77be4b13 | 466 | mutex_lock(&slab_mutex); |
686d550d | 467 | |
794b1248 | 468 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 469 | if (err) { |
3965fc36 | 470 | goto out_unlock; |
3aa24f51 | 471 | } |
686d550d | 472 | |
e70954fd TG |
473 | /* Refuse requests with allocator specific flags */ |
474 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
475 | err = -EINVAL; | |
476 | goto out_unlock; | |
477 | } | |
478 | ||
d8843922 GC |
479 | /* |
480 | * Some allocators will constraint the set of valid flags to a subset | |
481 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
482 | * case, and we'll just provide them with a sanitized version of the | |
483 | * passed flags. | |
484 | */ | |
485 | flags &= CACHE_CREATE_MASK; | |
686d550d | 486 | |
8eb8284b DW |
487 | /* Fail closed on bad usersize of useroffset values. */ |
488 | if (WARN_ON(!usersize && useroffset) || | |
489 | WARN_ON(size < usersize || size - usersize < useroffset)) | |
490 | usersize = useroffset = 0; | |
491 | ||
492 | if (!usersize) | |
493 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
794b1248 | 494 | if (s) |
3965fc36 | 495 | goto out_unlock; |
2633d7a0 | 496 | |
3dec16ea | 497 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
498 | if (!cache_name) { |
499 | err = -ENOMEM; | |
500 | goto out_unlock; | |
501 | } | |
7c9adf5a | 502 | |
c9a77a79 VD |
503 | s = create_cache(cache_name, size, size, |
504 | calculate_alignment(flags, align, size), | |
8eb8284b | 505 | flags, useroffset, usersize, ctor, NULL, NULL); |
794b1248 VD |
506 | if (IS_ERR(s)) { |
507 | err = PTR_ERR(s); | |
3dec16ea | 508 | kfree_const(cache_name); |
794b1248 | 509 | } |
3965fc36 VD |
510 | |
511 | out_unlock: | |
20cea968 | 512 | mutex_unlock(&slab_mutex); |
03afc0e2 | 513 | |
05257a1a | 514 | memcg_put_cache_ids(); |
03afc0e2 | 515 | put_online_mems(); |
20cea968 CL |
516 | put_online_cpus(); |
517 | ||
ba3253c7 | 518 | if (err) { |
686d550d CL |
519 | if (flags & SLAB_PANIC) |
520 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
521 | name, err); | |
522 | else { | |
1170532b | 523 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
524 | name, err); |
525 | dump_stack(); | |
526 | } | |
686d550d CL |
527 | return NULL; |
528 | } | |
039363f3 CL |
529 | return s; |
530 | } | |
8eb8284b DW |
531 | EXPORT_SYMBOL(kmem_cache_create_usercopy); |
532 | ||
533 | struct kmem_cache * | |
534 | kmem_cache_create(const char *name, size_t size, size_t align, | |
535 | slab_flags_t flags, void (*ctor)(void *)) | |
536 | { | |
6d07d1cd | 537 | return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, |
8eb8284b DW |
538 | ctor); |
539 | } | |
794b1248 | 540 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 541 | |
657dc2f9 | 542 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 543 | { |
657dc2f9 TH |
544 | LIST_HEAD(to_destroy); |
545 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 546 | |
657dc2f9 | 547 | /* |
5f0d5a3a | 548 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
657dc2f9 TH |
549 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
550 | * through RCU and and the associated kmem_cache are dereferenced | |
551 | * while freeing the pages, so the kmem_caches should be freed only | |
552 | * after the pending RCU operations are finished. As rcu_barrier() | |
553 | * is a pretty slow operation, we batch all pending destructions | |
554 | * asynchronously. | |
555 | */ | |
556 | mutex_lock(&slab_mutex); | |
557 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
558 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 559 | |
657dc2f9 TH |
560 | if (list_empty(&to_destroy)) |
561 | return; | |
562 | ||
563 | rcu_barrier(); | |
564 | ||
565 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
566 | #ifdef SLAB_SUPPORTS_SYSFS | |
567 | sysfs_slab_release(s); | |
568 | #else | |
569 | slab_kmem_cache_release(s); | |
570 | #endif | |
571 | } | |
d5b3cf71 VD |
572 | } |
573 | ||
657dc2f9 | 574 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 575 | { |
f9fa1d91 GT |
576 | /* free asan quarantined objects */ |
577 | kasan_cache_shutdown(s); | |
578 | ||
657dc2f9 TH |
579 | if (__kmem_cache_shutdown(s) != 0) |
580 | return -EBUSY; | |
d5b3cf71 | 581 | |
510ded33 | 582 | memcg_unlink_cache(s); |
657dc2f9 | 583 | list_del(&s->list); |
d5b3cf71 | 584 | |
5f0d5a3a | 585 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
657dc2f9 TH |
586 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
587 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
588 | } else { | |
d5b3cf71 | 589 | #ifdef SLAB_SUPPORTS_SYSFS |
bf5eb3de | 590 | sysfs_slab_release(s); |
d5b3cf71 VD |
591 | #else |
592 | slab_kmem_cache_release(s); | |
593 | #endif | |
594 | } | |
657dc2f9 TH |
595 | |
596 | return 0; | |
d5b3cf71 VD |
597 | } |
598 | ||
127424c8 | 599 | #if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB) |
794b1248 | 600 | /* |
776ed0f0 | 601 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
602 | * @memcg: The memory cgroup the new cache is for. |
603 | * @root_cache: The parent of the new cache. | |
604 | * | |
605 | * This function attempts to create a kmem cache that will serve allocation | |
606 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
607 | * from its parent. | |
608 | */ | |
d5b3cf71 VD |
609 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
610 | struct kmem_cache *root_cache) | |
2633d7a0 | 611 | { |
3e0350a3 | 612 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
33398cf2 | 613 | struct cgroup_subsys_state *css = &memcg->css; |
f7ce3190 | 614 | struct memcg_cache_array *arr; |
bd673145 | 615 | struct kmem_cache *s = NULL; |
794b1248 | 616 | char *cache_name; |
f7ce3190 | 617 | int idx; |
794b1248 VD |
618 | |
619 | get_online_cpus(); | |
03afc0e2 VD |
620 | get_online_mems(); |
621 | ||
794b1248 VD |
622 | mutex_lock(&slab_mutex); |
623 | ||
2a4db7eb | 624 | /* |
567e9ab2 | 625 | * The memory cgroup could have been offlined while the cache |
2a4db7eb VD |
626 | * creation work was pending. |
627 | */ | |
b6ecd2de | 628 | if (memcg->kmem_state != KMEM_ONLINE) |
2a4db7eb VD |
629 | goto out_unlock; |
630 | ||
f7ce3190 VD |
631 | idx = memcg_cache_id(memcg); |
632 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
633 | lockdep_is_held(&slab_mutex)); | |
634 | ||
d5b3cf71 VD |
635 | /* |
636 | * Since per-memcg caches are created asynchronously on first | |
637 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
638 | * create the same cache, but only one of them may succeed. | |
639 | */ | |
f7ce3190 | 640 | if (arr->entries[idx]) |
d5b3cf71 VD |
641 | goto out_unlock; |
642 | ||
f1008365 | 643 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
73f576c0 JW |
644 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, |
645 | css->serial_nr, memcg_name_buf); | |
794b1248 VD |
646 | if (!cache_name) |
647 | goto out_unlock; | |
648 | ||
c9a77a79 VD |
649 | s = create_cache(cache_name, root_cache->object_size, |
650 | root_cache->size, root_cache->align, | |
f773e36d | 651 | root_cache->flags & CACHE_CREATE_MASK, |
8eb8284b | 652 | root_cache->useroffset, root_cache->usersize, |
f773e36d | 653 | root_cache->ctor, memcg, root_cache); |
d5b3cf71 VD |
654 | /* |
655 | * If we could not create a memcg cache, do not complain, because | |
656 | * that's not critical at all as we can always proceed with the root | |
657 | * cache. | |
658 | */ | |
bd673145 | 659 | if (IS_ERR(s)) { |
794b1248 | 660 | kfree(cache_name); |
d5b3cf71 | 661 | goto out_unlock; |
bd673145 | 662 | } |
794b1248 | 663 | |
d5b3cf71 VD |
664 | /* |
665 | * Since readers won't lock (see cache_from_memcg_idx()), we need a | |
666 | * barrier here to ensure nobody will see the kmem_cache partially | |
667 | * initialized. | |
668 | */ | |
669 | smp_wmb(); | |
f7ce3190 | 670 | arr->entries[idx] = s; |
d5b3cf71 | 671 | |
794b1248 VD |
672 | out_unlock: |
673 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
674 | |
675 | put_online_mems(); | |
794b1248 | 676 | put_online_cpus(); |
2633d7a0 | 677 | } |
b8529907 | 678 | |
01fb58bc TH |
679 | static void kmemcg_deactivate_workfn(struct work_struct *work) |
680 | { | |
681 | struct kmem_cache *s = container_of(work, struct kmem_cache, | |
682 | memcg_params.deact_work); | |
683 | ||
684 | get_online_cpus(); | |
685 | get_online_mems(); | |
686 | ||
687 | mutex_lock(&slab_mutex); | |
688 | ||
689 | s->memcg_params.deact_fn(s); | |
690 | ||
691 | mutex_unlock(&slab_mutex); | |
692 | ||
693 | put_online_mems(); | |
694 | put_online_cpus(); | |
695 | ||
696 | /* done, put the ref from slab_deactivate_memcg_cache_rcu_sched() */ | |
697 | css_put(&s->memcg_params.memcg->css); | |
698 | } | |
699 | ||
700 | static void kmemcg_deactivate_rcufn(struct rcu_head *head) | |
701 | { | |
702 | struct kmem_cache *s = container_of(head, struct kmem_cache, | |
703 | memcg_params.deact_rcu_head); | |
704 | ||
705 | /* | |
706 | * We need to grab blocking locks. Bounce to ->deact_work. The | |
707 | * work item shares the space with the RCU head and can't be | |
708 | * initialized eariler. | |
709 | */ | |
710 | INIT_WORK(&s->memcg_params.deact_work, kmemcg_deactivate_workfn); | |
17cc4dfe | 711 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.deact_work); |
01fb58bc TH |
712 | } |
713 | ||
714 | /** | |
715 | * slab_deactivate_memcg_cache_rcu_sched - schedule deactivation after a | |
716 | * sched RCU grace period | |
717 | * @s: target kmem_cache | |
718 | * @deact_fn: deactivation function to call | |
719 | * | |
720 | * Schedule @deact_fn to be invoked with online cpus, mems and slab_mutex | |
721 | * held after a sched RCU grace period. The slab is guaranteed to stay | |
722 | * alive until @deact_fn is finished. This is to be used from | |
723 | * __kmemcg_cache_deactivate(). | |
724 | */ | |
725 | void slab_deactivate_memcg_cache_rcu_sched(struct kmem_cache *s, | |
726 | void (*deact_fn)(struct kmem_cache *)) | |
727 | { | |
728 | if (WARN_ON_ONCE(is_root_cache(s)) || | |
729 | WARN_ON_ONCE(s->memcg_params.deact_fn)) | |
730 | return; | |
731 | ||
732 | /* pin memcg so that @s doesn't get destroyed in the middle */ | |
733 | css_get(&s->memcg_params.memcg->css); | |
734 | ||
735 | s->memcg_params.deact_fn = deact_fn; | |
736 | call_rcu_sched(&s->memcg_params.deact_rcu_head, kmemcg_deactivate_rcufn); | |
737 | } | |
738 | ||
2a4db7eb VD |
739 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg) |
740 | { | |
741 | int idx; | |
742 | struct memcg_cache_array *arr; | |
d6e0b7fa | 743 | struct kmem_cache *s, *c; |
2a4db7eb VD |
744 | |
745 | idx = memcg_cache_id(memcg); | |
746 | ||
d6e0b7fa VD |
747 | get_online_cpus(); |
748 | get_online_mems(); | |
749 | ||
2a4db7eb | 750 | mutex_lock(&slab_mutex); |
510ded33 | 751 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
2a4db7eb VD |
752 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, |
753 | lockdep_is_held(&slab_mutex)); | |
d6e0b7fa VD |
754 | c = arr->entries[idx]; |
755 | if (!c) | |
756 | continue; | |
757 | ||
c9fc5864 | 758 | __kmemcg_cache_deactivate(c); |
2a4db7eb VD |
759 | arr->entries[idx] = NULL; |
760 | } | |
761 | mutex_unlock(&slab_mutex); | |
d6e0b7fa VD |
762 | |
763 | put_online_mems(); | |
764 | put_online_cpus(); | |
2a4db7eb VD |
765 | } |
766 | ||
d5b3cf71 | 767 | void memcg_destroy_kmem_caches(struct mem_cgroup *memcg) |
b8529907 | 768 | { |
d5b3cf71 | 769 | struct kmem_cache *s, *s2; |
b8529907 | 770 | |
d5b3cf71 VD |
771 | get_online_cpus(); |
772 | get_online_mems(); | |
b8529907 | 773 | |
b8529907 | 774 | mutex_lock(&slab_mutex); |
bc2791f8 TH |
775 | list_for_each_entry_safe(s, s2, &memcg->kmem_caches, |
776 | memcg_params.kmem_caches_node) { | |
d5b3cf71 VD |
777 | /* |
778 | * The cgroup is about to be freed and therefore has no charges | |
779 | * left. Hence, all its caches must be empty by now. | |
780 | */ | |
657dc2f9 | 781 | BUG_ON(shutdown_cache(s)); |
d5b3cf71 VD |
782 | } |
783 | mutex_unlock(&slab_mutex); | |
b8529907 | 784 | |
d5b3cf71 VD |
785 | put_online_mems(); |
786 | put_online_cpus(); | |
b8529907 | 787 | } |
d60fdcc9 | 788 | |
657dc2f9 | 789 | static int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
790 | { |
791 | struct memcg_cache_array *arr; | |
792 | struct kmem_cache *c, *c2; | |
793 | LIST_HEAD(busy); | |
794 | int i; | |
795 | ||
796 | BUG_ON(!is_root_cache(s)); | |
797 | ||
798 | /* | |
799 | * First, shutdown active caches, i.e. caches that belong to online | |
800 | * memory cgroups. | |
801 | */ | |
802 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
803 | lockdep_is_held(&slab_mutex)); | |
804 | for_each_memcg_cache_index(i) { | |
805 | c = arr->entries[i]; | |
806 | if (!c) | |
807 | continue; | |
657dc2f9 | 808 | if (shutdown_cache(c)) |
d60fdcc9 VD |
809 | /* |
810 | * The cache still has objects. Move it to a temporary | |
811 | * list so as not to try to destroy it for a second | |
812 | * time while iterating over inactive caches below. | |
813 | */ | |
9eeadc8b | 814 | list_move(&c->memcg_params.children_node, &busy); |
d60fdcc9 VD |
815 | else |
816 | /* | |
817 | * The cache is empty and will be destroyed soon. Clear | |
818 | * the pointer to it in the memcg_caches array so that | |
819 | * it will never be accessed even if the root cache | |
820 | * stays alive. | |
821 | */ | |
822 | arr->entries[i] = NULL; | |
823 | } | |
824 | ||
825 | /* | |
826 | * Second, shutdown all caches left from memory cgroups that are now | |
827 | * offline. | |
828 | */ | |
9eeadc8b TH |
829 | list_for_each_entry_safe(c, c2, &s->memcg_params.children, |
830 | memcg_params.children_node) | |
657dc2f9 | 831 | shutdown_cache(c); |
d60fdcc9 | 832 | |
9eeadc8b | 833 | list_splice(&busy, &s->memcg_params.children); |
d60fdcc9 VD |
834 | |
835 | /* | |
836 | * A cache being destroyed must be empty. In particular, this means | |
837 | * that all per memcg caches attached to it must be empty too. | |
838 | */ | |
9eeadc8b | 839 | if (!list_empty(&s->memcg_params.children)) |
d60fdcc9 VD |
840 | return -EBUSY; |
841 | return 0; | |
842 | } | |
843 | #else | |
657dc2f9 | 844 | static inline int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
845 | { |
846 | return 0; | |
847 | } | |
127424c8 | 848 | #endif /* CONFIG_MEMCG && !CONFIG_SLOB */ |
97d06609 | 849 | |
41a21285 CL |
850 | void slab_kmem_cache_release(struct kmem_cache *s) |
851 | { | |
52b4b950 | 852 | __kmem_cache_release(s); |
f7ce3190 | 853 | destroy_memcg_params(s); |
3dec16ea | 854 | kfree_const(s->name); |
41a21285 CL |
855 | kmem_cache_free(kmem_cache, s); |
856 | } | |
857 | ||
945cf2b6 CL |
858 | void kmem_cache_destroy(struct kmem_cache *s) |
859 | { | |
d60fdcc9 | 860 | int err; |
d5b3cf71 | 861 | |
3942d299 SS |
862 | if (unlikely(!s)) |
863 | return; | |
864 | ||
945cf2b6 | 865 | get_online_cpus(); |
03afc0e2 VD |
866 | get_online_mems(); |
867 | ||
945cf2b6 | 868 | mutex_lock(&slab_mutex); |
b8529907 | 869 | |
945cf2b6 | 870 | s->refcount--; |
b8529907 VD |
871 | if (s->refcount) |
872 | goto out_unlock; | |
873 | ||
657dc2f9 | 874 | err = shutdown_memcg_caches(s); |
d60fdcc9 | 875 | if (!err) |
657dc2f9 | 876 | err = shutdown_cache(s); |
b8529907 | 877 | |
cd918c55 | 878 | if (err) { |
756a025f JP |
879 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
880 | s->name); | |
cd918c55 VD |
881 | dump_stack(); |
882 | } | |
b8529907 VD |
883 | out_unlock: |
884 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 885 | |
03afc0e2 | 886 | put_online_mems(); |
945cf2b6 CL |
887 | put_online_cpus(); |
888 | } | |
889 | EXPORT_SYMBOL(kmem_cache_destroy); | |
890 | ||
03afc0e2 VD |
891 | /** |
892 | * kmem_cache_shrink - Shrink a cache. | |
893 | * @cachep: The cache to shrink. | |
894 | * | |
895 | * Releases as many slabs as possible for a cache. | |
896 | * To help debugging, a zero exit status indicates all slabs were released. | |
897 | */ | |
898 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
899 | { | |
900 | int ret; | |
901 | ||
902 | get_online_cpus(); | |
903 | get_online_mems(); | |
55834c59 | 904 | kasan_cache_shrink(cachep); |
c9fc5864 | 905 | ret = __kmem_cache_shrink(cachep); |
03afc0e2 VD |
906 | put_online_mems(); |
907 | put_online_cpus(); | |
908 | return ret; | |
909 | } | |
910 | EXPORT_SYMBOL(kmem_cache_shrink); | |
911 | ||
fda90124 | 912 | bool slab_is_available(void) |
97d06609 CL |
913 | { |
914 | return slab_state >= UP; | |
915 | } | |
b7454ad3 | 916 | |
45530c44 CL |
917 | #ifndef CONFIG_SLOB |
918 | /* Create a cache during boot when no slab services are available yet */ | |
919 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
8eb8284b | 920 | slab_flags_t flags, size_t useroffset, size_t usersize) |
45530c44 CL |
921 | { |
922 | int err; | |
923 | ||
924 | s->name = name; | |
925 | s->size = s->object_size = size; | |
45906855 | 926 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
8eb8284b DW |
927 | s->useroffset = useroffset; |
928 | s->usersize = usersize; | |
f7ce3190 VD |
929 | |
930 | slab_init_memcg_params(s); | |
931 | ||
45530c44 CL |
932 | err = __kmem_cache_create(s, flags); |
933 | ||
934 | if (err) | |
31ba7346 | 935 | panic("Creation of kmalloc slab %s size=%zu failed. Reason %d\n", |
45530c44 CL |
936 | name, size, err); |
937 | ||
938 | s->refcount = -1; /* Exempt from merging for now */ | |
939 | } | |
940 | ||
941 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
6c0c21ad DW |
942 | slab_flags_t flags, size_t useroffset, |
943 | size_t usersize) | |
45530c44 CL |
944 | { |
945 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
946 | ||
947 | if (!s) | |
948 | panic("Out of memory when creating slab %s\n", name); | |
949 | ||
6c0c21ad | 950 | create_boot_cache(s, name, size, flags, useroffset, usersize); |
45530c44 | 951 | list_add(&s->list, &slab_caches); |
510ded33 | 952 | memcg_link_cache(s); |
45530c44 CL |
953 | s->refcount = 1; |
954 | return s; | |
955 | } | |
956 | ||
9425c58e CL |
957 | struct kmem_cache *kmalloc_caches[KMALLOC_SHIFT_HIGH + 1]; |
958 | EXPORT_SYMBOL(kmalloc_caches); | |
959 | ||
960 | #ifdef CONFIG_ZONE_DMA | |
961 | struct kmem_cache *kmalloc_dma_caches[KMALLOC_SHIFT_HIGH + 1]; | |
962 | EXPORT_SYMBOL(kmalloc_dma_caches); | |
963 | #endif | |
964 | ||
2c59dd65 CL |
965 | /* |
966 | * Conversion table for small slabs sizes / 8 to the index in the | |
967 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
968 | * of two cache sizes there. The size of larger slabs can be determined using | |
969 | * fls. | |
970 | */ | |
971 | static s8 size_index[24] = { | |
972 | 3, /* 8 */ | |
973 | 4, /* 16 */ | |
974 | 5, /* 24 */ | |
975 | 5, /* 32 */ | |
976 | 6, /* 40 */ | |
977 | 6, /* 48 */ | |
978 | 6, /* 56 */ | |
979 | 6, /* 64 */ | |
980 | 1, /* 72 */ | |
981 | 1, /* 80 */ | |
982 | 1, /* 88 */ | |
983 | 1, /* 96 */ | |
984 | 7, /* 104 */ | |
985 | 7, /* 112 */ | |
986 | 7, /* 120 */ | |
987 | 7, /* 128 */ | |
988 | 2, /* 136 */ | |
989 | 2, /* 144 */ | |
990 | 2, /* 152 */ | |
991 | 2, /* 160 */ | |
992 | 2, /* 168 */ | |
993 | 2, /* 176 */ | |
994 | 2, /* 184 */ | |
995 | 2 /* 192 */ | |
996 | }; | |
997 | ||
998 | static inline int size_index_elem(size_t bytes) | |
999 | { | |
1000 | return (bytes - 1) / 8; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Find the kmem_cache structure that serves a given size of | |
1005 | * allocation | |
1006 | */ | |
1007 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
1008 | { | |
1009 | int index; | |
1010 | ||
9de1bc87 | 1011 | if (unlikely(size > KMALLOC_MAX_SIZE)) { |
907985f4 | 1012 | WARN_ON_ONCE(!(flags & __GFP_NOWARN)); |
6286ae97 | 1013 | return NULL; |
907985f4 | 1014 | } |
6286ae97 | 1015 | |
2c59dd65 CL |
1016 | if (size <= 192) { |
1017 | if (!size) | |
1018 | return ZERO_SIZE_PTR; | |
1019 | ||
1020 | index = size_index[size_index_elem(size)]; | |
1021 | } else | |
1022 | index = fls(size - 1); | |
1023 | ||
1024 | #ifdef CONFIG_ZONE_DMA | |
b1e05416 | 1025 | if (unlikely((flags & GFP_DMA))) |
2c59dd65 CL |
1026 | return kmalloc_dma_caches[index]; |
1027 | ||
1028 | #endif | |
1029 | return kmalloc_caches[index]; | |
1030 | } | |
1031 | ||
4066c33d GG |
1032 | /* |
1033 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
1034 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
1035 | * kmalloc-67108864. | |
1036 | */ | |
af3b5f87 | 1037 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
4066c33d GG |
1038 | {NULL, 0}, {"kmalloc-96", 96}, |
1039 | {"kmalloc-192", 192}, {"kmalloc-8", 8}, | |
1040 | {"kmalloc-16", 16}, {"kmalloc-32", 32}, | |
1041 | {"kmalloc-64", 64}, {"kmalloc-128", 128}, | |
1042 | {"kmalloc-256", 256}, {"kmalloc-512", 512}, | |
1043 | {"kmalloc-1024", 1024}, {"kmalloc-2048", 2048}, | |
1044 | {"kmalloc-4096", 4096}, {"kmalloc-8192", 8192}, | |
1045 | {"kmalloc-16384", 16384}, {"kmalloc-32768", 32768}, | |
1046 | {"kmalloc-65536", 65536}, {"kmalloc-131072", 131072}, | |
1047 | {"kmalloc-262144", 262144}, {"kmalloc-524288", 524288}, | |
1048 | {"kmalloc-1048576", 1048576}, {"kmalloc-2097152", 2097152}, | |
1049 | {"kmalloc-4194304", 4194304}, {"kmalloc-8388608", 8388608}, | |
1050 | {"kmalloc-16777216", 16777216}, {"kmalloc-33554432", 33554432}, | |
1051 | {"kmalloc-67108864", 67108864} | |
1052 | }; | |
1053 | ||
f97d5f63 | 1054 | /* |
34cc6990 DS |
1055 | * Patch up the size_index table if we have strange large alignment |
1056 | * requirements for the kmalloc array. This is only the case for | |
1057 | * MIPS it seems. The standard arches will not generate any code here. | |
1058 | * | |
1059 | * Largest permitted alignment is 256 bytes due to the way we | |
1060 | * handle the index determination for the smaller caches. | |
1061 | * | |
1062 | * Make sure that nothing crazy happens if someone starts tinkering | |
1063 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 1064 | */ |
34cc6990 | 1065 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 CL |
1066 | { |
1067 | int i; | |
1068 | ||
2c59dd65 CL |
1069 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
1070 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
1071 | ||
1072 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
1073 | int elem = size_index_elem(i); | |
1074 | ||
1075 | if (elem >= ARRAY_SIZE(size_index)) | |
1076 | break; | |
1077 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
1078 | } | |
1079 | ||
1080 | if (KMALLOC_MIN_SIZE >= 64) { | |
1081 | /* | |
1082 | * The 96 byte size cache is not used if the alignment | |
1083 | * is 64 byte. | |
1084 | */ | |
1085 | for (i = 64 + 8; i <= 96; i += 8) | |
1086 | size_index[size_index_elem(i)] = 7; | |
1087 | ||
1088 | } | |
1089 | ||
1090 | if (KMALLOC_MIN_SIZE >= 128) { | |
1091 | /* | |
1092 | * The 192 byte sized cache is not used if the alignment | |
1093 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
1094 | * instead. | |
1095 | */ | |
1096 | for (i = 128 + 8; i <= 192; i += 8) | |
1097 | size_index[size_index_elem(i)] = 8; | |
1098 | } | |
34cc6990 DS |
1099 | } |
1100 | ||
d50112ed | 1101 | static void __init new_kmalloc_cache(int idx, slab_flags_t flags) |
a9730fca CL |
1102 | { |
1103 | kmalloc_caches[idx] = create_kmalloc_cache(kmalloc_info[idx].name, | |
6c0c21ad DW |
1104 | kmalloc_info[idx].size, flags, 0, |
1105 | kmalloc_info[idx].size); | |
a9730fca CL |
1106 | } |
1107 | ||
34cc6990 DS |
1108 | /* |
1109 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
1110 | * may already have been created because they were needed to | |
1111 | * enable allocations for slab creation. | |
1112 | */ | |
d50112ed | 1113 | void __init create_kmalloc_caches(slab_flags_t flags) |
34cc6990 DS |
1114 | { |
1115 | int i; | |
1116 | ||
a9730fca CL |
1117 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { |
1118 | if (!kmalloc_caches[i]) | |
1119 | new_kmalloc_cache(i, flags); | |
f97d5f63 | 1120 | |
956e46ef | 1121 | /* |
a9730fca CL |
1122 | * Caches that are not of the two-to-the-power-of size. |
1123 | * These have to be created immediately after the | |
1124 | * earlier power of two caches | |
956e46ef | 1125 | */ |
a9730fca CL |
1126 | if (KMALLOC_MIN_SIZE <= 32 && !kmalloc_caches[1] && i == 6) |
1127 | new_kmalloc_cache(1, flags); | |
1128 | if (KMALLOC_MIN_SIZE <= 64 && !kmalloc_caches[2] && i == 7) | |
1129 | new_kmalloc_cache(2, flags); | |
8a965b3b CL |
1130 | } |
1131 | ||
f97d5f63 CL |
1132 | /* Kmalloc array is now usable */ |
1133 | slab_state = UP; | |
1134 | ||
f97d5f63 CL |
1135 | #ifdef CONFIG_ZONE_DMA |
1136 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1137 | struct kmem_cache *s = kmalloc_caches[i]; | |
1138 | ||
1139 | if (s) { | |
1140 | int size = kmalloc_size(i); | |
1141 | char *n = kasprintf(GFP_NOWAIT, | |
1142 | "dma-kmalloc-%d", size); | |
1143 | ||
1144 | BUG_ON(!n); | |
1145 | kmalloc_dma_caches[i] = create_kmalloc_cache(n, | |
6c0c21ad | 1146 | size, SLAB_CACHE_DMA | flags, 0, 0); |
f97d5f63 CL |
1147 | } |
1148 | } | |
1149 | #endif | |
1150 | } | |
45530c44 CL |
1151 | #endif /* !CONFIG_SLOB */ |
1152 | ||
cea371f4 VD |
1153 | /* |
1154 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1155 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1156 | * know the allocation order to free the pages properly in kfree. | |
1157 | */ | |
52383431 VD |
1158 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
1159 | { | |
1160 | void *ret; | |
1161 | struct page *page; | |
1162 | ||
1163 | flags |= __GFP_COMP; | |
4949148a | 1164 | page = alloc_pages(flags, order); |
52383431 VD |
1165 | ret = page ? page_address(page) : NULL; |
1166 | kmemleak_alloc(ret, size, 1, flags); | |
505f5dcb | 1167 | kasan_kmalloc_large(ret, size, flags); |
52383431 VD |
1168 | return ret; |
1169 | } | |
1170 | EXPORT_SYMBOL(kmalloc_order); | |
1171 | ||
f1b6eb6e CL |
1172 | #ifdef CONFIG_TRACING |
1173 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1174 | { | |
1175 | void *ret = kmalloc_order(size, flags, order); | |
1176 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1177 | return ret; | |
1178 | } | |
1179 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1180 | #endif | |
45530c44 | 1181 | |
7c00fce9 TG |
1182 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1183 | /* Randomize a generic freelist */ | |
1184 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
1185 | size_t count) | |
1186 | { | |
1187 | size_t i; | |
1188 | unsigned int rand; | |
1189 | ||
1190 | for (i = 0; i < count; i++) | |
1191 | list[i] = i; | |
1192 | ||
1193 | /* Fisher-Yates shuffle */ | |
1194 | for (i = count - 1; i > 0; i--) { | |
1195 | rand = prandom_u32_state(state); | |
1196 | rand %= (i + 1); | |
1197 | swap(list[i], list[rand]); | |
1198 | } | |
1199 | } | |
1200 | ||
1201 | /* Create a random sequence per cache */ | |
1202 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1203 | gfp_t gfp) | |
1204 | { | |
1205 | struct rnd_state state; | |
1206 | ||
1207 | if (count < 2 || cachep->random_seq) | |
1208 | return 0; | |
1209 | ||
1210 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1211 | if (!cachep->random_seq) | |
1212 | return -ENOMEM; | |
1213 | ||
1214 | /* Get best entropy at this stage of boot */ | |
1215 | prandom_seed_state(&state, get_random_long()); | |
1216 | ||
1217 | freelist_randomize(&state, cachep->random_seq, count); | |
1218 | return 0; | |
1219 | } | |
1220 | ||
1221 | /* Destroy the per-cache random freelist sequence */ | |
1222 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1223 | { | |
1224 | kfree(cachep->random_seq); | |
1225 | cachep->random_seq = NULL; | |
1226 | } | |
1227 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1228 | ||
5b365771 | 1229 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
e9b4db2b WL |
1230 | #ifdef CONFIG_SLAB |
1231 | #define SLABINFO_RIGHTS (S_IWUSR | S_IRUSR) | |
1232 | #else | |
1233 | #define SLABINFO_RIGHTS S_IRUSR | |
1234 | #endif | |
1235 | ||
b047501c | 1236 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1237 | { |
1238 | /* | |
1239 | * Output format version, so at least we can change it | |
1240 | * without _too_ many complaints. | |
1241 | */ | |
1242 | #ifdef CONFIG_DEBUG_SLAB | |
1243 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1244 | #else | |
1245 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1246 | #endif | |
756a025f | 1247 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1248 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1249 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1250 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1251 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1252 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1253 | #endif | |
1254 | seq_putc(m, '\n'); | |
1255 | } | |
1256 | ||
1df3b26f | 1257 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1258 | { |
b7454ad3 | 1259 | mutex_lock(&slab_mutex); |
510ded33 | 1260 | return seq_list_start(&slab_root_caches, *pos); |
b7454ad3 GC |
1261 | } |
1262 | ||
276a2439 | 1263 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 | 1264 | { |
510ded33 | 1265 | return seq_list_next(p, &slab_root_caches, pos); |
b7454ad3 GC |
1266 | } |
1267 | ||
276a2439 | 1268 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1269 | { |
1270 | mutex_unlock(&slab_mutex); | |
1271 | } | |
1272 | ||
749c5415 GC |
1273 | static void |
1274 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1275 | { | |
1276 | struct kmem_cache *c; | |
1277 | struct slabinfo sinfo; | |
749c5415 GC |
1278 | |
1279 | if (!is_root_cache(s)) | |
1280 | return; | |
1281 | ||
426589f5 | 1282 | for_each_memcg_cache(c, s) { |
749c5415 GC |
1283 | memset(&sinfo, 0, sizeof(sinfo)); |
1284 | get_slabinfo(c, &sinfo); | |
1285 | ||
1286 | info->active_slabs += sinfo.active_slabs; | |
1287 | info->num_slabs += sinfo.num_slabs; | |
1288 | info->shared_avail += sinfo.shared_avail; | |
1289 | info->active_objs += sinfo.active_objs; | |
1290 | info->num_objs += sinfo.num_objs; | |
1291 | } | |
1292 | } | |
1293 | ||
b047501c | 1294 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1295 | { |
0d7561c6 GC |
1296 | struct slabinfo sinfo; |
1297 | ||
1298 | memset(&sinfo, 0, sizeof(sinfo)); | |
1299 | get_slabinfo(s, &sinfo); | |
1300 | ||
749c5415 GC |
1301 | memcg_accumulate_slabinfo(s, &sinfo); |
1302 | ||
0d7561c6 | 1303 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 1304 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1305 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1306 | ||
1307 | seq_printf(m, " : tunables %4u %4u %4u", | |
1308 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1309 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1310 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1311 | slabinfo_show_stats(m, s); | |
1312 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1313 | } |
1314 | ||
1df3b26f | 1315 | static int slab_show(struct seq_file *m, void *p) |
749c5415 | 1316 | { |
510ded33 | 1317 | struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node); |
749c5415 | 1318 | |
510ded33 | 1319 | if (p == slab_root_caches.next) |
1df3b26f | 1320 | print_slabinfo_header(m); |
510ded33 | 1321 | cache_show(s, m); |
b047501c VD |
1322 | return 0; |
1323 | } | |
1324 | ||
852d8be0 YS |
1325 | void dump_unreclaimable_slab(void) |
1326 | { | |
1327 | struct kmem_cache *s, *s2; | |
1328 | struct slabinfo sinfo; | |
1329 | ||
1330 | /* | |
1331 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1332 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1333 | * risk of crash. | |
1334 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1335 | * without acquiring the mutex. | |
1336 | */ | |
1337 | if (!mutex_trylock(&slab_mutex)) { | |
1338 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1339 | return; | |
1340 | } | |
1341 | ||
1342 | pr_info("Unreclaimable slab info:\n"); | |
1343 | pr_info("Name Used Total\n"); | |
1344 | ||
1345 | list_for_each_entry_safe(s, s2, &slab_caches, list) { | |
1346 | if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT)) | |
1347 | continue; | |
1348 | ||
1349 | get_slabinfo(s, &sinfo); | |
1350 | ||
1351 | if (sinfo.num_objs > 0) | |
1352 | pr_info("%-17s %10luKB %10luKB\n", cache_name(s), | |
1353 | (sinfo.active_objs * s->size) / 1024, | |
1354 | (sinfo.num_objs * s->size) / 1024); | |
1355 | } | |
1356 | mutex_unlock(&slab_mutex); | |
1357 | } | |
1358 | ||
5b365771 | 1359 | #if defined(CONFIG_MEMCG) |
bc2791f8 TH |
1360 | void *memcg_slab_start(struct seq_file *m, loff_t *pos) |
1361 | { | |
1362 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1363 | ||
1364 | mutex_lock(&slab_mutex); | |
1365 | return seq_list_start(&memcg->kmem_caches, *pos); | |
1366 | } | |
1367 | ||
1368 | void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1369 | { | |
1370 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
1371 | ||
1372 | return seq_list_next(p, &memcg->kmem_caches, pos); | |
1373 | } | |
1374 | ||
1375 | void memcg_slab_stop(struct seq_file *m, void *p) | |
1376 | { | |
1377 | mutex_unlock(&slab_mutex); | |
1378 | } | |
1379 | ||
b047501c VD |
1380 | int memcg_slab_show(struct seq_file *m, void *p) |
1381 | { | |
bc2791f8 TH |
1382 | struct kmem_cache *s = list_entry(p, struct kmem_cache, |
1383 | memcg_params.kmem_caches_node); | |
b047501c VD |
1384 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
1385 | ||
bc2791f8 | 1386 | if (p == memcg->kmem_caches.next) |
b047501c | 1387 | print_slabinfo_header(m); |
bc2791f8 | 1388 | cache_show(s, m); |
b047501c | 1389 | return 0; |
749c5415 | 1390 | } |
b047501c | 1391 | #endif |
749c5415 | 1392 | |
b7454ad3 GC |
1393 | /* |
1394 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1395 | * | |
1396 | * Output layout: | |
1397 | * cache-name | |
1398 | * num-active-objs | |
1399 | * total-objs | |
1400 | * object size | |
1401 | * num-active-slabs | |
1402 | * total-slabs | |
1403 | * num-pages-per-slab | |
1404 | * + further values on SMP and with statistics enabled | |
1405 | */ | |
1406 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1407 | .start = slab_start, |
276a2439 WL |
1408 | .next = slab_next, |
1409 | .stop = slab_stop, | |
1df3b26f | 1410 | .show = slab_show, |
b7454ad3 GC |
1411 | }; |
1412 | ||
1413 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1414 | { | |
1415 | return seq_open(file, &slabinfo_op); | |
1416 | } | |
1417 | ||
1418 | static const struct file_operations proc_slabinfo_operations = { | |
1419 | .open = slabinfo_open, | |
1420 | .read = seq_read, | |
1421 | .write = slabinfo_write, | |
1422 | .llseek = seq_lseek, | |
1423 | .release = seq_release, | |
1424 | }; | |
1425 | ||
1426 | static int __init slab_proc_init(void) | |
1427 | { | |
e9b4db2b WL |
1428 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
1429 | &proc_slabinfo_operations); | |
b7454ad3 GC |
1430 | return 0; |
1431 | } | |
1432 | module_init(slab_proc_init); | |
5b365771 | 1433 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ |
928cec9c AR |
1434 | |
1435 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1436 | gfp_t flags) | |
1437 | { | |
1438 | void *ret; | |
1439 | size_t ks = 0; | |
1440 | ||
1441 | if (p) | |
1442 | ks = ksize(p); | |
1443 | ||
0316bec2 | 1444 | if (ks >= new_size) { |
505f5dcb | 1445 | kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1446 | return (void *)p; |
0316bec2 | 1447 | } |
928cec9c AR |
1448 | |
1449 | ret = kmalloc_track_caller(new_size, flags); | |
1450 | if (ret && p) | |
1451 | memcpy(ret, p, ks); | |
1452 | ||
1453 | return ret; | |
1454 | } | |
1455 | ||
1456 | /** | |
1457 | * __krealloc - like krealloc() but don't free @p. | |
1458 | * @p: object to reallocate memory for. | |
1459 | * @new_size: how many bytes of memory are required. | |
1460 | * @flags: the type of memory to allocate. | |
1461 | * | |
1462 | * This function is like krealloc() except it never frees the originally | |
1463 | * allocated buffer. Use this if you don't want to free the buffer immediately | |
1464 | * like, for example, with RCU. | |
1465 | */ | |
1466 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) | |
1467 | { | |
1468 | if (unlikely(!new_size)) | |
1469 | return ZERO_SIZE_PTR; | |
1470 | ||
1471 | return __do_krealloc(p, new_size, flags); | |
1472 | ||
1473 | } | |
1474 | EXPORT_SYMBOL(__krealloc); | |
1475 | ||
1476 | /** | |
1477 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1478 | * @p: object to reallocate memory for. | |
1479 | * @new_size: how many bytes of memory are required. | |
1480 | * @flags: the type of memory to allocate. | |
1481 | * | |
1482 | * The contents of the object pointed to are preserved up to the | |
1483 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1484 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1485 | * %NULL pointer, the object pointed to is freed. | |
1486 | */ | |
1487 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1488 | { | |
1489 | void *ret; | |
1490 | ||
1491 | if (unlikely(!new_size)) { | |
1492 | kfree(p); | |
1493 | return ZERO_SIZE_PTR; | |
1494 | } | |
1495 | ||
1496 | ret = __do_krealloc(p, new_size, flags); | |
1497 | if (ret && p != ret) | |
1498 | kfree(p); | |
1499 | ||
1500 | return ret; | |
1501 | } | |
1502 | EXPORT_SYMBOL(krealloc); | |
1503 | ||
1504 | /** | |
1505 | * kzfree - like kfree but zero memory | |
1506 | * @p: object to free memory of | |
1507 | * | |
1508 | * The memory of the object @p points to is zeroed before freed. | |
1509 | * If @p is %NULL, kzfree() does nothing. | |
1510 | * | |
1511 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1512 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1513 | * careful when using this function in performance sensitive code. | |
1514 | */ | |
1515 | void kzfree(const void *p) | |
1516 | { | |
1517 | size_t ks; | |
1518 | void *mem = (void *)p; | |
1519 | ||
1520 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1521 | return; | |
1522 | ks = ksize(mem); | |
1523 | memset(mem, 0, ks); | |
1524 | kfree(mem); | |
1525 | } | |
1526 | EXPORT_SYMBOL(kzfree); | |
1527 | ||
1528 | /* Tracepoints definitions. */ | |
1529 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1530 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1531 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1532 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1533 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1534 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); |