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> | |
1c99ba29 | 13 | #include <linux/cache.h> |
039363f3 CL |
14 | #include <linux/compiler.h> |
15 | #include <linux/module.h> | |
20cea968 CL |
16 | #include <linux/cpu.h> |
17 | #include <linux/uaccess.h> | |
b7454ad3 GC |
18 | #include <linux/seq_file.h> |
19 | #include <linux/proc_fs.h> | |
fcf8a1e4 | 20 | #include <linux/debugfs.h> |
039363f3 CL |
21 | #include <asm/cacheflush.h> |
22 | #include <asm/tlbflush.h> | |
23 | #include <asm/page.h> | |
2633d7a0 | 24 | #include <linux/memcontrol.h> |
928cec9c AR |
25 | |
26 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 27 | #include <trace/events/kmem.h> |
039363f3 | 28 | |
97d06609 CL |
29 | #include "slab.h" |
30 | ||
31 | enum slab_state slab_state; | |
18004c5d CL |
32 | LIST_HEAD(slab_caches); |
33 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 34 | struct kmem_cache *kmem_cache; |
97d06609 | 35 | |
2d891fbc KC |
36 | #ifdef CONFIG_HARDENED_USERCOPY |
37 | bool usercopy_fallback __ro_after_init = | |
38 | IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK); | |
39 | module_param(usercopy_fallback, bool, 0400); | |
40 | MODULE_PARM_DESC(usercopy_fallback, | |
41 | "WARN instead of reject usercopy whitelist violations"); | |
42 | #endif | |
43 | ||
657dc2f9 TH |
44 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
45 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
46 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
47 | slab_caches_to_rcu_destroy_workfn); | |
48 | ||
423c929c JK |
49 | /* |
50 | * Set of flags that will prevent slab merging | |
51 | */ | |
52 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
5f0d5a3a | 53 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ |
7ed2f9e6 | 54 | SLAB_FAILSLAB | SLAB_KASAN) |
423c929c | 55 | |
230e9fc2 | 56 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
6d6ea1e9 | 57 | SLAB_CACHE_DMA32 | SLAB_ACCOUNT) |
423c929c JK |
58 | |
59 | /* | |
60 | * Merge control. If this is set then no merging of slab caches will occur. | |
423c929c | 61 | */ |
7660a6fd | 62 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); |
423c929c JK |
63 | |
64 | static int __init setup_slab_nomerge(char *str) | |
65 | { | |
7660a6fd | 66 | slab_nomerge = true; |
423c929c JK |
67 | return 1; |
68 | } | |
69 | ||
70 | #ifdef CONFIG_SLUB | |
71 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
72 | #endif | |
73 | ||
74 | __setup("slab_nomerge", setup_slab_nomerge); | |
75 | ||
07f361b2 JK |
76 | /* |
77 | * Determine the size of a slab object | |
78 | */ | |
79 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
80 | { | |
81 | return s->object_size; | |
82 | } | |
83 | EXPORT_SYMBOL(kmem_cache_size); | |
84 | ||
77be4b13 | 85 | #ifdef CONFIG_DEBUG_VM |
f4957d5b | 86 | static int kmem_cache_sanity_check(const char *name, unsigned int size) |
039363f3 | 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 | 94 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
77be4b13 SK |
95 | return 0; |
96 | } | |
97 | #else | |
f4957d5b | 98 | static inline int kmem_cache_sanity_check(const char *name, unsigned int size) |
77be4b13 SK |
99 | { |
100 | return 0; | |
101 | } | |
20cea968 CL |
102 | #endif |
103 | ||
484748f0 CL |
104 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
105 | { | |
106 | size_t i; | |
107 | ||
ca257195 JDB |
108 | for (i = 0; i < nr; i++) { |
109 | if (s) | |
110 | kmem_cache_free(s, p[i]); | |
111 | else | |
112 | kfree(p[i]); | |
113 | } | |
484748f0 CL |
114 | } |
115 | ||
865762a8 | 116 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
117 | void **p) |
118 | { | |
119 | size_t i; | |
120 | ||
121 | for (i = 0; i < nr; i++) { | |
122 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
123 | if (!x) { | |
124 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 125 | return 0; |
484748f0 CL |
126 | } |
127 | } | |
865762a8 | 128 | return i; |
484748f0 CL |
129 | } |
130 | ||
84c07d11 | 131 | #ifdef CONFIG_MEMCG_KMEM |
510ded33 TH |
132 | |
133 | LIST_HEAD(slab_root_caches); | |
63b02ef7 | 134 | static DEFINE_SPINLOCK(memcg_kmem_wq_lock); |
510ded33 | 135 | |
f0a3a24b RG |
136 | static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref); |
137 | ||
f7ce3190 | 138 | void slab_init_memcg_params(struct kmem_cache *s) |
33a690c4 | 139 | { |
9eeadc8b | 140 | s->memcg_params.root_cache = NULL; |
f7ce3190 | 141 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
9eeadc8b | 142 | INIT_LIST_HEAD(&s->memcg_params.children); |
92ee383f | 143 | s->memcg_params.dying = false; |
f7ce3190 VD |
144 | } |
145 | ||
146 | static int init_memcg_params(struct kmem_cache *s, | |
c03914b7 | 147 | struct kmem_cache *root_cache) |
f7ce3190 VD |
148 | { |
149 | struct memcg_cache_array *arr; | |
33a690c4 | 150 | |
9eeadc8b | 151 | if (root_cache) { |
f0a3a24b RG |
152 | int ret = percpu_ref_init(&s->memcg_params.refcnt, |
153 | kmemcg_cache_shutdown, | |
154 | 0, GFP_KERNEL); | |
155 | if (ret) | |
156 | return ret; | |
157 | ||
f7ce3190 | 158 | s->memcg_params.root_cache = root_cache; |
9eeadc8b | 159 | INIT_LIST_HEAD(&s->memcg_params.children_node); |
bc2791f8 | 160 | INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node); |
33a690c4 | 161 | return 0; |
f7ce3190 | 162 | } |
33a690c4 | 163 | |
f7ce3190 | 164 | slab_init_memcg_params(s); |
33a690c4 | 165 | |
f7ce3190 VD |
166 | if (!memcg_nr_cache_ids) |
167 | return 0; | |
33a690c4 | 168 | |
f80c7dab JW |
169 | arr = kvzalloc(sizeof(struct memcg_cache_array) + |
170 | memcg_nr_cache_ids * sizeof(void *), | |
171 | GFP_KERNEL); | |
f7ce3190 VD |
172 | if (!arr) |
173 | return -ENOMEM; | |
33a690c4 | 174 | |
f7ce3190 | 175 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); |
33a690c4 VD |
176 | return 0; |
177 | } | |
178 | ||
f7ce3190 | 179 | static void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 | 180 | { |
b749ecfa | 181 | if (is_root_cache(s)) { |
f80c7dab | 182 | kvfree(rcu_access_pointer(s->memcg_params.memcg_caches)); |
b749ecfa RG |
183 | } else { |
184 | mem_cgroup_put(s->memcg_params.memcg); | |
185 | WRITE_ONCE(s->memcg_params.memcg, NULL); | |
f0a3a24b | 186 | percpu_ref_exit(&s->memcg_params.refcnt); |
b749ecfa | 187 | } |
f80c7dab JW |
188 | } |
189 | ||
190 | static void free_memcg_params(struct rcu_head *rcu) | |
191 | { | |
192 | struct memcg_cache_array *old; | |
193 | ||
194 | old = container_of(rcu, struct memcg_cache_array, rcu); | |
195 | kvfree(old); | |
33a690c4 VD |
196 | } |
197 | ||
f7ce3190 | 198 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
6f817f4c | 199 | { |
f7ce3190 | 200 | struct memcg_cache_array *old, *new; |
6f817f4c | 201 | |
f80c7dab JW |
202 | new = kvzalloc(sizeof(struct memcg_cache_array) + |
203 | new_array_size * sizeof(void *), GFP_KERNEL); | |
f7ce3190 | 204 | if (!new) |
6f817f4c VD |
205 | return -ENOMEM; |
206 | ||
f7ce3190 VD |
207 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
208 | lockdep_is_held(&slab_mutex)); | |
209 | if (old) | |
210 | memcpy(new->entries, old->entries, | |
211 | memcg_nr_cache_ids * sizeof(void *)); | |
6f817f4c | 212 | |
f7ce3190 VD |
213 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
214 | if (old) | |
f80c7dab | 215 | call_rcu(&old->rcu, free_memcg_params); |
6f817f4c VD |
216 | return 0; |
217 | } | |
218 | ||
55007d84 GC |
219 | int memcg_update_all_caches(int num_memcgs) |
220 | { | |
221 | struct kmem_cache *s; | |
222 | int ret = 0; | |
55007d84 | 223 | |
05257a1a | 224 | mutex_lock(&slab_mutex); |
510ded33 | 225 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
f7ce3190 | 226 | ret = update_memcg_params(s, num_memcgs); |
55007d84 | 227 | /* |
55007d84 GC |
228 | * Instead of freeing the memory, we'll just leave the caches |
229 | * up to this point in an updated state. | |
230 | */ | |
231 | if (ret) | |
05257a1a | 232 | break; |
55007d84 | 233 | } |
55007d84 GC |
234 | mutex_unlock(&slab_mutex); |
235 | return ret; | |
236 | } | |
657dc2f9 | 237 | |
c03914b7 | 238 | void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg) |
657dc2f9 | 239 | { |
510ded33 TH |
240 | if (is_root_cache(s)) { |
241 | list_add(&s->root_caches_node, &slab_root_caches); | |
242 | } else { | |
f0a3a24b | 243 | css_get(&memcg->css); |
c03914b7 | 244 | s->memcg_params.memcg = memcg; |
510ded33 TH |
245 | list_add(&s->memcg_params.children_node, |
246 | &s->memcg_params.root_cache->memcg_params.children); | |
247 | list_add(&s->memcg_params.kmem_caches_node, | |
248 | &s->memcg_params.memcg->kmem_caches); | |
249 | } | |
250 | } | |
251 | ||
252 | static void memcg_unlink_cache(struct kmem_cache *s) | |
253 | { | |
254 | if (is_root_cache(s)) { | |
255 | list_del(&s->root_caches_node); | |
256 | } else { | |
257 | list_del(&s->memcg_params.children_node); | |
258 | list_del(&s->memcg_params.kmem_caches_node); | |
259 | } | |
657dc2f9 | 260 | } |
33a690c4 | 261 | #else |
f7ce3190 | 262 | static inline int init_memcg_params(struct kmem_cache *s, |
c03914b7 | 263 | struct kmem_cache *root_cache) |
33a690c4 VD |
264 | { |
265 | return 0; | |
266 | } | |
267 | ||
f7ce3190 | 268 | static inline void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 VD |
269 | { |
270 | } | |
657dc2f9 | 271 | |
510ded33 | 272 | static inline void memcg_unlink_cache(struct kmem_cache *s) |
657dc2f9 TH |
273 | { |
274 | } | |
84c07d11 | 275 | #endif /* CONFIG_MEMCG_KMEM */ |
55007d84 | 276 | |
692ae74a BL |
277 | /* |
278 | * Figure out what the alignment of the objects will be given a set of | |
279 | * flags, a user specified alignment and the size of the objects. | |
280 | */ | |
f4957d5b AD |
281 | static unsigned int calculate_alignment(slab_flags_t flags, |
282 | unsigned int align, unsigned int size) | |
692ae74a BL |
283 | { |
284 | /* | |
285 | * If the user wants hardware cache aligned objects then follow that | |
286 | * suggestion if the object is sufficiently large. | |
287 | * | |
288 | * The hardware cache alignment cannot override the specified | |
289 | * alignment though. If that is greater then use it. | |
290 | */ | |
291 | if (flags & SLAB_HWCACHE_ALIGN) { | |
f4957d5b | 292 | unsigned int ralign; |
692ae74a BL |
293 | |
294 | ralign = cache_line_size(); | |
295 | while (size <= ralign / 2) | |
296 | ralign /= 2; | |
297 | align = max(align, ralign); | |
298 | } | |
299 | ||
300 | if (align < ARCH_SLAB_MINALIGN) | |
301 | align = ARCH_SLAB_MINALIGN; | |
302 | ||
303 | return ALIGN(align, sizeof(void *)); | |
304 | } | |
305 | ||
423c929c JK |
306 | /* |
307 | * Find a mergeable slab cache | |
308 | */ | |
309 | int slab_unmergeable(struct kmem_cache *s) | |
310 | { | |
311 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
312 | return 1; | |
313 | ||
314 | if (!is_root_cache(s)) | |
315 | return 1; | |
316 | ||
317 | if (s->ctor) | |
318 | return 1; | |
319 | ||
8eb8284b DW |
320 | if (s->usersize) |
321 | return 1; | |
322 | ||
423c929c JK |
323 | /* |
324 | * We may have set a slab to be unmergeable during bootstrap. | |
325 | */ | |
326 | if (s->refcount < 0) | |
327 | return 1; | |
328 | ||
d38a2b7a MS |
329 | #ifdef CONFIG_MEMCG_KMEM |
330 | /* | |
331 | * Skip the dying kmem_cache. | |
332 | */ | |
333 | if (s->memcg_params.dying) | |
334 | return 1; | |
335 | #endif | |
336 | ||
423c929c JK |
337 | return 0; |
338 | } | |
339 | ||
f4957d5b | 340 | struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, |
d50112ed | 341 | slab_flags_t flags, const char *name, void (*ctor)(void *)) |
423c929c JK |
342 | { |
343 | struct kmem_cache *s; | |
344 | ||
c6e28895 | 345 | if (slab_nomerge) |
423c929c JK |
346 | return NULL; |
347 | ||
348 | if (ctor) | |
349 | return NULL; | |
350 | ||
351 | size = ALIGN(size, sizeof(void *)); | |
352 | align = calculate_alignment(flags, align, size); | |
353 | size = ALIGN(size, align); | |
354 | flags = kmem_cache_flags(size, flags, name, NULL); | |
355 | ||
c6e28895 GM |
356 | if (flags & SLAB_NEVER_MERGE) |
357 | return NULL; | |
358 | ||
510ded33 | 359 | list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) { |
423c929c JK |
360 | if (slab_unmergeable(s)) |
361 | continue; | |
362 | ||
363 | if (size > s->size) | |
364 | continue; | |
365 | ||
366 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
367 | continue; | |
368 | /* | |
369 | * Check if alignment is compatible. | |
370 | * Courtesy of Adrian Drzewiecki | |
371 | */ | |
372 | if ((s->size & ~(align - 1)) != s->size) | |
373 | continue; | |
374 | ||
375 | if (s->size - size >= sizeof(void *)) | |
376 | continue; | |
377 | ||
95069ac8 JK |
378 | if (IS_ENABLED(CONFIG_SLAB) && align && |
379 | (align > s->align || s->align % align)) | |
380 | continue; | |
381 | ||
423c929c JK |
382 | return s; |
383 | } | |
384 | return NULL; | |
385 | } | |
386 | ||
c9a77a79 | 387 | static struct kmem_cache *create_cache(const char *name, |
613a5eb5 | 388 | unsigned int object_size, unsigned int align, |
7bbdb81e AD |
389 | slab_flags_t flags, unsigned int useroffset, |
390 | unsigned int usersize, void (*ctor)(void *), | |
c9a77a79 | 391 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
794b1248 VD |
392 | { |
393 | struct kmem_cache *s; | |
394 | int err; | |
395 | ||
8eb8284b DW |
396 | if (WARN_ON(useroffset + usersize > object_size)) |
397 | useroffset = usersize = 0; | |
398 | ||
794b1248 VD |
399 | err = -ENOMEM; |
400 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
401 | if (!s) | |
402 | goto out; | |
403 | ||
404 | s->name = name; | |
613a5eb5 | 405 | s->size = s->object_size = object_size; |
794b1248 VD |
406 | s->align = align; |
407 | s->ctor = ctor; | |
8eb8284b DW |
408 | s->useroffset = useroffset; |
409 | s->usersize = usersize; | |
794b1248 | 410 | |
c03914b7 | 411 | err = init_memcg_params(s, root_cache); |
794b1248 VD |
412 | if (err) |
413 | goto out_free_cache; | |
414 | ||
415 | err = __kmem_cache_create(s, flags); | |
416 | if (err) | |
417 | goto out_free_cache; | |
418 | ||
419 | s->refcount = 1; | |
420 | list_add(&s->list, &slab_caches); | |
c03914b7 | 421 | memcg_link_cache(s, memcg); |
794b1248 VD |
422 | out: |
423 | if (err) | |
424 | return ERR_PTR(err); | |
425 | return s; | |
426 | ||
427 | out_free_cache: | |
f7ce3190 | 428 | destroy_memcg_params(s); |
7c4da061 | 429 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
430 | goto out; |
431 | } | |
45906855 | 432 | |
f496990f MR |
433 | /** |
434 | * kmem_cache_create_usercopy - Create a cache with a region suitable | |
435 | * for copying to userspace | |
77be4b13 SK |
436 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
437 | * @size: The size of objects to be created in this cache. | |
438 | * @align: The required alignment for the objects. | |
439 | * @flags: SLAB flags | |
8eb8284b DW |
440 | * @useroffset: Usercopy region offset |
441 | * @usersize: Usercopy region size | |
77be4b13 SK |
442 | * @ctor: A constructor for the objects. |
443 | * | |
77be4b13 SK |
444 | * Cannot be called within a interrupt, but can be interrupted. |
445 | * The @ctor is run when new pages are allocated by the cache. | |
446 | * | |
447 | * The flags are | |
448 | * | |
449 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
450 | * to catch references to uninitialised memory. | |
451 | * | |
f496990f | 452 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check |
77be4b13 SK |
453 | * for buffer overruns. |
454 | * | |
455 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
456 | * cacheline. This can be beneficial if you're counting cycles as closely | |
457 | * as davem. | |
f496990f MR |
458 | * |
459 | * Return: a pointer to the cache on success, NULL on failure. | |
77be4b13 | 460 | */ |
2633d7a0 | 461 | struct kmem_cache * |
f4957d5b AD |
462 | kmem_cache_create_usercopy(const char *name, |
463 | unsigned int size, unsigned int align, | |
7bbdb81e AD |
464 | slab_flags_t flags, |
465 | unsigned int useroffset, unsigned int usersize, | |
8eb8284b | 466 | void (*ctor)(void *)) |
77be4b13 | 467 | { |
40911a79 | 468 | struct kmem_cache *s = NULL; |
3dec16ea | 469 | const char *cache_name; |
3965fc36 | 470 | int err; |
039363f3 | 471 | |
77be4b13 | 472 | get_online_cpus(); |
03afc0e2 | 473 | get_online_mems(); |
05257a1a | 474 | memcg_get_cache_ids(); |
03afc0e2 | 475 | |
77be4b13 | 476 | mutex_lock(&slab_mutex); |
686d550d | 477 | |
794b1248 | 478 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 479 | if (err) { |
3965fc36 | 480 | goto out_unlock; |
3aa24f51 | 481 | } |
686d550d | 482 | |
e70954fd TG |
483 | /* Refuse requests with allocator specific flags */ |
484 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
485 | err = -EINVAL; | |
486 | goto out_unlock; | |
487 | } | |
488 | ||
d8843922 GC |
489 | /* |
490 | * Some allocators will constraint the set of valid flags to a subset | |
491 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
492 | * case, and we'll just provide them with a sanitized version of the | |
493 | * passed flags. | |
494 | */ | |
495 | flags &= CACHE_CREATE_MASK; | |
686d550d | 496 | |
8eb8284b DW |
497 | /* Fail closed on bad usersize of useroffset values. */ |
498 | if (WARN_ON(!usersize && useroffset) || | |
499 | WARN_ON(size < usersize || size - usersize < useroffset)) | |
500 | usersize = useroffset = 0; | |
501 | ||
502 | if (!usersize) | |
503 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
794b1248 | 504 | if (s) |
3965fc36 | 505 | goto out_unlock; |
2633d7a0 | 506 | |
3dec16ea | 507 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
508 | if (!cache_name) { |
509 | err = -ENOMEM; | |
510 | goto out_unlock; | |
511 | } | |
7c9adf5a | 512 | |
613a5eb5 | 513 | s = create_cache(cache_name, size, |
c9a77a79 | 514 | calculate_alignment(flags, align, size), |
8eb8284b | 515 | flags, useroffset, usersize, ctor, NULL, NULL); |
794b1248 VD |
516 | if (IS_ERR(s)) { |
517 | err = PTR_ERR(s); | |
3dec16ea | 518 | kfree_const(cache_name); |
794b1248 | 519 | } |
3965fc36 VD |
520 | |
521 | out_unlock: | |
20cea968 | 522 | mutex_unlock(&slab_mutex); |
03afc0e2 | 523 | |
05257a1a | 524 | memcg_put_cache_ids(); |
03afc0e2 | 525 | put_online_mems(); |
20cea968 CL |
526 | put_online_cpus(); |
527 | ||
ba3253c7 | 528 | if (err) { |
686d550d CL |
529 | if (flags & SLAB_PANIC) |
530 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
531 | name, err); | |
532 | else { | |
1170532b | 533 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
534 | name, err); |
535 | dump_stack(); | |
536 | } | |
686d550d CL |
537 | return NULL; |
538 | } | |
039363f3 CL |
539 | return s; |
540 | } | |
8eb8284b DW |
541 | EXPORT_SYMBOL(kmem_cache_create_usercopy); |
542 | ||
f496990f MR |
543 | /** |
544 | * kmem_cache_create - Create a cache. | |
545 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
546 | * @size: The size of objects to be created in this cache. | |
547 | * @align: The required alignment for the objects. | |
548 | * @flags: SLAB flags | |
549 | * @ctor: A constructor for the objects. | |
550 | * | |
551 | * Cannot be called within a interrupt, but can be interrupted. | |
552 | * The @ctor is run when new pages are allocated by the cache. | |
553 | * | |
554 | * The flags are | |
555 | * | |
556 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
557 | * to catch references to uninitialised memory. | |
558 | * | |
559 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check | |
560 | * for buffer overruns. | |
561 | * | |
562 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
563 | * cacheline. This can be beneficial if you're counting cycles as closely | |
564 | * as davem. | |
565 | * | |
566 | * Return: a pointer to the cache on success, NULL on failure. | |
567 | */ | |
8eb8284b | 568 | struct kmem_cache * |
f4957d5b | 569 | kmem_cache_create(const char *name, unsigned int size, unsigned int align, |
8eb8284b DW |
570 | slab_flags_t flags, void (*ctor)(void *)) |
571 | { | |
6d07d1cd | 572 | return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, |
8eb8284b DW |
573 | ctor); |
574 | } | |
794b1248 | 575 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 576 | |
657dc2f9 | 577 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 578 | { |
657dc2f9 TH |
579 | LIST_HEAD(to_destroy); |
580 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 581 | |
657dc2f9 | 582 | /* |
5f0d5a3a | 583 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
657dc2f9 TH |
584 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
585 | * through RCU and and the associated kmem_cache are dereferenced | |
586 | * while freeing the pages, so the kmem_caches should be freed only | |
587 | * after the pending RCU operations are finished. As rcu_barrier() | |
588 | * is a pretty slow operation, we batch all pending destructions | |
589 | * asynchronously. | |
590 | */ | |
591 | mutex_lock(&slab_mutex); | |
592 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
593 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 594 | |
657dc2f9 TH |
595 | if (list_empty(&to_destroy)) |
596 | return; | |
597 | ||
598 | rcu_barrier(); | |
599 | ||
600 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
601 | #ifdef SLAB_SUPPORTS_SYSFS | |
602 | sysfs_slab_release(s); | |
603 | #else | |
604 | slab_kmem_cache_release(s); | |
605 | #endif | |
606 | } | |
d5b3cf71 VD |
607 | } |
608 | ||
657dc2f9 | 609 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 610 | { |
f9fa1d91 GT |
611 | /* free asan quarantined objects */ |
612 | kasan_cache_shutdown(s); | |
613 | ||
657dc2f9 TH |
614 | if (__kmem_cache_shutdown(s) != 0) |
615 | return -EBUSY; | |
d5b3cf71 | 616 | |
510ded33 | 617 | memcg_unlink_cache(s); |
657dc2f9 | 618 | list_del(&s->list); |
d5b3cf71 | 619 | |
5f0d5a3a | 620 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
d50d82fa MP |
621 | #ifdef SLAB_SUPPORTS_SYSFS |
622 | sysfs_slab_unlink(s); | |
623 | #endif | |
657dc2f9 TH |
624 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
625 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
626 | } else { | |
d5b3cf71 | 627 | #ifdef SLAB_SUPPORTS_SYSFS |
d50d82fa | 628 | sysfs_slab_unlink(s); |
bf5eb3de | 629 | sysfs_slab_release(s); |
d5b3cf71 VD |
630 | #else |
631 | slab_kmem_cache_release(s); | |
632 | #endif | |
633 | } | |
657dc2f9 TH |
634 | |
635 | return 0; | |
d5b3cf71 VD |
636 | } |
637 | ||
84c07d11 | 638 | #ifdef CONFIG_MEMCG_KMEM |
794b1248 | 639 | /* |
776ed0f0 | 640 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
641 | * @memcg: The memory cgroup the new cache is for. |
642 | * @root_cache: The parent of the new cache. | |
643 | * | |
644 | * This function attempts to create a kmem cache that will serve allocation | |
645 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
646 | * from its parent. | |
647 | */ | |
d5b3cf71 VD |
648 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
649 | struct kmem_cache *root_cache) | |
2633d7a0 | 650 | { |
3e0350a3 | 651 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
33398cf2 | 652 | struct cgroup_subsys_state *css = &memcg->css; |
f7ce3190 | 653 | struct memcg_cache_array *arr; |
bd673145 | 654 | struct kmem_cache *s = NULL; |
794b1248 | 655 | char *cache_name; |
f7ce3190 | 656 | int idx; |
794b1248 VD |
657 | |
658 | get_online_cpus(); | |
03afc0e2 VD |
659 | get_online_mems(); |
660 | ||
794b1248 VD |
661 | mutex_lock(&slab_mutex); |
662 | ||
2a4db7eb | 663 | /* |
567e9ab2 | 664 | * The memory cgroup could have been offlined while the cache |
2a4db7eb VD |
665 | * creation work was pending. |
666 | */ | |
57033297 | 667 | if (memcg->kmem_state != KMEM_ONLINE) |
2a4db7eb VD |
668 | goto out_unlock; |
669 | ||
f7ce3190 VD |
670 | idx = memcg_cache_id(memcg); |
671 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
672 | lockdep_is_held(&slab_mutex)); | |
673 | ||
d5b3cf71 VD |
674 | /* |
675 | * Since per-memcg caches are created asynchronously on first | |
676 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
677 | * create the same cache, but only one of them may succeed. | |
678 | */ | |
f7ce3190 | 679 | if (arr->entries[idx]) |
d5b3cf71 VD |
680 | goto out_unlock; |
681 | ||
f1008365 | 682 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
73f576c0 JW |
683 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, |
684 | css->serial_nr, memcg_name_buf); | |
794b1248 VD |
685 | if (!cache_name) |
686 | goto out_unlock; | |
687 | ||
c9a77a79 | 688 | s = create_cache(cache_name, root_cache->object_size, |
613a5eb5 | 689 | root_cache->align, |
f773e36d | 690 | root_cache->flags & CACHE_CREATE_MASK, |
8eb8284b | 691 | root_cache->useroffset, root_cache->usersize, |
f773e36d | 692 | root_cache->ctor, memcg, root_cache); |
d5b3cf71 VD |
693 | /* |
694 | * If we could not create a memcg cache, do not complain, because | |
695 | * that's not critical at all as we can always proceed with the root | |
696 | * cache. | |
697 | */ | |
bd673145 | 698 | if (IS_ERR(s)) { |
794b1248 | 699 | kfree(cache_name); |
d5b3cf71 | 700 | goto out_unlock; |
bd673145 | 701 | } |
794b1248 | 702 | |
d5b3cf71 | 703 | /* |
f0a3a24b | 704 | * Since readers won't lock (see memcg_kmem_get_cache()), we need a |
d5b3cf71 VD |
705 | * barrier here to ensure nobody will see the kmem_cache partially |
706 | * initialized. | |
707 | */ | |
708 | smp_wmb(); | |
f7ce3190 | 709 | arr->entries[idx] = s; |
d5b3cf71 | 710 | |
794b1248 VD |
711 | out_unlock: |
712 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
713 | |
714 | put_online_mems(); | |
794b1248 | 715 | put_online_cpus(); |
2633d7a0 | 716 | } |
b8529907 | 717 | |
0b14e8aa | 718 | static void kmemcg_workfn(struct work_struct *work) |
01fb58bc TH |
719 | { |
720 | struct kmem_cache *s = container_of(work, struct kmem_cache, | |
0b14e8aa | 721 | memcg_params.work); |
01fb58bc TH |
722 | |
723 | get_online_cpus(); | |
724 | get_online_mems(); | |
725 | ||
726 | mutex_lock(&slab_mutex); | |
0b14e8aa | 727 | s->memcg_params.work_fn(s); |
01fb58bc TH |
728 | mutex_unlock(&slab_mutex); |
729 | ||
730 | put_online_mems(); | |
731 | put_online_cpus(); | |
01fb58bc TH |
732 | } |
733 | ||
0b14e8aa | 734 | static void kmemcg_rcufn(struct rcu_head *head) |
01fb58bc TH |
735 | { |
736 | struct kmem_cache *s = container_of(head, struct kmem_cache, | |
0b14e8aa | 737 | memcg_params.rcu_head); |
01fb58bc TH |
738 | |
739 | /* | |
0b14e8aa | 740 | * We need to grab blocking locks. Bounce to ->work. The |
01fb58bc | 741 | * work item shares the space with the RCU head and can't be |
b991cee5 | 742 | * initialized earlier. |
01fb58bc | 743 | */ |
0b14e8aa RG |
744 | INIT_WORK(&s->memcg_params.work, kmemcg_workfn); |
745 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.work); | |
01fb58bc TH |
746 | } |
747 | ||
f0a3a24b RG |
748 | static void kmemcg_cache_shutdown_fn(struct kmem_cache *s) |
749 | { | |
750 | WARN_ON(shutdown_cache(s)); | |
751 | } | |
752 | ||
753 | static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref) | |
754 | { | |
755 | struct kmem_cache *s = container_of(percpu_ref, struct kmem_cache, | |
756 | memcg_params.refcnt); | |
757 | unsigned long flags; | |
758 | ||
759 | spin_lock_irqsave(&memcg_kmem_wq_lock, flags); | |
760 | if (s->memcg_params.root_cache->memcg_params.dying) | |
761 | goto unlock; | |
762 | ||
763 | s->memcg_params.work_fn = kmemcg_cache_shutdown_fn; | |
764 | INIT_WORK(&s->memcg_params.work, kmemcg_workfn); | |
765 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.work); | |
766 | ||
767 | unlock: | |
768 | spin_unlock_irqrestore(&memcg_kmem_wq_lock, flags); | |
769 | } | |
770 | ||
771 | static void kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s) | |
772 | { | |
773 | __kmemcg_cache_deactivate_after_rcu(s); | |
774 | percpu_ref_kill(&s->memcg_params.refcnt); | |
775 | } | |
776 | ||
43486694 | 777 | static void kmemcg_cache_deactivate(struct kmem_cache *s) |
01fb58bc | 778 | { |
f0a3a24b | 779 | if (WARN_ON_ONCE(is_root_cache(s))) |
01fb58bc TH |
780 | return; |
781 | ||
43486694 | 782 | __kmemcg_cache_deactivate(s); |
fcf8a1e4 | 783 | s->flags |= SLAB_DEACTIVATED; |
43486694 | 784 | |
63b02ef7 RG |
785 | /* |
786 | * memcg_kmem_wq_lock is used to synchronize memcg_params.dying | |
787 | * flag and make sure that no new kmem_cache deactivation tasks | |
788 | * are queued (see flush_memcg_workqueue() ). | |
789 | */ | |
790 | spin_lock_irq(&memcg_kmem_wq_lock); | |
92ee383f | 791 | if (s->memcg_params.root_cache->memcg_params.dying) |
63b02ef7 | 792 | goto unlock; |
92ee383f | 793 | |
f0a3a24b | 794 | s->memcg_params.work_fn = kmemcg_cache_deactivate_after_rcu; |
0b14e8aa | 795 | call_rcu(&s->memcg_params.rcu_head, kmemcg_rcufn); |
63b02ef7 RG |
796 | unlock: |
797 | spin_unlock_irq(&memcg_kmem_wq_lock); | |
01fb58bc TH |
798 | } |
799 | ||
fb2f2b0a RG |
800 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg, |
801 | struct mem_cgroup *parent) | |
2a4db7eb VD |
802 | { |
803 | int idx; | |
804 | struct memcg_cache_array *arr; | |
d6e0b7fa | 805 | struct kmem_cache *s, *c; |
fb2f2b0a | 806 | unsigned int nr_reparented; |
2a4db7eb VD |
807 | |
808 | idx = memcg_cache_id(memcg); | |
809 | ||
d6e0b7fa VD |
810 | get_online_cpus(); |
811 | get_online_mems(); | |
812 | ||
2a4db7eb | 813 | mutex_lock(&slab_mutex); |
510ded33 | 814 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
2a4db7eb VD |
815 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, |
816 | lockdep_is_held(&slab_mutex)); | |
d6e0b7fa VD |
817 | c = arr->entries[idx]; |
818 | if (!c) | |
819 | continue; | |
820 | ||
43486694 | 821 | kmemcg_cache_deactivate(c); |
2a4db7eb VD |
822 | arr->entries[idx] = NULL; |
823 | } | |
fb2f2b0a RG |
824 | nr_reparented = 0; |
825 | list_for_each_entry(s, &memcg->kmem_caches, | |
826 | memcg_params.kmem_caches_node) { | |
827 | WRITE_ONCE(s->memcg_params.memcg, parent); | |
828 | css_put(&memcg->css); | |
829 | nr_reparented++; | |
830 | } | |
831 | if (nr_reparented) { | |
832 | list_splice_init(&memcg->kmem_caches, | |
833 | &parent->kmem_caches); | |
834 | css_get_many(&parent->css, nr_reparented); | |
835 | } | |
2a4db7eb | 836 | mutex_unlock(&slab_mutex); |
d6e0b7fa VD |
837 | |
838 | put_online_mems(); | |
839 | put_online_cpus(); | |
2a4db7eb VD |
840 | } |
841 | ||
657dc2f9 | 842 | static int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
843 | { |
844 | struct memcg_cache_array *arr; | |
845 | struct kmem_cache *c, *c2; | |
846 | LIST_HEAD(busy); | |
847 | int i; | |
848 | ||
849 | BUG_ON(!is_root_cache(s)); | |
850 | ||
851 | /* | |
852 | * First, shutdown active caches, i.e. caches that belong to online | |
853 | * memory cgroups. | |
854 | */ | |
855 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
856 | lockdep_is_held(&slab_mutex)); | |
857 | for_each_memcg_cache_index(i) { | |
858 | c = arr->entries[i]; | |
859 | if (!c) | |
860 | continue; | |
657dc2f9 | 861 | if (shutdown_cache(c)) |
d60fdcc9 VD |
862 | /* |
863 | * The cache still has objects. Move it to a temporary | |
864 | * list so as not to try to destroy it for a second | |
865 | * time while iterating over inactive caches below. | |
866 | */ | |
9eeadc8b | 867 | list_move(&c->memcg_params.children_node, &busy); |
d60fdcc9 VD |
868 | else |
869 | /* | |
870 | * The cache is empty and will be destroyed soon. Clear | |
871 | * the pointer to it in the memcg_caches array so that | |
872 | * it will never be accessed even if the root cache | |
873 | * stays alive. | |
874 | */ | |
875 | arr->entries[i] = NULL; | |
876 | } | |
877 | ||
878 | /* | |
879 | * Second, shutdown all caches left from memory cgroups that are now | |
880 | * offline. | |
881 | */ | |
9eeadc8b TH |
882 | list_for_each_entry_safe(c, c2, &s->memcg_params.children, |
883 | memcg_params.children_node) | |
657dc2f9 | 884 | shutdown_cache(c); |
d60fdcc9 | 885 | |
9eeadc8b | 886 | list_splice(&busy, &s->memcg_params.children); |
d60fdcc9 VD |
887 | |
888 | /* | |
889 | * A cache being destroyed must be empty. In particular, this means | |
890 | * that all per memcg caches attached to it must be empty too. | |
891 | */ | |
9eeadc8b | 892 | if (!list_empty(&s->memcg_params.children)) |
d60fdcc9 VD |
893 | return -EBUSY; |
894 | return 0; | |
895 | } | |
92ee383f | 896 | |
d38a2b7a | 897 | static void memcg_set_kmem_cache_dying(struct kmem_cache *s) |
92ee383f | 898 | { |
63b02ef7 | 899 | spin_lock_irq(&memcg_kmem_wq_lock); |
92ee383f | 900 | s->memcg_params.dying = true; |
63b02ef7 | 901 | spin_unlock_irq(&memcg_kmem_wq_lock); |
d38a2b7a | 902 | } |
92ee383f | 903 | |
d38a2b7a MS |
904 | static void flush_memcg_workqueue(struct kmem_cache *s) |
905 | { | |
92ee383f | 906 | /* |
43486694 | 907 | * SLAB and SLUB deactivate the kmem_caches through call_rcu. Make |
92ee383f SB |
908 | * sure all registered rcu callbacks have been invoked. |
909 | */ | |
43486694 | 910 | rcu_barrier(); |
92ee383f SB |
911 | |
912 | /* | |
913 | * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB | |
914 | * deactivates the memcg kmem_caches through workqueue. Make sure all | |
915 | * previous workitems on workqueue are processed. | |
916 | */ | |
2fe20210 AH |
917 | if (likely(memcg_kmem_cache_wq)) |
918 | flush_workqueue(memcg_kmem_cache_wq); | |
a264df74 RG |
919 | |
920 | /* | |
921 | * If we're racing with children kmem_cache deactivation, it might | |
922 | * take another rcu grace period to complete their destruction. | |
923 | * At this moment the corresponding percpu_ref_kill() call should be | |
924 | * done, but it might take another rcu grace period to complete | |
925 | * switching to the atomic mode. | |
926 | * Please, note that we check without grabbing the slab_mutex. It's safe | |
927 | * because at this moment the children list can't grow. | |
928 | */ | |
929 | if (!list_empty(&s->memcg_params.children)) | |
930 | rcu_barrier(); | |
92ee383f | 931 | } |
d60fdcc9 | 932 | #else |
657dc2f9 | 933 | static inline int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
934 | { |
935 | return 0; | |
936 | } | |
84c07d11 | 937 | #endif /* CONFIG_MEMCG_KMEM */ |
97d06609 | 938 | |
41a21285 CL |
939 | void slab_kmem_cache_release(struct kmem_cache *s) |
940 | { | |
52b4b950 | 941 | __kmem_cache_release(s); |
f7ce3190 | 942 | destroy_memcg_params(s); |
3dec16ea | 943 | kfree_const(s->name); |
41a21285 CL |
944 | kmem_cache_free(kmem_cache, s); |
945 | } | |
946 | ||
945cf2b6 CL |
947 | void kmem_cache_destroy(struct kmem_cache *s) |
948 | { | |
d60fdcc9 | 949 | int err; |
d5b3cf71 | 950 | |
3942d299 SS |
951 | if (unlikely(!s)) |
952 | return; | |
953 | ||
945cf2b6 | 954 | get_online_cpus(); |
03afc0e2 VD |
955 | get_online_mems(); |
956 | ||
945cf2b6 | 957 | mutex_lock(&slab_mutex); |
b8529907 | 958 | |
945cf2b6 | 959 | s->refcount--; |
b8529907 VD |
960 | if (s->refcount) |
961 | goto out_unlock; | |
962 | ||
d38a2b7a MS |
963 | #ifdef CONFIG_MEMCG_KMEM |
964 | memcg_set_kmem_cache_dying(s); | |
965 | ||
966 | mutex_unlock(&slab_mutex); | |
967 | ||
968 | put_online_mems(); | |
969 | put_online_cpus(); | |
970 | ||
971 | flush_memcg_workqueue(s); | |
972 | ||
973 | get_online_cpus(); | |
974 | get_online_mems(); | |
975 | ||
976 | mutex_lock(&slab_mutex); | |
977 | #endif | |
978 | ||
657dc2f9 | 979 | err = shutdown_memcg_caches(s); |
d60fdcc9 | 980 | if (!err) |
657dc2f9 | 981 | err = shutdown_cache(s); |
b8529907 | 982 | |
cd918c55 | 983 | if (err) { |
756a025f JP |
984 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
985 | s->name); | |
cd918c55 VD |
986 | dump_stack(); |
987 | } | |
b8529907 VD |
988 | out_unlock: |
989 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 990 | |
03afc0e2 | 991 | put_online_mems(); |
945cf2b6 CL |
992 | put_online_cpus(); |
993 | } | |
994 | EXPORT_SYMBOL(kmem_cache_destroy); | |
995 | ||
03afc0e2 VD |
996 | /** |
997 | * kmem_cache_shrink - Shrink a cache. | |
998 | * @cachep: The cache to shrink. | |
999 | * | |
1000 | * Releases as many slabs as possible for a cache. | |
1001 | * To help debugging, a zero exit status indicates all slabs were released. | |
a862f68a MR |
1002 | * |
1003 | * Return: %0 if all slabs were released, non-zero otherwise | |
03afc0e2 VD |
1004 | */ |
1005 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
1006 | { | |
1007 | int ret; | |
1008 | ||
1009 | get_online_cpus(); | |
1010 | get_online_mems(); | |
55834c59 | 1011 | kasan_cache_shrink(cachep); |
c9fc5864 | 1012 | ret = __kmem_cache_shrink(cachep); |
03afc0e2 VD |
1013 | put_online_mems(); |
1014 | put_online_cpus(); | |
1015 | return ret; | |
1016 | } | |
1017 | EXPORT_SYMBOL(kmem_cache_shrink); | |
1018 | ||
04f768a3 WL |
1019 | /** |
1020 | * kmem_cache_shrink_all - shrink a cache and all memcg caches for root cache | |
1021 | * @s: The cache pointer | |
1022 | */ | |
1023 | void kmem_cache_shrink_all(struct kmem_cache *s) | |
1024 | { | |
1025 | struct kmem_cache *c; | |
1026 | ||
1027 | if (!IS_ENABLED(CONFIG_MEMCG_KMEM) || !is_root_cache(s)) { | |
1028 | kmem_cache_shrink(s); | |
1029 | return; | |
1030 | } | |
1031 | ||
1032 | get_online_cpus(); | |
1033 | get_online_mems(); | |
1034 | kasan_cache_shrink(s); | |
1035 | __kmem_cache_shrink(s); | |
1036 | ||
1037 | /* | |
1038 | * We have to take the slab_mutex to protect from the memcg list | |
1039 | * modification. | |
1040 | */ | |
1041 | mutex_lock(&slab_mutex); | |
1042 | for_each_memcg_cache(c, s) { | |
1043 | /* | |
1044 | * Don't need to shrink deactivated memcg caches. | |
1045 | */ | |
1046 | if (s->flags & SLAB_DEACTIVATED) | |
1047 | continue; | |
1048 | kasan_cache_shrink(c); | |
1049 | __kmem_cache_shrink(c); | |
1050 | } | |
1051 | mutex_unlock(&slab_mutex); | |
1052 | put_online_mems(); | |
1053 | put_online_cpus(); | |
1054 | } | |
1055 | ||
fda90124 | 1056 | bool slab_is_available(void) |
97d06609 CL |
1057 | { |
1058 | return slab_state >= UP; | |
1059 | } | |
b7454ad3 | 1060 | |
45530c44 CL |
1061 | #ifndef CONFIG_SLOB |
1062 | /* Create a cache during boot when no slab services are available yet */ | |
361d575e AD |
1063 | void __init create_boot_cache(struct kmem_cache *s, const char *name, |
1064 | unsigned int size, slab_flags_t flags, | |
1065 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
1066 | { |
1067 | int err; | |
59bb4798 | 1068 | unsigned int align = ARCH_KMALLOC_MINALIGN; |
45530c44 CL |
1069 | |
1070 | s->name = name; | |
1071 | s->size = s->object_size = size; | |
59bb4798 VB |
1072 | |
1073 | /* | |
1074 | * For power of two sizes, guarantee natural alignment for kmalloc | |
1075 | * caches, regardless of SL*B debugging options. | |
1076 | */ | |
1077 | if (is_power_of_2(size)) | |
1078 | align = max(align, size); | |
1079 | s->align = calculate_alignment(flags, align, size); | |
1080 | ||
8eb8284b DW |
1081 | s->useroffset = useroffset; |
1082 | s->usersize = usersize; | |
f7ce3190 VD |
1083 | |
1084 | slab_init_memcg_params(s); | |
1085 | ||
45530c44 CL |
1086 | err = __kmem_cache_create(s, flags); |
1087 | ||
1088 | if (err) | |
361d575e | 1089 | panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", |
45530c44 CL |
1090 | name, size, err); |
1091 | ||
1092 | s->refcount = -1; /* Exempt from merging for now */ | |
1093 | } | |
1094 | ||
55de8b9c AD |
1095 | struct kmem_cache *__init create_kmalloc_cache(const char *name, |
1096 | unsigned int size, slab_flags_t flags, | |
1097 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
1098 | { |
1099 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
1100 | ||
1101 | if (!s) | |
1102 | panic("Out of memory when creating slab %s\n", name); | |
1103 | ||
6c0c21ad | 1104 | create_boot_cache(s, name, size, flags, useroffset, usersize); |
45530c44 | 1105 | list_add(&s->list, &slab_caches); |
c03914b7 | 1106 | memcg_link_cache(s, NULL); |
45530c44 CL |
1107 | s->refcount = 1; |
1108 | return s; | |
1109 | } | |
1110 | ||
cc252eae | 1111 | struct kmem_cache * |
a07057dc AB |
1112 | kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = |
1113 | { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; | |
9425c58e CL |
1114 | EXPORT_SYMBOL(kmalloc_caches); |
1115 | ||
2c59dd65 CL |
1116 | /* |
1117 | * Conversion table for small slabs sizes / 8 to the index in the | |
1118 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
1119 | * of two cache sizes there. The size of larger slabs can be determined using | |
1120 | * fls. | |
1121 | */ | |
d5f86655 | 1122 | static u8 size_index[24] __ro_after_init = { |
2c59dd65 CL |
1123 | 3, /* 8 */ |
1124 | 4, /* 16 */ | |
1125 | 5, /* 24 */ | |
1126 | 5, /* 32 */ | |
1127 | 6, /* 40 */ | |
1128 | 6, /* 48 */ | |
1129 | 6, /* 56 */ | |
1130 | 6, /* 64 */ | |
1131 | 1, /* 72 */ | |
1132 | 1, /* 80 */ | |
1133 | 1, /* 88 */ | |
1134 | 1, /* 96 */ | |
1135 | 7, /* 104 */ | |
1136 | 7, /* 112 */ | |
1137 | 7, /* 120 */ | |
1138 | 7, /* 128 */ | |
1139 | 2, /* 136 */ | |
1140 | 2, /* 144 */ | |
1141 | 2, /* 152 */ | |
1142 | 2, /* 160 */ | |
1143 | 2, /* 168 */ | |
1144 | 2, /* 176 */ | |
1145 | 2, /* 184 */ | |
1146 | 2 /* 192 */ | |
1147 | }; | |
1148 | ||
ac914d08 | 1149 | static inline unsigned int size_index_elem(unsigned int bytes) |
2c59dd65 CL |
1150 | { |
1151 | return (bytes - 1) / 8; | |
1152 | } | |
1153 | ||
1154 | /* | |
1155 | * Find the kmem_cache structure that serves a given size of | |
1156 | * allocation | |
1157 | */ | |
1158 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
1159 | { | |
d5f86655 | 1160 | unsigned int index; |
2c59dd65 CL |
1161 | |
1162 | if (size <= 192) { | |
1163 | if (!size) | |
1164 | return ZERO_SIZE_PTR; | |
1165 | ||
1166 | index = size_index[size_index_elem(size)]; | |
61448479 | 1167 | } else { |
221d7da6 | 1168 | if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) |
61448479 | 1169 | return NULL; |
2c59dd65 | 1170 | index = fls(size - 1); |
61448479 | 1171 | } |
2c59dd65 | 1172 | |
cc252eae | 1173 | return kmalloc_caches[kmalloc_type(flags)][index]; |
2c59dd65 CL |
1174 | } |
1175 | ||
cb5d9fb3 PL |
1176 | #ifdef CONFIG_ZONE_DMA |
1177 | #define INIT_KMALLOC_INFO(__size, __short_size) \ | |
1178 | { \ | |
1179 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ | |
1180 | .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ | |
1181 | .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \ | |
1182 | .size = __size, \ | |
1183 | } | |
1184 | #else | |
1185 | #define INIT_KMALLOC_INFO(__size, __short_size) \ | |
1186 | { \ | |
1187 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ | |
1188 | .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ | |
1189 | .size = __size, \ | |
1190 | } | |
1191 | #endif | |
1192 | ||
4066c33d GG |
1193 | /* |
1194 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
1195 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
1196 | * kmalloc-67108864. | |
1197 | */ | |
af3b5f87 | 1198 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
cb5d9fb3 PL |
1199 | INIT_KMALLOC_INFO(0, 0), |
1200 | INIT_KMALLOC_INFO(96, 96), | |
1201 | INIT_KMALLOC_INFO(192, 192), | |
1202 | INIT_KMALLOC_INFO(8, 8), | |
1203 | INIT_KMALLOC_INFO(16, 16), | |
1204 | INIT_KMALLOC_INFO(32, 32), | |
1205 | INIT_KMALLOC_INFO(64, 64), | |
1206 | INIT_KMALLOC_INFO(128, 128), | |
1207 | INIT_KMALLOC_INFO(256, 256), | |
1208 | INIT_KMALLOC_INFO(512, 512), | |
1209 | INIT_KMALLOC_INFO(1024, 1k), | |
1210 | INIT_KMALLOC_INFO(2048, 2k), | |
1211 | INIT_KMALLOC_INFO(4096, 4k), | |
1212 | INIT_KMALLOC_INFO(8192, 8k), | |
1213 | INIT_KMALLOC_INFO(16384, 16k), | |
1214 | INIT_KMALLOC_INFO(32768, 32k), | |
1215 | INIT_KMALLOC_INFO(65536, 64k), | |
1216 | INIT_KMALLOC_INFO(131072, 128k), | |
1217 | INIT_KMALLOC_INFO(262144, 256k), | |
1218 | INIT_KMALLOC_INFO(524288, 512k), | |
1219 | INIT_KMALLOC_INFO(1048576, 1M), | |
1220 | INIT_KMALLOC_INFO(2097152, 2M), | |
1221 | INIT_KMALLOC_INFO(4194304, 4M), | |
1222 | INIT_KMALLOC_INFO(8388608, 8M), | |
1223 | INIT_KMALLOC_INFO(16777216, 16M), | |
1224 | INIT_KMALLOC_INFO(33554432, 32M), | |
1225 | INIT_KMALLOC_INFO(67108864, 64M) | |
4066c33d GG |
1226 | }; |
1227 | ||
f97d5f63 | 1228 | /* |
34cc6990 DS |
1229 | * Patch up the size_index table if we have strange large alignment |
1230 | * requirements for the kmalloc array. This is only the case for | |
1231 | * MIPS it seems. The standard arches will not generate any code here. | |
1232 | * | |
1233 | * Largest permitted alignment is 256 bytes due to the way we | |
1234 | * handle the index determination for the smaller caches. | |
1235 | * | |
1236 | * Make sure that nothing crazy happens if someone starts tinkering | |
1237 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 1238 | */ |
34cc6990 | 1239 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 | 1240 | { |
ac914d08 | 1241 | unsigned int i; |
f97d5f63 | 1242 | |
2c59dd65 CL |
1243 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
1244 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
1245 | ||
1246 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
ac914d08 | 1247 | unsigned int elem = size_index_elem(i); |
2c59dd65 CL |
1248 | |
1249 | if (elem >= ARRAY_SIZE(size_index)) | |
1250 | break; | |
1251 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
1252 | } | |
1253 | ||
1254 | if (KMALLOC_MIN_SIZE >= 64) { | |
1255 | /* | |
1256 | * The 96 byte size cache is not used if the alignment | |
1257 | * is 64 byte. | |
1258 | */ | |
1259 | for (i = 64 + 8; i <= 96; i += 8) | |
1260 | size_index[size_index_elem(i)] = 7; | |
1261 | ||
1262 | } | |
1263 | ||
1264 | if (KMALLOC_MIN_SIZE >= 128) { | |
1265 | /* | |
1266 | * The 192 byte sized cache is not used if the alignment | |
1267 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
1268 | * instead. | |
1269 | */ | |
1270 | for (i = 128 + 8; i <= 192; i += 8) | |
1271 | size_index[size_index_elem(i)] = 8; | |
1272 | } | |
34cc6990 DS |
1273 | } |
1274 | ||
1291523f | 1275 | static void __init |
13657d0a | 1276 | new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) |
a9730fca | 1277 | { |
cb5d9fb3 | 1278 | if (type == KMALLOC_RECLAIM) |
1291523f | 1279 | flags |= SLAB_RECLAIM_ACCOUNT; |
1291523f | 1280 | |
cb5d9fb3 PL |
1281 | kmalloc_caches[type][idx] = create_kmalloc_cache( |
1282 | kmalloc_info[idx].name[type], | |
6c0c21ad DW |
1283 | kmalloc_info[idx].size, flags, 0, |
1284 | kmalloc_info[idx].size); | |
a9730fca CL |
1285 | } |
1286 | ||
34cc6990 DS |
1287 | /* |
1288 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
1289 | * may already have been created because they were needed to | |
1290 | * enable allocations for slab creation. | |
1291 | */ | |
d50112ed | 1292 | void __init create_kmalloc_caches(slab_flags_t flags) |
34cc6990 | 1293 | { |
13657d0a PL |
1294 | int i; |
1295 | enum kmalloc_cache_type type; | |
34cc6990 | 1296 | |
1291523f VB |
1297 | for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { |
1298 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1299 | if (!kmalloc_caches[type][i]) | |
1300 | new_kmalloc_cache(i, type, flags); | |
f97d5f63 | 1301 | |
1291523f VB |
1302 | /* |
1303 | * Caches that are not of the two-to-the-power-of size. | |
1304 | * These have to be created immediately after the | |
1305 | * earlier power of two caches | |
1306 | */ | |
1307 | if (KMALLOC_MIN_SIZE <= 32 && i == 6 && | |
1308 | !kmalloc_caches[type][1]) | |
1309 | new_kmalloc_cache(1, type, flags); | |
1310 | if (KMALLOC_MIN_SIZE <= 64 && i == 7 && | |
1311 | !kmalloc_caches[type][2]) | |
1312 | new_kmalloc_cache(2, type, flags); | |
1313 | } | |
8a965b3b CL |
1314 | } |
1315 | ||
f97d5f63 CL |
1316 | /* Kmalloc array is now usable */ |
1317 | slab_state = UP; | |
1318 | ||
f97d5f63 CL |
1319 | #ifdef CONFIG_ZONE_DMA |
1320 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
cc252eae | 1321 | struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; |
f97d5f63 CL |
1322 | |
1323 | if (s) { | |
cc252eae | 1324 | kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( |
cb5d9fb3 | 1325 | kmalloc_info[i].name[KMALLOC_DMA], |
dc0a7f75 | 1326 | kmalloc_info[i].size, |
49f2d241 VB |
1327 | SLAB_CACHE_DMA | flags, 0, |
1328 | kmalloc_info[i].size); | |
f97d5f63 CL |
1329 | } |
1330 | } | |
1331 | #endif | |
1332 | } | |
45530c44 CL |
1333 | #endif /* !CONFIG_SLOB */ |
1334 | ||
cea371f4 VD |
1335 | /* |
1336 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1337 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1338 | * know the allocation order to free the pages properly in kfree. | |
1339 | */ | |
52383431 VD |
1340 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
1341 | { | |
6a486c0a | 1342 | void *ret = NULL; |
52383431 VD |
1343 | struct page *page; |
1344 | ||
1345 | flags |= __GFP_COMP; | |
4949148a | 1346 | page = alloc_pages(flags, order); |
6a486c0a VB |
1347 | if (likely(page)) { |
1348 | ret = page_address(page); | |
1349 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE, | |
1350 | 1 << order); | |
1351 | } | |
0116523c | 1352 | ret = kasan_kmalloc_large(ret, size, flags); |
a2f77575 | 1353 | /* As ret might get tagged, call kmemleak hook after KASAN. */ |
53128245 | 1354 | kmemleak_alloc(ret, size, 1, flags); |
52383431 VD |
1355 | return ret; |
1356 | } | |
1357 | EXPORT_SYMBOL(kmalloc_order); | |
1358 | ||
f1b6eb6e CL |
1359 | #ifdef CONFIG_TRACING |
1360 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1361 | { | |
1362 | void *ret = kmalloc_order(size, flags, order); | |
1363 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1364 | return ret; | |
1365 | } | |
1366 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1367 | #endif | |
45530c44 | 1368 | |
7c00fce9 TG |
1369 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1370 | /* Randomize a generic freelist */ | |
1371 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
302d55d5 | 1372 | unsigned int count) |
7c00fce9 | 1373 | { |
7c00fce9 | 1374 | unsigned int rand; |
302d55d5 | 1375 | unsigned int i; |
7c00fce9 TG |
1376 | |
1377 | for (i = 0; i < count; i++) | |
1378 | list[i] = i; | |
1379 | ||
1380 | /* Fisher-Yates shuffle */ | |
1381 | for (i = count - 1; i > 0; i--) { | |
1382 | rand = prandom_u32_state(state); | |
1383 | rand %= (i + 1); | |
1384 | swap(list[i], list[rand]); | |
1385 | } | |
1386 | } | |
1387 | ||
1388 | /* Create a random sequence per cache */ | |
1389 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1390 | gfp_t gfp) | |
1391 | { | |
1392 | struct rnd_state state; | |
1393 | ||
1394 | if (count < 2 || cachep->random_seq) | |
1395 | return 0; | |
1396 | ||
1397 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1398 | if (!cachep->random_seq) | |
1399 | return -ENOMEM; | |
1400 | ||
1401 | /* Get best entropy at this stage of boot */ | |
1402 | prandom_seed_state(&state, get_random_long()); | |
1403 | ||
1404 | freelist_randomize(&state, cachep->random_seq, count); | |
1405 | return 0; | |
1406 | } | |
1407 | ||
1408 | /* Destroy the per-cache random freelist sequence */ | |
1409 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1410 | { | |
1411 | kfree(cachep->random_seq); | |
1412 | cachep->random_seq = NULL; | |
1413 | } | |
1414 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1415 | ||
5b365771 | 1416 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
e9b4db2b | 1417 | #ifdef CONFIG_SLAB |
0825a6f9 | 1418 | #define SLABINFO_RIGHTS (0600) |
e9b4db2b | 1419 | #else |
0825a6f9 | 1420 | #define SLABINFO_RIGHTS (0400) |
e9b4db2b WL |
1421 | #endif |
1422 | ||
b047501c | 1423 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1424 | { |
1425 | /* | |
1426 | * Output format version, so at least we can change it | |
1427 | * without _too_ many complaints. | |
1428 | */ | |
1429 | #ifdef CONFIG_DEBUG_SLAB | |
1430 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1431 | #else | |
1432 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1433 | #endif | |
756a025f | 1434 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1435 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1436 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1437 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1438 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1439 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1440 | #endif | |
1441 | seq_putc(m, '\n'); | |
1442 | } | |
1443 | ||
1df3b26f | 1444 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1445 | { |
b7454ad3 | 1446 | mutex_lock(&slab_mutex); |
510ded33 | 1447 | return seq_list_start(&slab_root_caches, *pos); |
b7454ad3 GC |
1448 | } |
1449 | ||
276a2439 | 1450 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 | 1451 | { |
510ded33 | 1452 | return seq_list_next(p, &slab_root_caches, pos); |
b7454ad3 GC |
1453 | } |
1454 | ||
276a2439 | 1455 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1456 | { |
1457 | mutex_unlock(&slab_mutex); | |
1458 | } | |
1459 | ||
749c5415 GC |
1460 | static void |
1461 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1462 | { | |
1463 | struct kmem_cache *c; | |
1464 | struct slabinfo sinfo; | |
749c5415 GC |
1465 | |
1466 | if (!is_root_cache(s)) | |
1467 | return; | |
1468 | ||
426589f5 | 1469 | for_each_memcg_cache(c, s) { |
749c5415 GC |
1470 | memset(&sinfo, 0, sizeof(sinfo)); |
1471 | get_slabinfo(c, &sinfo); | |
1472 | ||
1473 | info->active_slabs += sinfo.active_slabs; | |
1474 | info->num_slabs += sinfo.num_slabs; | |
1475 | info->shared_avail += sinfo.shared_avail; | |
1476 | info->active_objs += sinfo.active_objs; | |
1477 | info->num_objs += sinfo.num_objs; | |
1478 | } | |
1479 | } | |
1480 | ||
b047501c | 1481 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1482 | { |
0d7561c6 GC |
1483 | struct slabinfo sinfo; |
1484 | ||
1485 | memset(&sinfo, 0, sizeof(sinfo)); | |
1486 | get_slabinfo(s, &sinfo); | |
1487 | ||
749c5415 GC |
1488 | memcg_accumulate_slabinfo(s, &sinfo); |
1489 | ||
0d7561c6 | 1490 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 1491 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1492 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1493 | ||
1494 | seq_printf(m, " : tunables %4u %4u %4u", | |
1495 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1496 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1497 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1498 | slabinfo_show_stats(m, s); | |
1499 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1500 | } |
1501 | ||
1df3b26f | 1502 | static int slab_show(struct seq_file *m, void *p) |
749c5415 | 1503 | { |
510ded33 | 1504 | struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node); |
749c5415 | 1505 | |
510ded33 | 1506 | if (p == slab_root_caches.next) |
1df3b26f | 1507 | print_slabinfo_header(m); |
510ded33 | 1508 | cache_show(s, m); |
b047501c VD |
1509 | return 0; |
1510 | } | |
1511 | ||
852d8be0 YS |
1512 | void dump_unreclaimable_slab(void) |
1513 | { | |
1514 | struct kmem_cache *s, *s2; | |
1515 | struct slabinfo sinfo; | |
1516 | ||
1517 | /* | |
1518 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1519 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1520 | * risk of crash. | |
1521 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1522 | * without acquiring the mutex. | |
1523 | */ | |
1524 | if (!mutex_trylock(&slab_mutex)) { | |
1525 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1526 | return; | |
1527 | } | |
1528 | ||
1529 | pr_info("Unreclaimable slab info:\n"); | |
1530 | pr_info("Name Used Total\n"); | |
1531 | ||
1532 | list_for_each_entry_safe(s, s2, &slab_caches, list) { | |
1533 | if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT)) | |
1534 | continue; | |
1535 | ||
1536 | get_slabinfo(s, &sinfo); | |
1537 | ||
1538 | if (sinfo.num_objs > 0) | |
1539 | pr_info("%-17s %10luKB %10luKB\n", cache_name(s), | |
1540 | (sinfo.active_objs * s->size) / 1024, | |
1541 | (sinfo.num_objs * s->size) / 1024); | |
1542 | } | |
1543 | mutex_unlock(&slab_mutex); | |
1544 | } | |
1545 | ||
a87425a3 | 1546 | #if defined(CONFIG_MEMCG_KMEM) |
bc2791f8 TH |
1547 | void *memcg_slab_start(struct seq_file *m, loff_t *pos) |
1548 | { | |
aa9694bb | 1549 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
bc2791f8 TH |
1550 | |
1551 | mutex_lock(&slab_mutex); | |
1552 | return seq_list_start(&memcg->kmem_caches, *pos); | |
1553 | } | |
1554 | ||
1555 | void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1556 | { | |
aa9694bb | 1557 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
bc2791f8 TH |
1558 | |
1559 | return seq_list_next(p, &memcg->kmem_caches, pos); | |
1560 | } | |
1561 | ||
1562 | void memcg_slab_stop(struct seq_file *m, void *p) | |
1563 | { | |
1564 | mutex_unlock(&slab_mutex); | |
1565 | } | |
1566 | ||
b047501c VD |
1567 | int memcg_slab_show(struct seq_file *m, void *p) |
1568 | { | |
bc2791f8 TH |
1569 | struct kmem_cache *s = list_entry(p, struct kmem_cache, |
1570 | memcg_params.kmem_caches_node); | |
aa9694bb | 1571 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
b047501c | 1572 | |
bc2791f8 | 1573 | if (p == memcg->kmem_caches.next) |
b047501c | 1574 | print_slabinfo_header(m); |
bc2791f8 | 1575 | cache_show(s, m); |
b047501c | 1576 | return 0; |
749c5415 | 1577 | } |
b047501c | 1578 | #endif |
749c5415 | 1579 | |
b7454ad3 GC |
1580 | /* |
1581 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1582 | * | |
1583 | * Output layout: | |
1584 | * cache-name | |
1585 | * num-active-objs | |
1586 | * total-objs | |
1587 | * object size | |
1588 | * num-active-slabs | |
1589 | * total-slabs | |
1590 | * num-pages-per-slab | |
1591 | * + further values on SMP and with statistics enabled | |
1592 | */ | |
1593 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1594 | .start = slab_start, |
276a2439 WL |
1595 | .next = slab_next, |
1596 | .stop = slab_stop, | |
1df3b26f | 1597 | .show = slab_show, |
b7454ad3 GC |
1598 | }; |
1599 | ||
1600 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1601 | { | |
1602 | return seq_open(file, &slabinfo_op); | |
1603 | } | |
1604 | ||
97a32539 | 1605 | static const struct proc_ops slabinfo_proc_ops = { |
d919b33d | 1606 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
1607 | .proc_open = slabinfo_open, |
1608 | .proc_read = seq_read, | |
1609 | .proc_write = slabinfo_write, | |
1610 | .proc_lseek = seq_lseek, | |
1611 | .proc_release = seq_release, | |
b7454ad3 GC |
1612 | }; |
1613 | ||
1614 | static int __init slab_proc_init(void) | |
1615 | { | |
97a32539 | 1616 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); |
b7454ad3 GC |
1617 | return 0; |
1618 | } | |
1619 | module_init(slab_proc_init); | |
fcf8a1e4 WL |
1620 | |
1621 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_MEMCG_KMEM) | |
1622 | /* | |
1623 | * Display information about kmem caches that have child memcg caches. | |
1624 | */ | |
1625 | static int memcg_slabinfo_show(struct seq_file *m, void *unused) | |
1626 | { | |
1627 | struct kmem_cache *s, *c; | |
1628 | struct slabinfo sinfo; | |
1629 | ||
1630 | mutex_lock(&slab_mutex); | |
1631 | seq_puts(m, "# <name> <css_id[:dead|deact]> <active_objs> <num_objs>"); | |
1632 | seq_puts(m, " <active_slabs> <num_slabs>\n"); | |
1633 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { | |
1634 | /* | |
1635 | * Skip kmem caches that don't have any memcg children. | |
1636 | */ | |
1637 | if (list_empty(&s->memcg_params.children)) | |
1638 | continue; | |
1639 | ||
1640 | memset(&sinfo, 0, sizeof(sinfo)); | |
1641 | get_slabinfo(s, &sinfo); | |
1642 | seq_printf(m, "%-17s root %6lu %6lu %6lu %6lu\n", | |
1643 | cache_name(s), sinfo.active_objs, sinfo.num_objs, | |
1644 | sinfo.active_slabs, sinfo.num_slabs); | |
1645 | ||
1646 | for_each_memcg_cache(c, s) { | |
1647 | struct cgroup_subsys_state *css; | |
1648 | char *status = ""; | |
1649 | ||
1650 | css = &c->memcg_params.memcg->css; | |
1651 | if (!(css->flags & CSS_ONLINE)) | |
1652 | status = ":dead"; | |
1653 | else if (c->flags & SLAB_DEACTIVATED) | |
1654 | status = ":deact"; | |
1655 | ||
1656 | memset(&sinfo, 0, sizeof(sinfo)); | |
1657 | get_slabinfo(c, &sinfo); | |
1658 | seq_printf(m, "%-17s %4d%-6s %6lu %6lu %6lu %6lu\n", | |
1659 | cache_name(c), css->id, status, | |
1660 | sinfo.active_objs, sinfo.num_objs, | |
1661 | sinfo.active_slabs, sinfo.num_slabs); | |
1662 | } | |
1663 | } | |
1664 | mutex_unlock(&slab_mutex); | |
1665 | return 0; | |
1666 | } | |
1667 | DEFINE_SHOW_ATTRIBUTE(memcg_slabinfo); | |
1668 | ||
1669 | static int __init memcg_slabinfo_init(void) | |
1670 | { | |
1671 | debugfs_create_file("memcg_slabinfo", S_IFREG | S_IRUGO, | |
1672 | NULL, NULL, &memcg_slabinfo_fops); | |
1673 | return 0; | |
1674 | } | |
1675 | ||
1676 | late_initcall(memcg_slabinfo_init); | |
1677 | #endif /* CONFIG_DEBUG_FS && CONFIG_MEMCG_KMEM */ | |
5b365771 | 1678 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ |
928cec9c AR |
1679 | |
1680 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1681 | gfp_t flags) | |
1682 | { | |
1683 | void *ret; | |
1684 | size_t ks = 0; | |
1685 | ||
1686 | if (p) | |
1687 | ks = ksize(p); | |
1688 | ||
0316bec2 | 1689 | if (ks >= new_size) { |
0116523c | 1690 | p = kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1691 | return (void *)p; |
0316bec2 | 1692 | } |
928cec9c AR |
1693 | |
1694 | ret = kmalloc_track_caller(new_size, flags); | |
1695 | if (ret && p) | |
1696 | memcpy(ret, p, ks); | |
1697 | ||
1698 | return ret; | |
1699 | } | |
1700 | ||
928cec9c AR |
1701 | /** |
1702 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1703 | * @p: object to reallocate memory for. | |
1704 | * @new_size: how many bytes of memory are required. | |
1705 | * @flags: the type of memory to allocate. | |
1706 | * | |
1707 | * The contents of the object pointed to are preserved up to the | |
1708 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1709 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1710 | * %NULL pointer, the object pointed to is freed. | |
a862f68a MR |
1711 | * |
1712 | * Return: pointer to the allocated memory or %NULL in case of error | |
928cec9c AR |
1713 | */ |
1714 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1715 | { | |
1716 | void *ret; | |
1717 | ||
1718 | if (unlikely(!new_size)) { | |
1719 | kfree(p); | |
1720 | return ZERO_SIZE_PTR; | |
1721 | } | |
1722 | ||
1723 | ret = __do_krealloc(p, new_size, flags); | |
772a2fa5 | 1724 | if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) |
928cec9c AR |
1725 | kfree(p); |
1726 | ||
1727 | return ret; | |
1728 | } | |
1729 | EXPORT_SYMBOL(krealloc); | |
1730 | ||
1731 | /** | |
1732 | * kzfree - like kfree but zero memory | |
1733 | * @p: object to free memory of | |
1734 | * | |
1735 | * The memory of the object @p points to is zeroed before freed. | |
1736 | * If @p is %NULL, kzfree() does nothing. | |
1737 | * | |
1738 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1739 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1740 | * careful when using this function in performance sensitive code. | |
1741 | */ | |
1742 | void kzfree(const void *p) | |
1743 | { | |
1744 | size_t ks; | |
1745 | void *mem = (void *)p; | |
1746 | ||
1747 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1748 | return; | |
1749 | ks = ksize(mem); | |
8982ae52 | 1750 | memzero_explicit(mem, ks); |
928cec9c AR |
1751 | kfree(mem); |
1752 | } | |
1753 | EXPORT_SYMBOL(kzfree); | |
1754 | ||
10d1f8cb ME |
1755 | /** |
1756 | * ksize - get the actual amount of memory allocated for a given object | |
1757 | * @objp: Pointer to the object | |
1758 | * | |
1759 | * kmalloc may internally round up allocations and return more memory | |
1760 | * than requested. ksize() can be used to determine the actual amount of | |
1761 | * memory allocated. The caller may use this additional memory, even though | |
1762 | * a smaller amount of memory was initially specified with the kmalloc call. | |
1763 | * The caller must guarantee that objp points to a valid object previously | |
1764 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
1765 | * must not be freed during the duration of the call. | |
1766 | * | |
1767 | * Return: size of the actual memory used by @objp in bytes | |
1768 | */ | |
1769 | size_t ksize(const void *objp) | |
1770 | { | |
0d4ca4c9 ME |
1771 | size_t size; |
1772 | ||
1773 | if (WARN_ON_ONCE(!objp)) | |
1774 | return 0; | |
1775 | /* | |
1776 | * We need to check that the pointed to object is valid, and only then | |
1777 | * unpoison the shadow memory below. We use __kasan_check_read(), to | |
1778 | * generate a more useful report at the time ksize() is called (rather | |
1779 | * than later where behaviour is undefined due to potential | |
1780 | * use-after-free or double-free). | |
1781 | * | |
1782 | * If the pointed to memory is invalid we return 0, to avoid users of | |
1783 | * ksize() writing to and potentially corrupting the memory region. | |
1784 | * | |
1785 | * We want to perform the check before __ksize(), to avoid potentially | |
1786 | * crashing in __ksize() due to accessing invalid metadata. | |
1787 | */ | |
1788 | if (unlikely(objp == ZERO_SIZE_PTR) || !__kasan_check_read(objp, 1)) | |
1789 | return 0; | |
1790 | ||
1791 | size = __ksize(objp); | |
10d1f8cb ME |
1792 | /* |
1793 | * We assume that ksize callers could use whole allocated area, | |
1794 | * so we need to unpoison this area. | |
1795 | */ | |
1796 | kasan_unpoison_shadow(objp, size); | |
1797 | return size; | |
1798 | } | |
1799 | EXPORT_SYMBOL(ksize); | |
1800 | ||
928cec9c AR |
1801 | /* Tracepoints definitions. */ |
1802 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1803 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1804 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1805 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1806 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1807 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); | |
4f6923fb HM |
1808 | |
1809 | int should_failslab(struct kmem_cache *s, gfp_t gfpflags) | |
1810 | { | |
1811 | if (__should_failslab(s, gfpflags)) | |
1812 | return -ENOMEM; | |
1813 | return 0; | |
1814 | } | |
1815 | ALLOW_ERROR_INJECTION(should_failslab, ERRNO); |