]>
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> | |
039363f3 CL |
20 | #include <asm/cacheflush.h> |
21 | #include <asm/tlbflush.h> | |
22 | #include <asm/page.h> | |
2633d7a0 | 23 | #include <linux/memcontrol.h> |
928cec9c AR |
24 | |
25 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 26 | #include <trace/events/kmem.h> |
039363f3 | 27 | |
97d06609 CL |
28 | #include "slab.h" |
29 | ||
30 | enum slab_state slab_state; | |
18004c5d CL |
31 | LIST_HEAD(slab_caches); |
32 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 33 | struct kmem_cache *kmem_cache; |
97d06609 | 34 | |
2d891fbc KC |
35 | #ifdef CONFIG_HARDENED_USERCOPY |
36 | bool usercopy_fallback __ro_after_init = | |
37 | IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK); | |
38 | module_param(usercopy_fallback, bool, 0400); | |
39 | MODULE_PARM_DESC(usercopy_fallback, | |
40 | "WARN instead of reject usercopy whitelist violations"); | |
41 | #endif | |
42 | ||
657dc2f9 TH |
43 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
44 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
45 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
46 | slab_caches_to_rcu_destroy_workfn); | |
47 | ||
423c929c JK |
48 | /* |
49 | * Set of flags that will prevent slab merging | |
50 | */ | |
51 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
5f0d5a3a | 52 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ |
7ed2f9e6 | 53 | SLAB_FAILSLAB | SLAB_KASAN) |
423c929c | 54 | |
230e9fc2 | 55 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
6d6ea1e9 | 56 | SLAB_CACHE_DMA32 | SLAB_ACCOUNT) |
423c929c JK |
57 | |
58 | /* | |
59 | * Merge control. If this is set then no merging of slab caches will occur. | |
423c929c | 60 | */ |
7660a6fd | 61 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); |
423c929c JK |
62 | |
63 | static int __init setup_slab_nomerge(char *str) | |
64 | { | |
7660a6fd | 65 | slab_nomerge = true; |
423c929c JK |
66 | return 1; |
67 | } | |
68 | ||
69 | #ifdef CONFIG_SLUB | |
70 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
71 | #endif | |
72 | ||
73 | __setup("slab_nomerge", setup_slab_nomerge); | |
74 | ||
07f361b2 JK |
75 | /* |
76 | * Determine the size of a slab object | |
77 | */ | |
78 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
79 | { | |
80 | return s->object_size; | |
81 | } | |
82 | EXPORT_SYMBOL(kmem_cache_size); | |
83 | ||
77be4b13 | 84 | #ifdef CONFIG_DEBUG_VM |
f4957d5b | 85 | static int kmem_cache_sanity_check(const char *name, unsigned int size) |
039363f3 | 86 | { |
039363f3 CL |
87 | if (!name || in_interrupt() || size < sizeof(void *) || |
88 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
89 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
90 | return -EINVAL; | |
039363f3 | 91 | } |
b920536a | 92 | |
20cea968 | 93 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
77be4b13 SK |
94 | return 0; |
95 | } | |
96 | #else | |
f4957d5b | 97 | static inline int kmem_cache_sanity_check(const char *name, unsigned int size) |
77be4b13 SK |
98 | { |
99 | return 0; | |
100 | } | |
20cea968 CL |
101 | #endif |
102 | ||
484748f0 CL |
103 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
104 | { | |
105 | size_t i; | |
106 | ||
ca257195 JDB |
107 | for (i = 0; i < nr; i++) { |
108 | if (s) | |
109 | kmem_cache_free(s, p[i]); | |
110 | else | |
111 | kfree(p[i]); | |
112 | } | |
484748f0 CL |
113 | } |
114 | ||
865762a8 | 115 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
116 | void **p) |
117 | { | |
118 | size_t i; | |
119 | ||
120 | for (i = 0; i < nr; i++) { | |
121 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
122 | if (!x) { | |
123 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 124 | return 0; |
484748f0 CL |
125 | } |
126 | } | |
865762a8 | 127 | return i; |
484748f0 CL |
128 | } |
129 | ||
84c07d11 | 130 | #ifdef CONFIG_MEMCG_KMEM |
510ded33 TH |
131 | |
132 | LIST_HEAD(slab_root_caches); | |
63b02ef7 | 133 | static DEFINE_SPINLOCK(memcg_kmem_wq_lock); |
510ded33 | 134 | |
f0a3a24b RG |
135 | static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref); |
136 | ||
f7ce3190 | 137 | void slab_init_memcg_params(struct kmem_cache *s) |
33a690c4 | 138 | { |
9eeadc8b | 139 | s->memcg_params.root_cache = NULL; |
f7ce3190 | 140 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
9eeadc8b | 141 | INIT_LIST_HEAD(&s->memcg_params.children); |
92ee383f | 142 | s->memcg_params.dying = false; |
f7ce3190 VD |
143 | } |
144 | ||
145 | static int init_memcg_params(struct kmem_cache *s, | |
c03914b7 | 146 | struct kmem_cache *root_cache) |
f7ce3190 VD |
147 | { |
148 | struct memcg_cache_array *arr; | |
33a690c4 | 149 | |
9eeadc8b | 150 | if (root_cache) { |
f0a3a24b RG |
151 | int ret = percpu_ref_init(&s->memcg_params.refcnt, |
152 | kmemcg_cache_shutdown, | |
153 | 0, GFP_KERNEL); | |
154 | if (ret) | |
155 | return ret; | |
156 | ||
f7ce3190 | 157 | s->memcg_params.root_cache = root_cache; |
9eeadc8b | 158 | INIT_LIST_HEAD(&s->memcg_params.children_node); |
bc2791f8 | 159 | INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node); |
33a690c4 | 160 | return 0; |
f7ce3190 | 161 | } |
33a690c4 | 162 | |
f7ce3190 | 163 | slab_init_memcg_params(s); |
33a690c4 | 164 | |
f7ce3190 VD |
165 | if (!memcg_nr_cache_ids) |
166 | return 0; | |
33a690c4 | 167 | |
f80c7dab JW |
168 | arr = kvzalloc(sizeof(struct memcg_cache_array) + |
169 | memcg_nr_cache_ids * sizeof(void *), | |
170 | GFP_KERNEL); | |
f7ce3190 VD |
171 | if (!arr) |
172 | return -ENOMEM; | |
33a690c4 | 173 | |
f7ce3190 | 174 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); |
33a690c4 VD |
175 | return 0; |
176 | } | |
177 | ||
f7ce3190 | 178 | static void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 | 179 | { |
f7ce3190 | 180 | if (is_root_cache(s)) |
f80c7dab | 181 | kvfree(rcu_access_pointer(s->memcg_params.memcg_caches)); |
f0a3a24b RG |
182 | else |
183 | percpu_ref_exit(&s->memcg_params.refcnt); | |
f80c7dab JW |
184 | } |
185 | ||
186 | static void free_memcg_params(struct rcu_head *rcu) | |
187 | { | |
188 | struct memcg_cache_array *old; | |
189 | ||
190 | old = container_of(rcu, struct memcg_cache_array, rcu); | |
191 | kvfree(old); | |
33a690c4 VD |
192 | } |
193 | ||
f7ce3190 | 194 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
6f817f4c | 195 | { |
f7ce3190 | 196 | struct memcg_cache_array *old, *new; |
6f817f4c | 197 | |
f80c7dab JW |
198 | new = kvzalloc(sizeof(struct memcg_cache_array) + |
199 | new_array_size * sizeof(void *), GFP_KERNEL); | |
f7ce3190 | 200 | if (!new) |
6f817f4c VD |
201 | return -ENOMEM; |
202 | ||
f7ce3190 VD |
203 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
204 | lockdep_is_held(&slab_mutex)); | |
205 | if (old) | |
206 | memcpy(new->entries, old->entries, | |
207 | memcg_nr_cache_ids * sizeof(void *)); | |
6f817f4c | 208 | |
f7ce3190 VD |
209 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
210 | if (old) | |
f80c7dab | 211 | call_rcu(&old->rcu, free_memcg_params); |
6f817f4c VD |
212 | return 0; |
213 | } | |
214 | ||
55007d84 GC |
215 | int memcg_update_all_caches(int num_memcgs) |
216 | { | |
217 | struct kmem_cache *s; | |
218 | int ret = 0; | |
55007d84 | 219 | |
05257a1a | 220 | mutex_lock(&slab_mutex); |
510ded33 | 221 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
f7ce3190 | 222 | ret = update_memcg_params(s, num_memcgs); |
55007d84 | 223 | /* |
55007d84 GC |
224 | * Instead of freeing the memory, we'll just leave the caches |
225 | * up to this point in an updated state. | |
226 | */ | |
227 | if (ret) | |
05257a1a | 228 | break; |
55007d84 | 229 | } |
55007d84 GC |
230 | mutex_unlock(&slab_mutex); |
231 | return ret; | |
232 | } | |
657dc2f9 | 233 | |
c03914b7 | 234 | void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg) |
657dc2f9 | 235 | { |
510ded33 TH |
236 | if (is_root_cache(s)) { |
237 | list_add(&s->root_caches_node, &slab_root_caches); | |
238 | } else { | |
f0a3a24b | 239 | css_get(&memcg->css); |
c03914b7 | 240 | s->memcg_params.memcg = memcg; |
510ded33 TH |
241 | list_add(&s->memcg_params.children_node, |
242 | &s->memcg_params.root_cache->memcg_params.children); | |
243 | list_add(&s->memcg_params.kmem_caches_node, | |
244 | &s->memcg_params.memcg->kmem_caches); | |
245 | } | |
246 | } | |
247 | ||
248 | static void memcg_unlink_cache(struct kmem_cache *s) | |
249 | { | |
250 | if (is_root_cache(s)) { | |
251 | list_del(&s->root_caches_node); | |
252 | } else { | |
253 | list_del(&s->memcg_params.children_node); | |
254 | list_del(&s->memcg_params.kmem_caches_node); | |
f0a3a24b | 255 | css_put(&s->memcg_params.memcg->css); |
510ded33 | 256 | } |
657dc2f9 | 257 | } |
33a690c4 | 258 | #else |
f7ce3190 | 259 | static inline int init_memcg_params(struct kmem_cache *s, |
c03914b7 | 260 | struct kmem_cache *root_cache) |
33a690c4 VD |
261 | { |
262 | return 0; | |
263 | } | |
264 | ||
f7ce3190 | 265 | static inline void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 VD |
266 | { |
267 | } | |
657dc2f9 | 268 | |
510ded33 | 269 | static inline void memcg_unlink_cache(struct kmem_cache *s) |
657dc2f9 TH |
270 | { |
271 | } | |
84c07d11 | 272 | #endif /* CONFIG_MEMCG_KMEM */ |
55007d84 | 273 | |
692ae74a BL |
274 | /* |
275 | * Figure out what the alignment of the objects will be given a set of | |
276 | * flags, a user specified alignment and the size of the objects. | |
277 | */ | |
f4957d5b AD |
278 | static unsigned int calculate_alignment(slab_flags_t flags, |
279 | unsigned int align, unsigned int size) | |
692ae74a BL |
280 | { |
281 | /* | |
282 | * If the user wants hardware cache aligned objects then follow that | |
283 | * suggestion if the object is sufficiently large. | |
284 | * | |
285 | * The hardware cache alignment cannot override the specified | |
286 | * alignment though. If that is greater then use it. | |
287 | */ | |
288 | if (flags & SLAB_HWCACHE_ALIGN) { | |
f4957d5b | 289 | unsigned int ralign; |
692ae74a BL |
290 | |
291 | ralign = cache_line_size(); | |
292 | while (size <= ralign / 2) | |
293 | ralign /= 2; | |
294 | align = max(align, ralign); | |
295 | } | |
296 | ||
297 | if (align < ARCH_SLAB_MINALIGN) | |
298 | align = ARCH_SLAB_MINALIGN; | |
299 | ||
300 | return ALIGN(align, sizeof(void *)); | |
301 | } | |
302 | ||
423c929c JK |
303 | /* |
304 | * Find a mergeable slab cache | |
305 | */ | |
306 | int slab_unmergeable(struct kmem_cache *s) | |
307 | { | |
308 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
309 | return 1; | |
310 | ||
311 | if (!is_root_cache(s)) | |
312 | return 1; | |
313 | ||
314 | if (s->ctor) | |
315 | return 1; | |
316 | ||
8eb8284b DW |
317 | if (s->usersize) |
318 | return 1; | |
319 | ||
423c929c JK |
320 | /* |
321 | * We may have set a slab to be unmergeable during bootstrap. | |
322 | */ | |
323 | if (s->refcount < 0) | |
324 | return 1; | |
325 | ||
326 | return 0; | |
327 | } | |
328 | ||
f4957d5b | 329 | struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, |
d50112ed | 330 | slab_flags_t flags, const char *name, void (*ctor)(void *)) |
423c929c JK |
331 | { |
332 | struct kmem_cache *s; | |
333 | ||
c6e28895 | 334 | if (slab_nomerge) |
423c929c JK |
335 | return NULL; |
336 | ||
337 | if (ctor) | |
338 | return NULL; | |
339 | ||
340 | size = ALIGN(size, sizeof(void *)); | |
341 | align = calculate_alignment(flags, align, size); | |
342 | size = ALIGN(size, align); | |
343 | flags = kmem_cache_flags(size, flags, name, NULL); | |
344 | ||
c6e28895 GM |
345 | if (flags & SLAB_NEVER_MERGE) |
346 | return NULL; | |
347 | ||
510ded33 | 348 | list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) { |
423c929c JK |
349 | if (slab_unmergeable(s)) |
350 | continue; | |
351 | ||
352 | if (size > s->size) | |
353 | continue; | |
354 | ||
355 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
356 | continue; | |
357 | /* | |
358 | * Check if alignment is compatible. | |
359 | * Courtesy of Adrian Drzewiecki | |
360 | */ | |
361 | if ((s->size & ~(align - 1)) != s->size) | |
362 | continue; | |
363 | ||
364 | if (s->size - size >= sizeof(void *)) | |
365 | continue; | |
366 | ||
95069ac8 JK |
367 | if (IS_ENABLED(CONFIG_SLAB) && align && |
368 | (align > s->align || s->align % align)) | |
369 | continue; | |
370 | ||
423c929c JK |
371 | return s; |
372 | } | |
373 | return NULL; | |
374 | } | |
375 | ||
c9a77a79 | 376 | static struct kmem_cache *create_cache(const char *name, |
613a5eb5 | 377 | unsigned int object_size, unsigned int align, |
7bbdb81e AD |
378 | slab_flags_t flags, unsigned int useroffset, |
379 | unsigned int usersize, void (*ctor)(void *), | |
c9a77a79 | 380 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
794b1248 VD |
381 | { |
382 | struct kmem_cache *s; | |
383 | int err; | |
384 | ||
8eb8284b DW |
385 | if (WARN_ON(useroffset + usersize > object_size)) |
386 | useroffset = usersize = 0; | |
387 | ||
794b1248 VD |
388 | err = -ENOMEM; |
389 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
390 | if (!s) | |
391 | goto out; | |
392 | ||
393 | s->name = name; | |
613a5eb5 | 394 | s->size = s->object_size = object_size; |
794b1248 VD |
395 | s->align = align; |
396 | s->ctor = ctor; | |
8eb8284b DW |
397 | s->useroffset = useroffset; |
398 | s->usersize = usersize; | |
794b1248 | 399 | |
c03914b7 | 400 | err = init_memcg_params(s, root_cache); |
794b1248 VD |
401 | if (err) |
402 | goto out_free_cache; | |
403 | ||
404 | err = __kmem_cache_create(s, flags); | |
405 | if (err) | |
406 | goto out_free_cache; | |
407 | ||
408 | s->refcount = 1; | |
409 | list_add(&s->list, &slab_caches); | |
c03914b7 | 410 | memcg_link_cache(s, memcg); |
794b1248 VD |
411 | out: |
412 | if (err) | |
413 | return ERR_PTR(err); | |
414 | return s; | |
415 | ||
416 | out_free_cache: | |
f7ce3190 | 417 | destroy_memcg_params(s); |
7c4da061 | 418 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
419 | goto out; |
420 | } | |
45906855 | 421 | |
f496990f MR |
422 | /** |
423 | * kmem_cache_create_usercopy - Create a cache with a region suitable | |
424 | * for copying to userspace | |
77be4b13 SK |
425 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
426 | * @size: The size of objects to be created in this cache. | |
427 | * @align: The required alignment for the objects. | |
428 | * @flags: SLAB flags | |
8eb8284b DW |
429 | * @useroffset: Usercopy region offset |
430 | * @usersize: Usercopy region size | |
77be4b13 SK |
431 | * @ctor: A constructor for the objects. |
432 | * | |
77be4b13 SK |
433 | * Cannot be called within a interrupt, but can be interrupted. |
434 | * The @ctor is run when new pages are allocated by the cache. | |
435 | * | |
436 | * The flags are | |
437 | * | |
438 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
439 | * to catch references to uninitialised memory. | |
440 | * | |
f496990f | 441 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check |
77be4b13 SK |
442 | * for buffer overruns. |
443 | * | |
444 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
445 | * cacheline. This can be beneficial if you're counting cycles as closely | |
446 | * as davem. | |
f496990f MR |
447 | * |
448 | * Return: a pointer to the cache on success, NULL on failure. | |
77be4b13 | 449 | */ |
2633d7a0 | 450 | struct kmem_cache * |
f4957d5b AD |
451 | kmem_cache_create_usercopy(const char *name, |
452 | unsigned int size, unsigned int align, | |
7bbdb81e AD |
453 | slab_flags_t flags, |
454 | unsigned int useroffset, unsigned int usersize, | |
8eb8284b | 455 | void (*ctor)(void *)) |
77be4b13 | 456 | { |
40911a79 | 457 | struct kmem_cache *s = NULL; |
3dec16ea | 458 | const char *cache_name; |
3965fc36 | 459 | int err; |
039363f3 | 460 | |
77be4b13 | 461 | get_online_cpus(); |
03afc0e2 | 462 | get_online_mems(); |
05257a1a | 463 | memcg_get_cache_ids(); |
03afc0e2 | 464 | |
77be4b13 | 465 | mutex_lock(&slab_mutex); |
686d550d | 466 | |
794b1248 | 467 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 468 | if (err) { |
3965fc36 | 469 | goto out_unlock; |
3aa24f51 | 470 | } |
686d550d | 471 | |
e70954fd TG |
472 | /* Refuse requests with allocator specific flags */ |
473 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
474 | err = -EINVAL; | |
475 | goto out_unlock; | |
476 | } | |
477 | ||
d8843922 GC |
478 | /* |
479 | * Some allocators will constraint the set of valid flags to a subset | |
480 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
481 | * case, and we'll just provide them with a sanitized version of the | |
482 | * passed flags. | |
483 | */ | |
484 | flags &= CACHE_CREATE_MASK; | |
686d550d | 485 | |
8eb8284b DW |
486 | /* Fail closed on bad usersize of useroffset values. */ |
487 | if (WARN_ON(!usersize && useroffset) || | |
488 | WARN_ON(size < usersize || size - usersize < useroffset)) | |
489 | usersize = useroffset = 0; | |
490 | ||
491 | if (!usersize) | |
492 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
794b1248 | 493 | if (s) |
3965fc36 | 494 | goto out_unlock; |
2633d7a0 | 495 | |
3dec16ea | 496 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
497 | if (!cache_name) { |
498 | err = -ENOMEM; | |
499 | goto out_unlock; | |
500 | } | |
7c9adf5a | 501 | |
613a5eb5 | 502 | s = create_cache(cache_name, size, |
c9a77a79 | 503 | calculate_alignment(flags, align, size), |
8eb8284b | 504 | flags, useroffset, usersize, ctor, NULL, NULL); |
794b1248 VD |
505 | if (IS_ERR(s)) { |
506 | err = PTR_ERR(s); | |
3dec16ea | 507 | kfree_const(cache_name); |
794b1248 | 508 | } |
3965fc36 VD |
509 | |
510 | out_unlock: | |
20cea968 | 511 | mutex_unlock(&slab_mutex); |
03afc0e2 | 512 | |
05257a1a | 513 | memcg_put_cache_ids(); |
03afc0e2 | 514 | put_online_mems(); |
20cea968 CL |
515 | put_online_cpus(); |
516 | ||
ba3253c7 | 517 | if (err) { |
686d550d CL |
518 | if (flags & SLAB_PANIC) |
519 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
520 | name, err); | |
521 | else { | |
1170532b | 522 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
523 | name, err); |
524 | dump_stack(); | |
525 | } | |
686d550d CL |
526 | return NULL; |
527 | } | |
039363f3 CL |
528 | return s; |
529 | } | |
8eb8284b DW |
530 | EXPORT_SYMBOL(kmem_cache_create_usercopy); |
531 | ||
f496990f MR |
532 | /** |
533 | * kmem_cache_create - Create a cache. | |
534 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
535 | * @size: The size of objects to be created in this cache. | |
536 | * @align: The required alignment for the objects. | |
537 | * @flags: SLAB flags | |
538 | * @ctor: A constructor for the objects. | |
539 | * | |
540 | * Cannot be called within a interrupt, but can be interrupted. | |
541 | * The @ctor is run when new pages are allocated by the cache. | |
542 | * | |
543 | * The flags are | |
544 | * | |
545 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
546 | * to catch references to uninitialised memory. | |
547 | * | |
548 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check | |
549 | * for buffer overruns. | |
550 | * | |
551 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
552 | * cacheline. This can be beneficial if you're counting cycles as closely | |
553 | * as davem. | |
554 | * | |
555 | * Return: a pointer to the cache on success, NULL on failure. | |
556 | */ | |
8eb8284b | 557 | struct kmem_cache * |
f4957d5b | 558 | kmem_cache_create(const char *name, unsigned int size, unsigned int align, |
8eb8284b DW |
559 | slab_flags_t flags, void (*ctor)(void *)) |
560 | { | |
6d07d1cd | 561 | return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, |
8eb8284b DW |
562 | ctor); |
563 | } | |
794b1248 | 564 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 565 | |
657dc2f9 | 566 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 567 | { |
657dc2f9 TH |
568 | LIST_HEAD(to_destroy); |
569 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 570 | |
657dc2f9 | 571 | /* |
5f0d5a3a | 572 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
657dc2f9 TH |
573 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
574 | * through RCU and and the associated kmem_cache are dereferenced | |
575 | * while freeing the pages, so the kmem_caches should be freed only | |
576 | * after the pending RCU operations are finished. As rcu_barrier() | |
577 | * is a pretty slow operation, we batch all pending destructions | |
578 | * asynchronously. | |
579 | */ | |
580 | mutex_lock(&slab_mutex); | |
581 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
582 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 583 | |
657dc2f9 TH |
584 | if (list_empty(&to_destroy)) |
585 | return; | |
586 | ||
587 | rcu_barrier(); | |
588 | ||
589 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
590 | #ifdef SLAB_SUPPORTS_SYSFS | |
591 | sysfs_slab_release(s); | |
592 | #else | |
593 | slab_kmem_cache_release(s); | |
594 | #endif | |
595 | } | |
d5b3cf71 VD |
596 | } |
597 | ||
657dc2f9 | 598 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 599 | { |
f9fa1d91 GT |
600 | /* free asan quarantined objects */ |
601 | kasan_cache_shutdown(s); | |
602 | ||
657dc2f9 TH |
603 | if (__kmem_cache_shutdown(s) != 0) |
604 | return -EBUSY; | |
d5b3cf71 | 605 | |
510ded33 | 606 | memcg_unlink_cache(s); |
657dc2f9 | 607 | list_del(&s->list); |
d5b3cf71 | 608 | |
5f0d5a3a | 609 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
d50d82fa MP |
610 | #ifdef SLAB_SUPPORTS_SYSFS |
611 | sysfs_slab_unlink(s); | |
612 | #endif | |
657dc2f9 TH |
613 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
614 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
615 | } else { | |
d5b3cf71 | 616 | #ifdef SLAB_SUPPORTS_SYSFS |
d50d82fa | 617 | sysfs_slab_unlink(s); |
bf5eb3de | 618 | sysfs_slab_release(s); |
d5b3cf71 VD |
619 | #else |
620 | slab_kmem_cache_release(s); | |
621 | #endif | |
622 | } | |
657dc2f9 TH |
623 | |
624 | return 0; | |
d5b3cf71 VD |
625 | } |
626 | ||
84c07d11 | 627 | #ifdef CONFIG_MEMCG_KMEM |
794b1248 | 628 | /* |
776ed0f0 | 629 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
630 | * @memcg: The memory cgroup the new cache is for. |
631 | * @root_cache: The parent of the new cache. | |
632 | * | |
633 | * This function attempts to create a kmem cache that will serve allocation | |
634 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
635 | * from its parent. | |
636 | */ | |
d5b3cf71 VD |
637 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
638 | struct kmem_cache *root_cache) | |
2633d7a0 | 639 | { |
3e0350a3 | 640 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
33398cf2 | 641 | struct cgroup_subsys_state *css = &memcg->css; |
f7ce3190 | 642 | struct memcg_cache_array *arr; |
bd673145 | 643 | struct kmem_cache *s = NULL; |
794b1248 | 644 | char *cache_name; |
f7ce3190 | 645 | int idx; |
794b1248 VD |
646 | |
647 | get_online_cpus(); | |
03afc0e2 VD |
648 | get_online_mems(); |
649 | ||
794b1248 VD |
650 | mutex_lock(&slab_mutex); |
651 | ||
2a4db7eb | 652 | /* |
567e9ab2 | 653 | * The memory cgroup could have been offlined while the cache |
2a4db7eb VD |
654 | * creation work was pending. |
655 | */ | |
57033297 | 656 | if (memcg->kmem_state != KMEM_ONLINE) |
2a4db7eb VD |
657 | goto out_unlock; |
658 | ||
f7ce3190 VD |
659 | idx = memcg_cache_id(memcg); |
660 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
661 | lockdep_is_held(&slab_mutex)); | |
662 | ||
d5b3cf71 VD |
663 | /* |
664 | * Since per-memcg caches are created asynchronously on first | |
665 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
666 | * create the same cache, but only one of them may succeed. | |
667 | */ | |
f7ce3190 | 668 | if (arr->entries[idx]) |
d5b3cf71 VD |
669 | goto out_unlock; |
670 | ||
f1008365 | 671 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
73f576c0 JW |
672 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, |
673 | css->serial_nr, memcg_name_buf); | |
794b1248 VD |
674 | if (!cache_name) |
675 | goto out_unlock; | |
676 | ||
c9a77a79 | 677 | s = create_cache(cache_name, root_cache->object_size, |
613a5eb5 | 678 | root_cache->align, |
f773e36d | 679 | root_cache->flags & CACHE_CREATE_MASK, |
8eb8284b | 680 | root_cache->useroffset, root_cache->usersize, |
f773e36d | 681 | root_cache->ctor, memcg, root_cache); |
d5b3cf71 VD |
682 | /* |
683 | * If we could not create a memcg cache, do not complain, because | |
684 | * that's not critical at all as we can always proceed with the root | |
685 | * cache. | |
686 | */ | |
bd673145 | 687 | if (IS_ERR(s)) { |
794b1248 | 688 | kfree(cache_name); |
d5b3cf71 | 689 | goto out_unlock; |
bd673145 | 690 | } |
794b1248 | 691 | |
d5b3cf71 | 692 | /* |
f0a3a24b | 693 | * Since readers won't lock (see memcg_kmem_get_cache()), we need a |
d5b3cf71 VD |
694 | * barrier here to ensure nobody will see the kmem_cache partially |
695 | * initialized. | |
696 | */ | |
697 | smp_wmb(); | |
f7ce3190 | 698 | arr->entries[idx] = s; |
d5b3cf71 | 699 | |
794b1248 VD |
700 | out_unlock: |
701 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
702 | |
703 | put_online_mems(); | |
794b1248 | 704 | put_online_cpus(); |
2633d7a0 | 705 | } |
b8529907 | 706 | |
0b14e8aa | 707 | static void kmemcg_workfn(struct work_struct *work) |
01fb58bc TH |
708 | { |
709 | struct kmem_cache *s = container_of(work, struct kmem_cache, | |
0b14e8aa | 710 | memcg_params.work); |
01fb58bc TH |
711 | |
712 | get_online_cpus(); | |
713 | get_online_mems(); | |
714 | ||
715 | mutex_lock(&slab_mutex); | |
0b14e8aa | 716 | s->memcg_params.work_fn(s); |
01fb58bc TH |
717 | mutex_unlock(&slab_mutex); |
718 | ||
719 | put_online_mems(); | |
720 | put_online_cpus(); | |
01fb58bc TH |
721 | } |
722 | ||
0b14e8aa | 723 | static void kmemcg_rcufn(struct rcu_head *head) |
01fb58bc TH |
724 | { |
725 | struct kmem_cache *s = container_of(head, struct kmem_cache, | |
0b14e8aa | 726 | memcg_params.rcu_head); |
01fb58bc TH |
727 | |
728 | /* | |
0b14e8aa | 729 | * We need to grab blocking locks. Bounce to ->work. The |
01fb58bc TH |
730 | * work item shares the space with the RCU head and can't be |
731 | * initialized eariler. | |
732 | */ | |
0b14e8aa RG |
733 | INIT_WORK(&s->memcg_params.work, kmemcg_workfn); |
734 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.work); | |
01fb58bc TH |
735 | } |
736 | ||
f0a3a24b RG |
737 | static void kmemcg_cache_shutdown_fn(struct kmem_cache *s) |
738 | { | |
739 | WARN_ON(shutdown_cache(s)); | |
740 | } | |
741 | ||
742 | static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref) | |
743 | { | |
744 | struct kmem_cache *s = container_of(percpu_ref, struct kmem_cache, | |
745 | memcg_params.refcnt); | |
746 | unsigned long flags; | |
747 | ||
748 | spin_lock_irqsave(&memcg_kmem_wq_lock, flags); | |
749 | if (s->memcg_params.root_cache->memcg_params.dying) | |
750 | goto unlock; | |
751 | ||
752 | s->memcg_params.work_fn = kmemcg_cache_shutdown_fn; | |
753 | INIT_WORK(&s->memcg_params.work, kmemcg_workfn); | |
754 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.work); | |
755 | ||
756 | unlock: | |
757 | spin_unlock_irqrestore(&memcg_kmem_wq_lock, flags); | |
758 | } | |
759 | ||
760 | static void kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s) | |
761 | { | |
762 | __kmemcg_cache_deactivate_after_rcu(s); | |
763 | percpu_ref_kill(&s->memcg_params.refcnt); | |
764 | } | |
765 | ||
43486694 | 766 | static void kmemcg_cache_deactivate(struct kmem_cache *s) |
01fb58bc | 767 | { |
f0a3a24b | 768 | if (WARN_ON_ONCE(is_root_cache(s))) |
01fb58bc TH |
769 | return; |
770 | ||
43486694 RG |
771 | __kmemcg_cache_deactivate(s); |
772 | ||
63b02ef7 RG |
773 | /* |
774 | * memcg_kmem_wq_lock is used to synchronize memcg_params.dying | |
775 | * flag and make sure that no new kmem_cache deactivation tasks | |
776 | * are queued (see flush_memcg_workqueue() ). | |
777 | */ | |
778 | spin_lock_irq(&memcg_kmem_wq_lock); | |
92ee383f | 779 | if (s->memcg_params.root_cache->memcg_params.dying) |
63b02ef7 | 780 | goto unlock; |
92ee383f | 781 | |
f0a3a24b | 782 | s->memcg_params.work_fn = kmemcg_cache_deactivate_after_rcu; |
0b14e8aa | 783 | call_rcu(&s->memcg_params.rcu_head, kmemcg_rcufn); |
63b02ef7 RG |
784 | unlock: |
785 | spin_unlock_irq(&memcg_kmem_wq_lock); | |
01fb58bc TH |
786 | } |
787 | ||
2a4db7eb VD |
788 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg) |
789 | { | |
790 | int idx; | |
791 | struct memcg_cache_array *arr; | |
d6e0b7fa | 792 | struct kmem_cache *s, *c; |
2a4db7eb VD |
793 | |
794 | idx = memcg_cache_id(memcg); | |
795 | ||
d6e0b7fa VD |
796 | get_online_cpus(); |
797 | get_online_mems(); | |
798 | ||
2a4db7eb | 799 | mutex_lock(&slab_mutex); |
510ded33 | 800 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
2a4db7eb VD |
801 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, |
802 | lockdep_is_held(&slab_mutex)); | |
d6e0b7fa VD |
803 | c = arr->entries[idx]; |
804 | if (!c) | |
805 | continue; | |
806 | ||
43486694 | 807 | kmemcg_cache_deactivate(c); |
2a4db7eb VD |
808 | arr->entries[idx] = NULL; |
809 | } | |
810 | mutex_unlock(&slab_mutex); | |
d6e0b7fa VD |
811 | |
812 | put_online_mems(); | |
813 | put_online_cpus(); | |
2a4db7eb VD |
814 | } |
815 | ||
657dc2f9 | 816 | static int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
817 | { |
818 | struct memcg_cache_array *arr; | |
819 | struct kmem_cache *c, *c2; | |
820 | LIST_HEAD(busy); | |
821 | int i; | |
822 | ||
823 | BUG_ON(!is_root_cache(s)); | |
824 | ||
825 | /* | |
826 | * First, shutdown active caches, i.e. caches that belong to online | |
827 | * memory cgroups. | |
828 | */ | |
829 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
830 | lockdep_is_held(&slab_mutex)); | |
831 | for_each_memcg_cache_index(i) { | |
832 | c = arr->entries[i]; | |
833 | if (!c) | |
834 | continue; | |
657dc2f9 | 835 | if (shutdown_cache(c)) |
d60fdcc9 VD |
836 | /* |
837 | * The cache still has objects. Move it to a temporary | |
838 | * list so as not to try to destroy it for a second | |
839 | * time while iterating over inactive caches below. | |
840 | */ | |
9eeadc8b | 841 | list_move(&c->memcg_params.children_node, &busy); |
d60fdcc9 VD |
842 | else |
843 | /* | |
844 | * The cache is empty and will be destroyed soon. Clear | |
845 | * the pointer to it in the memcg_caches array so that | |
846 | * it will never be accessed even if the root cache | |
847 | * stays alive. | |
848 | */ | |
849 | arr->entries[i] = NULL; | |
850 | } | |
851 | ||
852 | /* | |
853 | * Second, shutdown all caches left from memory cgroups that are now | |
854 | * offline. | |
855 | */ | |
9eeadc8b TH |
856 | list_for_each_entry_safe(c, c2, &s->memcg_params.children, |
857 | memcg_params.children_node) | |
657dc2f9 | 858 | shutdown_cache(c); |
d60fdcc9 | 859 | |
9eeadc8b | 860 | list_splice(&busy, &s->memcg_params.children); |
d60fdcc9 VD |
861 | |
862 | /* | |
863 | * A cache being destroyed must be empty. In particular, this means | |
864 | * that all per memcg caches attached to it must be empty too. | |
865 | */ | |
9eeadc8b | 866 | if (!list_empty(&s->memcg_params.children)) |
d60fdcc9 VD |
867 | return -EBUSY; |
868 | return 0; | |
869 | } | |
92ee383f SB |
870 | |
871 | static void flush_memcg_workqueue(struct kmem_cache *s) | |
872 | { | |
63b02ef7 | 873 | spin_lock_irq(&memcg_kmem_wq_lock); |
92ee383f | 874 | s->memcg_params.dying = true; |
63b02ef7 | 875 | spin_unlock_irq(&memcg_kmem_wq_lock); |
92ee383f SB |
876 | |
877 | /* | |
43486694 | 878 | * SLAB and SLUB deactivate the kmem_caches through call_rcu. Make |
92ee383f SB |
879 | * sure all registered rcu callbacks have been invoked. |
880 | */ | |
43486694 | 881 | rcu_barrier(); |
92ee383f SB |
882 | |
883 | /* | |
884 | * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB | |
885 | * deactivates the memcg kmem_caches through workqueue. Make sure all | |
886 | * previous workitems on workqueue are processed. | |
887 | */ | |
888 | flush_workqueue(memcg_kmem_cache_wq); | |
889 | } | |
d60fdcc9 | 890 | #else |
657dc2f9 | 891 | static inline int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
892 | { |
893 | return 0; | |
894 | } | |
92ee383f SB |
895 | |
896 | static inline void flush_memcg_workqueue(struct kmem_cache *s) | |
897 | { | |
898 | } | |
84c07d11 | 899 | #endif /* CONFIG_MEMCG_KMEM */ |
97d06609 | 900 | |
41a21285 CL |
901 | void slab_kmem_cache_release(struct kmem_cache *s) |
902 | { | |
52b4b950 | 903 | __kmem_cache_release(s); |
f7ce3190 | 904 | destroy_memcg_params(s); |
3dec16ea | 905 | kfree_const(s->name); |
41a21285 CL |
906 | kmem_cache_free(kmem_cache, s); |
907 | } | |
908 | ||
945cf2b6 CL |
909 | void kmem_cache_destroy(struct kmem_cache *s) |
910 | { | |
d60fdcc9 | 911 | int err; |
d5b3cf71 | 912 | |
3942d299 SS |
913 | if (unlikely(!s)) |
914 | return; | |
915 | ||
92ee383f SB |
916 | flush_memcg_workqueue(s); |
917 | ||
945cf2b6 | 918 | get_online_cpus(); |
03afc0e2 VD |
919 | get_online_mems(); |
920 | ||
945cf2b6 | 921 | mutex_lock(&slab_mutex); |
b8529907 | 922 | |
945cf2b6 | 923 | s->refcount--; |
b8529907 VD |
924 | if (s->refcount) |
925 | goto out_unlock; | |
926 | ||
657dc2f9 | 927 | err = shutdown_memcg_caches(s); |
d60fdcc9 | 928 | if (!err) |
657dc2f9 | 929 | err = shutdown_cache(s); |
b8529907 | 930 | |
cd918c55 | 931 | if (err) { |
756a025f JP |
932 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
933 | s->name); | |
cd918c55 VD |
934 | dump_stack(); |
935 | } | |
b8529907 VD |
936 | out_unlock: |
937 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 938 | |
03afc0e2 | 939 | put_online_mems(); |
945cf2b6 CL |
940 | put_online_cpus(); |
941 | } | |
942 | EXPORT_SYMBOL(kmem_cache_destroy); | |
943 | ||
03afc0e2 VD |
944 | /** |
945 | * kmem_cache_shrink - Shrink a cache. | |
946 | * @cachep: The cache to shrink. | |
947 | * | |
948 | * Releases as many slabs as possible for a cache. | |
949 | * To help debugging, a zero exit status indicates all slabs were released. | |
a862f68a MR |
950 | * |
951 | * Return: %0 if all slabs were released, non-zero otherwise | |
03afc0e2 VD |
952 | */ |
953 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
954 | { | |
955 | int ret; | |
956 | ||
957 | get_online_cpus(); | |
958 | get_online_mems(); | |
55834c59 | 959 | kasan_cache_shrink(cachep); |
c9fc5864 | 960 | ret = __kmem_cache_shrink(cachep); |
03afc0e2 VD |
961 | put_online_mems(); |
962 | put_online_cpus(); | |
963 | return ret; | |
964 | } | |
965 | EXPORT_SYMBOL(kmem_cache_shrink); | |
966 | ||
fda90124 | 967 | bool slab_is_available(void) |
97d06609 CL |
968 | { |
969 | return slab_state >= UP; | |
970 | } | |
b7454ad3 | 971 | |
45530c44 CL |
972 | #ifndef CONFIG_SLOB |
973 | /* Create a cache during boot when no slab services are available yet */ | |
361d575e AD |
974 | void __init create_boot_cache(struct kmem_cache *s, const char *name, |
975 | unsigned int size, slab_flags_t flags, | |
976 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
977 | { |
978 | int err; | |
979 | ||
980 | s->name = name; | |
981 | s->size = s->object_size = size; | |
45906855 | 982 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
8eb8284b DW |
983 | s->useroffset = useroffset; |
984 | s->usersize = usersize; | |
f7ce3190 VD |
985 | |
986 | slab_init_memcg_params(s); | |
987 | ||
45530c44 CL |
988 | err = __kmem_cache_create(s, flags); |
989 | ||
990 | if (err) | |
361d575e | 991 | panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", |
45530c44 CL |
992 | name, size, err); |
993 | ||
994 | s->refcount = -1; /* Exempt from merging for now */ | |
995 | } | |
996 | ||
55de8b9c AD |
997 | struct kmem_cache *__init create_kmalloc_cache(const char *name, |
998 | unsigned int size, slab_flags_t flags, | |
999 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
1000 | { |
1001 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
1002 | ||
1003 | if (!s) | |
1004 | panic("Out of memory when creating slab %s\n", name); | |
1005 | ||
6c0c21ad | 1006 | create_boot_cache(s, name, size, flags, useroffset, usersize); |
45530c44 | 1007 | list_add(&s->list, &slab_caches); |
c03914b7 | 1008 | memcg_link_cache(s, NULL); |
45530c44 CL |
1009 | s->refcount = 1; |
1010 | return s; | |
1011 | } | |
1012 | ||
cc252eae VB |
1013 | struct kmem_cache * |
1014 | kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init; | |
9425c58e CL |
1015 | EXPORT_SYMBOL(kmalloc_caches); |
1016 | ||
2c59dd65 CL |
1017 | /* |
1018 | * Conversion table for small slabs sizes / 8 to the index in the | |
1019 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
1020 | * of two cache sizes there. The size of larger slabs can be determined using | |
1021 | * fls. | |
1022 | */ | |
d5f86655 | 1023 | static u8 size_index[24] __ro_after_init = { |
2c59dd65 CL |
1024 | 3, /* 8 */ |
1025 | 4, /* 16 */ | |
1026 | 5, /* 24 */ | |
1027 | 5, /* 32 */ | |
1028 | 6, /* 40 */ | |
1029 | 6, /* 48 */ | |
1030 | 6, /* 56 */ | |
1031 | 6, /* 64 */ | |
1032 | 1, /* 72 */ | |
1033 | 1, /* 80 */ | |
1034 | 1, /* 88 */ | |
1035 | 1, /* 96 */ | |
1036 | 7, /* 104 */ | |
1037 | 7, /* 112 */ | |
1038 | 7, /* 120 */ | |
1039 | 7, /* 128 */ | |
1040 | 2, /* 136 */ | |
1041 | 2, /* 144 */ | |
1042 | 2, /* 152 */ | |
1043 | 2, /* 160 */ | |
1044 | 2, /* 168 */ | |
1045 | 2, /* 176 */ | |
1046 | 2, /* 184 */ | |
1047 | 2 /* 192 */ | |
1048 | }; | |
1049 | ||
ac914d08 | 1050 | static inline unsigned int size_index_elem(unsigned int bytes) |
2c59dd65 CL |
1051 | { |
1052 | return (bytes - 1) / 8; | |
1053 | } | |
1054 | ||
1055 | /* | |
1056 | * Find the kmem_cache structure that serves a given size of | |
1057 | * allocation | |
1058 | */ | |
1059 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
1060 | { | |
d5f86655 | 1061 | unsigned int index; |
2c59dd65 CL |
1062 | |
1063 | if (size <= 192) { | |
1064 | if (!size) | |
1065 | return ZERO_SIZE_PTR; | |
1066 | ||
1067 | index = size_index[size_index_elem(size)]; | |
61448479 | 1068 | } else { |
221d7da6 | 1069 | if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) |
61448479 | 1070 | return NULL; |
2c59dd65 | 1071 | index = fls(size - 1); |
61448479 | 1072 | } |
2c59dd65 | 1073 | |
cc252eae | 1074 | return kmalloc_caches[kmalloc_type(flags)][index]; |
2c59dd65 CL |
1075 | } |
1076 | ||
4066c33d GG |
1077 | /* |
1078 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
1079 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
1080 | * kmalloc-67108864. | |
1081 | */ | |
af3b5f87 | 1082 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
4066c33d GG |
1083 | {NULL, 0}, {"kmalloc-96", 96}, |
1084 | {"kmalloc-192", 192}, {"kmalloc-8", 8}, | |
1085 | {"kmalloc-16", 16}, {"kmalloc-32", 32}, | |
1086 | {"kmalloc-64", 64}, {"kmalloc-128", 128}, | |
1087 | {"kmalloc-256", 256}, {"kmalloc-512", 512}, | |
f0d77874 VB |
1088 | {"kmalloc-1k", 1024}, {"kmalloc-2k", 2048}, |
1089 | {"kmalloc-4k", 4096}, {"kmalloc-8k", 8192}, | |
1090 | {"kmalloc-16k", 16384}, {"kmalloc-32k", 32768}, | |
1091 | {"kmalloc-64k", 65536}, {"kmalloc-128k", 131072}, | |
1092 | {"kmalloc-256k", 262144}, {"kmalloc-512k", 524288}, | |
1093 | {"kmalloc-1M", 1048576}, {"kmalloc-2M", 2097152}, | |
1094 | {"kmalloc-4M", 4194304}, {"kmalloc-8M", 8388608}, | |
1095 | {"kmalloc-16M", 16777216}, {"kmalloc-32M", 33554432}, | |
1096 | {"kmalloc-64M", 67108864} | |
4066c33d GG |
1097 | }; |
1098 | ||
f97d5f63 | 1099 | /* |
34cc6990 DS |
1100 | * Patch up the size_index table if we have strange large alignment |
1101 | * requirements for the kmalloc array. This is only the case for | |
1102 | * MIPS it seems. The standard arches will not generate any code here. | |
1103 | * | |
1104 | * Largest permitted alignment is 256 bytes due to the way we | |
1105 | * handle the index determination for the smaller caches. | |
1106 | * | |
1107 | * Make sure that nothing crazy happens if someone starts tinkering | |
1108 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 1109 | */ |
34cc6990 | 1110 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 | 1111 | { |
ac914d08 | 1112 | unsigned int i; |
f97d5f63 | 1113 | |
2c59dd65 CL |
1114 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
1115 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
1116 | ||
1117 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
ac914d08 | 1118 | unsigned int elem = size_index_elem(i); |
2c59dd65 CL |
1119 | |
1120 | if (elem >= ARRAY_SIZE(size_index)) | |
1121 | break; | |
1122 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
1123 | } | |
1124 | ||
1125 | if (KMALLOC_MIN_SIZE >= 64) { | |
1126 | /* | |
1127 | * The 96 byte size cache is not used if the alignment | |
1128 | * is 64 byte. | |
1129 | */ | |
1130 | for (i = 64 + 8; i <= 96; i += 8) | |
1131 | size_index[size_index_elem(i)] = 7; | |
1132 | ||
1133 | } | |
1134 | ||
1135 | if (KMALLOC_MIN_SIZE >= 128) { | |
1136 | /* | |
1137 | * The 192 byte sized cache is not used if the alignment | |
1138 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
1139 | * instead. | |
1140 | */ | |
1141 | for (i = 128 + 8; i <= 192; i += 8) | |
1142 | size_index[size_index_elem(i)] = 8; | |
1143 | } | |
34cc6990 DS |
1144 | } |
1145 | ||
f0d77874 VB |
1146 | static const char * |
1147 | kmalloc_cache_name(const char *prefix, unsigned int size) | |
1148 | { | |
1149 | ||
1150 | static const char units[3] = "\0kM"; | |
1151 | int idx = 0; | |
1152 | ||
1153 | while (size >= 1024 && (size % 1024 == 0)) { | |
1154 | size /= 1024; | |
1155 | idx++; | |
1156 | } | |
1157 | ||
1158 | return kasprintf(GFP_NOWAIT, "%s-%u%c", prefix, size, units[idx]); | |
1159 | } | |
1160 | ||
1291523f VB |
1161 | static void __init |
1162 | new_kmalloc_cache(int idx, int type, slab_flags_t flags) | |
a9730fca | 1163 | { |
1291523f VB |
1164 | const char *name; |
1165 | ||
1166 | if (type == KMALLOC_RECLAIM) { | |
1167 | flags |= SLAB_RECLAIM_ACCOUNT; | |
f0d77874 | 1168 | name = kmalloc_cache_name("kmalloc-rcl", |
1291523f VB |
1169 | kmalloc_info[idx].size); |
1170 | BUG_ON(!name); | |
1171 | } else { | |
1172 | name = kmalloc_info[idx].name; | |
1173 | } | |
1174 | ||
1175 | kmalloc_caches[type][idx] = create_kmalloc_cache(name, | |
6c0c21ad DW |
1176 | kmalloc_info[idx].size, flags, 0, |
1177 | kmalloc_info[idx].size); | |
a9730fca CL |
1178 | } |
1179 | ||
34cc6990 DS |
1180 | /* |
1181 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
1182 | * may already have been created because they were needed to | |
1183 | * enable allocations for slab creation. | |
1184 | */ | |
d50112ed | 1185 | void __init create_kmalloc_caches(slab_flags_t flags) |
34cc6990 | 1186 | { |
1291523f | 1187 | int i, type; |
34cc6990 | 1188 | |
1291523f VB |
1189 | for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { |
1190 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1191 | if (!kmalloc_caches[type][i]) | |
1192 | new_kmalloc_cache(i, type, flags); | |
f97d5f63 | 1193 | |
1291523f VB |
1194 | /* |
1195 | * Caches that are not of the two-to-the-power-of size. | |
1196 | * These have to be created immediately after the | |
1197 | * earlier power of two caches | |
1198 | */ | |
1199 | if (KMALLOC_MIN_SIZE <= 32 && i == 6 && | |
1200 | !kmalloc_caches[type][1]) | |
1201 | new_kmalloc_cache(1, type, flags); | |
1202 | if (KMALLOC_MIN_SIZE <= 64 && i == 7 && | |
1203 | !kmalloc_caches[type][2]) | |
1204 | new_kmalloc_cache(2, type, flags); | |
1205 | } | |
8a965b3b CL |
1206 | } |
1207 | ||
f97d5f63 CL |
1208 | /* Kmalloc array is now usable */ |
1209 | slab_state = UP; | |
1210 | ||
f97d5f63 CL |
1211 | #ifdef CONFIG_ZONE_DMA |
1212 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
cc252eae | 1213 | struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; |
f97d5f63 CL |
1214 | |
1215 | if (s) { | |
0be70327 | 1216 | unsigned int size = kmalloc_size(i); |
f0d77874 | 1217 | const char *n = kmalloc_cache_name("dma-kmalloc", size); |
f97d5f63 CL |
1218 | |
1219 | BUG_ON(!n); | |
cc252eae VB |
1220 | kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( |
1221 | n, size, SLAB_CACHE_DMA | flags, 0, 0); | |
f97d5f63 CL |
1222 | } |
1223 | } | |
1224 | #endif | |
1225 | } | |
45530c44 CL |
1226 | #endif /* !CONFIG_SLOB */ |
1227 | ||
cea371f4 VD |
1228 | /* |
1229 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1230 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1231 | * know the allocation order to free the pages properly in kfree. | |
1232 | */ | |
52383431 VD |
1233 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
1234 | { | |
1235 | void *ret; | |
1236 | struct page *page; | |
1237 | ||
1238 | flags |= __GFP_COMP; | |
4949148a | 1239 | page = alloc_pages(flags, order); |
52383431 | 1240 | ret = page ? page_address(page) : NULL; |
0116523c | 1241 | ret = kasan_kmalloc_large(ret, size, flags); |
a2f77575 | 1242 | /* As ret might get tagged, call kmemleak hook after KASAN. */ |
53128245 | 1243 | kmemleak_alloc(ret, size, 1, flags); |
52383431 VD |
1244 | return ret; |
1245 | } | |
1246 | EXPORT_SYMBOL(kmalloc_order); | |
1247 | ||
f1b6eb6e CL |
1248 | #ifdef CONFIG_TRACING |
1249 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1250 | { | |
1251 | void *ret = kmalloc_order(size, flags, order); | |
1252 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1253 | return ret; | |
1254 | } | |
1255 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1256 | #endif | |
45530c44 | 1257 | |
7c00fce9 TG |
1258 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1259 | /* Randomize a generic freelist */ | |
1260 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
302d55d5 | 1261 | unsigned int count) |
7c00fce9 | 1262 | { |
7c00fce9 | 1263 | unsigned int rand; |
302d55d5 | 1264 | unsigned int i; |
7c00fce9 TG |
1265 | |
1266 | for (i = 0; i < count; i++) | |
1267 | list[i] = i; | |
1268 | ||
1269 | /* Fisher-Yates shuffle */ | |
1270 | for (i = count - 1; i > 0; i--) { | |
1271 | rand = prandom_u32_state(state); | |
1272 | rand %= (i + 1); | |
1273 | swap(list[i], list[rand]); | |
1274 | } | |
1275 | } | |
1276 | ||
1277 | /* Create a random sequence per cache */ | |
1278 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1279 | gfp_t gfp) | |
1280 | { | |
1281 | struct rnd_state state; | |
1282 | ||
1283 | if (count < 2 || cachep->random_seq) | |
1284 | return 0; | |
1285 | ||
1286 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1287 | if (!cachep->random_seq) | |
1288 | return -ENOMEM; | |
1289 | ||
1290 | /* Get best entropy at this stage of boot */ | |
1291 | prandom_seed_state(&state, get_random_long()); | |
1292 | ||
1293 | freelist_randomize(&state, cachep->random_seq, count); | |
1294 | return 0; | |
1295 | } | |
1296 | ||
1297 | /* Destroy the per-cache random freelist sequence */ | |
1298 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1299 | { | |
1300 | kfree(cachep->random_seq); | |
1301 | cachep->random_seq = NULL; | |
1302 | } | |
1303 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1304 | ||
5b365771 | 1305 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
e9b4db2b | 1306 | #ifdef CONFIG_SLAB |
0825a6f9 | 1307 | #define SLABINFO_RIGHTS (0600) |
e9b4db2b | 1308 | #else |
0825a6f9 | 1309 | #define SLABINFO_RIGHTS (0400) |
e9b4db2b WL |
1310 | #endif |
1311 | ||
b047501c | 1312 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1313 | { |
1314 | /* | |
1315 | * Output format version, so at least we can change it | |
1316 | * without _too_ many complaints. | |
1317 | */ | |
1318 | #ifdef CONFIG_DEBUG_SLAB | |
1319 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1320 | #else | |
1321 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1322 | #endif | |
756a025f | 1323 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1324 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1325 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1326 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1327 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1328 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1329 | #endif | |
1330 | seq_putc(m, '\n'); | |
1331 | } | |
1332 | ||
1df3b26f | 1333 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1334 | { |
b7454ad3 | 1335 | mutex_lock(&slab_mutex); |
510ded33 | 1336 | return seq_list_start(&slab_root_caches, *pos); |
b7454ad3 GC |
1337 | } |
1338 | ||
276a2439 | 1339 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 | 1340 | { |
510ded33 | 1341 | return seq_list_next(p, &slab_root_caches, pos); |
b7454ad3 GC |
1342 | } |
1343 | ||
276a2439 | 1344 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1345 | { |
1346 | mutex_unlock(&slab_mutex); | |
1347 | } | |
1348 | ||
749c5415 GC |
1349 | static void |
1350 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1351 | { | |
1352 | struct kmem_cache *c; | |
1353 | struct slabinfo sinfo; | |
749c5415 GC |
1354 | |
1355 | if (!is_root_cache(s)) | |
1356 | return; | |
1357 | ||
426589f5 | 1358 | for_each_memcg_cache(c, s) { |
749c5415 GC |
1359 | memset(&sinfo, 0, sizeof(sinfo)); |
1360 | get_slabinfo(c, &sinfo); | |
1361 | ||
1362 | info->active_slabs += sinfo.active_slabs; | |
1363 | info->num_slabs += sinfo.num_slabs; | |
1364 | info->shared_avail += sinfo.shared_avail; | |
1365 | info->active_objs += sinfo.active_objs; | |
1366 | info->num_objs += sinfo.num_objs; | |
1367 | } | |
1368 | } | |
1369 | ||
b047501c | 1370 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1371 | { |
0d7561c6 GC |
1372 | struct slabinfo sinfo; |
1373 | ||
1374 | memset(&sinfo, 0, sizeof(sinfo)); | |
1375 | get_slabinfo(s, &sinfo); | |
1376 | ||
749c5415 GC |
1377 | memcg_accumulate_slabinfo(s, &sinfo); |
1378 | ||
0d7561c6 | 1379 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 1380 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1381 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1382 | ||
1383 | seq_printf(m, " : tunables %4u %4u %4u", | |
1384 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1385 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1386 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1387 | slabinfo_show_stats(m, s); | |
1388 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1389 | } |
1390 | ||
1df3b26f | 1391 | static int slab_show(struct seq_file *m, void *p) |
749c5415 | 1392 | { |
510ded33 | 1393 | struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node); |
749c5415 | 1394 | |
510ded33 | 1395 | if (p == slab_root_caches.next) |
1df3b26f | 1396 | print_slabinfo_header(m); |
510ded33 | 1397 | cache_show(s, m); |
b047501c VD |
1398 | return 0; |
1399 | } | |
1400 | ||
852d8be0 YS |
1401 | void dump_unreclaimable_slab(void) |
1402 | { | |
1403 | struct kmem_cache *s, *s2; | |
1404 | struct slabinfo sinfo; | |
1405 | ||
1406 | /* | |
1407 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1408 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1409 | * risk of crash. | |
1410 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1411 | * without acquiring the mutex. | |
1412 | */ | |
1413 | if (!mutex_trylock(&slab_mutex)) { | |
1414 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1415 | return; | |
1416 | } | |
1417 | ||
1418 | pr_info("Unreclaimable slab info:\n"); | |
1419 | pr_info("Name Used Total\n"); | |
1420 | ||
1421 | list_for_each_entry_safe(s, s2, &slab_caches, list) { | |
1422 | if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT)) | |
1423 | continue; | |
1424 | ||
1425 | get_slabinfo(s, &sinfo); | |
1426 | ||
1427 | if (sinfo.num_objs > 0) | |
1428 | pr_info("%-17s %10luKB %10luKB\n", cache_name(s), | |
1429 | (sinfo.active_objs * s->size) / 1024, | |
1430 | (sinfo.num_objs * s->size) / 1024); | |
1431 | } | |
1432 | mutex_unlock(&slab_mutex); | |
1433 | } | |
1434 | ||
5b365771 | 1435 | #if defined(CONFIG_MEMCG) |
bc2791f8 TH |
1436 | void *memcg_slab_start(struct seq_file *m, loff_t *pos) |
1437 | { | |
aa9694bb | 1438 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
bc2791f8 TH |
1439 | |
1440 | mutex_lock(&slab_mutex); | |
1441 | return seq_list_start(&memcg->kmem_caches, *pos); | |
1442 | } | |
1443 | ||
1444 | void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1445 | { | |
aa9694bb | 1446 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
bc2791f8 TH |
1447 | |
1448 | return seq_list_next(p, &memcg->kmem_caches, pos); | |
1449 | } | |
1450 | ||
1451 | void memcg_slab_stop(struct seq_file *m, void *p) | |
1452 | { | |
1453 | mutex_unlock(&slab_mutex); | |
1454 | } | |
1455 | ||
b047501c VD |
1456 | int memcg_slab_show(struct seq_file *m, void *p) |
1457 | { | |
bc2791f8 TH |
1458 | struct kmem_cache *s = list_entry(p, struct kmem_cache, |
1459 | memcg_params.kmem_caches_node); | |
aa9694bb | 1460 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
b047501c | 1461 | |
bc2791f8 | 1462 | if (p == memcg->kmem_caches.next) |
b047501c | 1463 | print_slabinfo_header(m); |
bc2791f8 | 1464 | cache_show(s, m); |
b047501c | 1465 | return 0; |
749c5415 | 1466 | } |
b047501c | 1467 | #endif |
749c5415 | 1468 | |
b7454ad3 GC |
1469 | /* |
1470 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1471 | * | |
1472 | * Output layout: | |
1473 | * cache-name | |
1474 | * num-active-objs | |
1475 | * total-objs | |
1476 | * object size | |
1477 | * num-active-slabs | |
1478 | * total-slabs | |
1479 | * num-pages-per-slab | |
1480 | * + further values on SMP and with statistics enabled | |
1481 | */ | |
1482 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1483 | .start = slab_start, |
276a2439 WL |
1484 | .next = slab_next, |
1485 | .stop = slab_stop, | |
1df3b26f | 1486 | .show = slab_show, |
b7454ad3 GC |
1487 | }; |
1488 | ||
1489 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1490 | { | |
1491 | return seq_open(file, &slabinfo_op); | |
1492 | } | |
1493 | ||
1494 | static const struct file_operations proc_slabinfo_operations = { | |
1495 | .open = slabinfo_open, | |
1496 | .read = seq_read, | |
1497 | .write = slabinfo_write, | |
1498 | .llseek = seq_lseek, | |
1499 | .release = seq_release, | |
1500 | }; | |
1501 | ||
1502 | static int __init slab_proc_init(void) | |
1503 | { | |
e9b4db2b WL |
1504 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, |
1505 | &proc_slabinfo_operations); | |
b7454ad3 GC |
1506 | return 0; |
1507 | } | |
1508 | module_init(slab_proc_init); | |
5b365771 | 1509 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ |
928cec9c AR |
1510 | |
1511 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1512 | gfp_t flags) | |
1513 | { | |
1514 | void *ret; | |
1515 | size_t ks = 0; | |
1516 | ||
1517 | if (p) | |
1518 | ks = ksize(p); | |
1519 | ||
0316bec2 | 1520 | if (ks >= new_size) { |
0116523c | 1521 | p = kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1522 | return (void *)p; |
0316bec2 | 1523 | } |
928cec9c AR |
1524 | |
1525 | ret = kmalloc_track_caller(new_size, flags); | |
1526 | if (ret && p) | |
1527 | memcpy(ret, p, ks); | |
1528 | ||
1529 | return ret; | |
1530 | } | |
1531 | ||
1532 | /** | |
1533 | * __krealloc - like krealloc() but don't free @p. | |
1534 | * @p: object to reallocate memory for. | |
1535 | * @new_size: how many bytes of memory are required. | |
1536 | * @flags: the type of memory to allocate. | |
1537 | * | |
1538 | * This function is like krealloc() except it never frees the originally | |
1539 | * allocated buffer. Use this if you don't want to free the buffer immediately | |
1540 | * like, for example, with RCU. | |
a862f68a MR |
1541 | * |
1542 | * Return: pointer to the allocated memory or %NULL in case of error | |
928cec9c AR |
1543 | */ |
1544 | void *__krealloc(const void *p, size_t new_size, gfp_t flags) | |
1545 | { | |
1546 | if (unlikely(!new_size)) | |
1547 | return ZERO_SIZE_PTR; | |
1548 | ||
1549 | return __do_krealloc(p, new_size, flags); | |
1550 | ||
1551 | } | |
1552 | EXPORT_SYMBOL(__krealloc); | |
1553 | ||
1554 | /** | |
1555 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1556 | * @p: object to reallocate memory for. | |
1557 | * @new_size: how many bytes of memory are required. | |
1558 | * @flags: the type of memory to allocate. | |
1559 | * | |
1560 | * The contents of the object pointed to are preserved up to the | |
1561 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1562 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1563 | * %NULL pointer, the object pointed to is freed. | |
a862f68a MR |
1564 | * |
1565 | * Return: pointer to the allocated memory or %NULL in case of error | |
928cec9c AR |
1566 | */ |
1567 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1568 | { | |
1569 | void *ret; | |
1570 | ||
1571 | if (unlikely(!new_size)) { | |
1572 | kfree(p); | |
1573 | return ZERO_SIZE_PTR; | |
1574 | } | |
1575 | ||
1576 | ret = __do_krealloc(p, new_size, flags); | |
772a2fa5 | 1577 | if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) |
928cec9c AR |
1578 | kfree(p); |
1579 | ||
1580 | return ret; | |
1581 | } | |
1582 | EXPORT_SYMBOL(krealloc); | |
1583 | ||
1584 | /** | |
1585 | * kzfree - like kfree but zero memory | |
1586 | * @p: object to free memory of | |
1587 | * | |
1588 | * The memory of the object @p points to is zeroed before freed. | |
1589 | * If @p is %NULL, kzfree() does nothing. | |
1590 | * | |
1591 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1592 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1593 | * careful when using this function in performance sensitive code. | |
1594 | */ | |
1595 | void kzfree(const void *p) | |
1596 | { | |
1597 | size_t ks; | |
1598 | void *mem = (void *)p; | |
1599 | ||
1600 | if (unlikely(ZERO_OR_NULL_PTR(mem))) | |
1601 | return; | |
1602 | ks = ksize(mem); | |
1603 | memset(mem, 0, ks); | |
1604 | kfree(mem); | |
1605 | } | |
1606 | EXPORT_SYMBOL(kzfree); | |
1607 | ||
10d1f8cb ME |
1608 | /** |
1609 | * ksize - get the actual amount of memory allocated for a given object | |
1610 | * @objp: Pointer to the object | |
1611 | * | |
1612 | * kmalloc may internally round up allocations and return more memory | |
1613 | * than requested. ksize() can be used to determine the actual amount of | |
1614 | * memory allocated. The caller may use this additional memory, even though | |
1615 | * a smaller amount of memory was initially specified with the kmalloc call. | |
1616 | * The caller must guarantee that objp points to a valid object previously | |
1617 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
1618 | * must not be freed during the duration of the call. | |
1619 | * | |
1620 | * Return: size of the actual memory used by @objp in bytes | |
1621 | */ | |
1622 | size_t ksize(const void *objp) | |
1623 | { | |
0d4ca4c9 ME |
1624 | size_t size; |
1625 | ||
1626 | if (WARN_ON_ONCE(!objp)) | |
1627 | return 0; | |
1628 | /* | |
1629 | * We need to check that the pointed to object is valid, and only then | |
1630 | * unpoison the shadow memory below. We use __kasan_check_read(), to | |
1631 | * generate a more useful report at the time ksize() is called (rather | |
1632 | * than later where behaviour is undefined due to potential | |
1633 | * use-after-free or double-free). | |
1634 | * | |
1635 | * If the pointed to memory is invalid we return 0, to avoid users of | |
1636 | * ksize() writing to and potentially corrupting the memory region. | |
1637 | * | |
1638 | * We want to perform the check before __ksize(), to avoid potentially | |
1639 | * crashing in __ksize() due to accessing invalid metadata. | |
1640 | */ | |
1641 | if (unlikely(objp == ZERO_SIZE_PTR) || !__kasan_check_read(objp, 1)) | |
1642 | return 0; | |
1643 | ||
1644 | size = __ksize(objp); | |
10d1f8cb ME |
1645 | /* |
1646 | * We assume that ksize callers could use whole allocated area, | |
1647 | * so we need to unpoison this area. | |
1648 | */ | |
1649 | kasan_unpoison_shadow(objp, size); | |
1650 | return size; | |
1651 | } | |
1652 | EXPORT_SYMBOL(ksize); | |
1653 | ||
928cec9c AR |
1654 | /* Tracepoints definitions. */ |
1655 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1656 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1657 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1658 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1659 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1660 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); | |
4f6923fb HM |
1661 | |
1662 | int should_failslab(struct kmem_cache *s, gfp_t gfpflags) | |
1663 | { | |
1664 | if (__should_failslab(s, gfpflags)) | |
1665 | return -ENOMEM; | |
1666 | return 0; | |
1667 | } | |
1668 | ALLOW_ERROR_INJECTION(should_failslab, ERRNO); |