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