Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
039363f3 CL |
2 | /* |
3 | * Slab allocator functions that are independent of the allocator strategy | |
4 | * | |
5 | * (C) 2012 Christoph Lameter <cl@linux.com> | |
6 | */ | |
7 | #include <linux/slab.h> | |
8 | ||
9 | #include <linux/mm.h> | |
10 | #include <linux/poison.h> | |
11 | #include <linux/interrupt.h> | |
12 | #include <linux/memory.h> | |
1c99ba29 | 13 | #include <linux/cache.h> |
039363f3 CL |
14 | #include <linux/compiler.h> |
15 | #include <linux/module.h> | |
20cea968 CL |
16 | #include <linux/cpu.h> |
17 | #include <linux/uaccess.h> | |
b7454ad3 GC |
18 | #include <linux/seq_file.h> |
19 | #include <linux/proc_fs.h> | |
fcf8a1e4 | 20 | #include <linux/debugfs.h> |
039363f3 CL |
21 | #include <asm/cacheflush.h> |
22 | #include <asm/tlbflush.h> | |
23 | #include <asm/page.h> | |
2633d7a0 | 24 | #include <linux/memcontrol.h> |
928cec9c AR |
25 | |
26 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 27 | #include <trace/events/kmem.h> |
039363f3 | 28 | |
44405099 LL |
29 | #include "internal.h" |
30 | ||
97d06609 CL |
31 | #include "slab.h" |
32 | ||
33 | enum slab_state slab_state; | |
18004c5d CL |
34 | LIST_HEAD(slab_caches); |
35 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 36 | struct kmem_cache *kmem_cache; |
97d06609 | 37 | |
2d891fbc KC |
38 | #ifdef CONFIG_HARDENED_USERCOPY |
39 | bool usercopy_fallback __ro_after_init = | |
40 | IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK); | |
41 | module_param(usercopy_fallback, bool, 0400); | |
42 | MODULE_PARM_DESC(usercopy_fallback, | |
43 | "WARN instead of reject usercopy whitelist violations"); | |
44 | #endif | |
45 | ||
657dc2f9 TH |
46 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
47 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
48 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
49 | slab_caches_to_rcu_destroy_workfn); | |
50 | ||
423c929c JK |
51 | /* |
52 | * Set of flags that will prevent slab merging | |
53 | */ | |
54 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
5f0d5a3a | 55 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ |
7ed2f9e6 | 56 | SLAB_FAILSLAB | SLAB_KASAN) |
423c929c | 57 | |
230e9fc2 | 58 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
6d6ea1e9 | 59 | SLAB_CACHE_DMA32 | SLAB_ACCOUNT) |
423c929c JK |
60 | |
61 | /* | |
62 | * Merge control. If this is set then no merging of slab caches will occur. | |
423c929c | 63 | */ |
7660a6fd | 64 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); |
423c929c JK |
65 | |
66 | static int __init setup_slab_nomerge(char *str) | |
67 | { | |
7660a6fd | 68 | slab_nomerge = true; |
423c929c JK |
69 | return 1; |
70 | } | |
71 | ||
72 | #ifdef CONFIG_SLUB | |
73 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
74 | #endif | |
75 | ||
76 | __setup("slab_nomerge", setup_slab_nomerge); | |
77 | ||
07f361b2 JK |
78 | /* |
79 | * Determine the size of a slab object | |
80 | */ | |
81 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
82 | { | |
83 | return s->object_size; | |
84 | } | |
85 | EXPORT_SYMBOL(kmem_cache_size); | |
86 | ||
77be4b13 | 87 | #ifdef CONFIG_DEBUG_VM |
f4957d5b | 88 | static int kmem_cache_sanity_check(const char *name, unsigned int size) |
039363f3 | 89 | { |
039363f3 CL |
90 | if (!name || in_interrupt() || size < sizeof(void *) || |
91 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
92 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
93 | return -EINVAL; | |
039363f3 | 94 | } |
b920536a | 95 | |
20cea968 | 96 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
77be4b13 SK |
97 | return 0; |
98 | } | |
99 | #else | |
f4957d5b | 100 | static inline int kmem_cache_sanity_check(const char *name, unsigned int size) |
77be4b13 SK |
101 | { |
102 | return 0; | |
103 | } | |
20cea968 CL |
104 | #endif |
105 | ||
484748f0 CL |
106 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
107 | { | |
108 | size_t i; | |
109 | ||
ca257195 JDB |
110 | for (i = 0; i < nr; i++) { |
111 | if (s) | |
112 | kmem_cache_free(s, p[i]); | |
113 | else | |
114 | kfree(p[i]); | |
115 | } | |
484748f0 CL |
116 | } |
117 | ||
865762a8 | 118 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
119 | void **p) |
120 | { | |
121 | size_t i; | |
122 | ||
123 | for (i = 0; i < nr; i++) { | |
124 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
125 | if (!x) { | |
126 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 127 | return 0; |
484748f0 CL |
128 | } |
129 | } | |
865762a8 | 130 | return i; |
484748f0 CL |
131 | } |
132 | ||
84c07d11 | 133 | #ifdef CONFIG_MEMCG_KMEM |
510ded33 TH |
134 | |
135 | LIST_HEAD(slab_root_caches); | |
63b02ef7 | 136 | static DEFINE_SPINLOCK(memcg_kmem_wq_lock); |
510ded33 | 137 | |
f0a3a24b RG |
138 | static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref); |
139 | ||
f7ce3190 | 140 | void slab_init_memcg_params(struct kmem_cache *s) |
33a690c4 | 141 | { |
9eeadc8b | 142 | s->memcg_params.root_cache = NULL; |
f7ce3190 | 143 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, NULL); |
9eeadc8b | 144 | INIT_LIST_HEAD(&s->memcg_params.children); |
92ee383f | 145 | s->memcg_params.dying = false; |
f7ce3190 VD |
146 | } |
147 | ||
148 | static int init_memcg_params(struct kmem_cache *s, | |
c03914b7 | 149 | struct kmem_cache *root_cache) |
f7ce3190 VD |
150 | { |
151 | struct memcg_cache_array *arr; | |
33a690c4 | 152 | |
9eeadc8b | 153 | if (root_cache) { |
f0a3a24b RG |
154 | int ret = percpu_ref_init(&s->memcg_params.refcnt, |
155 | kmemcg_cache_shutdown, | |
156 | 0, GFP_KERNEL); | |
157 | if (ret) | |
158 | return ret; | |
159 | ||
f7ce3190 | 160 | s->memcg_params.root_cache = root_cache; |
9eeadc8b | 161 | INIT_LIST_HEAD(&s->memcg_params.children_node); |
bc2791f8 | 162 | INIT_LIST_HEAD(&s->memcg_params.kmem_caches_node); |
33a690c4 | 163 | return 0; |
f7ce3190 | 164 | } |
33a690c4 | 165 | |
f7ce3190 | 166 | slab_init_memcg_params(s); |
33a690c4 | 167 | |
f7ce3190 VD |
168 | if (!memcg_nr_cache_ids) |
169 | return 0; | |
33a690c4 | 170 | |
f80c7dab JW |
171 | arr = kvzalloc(sizeof(struct memcg_cache_array) + |
172 | memcg_nr_cache_ids * sizeof(void *), | |
173 | GFP_KERNEL); | |
f7ce3190 VD |
174 | if (!arr) |
175 | return -ENOMEM; | |
33a690c4 | 176 | |
f7ce3190 | 177 | RCU_INIT_POINTER(s->memcg_params.memcg_caches, arr); |
33a690c4 VD |
178 | return 0; |
179 | } | |
180 | ||
f7ce3190 | 181 | static void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 | 182 | { |
b749ecfa | 183 | if (is_root_cache(s)) { |
f80c7dab | 184 | kvfree(rcu_access_pointer(s->memcg_params.memcg_caches)); |
b749ecfa RG |
185 | } else { |
186 | mem_cgroup_put(s->memcg_params.memcg); | |
187 | WRITE_ONCE(s->memcg_params.memcg, NULL); | |
f0a3a24b | 188 | percpu_ref_exit(&s->memcg_params.refcnt); |
b749ecfa | 189 | } |
f80c7dab JW |
190 | } |
191 | ||
192 | static void free_memcg_params(struct rcu_head *rcu) | |
193 | { | |
194 | struct memcg_cache_array *old; | |
195 | ||
196 | old = container_of(rcu, struct memcg_cache_array, rcu); | |
197 | kvfree(old); | |
33a690c4 VD |
198 | } |
199 | ||
f7ce3190 | 200 | static int update_memcg_params(struct kmem_cache *s, int new_array_size) |
6f817f4c | 201 | { |
f7ce3190 | 202 | struct memcg_cache_array *old, *new; |
6f817f4c | 203 | |
f80c7dab JW |
204 | new = kvzalloc(sizeof(struct memcg_cache_array) + |
205 | new_array_size * sizeof(void *), GFP_KERNEL); | |
f7ce3190 | 206 | if (!new) |
6f817f4c VD |
207 | return -ENOMEM; |
208 | ||
f7ce3190 VD |
209 | old = rcu_dereference_protected(s->memcg_params.memcg_caches, |
210 | lockdep_is_held(&slab_mutex)); | |
211 | if (old) | |
212 | memcpy(new->entries, old->entries, | |
213 | memcg_nr_cache_ids * sizeof(void *)); | |
6f817f4c | 214 | |
f7ce3190 VD |
215 | rcu_assign_pointer(s->memcg_params.memcg_caches, new); |
216 | if (old) | |
f80c7dab | 217 | call_rcu(&old->rcu, free_memcg_params); |
6f817f4c VD |
218 | return 0; |
219 | } | |
220 | ||
55007d84 GC |
221 | int memcg_update_all_caches(int num_memcgs) |
222 | { | |
223 | struct kmem_cache *s; | |
224 | int ret = 0; | |
55007d84 | 225 | |
05257a1a | 226 | mutex_lock(&slab_mutex); |
510ded33 | 227 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
f7ce3190 | 228 | ret = update_memcg_params(s, num_memcgs); |
55007d84 | 229 | /* |
55007d84 GC |
230 | * Instead of freeing the memory, we'll just leave the caches |
231 | * up to this point in an updated state. | |
232 | */ | |
233 | if (ret) | |
05257a1a | 234 | break; |
55007d84 | 235 | } |
55007d84 GC |
236 | mutex_unlock(&slab_mutex); |
237 | return ret; | |
238 | } | |
657dc2f9 | 239 | |
c03914b7 | 240 | void memcg_link_cache(struct kmem_cache *s, struct mem_cgroup *memcg) |
657dc2f9 | 241 | { |
510ded33 TH |
242 | if (is_root_cache(s)) { |
243 | list_add(&s->root_caches_node, &slab_root_caches); | |
244 | } else { | |
f0a3a24b | 245 | css_get(&memcg->css); |
c03914b7 | 246 | s->memcg_params.memcg = memcg; |
510ded33 TH |
247 | list_add(&s->memcg_params.children_node, |
248 | &s->memcg_params.root_cache->memcg_params.children); | |
249 | list_add(&s->memcg_params.kmem_caches_node, | |
250 | &s->memcg_params.memcg->kmem_caches); | |
251 | } | |
252 | } | |
253 | ||
254 | static void memcg_unlink_cache(struct kmem_cache *s) | |
255 | { | |
256 | if (is_root_cache(s)) { | |
257 | list_del(&s->root_caches_node); | |
258 | } else { | |
259 | list_del(&s->memcg_params.children_node); | |
260 | list_del(&s->memcg_params.kmem_caches_node); | |
261 | } | |
657dc2f9 | 262 | } |
33a690c4 | 263 | #else |
f7ce3190 | 264 | static inline int init_memcg_params(struct kmem_cache *s, |
c03914b7 | 265 | struct kmem_cache *root_cache) |
33a690c4 VD |
266 | { |
267 | return 0; | |
268 | } | |
269 | ||
f7ce3190 | 270 | static inline void destroy_memcg_params(struct kmem_cache *s) |
33a690c4 VD |
271 | { |
272 | } | |
657dc2f9 | 273 | |
510ded33 | 274 | static inline void memcg_unlink_cache(struct kmem_cache *s) |
657dc2f9 TH |
275 | { |
276 | } | |
84c07d11 | 277 | #endif /* CONFIG_MEMCG_KMEM */ |
55007d84 | 278 | |
692ae74a BL |
279 | /* |
280 | * Figure out what the alignment of the objects will be given a set of | |
281 | * flags, a user specified alignment and the size of the objects. | |
282 | */ | |
f4957d5b AD |
283 | static unsigned int calculate_alignment(slab_flags_t flags, |
284 | unsigned int align, unsigned int size) | |
692ae74a BL |
285 | { |
286 | /* | |
287 | * If the user wants hardware cache aligned objects then follow that | |
288 | * suggestion if the object is sufficiently large. | |
289 | * | |
290 | * The hardware cache alignment cannot override the specified | |
291 | * alignment though. If that is greater then use it. | |
292 | */ | |
293 | if (flags & SLAB_HWCACHE_ALIGN) { | |
f4957d5b | 294 | unsigned int ralign; |
692ae74a BL |
295 | |
296 | ralign = cache_line_size(); | |
297 | while (size <= ralign / 2) | |
298 | ralign /= 2; | |
299 | align = max(align, ralign); | |
300 | } | |
301 | ||
302 | if (align < ARCH_SLAB_MINALIGN) | |
303 | align = ARCH_SLAB_MINALIGN; | |
304 | ||
305 | return ALIGN(align, sizeof(void *)); | |
306 | } | |
307 | ||
423c929c JK |
308 | /* |
309 | * Find a mergeable slab cache | |
310 | */ | |
311 | int slab_unmergeable(struct kmem_cache *s) | |
312 | { | |
313 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
314 | return 1; | |
315 | ||
316 | if (!is_root_cache(s)) | |
317 | return 1; | |
318 | ||
319 | if (s->ctor) | |
320 | return 1; | |
321 | ||
8eb8284b DW |
322 | if (s->usersize) |
323 | return 1; | |
324 | ||
423c929c JK |
325 | /* |
326 | * We may have set a slab to be unmergeable during bootstrap. | |
327 | */ | |
328 | if (s->refcount < 0) | |
329 | return 1; | |
330 | ||
d38a2b7a MS |
331 | #ifdef CONFIG_MEMCG_KMEM |
332 | /* | |
333 | * Skip the dying kmem_cache. | |
334 | */ | |
335 | if (s->memcg_params.dying) | |
336 | return 1; | |
337 | #endif | |
338 | ||
423c929c JK |
339 | return 0; |
340 | } | |
341 | ||
f4957d5b | 342 | struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, |
d50112ed | 343 | slab_flags_t flags, const char *name, void (*ctor)(void *)) |
423c929c JK |
344 | { |
345 | struct kmem_cache *s; | |
346 | ||
c6e28895 | 347 | if (slab_nomerge) |
423c929c JK |
348 | return NULL; |
349 | ||
350 | if (ctor) | |
351 | return NULL; | |
352 | ||
353 | size = ALIGN(size, sizeof(void *)); | |
354 | align = calculate_alignment(flags, align, size); | |
355 | size = ALIGN(size, align); | |
356 | flags = kmem_cache_flags(size, flags, name, NULL); | |
357 | ||
c6e28895 GM |
358 | if (flags & SLAB_NEVER_MERGE) |
359 | return NULL; | |
360 | ||
510ded33 | 361 | list_for_each_entry_reverse(s, &slab_root_caches, root_caches_node) { |
423c929c JK |
362 | if (slab_unmergeable(s)) |
363 | continue; | |
364 | ||
365 | if (size > s->size) | |
366 | continue; | |
367 | ||
368 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
369 | continue; | |
370 | /* | |
371 | * Check if alignment is compatible. | |
372 | * Courtesy of Adrian Drzewiecki | |
373 | */ | |
374 | if ((s->size & ~(align - 1)) != s->size) | |
375 | continue; | |
376 | ||
377 | if (s->size - size >= sizeof(void *)) | |
378 | continue; | |
379 | ||
95069ac8 JK |
380 | if (IS_ENABLED(CONFIG_SLAB) && align && |
381 | (align > s->align || s->align % align)) | |
382 | continue; | |
383 | ||
423c929c JK |
384 | return s; |
385 | } | |
386 | return NULL; | |
387 | } | |
388 | ||
c9a77a79 | 389 | static struct kmem_cache *create_cache(const char *name, |
613a5eb5 | 390 | unsigned int object_size, unsigned int align, |
7bbdb81e AD |
391 | slab_flags_t flags, unsigned int useroffset, |
392 | unsigned int usersize, void (*ctor)(void *), | |
c9a77a79 | 393 | struct mem_cgroup *memcg, struct kmem_cache *root_cache) |
794b1248 VD |
394 | { |
395 | struct kmem_cache *s; | |
396 | int err; | |
397 | ||
8eb8284b DW |
398 | if (WARN_ON(useroffset + usersize > object_size)) |
399 | useroffset = usersize = 0; | |
400 | ||
794b1248 VD |
401 | err = -ENOMEM; |
402 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
403 | if (!s) | |
404 | goto out; | |
405 | ||
406 | s->name = name; | |
613a5eb5 | 407 | s->size = s->object_size = object_size; |
794b1248 VD |
408 | s->align = align; |
409 | s->ctor = ctor; | |
8eb8284b DW |
410 | s->useroffset = useroffset; |
411 | s->usersize = usersize; | |
794b1248 | 412 | |
c03914b7 | 413 | err = init_memcg_params(s, root_cache); |
794b1248 VD |
414 | if (err) |
415 | goto out_free_cache; | |
416 | ||
417 | err = __kmem_cache_create(s, flags); | |
418 | if (err) | |
419 | goto out_free_cache; | |
420 | ||
421 | s->refcount = 1; | |
422 | list_add(&s->list, &slab_caches); | |
c03914b7 | 423 | memcg_link_cache(s, memcg); |
794b1248 VD |
424 | out: |
425 | if (err) | |
426 | return ERR_PTR(err); | |
427 | return s; | |
428 | ||
429 | out_free_cache: | |
f7ce3190 | 430 | destroy_memcg_params(s); |
7c4da061 | 431 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
432 | goto out; |
433 | } | |
45906855 | 434 | |
f496990f MR |
435 | /** |
436 | * kmem_cache_create_usercopy - Create a cache with a region suitable | |
437 | * for copying to userspace | |
77be4b13 SK |
438 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
439 | * @size: The size of objects to be created in this cache. | |
440 | * @align: The required alignment for the objects. | |
441 | * @flags: SLAB flags | |
8eb8284b DW |
442 | * @useroffset: Usercopy region offset |
443 | * @usersize: Usercopy region size | |
77be4b13 SK |
444 | * @ctor: A constructor for the objects. |
445 | * | |
77be4b13 SK |
446 | * Cannot be called within a interrupt, but can be interrupted. |
447 | * The @ctor is run when new pages are allocated by the cache. | |
448 | * | |
449 | * The flags are | |
450 | * | |
451 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
452 | * to catch references to uninitialised memory. | |
453 | * | |
f496990f | 454 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check |
77be4b13 SK |
455 | * for buffer overruns. |
456 | * | |
457 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
458 | * cacheline. This can be beneficial if you're counting cycles as closely | |
459 | * as davem. | |
f496990f MR |
460 | * |
461 | * Return: a pointer to the cache on success, NULL on failure. | |
77be4b13 | 462 | */ |
2633d7a0 | 463 | struct kmem_cache * |
f4957d5b AD |
464 | kmem_cache_create_usercopy(const char *name, |
465 | unsigned int size, unsigned int align, | |
7bbdb81e AD |
466 | slab_flags_t flags, |
467 | unsigned int useroffset, unsigned int usersize, | |
8eb8284b | 468 | void (*ctor)(void *)) |
77be4b13 | 469 | { |
40911a79 | 470 | struct kmem_cache *s = NULL; |
3dec16ea | 471 | const char *cache_name; |
3965fc36 | 472 | int err; |
039363f3 | 473 | |
77be4b13 | 474 | get_online_cpus(); |
03afc0e2 | 475 | get_online_mems(); |
05257a1a | 476 | memcg_get_cache_ids(); |
03afc0e2 | 477 | |
77be4b13 | 478 | mutex_lock(&slab_mutex); |
686d550d | 479 | |
794b1248 | 480 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 481 | if (err) { |
3965fc36 | 482 | goto out_unlock; |
3aa24f51 | 483 | } |
686d550d | 484 | |
e70954fd TG |
485 | /* Refuse requests with allocator specific flags */ |
486 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
487 | err = -EINVAL; | |
488 | goto out_unlock; | |
489 | } | |
490 | ||
d8843922 GC |
491 | /* |
492 | * Some allocators will constraint the set of valid flags to a subset | |
493 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
494 | * case, and we'll just provide them with a sanitized version of the | |
495 | * passed flags. | |
496 | */ | |
497 | flags &= CACHE_CREATE_MASK; | |
686d550d | 498 | |
8eb8284b DW |
499 | /* Fail closed on bad usersize of useroffset values. */ |
500 | if (WARN_ON(!usersize && useroffset) || | |
501 | WARN_ON(size < usersize || size - usersize < useroffset)) | |
502 | usersize = useroffset = 0; | |
503 | ||
504 | if (!usersize) | |
505 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
794b1248 | 506 | if (s) |
3965fc36 | 507 | goto out_unlock; |
2633d7a0 | 508 | |
3dec16ea | 509 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
510 | if (!cache_name) { |
511 | err = -ENOMEM; | |
512 | goto out_unlock; | |
513 | } | |
7c9adf5a | 514 | |
613a5eb5 | 515 | s = create_cache(cache_name, size, |
c9a77a79 | 516 | calculate_alignment(flags, align, size), |
8eb8284b | 517 | flags, useroffset, usersize, ctor, NULL, NULL); |
794b1248 VD |
518 | if (IS_ERR(s)) { |
519 | err = PTR_ERR(s); | |
3dec16ea | 520 | kfree_const(cache_name); |
794b1248 | 521 | } |
3965fc36 VD |
522 | |
523 | out_unlock: | |
20cea968 | 524 | mutex_unlock(&slab_mutex); |
03afc0e2 | 525 | |
05257a1a | 526 | memcg_put_cache_ids(); |
03afc0e2 | 527 | put_online_mems(); |
20cea968 CL |
528 | put_online_cpus(); |
529 | ||
ba3253c7 | 530 | if (err) { |
686d550d CL |
531 | if (flags & SLAB_PANIC) |
532 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
533 | name, err); | |
534 | else { | |
1170532b | 535 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
536 | name, err); |
537 | dump_stack(); | |
538 | } | |
686d550d CL |
539 | return NULL; |
540 | } | |
039363f3 CL |
541 | return s; |
542 | } | |
8eb8284b DW |
543 | EXPORT_SYMBOL(kmem_cache_create_usercopy); |
544 | ||
f496990f MR |
545 | /** |
546 | * kmem_cache_create - Create a cache. | |
547 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
548 | * @size: The size of objects to be created in this cache. | |
549 | * @align: The required alignment for the objects. | |
550 | * @flags: SLAB flags | |
551 | * @ctor: A constructor for the objects. | |
552 | * | |
553 | * Cannot be called within a interrupt, but can be interrupted. | |
554 | * The @ctor is run when new pages are allocated by the cache. | |
555 | * | |
556 | * The flags are | |
557 | * | |
558 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
559 | * to catch references to uninitialised memory. | |
560 | * | |
561 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check | |
562 | * for buffer overruns. | |
563 | * | |
564 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
565 | * cacheline. This can be beneficial if you're counting cycles as closely | |
566 | * as davem. | |
567 | * | |
568 | * Return: a pointer to the cache on success, NULL on failure. | |
569 | */ | |
8eb8284b | 570 | struct kmem_cache * |
f4957d5b | 571 | kmem_cache_create(const char *name, unsigned int size, unsigned int align, |
8eb8284b DW |
572 | slab_flags_t flags, void (*ctor)(void *)) |
573 | { | |
6d07d1cd | 574 | return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, |
8eb8284b DW |
575 | ctor); |
576 | } | |
794b1248 | 577 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 578 | |
657dc2f9 | 579 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 580 | { |
657dc2f9 TH |
581 | LIST_HEAD(to_destroy); |
582 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 583 | |
657dc2f9 | 584 | /* |
5f0d5a3a | 585 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
657dc2f9 TH |
586 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
587 | * through RCU and and the associated kmem_cache are dereferenced | |
588 | * while freeing the pages, so the kmem_caches should be freed only | |
589 | * after the pending RCU operations are finished. As rcu_barrier() | |
590 | * is a pretty slow operation, we batch all pending destructions | |
591 | * asynchronously. | |
592 | */ | |
593 | mutex_lock(&slab_mutex); | |
594 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
595 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 596 | |
657dc2f9 TH |
597 | if (list_empty(&to_destroy)) |
598 | return; | |
599 | ||
600 | rcu_barrier(); | |
601 | ||
602 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
603 | #ifdef SLAB_SUPPORTS_SYSFS | |
604 | sysfs_slab_release(s); | |
605 | #else | |
606 | slab_kmem_cache_release(s); | |
607 | #endif | |
608 | } | |
d5b3cf71 VD |
609 | } |
610 | ||
657dc2f9 | 611 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 612 | { |
f9fa1d91 GT |
613 | /* free asan quarantined objects */ |
614 | kasan_cache_shutdown(s); | |
615 | ||
657dc2f9 TH |
616 | if (__kmem_cache_shutdown(s) != 0) |
617 | return -EBUSY; | |
d5b3cf71 | 618 | |
510ded33 | 619 | memcg_unlink_cache(s); |
657dc2f9 | 620 | list_del(&s->list); |
d5b3cf71 | 621 | |
5f0d5a3a | 622 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
d50d82fa MP |
623 | #ifdef SLAB_SUPPORTS_SYSFS |
624 | sysfs_slab_unlink(s); | |
625 | #endif | |
657dc2f9 TH |
626 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
627 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
628 | } else { | |
d5b3cf71 | 629 | #ifdef SLAB_SUPPORTS_SYSFS |
d50d82fa | 630 | sysfs_slab_unlink(s); |
bf5eb3de | 631 | sysfs_slab_release(s); |
d5b3cf71 VD |
632 | #else |
633 | slab_kmem_cache_release(s); | |
634 | #endif | |
635 | } | |
657dc2f9 TH |
636 | |
637 | return 0; | |
d5b3cf71 VD |
638 | } |
639 | ||
84c07d11 | 640 | #ifdef CONFIG_MEMCG_KMEM |
794b1248 | 641 | /* |
776ed0f0 | 642 | * memcg_create_kmem_cache - Create a cache for a memory cgroup. |
794b1248 VD |
643 | * @memcg: The memory cgroup the new cache is for. |
644 | * @root_cache: The parent of the new cache. | |
645 | * | |
646 | * This function attempts to create a kmem cache that will serve allocation | |
647 | * requests going from @memcg to @root_cache. The new cache inherits properties | |
648 | * from its parent. | |
649 | */ | |
d5b3cf71 VD |
650 | void memcg_create_kmem_cache(struct mem_cgroup *memcg, |
651 | struct kmem_cache *root_cache) | |
2633d7a0 | 652 | { |
3e0350a3 | 653 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by slab_mutex */ |
33398cf2 | 654 | struct cgroup_subsys_state *css = &memcg->css; |
f7ce3190 | 655 | struct memcg_cache_array *arr; |
bd673145 | 656 | struct kmem_cache *s = NULL; |
794b1248 | 657 | char *cache_name; |
f7ce3190 | 658 | int idx; |
794b1248 VD |
659 | |
660 | get_online_cpus(); | |
03afc0e2 VD |
661 | get_online_mems(); |
662 | ||
794b1248 VD |
663 | mutex_lock(&slab_mutex); |
664 | ||
2a4db7eb | 665 | /* |
567e9ab2 | 666 | * The memory cgroup could have been offlined while the cache |
2a4db7eb VD |
667 | * creation work was pending. |
668 | */ | |
57033297 | 669 | if (memcg->kmem_state != KMEM_ONLINE) |
2a4db7eb VD |
670 | goto out_unlock; |
671 | ||
f7ce3190 VD |
672 | idx = memcg_cache_id(memcg); |
673 | arr = rcu_dereference_protected(root_cache->memcg_params.memcg_caches, | |
674 | lockdep_is_held(&slab_mutex)); | |
675 | ||
d5b3cf71 VD |
676 | /* |
677 | * Since per-memcg caches are created asynchronously on first | |
678 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
679 | * create the same cache, but only one of them may succeed. | |
680 | */ | |
f7ce3190 | 681 | if (arr->entries[idx]) |
d5b3cf71 VD |
682 | goto out_unlock; |
683 | ||
f1008365 | 684 | cgroup_name(css->cgroup, memcg_name_buf, sizeof(memcg_name_buf)); |
73f576c0 JW |
685 | cache_name = kasprintf(GFP_KERNEL, "%s(%llu:%s)", root_cache->name, |
686 | css->serial_nr, memcg_name_buf); | |
794b1248 VD |
687 | if (!cache_name) |
688 | goto out_unlock; | |
689 | ||
c9a77a79 | 690 | s = create_cache(cache_name, root_cache->object_size, |
613a5eb5 | 691 | root_cache->align, |
f773e36d | 692 | root_cache->flags & CACHE_CREATE_MASK, |
8eb8284b | 693 | root_cache->useroffset, root_cache->usersize, |
f773e36d | 694 | root_cache->ctor, memcg, root_cache); |
d5b3cf71 VD |
695 | /* |
696 | * If we could not create a memcg cache, do not complain, because | |
697 | * that's not critical at all as we can always proceed with the root | |
698 | * cache. | |
699 | */ | |
bd673145 | 700 | if (IS_ERR(s)) { |
794b1248 | 701 | kfree(cache_name); |
d5b3cf71 | 702 | goto out_unlock; |
bd673145 | 703 | } |
794b1248 | 704 | |
d5b3cf71 | 705 | /* |
f0a3a24b | 706 | * Since readers won't lock (see memcg_kmem_get_cache()), we need a |
d5b3cf71 VD |
707 | * barrier here to ensure nobody will see the kmem_cache partially |
708 | * initialized. | |
709 | */ | |
710 | smp_wmb(); | |
f7ce3190 | 711 | arr->entries[idx] = s; |
d5b3cf71 | 712 | |
794b1248 VD |
713 | out_unlock: |
714 | mutex_unlock(&slab_mutex); | |
03afc0e2 VD |
715 | |
716 | put_online_mems(); | |
794b1248 | 717 | put_online_cpus(); |
2633d7a0 | 718 | } |
b8529907 | 719 | |
0b14e8aa | 720 | static void kmemcg_workfn(struct work_struct *work) |
01fb58bc TH |
721 | { |
722 | struct kmem_cache *s = container_of(work, struct kmem_cache, | |
0b14e8aa | 723 | memcg_params.work); |
01fb58bc TH |
724 | |
725 | get_online_cpus(); | |
726 | get_online_mems(); | |
727 | ||
728 | mutex_lock(&slab_mutex); | |
0b14e8aa | 729 | s->memcg_params.work_fn(s); |
01fb58bc TH |
730 | mutex_unlock(&slab_mutex); |
731 | ||
732 | put_online_mems(); | |
733 | put_online_cpus(); | |
01fb58bc TH |
734 | } |
735 | ||
0b14e8aa | 736 | static void kmemcg_rcufn(struct rcu_head *head) |
01fb58bc TH |
737 | { |
738 | struct kmem_cache *s = container_of(head, struct kmem_cache, | |
0b14e8aa | 739 | memcg_params.rcu_head); |
01fb58bc TH |
740 | |
741 | /* | |
0b14e8aa | 742 | * We need to grab blocking locks. Bounce to ->work. The |
01fb58bc | 743 | * work item shares the space with the RCU head and can't be |
b991cee5 | 744 | * initialized earlier. |
01fb58bc | 745 | */ |
0b14e8aa RG |
746 | INIT_WORK(&s->memcg_params.work, kmemcg_workfn); |
747 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.work); | |
01fb58bc TH |
748 | } |
749 | ||
f0a3a24b RG |
750 | static void kmemcg_cache_shutdown_fn(struct kmem_cache *s) |
751 | { | |
752 | WARN_ON(shutdown_cache(s)); | |
753 | } | |
754 | ||
755 | static void kmemcg_cache_shutdown(struct percpu_ref *percpu_ref) | |
756 | { | |
757 | struct kmem_cache *s = container_of(percpu_ref, struct kmem_cache, | |
758 | memcg_params.refcnt); | |
759 | unsigned long flags; | |
760 | ||
761 | spin_lock_irqsave(&memcg_kmem_wq_lock, flags); | |
762 | if (s->memcg_params.root_cache->memcg_params.dying) | |
763 | goto unlock; | |
764 | ||
765 | s->memcg_params.work_fn = kmemcg_cache_shutdown_fn; | |
766 | INIT_WORK(&s->memcg_params.work, kmemcg_workfn); | |
767 | queue_work(memcg_kmem_cache_wq, &s->memcg_params.work); | |
768 | ||
769 | unlock: | |
770 | spin_unlock_irqrestore(&memcg_kmem_wq_lock, flags); | |
771 | } | |
772 | ||
773 | static void kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s) | |
774 | { | |
775 | __kmemcg_cache_deactivate_after_rcu(s); | |
776 | percpu_ref_kill(&s->memcg_params.refcnt); | |
777 | } | |
778 | ||
43486694 | 779 | static void kmemcg_cache_deactivate(struct kmem_cache *s) |
01fb58bc | 780 | { |
f0a3a24b | 781 | if (WARN_ON_ONCE(is_root_cache(s))) |
01fb58bc TH |
782 | return; |
783 | ||
43486694 | 784 | __kmemcg_cache_deactivate(s); |
fcf8a1e4 | 785 | s->flags |= SLAB_DEACTIVATED; |
43486694 | 786 | |
63b02ef7 RG |
787 | /* |
788 | * memcg_kmem_wq_lock is used to synchronize memcg_params.dying | |
789 | * flag and make sure that no new kmem_cache deactivation tasks | |
790 | * are queued (see flush_memcg_workqueue() ). | |
791 | */ | |
792 | spin_lock_irq(&memcg_kmem_wq_lock); | |
92ee383f | 793 | if (s->memcg_params.root_cache->memcg_params.dying) |
63b02ef7 | 794 | goto unlock; |
92ee383f | 795 | |
f0a3a24b | 796 | s->memcg_params.work_fn = kmemcg_cache_deactivate_after_rcu; |
0b14e8aa | 797 | call_rcu(&s->memcg_params.rcu_head, kmemcg_rcufn); |
63b02ef7 RG |
798 | unlock: |
799 | spin_unlock_irq(&memcg_kmem_wq_lock); | |
01fb58bc TH |
800 | } |
801 | ||
fb2f2b0a RG |
802 | void memcg_deactivate_kmem_caches(struct mem_cgroup *memcg, |
803 | struct mem_cgroup *parent) | |
2a4db7eb VD |
804 | { |
805 | int idx; | |
806 | struct memcg_cache_array *arr; | |
d6e0b7fa | 807 | struct kmem_cache *s, *c; |
fb2f2b0a | 808 | unsigned int nr_reparented; |
2a4db7eb VD |
809 | |
810 | idx = memcg_cache_id(memcg); | |
811 | ||
d6e0b7fa VD |
812 | get_online_cpus(); |
813 | get_online_mems(); | |
814 | ||
2a4db7eb | 815 | mutex_lock(&slab_mutex); |
510ded33 | 816 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { |
2a4db7eb VD |
817 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, |
818 | lockdep_is_held(&slab_mutex)); | |
d6e0b7fa VD |
819 | c = arr->entries[idx]; |
820 | if (!c) | |
821 | continue; | |
822 | ||
43486694 | 823 | kmemcg_cache_deactivate(c); |
2a4db7eb VD |
824 | arr->entries[idx] = NULL; |
825 | } | |
fb2f2b0a RG |
826 | nr_reparented = 0; |
827 | list_for_each_entry(s, &memcg->kmem_caches, | |
828 | memcg_params.kmem_caches_node) { | |
829 | WRITE_ONCE(s->memcg_params.memcg, parent); | |
830 | css_put(&memcg->css); | |
831 | nr_reparented++; | |
832 | } | |
833 | if (nr_reparented) { | |
834 | list_splice_init(&memcg->kmem_caches, | |
835 | &parent->kmem_caches); | |
836 | css_get_many(&parent->css, nr_reparented); | |
837 | } | |
2a4db7eb | 838 | mutex_unlock(&slab_mutex); |
d6e0b7fa VD |
839 | |
840 | put_online_mems(); | |
841 | put_online_cpus(); | |
2a4db7eb VD |
842 | } |
843 | ||
657dc2f9 | 844 | static int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
845 | { |
846 | struct memcg_cache_array *arr; | |
847 | struct kmem_cache *c, *c2; | |
848 | LIST_HEAD(busy); | |
849 | int i; | |
850 | ||
851 | BUG_ON(!is_root_cache(s)); | |
852 | ||
853 | /* | |
854 | * First, shutdown active caches, i.e. caches that belong to online | |
855 | * memory cgroups. | |
856 | */ | |
857 | arr = rcu_dereference_protected(s->memcg_params.memcg_caches, | |
858 | lockdep_is_held(&slab_mutex)); | |
859 | for_each_memcg_cache_index(i) { | |
860 | c = arr->entries[i]; | |
861 | if (!c) | |
862 | continue; | |
657dc2f9 | 863 | if (shutdown_cache(c)) |
d60fdcc9 VD |
864 | /* |
865 | * The cache still has objects. Move it to a temporary | |
866 | * list so as not to try to destroy it for a second | |
867 | * time while iterating over inactive caches below. | |
868 | */ | |
9eeadc8b | 869 | list_move(&c->memcg_params.children_node, &busy); |
d60fdcc9 VD |
870 | else |
871 | /* | |
872 | * The cache is empty and will be destroyed soon. Clear | |
873 | * the pointer to it in the memcg_caches array so that | |
874 | * it will never be accessed even if the root cache | |
875 | * stays alive. | |
876 | */ | |
877 | arr->entries[i] = NULL; | |
878 | } | |
879 | ||
880 | /* | |
881 | * Second, shutdown all caches left from memory cgroups that are now | |
882 | * offline. | |
883 | */ | |
9eeadc8b TH |
884 | list_for_each_entry_safe(c, c2, &s->memcg_params.children, |
885 | memcg_params.children_node) | |
657dc2f9 | 886 | shutdown_cache(c); |
d60fdcc9 | 887 | |
9eeadc8b | 888 | list_splice(&busy, &s->memcg_params.children); |
d60fdcc9 VD |
889 | |
890 | /* | |
891 | * A cache being destroyed must be empty. In particular, this means | |
892 | * that all per memcg caches attached to it must be empty too. | |
893 | */ | |
9eeadc8b | 894 | if (!list_empty(&s->memcg_params.children)) |
d60fdcc9 VD |
895 | return -EBUSY; |
896 | return 0; | |
897 | } | |
92ee383f | 898 | |
d38a2b7a | 899 | static void memcg_set_kmem_cache_dying(struct kmem_cache *s) |
92ee383f | 900 | { |
63b02ef7 | 901 | spin_lock_irq(&memcg_kmem_wq_lock); |
92ee383f | 902 | s->memcg_params.dying = true; |
63b02ef7 | 903 | spin_unlock_irq(&memcg_kmem_wq_lock); |
d38a2b7a | 904 | } |
92ee383f | 905 | |
d38a2b7a MS |
906 | static void flush_memcg_workqueue(struct kmem_cache *s) |
907 | { | |
92ee383f | 908 | /* |
43486694 | 909 | * SLAB and SLUB deactivate the kmem_caches through call_rcu. Make |
92ee383f SB |
910 | * sure all registered rcu callbacks have been invoked. |
911 | */ | |
43486694 | 912 | rcu_barrier(); |
92ee383f SB |
913 | |
914 | /* | |
915 | * SLAB and SLUB create memcg kmem_caches through workqueue and SLUB | |
916 | * deactivates the memcg kmem_caches through workqueue. Make sure all | |
917 | * previous workitems on workqueue are processed. | |
918 | */ | |
2fe20210 AH |
919 | if (likely(memcg_kmem_cache_wq)) |
920 | flush_workqueue(memcg_kmem_cache_wq); | |
a264df74 RG |
921 | |
922 | /* | |
923 | * If we're racing with children kmem_cache deactivation, it might | |
924 | * take another rcu grace period to complete their destruction. | |
925 | * At this moment the corresponding percpu_ref_kill() call should be | |
926 | * done, but it might take another rcu grace period to complete | |
927 | * switching to the atomic mode. | |
928 | * Please, note that we check without grabbing the slab_mutex. It's safe | |
929 | * because at this moment the children list can't grow. | |
930 | */ | |
931 | if (!list_empty(&s->memcg_params.children)) | |
932 | rcu_barrier(); | |
92ee383f | 933 | } |
d60fdcc9 | 934 | #else |
657dc2f9 | 935 | static inline int shutdown_memcg_caches(struct kmem_cache *s) |
d60fdcc9 VD |
936 | { |
937 | return 0; | |
938 | } | |
84c07d11 | 939 | #endif /* CONFIG_MEMCG_KMEM */ |
97d06609 | 940 | |
41a21285 CL |
941 | void slab_kmem_cache_release(struct kmem_cache *s) |
942 | { | |
52b4b950 | 943 | __kmem_cache_release(s); |
f7ce3190 | 944 | destroy_memcg_params(s); |
3dec16ea | 945 | kfree_const(s->name); |
41a21285 CL |
946 | kmem_cache_free(kmem_cache, s); |
947 | } | |
948 | ||
945cf2b6 CL |
949 | void kmem_cache_destroy(struct kmem_cache *s) |
950 | { | |
d60fdcc9 | 951 | int err; |
d5b3cf71 | 952 | |
3942d299 SS |
953 | if (unlikely(!s)) |
954 | return; | |
955 | ||
945cf2b6 | 956 | get_online_cpus(); |
03afc0e2 VD |
957 | get_online_mems(); |
958 | ||
945cf2b6 | 959 | mutex_lock(&slab_mutex); |
b8529907 | 960 | |
945cf2b6 | 961 | s->refcount--; |
b8529907 VD |
962 | if (s->refcount) |
963 | goto out_unlock; | |
964 | ||
d38a2b7a MS |
965 | #ifdef CONFIG_MEMCG_KMEM |
966 | memcg_set_kmem_cache_dying(s); | |
967 | ||
968 | mutex_unlock(&slab_mutex); | |
969 | ||
970 | put_online_mems(); | |
971 | put_online_cpus(); | |
972 | ||
973 | flush_memcg_workqueue(s); | |
974 | ||
975 | get_online_cpus(); | |
976 | get_online_mems(); | |
977 | ||
978 | mutex_lock(&slab_mutex); | |
979 | #endif | |
980 | ||
657dc2f9 | 981 | err = shutdown_memcg_caches(s); |
d60fdcc9 | 982 | if (!err) |
657dc2f9 | 983 | err = shutdown_cache(s); |
b8529907 | 984 | |
cd918c55 | 985 | if (err) { |
756a025f JP |
986 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
987 | s->name); | |
cd918c55 VD |
988 | dump_stack(); |
989 | } | |
b8529907 VD |
990 | out_unlock: |
991 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 992 | |
03afc0e2 | 993 | put_online_mems(); |
945cf2b6 CL |
994 | put_online_cpus(); |
995 | } | |
996 | EXPORT_SYMBOL(kmem_cache_destroy); | |
997 | ||
03afc0e2 VD |
998 | /** |
999 | * kmem_cache_shrink - Shrink a cache. | |
1000 | * @cachep: The cache to shrink. | |
1001 | * | |
1002 | * Releases as many slabs as possible for a cache. | |
1003 | * To help debugging, a zero exit status indicates all slabs were released. | |
a862f68a MR |
1004 | * |
1005 | * Return: %0 if all slabs were released, non-zero otherwise | |
03afc0e2 VD |
1006 | */ |
1007 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
1008 | { | |
1009 | int ret; | |
1010 | ||
1011 | get_online_cpus(); | |
1012 | get_online_mems(); | |
55834c59 | 1013 | kasan_cache_shrink(cachep); |
c9fc5864 | 1014 | ret = __kmem_cache_shrink(cachep); |
03afc0e2 VD |
1015 | put_online_mems(); |
1016 | put_online_cpus(); | |
1017 | return ret; | |
1018 | } | |
1019 | EXPORT_SYMBOL(kmem_cache_shrink); | |
1020 | ||
04f768a3 WL |
1021 | /** |
1022 | * kmem_cache_shrink_all - shrink a cache and all memcg caches for root cache | |
1023 | * @s: The cache pointer | |
1024 | */ | |
1025 | void kmem_cache_shrink_all(struct kmem_cache *s) | |
1026 | { | |
1027 | struct kmem_cache *c; | |
1028 | ||
1029 | if (!IS_ENABLED(CONFIG_MEMCG_KMEM) || !is_root_cache(s)) { | |
1030 | kmem_cache_shrink(s); | |
1031 | return; | |
1032 | } | |
1033 | ||
1034 | get_online_cpus(); | |
1035 | get_online_mems(); | |
1036 | kasan_cache_shrink(s); | |
1037 | __kmem_cache_shrink(s); | |
1038 | ||
1039 | /* | |
1040 | * We have to take the slab_mutex to protect from the memcg list | |
1041 | * modification. | |
1042 | */ | |
1043 | mutex_lock(&slab_mutex); | |
1044 | for_each_memcg_cache(c, s) { | |
1045 | /* | |
1046 | * Don't need to shrink deactivated memcg caches. | |
1047 | */ | |
1048 | if (s->flags & SLAB_DEACTIVATED) | |
1049 | continue; | |
1050 | kasan_cache_shrink(c); | |
1051 | __kmem_cache_shrink(c); | |
1052 | } | |
1053 | mutex_unlock(&slab_mutex); | |
1054 | put_online_mems(); | |
1055 | put_online_cpus(); | |
1056 | } | |
1057 | ||
fda90124 | 1058 | bool slab_is_available(void) |
97d06609 CL |
1059 | { |
1060 | return slab_state >= UP; | |
1061 | } | |
b7454ad3 | 1062 | |
45530c44 CL |
1063 | #ifndef CONFIG_SLOB |
1064 | /* Create a cache during boot when no slab services are available yet */ | |
361d575e AD |
1065 | void __init create_boot_cache(struct kmem_cache *s, const char *name, |
1066 | unsigned int size, slab_flags_t flags, | |
1067 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
1068 | { |
1069 | int err; | |
59bb4798 | 1070 | unsigned int align = ARCH_KMALLOC_MINALIGN; |
45530c44 CL |
1071 | |
1072 | s->name = name; | |
1073 | s->size = s->object_size = size; | |
59bb4798 VB |
1074 | |
1075 | /* | |
1076 | * For power of two sizes, guarantee natural alignment for kmalloc | |
1077 | * caches, regardless of SL*B debugging options. | |
1078 | */ | |
1079 | if (is_power_of_2(size)) | |
1080 | align = max(align, size); | |
1081 | s->align = calculate_alignment(flags, align, size); | |
1082 | ||
8eb8284b DW |
1083 | s->useroffset = useroffset; |
1084 | s->usersize = usersize; | |
f7ce3190 VD |
1085 | |
1086 | slab_init_memcg_params(s); | |
1087 | ||
45530c44 CL |
1088 | err = __kmem_cache_create(s, flags); |
1089 | ||
1090 | if (err) | |
361d575e | 1091 | panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", |
45530c44 CL |
1092 | name, size, err); |
1093 | ||
1094 | s->refcount = -1; /* Exempt from merging for now */ | |
1095 | } | |
1096 | ||
55de8b9c AD |
1097 | struct kmem_cache *__init create_kmalloc_cache(const char *name, |
1098 | unsigned int size, slab_flags_t flags, | |
1099 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
1100 | { |
1101 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
1102 | ||
1103 | if (!s) | |
1104 | panic("Out of memory when creating slab %s\n", name); | |
1105 | ||
6c0c21ad | 1106 | create_boot_cache(s, name, size, flags, useroffset, usersize); |
45530c44 | 1107 | list_add(&s->list, &slab_caches); |
c03914b7 | 1108 | memcg_link_cache(s, NULL); |
45530c44 CL |
1109 | s->refcount = 1; |
1110 | return s; | |
1111 | } | |
1112 | ||
cc252eae | 1113 | struct kmem_cache * |
a07057dc AB |
1114 | kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = |
1115 | { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; | |
9425c58e CL |
1116 | EXPORT_SYMBOL(kmalloc_caches); |
1117 | ||
2c59dd65 CL |
1118 | /* |
1119 | * Conversion table for small slabs sizes / 8 to the index in the | |
1120 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
1121 | * of two cache sizes there. The size of larger slabs can be determined using | |
1122 | * fls. | |
1123 | */ | |
d5f86655 | 1124 | static u8 size_index[24] __ro_after_init = { |
2c59dd65 CL |
1125 | 3, /* 8 */ |
1126 | 4, /* 16 */ | |
1127 | 5, /* 24 */ | |
1128 | 5, /* 32 */ | |
1129 | 6, /* 40 */ | |
1130 | 6, /* 48 */ | |
1131 | 6, /* 56 */ | |
1132 | 6, /* 64 */ | |
1133 | 1, /* 72 */ | |
1134 | 1, /* 80 */ | |
1135 | 1, /* 88 */ | |
1136 | 1, /* 96 */ | |
1137 | 7, /* 104 */ | |
1138 | 7, /* 112 */ | |
1139 | 7, /* 120 */ | |
1140 | 7, /* 128 */ | |
1141 | 2, /* 136 */ | |
1142 | 2, /* 144 */ | |
1143 | 2, /* 152 */ | |
1144 | 2, /* 160 */ | |
1145 | 2, /* 168 */ | |
1146 | 2, /* 176 */ | |
1147 | 2, /* 184 */ | |
1148 | 2 /* 192 */ | |
1149 | }; | |
1150 | ||
ac914d08 | 1151 | static inline unsigned int size_index_elem(unsigned int bytes) |
2c59dd65 CL |
1152 | { |
1153 | return (bytes - 1) / 8; | |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * Find the kmem_cache structure that serves a given size of | |
1158 | * allocation | |
1159 | */ | |
1160 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
1161 | { | |
d5f86655 | 1162 | unsigned int index; |
2c59dd65 CL |
1163 | |
1164 | if (size <= 192) { | |
1165 | if (!size) | |
1166 | return ZERO_SIZE_PTR; | |
1167 | ||
1168 | index = size_index[size_index_elem(size)]; | |
61448479 | 1169 | } else { |
221d7da6 | 1170 | if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) |
61448479 | 1171 | return NULL; |
2c59dd65 | 1172 | index = fls(size - 1); |
61448479 | 1173 | } |
2c59dd65 | 1174 | |
cc252eae | 1175 | return kmalloc_caches[kmalloc_type(flags)][index]; |
2c59dd65 CL |
1176 | } |
1177 | ||
cb5d9fb3 PL |
1178 | #ifdef CONFIG_ZONE_DMA |
1179 | #define INIT_KMALLOC_INFO(__size, __short_size) \ | |
1180 | { \ | |
1181 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ | |
1182 | .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ | |
1183 | .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \ | |
1184 | .size = __size, \ | |
1185 | } | |
1186 | #else | |
1187 | #define INIT_KMALLOC_INFO(__size, __short_size) \ | |
1188 | { \ | |
1189 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ | |
1190 | .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ | |
1191 | .size = __size, \ | |
1192 | } | |
1193 | #endif | |
1194 | ||
4066c33d GG |
1195 | /* |
1196 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
1197 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
1198 | * kmalloc-67108864. | |
1199 | */ | |
af3b5f87 | 1200 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
cb5d9fb3 PL |
1201 | INIT_KMALLOC_INFO(0, 0), |
1202 | INIT_KMALLOC_INFO(96, 96), | |
1203 | INIT_KMALLOC_INFO(192, 192), | |
1204 | INIT_KMALLOC_INFO(8, 8), | |
1205 | INIT_KMALLOC_INFO(16, 16), | |
1206 | INIT_KMALLOC_INFO(32, 32), | |
1207 | INIT_KMALLOC_INFO(64, 64), | |
1208 | INIT_KMALLOC_INFO(128, 128), | |
1209 | INIT_KMALLOC_INFO(256, 256), | |
1210 | INIT_KMALLOC_INFO(512, 512), | |
1211 | INIT_KMALLOC_INFO(1024, 1k), | |
1212 | INIT_KMALLOC_INFO(2048, 2k), | |
1213 | INIT_KMALLOC_INFO(4096, 4k), | |
1214 | INIT_KMALLOC_INFO(8192, 8k), | |
1215 | INIT_KMALLOC_INFO(16384, 16k), | |
1216 | INIT_KMALLOC_INFO(32768, 32k), | |
1217 | INIT_KMALLOC_INFO(65536, 64k), | |
1218 | INIT_KMALLOC_INFO(131072, 128k), | |
1219 | INIT_KMALLOC_INFO(262144, 256k), | |
1220 | INIT_KMALLOC_INFO(524288, 512k), | |
1221 | INIT_KMALLOC_INFO(1048576, 1M), | |
1222 | INIT_KMALLOC_INFO(2097152, 2M), | |
1223 | INIT_KMALLOC_INFO(4194304, 4M), | |
1224 | INIT_KMALLOC_INFO(8388608, 8M), | |
1225 | INIT_KMALLOC_INFO(16777216, 16M), | |
1226 | INIT_KMALLOC_INFO(33554432, 32M), | |
1227 | INIT_KMALLOC_INFO(67108864, 64M) | |
4066c33d GG |
1228 | }; |
1229 | ||
f97d5f63 | 1230 | /* |
34cc6990 DS |
1231 | * Patch up the size_index table if we have strange large alignment |
1232 | * requirements for the kmalloc array. This is only the case for | |
1233 | * MIPS it seems. The standard arches will not generate any code here. | |
1234 | * | |
1235 | * Largest permitted alignment is 256 bytes due to the way we | |
1236 | * handle the index determination for the smaller caches. | |
1237 | * | |
1238 | * Make sure that nothing crazy happens if someone starts tinkering | |
1239 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 1240 | */ |
34cc6990 | 1241 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 | 1242 | { |
ac914d08 | 1243 | unsigned int i; |
f97d5f63 | 1244 | |
2c59dd65 CL |
1245 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
1246 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
1247 | ||
1248 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
ac914d08 | 1249 | unsigned int elem = size_index_elem(i); |
2c59dd65 CL |
1250 | |
1251 | if (elem >= ARRAY_SIZE(size_index)) | |
1252 | break; | |
1253 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
1254 | } | |
1255 | ||
1256 | if (KMALLOC_MIN_SIZE >= 64) { | |
1257 | /* | |
1258 | * The 96 byte size cache is not used if the alignment | |
1259 | * is 64 byte. | |
1260 | */ | |
1261 | for (i = 64 + 8; i <= 96; i += 8) | |
1262 | size_index[size_index_elem(i)] = 7; | |
1263 | ||
1264 | } | |
1265 | ||
1266 | if (KMALLOC_MIN_SIZE >= 128) { | |
1267 | /* | |
1268 | * The 192 byte sized cache is not used if the alignment | |
1269 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
1270 | * instead. | |
1271 | */ | |
1272 | for (i = 128 + 8; i <= 192; i += 8) | |
1273 | size_index[size_index_elem(i)] = 8; | |
1274 | } | |
34cc6990 DS |
1275 | } |
1276 | ||
1291523f | 1277 | static void __init |
13657d0a | 1278 | new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) |
a9730fca | 1279 | { |
cb5d9fb3 | 1280 | if (type == KMALLOC_RECLAIM) |
1291523f | 1281 | flags |= SLAB_RECLAIM_ACCOUNT; |
1291523f | 1282 | |
cb5d9fb3 PL |
1283 | kmalloc_caches[type][idx] = create_kmalloc_cache( |
1284 | kmalloc_info[idx].name[type], | |
6c0c21ad DW |
1285 | kmalloc_info[idx].size, flags, 0, |
1286 | kmalloc_info[idx].size); | |
a9730fca CL |
1287 | } |
1288 | ||
34cc6990 DS |
1289 | /* |
1290 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
1291 | * may already have been created because they were needed to | |
1292 | * enable allocations for slab creation. | |
1293 | */ | |
d50112ed | 1294 | void __init create_kmalloc_caches(slab_flags_t flags) |
34cc6990 | 1295 | { |
13657d0a PL |
1296 | int i; |
1297 | enum kmalloc_cache_type type; | |
34cc6990 | 1298 | |
1291523f VB |
1299 | for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { |
1300 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { | |
1301 | if (!kmalloc_caches[type][i]) | |
1302 | new_kmalloc_cache(i, type, flags); | |
f97d5f63 | 1303 | |
1291523f VB |
1304 | /* |
1305 | * Caches that are not of the two-to-the-power-of size. | |
1306 | * These have to be created immediately after the | |
1307 | * earlier power of two caches | |
1308 | */ | |
1309 | if (KMALLOC_MIN_SIZE <= 32 && i == 6 && | |
1310 | !kmalloc_caches[type][1]) | |
1311 | new_kmalloc_cache(1, type, flags); | |
1312 | if (KMALLOC_MIN_SIZE <= 64 && i == 7 && | |
1313 | !kmalloc_caches[type][2]) | |
1314 | new_kmalloc_cache(2, type, flags); | |
1315 | } | |
8a965b3b CL |
1316 | } |
1317 | ||
f97d5f63 CL |
1318 | /* Kmalloc array is now usable */ |
1319 | slab_state = UP; | |
1320 | ||
f97d5f63 CL |
1321 | #ifdef CONFIG_ZONE_DMA |
1322 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
cc252eae | 1323 | struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; |
f97d5f63 CL |
1324 | |
1325 | if (s) { | |
cc252eae | 1326 | kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( |
cb5d9fb3 | 1327 | kmalloc_info[i].name[KMALLOC_DMA], |
dc0a7f75 | 1328 | kmalloc_info[i].size, |
49f2d241 VB |
1329 | SLAB_CACHE_DMA | flags, 0, |
1330 | kmalloc_info[i].size); | |
f97d5f63 CL |
1331 | } |
1332 | } | |
1333 | #endif | |
1334 | } | |
45530c44 CL |
1335 | #endif /* !CONFIG_SLOB */ |
1336 | ||
44405099 LL |
1337 | gfp_t kmalloc_fix_flags(gfp_t flags) |
1338 | { | |
1339 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; | |
1340 | ||
1341 | flags &= ~GFP_SLAB_BUG_MASK; | |
1342 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1343 | invalid_mask, &invalid_mask, flags, &flags); | |
1344 | dump_stack(); | |
1345 | ||
1346 | return flags; | |
1347 | } | |
1348 | ||
cea371f4 VD |
1349 | /* |
1350 | * To avoid unnecessary overhead, we pass through large allocation requests | |
1351 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
1352 | * know the allocation order to free the pages properly in kfree. | |
1353 | */ | |
52383431 VD |
1354 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
1355 | { | |
6a486c0a | 1356 | void *ret = NULL; |
52383431 VD |
1357 | struct page *page; |
1358 | ||
44405099 LL |
1359 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1360 | flags = kmalloc_fix_flags(flags); | |
1361 | ||
52383431 | 1362 | flags |= __GFP_COMP; |
4949148a | 1363 | page = alloc_pages(flags, order); |
6a486c0a VB |
1364 | if (likely(page)) { |
1365 | ret = page_address(page); | |
1366 | mod_node_page_state(page_pgdat(page), NR_SLAB_UNRECLAIMABLE, | |
1367 | 1 << order); | |
1368 | } | |
0116523c | 1369 | ret = kasan_kmalloc_large(ret, size, flags); |
a2f77575 | 1370 | /* As ret might get tagged, call kmemleak hook after KASAN. */ |
53128245 | 1371 | kmemleak_alloc(ret, size, 1, flags); |
52383431 VD |
1372 | return ret; |
1373 | } | |
1374 | EXPORT_SYMBOL(kmalloc_order); | |
1375 | ||
f1b6eb6e CL |
1376 | #ifdef CONFIG_TRACING |
1377 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
1378 | { | |
1379 | void *ret = kmalloc_order(size, flags, order); | |
1380 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
1381 | return ret; | |
1382 | } | |
1383 | EXPORT_SYMBOL(kmalloc_order_trace); | |
1384 | #endif | |
45530c44 | 1385 | |
7c00fce9 TG |
1386 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1387 | /* Randomize a generic freelist */ | |
1388 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
302d55d5 | 1389 | unsigned int count) |
7c00fce9 | 1390 | { |
7c00fce9 | 1391 | unsigned int rand; |
302d55d5 | 1392 | unsigned int i; |
7c00fce9 TG |
1393 | |
1394 | for (i = 0; i < count; i++) | |
1395 | list[i] = i; | |
1396 | ||
1397 | /* Fisher-Yates shuffle */ | |
1398 | for (i = count - 1; i > 0; i--) { | |
1399 | rand = prandom_u32_state(state); | |
1400 | rand %= (i + 1); | |
1401 | swap(list[i], list[rand]); | |
1402 | } | |
1403 | } | |
1404 | ||
1405 | /* Create a random sequence per cache */ | |
1406 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1407 | gfp_t gfp) | |
1408 | { | |
1409 | struct rnd_state state; | |
1410 | ||
1411 | if (count < 2 || cachep->random_seq) | |
1412 | return 0; | |
1413 | ||
1414 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1415 | if (!cachep->random_seq) | |
1416 | return -ENOMEM; | |
1417 | ||
1418 | /* Get best entropy at this stage of boot */ | |
1419 | prandom_seed_state(&state, get_random_long()); | |
1420 | ||
1421 | freelist_randomize(&state, cachep->random_seq, count); | |
1422 | return 0; | |
1423 | } | |
1424 | ||
1425 | /* Destroy the per-cache random freelist sequence */ | |
1426 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1427 | { | |
1428 | kfree(cachep->random_seq); | |
1429 | cachep->random_seq = NULL; | |
1430 | } | |
1431 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1432 | ||
5b365771 | 1433 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
e9b4db2b | 1434 | #ifdef CONFIG_SLAB |
0825a6f9 | 1435 | #define SLABINFO_RIGHTS (0600) |
e9b4db2b | 1436 | #else |
0825a6f9 | 1437 | #define SLABINFO_RIGHTS (0400) |
e9b4db2b WL |
1438 | #endif |
1439 | ||
b047501c | 1440 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1441 | { |
1442 | /* | |
1443 | * Output format version, so at least we can change it | |
1444 | * without _too_ many complaints. | |
1445 | */ | |
1446 | #ifdef CONFIG_DEBUG_SLAB | |
1447 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1448 | #else | |
1449 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1450 | #endif | |
756a025f | 1451 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1452 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1453 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1454 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1455 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1456 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1457 | #endif | |
1458 | seq_putc(m, '\n'); | |
1459 | } | |
1460 | ||
1df3b26f | 1461 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1462 | { |
b7454ad3 | 1463 | mutex_lock(&slab_mutex); |
510ded33 | 1464 | return seq_list_start(&slab_root_caches, *pos); |
b7454ad3 GC |
1465 | } |
1466 | ||
276a2439 | 1467 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 | 1468 | { |
510ded33 | 1469 | return seq_list_next(p, &slab_root_caches, pos); |
b7454ad3 GC |
1470 | } |
1471 | ||
276a2439 | 1472 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1473 | { |
1474 | mutex_unlock(&slab_mutex); | |
1475 | } | |
1476 | ||
749c5415 GC |
1477 | static void |
1478 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
1479 | { | |
1480 | struct kmem_cache *c; | |
1481 | struct slabinfo sinfo; | |
749c5415 GC |
1482 | |
1483 | if (!is_root_cache(s)) | |
1484 | return; | |
1485 | ||
426589f5 | 1486 | for_each_memcg_cache(c, s) { |
749c5415 GC |
1487 | memset(&sinfo, 0, sizeof(sinfo)); |
1488 | get_slabinfo(c, &sinfo); | |
1489 | ||
1490 | info->active_slabs += sinfo.active_slabs; | |
1491 | info->num_slabs += sinfo.num_slabs; | |
1492 | info->shared_avail += sinfo.shared_avail; | |
1493 | info->active_objs += sinfo.active_objs; | |
1494 | info->num_objs += sinfo.num_objs; | |
1495 | } | |
1496 | } | |
1497 | ||
b047501c | 1498 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1499 | { |
0d7561c6 GC |
1500 | struct slabinfo sinfo; |
1501 | ||
1502 | memset(&sinfo, 0, sizeof(sinfo)); | |
1503 | get_slabinfo(s, &sinfo); | |
1504 | ||
749c5415 GC |
1505 | memcg_accumulate_slabinfo(s, &sinfo); |
1506 | ||
0d7561c6 | 1507 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 1508 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1509 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1510 | ||
1511 | seq_printf(m, " : tunables %4u %4u %4u", | |
1512 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1513 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1514 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1515 | slabinfo_show_stats(m, s); | |
1516 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1517 | } |
1518 | ||
1df3b26f | 1519 | static int slab_show(struct seq_file *m, void *p) |
749c5415 | 1520 | { |
510ded33 | 1521 | struct kmem_cache *s = list_entry(p, struct kmem_cache, root_caches_node); |
749c5415 | 1522 | |
510ded33 | 1523 | if (p == slab_root_caches.next) |
1df3b26f | 1524 | print_slabinfo_header(m); |
510ded33 | 1525 | cache_show(s, m); |
b047501c VD |
1526 | return 0; |
1527 | } | |
1528 | ||
852d8be0 YS |
1529 | void dump_unreclaimable_slab(void) |
1530 | { | |
1531 | struct kmem_cache *s, *s2; | |
1532 | struct slabinfo sinfo; | |
1533 | ||
1534 | /* | |
1535 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1536 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1537 | * risk of crash. | |
1538 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1539 | * without acquiring the mutex. | |
1540 | */ | |
1541 | if (!mutex_trylock(&slab_mutex)) { | |
1542 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1543 | return; | |
1544 | } | |
1545 | ||
1546 | pr_info("Unreclaimable slab info:\n"); | |
1547 | pr_info("Name Used Total\n"); | |
1548 | ||
1549 | list_for_each_entry_safe(s, s2, &slab_caches, list) { | |
1550 | if (!is_root_cache(s) || (s->flags & SLAB_RECLAIM_ACCOUNT)) | |
1551 | continue; | |
1552 | ||
1553 | get_slabinfo(s, &sinfo); | |
1554 | ||
1555 | if (sinfo.num_objs > 0) | |
1556 | pr_info("%-17s %10luKB %10luKB\n", cache_name(s), | |
1557 | (sinfo.active_objs * s->size) / 1024, | |
1558 | (sinfo.num_objs * s->size) / 1024); | |
1559 | } | |
1560 | mutex_unlock(&slab_mutex); | |
1561 | } | |
1562 | ||
a87425a3 | 1563 | #if defined(CONFIG_MEMCG_KMEM) |
bc2791f8 TH |
1564 | void *memcg_slab_start(struct seq_file *m, loff_t *pos) |
1565 | { | |
aa9694bb | 1566 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
bc2791f8 TH |
1567 | |
1568 | mutex_lock(&slab_mutex); | |
1569 | return seq_list_start(&memcg->kmem_caches, *pos); | |
1570 | } | |
1571 | ||
1572 | void *memcg_slab_next(struct seq_file *m, void *p, loff_t *pos) | |
1573 | { | |
aa9694bb | 1574 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
bc2791f8 TH |
1575 | |
1576 | return seq_list_next(p, &memcg->kmem_caches, pos); | |
1577 | } | |
1578 | ||
1579 | void memcg_slab_stop(struct seq_file *m, void *p) | |
1580 | { | |
1581 | mutex_unlock(&slab_mutex); | |
1582 | } | |
1583 | ||
b047501c VD |
1584 | int memcg_slab_show(struct seq_file *m, void *p) |
1585 | { | |
bc2791f8 TH |
1586 | struct kmem_cache *s = list_entry(p, struct kmem_cache, |
1587 | memcg_params.kmem_caches_node); | |
aa9694bb | 1588 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
b047501c | 1589 | |
bc2791f8 | 1590 | if (p == memcg->kmem_caches.next) |
b047501c | 1591 | print_slabinfo_header(m); |
bc2791f8 | 1592 | cache_show(s, m); |
b047501c | 1593 | return 0; |
749c5415 | 1594 | } |
b047501c | 1595 | #endif |
749c5415 | 1596 | |
b7454ad3 GC |
1597 | /* |
1598 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1599 | * | |
1600 | * Output layout: | |
1601 | * cache-name | |
1602 | * num-active-objs | |
1603 | * total-objs | |
1604 | * object size | |
1605 | * num-active-slabs | |
1606 | * total-slabs | |
1607 | * num-pages-per-slab | |
1608 | * + further values on SMP and with statistics enabled | |
1609 | */ | |
1610 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1611 | .start = slab_start, |
276a2439 WL |
1612 | .next = slab_next, |
1613 | .stop = slab_stop, | |
1df3b26f | 1614 | .show = slab_show, |
b7454ad3 GC |
1615 | }; |
1616 | ||
1617 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1618 | { | |
1619 | return seq_open(file, &slabinfo_op); | |
1620 | } | |
1621 | ||
97a32539 | 1622 | static const struct proc_ops slabinfo_proc_ops = { |
d919b33d | 1623 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
1624 | .proc_open = slabinfo_open, |
1625 | .proc_read = seq_read, | |
1626 | .proc_write = slabinfo_write, | |
1627 | .proc_lseek = seq_lseek, | |
1628 | .proc_release = seq_release, | |
b7454ad3 GC |
1629 | }; |
1630 | ||
1631 | static int __init slab_proc_init(void) | |
1632 | { | |
97a32539 | 1633 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); |
b7454ad3 GC |
1634 | return 0; |
1635 | } | |
1636 | module_init(slab_proc_init); | |
fcf8a1e4 WL |
1637 | |
1638 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_MEMCG_KMEM) | |
1639 | /* | |
1640 | * Display information about kmem caches that have child memcg caches. | |
1641 | */ | |
1642 | static int memcg_slabinfo_show(struct seq_file *m, void *unused) | |
1643 | { | |
1644 | struct kmem_cache *s, *c; | |
1645 | struct slabinfo sinfo; | |
1646 | ||
1647 | mutex_lock(&slab_mutex); | |
1648 | seq_puts(m, "# <name> <css_id[:dead|deact]> <active_objs> <num_objs>"); | |
1649 | seq_puts(m, " <active_slabs> <num_slabs>\n"); | |
1650 | list_for_each_entry(s, &slab_root_caches, root_caches_node) { | |
1651 | /* | |
1652 | * Skip kmem caches that don't have any memcg children. | |
1653 | */ | |
1654 | if (list_empty(&s->memcg_params.children)) | |
1655 | continue; | |
1656 | ||
1657 | memset(&sinfo, 0, sizeof(sinfo)); | |
1658 | get_slabinfo(s, &sinfo); | |
1659 | seq_printf(m, "%-17s root %6lu %6lu %6lu %6lu\n", | |
1660 | cache_name(s), sinfo.active_objs, sinfo.num_objs, | |
1661 | sinfo.active_slabs, sinfo.num_slabs); | |
1662 | ||
1663 | for_each_memcg_cache(c, s) { | |
1664 | struct cgroup_subsys_state *css; | |
1665 | char *status = ""; | |
1666 | ||
1667 | css = &c->memcg_params.memcg->css; | |
1668 | if (!(css->flags & CSS_ONLINE)) | |
1669 | status = ":dead"; | |
1670 | else if (c->flags & SLAB_DEACTIVATED) | |
1671 | status = ":deact"; | |
1672 | ||
1673 | memset(&sinfo, 0, sizeof(sinfo)); | |
1674 | get_slabinfo(c, &sinfo); | |
1675 | seq_printf(m, "%-17s %4d%-6s %6lu %6lu %6lu %6lu\n", | |
1676 | cache_name(c), css->id, status, | |
1677 | sinfo.active_objs, sinfo.num_objs, | |
1678 | sinfo.active_slabs, sinfo.num_slabs); | |
1679 | } | |
1680 | } | |
1681 | mutex_unlock(&slab_mutex); | |
1682 | return 0; | |
1683 | } | |
1684 | DEFINE_SHOW_ATTRIBUTE(memcg_slabinfo); | |
1685 | ||
1686 | static int __init memcg_slabinfo_init(void) | |
1687 | { | |
1688 | debugfs_create_file("memcg_slabinfo", S_IFREG | S_IRUGO, | |
1689 | NULL, NULL, &memcg_slabinfo_fops); | |
1690 | return 0; | |
1691 | } | |
1692 | ||
1693 | late_initcall(memcg_slabinfo_init); | |
1694 | #endif /* CONFIG_DEBUG_FS && CONFIG_MEMCG_KMEM */ | |
5b365771 | 1695 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ |
928cec9c AR |
1696 | |
1697 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1698 | gfp_t flags) | |
1699 | { | |
1700 | void *ret; | |
fa9ba3aa | 1701 | size_t ks; |
928cec9c | 1702 | |
fa9ba3aa | 1703 | ks = ksize(p); |
928cec9c | 1704 | |
0316bec2 | 1705 | if (ks >= new_size) { |
0116523c | 1706 | p = kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1707 | return (void *)p; |
0316bec2 | 1708 | } |
928cec9c AR |
1709 | |
1710 | ret = kmalloc_track_caller(new_size, flags); | |
1711 | if (ret && p) | |
1712 | memcpy(ret, p, ks); | |
1713 | ||
1714 | return ret; | |
1715 | } | |
1716 | ||
928cec9c AR |
1717 | /** |
1718 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1719 | * @p: object to reallocate memory for. | |
1720 | * @new_size: how many bytes of memory are required. | |
1721 | * @flags: the type of memory to allocate. | |
1722 | * | |
1723 | * The contents of the object pointed to are preserved up to the | |
1724 | * lesser of the new and old sizes. If @p is %NULL, krealloc() | |
1725 | * behaves exactly like kmalloc(). If @new_size is 0 and @p is not a | |
1726 | * %NULL pointer, the object pointed to is freed. | |
a862f68a MR |
1727 | * |
1728 | * Return: pointer to the allocated memory or %NULL in case of error | |
928cec9c AR |
1729 | */ |
1730 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1731 | { | |
1732 | void *ret; | |
1733 | ||
1734 | if (unlikely(!new_size)) { | |
1735 | kfree(p); | |
1736 | return ZERO_SIZE_PTR; | |
1737 | } | |
1738 | ||
1739 | ret = __do_krealloc(p, new_size, flags); | |
772a2fa5 | 1740 | if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) |
928cec9c AR |
1741 | kfree(p); |
1742 | ||
1743 | return ret; | |
1744 | } | |
1745 | EXPORT_SYMBOL(krealloc); | |
1746 | ||
1747 | /** | |
453431a5 | 1748 | * kfree_sensitive - Clear sensitive information in memory before freeing |
928cec9c AR |
1749 | * @p: object to free memory of |
1750 | * | |
1751 | * The memory of the object @p points to is zeroed before freed. | |
453431a5 | 1752 | * If @p is %NULL, kfree_sensitive() does nothing. |
928cec9c AR |
1753 | * |
1754 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1755 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1756 | * careful when using this function in performance sensitive code. | |
1757 | */ | |
453431a5 | 1758 | void kfree_sensitive(const void *p) |
928cec9c AR |
1759 | { |
1760 | size_t ks; | |
1761 | void *mem = (void *)p; | |
1762 | ||
928cec9c | 1763 | ks = ksize(mem); |
fa9ba3aa WK |
1764 | if (ks) |
1765 | memzero_explicit(mem, ks); | |
928cec9c AR |
1766 | kfree(mem); |
1767 | } | |
453431a5 | 1768 | EXPORT_SYMBOL(kfree_sensitive); |
928cec9c | 1769 | |
10d1f8cb ME |
1770 | /** |
1771 | * ksize - get the actual amount of memory allocated for a given object | |
1772 | * @objp: Pointer to the object | |
1773 | * | |
1774 | * kmalloc may internally round up allocations and return more memory | |
1775 | * than requested. ksize() can be used to determine the actual amount of | |
1776 | * memory allocated. The caller may use this additional memory, even though | |
1777 | * a smaller amount of memory was initially specified with the kmalloc call. | |
1778 | * The caller must guarantee that objp points to a valid object previously | |
1779 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
1780 | * must not be freed during the duration of the call. | |
1781 | * | |
1782 | * Return: size of the actual memory used by @objp in bytes | |
1783 | */ | |
1784 | size_t ksize(const void *objp) | |
1785 | { | |
0d4ca4c9 ME |
1786 | size_t size; |
1787 | ||
0d4ca4c9 ME |
1788 | /* |
1789 | * We need to check that the pointed to object is valid, and only then | |
1790 | * unpoison the shadow memory below. We use __kasan_check_read(), to | |
1791 | * generate a more useful report at the time ksize() is called (rather | |
1792 | * than later where behaviour is undefined due to potential | |
1793 | * use-after-free or double-free). | |
1794 | * | |
1795 | * If the pointed to memory is invalid we return 0, to avoid users of | |
1796 | * ksize() writing to and potentially corrupting the memory region. | |
1797 | * | |
1798 | * We want to perform the check before __ksize(), to avoid potentially | |
1799 | * crashing in __ksize() due to accessing invalid metadata. | |
1800 | */ | |
fa9ba3aa | 1801 | if (unlikely(ZERO_OR_NULL_PTR(objp)) || !__kasan_check_read(objp, 1)) |
0d4ca4c9 ME |
1802 | return 0; |
1803 | ||
1804 | size = __ksize(objp); | |
10d1f8cb ME |
1805 | /* |
1806 | * We assume that ksize callers could use whole allocated area, | |
1807 | * so we need to unpoison this area. | |
1808 | */ | |
1809 | kasan_unpoison_shadow(objp, size); | |
1810 | return size; | |
1811 | } | |
1812 | EXPORT_SYMBOL(ksize); | |
1813 | ||
928cec9c AR |
1814 | /* Tracepoints definitions. */ |
1815 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1816 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1817 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1818 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1819 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1820 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); | |
4f6923fb HM |
1821 | |
1822 | int should_failslab(struct kmem_cache *s, gfp_t gfpflags) | |
1823 | { | |
1824 | if (__should_failslab(s, gfpflags)) | |
1825 | return -ENOMEM; | |
1826 | return 0; | |
1827 | } | |
1828 | ALLOW_ERROR_INJECTION(should_failslab, ERRNO); |