]>
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 | 14 | #include <linux/compiler.h> |
d3fb45f3 | 15 | #include <linux/kfence.h> |
039363f3 | 16 | #include <linux/module.h> |
20cea968 CL |
17 | #include <linux/cpu.h> |
18 | #include <linux/uaccess.h> | |
b7454ad3 GC |
19 | #include <linux/seq_file.h> |
20 | #include <linux/proc_fs.h> | |
fcf8a1e4 | 21 | #include <linux/debugfs.h> |
e86f8b09 | 22 | #include <linux/kasan.h> |
039363f3 CL |
23 | #include <asm/cacheflush.h> |
24 | #include <asm/tlbflush.h> | |
25 | #include <asm/page.h> | |
2633d7a0 | 26 | #include <linux/memcontrol.h> |
928cec9c AR |
27 | |
28 | #define CREATE_TRACE_POINTS | |
f1b6eb6e | 29 | #include <trace/events/kmem.h> |
039363f3 | 30 | |
44405099 LL |
31 | #include "internal.h" |
32 | ||
97d06609 CL |
33 | #include "slab.h" |
34 | ||
35 | enum slab_state slab_state; | |
18004c5d CL |
36 | LIST_HEAD(slab_caches); |
37 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 38 | struct kmem_cache *kmem_cache; |
97d06609 | 39 | |
2d891fbc KC |
40 | #ifdef CONFIG_HARDENED_USERCOPY |
41 | bool usercopy_fallback __ro_after_init = | |
42 | IS_ENABLED(CONFIG_HARDENED_USERCOPY_FALLBACK); | |
43 | module_param(usercopy_fallback, bool, 0400); | |
44 | MODULE_PARM_DESC(usercopy_fallback, | |
45 | "WARN instead of reject usercopy whitelist violations"); | |
46 | #endif | |
47 | ||
657dc2f9 TH |
48 | static LIST_HEAD(slab_caches_to_rcu_destroy); |
49 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work); | |
50 | static DECLARE_WORK(slab_caches_to_rcu_destroy_work, | |
51 | slab_caches_to_rcu_destroy_workfn); | |
52 | ||
423c929c JK |
53 | /* |
54 | * Set of flags that will prevent slab merging | |
55 | */ | |
56 | #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
5f0d5a3a | 57 | SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \ |
e86f8b09 | 58 | SLAB_FAILSLAB | kasan_never_merge()) |
423c929c | 59 | |
230e9fc2 | 60 | #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \ |
6d6ea1e9 | 61 | SLAB_CACHE_DMA32 | SLAB_ACCOUNT) |
423c929c JK |
62 | |
63 | /* | |
64 | * Merge control. If this is set then no merging of slab caches will occur. | |
423c929c | 65 | */ |
7660a6fd | 66 | static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT); |
423c929c JK |
67 | |
68 | static int __init setup_slab_nomerge(char *str) | |
69 | { | |
7660a6fd | 70 | slab_nomerge = true; |
423c929c JK |
71 | return 1; |
72 | } | |
73 | ||
82edd9d5 RA |
74 | static int __init setup_slab_merge(char *str) |
75 | { | |
76 | slab_nomerge = false; | |
77 | return 1; | |
78 | } | |
79 | ||
423c929c JK |
80 | #ifdef CONFIG_SLUB |
81 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
82edd9d5 | 82 | __setup_param("slub_merge", slub_merge, setup_slab_merge, 0); |
423c929c JK |
83 | #endif |
84 | ||
85 | __setup("slab_nomerge", setup_slab_nomerge); | |
82edd9d5 | 86 | __setup("slab_merge", setup_slab_merge); |
423c929c | 87 | |
07f361b2 JK |
88 | /* |
89 | * Determine the size of a slab object | |
90 | */ | |
91 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
92 | { | |
93 | return s->object_size; | |
94 | } | |
95 | EXPORT_SYMBOL(kmem_cache_size); | |
96 | ||
77be4b13 | 97 | #ifdef CONFIG_DEBUG_VM |
f4957d5b | 98 | static int kmem_cache_sanity_check(const char *name, unsigned int size) |
039363f3 | 99 | { |
74c1d3e0 | 100 | if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) { |
77be4b13 SK |
101 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
102 | return -EINVAL; | |
039363f3 | 103 | } |
b920536a | 104 | |
20cea968 | 105 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
77be4b13 SK |
106 | return 0; |
107 | } | |
108 | #else | |
f4957d5b | 109 | static inline int kmem_cache_sanity_check(const char *name, unsigned int size) |
77be4b13 SK |
110 | { |
111 | return 0; | |
112 | } | |
20cea968 CL |
113 | #endif |
114 | ||
484748f0 CL |
115 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
116 | { | |
117 | size_t i; | |
118 | ||
ca257195 JDB |
119 | for (i = 0; i < nr; i++) { |
120 | if (s) | |
121 | kmem_cache_free(s, p[i]); | |
122 | else | |
123 | kfree(p[i]); | |
124 | } | |
484748f0 CL |
125 | } |
126 | ||
865762a8 | 127 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
128 | void **p) |
129 | { | |
130 | size_t i; | |
131 | ||
132 | for (i = 0; i < nr; i++) { | |
133 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
134 | if (!x) { | |
135 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 136 | return 0; |
484748f0 CL |
137 | } |
138 | } | |
865762a8 | 139 | return i; |
484748f0 CL |
140 | } |
141 | ||
692ae74a BL |
142 | /* |
143 | * Figure out what the alignment of the objects will be given a set of | |
144 | * flags, a user specified alignment and the size of the objects. | |
145 | */ | |
f4957d5b AD |
146 | static unsigned int calculate_alignment(slab_flags_t flags, |
147 | unsigned int align, unsigned int size) | |
692ae74a BL |
148 | { |
149 | /* | |
150 | * If the user wants hardware cache aligned objects then follow that | |
151 | * suggestion if the object is sufficiently large. | |
152 | * | |
153 | * The hardware cache alignment cannot override the specified | |
154 | * alignment though. If that is greater then use it. | |
155 | */ | |
156 | if (flags & SLAB_HWCACHE_ALIGN) { | |
f4957d5b | 157 | unsigned int ralign; |
692ae74a BL |
158 | |
159 | ralign = cache_line_size(); | |
160 | while (size <= ralign / 2) | |
161 | ralign /= 2; | |
162 | align = max(align, ralign); | |
163 | } | |
164 | ||
165 | if (align < ARCH_SLAB_MINALIGN) | |
166 | align = ARCH_SLAB_MINALIGN; | |
167 | ||
168 | return ALIGN(align, sizeof(void *)); | |
169 | } | |
170 | ||
423c929c JK |
171 | /* |
172 | * Find a mergeable slab cache | |
173 | */ | |
174 | int slab_unmergeable(struct kmem_cache *s) | |
175 | { | |
176 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
177 | return 1; | |
178 | ||
423c929c JK |
179 | if (s->ctor) |
180 | return 1; | |
181 | ||
8eb8284b DW |
182 | if (s->usersize) |
183 | return 1; | |
184 | ||
423c929c JK |
185 | /* |
186 | * We may have set a slab to be unmergeable during bootstrap. | |
187 | */ | |
188 | if (s->refcount < 0) | |
189 | return 1; | |
190 | ||
191 | return 0; | |
192 | } | |
193 | ||
f4957d5b | 194 | struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, |
d50112ed | 195 | slab_flags_t flags, const char *name, void (*ctor)(void *)) |
423c929c JK |
196 | { |
197 | struct kmem_cache *s; | |
198 | ||
c6e28895 | 199 | if (slab_nomerge) |
423c929c JK |
200 | return NULL; |
201 | ||
202 | if (ctor) | |
203 | return NULL; | |
204 | ||
205 | size = ALIGN(size, sizeof(void *)); | |
206 | align = calculate_alignment(flags, align, size); | |
207 | size = ALIGN(size, align); | |
37540008 | 208 | flags = kmem_cache_flags(size, flags, name); |
423c929c | 209 | |
c6e28895 GM |
210 | if (flags & SLAB_NEVER_MERGE) |
211 | return NULL; | |
212 | ||
c7094406 | 213 | list_for_each_entry_reverse(s, &slab_caches, list) { |
423c929c JK |
214 | if (slab_unmergeable(s)) |
215 | continue; | |
216 | ||
217 | if (size > s->size) | |
218 | continue; | |
219 | ||
220 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
221 | continue; | |
222 | /* | |
223 | * Check if alignment is compatible. | |
224 | * Courtesy of Adrian Drzewiecki | |
225 | */ | |
226 | if ((s->size & ~(align - 1)) != s->size) | |
227 | continue; | |
228 | ||
229 | if (s->size - size >= sizeof(void *)) | |
230 | continue; | |
231 | ||
95069ac8 JK |
232 | if (IS_ENABLED(CONFIG_SLAB) && align && |
233 | (align > s->align || s->align % align)) | |
234 | continue; | |
235 | ||
423c929c JK |
236 | return s; |
237 | } | |
238 | return NULL; | |
239 | } | |
240 | ||
c9a77a79 | 241 | static struct kmem_cache *create_cache(const char *name, |
613a5eb5 | 242 | unsigned int object_size, unsigned int align, |
7bbdb81e AD |
243 | slab_flags_t flags, unsigned int useroffset, |
244 | unsigned int usersize, void (*ctor)(void *), | |
9855609b | 245 | struct kmem_cache *root_cache) |
794b1248 VD |
246 | { |
247 | struct kmem_cache *s; | |
248 | int err; | |
249 | ||
8eb8284b DW |
250 | if (WARN_ON(useroffset + usersize > object_size)) |
251 | useroffset = usersize = 0; | |
252 | ||
794b1248 VD |
253 | err = -ENOMEM; |
254 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
255 | if (!s) | |
256 | goto out; | |
257 | ||
258 | s->name = name; | |
613a5eb5 | 259 | s->size = s->object_size = object_size; |
794b1248 VD |
260 | s->align = align; |
261 | s->ctor = ctor; | |
8eb8284b DW |
262 | s->useroffset = useroffset; |
263 | s->usersize = usersize; | |
794b1248 | 264 | |
794b1248 VD |
265 | err = __kmem_cache_create(s, flags); |
266 | if (err) | |
267 | goto out_free_cache; | |
268 | ||
269 | s->refcount = 1; | |
270 | list_add(&s->list, &slab_caches); | |
794b1248 VD |
271 | out: |
272 | if (err) | |
273 | return ERR_PTR(err); | |
274 | return s; | |
275 | ||
276 | out_free_cache: | |
7c4da061 | 277 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
278 | goto out; |
279 | } | |
45906855 | 280 | |
f496990f MR |
281 | /** |
282 | * kmem_cache_create_usercopy - Create a cache with a region suitable | |
283 | * for copying to userspace | |
77be4b13 SK |
284 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
285 | * @size: The size of objects to be created in this cache. | |
286 | * @align: The required alignment for the objects. | |
287 | * @flags: SLAB flags | |
8eb8284b DW |
288 | * @useroffset: Usercopy region offset |
289 | * @usersize: Usercopy region size | |
77be4b13 SK |
290 | * @ctor: A constructor for the objects. |
291 | * | |
77be4b13 SK |
292 | * Cannot be called within a interrupt, but can be interrupted. |
293 | * The @ctor is run when new pages are allocated by the cache. | |
294 | * | |
295 | * The flags are | |
296 | * | |
297 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
298 | * to catch references to uninitialised memory. | |
299 | * | |
f496990f | 300 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check |
77be4b13 SK |
301 | * for buffer overruns. |
302 | * | |
303 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
304 | * cacheline. This can be beneficial if you're counting cycles as closely | |
305 | * as davem. | |
f496990f MR |
306 | * |
307 | * Return: a pointer to the cache on success, NULL on failure. | |
77be4b13 | 308 | */ |
2633d7a0 | 309 | struct kmem_cache * |
f4957d5b AD |
310 | kmem_cache_create_usercopy(const char *name, |
311 | unsigned int size, unsigned int align, | |
7bbdb81e AD |
312 | slab_flags_t flags, |
313 | unsigned int useroffset, unsigned int usersize, | |
8eb8284b | 314 | void (*ctor)(void *)) |
77be4b13 | 315 | { |
40911a79 | 316 | struct kmem_cache *s = NULL; |
3dec16ea | 317 | const char *cache_name; |
3965fc36 | 318 | int err; |
039363f3 | 319 | |
afe0c26d VB |
320 | #ifdef CONFIG_SLUB_DEBUG |
321 | /* | |
322 | * If no slub_debug was enabled globally, the static key is not yet | |
323 | * enabled by setup_slub_debug(). Enable it if the cache is being | |
324 | * created with any of the debugging flags passed explicitly. | |
325 | */ | |
326 | if (flags & SLAB_DEBUG_FLAGS) | |
327 | static_branch_enable(&slub_debug_enabled); | |
328 | #endif | |
329 | ||
77be4b13 | 330 | mutex_lock(&slab_mutex); |
686d550d | 331 | |
794b1248 | 332 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 333 | if (err) { |
3965fc36 | 334 | goto out_unlock; |
3aa24f51 | 335 | } |
686d550d | 336 | |
e70954fd TG |
337 | /* Refuse requests with allocator specific flags */ |
338 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
339 | err = -EINVAL; | |
340 | goto out_unlock; | |
341 | } | |
342 | ||
d8843922 GC |
343 | /* |
344 | * Some allocators will constraint the set of valid flags to a subset | |
345 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
346 | * case, and we'll just provide them with a sanitized version of the | |
347 | * passed flags. | |
348 | */ | |
349 | flags &= CACHE_CREATE_MASK; | |
686d550d | 350 | |
8eb8284b DW |
351 | /* Fail closed on bad usersize of useroffset values. */ |
352 | if (WARN_ON(!usersize && useroffset) || | |
353 | WARN_ON(size < usersize || size - usersize < useroffset)) | |
354 | usersize = useroffset = 0; | |
355 | ||
356 | if (!usersize) | |
357 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
794b1248 | 358 | if (s) |
3965fc36 | 359 | goto out_unlock; |
2633d7a0 | 360 | |
3dec16ea | 361 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
362 | if (!cache_name) { |
363 | err = -ENOMEM; | |
364 | goto out_unlock; | |
365 | } | |
7c9adf5a | 366 | |
613a5eb5 | 367 | s = create_cache(cache_name, size, |
c9a77a79 | 368 | calculate_alignment(flags, align, size), |
9855609b | 369 | flags, useroffset, usersize, ctor, NULL); |
794b1248 VD |
370 | if (IS_ERR(s)) { |
371 | err = PTR_ERR(s); | |
3dec16ea | 372 | kfree_const(cache_name); |
794b1248 | 373 | } |
3965fc36 VD |
374 | |
375 | out_unlock: | |
20cea968 | 376 | mutex_unlock(&slab_mutex); |
03afc0e2 | 377 | |
ba3253c7 | 378 | if (err) { |
686d550d | 379 | if (flags & SLAB_PANIC) |
4acaa7d5 | 380 | panic("%s: Failed to create slab '%s'. Error %d\n", |
381 | __func__, name, err); | |
686d550d | 382 | else { |
4acaa7d5 | 383 | pr_warn("%s(%s) failed with error %d\n", |
384 | __func__, name, err); | |
686d550d CL |
385 | dump_stack(); |
386 | } | |
686d550d CL |
387 | return NULL; |
388 | } | |
039363f3 CL |
389 | return s; |
390 | } | |
8eb8284b DW |
391 | EXPORT_SYMBOL(kmem_cache_create_usercopy); |
392 | ||
f496990f MR |
393 | /** |
394 | * kmem_cache_create - Create a cache. | |
395 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
396 | * @size: The size of objects to be created in this cache. | |
397 | * @align: The required alignment for the objects. | |
398 | * @flags: SLAB flags | |
399 | * @ctor: A constructor for the objects. | |
400 | * | |
401 | * Cannot be called within a interrupt, but can be interrupted. | |
402 | * The @ctor is run when new pages are allocated by the cache. | |
403 | * | |
404 | * The flags are | |
405 | * | |
406 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
407 | * to catch references to uninitialised memory. | |
408 | * | |
409 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check | |
410 | * for buffer overruns. | |
411 | * | |
412 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
413 | * cacheline. This can be beneficial if you're counting cycles as closely | |
414 | * as davem. | |
415 | * | |
416 | * Return: a pointer to the cache on success, NULL on failure. | |
417 | */ | |
8eb8284b | 418 | struct kmem_cache * |
f4957d5b | 419 | kmem_cache_create(const char *name, unsigned int size, unsigned int align, |
8eb8284b DW |
420 | slab_flags_t flags, void (*ctor)(void *)) |
421 | { | |
6d07d1cd | 422 | return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, |
8eb8284b DW |
423 | ctor); |
424 | } | |
794b1248 | 425 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 426 | |
657dc2f9 | 427 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 428 | { |
657dc2f9 TH |
429 | LIST_HEAD(to_destroy); |
430 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 431 | |
657dc2f9 | 432 | /* |
5f0d5a3a | 433 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
657dc2f9 | 434 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
081a06fa | 435 | * through RCU and the associated kmem_cache are dereferenced |
657dc2f9 TH |
436 | * while freeing the pages, so the kmem_caches should be freed only |
437 | * after the pending RCU operations are finished. As rcu_barrier() | |
438 | * is a pretty slow operation, we batch all pending destructions | |
439 | * asynchronously. | |
440 | */ | |
441 | mutex_lock(&slab_mutex); | |
442 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
443 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 444 | |
657dc2f9 TH |
445 | if (list_empty(&to_destroy)) |
446 | return; | |
447 | ||
448 | rcu_barrier(); | |
449 | ||
450 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
64dd6849 | 451 | debugfs_slab_release(s); |
d3fb45f3 | 452 | kfence_shutdown_cache(s); |
657dc2f9 TH |
453 | #ifdef SLAB_SUPPORTS_SYSFS |
454 | sysfs_slab_release(s); | |
455 | #else | |
456 | slab_kmem_cache_release(s); | |
457 | #endif | |
458 | } | |
d5b3cf71 VD |
459 | } |
460 | ||
657dc2f9 | 461 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 462 | { |
f9fa1d91 GT |
463 | /* free asan quarantined objects */ |
464 | kasan_cache_shutdown(s); | |
465 | ||
657dc2f9 TH |
466 | if (__kmem_cache_shutdown(s) != 0) |
467 | return -EBUSY; | |
d5b3cf71 | 468 | |
657dc2f9 | 469 | list_del(&s->list); |
d5b3cf71 | 470 | |
5f0d5a3a | 471 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
d50d82fa MP |
472 | #ifdef SLAB_SUPPORTS_SYSFS |
473 | sysfs_slab_unlink(s); | |
474 | #endif | |
657dc2f9 TH |
475 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
476 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
477 | } else { | |
d3fb45f3 | 478 | kfence_shutdown_cache(s); |
64dd6849 | 479 | debugfs_slab_release(s); |
d5b3cf71 | 480 | #ifdef SLAB_SUPPORTS_SYSFS |
d50d82fa | 481 | sysfs_slab_unlink(s); |
bf5eb3de | 482 | sysfs_slab_release(s); |
d5b3cf71 VD |
483 | #else |
484 | slab_kmem_cache_release(s); | |
485 | #endif | |
486 | } | |
657dc2f9 TH |
487 | |
488 | return 0; | |
d5b3cf71 VD |
489 | } |
490 | ||
41a21285 CL |
491 | void slab_kmem_cache_release(struct kmem_cache *s) |
492 | { | |
52b4b950 | 493 | __kmem_cache_release(s); |
3dec16ea | 494 | kfree_const(s->name); |
41a21285 CL |
495 | kmem_cache_free(kmem_cache, s); |
496 | } | |
497 | ||
945cf2b6 CL |
498 | void kmem_cache_destroy(struct kmem_cache *s) |
499 | { | |
d60fdcc9 | 500 | int err; |
d5b3cf71 | 501 | |
3942d299 SS |
502 | if (unlikely(!s)) |
503 | return; | |
504 | ||
5a836bf6 | 505 | cpus_read_lock(); |
945cf2b6 | 506 | mutex_lock(&slab_mutex); |
b8529907 | 507 | |
945cf2b6 | 508 | s->refcount--; |
b8529907 VD |
509 | if (s->refcount) |
510 | goto out_unlock; | |
511 | ||
10befea9 | 512 | err = shutdown_cache(s); |
cd918c55 | 513 | if (err) { |
4acaa7d5 | 514 | pr_err("%s %s: Slab cache still has objects\n", |
515 | __func__, s->name); | |
cd918c55 VD |
516 | dump_stack(); |
517 | } | |
b8529907 VD |
518 | out_unlock: |
519 | mutex_unlock(&slab_mutex); | |
5a836bf6 | 520 | cpus_read_unlock(); |
945cf2b6 CL |
521 | } |
522 | EXPORT_SYMBOL(kmem_cache_destroy); | |
523 | ||
03afc0e2 VD |
524 | /** |
525 | * kmem_cache_shrink - Shrink a cache. | |
526 | * @cachep: The cache to shrink. | |
527 | * | |
528 | * Releases as many slabs as possible for a cache. | |
529 | * To help debugging, a zero exit status indicates all slabs were released. | |
a862f68a MR |
530 | * |
531 | * Return: %0 if all slabs were released, non-zero otherwise | |
03afc0e2 VD |
532 | */ |
533 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
534 | { | |
535 | int ret; | |
536 | ||
7e1fa93d | 537 | |
55834c59 | 538 | kasan_cache_shrink(cachep); |
c9fc5864 | 539 | ret = __kmem_cache_shrink(cachep); |
7e1fa93d | 540 | |
03afc0e2 VD |
541 | return ret; |
542 | } | |
543 | EXPORT_SYMBOL(kmem_cache_shrink); | |
544 | ||
fda90124 | 545 | bool slab_is_available(void) |
97d06609 CL |
546 | { |
547 | return slab_state >= UP; | |
548 | } | |
b7454ad3 | 549 | |
5bb1bb35 | 550 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
551 | /** |
552 | * kmem_valid_obj - does the pointer reference a valid slab object? | |
553 | * @object: pointer to query. | |
554 | * | |
555 | * Return: %true if the pointer is to a not-yet-freed object from | |
556 | * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer | |
557 | * is to an already-freed object, and %false otherwise. | |
558 | */ | |
559 | bool kmem_valid_obj(void *object) | |
560 | { | |
561 | struct page *page; | |
562 | ||
563 | /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ | |
564 | if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) | |
565 | return false; | |
566 | page = virt_to_head_page(object); | |
567 | return PageSlab(page); | |
568 | } | |
0d3dd2c8 | 569 | EXPORT_SYMBOL_GPL(kmem_valid_obj); |
8e7f37f2 | 570 | |
a9c7bedb ME |
571 | static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page) |
572 | { | |
573 | if (__kfence_obj_info(kpp, object, page)) | |
574 | return; | |
575 | __kmem_obj_info(kpp, object, page); | |
576 | } | |
577 | ||
8e7f37f2 PM |
578 | /** |
579 | * kmem_dump_obj - Print available slab provenance information | |
580 | * @object: slab object for which to find provenance information. | |
581 | * | |
582 | * This function uses pr_cont(), so that the caller is expected to have | |
583 | * printed out whatever preamble is appropriate. The provenance information | |
584 | * depends on the type of object and on how much debugging is enabled. | |
585 | * For a slab-cache object, the fact that it is a slab object is printed, | |
586 | * and, if available, the slab name, return address, and stack trace from | |
e548eaa1 | 587 | * the allocation and last free path of that object. |
8e7f37f2 PM |
588 | * |
589 | * This function will splat if passed a pointer to a non-slab object. | |
590 | * If you are not sure what type of object you have, you should instead | |
591 | * use mem_dump_obj(). | |
592 | */ | |
593 | void kmem_dump_obj(void *object) | |
594 | { | |
595 | char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; | |
596 | int i; | |
597 | struct page *page; | |
598 | unsigned long ptroffset; | |
599 | struct kmem_obj_info kp = { }; | |
600 | ||
601 | if (WARN_ON_ONCE(!virt_addr_valid(object))) | |
602 | return; | |
603 | page = virt_to_head_page(object); | |
604 | if (WARN_ON_ONCE(!PageSlab(page))) { | |
605 | pr_cont(" non-slab memory.\n"); | |
606 | return; | |
607 | } | |
608 | kmem_obj_info(&kp, object, page); | |
609 | if (kp.kp_slab_cache) | |
610 | pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); | |
611 | else | |
612 | pr_cont(" slab%s", cp); | |
a9c7bedb ME |
613 | if (is_kfence_address(object)) |
614 | pr_cont(" (kfence)"); | |
8e7f37f2 PM |
615 | if (kp.kp_objp) |
616 | pr_cont(" start %px", kp.kp_objp); | |
617 | if (kp.kp_data_offset) | |
618 | pr_cont(" data offset %lu", kp.kp_data_offset); | |
619 | if (kp.kp_objp) { | |
620 | ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; | |
621 | pr_cont(" pointer offset %lu", ptroffset); | |
622 | } | |
623 | if (kp.kp_slab_cache && kp.kp_slab_cache->usersize) | |
624 | pr_cont(" size %u", kp.kp_slab_cache->usersize); | |
625 | if (kp.kp_ret) | |
626 | pr_cont(" allocated at %pS\n", kp.kp_ret); | |
627 | else | |
628 | pr_cont("\n"); | |
629 | for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { | |
630 | if (!kp.kp_stack[i]) | |
631 | break; | |
632 | pr_info(" %pS\n", kp.kp_stack[i]); | |
633 | } | |
e548eaa1 MS |
634 | |
635 | if (kp.kp_free_stack[0]) | |
636 | pr_cont(" Free path:\n"); | |
637 | ||
638 | for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) { | |
639 | if (!kp.kp_free_stack[i]) | |
640 | break; | |
641 | pr_info(" %pS\n", kp.kp_free_stack[i]); | |
642 | } | |
643 | ||
8e7f37f2 | 644 | } |
0d3dd2c8 | 645 | EXPORT_SYMBOL_GPL(kmem_dump_obj); |
5bb1bb35 | 646 | #endif |
8e7f37f2 | 647 | |
45530c44 CL |
648 | #ifndef CONFIG_SLOB |
649 | /* Create a cache during boot when no slab services are available yet */ | |
361d575e AD |
650 | void __init create_boot_cache(struct kmem_cache *s, const char *name, |
651 | unsigned int size, slab_flags_t flags, | |
652 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
653 | { |
654 | int err; | |
59bb4798 | 655 | unsigned int align = ARCH_KMALLOC_MINALIGN; |
45530c44 CL |
656 | |
657 | s->name = name; | |
658 | s->size = s->object_size = size; | |
59bb4798 VB |
659 | |
660 | /* | |
661 | * For power of two sizes, guarantee natural alignment for kmalloc | |
662 | * caches, regardless of SL*B debugging options. | |
663 | */ | |
664 | if (is_power_of_2(size)) | |
665 | align = max(align, size); | |
666 | s->align = calculate_alignment(flags, align, size); | |
667 | ||
8eb8284b DW |
668 | s->useroffset = useroffset; |
669 | s->usersize = usersize; | |
f7ce3190 | 670 | |
45530c44 CL |
671 | err = __kmem_cache_create(s, flags); |
672 | ||
673 | if (err) | |
361d575e | 674 | panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", |
45530c44 CL |
675 | name, size, err); |
676 | ||
677 | s->refcount = -1; /* Exempt from merging for now */ | |
678 | } | |
679 | ||
55de8b9c AD |
680 | struct kmem_cache *__init create_kmalloc_cache(const char *name, |
681 | unsigned int size, slab_flags_t flags, | |
682 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
683 | { |
684 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
685 | ||
686 | if (!s) | |
687 | panic("Out of memory when creating slab %s\n", name); | |
688 | ||
6c0c21ad | 689 | create_boot_cache(s, name, size, flags, useroffset, usersize); |
92850134 | 690 | kasan_cache_create_kmalloc(s); |
45530c44 CL |
691 | list_add(&s->list, &slab_caches); |
692 | s->refcount = 1; | |
693 | return s; | |
694 | } | |
695 | ||
cc252eae | 696 | struct kmem_cache * |
a07057dc AB |
697 | kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = |
698 | { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; | |
9425c58e CL |
699 | EXPORT_SYMBOL(kmalloc_caches); |
700 | ||
2c59dd65 CL |
701 | /* |
702 | * Conversion table for small slabs sizes / 8 to the index in the | |
703 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
704 | * of two cache sizes there. The size of larger slabs can be determined using | |
705 | * fls. | |
706 | */ | |
d5f86655 | 707 | static u8 size_index[24] __ro_after_init = { |
2c59dd65 CL |
708 | 3, /* 8 */ |
709 | 4, /* 16 */ | |
710 | 5, /* 24 */ | |
711 | 5, /* 32 */ | |
712 | 6, /* 40 */ | |
713 | 6, /* 48 */ | |
714 | 6, /* 56 */ | |
715 | 6, /* 64 */ | |
716 | 1, /* 72 */ | |
717 | 1, /* 80 */ | |
718 | 1, /* 88 */ | |
719 | 1, /* 96 */ | |
720 | 7, /* 104 */ | |
721 | 7, /* 112 */ | |
722 | 7, /* 120 */ | |
723 | 7, /* 128 */ | |
724 | 2, /* 136 */ | |
725 | 2, /* 144 */ | |
726 | 2, /* 152 */ | |
727 | 2, /* 160 */ | |
728 | 2, /* 168 */ | |
729 | 2, /* 176 */ | |
730 | 2, /* 184 */ | |
731 | 2 /* 192 */ | |
732 | }; | |
733 | ||
ac914d08 | 734 | static inline unsigned int size_index_elem(unsigned int bytes) |
2c59dd65 CL |
735 | { |
736 | return (bytes - 1) / 8; | |
737 | } | |
738 | ||
739 | /* | |
740 | * Find the kmem_cache structure that serves a given size of | |
741 | * allocation | |
742 | */ | |
743 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
744 | { | |
d5f86655 | 745 | unsigned int index; |
2c59dd65 CL |
746 | |
747 | if (size <= 192) { | |
748 | if (!size) | |
749 | return ZERO_SIZE_PTR; | |
750 | ||
751 | index = size_index[size_index_elem(size)]; | |
61448479 | 752 | } else { |
221d7da6 | 753 | if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) |
61448479 | 754 | return NULL; |
2c59dd65 | 755 | index = fls(size - 1); |
61448479 | 756 | } |
2c59dd65 | 757 | |
cc252eae | 758 | return kmalloc_caches[kmalloc_type(flags)][index]; |
2c59dd65 CL |
759 | } |
760 | ||
cb5d9fb3 | 761 | #ifdef CONFIG_ZONE_DMA |
494c1dfe WL |
762 | #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz, |
763 | #else | |
764 | #define KMALLOC_DMA_NAME(sz) | |
765 | #endif | |
766 | ||
767 | #ifdef CONFIG_MEMCG_KMEM | |
768 | #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz, | |
cb5d9fb3 | 769 | #else |
494c1dfe WL |
770 | #define KMALLOC_CGROUP_NAME(sz) |
771 | #endif | |
772 | ||
cb5d9fb3 PL |
773 | #define INIT_KMALLOC_INFO(__size, __short_size) \ |
774 | { \ | |
775 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ | |
776 | .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ | |
494c1dfe WL |
777 | KMALLOC_CGROUP_NAME(__short_size) \ |
778 | KMALLOC_DMA_NAME(__short_size) \ | |
cb5d9fb3 PL |
779 | .size = __size, \ |
780 | } | |
cb5d9fb3 | 781 | |
4066c33d GG |
782 | /* |
783 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
588c7fa0 HY |
784 | * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is |
785 | * kmalloc-32M. | |
4066c33d | 786 | */ |
af3b5f87 | 787 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
cb5d9fb3 PL |
788 | INIT_KMALLOC_INFO(0, 0), |
789 | INIT_KMALLOC_INFO(96, 96), | |
790 | INIT_KMALLOC_INFO(192, 192), | |
791 | INIT_KMALLOC_INFO(8, 8), | |
792 | INIT_KMALLOC_INFO(16, 16), | |
793 | INIT_KMALLOC_INFO(32, 32), | |
794 | INIT_KMALLOC_INFO(64, 64), | |
795 | INIT_KMALLOC_INFO(128, 128), | |
796 | INIT_KMALLOC_INFO(256, 256), | |
797 | INIT_KMALLOC_INFO(512, 512), | |
798 | INIT_KMALLOC_INFO(1024, 1k), | |
799 | INIT_KMALLOC_INFO(2048, 2k), | |
800 | INIT_KMALLOC_INFO(4096, 4k), | |
801 | INIT_KMALLOC_INFO(8192, 8k), | |
802 | INIT_KMALLOC_INFO(16384, 16k), | |
803 | INIT_KMALLOC_INFO(32768, 32k), | |
804 | INIT_KMALLOC_INFO(65536, 64k), | |
805 | INIT_KMALLOC_INFO(131072, 128k), | |
806 | INIT_KMALLOC_INFO(262144, 256k), | |
807 | INIT_KMALLOC_INFO(524288, 512k), | |
808 | INIT_KMALLOC_INFO(1048576, 1M), | |
809 | INIT_KMALLOC_INFO(2097152, 2M), | |
810 | INIT_KMALLOC_INFO(4194304, 4M), | |
811 | INIT_KMALLOC_INFO(8388608, 8M), | |
812 | INIT_KMALLOC_INFO(16777216, 16M), | |
588c7fa0 | 813 | INIT_KMALLOC_INFO(33554432, 32M) |
4066c33d GG |
814 | }; |
815 | ||
f97d5f63 | 816 | /* |
34cc6990 DS |
817 | * Patch up the size_index table if we have strange large alignment |
818 | * requirements for the kmalloc array. This is only the case for | |
819 | * MIPS it seems. The standard arches will not generate any code here. | |
820 | * | |
821 | * Largest permitted alignment is 256 bytes due to the way we | |
822 | * handle the index determination for the smaller caches. | |
823 | * | |
824 | * Make sure that nothing crazy happens if someone starts tinkering | |
825 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 826 | */ |
34cc6990 | 827 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 | 828 | { |
ac914d08 | 829 | unsigned int i; |
f97d5f63 | 830 | |
2c59dd65 CL |
831 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
832 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
833 | ||
834 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
ac914d08 | 835 | unsigned int elem = size_index_elem(i); |
2c59dd65 CL |
836 | |
837 | if (elem >= ARRAY_SIZE(size_index)) | |
838 | break; | |
839 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
840 | } | |
841 | ||
842 | if (KMALLOC_MIN_SIZE >= 64) { | |
843 | /* | |
844 | * The 96 byte size cache is not used if the alignment | |
845 | * is 64 byte. | |
846 | */ | |
847 | for (i = 64 + 8; i <= 96; i += 8) | |
848 | size_index[size_index_elem(i)] = 7; | |
849 | ||
850 | } | |
851 | ||
852 | if (KMALLOC_MIN_SIZE >= 128) { | |
853 | /* | |
854 | * The 192 byte sized cache is not used if the alignment | |
855 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
856 | * instead. | |
857 | */ | |
858 | for (i = 128 + 8; i <= 192; i += 8) | |
859 | size_index[size_index_elem(i)] = 8; | |
860 | } | |
34cc6990 DS |
861 | } |
862 | ||
1291523f | 863 | static void __init |
13657d0a | 864 | new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) |
a9730fca | 865 | { |
494c1dfe | 866 | if (type == KMALLOC_RECLAIM) { |
1291523f | 867 | flags |= SLAB_RECLAIM_ACCOUNT; |
494c1dfe WL |
868 | } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) { |
869 | if (cgroup_memory_nokmem) { | |
870 | kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx]; | |
871 | return; | |
872 | } | |
873 | flags |= SLAB_ACCOUNT; | |
874 | } | |
1291523f | 875 | |
cb5d9fb3 PL |
876 | kmalloc_caches[type][idx] = create_kmalloc_cache( |
877 | kmalloc_info[idx].name[type], | |
6c0c21ad DW |
878 | kmalloc_info[idx].size, flags, 0, |
879 | kmalloc_info[idx].size); | |
13e680fb WL |
880 | |
881 | /* | |
882 | * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for | |
883 | * KMALLOC_NORMAL caches. | |
884 | */ | |
885 | if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL)) | |
886 | kmalloc_caches[type][idx]->refcount = -1; | |
a9730fca CL |
887 | } |
888 | ||
34cc6990 DS |
889 | /* |
890 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
891 | * may already have been created because they were needed to | |
892 | * enable allocations for slab creation. | |
893 | */ | |
d50112ed | 894 | void __init create_kmalloc_caches(slab_flags_t flags) |
34cc6990 | 895 | { |
13657d0a PL |
896 | int i; |
897 | enum kmalloc_cache_type type; | |
34cc6990 | 898 | |
494c1dfe WL |
899 | /* |
900 | * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined | |
901 | */ | |
1291523f VB |
902 | for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { |
903 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { | |
904 | if (!kmalloc_caches[type][i]) | |
905 | new_kmalloc_cache(i, type, flags); | |
f97d5f63 | 906 | |
1291523f VB |
907 | /* |
908 | * Caches that are not of the two-to-the-power-of size. | |
909 | * These have to be created immediately after the | |
910 | * earlier power of two caches | |
911 | */ | |
912 | if (KMALLOC_MIN_SIZE <= 32 && i == 6 && | |
913 | !kmalloc_caches[type][1]) | |
914 | new_kmalloc_cache(1, type, flags); | |
915 | if (KMALLOC_MIN_SIZE <= 64 && i == 7 && | |
916 | !kmalloc_caches[type][2]) | |
917 | new_kmalloc_cache(2, type, flags); | |
918 | } | |
8a965b3b CL |
919 | } |
920 | ||
f97d5f63 CL |
921 | /* Kmalloc array is now usable */ |
922 | slab_state = UP; | |
923 | ||
f97d5f63 CL |
924 | #ifdef CONFIG_ZONE_DMA |
925 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
cc252eae | 926 | struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; |
f97d5f63 CL |
927 | |
928 | if (s) { | |
cc252eae | 929 | kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( |
cb5d9fb3 | 930 | kmalloc_info[i].name[KMALLOC_DMA], |
dc0a7f75 | 931 | kmalloc_info[i].size, |
49f2d241 VB |
932 | SLAB_CACHE_DMA | flags, 0, |
933 | kmalloc_info[i].size); | |
f97d5f63 CL |
934 | } |
935 | } | |
936 | #endif | |
937 | } | |
45530c44 CL |
938 | #endif /* !CONFIG_SLOB */ |
939 | ||
44405099 LL |
940 | gfp_t kmalloc_fix_flags(gfp_t flags) |
941 | { | |
942 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; | |
943 | ||
944 | flags &= ~GFP_SLAB_BUG_MASK; | |
945 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
946 | invalid_mask, &invalid_mask, flags, &flags); | |
947 | dump_stack(); | |
948 | ||
949 | return flags; | |
950 | } | |
951 | ||
cea371f4 VD |
952 | /* |
953 | * To avoid unnecessary overhead, we pass through large allocation requests | |
954 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
955 | * know the allocation order to free the pages properly in kfree. | |
956 | */ | |
52383431 VD |
957 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
958 | { | |
6a486c0a | 959 | void *ret = NULL; |
52383431 VD |
960 | struct page *page; |
961 | ||
44405099 LL |
962 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
963 | flags = kmalloc_fix_flags(flags); | |
964 | ||
52383431 | 965 | flags |= __GFP_COMP; |
4949148a | 966 | page = alloc_pages(flags, order); |
6a486c0a VB |
967 | if (likely(page)) { |
968 | ret = page_address(page); | |
96403bfe MS |
969 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
970 | PAGE_SIZE << order); | |
6a486c0a | 971 | } |
0116523c | 972 | ret = kasan_kmalloc_large(ret, size, flags); |
a2f77575 | 973 | /* As ret might get tagged, call kmemleak hook after KASAN. */ |
53128245 | 974 | kmemleak_alloc(ret, size, 1, flags); |
52383431 VD |
975 | return ret; |
976 | } | |
977 | EXPORT_SYMBOL(kmalloc_order); | |
978 | ||
f1b6eb6e CL |
979 | #ifdef CONFIG_TRACING |
980 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
981 | { | |
982 | void *ret = kmalloc_order(size, flags, order); | |
983 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
984 | return ret; | |
985 | } | |
986 | EXPORT_SYMBOL(kmalloc_order_trace); | |
987 | #endif | |
45530c44 | 988 | |
7c00fce9 TG |
989 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
990 | /* Randomize a generic freelist */ | |
991 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
302d55d5 | 992 | unsigned int count) |
7c00fce9 | 993 | { |
7c00fce9 | 994 | unsigned int rand; |
302d55d5 | 995 | unsigned int i; |
7c00fce9 TG |
996 | |
997 | for (i = 0; i < count; i++) | |
998 | list[i] = i; | |
999 | ||
1000 | /* Fisher-Yates shuffle */ | |
1001 | for (i = count - 1; i > 0; i--) { | |
1002 | rand = prandom_u32_state(state); | |
1003 | rand %= (i + 1); | |
1004 | swap(list[i], list[rand]); | |
1005 | } | |
1006 | } | |
1007 | ||
1008 | /* Create a random sequence per cache */ | |
1009 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
1010 | gfp_t gfp) | |
1011 | { | |
1012 | struct rnd_state state; | |
1013 | ||
1014 | if (count < 2 || cachep->random_seq) | |
1015 | return 0; | |
1016 | ||
1017 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
1018 | if (!cachep->random_seq) | |
1019 | return -ENOMEM; | |
1020 | ||
1021 | /* Get best entropy at this stage of boot */ | |
1022 | prandom_seed_state(&state, get_random_long()); | |
1023 | ||
1024 | freelist_randomize(&state, cachep->random_seq, count); | |
1025 | return 0; | |
1026 | } | |
1027 | ||
1028 | /* Destroy the per-cache random freelist sequence */ | |
1029 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
1030 | { | |
1031 | kfree(cachep->random_seq); | |
1032 | cachep->random_seq = NULL; | |
1033 | } | |
1034 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1035 | ||
5b365771 | 1036 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
e9b4db2b | 1037 | #ifdef CONFIG_SLAB |
0825a6f9 | 1038 | #define SLABINFO_RIGHTS (0600) |
e9b4db2b | 1039 | #else |
0825a6f9 | 1040 | #define SLABINFO_RIGHTS (0400) |
e9b4db2b WL |
1041 | #endif |
1042 | ||
b047501c | 1043 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
1044 | { |
1045 | /* | |
1046 | * Output format version, so at least we can change it | |
1047 | * without _too_ many complaints. | |
1048 | */ | |
1049 | #ifdef CONFIG_DEBUG_SLAB | |
1050 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
1051 | #else | |
1052 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
1053 | #endif | |
756a025f | 1054 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
1055 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
1056 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
1057 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 1058 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
1059 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
1060 | #endif | |
1061 | seq_putc(m, '\n'); | |
1062 | } | |
1063 | ||
1df3b26f | 1064 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1065 | { |
b7454ad3 | 1066 | mutex_lock(&slab_mutex); |
c7094406 | 1067 | return seq_list_start(&slab_caches, *pos); |
b7454ad3 GC |
1068 | } |
1069 | ||
276a2439 | 1070 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 | 1071 | { |
c7094406 | 1072 | return seq_list_next(p, &slab_caches, pos); |
b7454ad3 GC |
1073 | } |
1074 | ||
276a2439 | 1075 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1076 | { |
1077 | mutex_unlock(&slab_mutex); | |
1078 | } | |
1079 | ||
b047501c | 1080 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1081 | { |
0d7561c6 GC |
1082 | struct slabinfo sinfo; |
1083 | ||
1084 | memset(&sinfo, 0, sizeof(sinfo)); | |
1085 | get_slabinfo(s, &sinfo); | |
1086 | ||
1087 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", | |
10befea9 | 1088 | s->name, sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1089 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1090 | ||
1091 | seq_printf(m, " : tunables %4u %4u %4u", | |
1092 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1093 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1094 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1095 | slabinfo_show_stats(m, s); | |
1096 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1097 | } |
1098 | ||
1df3b26f | 1099 | static int slab_show(struct seq_file *m, void *p) |
749c5415 | 1100 | { |
c7094406 | 1101 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
749c5415 | 1102 | |
c7094406 | 1103 | if (p == slab_caches.next) |
1df3b26f | 1104 | print_slabinfo_header(m); |
10befea9 | 1105 | cache_show(s, m); |
b047501c VD |
1106 | return 0; |
1107 | } | |
1108 | ||
852d8be0 YS |
1109 | void dump_unreclaimable_slab(void) |
1110 | { | |
7714304f | 1111 | struct kmem_cache *s; |
852d8be0 YS |
1112 | struct slabinfo sinfo; |
1113 | ||
1114 | /* | |
1115 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1116 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1117 | * risk of crash. | |
1118 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1119 | * without acquiring the mutex. | |
1120 | */ | |
1121 | if (!mutex_trylock(&slab_mutex)) { | |
1122 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1123 | return; | |
1124 | } | |
1125 | ||
1126 | pr_info("Unreclaimable slab info:\n"); | |
1127 | pr_info("Name Used Total\n"); | |
1128 | ||
7714304f | 1129 | list_for_each_entry(s, &slab_caches, list) { |
10befea9 | 1130 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
852d8be0 YS |
1131 | continue; |
1132 | ||
1133 | get_slabinfo(s, &sinfo); | |
1134 | ||
1135 | if (sinfo.num_objs > 0) | |
10befea9 | 1136 | pr_info("%-17s %10luKB %10luKB\n", s->name, |
852d8be0 YS |
1137 | (sinfo.active_objs * s->size) / 1024, |
1138 | (sinfo.num_objs * s->size) / 1024); | |
1139 | } | |
1140 | mutex_unlock(&slab_mutex); | |
1141 | } | |
1142 | ||
a87425a3 | 1143 | #if defined(CONFIG_MEMCG_KMEM) |
b047501c VD |
1144 | int memcg_slab_show(struct seq_file *m, void *p) |
1145 | { | |
4330a26b RG |
1146 | /* |
1147 | * Deprecated. | |
1148 | * Please, take a look at tools/cgroup/slabinfo.py . | |
1149 | */ | |
b047501c | 1150 | return 0; |
749c5415 | 1151 | } |
b047501c | 1152 | #endif |
749c5415 | 1153 | |
b7454ad3 GC |
1154 | /* |
1155 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1156 | * | |
1157 | * Output layout: | |
1158 | * cache-name | |
1159 | * num-active-objs | |
1160 | * total-objs | |
1161 | * object size | |
1162 | * num-active-slabs | |
1163 | * total-slabs | |
1164 | * num-pages-per-slab | |
1165 | * + further values on SMP and with statistics enabled | |
1166 | */ | |
1167 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1168 | .start = slab_start, |
276a2439 WL |
1169 | .next = slab_next, |
1170 | .stop = slab_stop, | |
1df3b26f | 1171 | .show = slab_show, |
b7454ad3 GC |
1172 | }; |
1173 | ||
1174 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1175 | { | |
1176 | return seq_open(file, &slabinfo_op); | |
1177 | } | |
1178 | ||
97a32539 | 1179 | static const struct proc_ops slabinfo_proc_ops = { |
d919b33d | 1180 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
1181 | .proc_open = slabinfo_open, |
1182 | .proc_read = seq_read, | |
1183 | .proc_write = slabinfo_write, | |
1184 | .proc_lseek = seq_lseek, | |
1185 | .proc_release = seq_release, | |
b7454ad3 GC |
1186 | }; |
1187 | ||
1188 | static int __init slab_proc_init(void) | |
1189 | { | |
97a32539 | 1190 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); |
b7454ad3 GC |
1191 | return 0; |
1192 | } | |
1193 | module_init(slab_proc_init); | |
fcf8a1e4 | 1194 | |
5b365771 | 1195 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ |
928cec9c AR |
1196 | |
1197 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1198 | gfp_t flags) | |
1199 | { | |
1200 | void *ret; | |
fa9ba3aa | 1201 | size_t ks; |
928cec9c | 1202 | |
d12d9ad8 AK |
1203 | /* Don't use instrumented ksize to allow precise KASAN poisoning. */ |
1204 | if (likely(!ZERO_OR_NULL_PTR(p))) { | |
1205 | if (!kasan_check_byte(p)) | |
1206 | return NULL; | |
1207 | ks = kfence_ksize(p) ?: __ksize(p); | |
1208 | } else | |
1209 | ks = 0; | |
928cec9c | 1210 | |
d12d9ad8 | 1211 | /* If the object still fits, repoison it precisely. */ |
0316bec2 | 1212 | if (ks >= new_size) { |
0116523c | 1213 | p = kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1214 | return (void *)p; |
0316bec2 | 1215 | } |
928cec9c AR |
1216 | |
1217 | ret = kmalloc_track_caller(new_size, flags); | |
d12d9ad8 AK |
1218 | if (ret && p) { |
1219 | /* Disable KASAN checks as the object's redzone is accessed. */ | |
1220 | kasan_disable_current(); | |
1221 | memcpy(ret, kasan_reset_tag(p), ks); | |
1222 | kasan_enable_current(); | |
1223 | } | |
928cec9c AR |
1224 | |
1225 | return ret; | |
1226 | } | |
1227 | ||
928cec9c AR |
1228 | /** |
1229 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1230 | * @p: object to reallocate memory for. | |
1231 | * @new_size: how many bytes of memory are required. | |
1232 | * @flags: the type of memory to allocate. | |
1233 | * | |
1234 | * The contents of the object pointed to are preserved up to the | |
15d5de49 BG |
1235 | * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). |
1236 | * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size | |
1237 | * is 0 and @p is not a %NULL pointer, the object pointed to is freed. | |
a862f68a MR |
1238 | * |
1239 | * Return: pointer to the allocated memory or %NULL in case of error | |
928cec9c AR |
1240 | */ |
1241 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1242 | { | |
1243 | void *ret; | |
1244 | ||
1245 | if (unlikely(!new_size)) { | |
1246 | kfree(p); | |
1247 | return ZERO_SIZE_PTR; | |
1248 | } | |
1249 | ||
1250 | ret = __do_krealloc(p, new_size, flags); | |
772a2fa5 | 1251 | if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) |
928cec9c AR |
1252 | kfree(p); |
1253 | ||
1254 | return ret; | |
1255 | } | |
1256 | EXPORT_SYMBOL(krealloc); | |
1257 | ||
1258 | /** | |
453431a5 | 1259 | * kfree_sensitive - Clear sensitive information in memory before freeing |
928cec9c AR |
1260 | * @p: object to free memory of |
1261 | * | |
1262 | * The memory of the object @p points to is zeroed before freed. | |
453431a5 | 1263 | * If @p is %NULL, kfree_sensitive() does nothing. |
928cec9c AR |
1264 | * |
1265 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1266 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1267 | * careful when using this function in performance sensitive code. | |
1268 | */ | |
453431a5 | 1269 | void kfree_sensitive(const void *p) |
928cec9c AR |
1270 | { |
1271 | size_t ks; | |
1272 | void *mem = (void *)p; | |
1273 | ||
928cec9c | 1274 | ks = ksize(mem); |
fa9ba3aa WK |
1275 | if (ks) |
1276 | memzero_explicit(mem, ks); | |
928cec9c AR |
1277 | kfree(mem); |
1278 | } | |
453431a5 | 1279 | EXPORT_SYMBOL(kfree_sensitive); |
928cec9c | 1280 | |
10d1f8cb ME |
1281 | /** |
1282 | * ksize - get the actual amount of memory allocated for a given object | |
1283 | * @objp: Pointer to the object | |
1284 | * | |
1285 | * kmalloc may internally round up allocations and return more memory | |
1286 | * than requested. ksize() can be used to determine the actual amount of | |
1287 | * memory allocated. The caller may use this additional memory, even though | |
1288 | * a smaller amount of memory was initially specified with the kmalloc call. | |
1289 | * The caller must guarantee that objp points to a valid object previously | |
1290 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
1291 | * must not be freed during the duration of the call. | |
1292 | * | |
1293 | * Return: size of the actual memory used by @objp in bytes | |
1294 | */ | |
1295 | size_t ksize(const void *objp) | |
1296 | { | |
0d4ca4c9 ME |
1297 | size_t size; |
1298 | ||
0d4ca4c9 | 1299 | /* |
611806b4 AK |
1300 | * We need to first check that the pointer to the object is valid, and |
1301 | * only then unpoison the memory. The report printed from ksize() is | |
1302 | * more useful, then when it's printed later when the behaviour could | |
1303 | * be undefined due to a potential use-after-free or double-free. | |
0d4ca4c9 | 1304 | * |
611806b4 AK |
1305 | * We use kasan_check_byte(), which is supported for the hardware |
1306 | * tag-based KASAN mode, unlike kasan_check_read/write(). | |
1307 | * | |
1308 | * If the pointed to memory is invalid, we return 0 to avoid users of | |
0d4ca4c9 ME |
1309 | * ksize() writing to and potentially corrupting the memory region. |
1310 | * | |
1311 | * We want to perform the check before __ksize(), to avoid potentially | |
1312 | * crashing in __ksize() due to accessing invalid metadata. | |
1313 | */ | |
611806b4 | 1314 | if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) |
0d4ca4c9 ME |
1315 | return 0; |
1316 | ||
d3fb45f3 | 1317 | size = kfence_ksize(objp) ?: __ksize(objp); |
10d1f8cb ME |
1318 | /* |
1319 | * We assume that ksize callers could use whole allocated area, | |
1320 | * so we need to unpoison this area. | |
1321 | */ | |
cebd0eb2 | 1322 | kasan_unpoison_range(objp, size); |
10d1f8cb ME |
1323 | return size; |
1324 | } | |
1325 | EXPORT_SYMBOL(ksize); | |
1326 | ||
928cec9c AR |
1327 | /* Tracepoints definitions. */ |
1328 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1329 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1330 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1331 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1332 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1333 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); | |
4f6923fb HM |
1334 | |
1335 | int should_failslab(struct kmem_cache *s, gfp_t gfpflags) | |
1336 | { | |
1337 | if (__should_failslab(s, gfpflags)) | |
1338 | return -ENOMEM; | |
1339 | return 0; | |
1340 | } | |
1341 | ALLOW_ERROR_INJECTION(should_failslab, ERRNO); |