]>
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 | ||
74 | #ifdef CONFIG_SLUB | |
75 | __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0); | |
76 | #endif | |
77 | ||
78 | __setup("slab_nomerge", setup_slab_nomerge); | |
79 | ||
07f361b2 JK |
80 | /* |
81 | * Determine the size of a slab object | |
82 | */ | |
83 | unsigned int kmem_cache_size(struct kmem_cache *s) | |
84 | { | |
85 | return s->object_size; | |
86 | } | |
87 | EXPORT_SYMBOL(kmem_cache_size); | |
88 | ||
77be4b13 | 89 | #ifdef CONFIG_DEBUG_VM |
f4957d5b | 90 | static int kmem_cache_sanity_check(const char *name, unsigned int size) |
039363f3 | 91 | { |
039363f3 CL |
92 | if (!name || in_interrupt() || size < sizeof(void *) || |
93 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
94 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
95 | return -EINVAL; | |
039363f3 | 96 | } |
b920536a | 97 | |
20cea968 | 98 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ |
77be4b13 SK |
99 | return 0; |
100 | } | |
101 | #else | |
f4957d5b | 102 | static inline int kmem_cache_sanity_check(const char *name, unsigned int size) |
77be4b13 SK |
103 | { |
104 | return 0; | |
105 | } | |
20cea968 CL |
106 | #endif |
107 | ||
484748f0 CL |
108 | void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p) |
109 | { | |
110 | size_t i; | |
111 | ||
ca257195 JDB |
112 | for (i = 0; i < nr; i++) { |
113 | if (s) | |
114 | kmem_cache_free(s, p[i]); | |
115 | else | |
116 | kfree(p[i]); | |
117 | } | |
484748f0 CL |
118 | } |
119 | ||
865762a8 | 120 | int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr, |
484748f0 CL |
121 | void **p) |
122 | { | |
123 | size_t i; | |
124 | ||
125 | for (i = 0; i < nr; i++) { | |
126 | void *x = p[i] = kmem_cache_alloc(s, flags); | |
127 | if (!x) { | |
128 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 129 | return 0; |
484748f0 CL |
130 | } |
131 | } | |
865762a8 | 132 | return i; |
484748f0 CL |
133 | } |
134 | ||
692ae74a BL |
135 | /* |
136 | * Figure out what the alignment of the objects will be given a set of | |
137 | * flags, a user specified alignment and the size of the objects. | |
138 | */ | |
f4957d5b AD |
139 | static unsigned int calculate_alignment(slab_flags_t flags, |
140 | unsigned int align, unsigned int size) | |
692ae74a BL |
141 | { |
142 | /* | |
143 | * If the user wants hardware cache aligned objects then follow that | |
144 | * suggestion if the object is sufficiently large. | |
145 | * | |
146 | * The hardware cache alignment cannot override the specified | |
147 | * alignment though. If that is greater then use it. | |
148 | */ | |
149 | if (flags & SLAB_HWCACHE_ALIGN) { | |
f4957d5b | 150 | unsigned int ralign; |
692ae74a BL |
151 | |
152 | ralign = cache_line_size(); | |
153 | while (size <= ralign / 2) | |
154 | ralign /= 2; | |
155 | align = max(align, ralign); | |
156 | } | |
157 | ||
158 | if (align < ARCH_SLAB_MINALIGN) | |
159 | align = ARCH_SLAB_MINALIGN; | |
160 | ||
161 | return ALIGN(align, sizeof(void *)); | |
162 | } | |
163 | ||
423c929c JK |
164 | /* |
165 | * Find a mergeable slab cache | |
166 | */ | |
167 | int slab_unmergeable(struct kmem_cache *s) | |
168 | { | |
169 | if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE)) | |
170 | return 1; | |
171 | ||
423c929c JK |
172 | if (s->ctor) |
173 | return 1; | |
174 | ||
8eb8284b DW |
175 | if (s->usersize) |
176 | return 1; | |
177 | ||
423c929c JK |
178 | /* |
179 | * We may have set a slab to be unmergeable during bootstrap. | |
180 | */ | |
181 | if (s->refcount < 0) | |
182 | return 1; | |
183 | ||
184 | return 0; | |
185 | } | |
186 | ||
f4957d5b | 187 | struct kmem_cache *find_mergeable(unsigned int size, unsigned int align, |
d50112ed | 188 | slab_flags_t flags, const char *name, void (*ctor)(void *)) |
423c929c JK |
189 | { |
190 | struct kmem_cache *s; | |
191 | ||
c6e28895 | 192 | if (slab_nomerge) |
423c929c JK |
193 | return NULL; |
194 | ||
195 | if (ctor) | |
196 | return NULL; | |
197 | ||
198 | size = ALIGN(size, sizeof(void *)); | |
199 | align = calculate_alignment(flags, align, size); | |
200 | size = ALIGN(size, align); | |
37540008 | 201 | flags = kmem_cache_flags(size, flags, name); |
423c929c | 202 | |
c6e28895 GM |
203 | if (flags & SLAB_NEVER_MERGE) |
204 | return NULL; | |
205 | ||
c7094406 | 206 | list_for_each_entry_reverse(s, &slab_caches, list) { |
423c929c JK |
207 | if (slab_unmergeable(s)) |
208 | continue; | |
209 | ||
210 | if (size > s->size) | |
211 | continue; | |
212 | ||
213 | if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME)) | |
214 | continue; | |
215 | /* | |
216 | * Check if alignment is compatible. | |
217 | * Courtesy of Adrian Drzewiecki | |
218 | */ | |
219 | if ((s->size & ~(align - 1)) != s->size) | |
220 | continue; | |
221 | ||
222 | if (s->size - size >= sizeof(void *)) | |
223 | continue; | |
224 | ||
95069ac8 JK |
225 | if (IS_ENABLED(CONFIG_SLAB) && align && |
226 | (align > s->align || s->align % align)) | |
227 | continue; | |
228 | ||
423c929c JK |
229 | return s; |
230 | } | |
231 | return NULL; | |
232 | } | |
233 | ||
c9a77a79 | 234 | static struct kmem_cache *create_cache(const char *name, |
613a5eb5 | 235 | unsigned int object_size, unsigned int align, |
7bbdb81e AD |
236 | slab_flags_t flags, unsigned int useroffset, |
237 | unsigned int usersize, void (*ctor)(void *), | |
9855609b | 238 | struct kmem_cache *root_cache) |
794b1248 VD |
239 | { |
240 | struct kmem_cache *s; | |
241 | int err; | |
242 | ||
8eb8284b DW |
243 | if (WARN_ON(useroffset + usersize > object_size)) |
244 | useroffset = usersize = 0; | |
245 | ||
794b1248 VD |
246 | err = -ENOMEM; |
247 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); | |
248 | if (!s) | |
249 | goto out; | |
250 | ||
251 | s->name = name; | |
613a5eb5 | 252 | s->size = s->object_size = object_size; |
794b1248 VD |
253 | s->align = align; |
254 | s->ctor = ctor; | |
8eb8284b DW |
255 | s->useroffset = useroffset; |
256 | s->usersize = usersize; | |
794b1248 | 257 | |
794b1248 VD |
258 | err = __kmem_cache_create(s, flags); |
259 | if (err) | |
260 | goto out_free_cache; | |
261 | ||
262 | s->refcount = 1; | |
263 | list_add(&s->list, &slab_caches); | |
794b1248 VD |
264 | out: |
265 | if (err) | |
266 | return ERR_PTR(err); | |
267 | return s; | |
268 | ||
269 | out_free_cache: | |
7c4da061 | 270 | kmem_cache_free(kmem_cache, s); |
794b1248 VD |
271 | goto out; |
272 | } | |
45906855 | 273 | |
f496990f MR |
274 | /** |
275 | * kmem_cache_create_usercopy - Create a cache with a region suitable | |
276 | * for copying to userspace | |
77be4b13 SK |
277 | * @name: A string which is used in /proc/slabinfo to identify this cache. |
278 | * @size: The size of objects to be created in this cache. | |
279 | * @align: The required alignment for the objects. | |
280 | * @flags: SLAB flags | |
8eb8284b DW |
281 | * @useroffset: Usercopy region offset |
282 | * @usersize: Usercopy region size | |
77be4b13 SK |
283 | * @ctor: A constructor for the objects. |
284 | * | |
77be4b13 SK |
285 | * Cannot be called within a interrupt, but can be interrupted. |
286 | * The @ctor is run when new pages are allocated by the cache. | |
287 | * | |
288 | * The flags are | |
289 | * | |
290 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
291 | * to catch references to uninitialised memory. | |
292 | * | |
f496990f | 293 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check |
77be4b13 SK |
294 | * for buffer overruns. |
295 | * | |
296 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
297 | * cacheline. This can be beneficial if you're counting cycles as closely | |
298 | * as davem. | |
f496990f MR |
299 | * |
300 | * Return: a pointer to the cache on success, NULL on failure. | |
77be4b13 | 301 | */ |
2633d7a0 | 302 | struct kmem_cache * |
f4957d5b AD |
303 | kmem_cache_create_usercopy(const char *name, |
304 | unsigned int size, unsigned int align, | |
7bbdb81e AD |
305 | slab_flags_t flags, |
306 | unsigned int useroffset, unsigned int usersize, | |
8eb8284b | 307 | void (*ctor)(void *)) |
77be4b13 | 308 | { |
40911a79 | 309 | struct kmem_cache *s = NULL; |
3dec16ea | 310 | const char *cache_name; |
3965fc36 | 311 | int err; |
039363f3 | 312 | |
77be4b13 | 313 | mutex_lock(&slab_mutex); |
686d550d | 314 | |
794b1248 | 315 | err = kmem_cache_sanity_check(name, size); |
3aa24f51 | 316 | if (err) { |
3965fc36 | 317 | goto out_unlock; |
3aa24f51 | 318 | } |
686d550d | 319 | |
e70954fd TG |
320 | /* Refuse requests with allocator specific flags */ |
321 | if (flags & ~SLAB_FLAGS_PERMITTED) { | |
322 | err = -EINVAL; | |
323 | goto out_unlock; | |
324 | } | |
325 | ||
d8843922 GC |
326 | /* |
327 | * Some allocators will constraint the set of valid flags to a subset | |
328 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
329 | * case, and we'll just provide them with a sanitized version of the | |
330 | * passed flags. | |
331 | */ | |
332 | flags &= CACHE_CREATE_MASK; | |
686d550d | 333 | |
8eb8284b DW |
334 | /* Fail closed on bad usersize of useroffset values. */ |
335 | if (WARN_ON(!usersize && useroffset) || | |
336 | WARN_ON(size < usersize || size - usersize < useroffset)) | |
337 | usersize = useroffset = 0; | |
338 | ||
339 | if (!usersize) | |
340 | s = __kmem_cache_alias(name, size, align, flags, ctor); | |
794b1248 | 341 | if (s) |
3965fc36 | 342 | goto out_unlock; |
2633d7a0 | 343 | |
3dec16ea | 344 | cache_name = kstrdup_const(name, GFP_KERNEL); |
794b1248 VD |
345 | if (!cache_name) { |
346 | err = -ENOMEM; | |
347 | goto out_unlock; | |
348 | } | |
7c9adf5a | 349 | |
613a5eb5 | 350 | s = create_cache(cache_name, size, |
c9a77a79 | 351 | calculate_alignment(flags, align, size), |
9855609b | 352 | flags, useroffset, usersize, ctor, NULL); |
794b1248 VD |
353 | if (IS_ERR(s)) { |
354 | err = PTR_ERR(s); | |
3dec16ea | 355 | kfree_const(cache_name); |
794b1248 | 356 | } |
3965fc36 VD |
357 | |
358 | out_unlock: | |
20cea968 | 359 | mutex_unlock(&slab_mutex); |
03afc0e2 | 360 | |
ba3253c7 | 361 | if (err) { |
686d550d CL |
362 | if (flags & SLAB_PANIC) |
363 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
364 | name, err); | |
365 | else { | |
1170532b | 366 | pr_warn("kmem_cache_create(%s) failed with error %d\n", |
686d550d CL |
367 | name, err); |
368 | dump_stack(); | |
369 | } | |
686d550d CL |
370 | return NULL; |
371 | } | |
039363f3 CL |
372 | return s; |
373 | } | |
8eb8284b DW |
374 | EXPORT_SYMBOL(kmem_cache_create_usercopy); |
375 | ||
f496990f MR |
376 | /** |
377 | * kmem_cache_create - Create a cache. | |
378 | * @name: A string which is used in /proc/slabinfo to identify this cache. | |
379 | * @size: The size of objects to be created in this cache. | |
380 | * @align: The required alignment for the objects. | |
381 | * @flags: SLAB flags | |
382 | * @ctor: A constructor for the objects. | |
383 | * | |
384 | * Cannot be called within a interrupt, but can be interrupted. | |
385 | * The @ctor is run when new pages are allocated by the cache. | |
386 | * | |
387 | * The flags are | |
388 | * | |
389 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
390 | * to catch references to uninitialised memory. | |
391 | * | |
392 | * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check | |
393 | * for buffer overruns. | |
394 | * | |
395 | * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware | |
396 | * cacheline. This can be beneficial if you're counting cycles as closely | |
397 | * as davem. | |
398 | * | |
399 | * Return: a pointer to the cache on success, NULL on failure. | |
400 | */ | |
8eb8284b | 401 | struct kmem_cache * |
f4957d5b | 402 | kmem_cache_create(const char *name, unsigned int size, unsigned int align, |
8eb8284b DW |
403 | slab_flags_t flags, void (*ctor)(void *)) |
404 | { | |
6d07d1cd | 405 | return kmem_cache_create_usercopy(name, size, align, flags, 0, 0, |
8eb8284b DW |
406 | ctor); |
407 | } | |
794b1248 | 408 | EXPORT_SYMBOL(kmem_cache_create); |
2633d7a0 | 409 | |
657dc2f9 | 410 | static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work) |
d5b3cf71 | 411 | { |
657dc2f9 TH |
412 | LIST_HEAD(to_destroy); |
413 | struct kmem_cache *s, *s2; | |
d5b3cf71 | 414 | |
657dc2f9 | 415 | /* |
5f0d5a3a | 416 | * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the |
657dc2f9 | 417 | * @slab_caches_to_rcu_destroy list. The slab pages are freed |
081a06fa | 418 | * through RCU and the associated kmem_cache are dereferenced |
657dc2f9 TH |
419 | * while freeing the pages, so the kmem_caches should be freed only |
420 | * after the pending RCU operations are finished. As rcu_barrier() | |
421 | * is a pretty slow operation, we batch all pending destructions | |
422 | * asynchronously. | |
423 | */ | |
424 | mutex_lock(&slab_mutex); | |
425 | list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy); | |
426 | mutex_unlock(&slab_mutex); | |
d5b3cf71 | 427 | |
657dc2f9 TH |
428 | if (list_empty(&to_destroy)) |
429 | return; | |
430 | ||
431 | rcu_barrier(); | |
432 | ||
433 | list_for_each_entry_safe(s, s2, &to_destroy, list) { | |
d3fb45f3 | 434 | kfence_shutdown_cache(s); |
657dc2f9 TH |
435 | #ifdef SLAB_SUPPORTS_SYSFS |
436 | sysfs_slab_release(s); | |
437 | #else | |
438 | slab_kmem_cache_release(s); | |
439 | #endif | |
440 | } | |
d5b3cf71 VD |
441 | } |
442 | ||
657dc2f9 | 443 | static int shutdown_cache(struct kmem_cache *s) |
d5b3cf71 | 444 | { |
f9fa1d91 GT |
445 | /* free asan quarantined objects */ |
446 | kasan_cache_shutdown(s); | |
447 | ||
657dc2f9 TH |
448 | if (__kmem_cache_shutdown(s) != 0) |
449 | return -EBUSY; | |
d5b3cf71 | 450 | |
657dc2f9 | 451 | list_del(&s->list); |
d5b3cf71 | 452 | |
5f0d5a3a | 453 | if (s->flags & SLAB_TYPESAFE_BY_RCU) { |
d50d82fa MP |
454 | #ifdef SLAB_SUPPORTS_SYSFS |
455 | sysfs_slab_unlink(s); | |
456 | #endif | |
657dc2f9 TH |
457 | list_add_tail(&s->list, &slab_caches_to_rcu_destroy); |
458 | schedule_work(&slab_caches_to_rcu_destroy_work); | |
459 | } else { | |
d3fb45f3 | 460 | kfence_shutdown_cache(s); |
d5b3cf71 | 461 | #ifdef SLAB_SUPPORTS_SYSFS |
d50d82fa | 462 | sysfs_slab_unlink(s); |
bf5eb3de | 463 | sysfs_slab_release(s); |
d5b3cf71 VD |
464 | #else |
465 | slab_kmem_cache_release(s); | |
466 | #endif | |
467 | } | |
657dc2f9 TH |
468 | |
469 | return 0; | |
d5b3cf71 VD |
470 | } |
471 | ||
41a21285 CL |
472 | void slab_kmem_cache_release(struct kmem_cache *s) |
473 | { | |
52b4b950 | 474 | __kmem_cache_release(s); |
3dec16ea | 475 | kfree_const(s->name); |
41a21285 CL |
476 | kmem_cache_free(kmem_cache, s); |
477 | } | |
478 | ||
945cf2b6 CL |
479 | void kmem_cache_destroy(struct kmem_cache *s) |
480 | { | |
d60fdcc9 | 481 | int err; |
d5b3cf71 | 482 | |
3942d299 SS |
483 | if (unlikely(!s)) |
484 | return; | |
485 | ||
945cf2b6 | 486 | mutex_lock(&slab_mutex); |
b8529907 | 487 | |
945cf2b6 | 488 | s->refcount--; |
b8529907 VD |
489 | if (s->refcount) |
490 | goto out_unlock; | |
491 | ||
10befea9 | 492 | err = shutdown_cache(s); |
cd918c55 | 493 | if (err) { |
756a025f JP |
494 | pr_err("kmem_cache_destroy %s: Slab cache still has objects\n", |
495 | s->name); | |
cd918c55 VD |
496 | dump_stack(); |
497 | } | |
b8529907 VD |
498 | out_unlock: |
499 | mutex_unlock(&slab_mutex); | |
945cf2b6 CL |
500 | } |
501 | EXPORT_SYMBOL(kmem_cache_destroy); | |
502 | ||
03afc0e2 VD |
503 | /** |
504 | * kmem_cache_shrink - Shrink a cache. | |
505 | * @cachep: The cache to shrink. | |
506 | * | |
507 | * Releases as many slabs as possible for a cache. | |
508 | * To help debugging, a zero exit status indicates all slabs were released. | |
a862f68a MR |
509 | * |
510 | * Return: %0 if all slabs were released, non-zero otherwise | |
03afc0e2 VD |
511 | */ |
512 | int kmem_cache_shrink(struct kmem_cache *cachep) | |
513 | { | |
514 | int ret; | |
515 | ||
7e1fa93d | 516 | |
55834c59 | 517 | kasan_cache_shrink(cachep); |
c9fc5864 | 518 | ret = __kmem_cache_shrink(cachep); |
7e1fa93d | 519 | |
03afc0e2 VD |
520 | return ret; |
521 | } | |
522 | EXPORT_SYMBOL(kmem_cache_shrink); | |
523 | ||
fda90124 | 524 | bool slab_is_available(void) |
97d06609 CL |
525 | { |
526 | return slab_state >= UP; | |
527 | } | |
b7454ad3 | 528 | |
8e7f37f2 PM |
529 | /** |
530 | * kmem_valid_obj - does the pointer reference a valid slab object? | |
531 | * @object: pointer to query. | |
532 | * | |
533 | * Return: %true if the pointer is to a not-yet-freed object from | |
534 | * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer | |
535 | * is to an already-freed object, and %false otherwise. | |
536 | */ | |
537 | bool kmem_valid_obj(void *object) | |
538 | { | |
539 | struct page *page; | |
540 | ||
541 | /* Some arches consider ZERO_SIZE_PTR to be a valid address. */ | |
542 | if (object < (void *)PAGE_SIZE || !virt_addr_valid(object)) | |
543 | return false; | |
544 | page = virt_to_head_page(object); | |
545 | return PageSlab(page); | |
546 | } | |
547 | ||
548 | /** | |
549 | * kmem_dump_obj - Print available slab provenance information | |
550 | * @object: slab object for which to find provenance information. | |
551 | * | |
552 | * This function uses pr_cont(), so that the caller is expected to have | |
553 | * printed out whatever preamble is appropriate. The provenance information | |
554 | * depends on the type of object and on how much debugging is enabled. | |
555 | * For a slab-cache object, the fact that it is a slab object is printed, | |
556 | * and, if available, the slab name, return address, and stack trace from | |
557 | * the allocation of that object. | |
558 | * | |
559 | * This function will splat if passed a pointer to a non-slab object. | |
560 | * If you are not sure what type of object you have, you should instead | |
561 | * use mem_dump_obj(). | |
562 | */ | |
563 | void kmem_dump_obj(void *object) | |
564 | { | |
565 | char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc"; | |
566 | int i; | |
567 | struct page *page; | |
568 | unsigned long ptroffset; | |
569 | struct kmem_obj_info kp = { }; | |
570 | ||
571 | if (WARN_ON_ONCE(!virt_addr_valid(object))) | |
572 | return; | |
573 | page = virt_to_head_page(object); | |
574 | if (WARN_ON_ONCE(!PageSlab(page))) { | |
575 | pr_cont(" non-slab memory.\n"); | |
576 | return; | |
577 | } | |
578 | kmem_obj_info(&kp, object, page); | |
579 | if (kp.kp_slab_cache) | |
580 | pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name); | |
581 | else | |
582 | pr_cont(" slab%s", cp); | |
583 | if (kp.kp_objp) | |
584 | pr_cont(" start %px", kp.kp_objp); | |
585 | if (kp.kp_data_offset) | |
586 | pr_cont(" data offset %lu", kp.kp_data_offset); | |
587 | if (kp.kp_objp) { | |
588 | ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset; | |
589 | pr_cont(" pointer offset %lu", ptroffset); | |
590 | } | |
591 | if (kp.kp_slab_cache && kp.kp_slab_cache->usersize) | |
592 | pr_cont(" size %u", kp.kp_slab_cache->usersize); | |
593 | if (kp.kp_ret) | |
594 | pr_cont(" allocated at %pS\n", kp.kp_ret); | |
595 | else | |
596 | pr_cont("\n"); | |
597 | for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) { | |
598 | if (!kp.kp_stack[i]) | |
599 | break; | |
600 | pr_info(" %pS\n", kp.kp_stack[i]); | |
601 | } | |
602 | } | |
603 | ||
45530c44 CL |
604 | #ifndef CONFIG_SLOB |
605 | /* Create a cache during boot when no slab services are available yet */ | |
361d575e AD |
606 | void __init create_boot_cache(struct kmem_cache *s, const char *name, |
607 | unsigned int size, slab_flags_t flags, | |
608 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
609 | { |
610 | int err; | |
59bb4798 | 611 | unsigned int align = ARCH_KMALLOC_MINALIGN; |
45530c44 CL |
612 | |
613 | s->name = name; | |
614 | s->size = s->object_size = size; | |
59bb4798 VB |
615 | |
616 | /* | |
617 | * For power of two sizes, guarantee natural alignment for kmalloc | |
618 | * caches, regardless of SL*B debugging options. | |
619 | */ | |
620 | if (is_power_of_2(size)) | |
621 | align = max(align, size); | |
622 | s->align = calculate_alignment(flags, align, size); | |
623 | ||
8eb8284b DW |
624 | s->useroffset = useroffset; |
625 | s->usersize = usersize; | |
f7ce3190 | 626 | |
45530c44 CL |
627 | err = __kmem_cache_create(s, flags); |
628 | ||
629 | if (err) | |
361d575e | 630 | panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n", |
45530c44 CL |
631 | name, size, err); |
632 | ||
633 | s->refcount = -1; /* Exempt from merging for now */ | |
634 | } | |
635 | ||
55de8b9c AD |
636 | struct kmem_cache *__init create_kmalloc_cache(const char *name, |
637 | unsigned int size, slab_flags_t flags, | |
638 | unsigned int useroffset, unsigned int usersize) | |
45530c44 CL |
639 | { |
640 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
641 | ||
642 | if (!s) | |
643 | panic("Out of memory when creating slab %s\n", name); | |
644 | ||
6c0c21ad | 645 | create_boot_cache(s, name, size, flags, useroffset, usersize); |
92850134 | 646 | kasan_cache_create_kmalloc(s); |
45530c44 CL |
647 | list_add(&s->list, &slab_caches); |
648 | s->refcount = 1; | |
649 | return s; | |
650 | } | |
651 | ||
cc252eae | 652 | struct kmem_cache * |
a07057dc AB |
653 | kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init = |
654 | { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ }; | |
9425c58e CL |
655 | EXPORT_SYMBOL(kmalloc_caches); |
656 | ||
2c59dd65 CL |
657 | /* |
658 | * Conversion table for small slabs sizes / 8 to the index in the | |
659 | * kmalloc array. This is necessary for slabs < 192 since we have non power | |
660 | * of two cache sizes there. The size of larger slabs can be determined using | |
661 | * fls. | |
662 | */ | |
d5f86655 | 663 | static u8 size_index[24] __ro_after_init = { |
2c59dd65 CL |
664 | 3, /* 8 */ |
665 | 4, /* 16 */ | |
666 | 5, /* 24 */ | |
667 | 5, /* 32 */ | |
668 | 6, /* 40 */ | |
669 | 6, /* 48 */ | |
670 | 6, /* 56 */ | |
671 | 6, /* 64 */ | |
672 | 1, /* 72 */ | |
673 | 1, /* 80 */ | |
674 | 1, /* 88 */ | |
675 | 1, /* 96 */ | |
676 | 7, /* 104 */ | |
677 | 7, /* 112 */ | |
678 | 7, /* 120 */ | |
679 | 7, /* 128 */ | |
680 | 2, /* 136 */ | |
681 | 2, /* 144 */ | |
682 | 2, /* 152 */ | |
683 | 2, /* 160 */ | |
684 | 2, /* 168 */ | |
685 | 2, /* 176 */ | |
686 | 2, /* 184 */ | |
687 | 2 /* 192 */ | |
688 | }; | |
689 | ||
ac914d08 | 690 | static inline unsigned int size_index_elem(unsigned int bytes) |
2c59dd65 CL |
691 | { |
692 | return (bytes - 1) / 8; | |
693 | } | |
694 | ||
695 | /* | |
696 | * Find the kmem_cache structure that serves a given size of | |
697 | * allocation | |
698 | */ | |
699 | struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags) | |
700 | { | |
d5f86655 | 701 | unsigned int index; |
2c59dd65 CL |
702 | |
703 | if (size <= 192) { | |
704 | if (!size) | |
705 | return ZERO_SIZE_PTR; | |
706 | ||
707 | index = size_index[size_index_elem(size)]; | |
61448479 | 708 | } else { |
221d7da6 | 709 | if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE)) |
61448479 | 710 | return NULL; |
2c59dd65 | 711 | index = fls(size - 1); |
61448479 | 712 | } |
2c59dd65 | 713 | |
cc252eae | 714 | return kmalloc_caches[kmalloc_type(flags)][index]; |
2c59dd65 CL |
715 | } |
716 | ||
cb5d9fb3 PL |
717 | #ifdef CONFIG_ZONE_DMA |
718 | #define INIT_KMALLOC_INFO(__size, __short_size) \ | |
719 | { \ | |
720 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ | |
721 | .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ | |
722 | .name[KMALLOC_DMA] = "dma-kmalloc-" #__short_size, \ | |
723 | .size = __size, \ | |
724 | } | |
725 | #else | |
726 | #define INIT_KMALLOC_INFO(__size, __short_size) \ | |
727 | { \ | |
728 | .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \ | |
729 | .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \ | |
730 | .size = __size, \ | |
731 | } | |
732 | #endif | |
733 | ||
4066c33d GG |
734 | /* |
735 | * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time. | |
736 | * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is | |
737 | * kmalloc-67108864. | |
738 | */ | |
af3b5f87 | 739 | const struct kmalloc_info_struct kmalloc_info[] __initconst = { |
cb5d9fb3 PL |
740 | INIT_KMALLOC_INFO(0, 0), |
741 | INIT_KMALLOC_INFO(96, 96), | |
742 | INIT_KMALLOC_INFO(192, 192), | |
743 | INIT_KMALLOC_INFO(8, 8), | |
744 | INIT_KMALLOC_INFO(16, 16), | |
745 | INIT_KMALLOC_INFO(32, 32), | |
746 | INIT_KMALLOC_INFO(64, 64), | |
747 | INIT_KMALLOC_INFO(128, 128), | |
748 | INIT_KMALLOC_INFO(256, 256), | |
749 | INIT_KMALLOC_INFO(512, 512), | |
750 | INIT_KMALLOC_INFO(1024, 1k), | |
751 | INIT_KMALLOC_INFO(2048, 2k), | |
752 | INIT_KMALLOC_INFO(4096, 4k), | |
753 | INIT_KMALLOC_INFO(8192, 8k), | |
754 | INIT_KMALLOC_INFO(16384, 16k), | |
755 | INIT_KMALLOC_INFO(32768, 32k), | |
756 | INIT_KMALLOC_INFO(65536, 64k), | |
757 | INIT_KMALLOC_INFO(131072, 128k), | |
758 | INIT_KMALLOC_INFO(262144, 256k), | |
759 | INIT_KMALLOC_INFO(524288, 512k), | |
760 | INIT_KMALLOC_INFO(1048576, 1M), | |
761 | INIT_KMALLOC_INFO(2097152, 2M), | |
762 | INIT_KMALLOC_INFO(4194304, 4M), | |
763 | INIT_KMALLOC_INFO(8388608, 8M), | |
764 | INIT_KMALLOC_INFO(16777216, 16M), | |
765 | INIT_KMALLOC_INFO(33554432, 32M), | |
766 | INIT_KMALLOC_INFO(67108864, 64M) | |
4066c33d GG |
767 | }; |
768 | ||
f97d5f63 | 769 | /* |
34cc6990 DS |
770 | * Patch up the size_index table if we have strange large alignment |
771 | * requirements for the kmalloc array. This is only the case for | |
772 | * MIPS it seems. The standard arches will not generate any code here. | |
773 | * | |
774 | * Largest permitted alignment is 256 bytes due to the way we | |
775 | * handle the index determination for the smaller caches. | |
776 | * | |
777 | * Make sure that nothing crazy happens if someone starts tinkering | |
778 | * around with ARCH_KMALLOC_MINALIGN | |
f97d5f63 | 779 | */ |
34cc6990 | 780 | void __init setup_kmalloc_cache_index_table(void) |
f97d5f63 | 781 | { |
ac914d08 | 782 | unsigned int i; |
f97d5f63 | 783 | |
2c59dd65 CL |
784 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 || |
785 | (KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1))); | |
786 | ||
787 | for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) { | |
ac914d08 | 788 | unsigned int elem = size_index_elem(i); |
2c59dd65 CL |
789 | |
790 | if (elem >= ARRAY_SIZE(size_index)) | |
791 | break; | |
792 | size_index[elem] = KMALLOC_SHIFT_LOW; | |
793 | } | |
794 | ||
795 | if (KMALLOC_MIN_SIZE >= 64) { | |
796 | /* | |
797 | * The 96 byte size cache is not used if the alignment | |
798 | * is 64 byte. | |
799 | */ | |
800 | for (i = 64 + 8; i <= 96; i += 8) | |
801 | size_index[size_index_elem(i)] = 7; | |
802 | ||
803 | } | |
804 | ||
805 | if (KMALLOC_MIN_SIZE >= 128) { | |
806 | /* | |
807 | * The 192 byte sized cache is not used if the alignment | |
808 | * is 128 byte. Redirect kmalloc to use the 256 byte cache | |
809 | * instead. | |
810 | */ | |
811 | for (i = 128 + 8; i <= 192; i += 8) | |
812 | size_index[size_index_elem(i)] = 8; | |
813 | } | |
34cc6990 DS |
814 | } |
815 | ||
1291523f | 816 | static void __init |
13657d0a | 817 | new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags) |
a9730fca | 818 | { |
cb5d9fb3 | 819 | if (type == KMALLOC_RECLAIM) |
1291523f | 820 | flags |= SLAB_RECLAIM_ACCOUNT; |
1291523f | 821 | |
cb5d9fb3 PL |
822 | kmalloc_caches[type][idx] = create_kmalloc_cache( |
823 | kmalloc_info[idx].name[type], | |
6c0c21ad DW |
824 | kmalloc_info[idx].size, flags, 0, |
825 | kmalloc_info[idx].size); | |
a9730fca CL |
826 | } |
827 | ||
34cc6990 DS |
828 | /* |
829 | * Create the kmalloc array. Some of the regular kmalloc arrays | |
830 | * may already have been created because they were needed to | |
831 | * enable allocations for slab creation. | |
832 | */ | |
d50112ed | 833 | void __init create_kmalloc_caches(slab_flags_t flags) |
34cc6990 | 834 | { |
13657d0a PL |
835 | int i; |
836 | enum kmalloc_cache_type type; | |
34cc6990 | 837 | |
1291523f VB |
838 | for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) { |
839 | for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) { | |
840 | if (!kmalloc_caches[type][i]) | |
841 | new_kmalloc_cache(i, type, flags); | |
f97d5f63 | 842 | |
1291523f VB |
843 | /* |
844 | * Caches that are not of the two-to-the-power-of size. | |
845 | * These have to be created immediately after the | |
846 | * earlier power of two caches | |
847 | */ | |
848 | if (KMALLOC_MIN_SIZE <= 32 && i == 6 && | |
849 | !kmalloc_caches[type][1]) | |
850 | new_kmalloc_cache(1, type, flags); | |
851 | if (KMALLOC_MIN_SIZE <= 64 && i == 7 && | |
852 | !kmalloc_caches[type][2]) | |
853 | new_kmalloc_cache(2, type, flags); | |
854 | } | |
8a965b3b CL |
855 | } |
856 | ||
f97d5f63 CL |
857 | /* Kmalloc array is now usable */ |
858 | slab_state = UP; | |
859 | ||
f97d5f63 CL |
860 | #ifdef CONFIG_ZONE_DMA |
861 | for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) { | |
cc252eae | 862 | struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i]; |
f97d5f63 CL |
863 | |
864 | if (s) { | |
cc252eae | 865 | kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache( |
cb5d9fb3 | 866 | kmalloc_info[i].name[KMALLOC_DMA], |
dc0a7f75 | 867 | kmalloc_info[i].size, |
49f2d241 VB |
868 | SLAB_CACHE_DMA | flags, 0, |
869 | kmalloc_info[i].size); | |
f97d5f63 CL |
870 | } |
871 | } | |
872 | #endif | |
873 | } | |
45530c44 CL |
874 | #endif /* !CONFIG_SLOB */ |
875 | ||
44405099 LL |
876 | gfp_t kmalloc_fix_flags(gfp_t flags) |
877 | { | |
878 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; | |
879 | ||
880 | flags &= ~GFP_SLAB_BUG_MASK; | |
881 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
882 | invalid_mask, &invalid_mask, flags, &flags); | |
883 | dump_stack(); | |
884 | ||
885 | return flags; | |
886 | } | |
887 | ||
cea371f4 VD |
888 | /* |
889 | * To avoid unnecessary overhead, we pass through large allocation requests | |
890 | * directly to the page allocator. We use __GFP_COMP, because we will need to | |
891 | * know the allocation order to free the pages properly in kfree. | |
892 | */ | |
52383431 VD |
893 | void *kmalloc_order(size_t size, gfp_t flags, unsigned int order) |
894 | { | |
6a486c0a | 895 | void *ret = NULL; |
52383431 VD |
896 | struct page *page; |
897 | ||
44405099 LL |
898 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
899 | flags = kmalloc_fix_flags(flags); | |
900 | ||
52383431 | 901 | flags |= __GFP_COMP; |
4949148a | 902 | page = alloc_pages(flags, order); |
6a486c0a VB |
903 | if (likely(page)) { |
904 | ret = page_address(page); | |
96403bfe MS |
905 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
906 | PAGE_SIZE << order); | |
6a486c0a | 907 | } |
0116523c | 908 | ret = kasan_kmalloc_large(ret, size, flags); |
a2f77575 | 909 | /* As ret might get tagged, call kmemleak hook after KASAN. */ |
53128245 | 910 | kmemleak_alloc(ret, size, 1, flags); |
52383431 VD |
911 | return ret; |
912 | } | |
913 | EXPORT_SYMBOL(kmalloc_order); | |
914 | ||
f1b6eb6e CL |
915 | #ifdef CONFIG_TRACING |
916 | void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order) | |
917 | { | |
918 | void *ret = kmalloc_order(size, flags, order); | |
919 | trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags); | |
920 | return ret; | |
921 | } | |
922 | EXPORT_SYMBOL(kmalloc_order_trace); | |
923 | #endif | |
45530c44 | 924 | |
7c00fce9 TG |
925 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
926 | /* Randomize a generic freelist */ | |
927 | static void freelist_randomize(struct rnd_state *state, unsigned int *list, | |
302d55d5 | 928 | unsigned int count) |
7c00fce9 | 929 | { |
7c00fce9 | 930 | unsigned int rand; |
302d55d5 | 931 | unsigned int i; |
7c00fce9 TG |
932 | |
933 | for (i = 0; i < count; i++) | |
934 | list[i] = i; | |
935 | ||
936 | /* Fisher-Yates shuffle */ | |
937 | for (i = count - 1; i > 0; i--) { | |
938 | rand = prandom_u32_state(state); | |
939 | rand %= (i + 1); | |
940 | swap(list[i], list[rand]); | |
941 | } | |
942 | } | |
943 | ||
944 | /* Create a random sequence per cache */ | |
945 | int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count, | |
946 | gfp_t gfp) | |
947 | { | |
948 | struct rnd_state state; | |
949 | ||
950 | if (count < 2 || cachep->random_seq) | |
951 | return 0; | |
952 | ||
953 | cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp); | |
954 | if (!cachep->random_seq) | |
955 | return -ENOMEM; | |
956 | ||
957 | /* Get best entropy at this stage of boot */ | |
958 | prandom_seed_state(&state, get_random_long()); | |
959 | ||
960 | freelist_randomize(&state, cachep->random_seq, count); | |
961 | return 0; | |
962 | } | |
963 | ||
964 | /* Destroy the per-cache random freelist sequence */ | |
965 | void cache_random_seq_destroy(struct kmem_cache *cachep) | |
966 | { | |
967 | kfree(cachep->random_seq); | |
968 | cachep->random_seq = NULL; | |
969 | } | |
970 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
971 | ||
5b365771 | 972 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
e9b4db2b | 973 | #ifdef CONFIG_SLAB |
0825a6f9 | 974 | #define SLABINFO_RIGHTS (0600) |
e9b4db2b | 975 | #else |
0825a6f9 | 976 | #define SLABINFO_RIGHTS (0400) |
e9b4db2b WL |
977 | #endif |
978 | ||
b047501c | 979 | static void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
980 | { |
981 | /* | |
982 | * Output format version, so at least we can change it | |
983 | * without _too_ many complaints. | |
984 | */ | |
985 | #ifdef CONFIG_DEBUG_SLAB | |
986 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
987 | #else | |
988 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
989 | #endif | |
756a025f | 990 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>"); |
bcee6e2a GC |
991 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); |
992 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
993 | #ifdef CONFIG_DEBUG_SLAB | |
756a025f | 994 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); |
bcee6e2a GC |
995 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); |
996 | #endif | |
997 | seq_putc(m, '\n'); | |
998 | } | |
999 | ||
1df3b26f | 1000 | void *slab_start(struct seq_file *m, loff_t *pos) |
b7454ad3 | 1001 | { |
b7454ad3 | 1002 | mutex_lock(&slab_mutex); |
c7094406 | 1003 | return seq_list_start(&slab_caches, *pos); |
b7454ad3 GC |
1004 | } |
1005 | ||
276a2439 | 1006 | void *slab_next(struct seq_file *m, void *p, loff_t *pos) |
b7454ad3 | 1007 | { |
c7094406 | 1008 | return seq_list_next(p, &slab_caches, pos); |
b7454ad3 GC |
1009 | } |
1010 | ||
276a2439 | 1011 | void slab_stop(struct seq_file *m, void *p) |
b7454ad3 GC |
1012 | { |
1013 | mutex_unlock(&slab_mutex); | |
1014 | } | |
1015 | ||
b047501c | 1016 | static void cache_show(struct kmem_cache *s, struct seq_file *m) |
b7454ad3 | 1017 | { |
0d7561c6 GC |
1018 | struct slabinfo sinfo; |
1019 | ||
1020 | memset(&sinfo, 0, sizeof(sinfo)); | |
1021 | get_slabinfo(s, &sinfo); | |
1022 | ||
1023 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", | |
10befea9 | 1024 | s->name, sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
1025 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
1026 | ||
1027 | seq_printf(m, " : tunables %4u %4u %4u", | |
1028 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
1029 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
1030 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
1031 | slabinfo_show_stats(m, s); | |
1032 | seq_putc(m, '\n'); | |
b7454ad3 GC |
1033 | } |
1034 | ||
1df3b26f | 1035 | static int slab_show(struct seq_file *m, void *p) |
749c5415 | 1036 | { |
c7094406 | 1037 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); |
749c5415 | 1038 | |
c7094406 | 1039 | if (p == slab_caches.next) |
1df3b26f | 1040 | print_slabinfo_header(m); |
10befea9 | 1041 | cache_show(s, m); |
b047501c VD |
1042 | return 0; |
1043 | } | |
1044 | ||
852d8be0 YS |
1045 | void dump_unreclaimable_slab(void) |
1046 | { | |
7714304f | 1047 | struct kmem_cache *s; |
852d8be0 YS |
1048 | struct slabinfo sinfo; |
1049 | ||
1050 | /* | |
1051 | * Here acquiring slab_mutex is risky since we don't prefer to get | |
1052 | * sleep in oom path. But, without mutex hold, it may introduce a | |
1053 | * risk of crash. | |
1054 | * Use mutex_trylock to protect the list traverse, dump nothing | |
1055 | * without acquiring the mutex. | |
1056 | */ | |
1057 | if (!mutex_trylock(&slab_mutex)) { | |
1058 | pr_warn("excessive unreclaimable slab but cannot dump stats\n"); | |
1059 | return; | |
1060 | } | |
1061 | ||
1062 | pr_info("Unreclaimable slab info:\n"); | |
1063 | pr_info("Name Used Total\n"); | |
1064 | ||
7714304f | 1065 | list_for_each_entry(s, &slab_caches, list) { |
10befea9 | 1066 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
852d8be0 YS |
1067 | continue; |
1068 | ||
1069 | get_slabinfo(s, &sinfo); | |
1070 | ||
1071 | if (sinfo.num_objs > 0) | |
10befea9 | 1072 | pr_info("%-17s %10luKB %10luKB\n", s->name, |
852d8be0 YS |
1073 | (sinfo.active_objs * s->size) / 1024, |
1074 | (sinfo.num_objs * s->size) / 1024); | |
1075 | } | |
1076 | mutex_unlock(&slab_mutex); | |
1077 | } | |
1078 | ||
a87425a3 | 1079 | #if defined(CONFIG_MEMCG_KMEM) |
b047501c VD |
1080 | int memcg_slab_show(struct seq_file *m, void *p) |
1081 | { | |
4330a26b RG |
1082 | /* |
1083 | * Deprecated. | |
1084 | * Please, take a look at tools/cgroup/slabinfo.py . | |
1085 | */ | |
b047501c | 1086 | return 0; |
749c5415 | 1087 | } |
b047501c | 1088 | #endif |
749c5415 | 1089 | |
b7454ad3 GC |
1090 | /* |
1091 | * slabinfo_op - iterator that generates /proc/slabinfo | |
1092 | * | |
1093 | * Output layout: | |
1094 | * cache-name | |
1095 | * num-active-objs | |
1096 | * total-objs | |
1097 | * object size | |
1098 | * num-active-slabs | |
1099 | * total-slabs | |
1100 | * num-pages-per-slab | |
1101 | * + further values on SMP and with statistics enabled | |
1102 | */ | |
1103 | static const struct seq_operations slabinfo_op = { | |
1df3b26f | 1104 | .start = slab_start, |
276a2439 WL |
1105 | .next = slab_next, |
1106 | .stop = slab_stop, | |
1df3b26f | 1107 | .show = slab_show, |
b7454ad3 GC |
1108 | }; |
1109 | ||
1110 | static int slabinfo_open(struct inode *inode, struct file *file) | |
1111 | { | |
1112 | return seq_open(file, &slabinfo_op); | |
1113 | } | |
1114 | ||
97a32539 | 1115 | static const struct proc_ops slabinfo_proc_ops = { |
d919b33d | 1116 | .proc_flags = PROC_ENTRY_PERMANENT, |
97a32539 AD |
1117 | .proc_open = slabinfo_open, |
1118 | .proc_read = seq_read, | |
1119 | .proc_write = slabinfo_write, | |
1120 | .proc_lseek = seq_lseek, | |
1121 | .proc_release = seq_release, | |
b7454ad3 GC |
1122 | }; |
1123 | ||
1124 | static int __init slab_proc_init(void) | |
1125 | { | |
97a32539 | 1126 | proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops); |
b7454ad3 GC |
1127 | return 0; |
1128 | } | |
1129 | module_init(slab_proc_init); | |
fcf8a1e4 | 1130 | |
5b365771 | 1131 | #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */ |
928cec9c AR |
1132 | |
1133 | static __always_inline void *__do_krealloc(const void *p, size_t new_size, | |
1134 | gfp_t flags) | |
1135 | { | |
1136 | void *ret; | |
fa9ba3aa | 1137 | size_t ks; |
928cec9c | 1138 | |
d12d9ad8 AK |
1139 | /* Don't use instrumented ksize to allow precise KASAN poisoning. */ |
1140 | if (likely(!ZERO_OR_NULL_PTR(p))) { | |
1141 | if (!kasan_check_byte(p)) | |
1142 | return NULL; | |
1143 | ks = kfence_ksize(p) ?: __ksize(p); | |
1144 | } else | |
1145 | ks = 0; | |
928cec9c | 1146 | |
d12d9ad8 | 1147 | /* If the object still fits, repoison it precisely. */ |
0316bec2 | 1148 | if (ks >= new_size) { |
0116523c | 1149 | p = kasan_krealloc((void *)p, new_size, flags); |
928cec9c | 1150 | return (void *)p; |
0316bec2 | 1151 | } |
928cec9c AR |
1152 | |
1153 | ret = kmalloc_track_caller(new_size, flags); | |
d12d9ad8 AK |
1154 | if (ret && p) { |
1155 | /* Disable KASAN checks as the object's redzone is accessed. */ | |
1156 | kasan_disable_current(); | |
1157 | memcpy(ret, kasan_reset_tag(p), ks); | |
1158 | kasan_enable_current(); | |
1159 | } | |
928cec9c AR |
1160 | |
1161 | return ret; | |
1162 | } | |
1163 | ||
928cec9c AR |
1164 | /** |
1165 | * krealloc - reallocate memory. The contents will remain unchanged. | |
1166 | * @p: object to reallocate memory for. | |
1167 | * @new_size: how many bytes of memory are required. | |
1168 | * @flags: the type of memory to allocate. | |
1169 | * | |
1170 | * The contents of the object pointed to are preserved up to the | |
15d5de49 BG |
1171 | * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored). |
1172 | * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size | |
1173 | * is 0 and @p is not a %NULL pointer, the object pointed to is freed. | |
a862f68a MR |
1174 | * |
1175 | * Return: pointer to the allocated memory or %NULL in case of error | |
928cec9c AR |
1176 | */ |
1177 | void *krealloc(const void *p, size_t new_size, gfp_t flags) | |
1178 | { | |
1179 | void *ret; | |
1180 | ||
1181 | if (unlikely(!new_size)) { | |
1182 | kfree(p); | |
1183 | return ZERO_SIZE_PTR; | |
1184 | } | |
1185 | ||
1186 | ret = __do_krealloc(p, new_size, flags); | |
772a2fa5 | 1187 | if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret)) |
928cec9c AR |
1188 | kfree(p); |
1189 | ||
1190 | return ret; | |
1191 | } | |
1192 | EXPORT_SYMBOL(krealloc); | |
1193 | ||
1194 | /** | |
453431a5 | 1195 | * kfree_sensitive - Clear sensitive information in memory before freeing |
928cec9c AR |
1196 | * @p: object to free memory of |
1197 | * | |
1198 | * The memory of the object @p points to is zeroed before freed. | |
453431a5 | 1199 | * If @p is %NULL, kfree_sensitive() does nothing. |
928cec9c AR |
1200 | * |
1201 | * Note: this function zeroes the whole allocated buffer which can be a good | |
1202 | * deal bigger than the requested buffer size passed to kmalloc(). So be | |
1203 | * careful when using this function in performance sensitive code. | |
1204 | */ | |
453431a5 | 1205 | void kfree_sensitive(const void *p) |
928cec9c AR |
1206 | { |
1207 | size_t ks; | |
1208 | void *mem = (void *)p; | |
1209 | ||
928cec9c | 1210 | ks = ksize(mem); |
fa9ba3aa WK |
1211 | if (ks) |
1212 | memzero_explicit(mem, ks); | |
928cec9c AR |
1213 | kfree(mem); |
1214 | } | |
453431a5 | 1215 | EXPORT_SYMBOL(kfree_sensitive); |
928cec9c | 1216 | |
10d1f8cb ME |
1217 | /** |
1218 | * ksize - get the actual amount of memory allocated for a given object | |
1219 | * @objp: Pointer to the object | |
1220 | * | |
1221 | * kmalloc may internally round up allocations and return more memory | |
1222 | * than requested. ksize() can be used to determine the actual amount of | |
1223 | * memory allocated. The caller may use this additional memory, even though | |
1224 | * a smaller amount of memory was initially specified with the kmalloc call. | |
1225 | * The caller must guarantee that objp points to a valid object previously | |
1226 | * allocated with either kmalloc() or kmem_cache_alloc(). The object | |
1227 | * must not be freed during the duration of the call. | |
1228 | * | |
1229 | * Return: size of the actual memory used by @objp in bytes | |
1230 | */ | |
1231 | size_t ksize(const void *objp) | |
1232 | { | |
0d4ca4c9 ME |
1233 | size_t size; |
1234 | ||
0d4ca4c9 | 1235 | /* |
611806b4 AK |
1236 | * We need to first check that the pointer to the object is valid, and |
1237 | * only then unpoison the memory. The report printed from ksize() is | |
1238 | * more useful, then when it's printed later when the behaviour could | |
1239 | * be undefined due to a potential use-after-free or double-free. | |
0d4ca4c9 | 1240 | * |
611806b4 AK |
1241 | * We use kasan_check_byte(), which is supported for the hardware |
1242 | * tag-based KASAN mode, unlike kasan_check_read/write(). | |
1243 | * | |
1244 | * If the pointed to memory is invalid, we return 0 to avoid users of | |
0d4ca4c9 ME |
1245 | * ksize() writing to and potentially corrupting the memory region. |
1246 | * | |
1247 | * We want to perform the check before __ksize(), to avoid potentially | |
1248 | * crashing in __ksize() due to accessing invalid metadata. | |
1249 | */ | |
611806b4 | 1250 | if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp)) |
0d4ca4c9 ME |
1251 | return 0; |
1252 | ||
d3fb45f3 | 1253 | size = kfence_ksize(objp) ?: __ksize(objp); |
10d1f8cb ME |
1254 | /* |
1255 | * We assume that ksize callers could use whole allocated area, | |
1256 | * so we need to unpoison this area. | |
1257 | */ | |
cebd0eb2 | 1258 | kasan_unpoison_range(objp, size); |
10d1f8cb ME |
1259 | return size; |
1260 | } | |
1261 | EXPORT_SYMBOL(ksize); | |
1262 | ||
928cec9c AR |
1263 | /* Tracepoints definitions. */ |
1264 | EXPORT_TRACEPOINT_SYMBOL(kmalloc); | |
1265 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc); | |
1266 | EXPORT_TRACEPOINT_SYMBOL(kmalloc_node); | |
1267 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node); | |
1268 | EXPORT_TRACEPOINT_SYMBOL(kfree); | |
1269 | EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free); | |
4f6923fb HM |
1270 | |
1271 | int should_failslab(struct kmem_cache *s, gfp_t gfpflags) | |
1272 | { | |
1273 | if (__should_failslab(s, gfpflags)) | |
1274 | return -ENOMEM; | |
1275 | return 0; | |
1276 | } | |
1277 | ALLOW_ERROR_INJECTION(should_failslab, ERRNO); |