Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
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>
13 : #include <linux/cache.h>
14 : #include <linux/compiler.h>
15 : #include <linux/kfence.h>
16 : #include <linux/module.h>
17 : #include <linux/cpu.h>
18 : #include <linux/uaccess.h>
19 : #include <linux/seq_file.h>
20 : #include <linux/proc_fs.h>
21 : #include <linux/debugfs.h>
22 : #include <linux/kasan.h>
23 : #include <asm/cacheflush.h>
24 : #include <asm/tlbflush.h>
25 : #include <asm/page.h>
26 : #include <linux/memcontrol.h>
27 :
28 : #define CREATE_TRACE_POINTS
29 : #include <trace/events/kmem.h>
30 :
31 : #include "internal.h"
32 :
33 : #include "slab.h"
34 :
35 : enum slab_state slab_state;
36 : LIST_HEAD(slab_caches);
37 : DEFINE_MUTEX(slab_mutex);
38 : struct kmem_cache *kmem_cache;
39 :
40 : static LIST_HEAD(slab_caches_to_rcu_destroy);
41 : static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
42 : static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
43 : slab_caches_to_rcu_destroy_workfn);
44 :
45 : /*
46 : * Set of flags that will prevent slab merging
47 : */
48 : #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
49 : SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
50 : SLAB_FAILSLAB | kasan_never_merge())
51 :
52 : #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
53 : SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
54 :
55 : /*
56 : * Merge control. If this is set then no merging of slab caches will occur.
57 : */
58 : static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
59 :
60 0 : static int __init setup_slab_nomerge(char *str)
61 : {
62 0 : slab_nomerge = true;
63 0 : return 1;
64 : }
65 :
66 0 : static int __init setup_slab_merge(char *str)
67 : {
68 0 : slab_nomerge = false;
69 0 : return 1;
70 : }
71 :
72 : #ifdef CONFIG_SLUB
73 : __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
74 : __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
75 : #endif
76 :
77 : __setup("slab_nomerge", setup_slab_nomerge);
78 : __setup("slab_merge", setup_slab_merge);
79 :
80 : /*
81 : * Determine the size of a slab object
82 : */
83 0 : unsigned int kmem_cache_size(struct kmem_cache *s)
84 : {
85 0 : return s->object_size;
86 : }
87 : EXPORT_SYMBOL(kmem_cache_size);
88 :
89 : #ifdef CONFIG_DEBUG_VM
90 : static int kmem_cache_sanity_check(const char *name, unsigned int size)
91 : {
92 : if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
93 : pr_err("kmem_cache_create(%s) integrity check failed\n", name);
94 : return -EINVAL;
95 : }
96 :
97 : WARN_ON(strchr(name, ' ')); /* It confuses parsers */
98 : return 0;
99 : }
100 : #else
101 : static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
102 : {
103 : return 0;
104 : }
105 : #endif
106 :
107 0 : void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
108 : {
109 : size_t i;
110 :
111 0 : for (i = 0; i < nr; i++) {
112 0 : if (s)
113 0 : kmem_cache_free(s, p[i]);
114 : else
115 0 : kfree(p[i]);
116 : }
117 0 : }
118 :
119 0 : int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
120 : void **p)
121 : {
122 : size_t i;
123 :
124 0 : for (i = 0; i < nr; i++) {
125 0 : void *x = p[i] = kmem_cache_alloc(s, flags);
126 0 : if (!x) {
127 0 : __kmem_cache_free_bulk(s, i, p);
128 0 : return 0;
129 : }
130 : }
131 0 : return i;
132 : }
133 :
134 : /*
135 : * Figure out what the alignment of the objects will be given a set of
136 : * flags, a user specified alignment and the size of the objects.
137 : */
138 : static unsigned int calculate_alignment(slab_flags_t flags,
139 : unsigned int align, unsigned int size)
140 : {
141 : /*
142 : * If the user wants hardware cache aligned objects then follow that
143 : * suggestion if the object is sufficiently large.
144 : *
145 : * The hardware cache alignment cannot override the specified
146 : * alignment though. If that is greater then use it.
147 : */
148 114 : if (flags & SLAB_HWCACHE_ALIGN) {
149 : unsigned int ralign;
150 :
151 37 : ralign = cache_line_size();
152 38 : while (size <= ralign / 2)
153 : ralign /= 2;
154 37 : align = max(align, ralign);
155 : }
156 :
157 114 : if (align < ARCH_SLAB_MINALIGN)
158 29 : align = ARCH_SLAB_MINALIGN;
159 :
160 114 : return ALIGN(align, sizeof(void *));
161 : }
162 :
163 : /*
164 : * Find a mergeable slab cache
165 : */
166 67 : int slab_unmergeable(struct kmem_cache *s)
167 : {
168 2080 : if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
169 : return 1;
170 :
171 2008 : if (s->ctor)
172 : return 1;
173 :
174 1879 : if (s->usersize)
175 : return 1;
176 :
177 : /*
178 : * We may have set a slab to be unmergeable during bootstrap.
179 : */
180 530 : if (s->refcount < 0)
181 : return 1;
182 :
183 24 : return 0;
184 : }
185 :
186 54 : struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
187 : slab_flags_t flags, const char *name, void (*ctor)(void *))
188 : {
189 : struct kmem_cache *s;
190 :
191 54 : if (slab_nomerge)
192 : return NULL;
193 :
194 54 : if (ctor)
195 : return NULL;
196 :
197 47 : size = ALIGN(size, sizeof(void *));
198 47 : align = calculate_alignment(flags, align, size);
199 47 : size = ALIGN(size, align);
200 47 : flags = kmem_cache_flags(size, flags, name);
201 :
202 47 : if (flags & SLAB_NEVER_MERGE)
203 : return NULL;
204 :
205 2037 : list_for_each_entry_reverse(s, &slab_caches, list) {
206 2013 : if (slab_unmergeable(s))
207 1599 : continue;
208 :
209 414 : if (size > s->size)
210 186 : continue;
211 :
212 228 : if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
213 13 : continue;
214 : /*
215 : * Check if alignment is compatible.
216 : * Courtesy of Adrian Drzewiecki
217 : */
218 215 : if ((s->size & ~(align - 1)) != s->size)
219 15 : continue;
220 :
221 200 : if (s->size - size >= sizeof(void *))
222 179 : continue;
223 :
224 : if (IS_ENABLED(CONFIG_SLAB) && align &&
225 : (align > s->align || s->align % align))
226 : continue;
227 :
228 : return s;
229 : }
230 : return NULL;
231 : }
232 :
233 39 : static struct kmem_cache *create_cache(const char *name,
234 : unsigned int object_size, unsigned int align,
235 : slab_flags_t flags, unsigned int useroffset,
236 : unsigned int usersize, void (*ctor)(void *),
237 : struct kmem_cache *root_cache)
238 : {
239 : struct kmem_cache *s;
240 : int err;
241 :
242 39 : if (WARN_ON(useroffset + usersize > object_size))
243 0 : useroffset = usersize = 0;
244 :
245 39 : err = -ENOMEM;
246 78 : s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
247 39 : if (!s)
248 : goto out;
249 :
250 39 : s->name = name;
251 39 : s->size = s->object_size = object_size;
252 39 : s->align = align;
253 39 : s->ctor = ctor;
254 39 : s->useroffset = useroffset;
255 39 : s->usersize = usersize;
256 :
257 39 : err = __kmem_cache_create(s, flags);
258 39 : if (err)
259 : goto out_free_cache;
260 :
261 39 : s->refcount = 1;
262 39 : list_add(&s->list, &slab_caches);
263 : out:
264 39 : if (err)
265 0 : return ERR_PTR(err);
266 : return s;
267 :
268 : out_free_cache:
269 0 : kmem_cache_free(kmem_cache, s);
270 : goto out;
271 : }
272 :
273 : /**
274 : * kmem_cache_create_usercopy - Create a cache with a region suitable
275 : * for copying to userspace
276 : * @name: A string which is used in /proc/slabinfo to identify this cache.
277 : * @size: The size of objects to be created in this cache.
278 : * @align: The required alignment for the objects.
279 : * @flags: SLAB flags
280 : * @useroffset: Usercopy region offset
281 : * @usersize: Usercopy region size
282 : * @ctor: A constructor for the objects.
283 : *
284 : * Cannot be called within a interrupt, but can be interrupted.
285 : * The @ctor is run when new pages are allocated by the cache.
286 : *
287 : * The flags are
288 : *
289 : * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
290 : * to catch references to uninitialised memory.
291 : *
292 : * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
293 : * for buffer overruns.
294 : *
295 : * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
296 : * cacheline. This can be beneficial if you're counting cycles as closely
297 : * as davem.
298 : *
299 : * Return: a pointer to the cache on success, NULL on failure.
300 : */
301 : struct kmem_cache *
302 60 : kmem_cache_create_usercopy(const char *name,
303 : unsigned int size, unsigned int align,
304 : slab_flags_t flags,
305 : unsigned int useroffset, unsigned int usersize,
306 : void (*ctor)(void *))
307 : {
308 60 : struct kmem_cache *s = NULL;
309 : const char *cache_name;
310 : int err;
311 :
312 : #ifdef CONFIG_SLUB_DEBUG
313 : /*
314 : * If no slub_debug was enabled globally, the static key is not yet
315 : * enabled by setup_slub_debug(). Enable it if the cache is being
316 : * created with any of the debugging flags passed explicitly.
317 : */
318 60 : if (flags & SLAB_DEBUG_FLAGS)
319 0 : static_branch_enable(&slub_debug_enabled);
320 : #endif
321 :
322 60 : mutex_lock(&slab_mutex);
323 :
324 60 : err = kmem_cache_sanity_check(name, size);
325 : if (err) {
326 : goto out_unlock;
327 : }
328 :
329 : /* Refuse requests with allocator specific flags */
330 60 : if (flags & ~SLAB_FLAGS_PERMITTED) {
331 : err = -EINVAL;
332 : goto out_unlock;
333 : }
334 :
335 : /*
336 : * Some allocators will constraint the set of valid flags to a subset
337 : * of all flags. We expect them to define CACHE_CREATE_MASK in this
338 : * case, and we'll just provide them with a sanitized version of the
339 : * passed flags.
340 : */
341 60 : flags &= CACHE_CREATE_MASK;
342 :
343 : /* Fail closed on bad usersize of useroffset values. */
344 120 : if (WARN_ON(!usersize && useroffset) ||
345 60 : WARN_ON(size < usersize || size - usersize < useroffset))
346 : usersize = useroffset = 0;
347 :
348 60 : if (!usersize)
349 54 : s = __kmem_cache_alias(name, size, align, flags, ctor);
350 60 : if (s)
351 : goto out_unlock;
352 :
353 39 : cache_name = kstrdup_const(name, GFP_KERNEL);
354 39 : if (!cache_name) {
355 : err = -ENOMEM;
356 : goto out_unlock;
357 : }
358 :
359 39 : s = create_cache(cache_name, size,
360 : calculate_alignment(flags, align, size),
361 : flags, useroffset, usersize, ctor, NULL);
362 39 : if (IS_ERR(s)) {
363 0 : err = PTR_ERR(s);
364 0 : kfree_const(cache_name);
365 : }
366 :
367 : out_unlock:
368 60 : mutex_unlock(&slab_mutex);
369 :
370 60 : if (err) {
371 0 : if (flags & SLAB_PANIC)
372 0 : panic("%s: Failed to create slab '%s'. Error %d\n",
373 : __func__, name, err);
374 : else {
375 0 : pr_warn("%s(%s) failed with error %d\n",
376 : __func__, name, err);
377 0 : dump_stack();
378 : }
379 0 : return NULL;
380 : }
381 : return s;
382 : }
383 : EXPORT_SYMBOL(kmem_cache_create_usercopy);
384 :
385 : /**
386 : * kmem_cache_create - Create a cache.
387 : * @name: A string which is used in /proc/slabinfo to identify this cache.
388 : * @size: The size of objects to be created in this cache.
389 : * @align: The required alignment for the objects.
390 : * @flags: SLAB flags
391 : * @ctor: A constructor for the objects.
392 : *
393 : * Cannot be called within a interrupt, but can be interrupted.
394 : * The @ctor is run when new pages are allocated by the cache.
395 : *
396 : * The flags are
397 : *
398 : * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
399 : * to catch references to uninitialised memory.
400 : *
401 : * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
402 : * for buffer overruns.
403 : *
404 : * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
405 : * cacheline. This can be beneficial if you're counting cycles as closely
406 : * as davem.
407 : *
408 : * Return: a pointer to the cache on success, NULL on failure.
409 : */
410 : struct kmem_cache *
411 54 : kmem_cache_create(const char *name, unsigned int size, unsigned int align,
412 : slab_flags_t flags, void (*ctor)(void *))
413 : {
414 54 : return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
415 : ctor);
416 : }
417 : EXPORT_SYMBOL(kmem_cache_create);
418 :
419 0 : static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
420 : {
421 0 : LIST_HEAD(to_destroy);
422 : struct kmem_cache *s, *s2;
423 :
424 : /*
425 : * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
426 : * @slab_caches_to_rcu_destroy list. The slab pages are freed
427 : * through RCU and the associated kmem_cache are dereferenced
428 : * while freeing the pages, so the kmem_caches should be freed only
429 : * after the pending RCU operations are finished. As rcu_barrier()
430 : * is a pretty slow operation, we batch all pending destructions
431 : * asynchronously.
432 : */
433 0 : mutex_lock(&slab_mutex);
434 0 : list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
435 0 : mutex_unlock(&slab_mutex);
436 :
437 0 : if (list_empty(&to_destroy))
438 0 : return;
439 :
440 0 : rcu_barrier();
441 :
442 0 : list_for_each_entry_safe(s, s2, &to_destroy, list) {
443 0 : debugfs_slab_release(s);
444 0 : kfence_shutdown_cache(s);
445 : #ifdef SLAB_SUPPORTS_SYSFS
446 0 : sysfs_slab_release(s);
447 : #else
448 : slab_kmem_cache_release(s);
449 : #endif
450 : }
451 : }
452 :
453 0 : static int shutdown_cache(struct kmem_cache *s)
454 : {
455 : /* free asan quarantined objects */
456 0 : kasan_cache_shutdown(s);
457 :
458 0 : if (__kmem_cache_shutdown(s) != 0)
459 : return -EBUSY;
460 :
461 0 : list_del(&s->list);
462 :
463 0 : if (s->flags & SLAB_TYPESAFE_BY_RCU) {
464 : #ifdef SLAB_SUPPORTS_SYSFS
465 0 : sysfs_slab_unlink(s);
466 : #endif
467 0 : list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
468 : schedule_work(&slab_caches_to_rcu_destroy_work);
469 : } else {
470 0 : kfence_shutdown_cache(s);
471 0 : debugfs_slab_release(s);
472 : #ifdef SLAB_SUPPORTS_SYSFS
473 0 : sysfs_slab_unlink(s);
474 0 : sysfs_slab_release(s);
475 : #else
476 : slab_kmem_cache_release(s);
477 : #endif
478 : }
479 :
480 : return 0;
481 : }
482 :
483 0 : void slab_kmem_cache_release(struct kmem_cache *s)
484 : {
485 0 : __kmem_cache_release(s);
486 0 : kfree_const(s->name);
487 0 : kmem_cache_free(kmem_cache, s);
488 0 : }
489 :
490 0 : void kmem_cache_destroy(struct kmem_cache *s)
491 : {
492 0 : if (unlikely(!s) || !kasan_check_byte(s))
493 : return;
494 :
495 : cpus_read_lock();
496 0 : mutex_lock(&slab_mutex);
497 :
498 0 : s->refcount--;
499 0 : if (s->refcount)
500 : goto out_unlock;
501 :
502 0 : WARN(shutdown_cache(s),
503 : "%s %s: Slab cache still has objects when called from %pS",
504 : __func__, s->name, (void *)_RET_IP_);
505 : out_unlock:
506 0 : mutex_unlock(&slab_mutex);
507 : cpus_read_unlock();
508 : }
509 : EXPORT_SYMBOL(kmem_cache_destroy);
510 :
511 : /**
512 : * kmem_cache_shrink - Shrink a cache.
513 : * @cachep: The cache to shrink.
514 : *
515 : * Releases as many slabs as possible for a cache.
516 : * To help debugging, a zero exit status indicates all slabs were released.
517 : *
518 : * Return: %0 if all slabs were released, non-zero otherwise
519 : */
520 0 : int kmem_cache_shrink(struct kmem_cache *cachep)
521 : {
522 : int ret;
523 :
524 :
525 0 : kasan_cache_shrink(cachep);
526 0 : ret = __kmem_cache_shrink(cachep);
527 :
528 0 : return ret;
529 : }
530 : EXPORT_SYMBOL(kmem_cache_shrink);
531 :
532 22 : bool slab_is_available(void)
533 : {
534 22 : return slab_state >= UP;
535 : }
536 :
537 : #ifdef CONFIG_PRINTK
538 : /**
539 : * kmem_valid_obj - does the pointer reference a valid slab object?
540 : * @object: pointer to query.
541 : *
542 : * Return: %true if the pointer is to a not-yet-freed object from
543 : * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
544 : * is to an already-freed object, and %false otherwise.
545 : */
546 0 : bool kmem_valid_obj(void *object)
547 : {
548 : struct folio *folio;
549 :
550 : /* Some arches consider ZERO_SIZE_PTR to be a valid address. */
551 0 : if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
552 : return false;
553 0 : folio = virt_to_folio(object);
554 0 : return folio_test_slab(folio);
555 : }
556 : EXPORT_SYMBOL_GPL(kmem_valid_obj);
557 :
558 : static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
559 : {
560 0 : if (__kfence_obj_info(kpp, object, slab))
561 : return;
562 0 : __kmem_obj_info(kpp, object, slab);
563 : }
564 :
565 : /**
566 : * kmem_dump_obj - Print available slab provenance information
567 : * @object: slab object for which to find provenance information.
568 : *
569 : * This function uses pr_cont(), so that the caller is expected to have
570 : * printed out whatever preamble is appropriate. The provenance information
571 : * depends on the type of object and on how much debugging is enabled.
572 : * For a slab-cache object, the fact that it is a slab object is printed,
573 : * and, if available, the slab name, return address, and stack trace from
574 : * the allocation and last free path of that object.
575 : *
576 : * This function will splat if passed a pointer to a non-slab object.
577 : * If you are not sure what type of object you have, you should instead
578 : * use mem_dump_obj().
579 : */
580 0 : void kmem_dump_obj(void *object)
581 : {
582 0 : char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
583 : int i;
584 : struct slab *slab;
585 : unsigned long ptroffset;
586 0 : struct kmem_obj_info kp = { };
587 :
588 0 : if (WARN_ON_ONCE(!virt_addr_valid(object)))
589 0 : return;
590 0 : slab = virt_to_slab(object);
591 0 : if (WARN_ON_ONCE(!slab)) {
592 0 : pr_cont(" non-slab memory.\n");
593 0 : return;
594 : }
595 0 : kmem_obj_info(&kp, object, slab);
596 0 : if (kp.kp_slab_cache)
597 0 : pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
598 : else
599 0 : pr_cont(" slab%s", cp);
600 0 : if (is_kfence_address(object))
601 : pr_cont(" (kfence)");
602 0 : if (kp.kp_objp)
603 0 : pr_cont(" start %px", kp.kp_objp);
604 0 : if (kp.kp_data_offset)
605 0 : pr_cont(" data offset %lu", kp.kp_data_offset);
606 0 : if (kp.kp_objp) {
607 0 : ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
608 0 : pr_cont(" pointer offset %lu", ptroffset);
609 : }
610 0 : if (kp.kp_slab_cache && kp.kp_slab_cache->usersize)
611 0 : pr_cont(" size %u", kp.kp_slab_cache->usersize);
612 0 : if (kp.kp_ret)
613 0 : pr_cont(" allocated at %pS\n", kp.kp_ret);
614 : else
615 0 : pr_cont("\n");
616 0 : for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
617 0 : if (!kp.kp_stack[i])
618 : break;
619 0 : pr_info(" %pS\n", kp.kp_stack[i]);
620 : }
621 :
622 0 : if (kp.kp_free_stack[0])
623 0 : pr_cont(" Free path:\n");
624 :
625 0 : for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
626 0 : if (!kp.kp_free_stack[i])
627 : break;
628 0 : pr_info(" %pS\n", kp.kp_free_stack[i]);
629 : }
630 :
631 : }
632 : EXPORT_SYMBOL_GPL(kmem_dump_obj);
633 : #endif
634 :
635 : #ifndef CONFIG_SLOB
636 : /* Create a cache during boot when no slab services are available yet */
637 28 : void __init create_boot_cache(struct kmem_cache *s, const char *name,
638 : unsigned int size, slab_flags_t flags,
639 : unsigned int useroffset, unsigned int usersize)
640 : {
641 : int err;
642 28 : unsigned int align = ARCH_KMALLOC_MINALIGN;
643 :
644 28 : s->name = name;
645 28 : s->size = s->object_size = size;
646 :
647 : /*
648 : * For power of two sizes, guarantee natural alignment for kmalloc
649 : * caches, regardless of SL*B debugging options.
650 : */
651 56 : if (is_power_of_2(size))
652 22 : align = max(align, size);
653 28 : s->align = calculate_alignment(flags, align, size);
654 :
655 28 : s->useroffset = useroffset;
656 28 : s->usersize = usersize;
657 :
658 28 : err = __kmem_cache_create(s, flags);
659 :
660 28 : if (err)
661 0 : panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
662 : name, size, err);
663 :
664 28 : s->refcount = -1; /* Exempt from merging for now */
665 28 : }
666 :
667 26 : struct kmem_cache *__init create_kmalloc_cache(const char *name,
668 : unsigned int size, slab_flags_t flags,
669 : unsigned int useroffset, unsigned int usersize)
670 : {
671 52 : struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
672 :
673 26 : if (!s)
674 0 : panic("Out of memory when creating slab %s\n", name);
675 :
676 26 : create_boot_cache(s, name, size, flags, useroffset, usersize);
677 : kasan_cache_create_kmalloc(s);
678 52 : list_add(&s->list, &slab_caches);
679 26 : s->refcount = 1;
680 26 : return s;
681 : }
682 :
683 : struct kmem_cache *
684 : kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
685 : { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
686 : EXPORT_SYMBOL(kmalloc_caches);
687 :
688 : /*
689 : * Conversion table for small slabs sizes / 8 to the index in the
690 : * kmalloc array. This is necessary for slabs < 192 since we have non power
691 : * of two cache sizes there. The size of larger slabs can be determined using
692 : * fls.
693 : */
694 : static u8 size_index[24] __ro_after_init = {
695 : 3, /* 8 */
696 : 4, /* 16 */
697 : 5, /* 24 */
698 : 5, /* 32 */
699 : 6, /* 40 */
700 : 6, /* 48 */
701 : 6, /* 56 */
702 : 6, /* 64 */
703 : 1, /* 72 */
704 : 1, /* 80 */
705 : 1, /* 88 */
706 : 1, /* 96 */
707 : 7, /* 104 */
708 : 7, /* 112 */
709 : 7, /* 120 */
710 : 7, /* 128 */
711 : 2, /* 136 */
712 : 2, /* 144 */
713 : 2, /* 152 */
714 : 2, /* 160 */
715 : 2, /* 168 */
716 : 2, /* 176 */
717 : 2, /* 184 */
718 : 2 /* 192 */
719 : };
720 :
721 : static inline unsigned int size_index_elem(unsigned int bytes)
722 : {
723 3327 : return (bytes - 1) / 8;
724 : }
725 :
726 : /*
727 : * Find the kmem_cache structure that serves a given size of
728 : * allocation
729 : */
730 3398 : struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
731 : {
732 : unsigned int index;
733 :
734 3398 : if (size <= 192) {
735 3327 : if (!size)
736 : return ZERO_SIZE_PTR;
737 :
738 6654 : index = size_index[size_index_elem(size)];
739 : } else {
740 71 : if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
741 : return NULL;
742 142 : index = fls(size - 1);
743 : }
744 :
745 3398 : return kmalloc_caches[kmalloc_type(flags)][index];
746 : }
747 :
748 : #ifdef CONFIG_ZONE_DMA
749 : #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
750 : #else
751 : #define KMALLOC_DMA_NAME(sz)
752 : #endif
753 :
754 : #ifdef CONFIG_MEMCG_KMEM
755 : #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
756 : #else
757 : #define KMALLOC_CGROUP_NAME(sz)
758 : #endif
759 :
760 : #define INIT_KMALLOC_INFO(__size, __short_size) \
761 : { \
762 : .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
763 : .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
764 : KMALLOC_CGROUP_NAME(__short_size) \
765 : KMALLOC_DMA_NAME(__short_size) \
766 : .size = __size, \
767 : }
768 :
769 : /*
770 : * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
771 : * kmalloc_index() supports up to 2^25=32MB, so the final entry of the table is
772 : * kmalloc-32M.
773 : */
774 : const struct kmalloc_info_struct kmalloc_info[] __initconst = {
775 : INIT_KMALLOC_INFO(0, 0),
776 : INIT_KMALLOC_INFO(96, 96),
777 : INIT_KMALLOC_INFO(192, 192),
778 : INIT_KMALLOC_INFO(8, 8),
779 : INIT_KMALLOC_INFO(16, 16),
780 : INIT_KMALLOC_INFO(32, 32),
781 : INIT_KMALLOC_INFO(64, 64),
782 : INIT_KMALLOC_INFO(128, 128),
783 : INIT_KMALLOC_INFO(256, 256),
784 : INIT_KMALLOC_INFO(512, 512),
785 : INIT_KMALLOC_INFO(1024, 1k),
786 : INIT_KMALLOC_INFO(2048, 2k),
787 : INIT_KMALLOC_INFO(4096, 4k),
788 : INIT_KMALLOC_INFO(8192, 8k),
789 : INIT_KMALLOC_INFO(16384, 16k),
790 : INIT_KMALLOC_INFO(32768, 32k),
791 : INIT_KMALLOC_INFO(65536, 64k),
792 : INIT_KMALLOC_INFO(131072, 128k),
793 : INIT_KMALLOC_INFO(262144, 256k),
794 : INIT_KMALLOC_INFO(524288, 512k),
795 : INIT_KMALLOC_INFO(1048576, 1M),
796 : INIT_KMALLOC_INFO(2097152, 2M),
797 : INIT_KMALLOC_INFO(4194304, 4M),
798 : INIT_KMALLOC_INFO(8388608, 8M),
799 : INIT_KMALLOC_INFO(16777216, 16M),
800 : INIT_KMALLOC_INFO(33554432, 32M)
801 : };
802 :
803 : /*
804 : * Patch up the size_index table if we have strange large alignment
805 : * requirements for the kmalloc array. This is only the case for
806 : * MIPS it seems. The standard arches will not generate any code here.
807 : *
808 : * Largest permitted alignment is 256 bytes due to the way we
809 : * handle the index determination for the smaller caches.
810 : *
811 : * Make sure that nothing crazy happens if someone starts tinkering
812 : * around with ARCH_KMALLOC_MINALIGN
813 : */
814 1 : void __init setup_kmalloc_cache_index_table(void)
815 : {
816 : unsigned int i;
817 :
818 1 : BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
819 : !is_power_of_2(KMALLOC_MIN_SIZE));
820 :
821 1 : for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
822 : unsigned int elem = size_index_elem(i);
823 :
824 : if (elem >= ARRAY_SIZE(size_index))
825 : break;
826 : size_index[elem] = KMALLOC_SHIFT_LOW;
827 : }
828 :
829 : if (KMALLOC_MIN_SIZE >= 64) {
830 : /*
831 : * The 96 byte sized cache is not used if the alignment
832 : * is 64 byte.
833 : */
834 : for (i = 64 + 8; i <= 96; i += 8)
835 : size_index[size_index_elem(i)] = 7;
836 :
837 : }
838 :
839 : if (KMALLOC_MIN_SIZE >= 128) {
840 : /*
841 : * The 192 byte sized cache is not used if the alignment
842 : * is 128 byte. Redirect kmalloc to use the 256 byte cache
843 : * instead.
844 : */
845 : for (i = 128 + 8; i <= 192; i += 8)
846 : size_index[size_index_elem(i)] = 8;
847 : }
848 1 : }
849 :
850 : static void __init
851 26 : new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
852 : {
853 26 : if (type == KMALLOC_RECLAIM) {
854 13 : flags |= SLAB_RECLAIM_ACCOUNT;
855 : } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
856 : if (mem_cgroup_kmem_disabled()) {
857 : kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
858 : return;
859 : }
860 : flags |= SLAB_ACCOUNT;
861 : }
862 :
863 26 : kmalloc_caches[type][idx] = create_kmalloc_cache(
864 : kmalloc_info[idx].name[type],
865 : kmalloc_info[idx].size, flags, 0,
866 : kmalloc_info[idx].size);
867 :
868 : /*
869 : * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
870 : * KMALLOC_NORMAL caches.
871 : */
872 : if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
873 : kmalloc_caches[type][idx]->refcount = -1;
874 : }
875 :
876 : /*
877 : * Create the kmalloc array. Some of the regular kmalloc arrays
878 : * may already have been created because they were needed to
879 : * enable allocations for slab creation.
880 : */
881 1 : void __init create_kmalloc_caches(slab_flags_t flags)
882 : {
883 : int i;
884 : enum kmalloc_cache_type type;
885 :
886 : /*
887 : * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
888 : */
889 3 : for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
890 22 : for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
891 22 : if (!kmalloc_caches[type][i])
892 22 : new_kmalloc_cache(i, type, flags);
893 :
894 : /*
895 : * Caches that are not of the two-to-the-power-of size.
896 : * These have to be created immediately after the
897 : * earlier power of two caches
898 : */
899 24 : if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
900 2 : !kmalloc_caches[type][1])
901 2 : new_kmalloc_cache(1, type, flags);
902 24 : if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
903 2 : !kmalloc_caches[type][2])
904 2 : new_kmalloc_cache(2, type, flags);
905 : }
906 : }
907 :
908 : /* Kmalloc array is now usable */
909 1 : slab_state = UP;
910 :
911 : #ifdef CONFIG_ZONE_DMA
912 : for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
913 : struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
914 :
915 : if (s) {
916 : kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
917 : kmalloc_info[i].name[KMALLOC_DMA],
918 : kmalloc_info[i].size,
919 : SLAB_CACHE_DMA | flags, 0,
920 : kmalloc_info[i].size);
921 : }
922 : }
923 : #endif
924 1 : }
925 : #endif /* !CONFIG_SLOB */
926 :
927 0 : gfp_t kmalloc_fix_flags(gfp_t flags)
928 : {
929 0 : gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
930 :
931 0 : flags &= ~GFP_SLAB_BUG_MASK;
932 0 : pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
933 : invalid_mask, &invalid_mask, flags, &flags);
934 0 : dump_stack();
935 :
936 0 : return flags;
937 : }
938 :
939 : /*
940 : * To avoid unnecessary overhead, we pass through large allocation requests
941 : * directly to the page allocator. We use __GFP_COMP, because we will need to
942 : * know the allocation order to free the pages properly in kfree.
943 : */
944 8 : void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
945 : {
946 8 : void *ret = NULL;
947 : struct page *page;
948 :
949 8 : if (unlikely(flags & GFP_SLAB_BUG_MASK))
950 0 : flags = kmalloc_fix_flags(flags);
951 :
952 8 : flags |= __GFP_COMP;
953 8 : page = alloc_pages(flags, order);
954 8 : if (likely(page)) {
955 8 : ret = page_address(page);
956 8 : mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
957 8 : PAGE_SIZE << order);
958 : }
959 8 : ret = kasan_kmalloc_large(ret, size, flags);
960 : /* As ret might get tagged, call kmemleak hook after KASAN. */
961 8 : kmemleak_alloc(ret, size, 1, flags);
962 8 : return ret;
963 : }
964 : EXPORT_SYMBOL(kmalloc_order);
965 :
966 : #ifdef CONFIG_TRACING
967 : void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
968 : {
969 : void *ret = kmalloc_order(size, flags, order);
970 : trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
971 : return ret;
972 : }
973 : EXPORT_SYMBOL(kmalloc_order_trace);
974 : #endif
975 :
976 : #ifdef CONFIG_SLAB_FREELIST_RANDOM
977 : /* Randomize a generic freelist */
978 : static void freelist_randomize(struct rnd_state *state, unsigned int *list,
979 : unsigned int count)
980 : {
981 : unsigned int rand;
982 : unsigned int i;
983 :
984 : for (i = 0; i < count; i++)
985 : list[i] = i;
986 :
987 : /* Fisher-Yates shuffle */
988 : for (i = count - 1; i > 0; i--) {
989 : rand = prandom_u32_state(state);
990 : rand %= (i + 1);
991 : swap(list[i], list[rand]);
992 : }
993 : }
994 :
995 : /* Create a random sequence per cache */
996 : int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
997 : gfp_t gfp)
998 : {
999 : struct rnd_state state;
1000 :
1001 : if (count < 2 || cachep->random_seq)
1002 : return 0;
1003 :
1004 : cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1005 : if (!cachep->random_seq)
1006 : return -ENOMEM;
1007 :
1008 : /* Get best entropy at this stage of boot */
1009 : prandom_seed_state(&state, get_random_long());
1010 :
1011 : freelist_randomize(&state, cachep->random_seq, count);
1012 : return 0;
1013 : }
1014 :
1015 : /* Destroy the per-cache random freelist sequence */
1016 : void cache_random_seq_destroy(struct kmem_cache *cachep)
1017 : {
1018 : kfree(cachep->random_seq);
1019 : cachep->random_seq = NULL;
1020 : }
1021 : #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1022 :
1023 : #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1024 : #ifdef CONFIG_SLAB
1025 : #define SLABINFO_RIGHTS (0600)
1026 : #else
1027 : #define SLABINFO_RIGHTS (0400)
1028 : #endif
1029 :
1030 0 : static void print_slabinfo_header(struct seq_file *m)
1031 : {
1032 : /*
1033 : * Output format version, so at least we can change it
1034 : * without _too_ many complaints.
1035 : */
1036 : #ifdef CONFIG_DEBUG_SLAB
1037 : seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1038 : #else
1039 0 : seq_puts(m, "slabinfo - version: 2.1\n");
1040 : #endif
1041 0 : seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1042 0 : seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1043 0 : seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1044 : #ifdef CONFIG_DEBUG_SLAB
1045 : seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1046 : seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1047 : #endif
1048 0 : seq_putc(m, '\n');
1049 0 : }
1050 :
1051 0 : static void *slab_start(struct seq_file *m, loff_t *pos)
1052 : {
1053 0 : mutex_lock(&slab_mutex);
1054 0 : return seq_list_start(&slab_caches, *pos);
1055 : }
1056 :
1057 0 : static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1058 : {
1059 0 : return seq_list_next(p, &slab_caches, pos);
1060 : }
1061 :
1062 0 : static void slab_stop(struct seq_file *m, void *p)
1063 : {
1064 0 : mutex_unlock(&slab_mutex);
1065 0 : }
1066 :
1067 0 : static void cache_show(struct kmem_cache *s, struct seq_file *m)
1068 : {
1069 : struct slabinfo sinfo;
1070 :
1071 0 : memset(&sinfo, 0, sizeof(sinfo));
1072 0 : get_slabinfo(s, &sinfo);
1073 :
1074 0 : seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1075 : s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1076 0 : sinfo.objects_per_slab, (1 << sinfo.cache_order));
1077 :
1078 0 : seq_printf(m, " : tunables %4u %4u %4u",
1079 : sinfo.limit, sinfo.batchcount, sinfo.shared);
1080 0 : seq_printf(m, " : slabdata %6lu %6lu %6lu",
1081 : sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1082 0 : slabinfo_show_stats(m, s);
1083 0 : seq_putc(m, '\n');
1084 0 : }
1085 :
1086 0 : static int slab_show(struct seq_file *m, void *p)
1087 : {
1088 0 : struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1089 :
1090 0 : if (p == slab_caches.next)
1091 0 : print_slabinfo_header(m);
1092 0 : cache_show(s, m);
1093 0 : return 0;
1094 : }
1095 :
1096 0 : void dump_unreclaimable_slab(void)
1097 : {
1098 : struct kmem_cache *s;
1099 : struct slabinfo sinfo;
1100 :
1101 : /*
1102 : * Here acquiring slab_mutex is risky since we don't prefer to get
1103 : * sleep in oom path. But, without mutex hold, it may introduce a
1104 : * risk of crash.
1105 : * Use mutex_trylock to protect the list traverse, dump nothing
1106 : * without acquiring the mutex.
1107 : */
1108 0 : if (!mutex_trylock(&slab_mutex)) {
1109 0 : pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1110 0 : return;
1111 : }
1112 :
1113 0 : pr_info("Unreclaimable slab info:\n");
1114 0 : pr_info("Name Used Total\n");
1115 :
1116 0 : list_for_each_entry(s, &slab_caches, list) {
1117 0 : if (s->flags & SLAB_RECLAIM_ACCOUNT)
1118 0 : continue;
1119 :
1120 0 : get_slabinfo(s, &sinfo);
1121 :
1122 0 : if (sinfo.num_objs > 0)
1123 0 : pr_info("%-17s %10luKB %10luKB\n", s->name,
1124 : (sinfo.active_objs * s->size) / 1024,
1125 : (sinfo.num_objs * s->size) / 1024);
1126 : }
1127 0 : mutex_unlock(&slab_mutex);
1128 : }
1129 :
1130 : /*
1131 : * slabinfo_op - iterator that generates /proc/slabinfo
1132 : *
1133 : * Output layout:
1134 : * cache-name
1135 : * num-active-objs
1136 : * total-objs
1137 : * object size
1138 : * num-active-slabs
1139 : * total-slabs
1140 : * num-pages-per-slab
1141 : * + further values on SMP and with statistics enabled
1142 : */
1143 : static const struct seq_operations slabinfo_op = {
1144 : .start = slab_start,
1145 : .next = slab_next,
1146 : .stop = slab_stop,
1147 : .show = slab_show,
1148 : };
1149 :
1150 0 : static int slabinfo_open(struct inode *inode, struct file *file)
1151 : {
1152 0 : return seq_open(file, &slabinfo_op);
1153 : }
1154 :
1155 : static const struct proc_ops slabinfo_proc_ops = {
1156 : .proc_flags = PROC_ENTRY_PERMANENT,
1157 : .proc_open = slabinfo_open,
1158 : .proc_read = seq_read,
1159 : .proc_write = slabinfo_write,
1160 : .proc_lseek = seq_lseek,
1161 : .proc_release = seq_release,
1162 : };
1163 :
1164 1 : static int __init slab_proc_init(void)
1165 : {
1166 1 : proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1167 1 : return 0;
1168 : }
1169 : module_init(slab_proc_init);
1170 :
1171 : #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1172 :
1173 : static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1174 : gfp_t flags)
1175 : {
1176 : void *ret;
1177 : size_t ks;
1178 :
1179 : /* Don't use instrumented ksize to allow precise KASAN poisoning. */
1180 228 : if (likely(!ZERO_OR_NULL_PTR(p))) {
1181 228 : if (!kasan_check_byte(p))
1182 : return NULL;
1183 228 : ks = kfence_ksize(p) ?: __ksize(p);
1184 : } else
1185 : ks = 0;
1186 :
1187 : /* If the object still fits, repoison it precisely. */
1188 228 : if (ks >= new_size) {
1189 : p = kasan_krealloc((void *)p, new_size, flags);
1190 : return (void *)p;
1191 : }
1192 :
1193 93 : ret = kmalloc_track_caller(new_size, flags);
1194 93 : if (ret && p) {
1195 : /* Disable KASAN checks as the object's redzone is accessed. */
1196 : kasan_disable_current();
1197 93 : memcpy(ret, kasan_reset_tag(p), ks);
1198 : kasan_enable_current();
1199 : }
1200 :
1201 : return ret;
1202 : }
1203 :
1204 : /**
1205 : * krealloc - reallocate memory. The contents will remain unchanged.
1206 : * @p: object to reallocate memory for.
1207 : * @new_size: how many bytes of memory are required.
1208 : * @flags: the type of memory to allocate.
1209 : *
1210 : * The contents of the object pointed to are preserved up to the
1211 : * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1212 : * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1213 : * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1214 : *
1215 : * Return: pointer to the allocated memory or %NULL in case of error
1216 : */
1217 228 : void *krealloc(const void *p, size_t new_size, gfp_t flags)
1218 : {
1219 : void *ret;
1220 :
1221 228 : if (unlikely(!new_size)) {
1222 0 : kfree(p);
1223 0 : return ZERO_SIZE_PTR;
1224 : }
1225 :
1226 228 : ret = __do_krealloc(p, new_size, flags);
1227 228 : if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1228 93 : kfree(p);
1229 :
1230 : return ret;
1231 : }
1232 : EXPORT_SYMBOL(krealloc);
1233 :
1234 : /**
1235 : * kfree_sensitive - Clear sensitive information in memory before freeing
1236 : * @p: object to free memory of
1237 : *
1238 : * The memory of the object @p points to is zeroed before freed.
1239 : * If @p is %NULL, kfree_sensitive() does nothing.
1240 : *
1241 : * Note: this function zeroes the whole allocated buffer which can be a good
1242 : * deal bigger than the requested buffer size passed to kmalloc(). So be
1243 : * careful when using this function in performance sensitive code.
1244 : */
1245 0 : void kfree_sensitive(const void *p)
1246 : {
1247 : size_t ks;
1248 0 : void *mem = (void *)p;
1249 :
1250 0 : ks = ksize(mem);
1251 0 : if (ks)
1252 : memzero_explicit(mem, ks);
1253 0 : kfree(mem);
1254 0 : }
1255 : EXPORT_SYMBOL(kfree_sensitive);
1256 :
1257 : /**
1258 : * ksize - get the actual amount of memory allocated for a given object
1259 : * @objp: Pointer to the object
1260 : *
1261 : * kmalloc may internally round up allocations and return more memory
1262 : * than requested. ksize() can be used to determine the actual amount of
1263 : * memory allocated. The caller may use this additional memory, even though
1264 : * a smaller amount of memory was initially specified with the kmalloc call.
1265 : * The caller must guarantee that objp points to a valid object previously
1266 : * allocated with either kmalloc() or kmem_cache_alloc(). The object
1267 : * must not be freed during the duration of the call.
1268 : *
1269 : * Return: size of the actual memory used by @objp in bytes
1270 : */
1271 0 : size_t ksize(const void *objp)
1272 : {
1273 : size_t size;
1274 :
1275 : /*
1276 : * We need to first check that the pointer to the object is valid, and
1277 : * only then unpoison the memory. The report printed from ksize() is
1278 : * more useful, then when it's printed later when the behaviour could
1279 : * be undefined due to a potential use-after-free or double-free.
1280 : *
1281 : * We use kasan_check_byte(), which is supported for the hardware
1282 : * tag-based KASAN mode, unlike kasan_check_read/write().
1283 : *
1284 : * If the pointed to memory is invalid, we return 0 to avoid users of
1285 : * ksize() writing to and potentially corrupting the memory region.
1286 : *
1287 : * We want to perform the check before __ksize(), to avoid potentially
1288 : * crashing in __ksize() due to accessing invalid metadata.
1289 : */
1290 0 : if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1291 : return 0;
1292 :
1293 0 : size = kfence_ksize(objp) ?: __ksize(objp);
1294 : /*
1295 : * We assume that ksize callers could use whole allocated area,
1296 : * so we need to unpoison this area.
1297 : */
1298 0 : kasan_unpoison_range(objp, size);
1299 0 : return size;
1300 : }
1301 : EXPORT_SYMBOL(ksize);
1302 :
1303 : /* Tracepoints definitions. */
1304 : EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1305 : EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1306 : EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1307 : EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1308 : EXPORT_TRACEPOINT_SYMBOL(kfree);
1309 : EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1310 :
1311 18327 : int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1312 : {
1313 18327 : if (__should_failslab(s, gfpflags))
1314 : return -ENOMEM;
1315 : return 0;
1316 : }
1317 : ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
|
