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/dma-mapping.h>
21 #include <linux/swiotlb.h>
22 #include <linux/proc_fs.h>
23 #include <linux/debugfs.h>
24 #include <linux/kasan.h>
25 #include <asm/cacheflush.h>
26 #include <asm/tlbflush.h>
27 #include <asm/page.h>
28 #include <linux/memcontrol.h>
29 #include <linux/stackdepot.h>
30
31 #include "internal.h"
32 #include "slab.h"
33
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/kmem.h>
36 #undef CREATE_TRACE_POINTS
37 #include <trace/hooks/mm.h>
38
39 enum slab_state slab_state;
40 LIST_HEAD(slab_caches);
41 DEFINE_MUTEX(slab_mutex);
42 struct kmem_cache *kmem_cache;
43
44 static LIST_HEAD(slab_caches_to_rcu_destroy);
45 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
46 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
47 slab_caches_to_rcu_destroy_workfn);
48
49 /*
50 * Set of flags that will prevent slab merging
51 */
52 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
53 SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
54 SLAB_FAILSLAB | SLAB_NO_MERGE | kasan_never_merge())
55
56 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
57 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
58
59 /*
60 * Merge control. If this is set then no merging of slab caches will occur.
61 */
62 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
63
setup_slab_nomerge(char * str)64 static int __init setup_slab_nomerge(char *str)
65 {
66 slab_nomerge = true;
67 return 1;
68 }
69
setup_slab_merge(char * str)70 static int __init setup_slab_merge(char *str)
71 {
72 slab_nomerge = false;
73 return 1;
74 }
75
76 #ifdef CONFIG_SLUB
77 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
78 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
79 #endif
80
81 __setup("slab_nomerge", setup_slab_nomerge);
82 __setup("slab_merge", setup_slab_merge);
83
84 /*
85 * Determine the size of a slab object
86 */
kmem_cache_size(struct kmem_cache * s)87 unsigned int kmem_cache_size(struct kmem_cache *s)
88 {
89 return s->object_size;
90 }
91 EXPORT_SYMBOL(kmem_cache_size);
92
93 #ifdef CONFIG_DEBUG_VM
kmem_cache_sanity_check(const char * name,unsigned int size)94 static int kmem_cache_sanity_check(const char *name, unsigned int size)
95 {
96 if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
97 pr_err("kmem_cache_create(%s) integrity check failed\n", name);
98 return -EINVAL;
99 }
100
101 WARN_ON(strchr(name, ' ')); /* It confuses parsers */
102 return 0;
103 }
104 #else
kmem_cache_sanity_check(const char * name,unsigned int size)105 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
106 {
107 return 0;
108 }
109 #endif
110
111 /*
112 * Figure out what the alignment of the objects will be given a set of
113 * flags, a user specified alignment and the size of the objects.
114 */
calculate_alignment(slab_flags_t flags,unsigned int align,unsigned int size)115 static unsigned int calculate_alignment(slab_flags_t flags,
116 unsigned int align, unsigned int size)
117 {
118 /*
119 * If the user wants hardware cache aligned objects then follow that
120 * suggestion if the object is sufficiently large.
121 *
122 * The hardware cache alignment cannot override the specified
123 * alignment though. If that is greater then use it.
124 */
125 if (flags & SLAB_HWCACHE_ALIGN) {
126 unsigned int ralign;
127
128 ralign = cache_line_size();
129 while (size <= ralign / 2)
130 ralign /= 2;
131 align = max(align, ralign);
132 }
133
134 align = max(align, arch_slab_minalign());
135
136 return ALIGN(align, sizeof(void *));
137 }
138
139 /*
140 * Find a mergeable slab cache
141 */
slab_unmergeable(struct kmem_cache * s)142 int slab_unmergeable(struct kmem_cache *s)
143 {
144 if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
145 return 1;
146
147 if (s->ctor)
148 return 1;
149
150 #ifdef CONFIG_HARDENED_USERCOPY
151 if (s->usersize)
152 return 1;
153 #endif
154
155 /*
156 * We may have set a slab to be unmergeable during bootstrap.
157 */
158 if (s->refcount < 0)
159 return 1;
160
161 return 0;
162 }
163
find_mergeable(unsigned int size,unsigned int align,slab_flags_t flags,const char * name,void (* ctor)(void *))164 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
165 slab_flags_t flags, const char *name, void (*ctor)(void *))
166 {
167 struct kmem_cache *s;
168
169 if (slab_nomerge)
170 return NULL;
171
172 if (ctor)
173 return NULL;
174
175 size = ALIGN(size, sizeof(void *));
176 align = calculate_alignment(flags, align, size);
177 size = ALIGN(size, align);
178 flags = kmem_cache_flags(size, flags, name);
179
180 if (flags & SLAB_NEVER_MERGE)
181 return NULL;
182
183 list_for_each_entry_reverse(s, &slab_caches, list) {
184 if (slab_unmergeable(s))
185 continue;
186
187 if (size > s->size)
188 continue;
189
190 if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
191 continue;
192 /*
193 * Check if alignment is compatible.
194 * Courtesy of Adrian Drzewiecki
195 */
196 if ((s->size & ~(align - 1)) != s->size)
197 continue;
198
199 if (s->size - size >= sizeof(void *))
200 continue;
201
202 if (IS_ENABLED(CONFIG_SLAB) && align &&
203 (align > s->align || s->align % align))
204 continue;
205
206 return s;
207 }
208 return NULL;
209 }
210
create_cache(const char * name,unsigned int object_size,unsigned int align,slab_flags_t flags,unsigned int useroffset,unsigned int usersize,void (* ctor)(void *),struct kmem_cache * root_cache)211 static struct kmem_cache *create_cache(const char *name,
212 unsigned int object_size, unsigned int align,
213 slab_flags_t flags, unsigned int useroffset,
214 unsigned int usersize, void (*ctor)(void *),
215 struct kmem_cache *root_cache)
216 {
217 struct kmem_cache *s;
218 int err;
219
220 if (WARN_ON(useroffset + usersize > object_size))
221 useroffset = usersize = 0;
222
223 err = -ENOMEM;
224 s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
225 if (!s)
226 goto out;
227
228 s->name = name;
229 s->size = s->object_size = object_size;
230 s->align = align;
231 s->ctor = ctor;
232 #ifdef CONFIG_HARDENED_USERCOPY
233 s->useroffset = useroffset;
234 s->usersize = usersize;
235 #endif
236
237 err = __kmem_cache_create(s, flags);
238 if (err)
239 goto out_free_cache;
240
241 s->refcount = 1;
242 list_add(&s->list, &slab_caches);
243 return s;
244
245 out_free_cache:
246 kmem_cache_free(kmem_cache, s);
247 out:
248 return ERR_PTR(err);
249 }
250
251 /**
252 * kmem_cache_create_usercopy - Create a cache with a region suitable
253 * for copying to userspace
254 * @name: A string which is used in /proc/slabinfo to identify this cache.
255 * @size: The size of objects to be created in this cache.
256 * @align: The required alignment for the objects.
257 * @flags: SLAB flags
258 * @useroffset: Usercopy region offset
259 * @usersize: Usercopy region size
260 * @ctor: A constructor for the objects.
261 *
262 * Cannot be called within a interrupt, but can be interrupted.
263 * The @ctor is run when new pages are allocated by the cache.
264 *
265 * The flags are
266 *
267 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
268 * to catch references to uninitialised memory.
269 *
270 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
271 * for buffer overruns.
272 *
273 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
274 * cacheline. This can be beneficial if you're counting cycles as closely
275 * as davem.
276 *
277 * Return: a pointer to the cache on success, NULL on failure.
278 */
279 struct kmem_cache *
kmem_cache_create_usercopy(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,unsigned int useroffset,unsigned int usersize,void (* ctor)(void *))280 kmem_cache_create_usercopy(const char *name,
281 unsigned int size, unsigned int align,
282 slab_flags_t flags,
283 unsigned int useroffset, unsigned int usersize,
284 void (*ctor)(void *))
285 {
286 struct kmem_cache *s = NULL;
287 const char *cache_name;
288 int err;
289
290 #ifdef CONFIG_SLUB_DEBUG
291 /*
292 * If no slub_debug was enabled globally, the static key is not yet
293 * enabled by setup_slub_debug(). Enable it if the cache is being
294 * created with any of the debugging flags passed explicitly.
295 * It's also possible that this is the first cache created with
296 * SLAB_STORE_USER and we should init stack_depot for it.
297 */
298 if (flags & SLAB_DEBUG_FLAGS)
299 static_branch_enable(&slub_debug_enabled);
300 if (flags & SLAB_STORE_USER)
301 stack_depot_init();
302 #endif
303
304 mutex_lock(&slab_mutex);
305
306 err = kmem_cache_sanity_check(name, size);
307 if (err) {
308 goto out_unlock;
309 }
310
311 /* Refuse requests with allocator specific flags */
312 if (flags & ~SLAB_FLAGS_PERMITTED) {
313 err = -EINVAL;
314 goto out_unlock;
315 }
316
317 /*
318 * Some allocators will constraint the set of valid flags to a subset
319 * of all flags. We expect them to define CACHE_CREATE_MASK in this
320 * case, and we'll just provide them with a sanitized version of the
321 * passed flags.
322 */
323 flags &= CACHE_CREATE_MASK;
324
325 /* Fail closed on bad usersize of useroffset values. */
326 if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
327 WARN_ON(!usersize && useroffset) ||
328 WARN_ON(size < usersize || size - usersize < useroffset))
329 usersize = useroffset = 0;
330
331 if (!usersize)
332 s = __kmem_cache_alias(name, size, align, flags, ctor);
333 if (s)
334 goto out_unlock;
335
336 cache_name = kstrdup_const(name, GFP_KERNEL);
337 if (!cache_name) {
338 err = -ENOMEM;
339 goto out_unlock;
340 }
341
342 s = create_cache(cache_name, size,
343 calculate_alignment(flags, align, size),
344 flags, useroffset, usersize, ctor, NULL);
345 if (IS_ERR(s)) {
346 err = PTR_ERR(s);
347 kfree_const(cache_name);
348 }
349
350 out_unlock:
351 mutex_unlock(&slab_mutex);
352
353 if (err) {
354 if (flags & SLAB_PANIC)
355 panic("%s: Failed to create slab '%s'. Error %d\n",
356 __func__, name, err);
357 else {
358 pr_warn("%s(%s) failed with error %d\n",
359 __func__, name, err);
360 dump_stack();
361 }
362 return NULL;
363 }
364 return s;
365 }
366 EXPORT_SYMBOL(kmem_cache_create_usercopy);
367
368 /**
369 * kmem_cache_create - Create a cache.
370 * @name: A string which is used in /proc/slabinfo to identify this cache.
371 * @size: The size of objects to be created in this cache.
372 * @align: The required alignment for the objects.
373 * @flags: SLAB flags
374 * @ctor: A constructor for the objects.
375 *
376 * Cannot be called within a interrupt, but can be interrupted.
377 * The @ctor is run when new pages are allocated by the cache.
378 *
379 * The flags are
380 *
381 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
382 * to catch references to uninitialised memory.
383 *
384 * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
385 * for buffer overruns.
386 *
387 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
388 * cacheline. This can be beneficial if you're counting cycles as closely
389 * as davem.
390 *
391 * Return: a pointer to the cache on success, NULL on failure.
392 */
393 struct kmem_cache *
kmem_cache_create(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))394 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
395 slab_flags_t flags, void (*ctor)(void *))
396 {
397 return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
398 ctor);
399 }
400 EXPORT_SYMBOL(kmem_cache_create);
401
402 #ifdef SLAB_SUPPORTS_SYSFS
403 /*
404 * For a given kmem_cache, kmem_cache_destroy() should only be called
405 * once or there will be a use-after-free problem. The actual deletion
406 * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
407 * protection. So they are now done without holding those locks.
408 *
409 * Note that there will be a slight delay in the deletion of sysfs files
410 * if kmem_cache_release() is called indrectly from a work function.
411 */
kmem_cache_release(struct kmem_cache * s)412 static void kmem_cache_release(struct kmem_cache *s)
413 {
414 sysfs_slab_unlink(s);
415 sysfs_slab_release(s);
416 }
417 #else
kmem_cache_release(struct kmem_cache * s)418 static void kmem_cache_release(struct kmem_cache *s)
419 {
420 slab_kmem_cache_release(s);
421 }
422 #endif
423
slab_caches_to_rcu_destroy_workfn(struct work_struct * work)424 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
425 {
426 LIST_HEAD(to_destroy);
427 struct kmem_cache *s, *s2;
428
429 /*
430 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
431 * @slab_caches_to_rcu_destroy list. The slab pages are freed
432 * through RCU and the associated kmem_cache are dereferenced
433 * while freeing the pages, so the kmem_caches should be freed only
434 * after the pending RCU operations are finished. As rcu_barrier()
435 * is a pretty slow operation, we batch all pending destructions
436 * asynchronously.
437 */
438 mutex_lock(&slab_mutex);
439 list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
440 mutex_unlock(&slab_mutex);
441
442 if (list_empty(&to_destroy))
443 return;
444
445 rcu_barrier();
446
447 list_for_each_entry_safe(s, s2, &to_destroy, list) {
448 debugfs_slab_release(s);
449 kfence_shutdown_cache(s);
450 kmem_cache_release(s);
451 }
452 }
453
shutdown_cache(struct kmem_cache * s)454 static int shutdown_cache(struct kmem_cache *s)
455 {
456 /* free asan quarantined objects */
457 kasan_cache_shutdown(s);
458
459 if (__kmem_cache_shutdown(s) != 0)
460 return -EBUSY;
461
462 list_del(&s->list);
463
464 if (s->flags & SLAB_TYPESAFE_BY_RCU) {
465 list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
466 schedule_work(&slab_caches_to_rcu_destroy_work);
467 } else {
468 kfence_shutdown_cache(s);
469 debugfs_slab_release(s);
470 }
471
472 return 0;
473 }
474
slab_kmem_cache_release(struct kmem_cache * s)475 void slab_kmem_cache_release(struct kmem_cache *s)
476 {
477 __kmem_cache_release(s);
478 kfree_const(s->name);
479 kmem_cache_free(kmem_cache, s);
480 }
481
kmem_cache_destroy(struct kmem_cache * s)482 void kmem_cache_destroy(struct kmem_cache *s)
483 {
484 int err = -EBUSY;
485 bool rcu_set;
486
487 if (unlikely(!s) || !kasan_check_byte(s))
488 return;
489
490 cpus_read_lock();
491 mutex_lock(&slab_mutex);
492
493 rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
494
495 s->refcount--;
496 if (s->refcount)
497 goto out_unlock;
498
499 err = shutdown_cache(s);
500 WARN(err, "%s %s: Slab cache still has objects when called from %pS",
501 __func__, s->name, (void *)_RET_IP_);
502 out_unlock:
503 mutex_unlock(&slab_mutex);
504 cpus_read_unlock();
505 if (!err && !rcu_set)
506 kmem_cache_release(s);
507 }
508 EXPORT_SYMBOL(kmem_cache_destroy);
509
510 /**
511 * kmem_cache_shrink - Shrink a cache.
512 * @cachep: The cache to shrink.
513 *
514 * Releases as many slabs as possible for a cache.
515 * To help debugging, a zero exit status indicates all slabs were released.
516 *
517 * Return: %0 if all slabs were released, non-zero otherwise
518 */
kmem_cache_shrink(struct kmem_cache * cachep)519 int kmem_cache_shrink(struct kmem_cache *cachep)
520 {
521 kasan_cache_shrink(cachep);
522
523 return __kmem_cache_shrink(cachep);
524 }
525 EXPORT_SYMBOL(kmem_cache_shrink);
526
slab_is_available(void)527 bool slab_is_available(void)
528 {
529 return slab_state >= UP;
530 }
531
532 #ifdef CONFIG_PRINTK
kmem_obj_info(struct kmem_obj_info * kpp,void * object,struct slab * slab)533 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
534 {
535 if (__kfence_obj_info(kpp, object, slab))
536 return;
537 __kmem_obj_info(kpp, object, slab);
538 }
539
540 /**
541 * kmem_dump_obj - Print available slab provenance information
542 * @object: slab object for which to find provenance information.
543 *
544 * This function uses pr_cont(), so that the caller is expected to have
545 * printed out whatever preamble is appropriate. The provenance information
546 * depends on the type of object and on how much debugging is enabled.
547 * For a slab-cache object, the fact that it is a slab object is printed,
548 * and, if available, the slab name, return address, and stack trace from
549 * the allocation and last free path of that object.
550 *
551 * Return: %true if the pointer is to a not-yet-freed object from
552 * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
553 * is to an already-freed object, and %false otherwise.
554 */
kmem_dump_obj(void * object)555 bool kmem_dump_obj(void *object)
556 {
557 char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
558 int i;
559 struct slab *slab;
560 unsigned long ptroffset;
561 struct kmem_obj_info kp = { };
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 slab = virt_to_slab(object);
567 if (!slab)
568 return false;
569
570 kmem_obj_info(&kp, object, slab);
571 if (kp.kp_slab_cache)
572 pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
573 else
574 pr_cont(" slab%s", cp);
575 if (is_kfence_address(object))
576 pr_cont(" (kfence)");
577 if (kp.kp_objp)
578 pr_cont(" start %px", kp.kp_objp);
579 if (kp.kp_data_offset)
580 pr_cont(" data offset %lu", kp.kp_data_offset);
581 if (kp.kp_objp) {
582 ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
583 pr_cont(" pointer offset %lu", ptroffset);
584 }
585 if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
586 pr_cont(" size %u", kp.kp_slab_cache->object_size);
587 if (kp.kp_ret)
588 pr_cont(" allocated at %pS\n", kp.kp_ret);
589 else
590 pr_cont("\n");
591 for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
592 if (!kp.kp_stack[i])
593 break;
594 pr_info(" %pS\n", kp.kp_stack[i]);
595 }
596
597 if (kp.kp_free_stack[0])
598 pr_cont(" Free path:\n");
599
600 for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
601 if (!kp.kp_free_stack[i])
602 break;
603 pr_info(" %pS\n", kp.kp_free_stack[i]);
604 }
605
606 return true;
607 }
608 EXPORT_SYMBOL_GPL(kmem_dump_obj);
609 #endif
610
611 /* Create a cache during boot when no slab services are available yet */
create_boot_cache(struct kmem_cache * s,const char * name,unsigned int size,slab_flags_t flags,unsigned int useroffset,unsigned int usersize)612 void __init create_boot_cache(struct kmem_cache *s, const char *name,
613 unsigned int size, slab_flags_t flags,
614 unsigned int useroffset, unsigned int usersize)
615 {
616 int err;
617 unsigned int align = ARCH_KMALLOC_MINALIGN;
618
619 s->name = name;
620 s->size = s->object_size = size;
621
622 /*
623 * For power of two sizes, guarantee natural alignment for kmalloc
624 * caches, regardless of SL*B debugging options.
625 */
626 if (is_power_of_2(size))
627 align = max(align, size);
628 s->align = calculate_alignment(flags, align, size);
629
630 #ifdef CONFIG_HARDENED_USERCOPY
631 s->useroffset = useroffset;
632 s->usersize = usersize;
633 #endif
634
635 err = __kmem_cache_create(s, flags);
636
637 if (err)
638 panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
639 name, size, err);
640
641 s->refcount = -1; /* Exempt from merging for now */
642 }
643
create_kmalloc_cache(const char * name,unsigned int size,slab_flags_t flags)644 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
645 unsigned int size,
646 slab_flags_t flags)
647 {
648 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
649
650 if (!s)
651 panic("Out of memory when creating slab %s\n", name);
652
653 create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
654 list_add(&s->list, &slab_caches);
655 s->refcount = 1;
656 return s;
657 }
658
659 struct kmem_cache *
660 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
661 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
662 EXPORT_SYMBOL(kmalloc_caches);
663
664 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
665 unsigned long random_kmalloc_seed __ro_after_init;
666 EXPORT_SYMBOL(random_kmalloc_seed);
667 #endif
668
669 /*
670 * Conversion table for small slabs sizes / 8 to the index in the
671 * kmalloc array. This is necessary for slabs < 192 since we have non power
672 * of two cache sizes there. The size of larger slabs can be determined using
673 * fls.
674 */
675 static u8 size_index[24] __ro_after_init = {
676 3, /* 8 */
677 4, /* 16 */
678 5, /* 24 */
679 5, /* 32 */
680 6, /* 40 */
681 6, /* 48 */
682 6, /* 56 */
683 6, /* 64 */
684 1, /* 72 */
685 1, /* 80 */
686 1, /* 88 */
687 1, /* 96 */
688 7, /* 104 */
689 7, /* 112 */
690 7, /* 120 */
691 7, /* 128 */
692 2, /* 136 */
693 2, /* 144 */
694 2, /* 152 */
695 2, /* 160 */
696 2, /* 168 */
697 2, /* 176 */
698 2, /* 184 */
699 2 /* 192 */
700 };
701
size_index_elem(unsigned int bytes)702 static inline unsigned int size_index_elem(unsigned int bytes)
703 {
704 return (bytes - 1) / 8;
705 }
706
707 /*
708 * Find the kmem_cache structure that serves a given size of
709 * allocation
710 */
kmalloc_slab(size_t size,gfp_t flags,unsigned long caller)711 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags, unsigned long caller)
712 {
713 unsigned int index;
714 struct kmem_cache *s = NULL;
715
716 if (size <= 192) {
717 if (!size)
718 return ZERO_SIZE_PTR;
719
720 index = size_index[size_index_elem(size)];
721 } else {
722 if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
723 return NULL;
724 index = fls(size - 1);
725 }
726
727 trace_android_vh_kmalloc_slab(index, flags, &s);
728 if (s)
729 return s;
730
731 return kmalloc_caches[kmalloc_type(flags, caller)][index];
732 }
733
kmalloc_size_roundup(size_t size)734 size_t kmalloc_size_roundup(size_t size)
735 {
736 if (size && size <= KMALLOC_MAX_CACHE_SIZE) {
737 /*
738 * The flags don't matter since size_index is common to all.
739 * Neither does the caller for just getting ->object_size.
740 */
741 return kmalloc_slab(size, GFP_KERNEL, 0)->object_size;
742 }
743
744 /* Above the smaller buckets, size is a multiple of page size. */
745 if (size && size <= KMALLOC_MAX_SIZE)
746 return PAGE_SIZE << get_order(size);
747
748 /*
749 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR
750 * and very large size - kmalloc() may fail.
751 */
752 return size;
753
754 }
755 EXPORT_SYMBOL(kmalloc_size_roundup);
756
757 #ifdef CONFIG_ZONE_DMA
758 #define KMALLOC_DMA_NAME(sz) .name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
759 #else
760 #define KMALLOC_DMA_NAME(sz)
761 #endif
762
763 #ifdef CONFIG_MEMCG_KMEM
764 #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
765 #else
766 #define KMALLOC_CGROUP_NAME(sz)
767 #endif
768
769 #ifndef CONFIG_SLUB_TINY
770 #define KMALLOC_RCL_NAME(sz) .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
771 #else
772 #define KMALLOC_RCL_NAME(sz)
773 #endif
774
775 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
776 #define __KMALLOC_RANDOM_CONCAT(a, b) a ## b
777 #define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz)
778 #define KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 1] = "kmalloc-rnd-01-" #sz,
779 #define KMA_RAND_2(sz) KMA_RAND_1(sz) .name[KMALLOC_RANDOM_START + 2] = "kmalloc-rnd-02-" #sz,
780 #define KMA_RAND_3(sz) KMA_RAND_2(sz) .name[KMALLOC_RANDOM_START + 3] = "kmalloc-rnd-03-" #sz,
781 #define KMA_RAND_4(sz) KMA_RAND_3(sz) .name[KMALLOC_RANDOM_START + 4] = "kmalloc-rnd-04-" #sz,
782 #define KMA_RAND_5(sz) KMA_RAND_4(sz) .name[KMALLOC_RANDOM_START + 5] = "kmalloc-rnd-05-" #sz,
783 #define KMA_RAND_6(sz) KMA_RAND_5(sz) .name[KMALLOC_RANDOM_START + 6] = "kmalloc-rnd-06-" #sz,
784 #define KMA_RAND_7(sz) KMA_RAND_6(sz) .name[KMALLOC_RANDOM_START + 7] = "kmalloc-rnd-07-" #sz,
785 #define KMA_RAND_8(sz) KMA_RAND_7(sz) .name[KMALLOC_RANDOM_START + 8] = "kmalloc-rnd-08-" #sz,
786 #define KMA_RAND_9(sz) KMA_RAND_8(sz) .name[KMALLOC_RANDOM_START + 9] = "kmalloc-rnd-09-" #sz,
787 #define KMA_RAND_10(sz) KMA_RAND_9(sz) .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz,
788 #define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz,
789 #define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz,
790 #define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz,
791 #define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz,
792 #define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz,
793 #else // CONFIG_RANDOM_KMALLOC_CACHES
794 #define KMALLOC_RANDOM_NAME(N, sz)
795 #endif
796
797 #define INIT_KMALLOC_INFO(__size, __short_size) \
798 { \
799 .name[KMALLOC_NORMAL] = "kmalloc-" #__short_size, \
800 KMALLOC_RCL_NAME(__short_size) \
801 KMALLOC_CGROUP_NAME(__short_size) \
802 KMALLOC_DMA_NAME(__short_size) \
803 KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size) \
804 .size = __size, \
805 }
806
807 /*
808 * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
809 * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
810 * kmalloc-2M.
811 */
812 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
813 INIT_KMALLOC_INFO(0, 0),
814 INIT_KMALLOC_INFO(96, 96),
815 INIT_KMALLOC_INFO(192, 192),
816 INIT_KMALLOC_INFO(8, 8),
817 INIT_KMALLOC_INFO(16, 16),
818 INIT_KMALLOC_INFO(32, 32),
819 INIT_KMALLOC_INFO(64, 64),
820 INIT_KMALLOC_INFO(128, 128),
821 INIT_KMALLOC_INFO(256, 256),
822 INIT_KMALLOC_INFO(512, 512),
823 INIT_KMALLOC_INFO(1024, 1k),
824 INIT_KMALLOC_INFO(2048, 2k),
825 INIT_KMALLOC_INFO(4096, 4k),
826 INIT_KMALLOC_INFO(8192, 8k),
827 INIT_KMALLOC_INFO(16384, 16k),
828 INIT_KMALLOC_INFO(32768, 32k),
829 INIT_KMALLOC_INFO(65536, 64k),
830 INIT_KMALLOC_INFO(131072, 128k),
831 INIT_KMALLOC_INFO(262144, 256k),
832 INIT_KMALLOC_INFO(524288, 512k),
833 INIT_KMALLOC_INFO(1048576, 1M),
834 INIT_KMALLOC_INFO(2097152, 2M)
835 };
836
837 /*
838 * Patch up the size_index table if we have strange large alignment
839 * requirements for the kmalloc array. This is only the case for
840 * MIPS it seems. The standard arches will not generate any code here.
841 *
842 * Largest permitted alignment is 256 bytes due to the way we
843 * handle the index determination for the smaller caches.
844 *
845 * Make sure that nothing crazy happens if someone starts tinkering
846 * around with ARCH_KMALLOC_MINALIGN
847 */
setup_kmalloc_cache_index_table(void)848 void __init setup_kmalloc_cache_index_table(void)
849 {
850 unsigned int i;
851
852 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
853 !is_power_of_2(KMALLOC_MIN_SIZE));
854
855 for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
856 unsigned int elem = size_index_elem(i);
857
858 if (elem >= ARRAY_SIZE(size_index))
859 break;
860 size_index[elem] = KMALLOC_SHIFT_LOW;
861 }
862
863 if (KMALLOC_MIN_SIZE >= 64) {
864 /*
865 * The 96 byte sized cache is not used if the alignment
866 * is 64 byte.
867 */
868 for (i = 64 + 8; i <= 96; i += 8)
869 size_index[size_index_elem(i)] = 7;
870
871 }
872
873 if (KMALLOC_MIN_SIZE >= 128) {
874 /*
875 * The 192 byte sized cache is not used if the alignment
876 * is 128 byte. Redirect kmalloc to use the 256 byte cache
877 * instead.
878 */
879 for (i = 128 + 8; i <= 192; i += 8)
880 size_index[size_index_elem(i)] = 8;
881 }
882 }
883
__kmalloc_minalign(void)884 static unsigned int __kmalloc_minalign(void)
885 {
886 unsigned int minalign = dma_get_cache_alignment();
887
888 if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) &&
889 is_swiotlb_allocated())
890 minalign = ARCH_KMALLOC_MINALIGN;
891
892 return max(minalign, arch_slab_minalign());
893 }
894
895 void __init
new_kmalloc_cache(int idx,enum kmalloc_cache_type type,slab_flags_t flags)896 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
897 {
898 unsigned int minalign = __kmalloc_minalign();
899 unsigned int aligned_size = kmalloc_info[idx].size;
900 int aligned_idx = idx;
901
902 if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
903 flags |= SLAB_RECLAIM_ACCOUNT;
904 } else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
905 if (mem_cgroup_kmem_disabled()) {
906 kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
907 return;
908 }
909 flags |= SLAB_ACCOUNT;
910 } else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
911 flags |= SLAB_CACHE_DMA;
912 }
913
914 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
915 if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END)
916 flags |= SLAB_NO_MERGE;
917 #endif
918
919 /*
920 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
921 * KMALLOC_NORMAL caches.
922 */
923 if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
924 flags |= SLAB_NO_MERGE;
925
926 if (minalign > ARCH_KMALLOC_MINALIGN) {
927 aligned_size = ALIGN(aligned_size, minalign);
928 aligned_idx = __kmalloc_index(aligned_size, false);
929 }
930
931 if (!kmalloc_caches[type][aligned_idx])
932 kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
933 kmalloc_info[aligned_idx].name[type],
934 aligned_size, flags);
935 if (idx != aligned_idx)
936 kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
937 }
938
939 /*
940 * Create the kmalloc array. Some of the regular kmalloc arrays
941 * may already have been created because they were needed to
942 * enable allocations for slab creation.
943 */
create_kmalloc_caches(slab_flags_t flags)944 void __init create_kmalloc_caches(slab_flags_t flags)
945 {
946 int i;
947 enum kmalloc_cache_type type;
948
949 /*
950 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
951 */
952 for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
953 for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
954 if (!kmalloc_caches[type][i])
955 new_kmalloc_cache(i, type, flags);
956
957 /*
958 * Caches that are not of the two-to-the-power-of size.
959 * These have to be created immediately after the
960 * earlier power of two caches
961 */
962 if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
963 !kmalloc_caches[type][1])
964 new_kmalloc_cache(1, type, flags);
965 if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
966 !kmalloc_caches[type][2])
967 new_kmalloc_cache(2, type, flags);
968 }
969 }
970 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
971 random_kmalloc_seed = get_random_u64();
972 #endif
973
974 /* Kmalloc array is now usable */
975 slab_state = UP;
976 }
977
free_large_kmalloc(struct folio * folio,void * object)978 void free_large_kmalloc(struct folio *folio, void *object)
979 {
980 unsigned int order = folio_order(folio);
981
982 if (WARN_ON_ONCE(order == 0))
983 pr_warn_once("object pointer: 0x%p\n", object);
984
985 kmemleak_free(object);
986 kasan_kfree_large(object);
987 kmsan_kfree_large(object);
988
989 mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
990 -(PAGE_SIZE << order));
991 __free_pages(folio_page(folio, 0), order);
992 }
993
994 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
995 static __always_inline
__do_kmalloc_node(size_t size,gfp_t flags,int node,unsigned long caller)996 void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
997 {
998 struct kmem_cache *s;
999 void *ret;
1000
1001 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
1002 ret = __kmalloc_large_node(size, flags, node);
1003 trace_kmalloc(caller, ret, size,
1004 PAGE_SIZE << get_order(size), flags, node);
1005 return ret;
1006 }
1007
1008 s = kmalloc_slab(size, flags, caller);
1009
1010 if (unlikely(ZERO_OR_NULL_PTR(s)))
1011 return s;
1012
1013 ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
1014 ret = kasan_kmalloc(s, ret, size, flags);
1015 trace_kmalloc(caller, ret, size, s->size, flags, node);
1016 return ret;
1017 }
1018
__kmalloc_node(size_t size,gfp_t flags,int node)1019 void *__kmalloc_node(size_t size, gfp_t flags, int node)
1020 {
1021 return __do_kmalloc_node(size, flags, node, _RET_IP_);
1022 }
1023 EXPORT_SYMBOL(__kmalloc_node);
1024
__kmalloc(size_t size,gfp_t flags)1025 void *__kmalloc(size_t size, gfp_t flags)
1026 {
1027 return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
1028 }
1029 EXPORT_SYMBOL(__kmalloc);
1030
__kmalloc_node_track_caller(size_t size,gfp_t flags,int node,unsigned long caller)1031 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
1032 int node, unsigned long caller)
1033 {
1034 return __do_kmalloc_node(size, flags, node, caller);
1035 }
1036 EXPORT_SYMBOL(__kmalloc_node_track_caller);
1037
1038 /**
1039 * kfree - free previously allocated memory
1040 * @object: pointer returned by kmalloc() or kmem_cache_alloc()
1041 *
1042 * If @object is NULL, no operation is performed.
1043 */
kfree(const void * object)1044 void kfree(const void *object)
1045 {
1046 struct folio *folio;
1047 struct slab *slab;
1048 struct kmem_cache *s;
1049
1050 trace_kfree(_RET_IP_, object);
1051
1052 if (unlikely(ZERO_OR_NULL_PTR(object)))
1053 return;
1054
1055 folio = virt_to_folio(object);
1056 if (unlikely(!folio_test_slab(folio))) {
1057 free_large_kmalloc(folio, (void *)object);
1058 return;
1059 }
1060
1061 slab = folio_slab(folio);
1062 s = slab->slab_cache;
1063 __kmem_cache_free(s, (void *)object, _RET_IP_);
1064 }
1065 EXPORT_SYMBOL(kfree);
1066
1067 /**
1068 * __ksize -- Report full size of underlying allocation
1069 * @object: pointer to the object
1070 *
1071 * This should only be used internally to query the true size of allocations.
1072 * It is not meant to be a way to discover the usable size of an allocation
1073 * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1074 * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1075 * and/or FORTIFY_SOURCE.
1076 *
1077 * Return: size of the actual memory used by @object in bytes
1078 */
__ksize(const void * object)1079 size_t __ksize(const void *object)
1080 {
1081 struct folio *folio;
1082
1083 if (unlikely(object == ZERO_SIZE_PTR))
1084 return 0;
1085
1086 folio = virt_to_folio(object);
1087
1088 if (unlikely(!folio_test_slab(folio))) {
1089 if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1090 return 0;
1091 if (WARN_ON(object != folio_address(folio)))
1092 return 0;
1093 return folio_size(folio);
1094 }
1095
1096 #ifdef CONFIG_SLUB_DEBUG
1097 skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1098 #endif
1099
1100 return slab_ksize(folio_slab(folio)->slab_cache);
1101 }
1102
kmalloc_trace(struct kmem_cache * s,gfp_t gfpflags,size_t size)1103 void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1104 {
1105 void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1106 size, _RET_IP_);
1107
1108 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1109
1110 ret = kasan_kmalloc(s, ret, size, gfpflags);
1111 return ret;
1112 }
1113 EXPORT_SYMBOL(kmalloc_trace);
1114
kmalloc_node_trace(struct kmem_cache * s,gfp_t gfpflags,int node,size_t size)1115 void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1116 int node, size_t size)
1117 {
1118 void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1119
1120 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1121
1122 ret = kasan_kmalloc(s, ret, size, gfpflags);
1123 return ret;
1124 }
1125 EXPORT_SYMBOL(kmalloc_node_trace);
1126
kmalloc_fix_flags(gfp_t flags)1127 gfp_t kmalloc_fix_flags(gfp_t flags)
1128 {
1129 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1130
1131 flags &= ~GFP_SLAB_BUG_MASK;
1132 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1133 invalid_mask, &invalid_mask, flags, &flags);
1134 dump_stack();
1135
1136 return flags;
1137 }
1138
1139 /*
1140 * To avoid unnecessary overhead, we pass through large allocation requests
1141 * directly to the page allocator. We use __GFP_COMP, because we will need to
1142 * know the allocation order to free the pages properly in kfree.
1143 */
1144
__kmalloc_large_node(size_t size,gfp_t flags,int node)1145 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1146 {
1147 struct page *page;
1148 void *ptr = NULL;
1149 unsigned int order = get_order(size);
1150
1151 if (unlikely(flags & GFP_SLAB_BUG_MASK))
1152 flags = kmalloc_fix_flags(flags);
1153
1154 flags |= __GFP_COMP;
1155 page = alloc_pages_node(node, flags, order);
1156 if (page) {
1157 ptr = page_address(page);
1158 mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1159 PAGE_SIZE << order);
1160 }
1161
1162 trace_android_vh_kmalloc_large_alloced(page, order, flags);
1163
1164 ptr = kasan_kmalloc_large(ptr, size, flags);
1165 /* As ptr might get tagged, call kmemleak hook after KASAN. */
1166 kmemleak_alloc(ptr, size, 1, flags);
1167 kmsan_kmalloc_large(ptr, size, flags);
1168
1169 return ptr;
1170 }
1171
kmalloc_large(size_t size,gfp_t flags)1172 void *kmalloc_large(size_t size, gfp_t flags)
1173 {
1174 void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1175
1176 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1177 flags, NUMA_NO_NODE);
1178 return ret;
1179 }
1180 EXPORT_SYMBOL(kmalloc_large);
1181
kmalloc_large_node(size_t size,gfp_t flags,int node)1182 void *kmalloc_large_node(size_t size, gfp_t flags, int node)
1183 {
1184 void *ret = __kmalloc_large_node(size, flags, node);
1185
1186 trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1187 flags, node);
1188 return ret;
1189 }
1190 EXPORT_SYMBOL(kmalloc_large_node);
1191
1192 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1193 /* Randomize a generic freelist */
freelist_randomize(unsigned int * list,unsigned int count)1194 static void freelist_randomize(unsigned int *list,
1195 unsigned int count)
1196 {
1197 unsigned int rand;
1198 unsigned int i;
1199
1200 for (i = 0; i < count; i++)
1201 list[i] = i;
1202
1203 /* Fisher-Yates shuffle */
1204 for (i = count - 1; i > 0; i--) {
1205 rand = get_random_u32_below(i + 1);
1206 swap(list[i], list[rand]);
1207 }
1208 }
1209
1210 /* Create a random sequence per cache */
cache_random_seq_create(struct kmem_cache * cachep,unsigned int count,gfp_t gfp)1211 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1212 gfp_t gfp)
1213 {
1214
1215 if (count < 2 || cachep->random_seq)
1216 return 0;
1217
1218 cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1219 if (!cachep->random_seq)
1220 return -ENOMEM;
1221
1222 freelist_randomize(cachep->random_seq, count);
1223 return 0;
1224 }
1225
1226 /* Destroy the per-cache random freelist sequence */
cache_random_seq_destroy(struct kmem_cache * cachep)1227 void cache_random_seq_destroy(struct kmem_cache *cachep)
1228 {
1229 kfree(cachep->random_seq);
1230 cachep->random_seq = NULL;
1231 }
1232 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1233
1234 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1235 #ifdef CONFIG_SLAB
1236 #define SLABINFO_RIGHTS (0600)
1237 #else
1238 #define SLABINFO_RIGHTS (0400)
1239 #endif
1240
print_slabinfo_header(struct seq_file * m)1241 static void print_slabinfo_header(struct seq_file *m)
1242 {
1243 /*
1244 * Output format version, so at least we can change it
1245 * without _too_ many complaints.
1246 */
1247 #ifdef CONFIG_DEBUG_SLAB
1248 seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1249 #else
1250 seq_puts(m, "slabinfo - version: 2.1\n");
1251 #endif
1252 seq_puts(m, "# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1253 seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1254 seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1255 #ifdef CONFIG_DEBUG_SLAB
1256 seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1257 seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1258 #endif
1259 trace_android_vh_print_slabinfo_header(m);
1260 seq_putc(m, '\n');
1261 }
1262
slab_start(struct seq_file * m,loff_t * pos)1263 static void *slab_start(struct seq_file *m, loff_t *pos)
1264 {
1265 mutex_lock(&slab_mutex);
1266 return seq_list_start(&slab_caches, *pos);
1267 }
1268
slab_next(struct seq_file * m,void * p,loff_t * pos)1269 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1270 {
1271 return seq_list_next(p, &slab_caches, pos);
1272 }
1273
slab_stop(struct seq_file * m,void * p)1274 static void slab_stop(struct seq_file *m, void *p)
1275 {
1276 mutex_unlock(&slab_mutex);
1277 }
1278
cache_show(struct kmem_cache * s,struct seq_file * m)1279 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1280 {
1281 struct slabinfo sinfo;
1282
1283 memset(&sinfo, 0, sizeof(sinfo));
1284 get_slabinfo(s, &sinfo);
1285
1286 seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1287 s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1288 sinfo.objects_per_slab, (1 << sinfo.cache_order));
1289
1290 seq_printf(m, " : tunables %4u %4u %4u",
1291 sinfo.limit, sinfo.batchcount, sinfo.shared);
1292 seq_printf(m, " : slabdata %6lu %6lu %6lu",
1293 sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1294 slabinfo_show_stats(m, s);
1295 trace_android_vh_cache_show(m, &sinfo, s);
1296 seq_putc(m, '\n');
1297 }
1298
slab_show(struct seq_file * m,void * p)1299 static int slab_show(struct seq_file *m, void *p)
1300 {
1301 struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1302
1303 if (p == slab_caches.next)
1304 print_slabinfo_header(m);
1305 cache_show(s, m);
1306 return 0;
1307 }
1308
dump_unreclaimable_slab(void)1309 void dump_unreclaimable_slab(void)
1310 {
1311 struct kmem_cache *s;
1312 struct slabinfo sinfo;
1313
1314 /*
1315 * Here acquiring slab_mutex is risky since we don't prefer to get
1316 * sleep in oom path. But, without mutex hold, it may introduce a
1317 * risk of crash.
1318 * Use mutex_trylock to protect the list traverse, dump nothing
1319 * without acquiring the mutex.
1320 */
1321 if (!mutex_trylock(&slab_mutex)) {
1322 pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1323 return;
1324 }
1325
1326 pr_info("Unreclaimable slab info:\n");
1327 pr_info("Name Used Total\n");
1328
1329 list_for_each_entry(s, &slab_caches, list) {
1330 if (s->flags & SLAB_RECLAIM_ACCOUNT)
1331 continue;
1332
1333 get_slabinfo(s, &sinfo);
1334
1335 if (sinfo.num_objs > 0)
1336 pr_info("%-17s %10luKB %10luKB\n", s->name,
1337 (sinfo.active_objs * s->size) / 1024,
1338 (sinfo.num_objs * s->size) / 1024);
1339 }
1340 mutex_unlock(&slab_mutex);
1341 }
1342
1343 /*
1344 * slabinfo_op - iterator that generates /proc/slabinfo
1345 *
1346 * Output layout:
1347 * cache-name
1348 * num-active-objs
1349 * total-objs
1350 * object size
1351 * num-active-slabs
1352 * total-slabs
1353 * num-pages-per-slab
1354 * + further values on SMP and with statistics enabled
1355 */
1356 static const struct seq_operations slabinfo_op = {
1357 .start = slab_start,
1358 .next = slab_next,
1359 .stop = slab_stop,
1360 .show = slab_show,
1361 };
1362
slabinfo_open(struct inode * inode,struct file * file)1363 static int slabinfo_open(struct inode *inode, struct file *file)
1364 {
1365 return seq_open(file, &slabinfo_op);
1366 }
1367
1368 static const struct proc_ops slabinfo_proc_ops = {
1369 .proc_flags = PROC_ENTRY_PERMANENT,
1370 .proc_open = slabinfo_open,
1371 .proc_read = seq_read,
1372 .proc_write = slabinfo_write,
1373 .proc_lseek = seq_lseek,
1374 .proc_release = seq_release,
1375 };
1376
slab_proc_init(void)1377 static int __init slab_proc_init(void)
1378 {
1379 proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1380 return 0;
1381 }
1382 module_init(slab_proc_init);
1383
1384 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1385
1386 static __always_inline __realloc_size(2) void *
__do_krealloc(const void * p,size_t new_size,gfp_t flags)1387 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
1388 {
1389 void *ret;
1390 size_t ks;
1391
1392 /* Check for double-free before calling ksize. */
1393 if (likely(!ZERO_OR_NULL_PTR(p))) {
1394 if (!kasan_check_byte(p))
1395 return NULL;
1396 ks = ksize(p);
1397 } else
1398 ks = 0;
1399
1400 /* If the object still fits, repoison it precisely. */
1401 if (ks >= new_size) {
1402 p = kasan_krealloc((void *)p, new_size, flags);
1403 return (void *)p;
1404 }
1405
1406 ret = kmalloc_track_caller(new_size, flags);
1407 if (ret && p) {
1408 /* Disable KASAN checks as the object's redzone is accessed. */
1409 kasan_disable_current();
1410 memcpy(ret, kasan_reset_tag(p), ks);
1411 kasan_enable_current();
1412 }
1413
1414 return ret;
1415 }
1416
1417 /**
1418 * krealloc - reallocate memory. The contents will remain unchanged.
1419 * @p: object to reallocate memory for.
1420 * @new_size: how many bytes of memory are required.
1421 * @flags: the type of memory to allocate.
1422 *
1423 * The contents of the object pointed to are preserved up to the
1424 * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1425 * If @p is %NULL, krealloc() behaves exactly like kmalloc(). If @new_size
1426 * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1427 *
1428 * Return: pointer to the allocated memory or %NULL in case of error
1429 */
krealloc(const void * p,size_t new_size,gfp_t flags)1430 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1431 {
1432 void *ret;
1433
1434 if (unlikely(!new_size)) {
1435 kfree(p);
1436 return ZERO_SIZE_PTR;
1437 }
1438
1439 ret = __do_krealloc(p, new_size, flags);
1440 if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1441 kfree(p);
1442
1443 return ret;
1444 }
1445 EXPORT_SYMBOL(krealloc);
1446
1447 /**
1448 * kfree_sensitive - Clear sensitive information in memory before freeing
1449 * @p: object to free memory of
1450 *
1451 * The memory of the object @p points to is zeroed before freed.
1452 * If @p is %NULL, kfree_sensitive() does nothing.
1453 *
1454 * Note: this function zeroes the whole allocated buffer which can be a good
1455 * deal bigger than the requested buffer size passed to kmalloc(). So be
1456 * careful when using this function in performance sensitive code.
1457 */
kfree_sensitive(const void * p)1458 void kfree_sensitive(const void *p)
1459 {
1460 size_t ks;
1461 void *mem = (void *)p;
1462
1463 ks = ksize(mem);
1464 if (ks) {
1465 kasan_unpoison_range(mem, ks);
1466 memzero_explicit(mem, ks);
1467 }
1468 kfree(mem);
1469 }
1470 EXPORT_SYMBOL(kfree_sensitive);
1471
ksize(const void * objp)1472 size_t ksize(const void *objp)
1473 {
1474 /*
1475 * We need to first check that the pointer to the object is valid.
1476 * The KASAN report printed from ksize() is more useful, then when
1477 * it's printed later when the behaviour could be undefined due to
1478 * a potential use-after-free or double-free.
1479 *
1480 * We use kasan_check_byte(), which is supported for the hardware
1481 * tag-based KASAN mode, unlike kasan_check_read/write().
1482 *
1483 * If the pointed to memory is invalid, we return 0 to avoid users of
1484 * ksize() writing to and potentially corrupting the memory region.
1485 *
1486 * We want to perform the check before __ksize(), to avoid potentially
1487 * crashing in __ksize() due to accessing invalid metadata.
1488 */
1489 if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1490 return 0;
1491
1492 return kfence_ksize(objp) ?: __ksize(objp);
1493 }
1494 EXPORT_SYMBOL(ksize);
1495
1496 /* Tracepoints definitions. */
1497 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1498 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1499 EXPORT_TRACEPOINT_SYMBOL(kfree);
1500 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1501
should_failslab(struct kmem_cache * s,gfp_t gfpflags)1502 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1503 {
1504 if (__should_failslab(s, gfpflags))
1505 return -ENOMEM;
1506 return 0;
1507 }
1508 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
1509