xref: /openbmc/linux/mm/slab_common.c (revision 28dce2c4)
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 #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 
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 
53 /*
54  * Set of flags that will prevent slab merging
55  */
56 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
57 		SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
58 		SLAB_FAILSLAB | kasan_never_merge())
59 
60 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
61 			 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
62 
63 /*
64  * Merge control. If this is set then no merging of slab caches will occur.
65  */
66 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
67 
68 static int __init setup_slab_nomerge(char *str)
69 {
70 	slab_nomerge = true;
71 	return 1;
72 }
73 
74 static int __init setup_slab_merge(char *str)
75 {
76 	slab_nomerge = false;
77 	return 1;
78 }
79 
80 #ifdef CONFIG_SLUB
81 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
82 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
83 #endif
84 
85 __setup("slab_nomerge", setup_slab_nomerge);
86 __setup("slab_merge", setup_slab_merge);
87 
88 /*
89  * Determine the size of a slab object
90  */
91 unsigned int kmem_cache_size(struct kmem_cache *s)
92 {
93 	return s->object_size;
94 }
95 EXPORT_SYMBOL(kmem_cache_size);
96 
97 #ifdef CONFIG_DEBUG_VM
98 static int kmem_cache_sanity_check(const char *name, unsigned int size)
99 {
100 	if (!name || in_interrupt() || size < sizeof(void *) ||
101 		size > KMALLOC_MAX_SIZE) {
102 		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
103 		return -EINVAL;
104 	}
105 
106 	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
107 	return 0;
108 }
109 #else
110 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
111 {
112 	return 0;
113 }
114 #endif
115 
116 void __kmem_cache_free_bulk(struct kmem_cache *s, size_t nr, void **p)
117 {
118 	size_t i;
119 
120 	for (i = 0; i < nr; i++) {
121 		if (s)
122 			kmem_cache_free(s, p[i]);
123 		else
124 			kfree(p[i]);
125 	}
126 }
127 
128 int __kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t nr,
129 								void **p)
130 {
131 	size_t i;
132 
133 	for (i = 0; i < nr; i++) {
134 		void *x = p[i] = kmem_cache_alloc(s, flags);
135 		if (!x) {
136 			__kmem_cache_free_bulk(s, i, p);
137 			return 0;
138 		}
139 	}
140 	return i;
141 }
142 
143 /*
144  * Figure out what the alignment of the objects will be given a set of
145  * flags, a user specified alignment and the size of the objects.
146  */
147 static unsigned int calculate_alignment(slab_flags_t flags,
148 		unsigned int align, unsigned int size)
149 {
150 	/*
151 	 * If the user wants hardware cache aligned objects then follow that
152 	 * suggestion if the object is sufficiently large.
153 	 *
154 	 * The hardware cache alignment cannot override the specified
155 	 * alignment though. If that is greater then use it.
156 	 */
157 	if (flags & SLAB_HWCACHE_ALIGN) {
158 		unsigned int ralign;
159 
160 		ralign = cache_line_size();
161 		while (size <= ralign / 2)
162 			ralign /= 2;
163 		align = max(align, ralign);
164 	}
165 
166 	if (align < ARCH_SLAB_MINALIGN)
167 		align = ARCH_SLAB_MINALIGN;
168 
169 	return ALIGN(align, sizeof(void *));
170 }
171 
172 /*
173  * Find a mergeable slab cache
174  */
175 int slab_unmergeable(struct kmem_cache *s)
176 {
177 	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
178 		return 1;
179 
180 	if (s->ctor)
181 		return 1;
182 
183 	if (s->usersize)
184 		return 1;
185 
186 	/*
187 	 * We may have set a slab to be unmergeable during bootstrap.
188 	 */
189 	if (s->refcount < 0)
190 		return 1;
191 
192 	return 0;
193 }
194 
195 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
196 		slab_flags_t flags, const char *name, void (*ctor)(void *))
197 {
198 	struct kmem_cache *s;
199 
200 	if (slab_nomerge)
201 		return NULL;
202 
203 	if (ctor)
204 		return NULL;
205 
206 	size = ALIGN(size, sizeof(void *));
207 	align = calculate_alignment(flags, align, size);
208 	size = ALIGN(size, align);
209 	flags = kmem_cache_flags(size, flags, name);
210 
211 	if (flags & SLAB_NEVER_MERGE)
212 		return NULL;
213 
214 	list_for_each_entry_reverse(s, &slab_caches, list) {
215 		if (slab_unmergeable(s))
216 			continue;
217 
218 		if (size > s->size)
219 			continue;
220 
221 		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
222 			continue;
223 		/*
224 		 * Check if alignment is compatible.
225 		 * Courtesy of Adrian Drzewiecki
226 		 */
227 		if ((s->size & ~(align - 1)) != s->size)
228 			continue;
229 
230 		if (s->size - size >= sizeof(void *))
231 			continue;
232 
233 		if (IS_ENABLED(CONFIG_SLAB) && align &&
234 			(align > s->align || s->align % align))
235 			continue;
236 
237 		return s;
238 	}
239 	return NULL;
240 }
241 
242 static struct kmem_cache *create_cache(const char *name,
243 		unsigned int object_size, unsigned int align,
244 		slab_flags_t flags, unsigned int useroffset,
245 		unsigned int usersize, void (*ctor)(void *),
246 		struct kmem_cache *root_cache)
247 {
248 	struct kmem_cache *s;
249 	int err;
250 
251 	if (WARN_ON(useroffset + usersize > object_size))
252 		useroffset = usersize = 0;
253 
254 	err = -ENOMEM;
255 	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
256 	if (!s)
257 		goto out;
258 
259 	s->name = name;
260 	s->size = s->object_size = object_size;
261 	s->align = align;
262 	s->ctor = ctor;
263 	s->useroffset = useroffset;
264 	s->usersize = usersize;
265 
266 	err = __kmem_cache_create(s, flags);
267 	if (err)
268 		goto out_free_cache;
269 
270 	s->refcount = 1;
271 	list_add(&s->list, &slab_caches);
272 out:
273 	if (err)
274 		return ERR_PTR(err);
275 	return s;
276 
277 out_free_cache:
278 	kmem_cache_free(kmem_cache, s);
279 	goto out;
280 }
281 
282 /**
283  * kmem_cache_create_usercopy - Create a cache with a region suitable
284  * for copying to userspace
285  * @name: A string which is used in /proc/slabinfo to identify this cache.
286  * @size: The size of objects to be created in this cache.
287  * @align: The required alignment for the objects.
288  * @flags: SLAB flags
289  * @useroffset: Usercopy region offset
290  * @usersize: Usercopy region size
291  * @ctor: A constructor for the objects.
292  *
293  * Cannot be called within a interrupt, but can be interrupted.
294  * The @ctor is run when new pages are allocated by the cache.
295  *
296  * The flags are
297  *
298  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
299  * to catch references to uninitialised memory.
300  *
301  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
302  * for buffer overruns.
303  *
304  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
305  * cacheline.  This can be beneficial if you're counting cycles as closely
306  * as davem.
307  *
308  * Return: a pointer to the cache on success, NULL on failure.
309  */
310 struct kmem_cache *
311 kmem_cache_create_usercopy(const char *name,
312 		  unsigned int size, unsigned int align,
313 		  slab_flags_t flags,
314 		  unsigned int useroffset, unsigned int usersize,
315 		  void (*ctor)(void *))
316 {
317 	struct kmem_cache *s = NULL;
318 	const char *cache_name;
319 	int err;
320 
321 #ifdef CONFIG_SLUB_DEBUG
322 	/*
323 	 * If no slub_debug was enabled globally, the static key is not yet
324 	 * enabled by setup_slub_debug(). Enable it if the cache is being
325 	 * created with any of the debugging flags passed explicitly.
326 	 */
327 	if (flags & SLAB_DEBUG_FLAGS)
328 		static_branch_enable(&slub_debug_enabled);
329 #endif
330 
331 	mutex_lock(&slab_mutex);
332 
333 	err = kmem_cache_sanity_check(name, size);
334 	if (err) {
335 		goto out_unlock;
336 	}
337 
338 	/* Refuse requests with allocator specific flags */
339 	if (flags & ~SLAB_FLAGS_PERMITTED) {
340 		err = -EINVAL;
341 		goto out_unlock;
342 	}
343 
344 	/*
345 	 * Some allocators will constraint the set of valid flags to a subset
346 	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
347 	 * case, and we'll just provide them with a sanitized version of the
348 	 * passed flags.
349 	 */
350 	flags &= CACHE_CREATE_MASK;
351 
352 	/* Fail closed on bad usersize of useroffset values. */
353 	if (WARN_ON(!usersize && useroffset) ||
354 	    WARN_ON(size < usersize || size - usersize < useroffset))
355 		usersize = useroffset = 0;
356 
357 	if (!usersize)
358 		s = __kmem_cache_alias(name, size, align, flags, ctor);
359 	if (s)
360 		goto out_unlock;
361 
362 	cache_name = kstrdup_const(name, GFP_KERNEL);
363 	if (!cache_name) {
364 		err = -ENOMEM;
365 		goto out_unlock;
366 	}
367 
368 	s = create_cache(cache_name, size,
369 			 calculate_alignment(flags, align, size),
370 			 flags, useroffset, usersize, ctor, NULL);
371 	if (IS_ERR(s)) {
372 		err = PTR_ERR(s);
373 		kfree_const(cache_name);
374 	}
375 
376 out_unlock:
377 	mutex_unlock(&slab_mutex);
378 
379 	if (err) {
380 		if (flags & SLAB_PANIC)
381 			panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
382 				name, err);
383 		else {
384 			pr_warn("kmem_cache_create(%s) failed with error %d\n",
385 				name, err);
386 			dump_stack();
387 		}
388 		return NULL;
389 	}
390 	return s;
391 }
392 EXPORT_SYMBOL(kmem_cache_create_usercopy);
393 
394 /**
395  * kmem_cache_create - Create a cache.
396  * @name: A string which is used in /proc/slabinfo to identify this cache.
397  * @size: The size of objects to be created in this cache.
398  * @align: The required alignment for the objects.
399  * @flags: SLAB flags
400  * @ctor: A constructor for the objects.
401  *
402  * Cannot be called within a interrupt, but can be interrupted.
403  * The @ctor is run when new pages are allocated by the cache.
404  *
405  * The flags are
406  *
407  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
408  * to catch references to uninitialised memory.
409  *
410  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
411  * for buffer overruns.
412  *
413  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
414  * cacheline.  This can be beneficial if you're counting cycles as closely
415  * as davem.
416  *
417  * Return: a pointer to the cache on success, NULL on failure.
418  */
419 struct kmem_cache *
420 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
421 		slab_flags_t flags, void (*ctor)(void *))
422 {
423 	return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
424 					  ctor);
425 }
426 EXPORT_SYMBOL(kmem_cache_create);
427 
428 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
429 {
430 	LIST_HEAD(to_destroy);
431 	struct kmem_cache *s, *s2;
432 
433 	/*
434 	 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
435 	 * @slab_caches_to_rcu_destroy list.  The slab pages are freed
436 	 * through RCU and the associated kmem_cache are dereferenced
437 	 * while freeing the pages, so the kmem_caches should be freed only
438 	 * after the pending RCU operations are finished.  As rcu_barrier()
439 	 * is a pretty slow operation, we batch all pending destructions
440 	 * asynchronously.
441 	 */
442 	mutex_lock(&slab_mutex);
443 	list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
444 	mutex_unlock(&slab_mutex);
445 
446 	if (list_empty(&to_destroy))
447 		return;
448 
449 	rcu_barrier();
450 
451 	list_for_each_entry_safe(s, s2, &to_destroy, list) {
452 		kfence_shutdown_cache(s);
453 #ifdef SLAB_SUPPORTS_SYSFS
454 		sysfs_slab_release(s);
455 #else
456 		slab_kmem_cache_release(s);
457 #endif
458 	}
459 }
460 
461 static int shutdown_cache(struct kmem_cache *s)
462 {
463 	/* free asan quarantined objects */
464 	kasan_cache_shutdown(s);
465 
466 	if (__kmem_cache_shutdown(s) != 0)
467 		return -EBUSY;
468 
469 	list_del(&s->list);
470 
471 	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
472 #ifdef SLAB_SUPPORTS_SYSFS
473 		sysfs_slab_unlink(s);
474 #endif
475 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
476 		schedule_work(&slab_caches_to_rcu_destroy_work);
477 	} else {
478 		kfence_shutdown_cache(s);
479 #ifdef SLAB_SUPPORTS_SYSFS
480 		sysfs_slab_unlink(s);
481 		sysfs_slab_release(s);
482 #else
483 		slab_kmem_cache_release(s);
484 #endif
485 	}
486 
487 	return 0;
488 }
489 
490 void slab_kmem_cache_release(struct kmem_cache *s)
491 {
492 	__kmem_cache_release(s);
493 	kfree_const(s->name);
494 	kmem_cache_free(kmem_cache, s);
495 }
496 
497 void kmem_cache_destroy(struct kmem_cache *s)
498 {
499 	int err;
500 
501 	if (unlikely(!s))
502 		return;
503 
504 	mutex_lock(&slab_mutex);
505 
506 	s->refcount--;
507 	if (s->refcount)
508 		goto out_unlock;
509 
510 	err = shutdown_cache(s);
511 	if (err) {
512 		pr_err("kmem_cache_destroy %s: Slab cache still has objects\n",
513 		       s->name);
514 		dump_stack();
515 	}
516 out_unlock:
517 	mutex_unlock(&slab_mutex);
518 }
519 EXPORT_SYMBOL(kmem_cache_destroy);
520 
521 /**
522  * kmem_cache_shrink - Shrink a cache.
523  * @cachep: The cache to shrink.
524  *
525  * Releases as many slabs as possible for a cache.
526  * To help debugging, a zero exit status indicates all slabs were released.
527  *
528  * Return: %0 if all slabs were released, non-zero otherwise
529  */
530 int kmem_cache_shrink(struct kmem_cache *cachep)
531 {
532 	int ret;
533 
534 
535 	kasan_cache_shrink(cachep);
536 	ret = __kmem_cache_shrink(cachep);
537 
538 	return ret;
539 }
540 EXPORT_SYMBOL(kmem_cache_shrink);
541 
542 bool slab_is_available(void)
543 {
544 	return slab_state >= UP;
545 }
546 
547 #ifdef CONFIG_PRINTK
548 /**
549  * kmem_valid_obj - does the pointer reference a valid slab object?
550  * @object: pointer to query.
551  *
552  * Return: %true if the pointer is to a not-yet-freed object from
553  * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
554  * is to an already-freed object, and %false otherwise.
555  */
556 bool kmem_valid_obj(void *object)
557 {
558 	struct page *page;
559 
560 	/* Some arches consider ZERO_SIZE_PTR to be a valid address. */
561 	if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
562 		return false;
563 	page = virt_to_head_page(object);
564 	return PageSlab(page);
565 }
566 EXPORT_SYMBOL_GPL(kmem_valid_obj);
567 
568 /**
569  * kmem_dump_obj - Print available slab provenance information
570  * @object: slab object for which to find provenance information.
571  *
572  * This function uses pr_cont(), so that the caller is expected to have
573  * printed out whatever preamble is appropriate.  The provenance information
574  * depends on the type of object and on how much debugging is enabled.
575  * For a slab-cache object, the fact that it is a slab object is printed,
576  * and, if available, the slab name, return address, and stack trace from
577  * the allocation of that object.
578  *
579  * This function will splat if passed a pointer to a non-slab object.
580  * If you are not sure what type of object you have, you should instead
581  * use mem_dump_obj().
582  */
583 void kmem_dump_obj(void *object)
584 {
585 	char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
586 	int i;
587 	struct page *page;
588 	unsigned long ptroffset;
589 	struct kmem_obj_info kp = { };
590 
591 	if (WARN_ON_ONCE(!virt_addr_valid(object)))
592 		return;
593 	page = virt_to_head_page(object);
594 	if (WARN_ON_ONCE(!PageSlab(page))) {
595 		pr_cont(" non-slab memory.\n");
596 		return;
597 	}
598 	kmem_obj_info(&kp, object, page);
599 	if (kp.kp_slab_cache)
600 		pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
601 	else
602 		pr_cont(" slab%s", cp);
603 	if (kp.kp_objp)
604 		pr_cont(" start %px", kp.kp_objp);
605 	if (kp.kp_data_offset)
606 		pr_cont(" data offset %lu", kp.kp_data_offset);
607 	if (kp.kp_objp) {
608 		ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
609 		pr_cont(" pointer offset %lu", ptroffset);
610 	}
611 	if (kp.kp_slab_cache && kp.kp_slab_cache->usersize)
612 		pr_cont(" size %u", kp.kp_slab_cache->usersize);
613 	if (kp.kp_ret)
614 		pr_cont(" allocated at %pS\n", kp.kp_ret);
615 	else
616 		pr_cont("\n");
617 	for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
618 		if (!kp.kp_stack[i])
619 			break;
620 		pr_info("    %pS\n", kp.kp_stack[i]);
621 	}
622 }
623 EXPORT_SYMBOL_GPL(kmem_dump_obj);
624 #endif
625 
626 #ifndef CONFIG_SLOB
627 /* Create a cache during boot when no slab services are available yet */
628 void __init create_boot_cache(struct kmem_cache *s, const char *name,
629 		unsigned int size, slab_flags_t flags,
630 		unsigned int useroffset, unsigned int usersize)
631 {
632 	int err;
633 	unsigned int align = ARCH_KMALLOC_MINALIGN;
634 
635 	s->name = name;
636 	s->size = s->object_size = size;
637 
638 	/*
639 	 * For power of two sizes, guarantee natural alignment for kmalloc
640 	 * caches, regardless of SL*B debugging options.
641 	 */
642 	if (is_power_of_2(size))
643 		align = max(align, size);
644 	s->align = calculate_alignment(flags, align, size);
645 
646 	s->useroffset = useroffset;
647 	s->usersize = usersize;
648 
649 	err = __kmem_cache_create(s, flags);
650 
651 	if (err)
652 		panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
653 					name, size, err);
654 
655 	s->refcount = -1;	/* Exempt from merging for now */
656 }
657 
658 struct kmem_cache *__init create_kmalloc_cache(const char *name,
659 		unsigned int size, slab_flags_t flags,
660 		unsigned int useroffset, unsigned int usersize)
661 {
662 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
663 
664 	if (!s)
665 		panic("Out of memory when creating slab %s\n", name);
666 
667 	create_boot_cache(s, name, size, flags, useroffset, usersize);
668 	kasan_cache_create_kmalloc(s);
669 	list_add(&s->list, &slab_caches);
670 	s->refcount = 1;
671 	return s;
672 }
673 
674 struct kmem_cache *
675 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
676 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
677 EXPORT_SYMBOL(kmalloc_caches);
678 
679 /*
680  * Conversion table for small slabs sizes / 8 to the index in the
681  * kmalloc array. This is necessary for slabs < 192 since we have non power
682  * of two cache sizes there. The size of larger slabs can be determined using
683  * fls.
684  */
685 static u8 size_index[24] __ro_after_init = {
686 	3,	/* 8 */
687 	4,	/* 16 */
688 	5,	/* 24 */
689 	5,	/* 32 */
690 	6,	/* 40 */
691 	6,	/* 48 */
692 	6,	/* 56 */
693 	6,	/* 64 */
694 	1,	/* 72 */
695 	1,	/* 80 */
696 	1,	/* 88 */
697 	1,	/* 96 */
698 	7,	/* 104 */
699 	7,	/* 112 */
700 	7,	/* 120 */
701 	7,	/* 128 */
702 	2,	/* 136 */
703 	2,	/* 144 */
704 	2,	/* 152 */
705 	2,	/* 160 */
706 	2,	/* 168 */
707 	2,	/* 176 */
708 	2,	/* 184 */
709 	2	/* 192 */
710 };
711 
712 static inline unsigned int size_index_elem(unsigned int bytes)
713 {
714 	return (bytes - 1) / 8;
715 }
716 
717 /*
718  * Find the kmem_cache structure that serves a given size of
719  * allocation
720  */
721 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
722 {
723 	unsigned int index;
724 
725 	if (size <= 192) {
726 		if (!size)
727 			return ZERO_SIZE_PTR;
728 
729 		index = size_index[size_index_elem(size)];
730 	} else {
731 		if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
732 			return NULL;
733 		index = fls(size - 1);
734 	}
735 
736 	return kmalloc_caches[kmalloc_type(flags)][index];
737 }
738 
739 #ifdef CONFIG_ZONE_DMA
740 #define INIT_KMALLOC_INFO(__size, __short_size)			\
741 {								\
742 	.name[KMALLOC_NORMAL]  = "kmalloc-" #__short_size,	\
743 	.name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size,	\
744 	.name[KMALLOC_DMA]     = "dma-kmalloc-" #__short_size,	\
745 	.size = __size,						\
746 }
747 #else
748 #define INIT_KMALLOC_INFO(__size, __short_size)			\
749 {								\
750 	.name[KMALLOC_NORMAL]  = "kmalloc-" #__short_size,	\
751 	.name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size,	\
752 	.size = __size,						\
753 }
754 #endif
755 
756 /*
757  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
758  * kmalloc_index() supports up to 2^26=64MB, so the final entry of the table is
759  * kmalloc-67108864.
760  */
761 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
762 	INIT_KMALLOC_INFO(0, 0),
763 	INIT_KMALLOC_INFO(96, 96),
764 	INIT_KMALLOC_INFO(192, 192),
765 	INIT_KMALLOC_INFO(8, 8),
766 	INIT_KMALLOC_INFO(16, 16),
767 	INIT_KMALLOC_INFO(32, 32),
768 	INIT_KMALLOC_INFO(64, 64),
769 	INIT_KMALLOC_INFO(128, 128),
770 	INIT_KMALLOC_INFO(256, 256),
771 	INIT_KMALLOC_INFO(512, 512),
772 	INIT_KMALLOC_INFO(1024, 1k),
773 	INIT_KMALLOC_INFO(2048, 2k),
774 	INIT_KMALLOC_INFO(4096, 4k),
775 	INIT_KMALLOC_INFO(8192, 8k),
776 	INIT_KMALLOC_INFO(16384, 16k),
777 	INIT_KMALLOC_INFO(32768, 32k),
778 	INIT_KMALLOC_INFO(65536, 64k),
779 	INIT_KMALLOC_INFO(131072, 128k),
780 	INIT_KMALLOC_INFO(262144, 256k),
781 	INIT_KMALLOC_INFO(524288, 512k),
782 	INIT_KMALLOC_INFO(1048576, 1M),
783 	INIT_KMALLOC_INFO(2097152, 2M),
784 	INIT_KMALLOC_INFO(4194304, 4M),
785 	INIT_KMALLOC_INFO(8388608, 8M),
786 	INIT_KMALLOC_INFO(16777216, 16M),
787 	INIT_KMALLOC_INFO(33554432, 32M),
788 	INIT_KMALLOC_INFO(67108864, 64M)
789 };
790 
791 /*
792  * Patch up the size_index table if we have strange large alignment
793  * requirements for the kmalloc array. This is only the case for
794  * MIPS it seems. The standard arches will not generate any code here.
795  *
796  * Largest permitted alignment is 256 bytes due to the way we
797  * handle the index determination for the smaller caches.
798  *
799  * Make sure that nothing crazy happens if someone starts tinkering
800  * around with ARCH_KMALLOC_MINALIGN
801  */
802 void __init setup_kmalloc_cache_index_table(void)
803 {
804 	unsigned int i;
805 
806 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
807 		(KMALLOC_MIN_SIZE & (KMALLOC_MIN_SIZE - 1)));
808 
809 	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
810 		unsigned int elem = size_index_elem(i);
811 
812 		if (elem >= ARRAY_SIZE(size_index))
813 			break;
814 		size_index[elem] = KMALLOC_SHIFT_LOW;
815 	}
816 
817 	if (KMALLOC_MIN_SIZE >= 64) {
818 		/*
819 		 * The 96 byte size cache is not used if the alignment
820 		 * is 64 byte.
821 		 */
822 		for (i = 64 + 8; i <= 96; i += 8)
823 			size_index[size_index_elem(i)] = 7;
824 
825 	}
826 
827 	if (KMALLOC_MIN_SIZE >= 128) {
828 		/*
829 		 * The 192 byte sized cache is not used if the alignment
830 		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
831 		 * instead.
832 		 */
833 		for (i = 128 + 8; i <= 192; i += 8)
834 			size_index[size_index_elem(i)] = 8;
835 	}
836 }
837 
838 static void __init
839 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
840 {
841 	if (type == KMALLOC_RECLAIM)
842 		flags |= SLAB_RECLAIM_ACCOUNT;
843 
844 	kmalloc_caches[type][idx] = create_kmalloc_cache(
845 					kmalloc_info[idx].name[type],
846 					kmalloc_info[idx].size, flags, 0,
847 					kmalloc_info[idx].size);
848 }
849 
850 /*
851  * Create the kmalloc array. Some of the regular kmalloc arrays
852  * may already have been created because they were needed to
853  * enable allocations for slab creation.
854  */
855 void __init create_kmalloc_caches(slab_flags_t flags)
856 {
857 	int i;
858 	enum kmalloc_cache_type type;
859 
860 	for (type = KMALLOC_NORMAL; type <= KMALLOC_RECLAIM; type++) {
861 		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
862 			if (!kmalloc_caches[type][i])
863 				new_kmalloc_cache(i, type, flags);
864 
865 			/*
866 			 * Caches that are not of the two-to-the-power-of size.
867 			 * These have to be created immediately after the
868 			 * earlier power of two caches
869 			 */
870 			if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
871 					!kmalloc_caches[type][1])
872 				new_kmalloc_cache(1, type, flags);
873 			if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
874 					!kmalloc_caches[type][2])
875 				new_kmalloc_cache(2, type, flags);
876 		}
877 	}
878 
879 	/* Kmalloc array is now usable */
880 	slab_state = UP;
881 
882 #ifdef CONFIG_ZONE_DMA
883 	for (i = 0; i <= KMALLOC_SHIFT_HIGH; i++) {
884 		struct kmem_cache *s = kmalloc_caches[KMALLOC_NORMAL][i];
885 
886 		if (s) {
887 			kmalloc_caches[KMALLOC_DMA][i] = create_kmalloc_cache(
888 				kmalloc_info[i].name[KMALLOC_DMA],
889 				kmalloc_info[i].size,
890 				SLAB_CACHE_DMA | flags, 0,
891 				kmalloc_info[i].size);
892 		}
893 	}
894 #endif
895 }
896 #endif /* !CONFIG_SLOB */
897 
898 gfp_t kmalloc_fix_flags(gfp_t flags)
899 {
900 	gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
901 
902 	flags &= ~GFP_SLAB_BUG_MASK;
903 	pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
904 			invalid_mask, &invalid_mask, flags, &flags);
905 	dump_stack();
906 
907 	return flags;
908 }
909 
910 /*
911  * To avoid unnecessary overhead, we pass through large allocation requests
912  * directly to the page allocator. We use __GFP_COMP, because we will need to
913  * know the allocation order to free the pages properly in kfree.
914  */
915 void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
916 {
917 	void *ret = NULL;
918 	struct page *page;
919 
920 	if (unlikely(flags & GFP_SLAB_BUG_MASK))
921 		flags = kmalloc_fix_flags(flags);
922 
923 	flags |= __GFP_COMP;
924 	page = alloc_pages(flags, order);
925 	if (likely(page)) {
926 		ret = page_address(page);
927 		mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
928 				      PAGE_SIZE << order);
929 	}
930 	ret = kasan_kmalloc_large(ret, size, flags);
931 	/* As ret might get tagged, call kmemleak hook after KASAN. */
932 	kmemleak_alloc(ret, size, 1, flags);
933 	return ret;
934 }
935 EXPORT_SYMBOL(kmalloc_order);
936 
937 #ifdef CONFIG_TRACING
938 void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
939 {
940 	void *ret = kmalloc_order(size, flags, order);
941 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
942 	return ret;
943 }
944 EXPORT_SYMBOL(kmalloc_order_trace);
945 #endif
946 
947 #ifdef CONFIG_SLAB_FREELIST_RANDOM
948 /* Randomize a generic freelist */
949 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
950 			       unsigned int count)
951 {
952 	unsigned int rand;
953 	unsigned int i;
954 
955 	for (i = 0; i < count; i++)
956 		list[i] = i;
957 
958 	/* Fisher-Yates shuffle */
959 	for (i = count - 1; i > 0; i--) {
960 		rand = prandom_u32_state(state);
961 		rand %= (i + 1);
962 		swap(list[i], list[rand]);
963 	}
964 }
965 
966 /* Create a random sequence per cache */
967 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
968 				    gfp_t gfp)
969 {
970 	struct rnd_state state;
971 
972 	if (count < 2 || cachep->random_seq)
973 		return 0;
974 
975 	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
976 	if (!cachep->random_seq)
977 		return -ENOMEM;
978 
979 	/* Get best entropy at this stage of boot */
980 	prandom_seed_state(&state, get_random_long());
981 
982 	freelist_randomize(&state, cachep->random_seq, count);
983 	return 0;
984 }
985 
986 /* Destroy the per-cache random freelist sequence */
987 void cache_random_seq_destroy(struct kmem_cache *cachep)
988 {
989 	kfree(cachep->random_seq);
990 	cachep->random_seq = NULL;
991 }
992 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
993 
994 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
995 #ifdef CONFIG_SLAB
996 #define SLABINFO_RIGHTS (0600)
997 #else
998 #define SLABINFO_RIGHTS (0400)
999 #endif
1000 
1001 static void print_slabinfo_header(struct seq_file *m)
1002 {
1003 	/*
1004 	 * Output format version, so at least we can change it
1005 	 * without _too_ many complaints.
1006 	 */
1007 #ifdef CONFIG_DEBUG_SLAB
1008 	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1009 #else
1010 	seq_puts(m, "slabinfo - version: 2.1\n");
1011 #endif
1012 	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1013 	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1014 	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1015 #ifdef CONFIG_DEBUG_SLAB
1016 	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1017 	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1018 #endif
1019 	seq_putc(m, '\n');
1020 }
1021 
1022 void *slab_start(struct seq_file *m, loff_t *pos)
1023 {
1024 	mutex_lock(&slab_mutex);
1025 	return seq_list_start(&slab_caches, *pos);
1026 }
1027 
1028 void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1029 {
1030 	return seq_list_next(p, &slab_caches, pos);
1031 }
1032 
1033 void slab_stop(struct seq_file *m, void *p)
1034 {
1035 	mutex_unlock(&slab_mutex);
1036 }
1037 
1038 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1039 {
1040 	struct slabinfo sinfo;
1041 
1042 	memset(&sinfo, 0, sizeof(sinfo));
1043 	get_slabinfo(s, &sinfo);
1044 
1045 	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1046 		   s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1047 		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1048 
1049 	seq_printf(m, " : tunables %4u %4u %4u",
1050 		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1051 	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1052 		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1053 	slabinfo_show_stats(m, s);
1054 	seq_putc(m, '\n');
1055 }
1056 
1057 static int slab_show(struct seq_file *m, void *p)
1058 {
1059 	struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1060 
1061 	if (p == slab_caches.next)
1062 		print_slabinfo_header(m);
1063 	cache_show(s, m);
1064 	return 0;
1065 }
1066 
1067 void dump_unreclaimable_slab(void)
1068 {
1069 	struct kmem_cache *s;
1070 	struct slabinfo sinfo;
1071 
1072 	/*
1073 	 * Here acquiring slab_mutex is risky since we don't prefer to get
1074 	 * sleep in oom path. But, without mutex hold, it may introduce a
1075 	 * risk of crash.
1076 	 * Use mutex_trylock to protect the list traverse, dump nothing
1077 	 * without acquiring the mutex.
1078 	 */
1079 	if (!mutex_trylock(&slab_mutex)) {
1080 		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1081 		return;
1082 	}
1083 
1084 	pr_info("Unreclaimable slab info:\n");
1085 	pr_info("Name                      Used          Total\n");
1086 
1087 	list_for_each_entry(s, &slab_caches, list) {
1088 		if (s->flags & SLAB_RECLAIM_ACCOUNT)
1089 			continue;
1090 
1091 		get_slabinfo(s, &sinfo);
1092 
1093 		if (sinfo.num_objs > 0)
1094 			pr_info("%-17s %10luKB %10luKB\n", s->name,
1095 				(sinfo.active_objs * s->size) / 1024,
1096 				(sinfo.num_objs * s->size) / 1024);
1097 	}
1098 	mutex_unlock(&slab_mutex);
1099 }
1100 
1101 #if defined(CONFIG_MEMCG_KMEM)
1102 int memcg_slab_show(struct seq_file *m, void *p)
1103 {
1104 	/*
1105 	 * Deprecated.
1106 	 * Please, take a look at tools/cgroup/slabinfo.py .
1107 	 */
1108 	return 0;
1109 }
1110 #endif
1111 
1112 /*
1113  * slabinfo_op - iterator that generates /proc/slabinfo
1114  *
1115  * Output layout:
1116  * cache-name
1117  * num-active-objs
1118  * total-objs
1119  * object size
1120  * num-active-slabs
1121  * total-slabs
1122  * num-pages-per-slab
1123  * + further values on SMP and with statistics enabled
1124  */
1125 static const struct seq_operations slabinfo_op = {
1126 	.start = slab_start,
1127 	.next = slab_next,
1128 	.stop = slab_stop,
1129 	.show = slab_show,
1130 };
1131 
1132 static int slabinfo_open(struct inode *inode, struct file *file)
1133 {
1134 	return seq_open(file, &slabinfo_op);
1135 }
1136 
1137 static const struct proc_ops slabinfo_proc_ops = {
1138 	.proc_flags	= PROC_ENTRY_PERMANENT,
1139 	.proc_open	= slabinfo_open,
1140 	.proc_read	= seq_read,
1141 	.proc_write	= slabinfo_write,
1142 	.proc_lseek	= seq_lseek,
1143 	.proc_release	= seq_release,
1144 };
1145 
1146 static int __init slab_proc_init(void)
1147 {
1148 	proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1149 	return 0;
1150 }
1151 module_init(slab_proc_init);
1152 
1153 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1154 
1155 static __always_inline void *__do_krealloc(const void *p, size_t new_size,
1156 					   gfp_t flags)
1157 {
1158 	void *ret;
1159 	size_t ks;
1160 
1161 	/* Don't use instrumented ksize to allow precise KASAN poisoning. */
1162 	if (likely(!ZERO_OR_NULL_PTR(p))) {
1163 		if (!kasan_check_byte(p))
1164 			return NULL;
1165 		ks = kfence_ksize(p) ?: __ksize(p);
1166 	} else
1167 		ks = 0;
1168 
1169 	/* If the object still fits, repoison it precisely. */
1170 	if (ks >= new_size) {
1171 		p = kasan_krealloc((void *)p, new_size, flags);
1172 		return (void *)p;
1173 	}
1174 
1175 	ret = kmalloc_track_caller(new_size, flags);
1176 	if (ret && p) {
1177 		/* Disable KASAN checks as the object's redzone is accessed. */
1178 		kasan_disable_current();
1179 		memcpy(ret, kasan_reset_tag(p), ks);
1180 		kasan_enable_current();
1181 	}
1182 
1183 	return ret;
1184 }
1185 
1186 /**
1187  * krealloc - reallocate memory. The contents will remain unchanged.
1188  * @p: object to reallocate memory for.
1189  * @new_size: how many bytes of memory are required.
1190  * @flags: the type of memory to allocate.
1191  *
1192  * The contents of the object pointed to are preserved up to the
1193  * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1194  * If @p is %NULL, krealloc() behaves exactly like kmalloc().  If @new_size
1195  * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1196  *
1197  * Return: pointer to the allocated memory or %NULL in case of error
1198  */
1199 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1200 {
1201 	void *ret;
1202 
1203 	if (unlikely(!new_size)) {
1204 		kfree(p);
1205 		return ZERO_SIZE_PTR;
1206 	}
1207 
1208 	ret = __do_krealloc(p, new_size, flags);
1209 	if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1210 		kfree(p);
1211 
1212 	return ret;
1213 }
1214 EXPORT_SYMBOL(krealloc);
1215 
1216 /**
1217  * kfree_sensitive - Clear sensitive information in memory before freeing
1218  * @p: object to free memory of
1219  *
1220  * The memory of the object @p points to is zeroed before freed.
1221  * If @p is %NULL, kfree_sensitive() does nothing.
1222  *
1223  * Note: this function zeroes the whole allocated buffer which can be a good
1224  * deal bigger than the requested buffer size passed to kmalloc(). So be
1225  * careful when using this function in performance sensitive code.
1226  */
1227 void kfree_sensitive(const void *p)
1228 {
1229 	size_t ks;
1230 	void *mem = (void *)p;
1231 
1232 	ks = ksize(mem);
1233 	if (ks)
1234 		memzero_explicit(mem, ks);
1235 	kfree(mem);
1236 }
1237 EXPORT_SYMBOL(kfree_sensitive);
1238 
1239 /**
1240  * ksize - get the actual amount of memory allocated for a given object
1241  * @objp: Pointer to the object
1242  *
1243  * kmalloc may internally round up allocations and return more memory
1244  * than requested. ksize() can be used to determine the actual amount of
1245  * memory allocated. The caller may use this additional memory, even though
1246  * a smaller amount of memory was initially specified with the kmalloc call.
1247  * The caller must guarantee that objp points to a valid object previously
1248  * allocated with either kmalloc() or kmem_cache_alloc(). The object
1249  * must not be freed during the duration of the call.
1250  *
1251  * Return: size of the actual memory used by @objp in bytes
1252  */
1253 size_t ksize(const void *objp)
1254 {
1255 	size_t size;
1256 
1257 	/*
1258 	 * We need to first check that the pointer to the object is valid, and
1259 	 * only then unpoison the memory. The report printed from ksize() is
1260 	 * more useful, then when it's printed later when the behaviour could
1261 	 * be undefined due to a potential use-after-free or double-free.
1262 	 *
1263 	 * We use kasan_check_byte(), which is supported for the hardware
1264 	 * tag-based KASAN mode, unlike kasan_check_read/write().
1265 	 *
1266 	 * If the pointed to memory is invalid, we return 0 to avoid users of
1267 	 * ksize() writing to and potentially corrupting the memory region.
1268 	 *
1269 	 * We want to perform the check before __ksize(), to avoid potentially
1270 	 * crashing in __ksize() due to accessing invalid metadata.
1271 	 */
1272 	if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1273 		return 0;
1274 
1275 	size = kfence_ksize(objp) ?: __ksize(objp);
1276 	/*
1277 	 * We assume that ksize callers could use whole allocated area,
1278 	 * so we need to unpoison this area.
1279 	 */
1280 	kasan_unpoison_range(objp, size);
1281 	return size;
1282 }
1283 EXPORT_SYMBOL(ksize);
1284 
1285 /* Tracepoints definitions. */
1286 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1287 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1288 EXPORT_TRACEPOINT_SYMBOL(kmalloc_node);
1289 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc_node);
1290 EXPORT_TRACEPOINT_SYMBOL(kfree);
1291 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1292 
1293 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1294 {
1295 	if (__should_failslab(s, gfpflags))
1296 		return -ENOMEM;
1297 	return 0;
1298 }
1299 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
1300