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 #include <linux/stackdepot.h>
28 
29 #include "internal.h"
30 #include "slab.h"
31 
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/kmem.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 
setup_slab_nomerge(char * str)60 static int __init setup_slab_nomerge(char *str)
61 {
62 	slab_nomerge = true;
63 	return 1;
64 }
65 
setup_slab_merge(char * str)66 static int __init setup_slab_merge(char *str)
67 {
68 	slab_nomerge = false;
69 	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  */
kmem_cache_size(struct kmem_cache * s)83 unsigned int kmem_cache_size(struct kmem_cache *s)
84 {
85 	return s->object_size;
86 }
87 EXPORT_SYMBOL(kmem_cache_size);
88 
89 #ifdef CONFIG_DEBUG_VM
kmem_cache_sanity_check(const char * name,unsigned int size)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
kmem_cache_sanity_check(const char * name,unsigned int size)101 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
102 {
103 	return 0;
104 }
105 #endif
106 
107 /*
108  * Figure out what the alignment of the objects will be given a set of
109  * flags, a user specified alignment and the size of the objects.
110  */
calculate_alignment(slab_flags_t flags,unsigned int align,unsigned int size)111 static unsigned int calculate_alignment(slab_flags_t flags,
112 		unsigned int align, unsigned int size)
113 {
114 	/*
115 	 * If the user wants hardware cache aligned objects then follow that
116 	 * suggestion if the object is sufficiently large.
117 	 *
118 	 * The hardware cache alignment cannot override the specified
119 	 * alignment though. If that is greater then use it.
120 	 */
121 	if (flags & SLAB_HWCACHE_ALIGN) {
122 		unsigned int ralign;
123 
124 		ralign = cache_line_size();
125 		while (size <= ralign / 2)
126 			ralign /= 2;
127 		align = max(align, ralign);
128 	}
129 
130 	align = max(align, arch_slab_minalign());
131 
132 	return ALIGN(align, sizeof(void *));
133 }
134 
135 /*
136  * Find a mergeable slab cache
137  */
slab_unmergeable(struct kmem_cache * s)138 int slab_unmergeable(struct kmem_cache *s)
139 {
140 	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
141 		return 1;
142 
143 	if (s->ctor)
144 		return 1;
145 
146 	if (s->usersize)
147 		return 1;
148 
149 	/*
150 	 * We may have set a slab to be unmergeable during bootstrap.
151 	 */
152 	if (s->refcount < 0)
153 		return 1;
154 
155 	return 0;
156 }
157 
find_mergeable(unsigned int size,unsigned int align,slab_flags_t flags,const char * name,void (* ctor)(void *))158 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
159 		slab_flags_t flags, const char *name, void (*ctor)(void *))
160 {
161 	struct kmem_cache *s;
162 
163 	if (slab_nomerge)
164 		return NULL;
165 
166 	if (ctor)
167 		return NULL;
168 
169 	size = ALIGN(size, sizeof(void *));
170 	align = calculate_alignment(flags, align, size);
171 	size = ALIGN(size, align);
172 	flags = kmem_cache_flags(size, flags, name);
173 
174 	if (flags & SLAB_NEVER_MERGE)
175 		return NULL;
176 
177 	list_for_each_entry_reverse(s, &slab_caches, list) {
178 		if (slab_unmergeable(s))
179 			continue;
180 
181 		if (size > s->size)
182 			continue;
183 
184 		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
185 			continue;
186 		/*
187 		 * Check if alignment is compatible.
188 		 * Courtesy of Adrian Drzewiecki
189 		 */
190 		if ((s->size & ~(align - 1)) != s->size)
191 			continue;
192 
193 		if (s->size - size >= sizeof(void *))
194 			continue;
195 
196 		if (IS_ENABLED(CONFIG_SLAB) && align &&
197 			(align > s->align || s->align % align))
198 			continue;
199 
200 		return s;
201 	}
202 	return NULL;
203 }
204 
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)205 static struct kmem_cache *create_cache(const char *name,
206 		unsigned int object_size, unsigned int align,
207 		slab_flags_t flags, unsigned int useroffset,
208 		unsigned int usersize, void (*ctor)(void *),
209 		struct kmem_cache *root_cache)
210 {
211 	struct kmem_cache *s;
212 	int err;
213 
214 	if (WARN_ON(useroffset + usersize > object_size))
215 		useroffset = usersize = 0;
216 
217 	err = -ENOMEM;
218 	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
219 	if (!s)
220 		goto out;
221 
222 	s->name = name;
223 	s->size = s->object_size = object_size;
224 	s->align = align;
225 	s->ctor = ctor;
226 	s->useroffset = useroffset;
227 	s->usersize = usersize;
228 
229 	err = __kmem_cache_create(s, flags);
230 	if (err)
231 		goto out_free_cache;
232 
233 	s->refcount = 1;
234 	list_add(&s->list, &slab_caches);
235 out:
236 	if (err)
237 		return ERR_PTR(err);
238 	return s;
239 
240 out_free_cache:
241 	kmem_cache_free(kmem_cache, s);
242 	goto out;
243 }
244 
245 /**
246  * kmem_cache_create_usercopy - Create a cache with a region suitable
247  * for copying to userspace
248  * @name: A string which is used in /proc/slabinfo to identify this cache.
249  * @size: The size of objects to be created in this cache.
250  * @align: The required alignment for the objects.
251  * @flags: SLAB flags
252  * @useroffset: Usercopy region offset
253  * @usersize: Usercopy region size
254  * @ctor: A constructor for the objects.
255  *
256  * Cannot be called within a interrupt, but can be interrupted.
257  * The @ctor is run when new pages are allocated by the cache.
258  *
259  * The flags are
260  *
261  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
262  * to catch references to uninitialised memory.
263  *
264  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
265  * for buffer overruns.
266  *
267  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
268  * cacheline.  This can be beneficial if you're counting cycles as closely
269  * as davem.
270  *
271  * Return: a pointer to the cache on success, NULL on failure.
272  */
273 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 *))274 kmem_cache_create_usercopy(const char *name,
275 		  unsigned int size, unsigned int align,
276 		  slab_flags_t flags,
277 		  unsigned int useroffset, unsigned int usersize,
278 		  void (*ctor)(void *))
279 {
280 	struct kmem_cache *s = NULL;
281 	const char *cache_name;
282 	int err;
283 
284 #ifdef CONFIG_SLUB_DEBUG
285 	/*
286 	 * If no slub_debug was enabled globally, the static key is not yet
287 	 * enabled by setup_slub_debug(). Enable it if the cache is being
288 	 * created with any of the debugging flags passed explicitly.
289 	 * It's also possible that this is the first cache created with
290 	 * SLAB_STORE_USER and we should init stack_depot for it.
291 	 */
292 	if (flags & SLAB_DEBUG_FLAGS)
293 		static_branch_enable(&slub_debug_enabled);
294 	if (flags & SLAB_STORE_USER)
295 		stack_depot_init();
296 #endif
297 
298 	mutex_lock(&slab_mutex);
299 
300 	err = kmem_cache_sanity_check(name, size);
301 	if (err) {
302 		goto out_unlock;
303 	}
304 
305 	/* Refuse requests with allocator specific flags */
306 	if (flags & ~SLAB_FLAGS_PERMITTED) {
307 		err = -EINVAL;
308 		goto out_unlock;
309 	}
310 
311 	/*
312 	 * Some allocators will constraint the set of valid flags to a subset
313 	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
314 	 * case, and we'll just provide them with a sanitized version of the
315 	 * passed flags.
316 	 */
317 	flags &= CACHE_CREATE_MASK;
318 
319 	/* Fail closed on bad usersize of useroffset values. */
320 	if (WARN_ON(!usersize && useroffset) ||
321 	    WARN_ON(size < usersize || size - usersize < useroffset))
322 		usersize = useroffset = 0;
323 
324 	if (!usersize)
325 		s = __kmem_cache_alias(name, size, align, flags, ctor);
326 	if (s)
327 		goto out_unlock;
328 
329 	cache_name = kstrdup_const(name, GFP_KERNEL);
330 	if (!cache_name) {
331 		err = -ENOMEM;
332 		goto out_unlock;
333 	}
334 
335 	s = create_cache(cache_name, size,
336 			 calculate_alignment(flags, align, size),
337 			 flags, useroffset, usersize, ctor, NULL);
338 	if (IS_ERR(s)) {
339 		err = PTR_ERR(s);
340 		kfree_const(cache_name);
341 	}
342 
343 out_unlock:
344 	mutex_unlock(&slab_mutex);
345 
346 	if (err) {
347 		if (flags & SLAB_PANIC)
348 			panic("%s: Failed to create slab '%s'. Error %d\n",
349 				__func__, name, err);
350 		else {
351 			pr_warn("%s(%s) failed with error %d\n",
352 				__func__, name, err);
353 			dump_stack();
354 		}
355 		return NULL;
356 	}
357 	return s;
358 }
359 EXPORT_SYMBOL(kmem_cache_create_usercopy);
360 
361 /**
362  * kmem_cache_create - Create a cache.
363  * @name: A string which is used in /proc/slabinfo to identify this cache.
364  * @size: The size of objects to be created in this cache.
365  * @align: The required alignment for the objects.
366  * @flags: SLAB flags
367  * @ctor: A constructor for the objects.
368  *
369  * Cannot be called within a interrupt, but can be interrupted.
370  * The @ctor is run when new pages are allocated by the cache.
371  *
372  * The flags are
373  *
374  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
375  * to catch references to uninitialised memory.
376  *
377  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
378  * for buffer overruns.
379  *
380  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
381  * cacheline.  This can be beneficial if you're counting cycles as closely
382  * as davem.
383  *
384  * Return: a pointer to the cache on success, NULL on failure.
385  */
386 struct kmem_cache *
kmem_cache_create(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))387 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
388 		slab_flags_t flags, void (*ctor)(void *))
389 {
390 	return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
391 					  ctor);
392 }
393 EXPORT_SYMBOL(kmem_cache_create);
394 
395 #ifdef SLAB_SUPPORTS_SYSFS
396 /*
397  * For a given kmem_cache, kmem_cache_destroy() should only be called
398  * once or there will be a use-after-free problem. The actual deletion
399  * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
400  * protection. So they are now done without holding those locks.
401  *
402  * Note that there will be a slight delay in the deletion of sysfs files
403  * if kmem_cache_release() is called indrectly from a work function.
404  */
kmem_cache_release(struct kmem_cache * s)405 static void kmem_cache_release(struct kmem_cache *s)
406 {
407 	sysfs_slab_unlink(s);
408 	sysfs_slab_release(s);
409 }
410 #else
kmem_cache_release(struct kmem_cache * s)411 static void kmem_cache_release(struct kmem_cache *s)
412 {
413 	slab_kmem_cache_release(s);
414 }
415 #endif
416 
slab_caches_to_rcu_destroy_workfn(struct work_struct * work)417 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
418 {
419 	LIST_HEAD(to_destroy);
420 	struct kmem_cache *s, *s2;
421 
422 	/*
423 	 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
424 	 * @slab_caches_to_rcu_destroy list.  The slab pages are freed
425 	 * through RCU and the associated kmem_cache are dereferenced
426 	 * while freeing the pages, so the kmem_caches should be freed only
427 	 * after the pending RCU operations are finished.  As rcu_barrier()
428 	 * is a pretty slow operation, we batch all pending destructions
429 	 * asynchronously.
430 	 */
431 	mutex_lock(&slab_mutex);
432 	list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
433 	mutex_unlock(&slab_mutex);
434 
435 	if (list_empty(&to_destroy))
436 		return;
437 
438 	rcu_barrier();
439 
440 	list_for_each_entry_safe(s, s2, &to_destroy, list) {
441 		debugfs_slab_release(s);
442 		kfence_shutdown_cache(s);
443 		kmem_cache_release(s);
444 	}
445 }
446 
shutdown_cache(struct kmem_cache * s)447 static int shutdown_cache(struct kmem_cache *s)
448 {
449 	/* free asan quarantined objects */
450 	kasan_cache_shutdown(s);
451 
452 	if (__kmem_cache_shutdown(s) != 0)
453 		return -EBUSY;
454 
455 	list_del(&s->list);
456 
457 	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
458 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
459 		schedule_work(&slab_caches_to_rcu_destroy_work);
460 	} else {
461 		kfence_shutdown_cache(s);
462 		debugfs_slab_release(s);
463 	}
464 
465 	return 0;
466 }
467 
slab_kmem_cache_release(struct kmem_cache * s)468 void slab_kmem_cache_release(struct kmem_cache *s)
469 {
470 	__kmem_cache_release(s);
471 	kfree_const(s->name);
472 	kmem_cache_free(kmem_cache, s);
473 }
474 
kmem_cache_destroy(struct kmem_cache * s)475 void kmem_cache_destroy(struct kmem_cache *s)
476 {
477 	int refcnt;
478 	bool rcu_set;
479 
480 	if (unlikely(!s) || !kasan_check_byte(s))
481 		return;
482 
483 	cpus_read_lock();
484 	mutex_lock(&slab_mutex);
485 
486 	rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
487 
488 	refcnt = --s->refcount;
489 	if (refcnt)
490 		goto out_unlock;
491 
492 	WARN(shutdown_cache(s),
493 	     "%s %s: Slab cache still has objects when called from %pS",
494 	     __func__, s->name, (void *)_RET_IP_);
495 out_unlock:
496 	mutex_unlock(&slab_mutex);
497 	cpus_read_unlock();
498 	if (!refcnt && !rcu_set)
499 		kmem_cache_release(s);
500 }
501 EXPORT_SYMBOL(kmem_cache_destroy);
502 
503 /**
504  * kmem_cache_shrink - Shrink a cache.
505  * @cachep: The cache to shrink.
506  *
507  * Releases as many slabs as possible for a cache.
508  * To help debugging, a zero exit status indicates all slabs were released.
509  *
510  * Return: %0 if all slabs were released, non-zero otherwise
511  */
kmem_cache_shrink(struct kmem_cache * cachep)512 int kmem_cache_shrink(struct kmem_cache *cachep)
513 {
514 	kasan_cache_shrink(cachep);
515 
516 	return __kmem_cache_shrink(cachep);
517 }
518 EXPORT_SYMBOL(kmem_cache_shrink);
519 
slab_is_available(void)520 bool slab_is_available(void)
521 {
522 	return slab_state >= UP;
523 }
524 
525 #ifdef CONFIG_PRINTK
526 /**
527  * kmem_valid_obj - does the pointer reference a valid slab object?
528  * @object: pointer to query.
529  *
530  * Return: %true if the pointer is to a not-yet-freed object from
531  * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
532  * is to an already-freed object, and %false otherwise.
533  */
kmem_valid_obj(void * object)534 bool kmem_valid_obj(void *object)
535 {
536 	struct folio *folio;
537 
538 	/* Some arches consider ZERO_SIZE_PTR to be a valid address. */
539 	if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
540 		return false;
541 	folio = virt_to_folio(object);
542 	return folio_test_slab(folio);
543 }
544 EXPORT_SYMBOL_GPL(kmem_valid_obj);
545 
kmem_obj_info(struct kmem_obj_info * kpp,void * object,struct slab * slab)546 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
547 {
548 	if (__kfence_obj_info(kpp, object, slab))
549 		return;
550 	__kmem_obj_info(kpp, object, slab);
551 }
552 
553 /**
554  * kmem_dump_obj - Print available slab provenance information
555  * @object: slab object for which to find provenance information.
556  *
557  * This function uses pr_cont(), so that the caller is expected to have
558  * printed out whatever preamble is appropriate.  The provenance information
559  * depends on the type of object and on how much debugging is enabled.
560  * For a slab-cache object, the fact that it is a slab object is printed,
561  * and, if available, the slab name, return address, and stack trace from
562  * the allocation and last free path of that object.
563  *
564  * This function will splat if passed a pointer to a non-slab object.
565  * If you are not sure what type of object you have, you should instead
566  * use mem_dump_obj().
567  */
kmem_dump_obj(void * object)568 void kmem_dump_obj(void *object)
569 {
570 	char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
571 	int i;
572 	struct slab *slab;
573 	unsigned long ptroffset;
574 	struct kmem_obj_info kp = { };
575 
576 	if (WARN_ON_ONCE(!virt_addr_valid(object)))
577 		return;
578 	slab = virt_to_slab(object);
579 	if (WARN_ON_ONCE(!slab)) {
580 		pr_cont(" non-slab memory.\n");
581 		return;
582 	}
583 	kmem_obj_info(&kp, object, slab);
584 	if (kp.kp_slab_cache)
585 		pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
586 	else
587 		pr_cont(" slab%s", cp);
588 	if (is_kfence_address(object))
589 		pr_cont(" (kfence)");
590 	if (kp.kp_objp)
591 		pr_cont(" start %px", kp.kp_objp);
592 	if (kp.kp_data_offset)
593 		pr_cont(" data offset %lu", kp.kp_data_offset);
594 	if (kp.kp_objp) {
595 		ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
596 		pr_cont(" pointer offset %lu", ptroffset);
597 	}
598 	if (kp.kp_slab_cache && kp.kp_slab_cache->usersize)
599 		pr_cont(" size %u", kp.kp_slab_cache->usersize);
600 	if (kp.kp_ret)
601 		pr_cont(" allocated at %pS\n", kp.kp_ret);
602 	else
603 		pr_cont("\n");
604 	for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
605 		if (!kp.kp_stack[i])
606 			break;
607 		pr_info("    %pS\n", kp.kp_stack[i]);
608 	}
609 
610 	if (kp.kp_free_stack[0])
611 		pr_cont(" Free path:\n");
612 
613 	for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
614 		if (!kp.kp_free_stack[i])
615 			break;
616 		pr_info("    %pS\n", kp.kp_free_stack[i]);
617 	}
618 
619 }
620 EXPORT_SYMBOL_GPL(kmem_dump_obj);
621 #endif
622 
623 #ifndef CONFIG_SLOB
624 /* 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)625 void __init create_boot_cache(struct kmem_cache *s, const char *name,
626 		unsigned int size, slab_flags_t flags,
627 		unsigned int useroffset, unsigned int usersize)
628 {
629 	int err;
630 	unsigned int align = ARCH_KMALLOC_MINALIGN;
631 
632 	s->name = name;
633 	s->size = s->object_size = size;
634 
635 	/*
636 	 * For power of two sizes, guarantee natural alignment for kmalloc
637 	 * caches, regardless of SL*B debugging options.
638 	 */
639 	if (is_power_of_2(size))
640 		align = max(align, size);
641 	s->align = calculate_alignment(flags, align, size);
642 
643 	s->useroffset = useroffset;
644 	s->usersize = usersize;
645 
646 	err = __kmem_cache_create(s, flags);
647 
648 	if (err)
649 		panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
650 					name, size, err);
651 
652 	s->refcount = -1;	/* Exempt from merging for now */
653 }
654 
create_kmalloc_cache(const char * name,unsigned int size,slab_flags_t flags,unsigned int useroffset,unsigned int usersize)655 struct kmem_cache *__init create_kmalloc_cache(const char *name,
656 		unsigned int size, slab_flags_t flags,
657 		unsigned int useroffset, unsigned int usersize)
658 {
659 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
660 
661 	if (!s)
662 		panic("Out of memory when creating slab %s\n", name);
663 
664 	create_boot_cache(s, name, size, flags | SLAB_KMALLOC, useroffset,
665 								usersize);
666 	kasan_cache_create_kmalloc(s);
667 	list_add(&s->list, &slab_caches);
668 	s->refcount = 1;
669 	return s;
670 }
671 
672 struct kmem_cache *
673 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
674 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
675 EXPORT_SYMBOL(kmalloc_caches);
676 
677 /*
678  * Conversion table for small slabs sizes / 8 to the index in the
679  * kmalloc array. This is necessary for slabs < 192 since we have non power
680  * of two cache sizes there. The size of larger slabs can be determined using
681  * fls.
682  */
683 static u8 size_index[24] __ro_after_init = {
684 	3,	/* 8 */
685 	4,	/* 16 */
686 	5,	/* 24 */
687 	5,	/* 32 */
688 	6,	/* 40 */
689 	6,	/* 48 */
690 	6,	/* 56 */
691 	6,	/* 64 */
692 	1,	/* 72 */
693 	1,	/* 80 */
694 	1,	/* 88 */
695 	1,	/* 96 */
696 	7,	/* 104 */
697 	7,	/* 112 */
698 	7,	/* 120 */
699 	7,	/* 128 */
700 	2,	/* 136 */
701 	2,	/* 144 */
702 	2,	/* 152 */
703 	2,	/* 160 */
704 	2,	/* 168 */
705 	2,	/* 176 */
706 	2,	/* 184 */
707 	2	/* 192 */
708 };
709 
size_index_elem(unsigned int bytes)710 static inline unsigned int size_index_elem(unsigned int bytes)
711 {
712 	return (bytes - 1) / 8;
713 }
714 
715 /*
716  * Find the kmem_cache structure that serves a given size of
717  * allocation
718  */
kmalloc_slab(size_t size,gfp_t flags)719 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags)
720 {
721 	unsigned int index;
722 
723 	if (size <= 192) {
724 		if (!size)
725 			return ZERO_SIZE_PTR;
726 
727 		index = size_index[size_index_elem(size)];
728 	} else {
729 		if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
730 			return NULL;
731 		index = fls(size - 1);
732 	}
733 
734 	return kmalloc_caches[kmalloc_type(flags)][index];
735 }
736 
kmalloc_size_roundup(size_t size)737 size_t kmalloc_size_roundup(size_t size)
738 {
739 	struct kmem_cache *c;
740 
741 	/* Short-circuit the 0 size case. */
742 	if (unlikely(size == 0))
743 		return 0;
744 	/* Short-circuit saturated "too-large" case. */
745 	if (unlikely(size == SIZE_MAX))
746 		return SIZE_MAX;
747 	/* Above the smaller buckets, size is a multiple of page size. */
748 	if (size > KMALLOC_MAX_CACHE_SIZE)
749 		return PAGE_SIZE << get_order(size);
750 
751 	/* The flags don't matter since size_index is common to all. */
752 	c = kmalloc_slab(size, GFP_KERNEL);
753 	return c ? c->object_size : 0;
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 #define INIT_KMALLOC_INFO(__size, __short_size)			\
770 {								\
771 	.name[KMALLOC_NORMAL]  = "kmalloc-" #__short_size,	\
772 	.name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size,	\
773 	KMALLOC_CGROUP_NAME(__short_size)			\
774 	KMALLOC_DMA_NAME(__short_size)				\
775 	.size = __size,						\
776 }
777 
778 /*
779  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
780  * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
781  * kmalloc-2M.
782  */
783 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
784 	INIT_KMALLOC_INFO(0, 0),
785 	INIT_KMALLOC_INFO(96, 96),
786 	INIT_KMALLOC_INFO(192, 192),
787 	INIT_KMALLOC_INFO(8, 8),
788 	INIT_KMALLOC_INFO(16, 16),
789 	INIT_KMALLOC_INFO(32, 32),
790 	INIT_KMALLOC_INFO(64, 64),
791 	INIT_KMALLOC_INFO(128, 128),
792 	INIT_KMALLOC_INFO(256, 256),
793 	INIT_KMALLOC_INFO(512, 512),
794 	INIT_KMALLOC_INFO(1024, 1k),
795 	INIT_KMALLOC_INFO(2048, 2k),
796 	INIT_KMALLOC_INFO(4096, 4k),
797 	INIT_KMALLOC_INFO(8192, 8k),
798 	INIT_KMALLOC_INFO(16384, 16k),
799 	INIT_KMALLOC_INFO(32768, 32k),
800 	INIT_KMALLOC_INFO(65536, 64k),
801 	INIT_KMALLOC_INFO(131072, 128k),
802 	INIT_KMALLOC_INFO(262144, 256k),
803 	INIT_KMALLOC_INFO(524288, 512k),
804 	INIT_KMALLOC_INFO(1048576, 1M),
805 	INIT_KMALLOC_INFO(2097152, 2M)
806 };
807 
808 /*
809  * Patch up the size_index table if we have strange large alignment
810  * requirements for the kmalloc array. This is only the case for
811  * MIPS it seems. The standard arches will not generate any code here.
812  *
813  * Largest permitted alignment is 256 bytes due to the way we
814  * handle the index determination for the smaller caches.
815  *
816  * Make sure that nothing crazy happens if someone starts tinkering
817  * around with ARCH_KMALLOC_MINALIGN
818  */
setup_kmalloc_cache_index_table(void)819 void __init setup_kmalloc_cache_index_table(void)
820 {
821 	unsigned int i;
822 
823 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
824 		!is_power_of_2(KMALLOC_MIN_SIZE));
825 
826 	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
827 		unsigned int elem = size_index_elem(i);
828 
829 		if (elem >= ARRAY_SIZE(size_index))
830 			break;
831 		size_index[elem] = KMALLOC_SHIFT_LOW;
832 	}
833 
834 	if (KMALLOC_MIN_SIZE >= 64) {
835 		/*
836 		 * The 96 byte sized cache is not used if the alignment
837 		 * is 64 byte.
838 		 */
839 		for (i = 64 + 8; i <= 96; i += 8)
840 			size_index[size_index_elem(i)] = 7;
841 
842 	}
843 
844 	if (KMALLOC_MIN_SIZE >= 128) {
845 		/*
846 		 * The 192 byte sized cache is not used if the alignment
847 		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
848 		 * instead.
849 		 */
850 		for (i = 128 + 8; i <= 192; i += 8)
851 			size_index[size_index_elem(i)] = 8;
852 	}
853 }
854 
855 static void __init
new_kmalloc_cache(int idx,enum kmalloc_cache_type type,slab_flags_t flags)856 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
857 {
858 	if (type == KMALLOC_RECLAIM) {
859 		flags |= SLAB_RECLAIM_ACCOUNT;
860 	} else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
861 		if (mem_cgroup_kmem_disabled()) {
862 			kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
863 			return;
864 		}
865 		flags |= SLAB_ACCOUNT;
866 	} else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
867 		flags |= SLAB_CACHE_DMA;
868 	}
869 
870 	kmalloc_caches[type][idx] = create_kmalloc_cache(
871 					kmalloc_info[idx].name[type],
872 					kmalloc_info[idx].size, flags, 0,
873 					kmalloc_info[idx].size);
874 
875 	/*
876 	 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
877 	 * KMALLOC_NORMAL caches.
878 	 */
879 	if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
880 		kmalloc_caches[type][idx]->refcount = -1;
881 }
882 
883 /*
884  * Create the kmalloc array. Some of the regular kmalloc arrays
885  * may already have been created because they were needed to
886  * enable allocations for slab creation.
887  */
create_kmalloc_caches(slab_flags_t flags)888 void __init create_kmalloc_caches(slab_flags_t flags)
889 {
890 	int i;
891 	enum kmalloc_cache_type type;
892 
893 	/*
894 	 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
895 	 */
896 	for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
897 		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
898 			if (!kmalloc_caches[type][i])
899 				new_kmalloc_cache(i, type, flags);
900 
901 			/*
902 			 * Caches that are not of the two-to-the-power-of size.
903 			 * These have to be created immediately after the
904 			 * earlier power of two caches
905 			 */
906 			if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
907 					!kmalloc_caches[type][1])
908 				new_kmalloc_cache(1, type, flags);
909 			if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
910 					!kmalloc_caches[type][2])
911 				new_kmalloc_cache(2, type, flags);
912 		}
913 	}
914 
915 	/* Kmalloc array is now usable */
916 	slab_state = UP;
917 }
918 
free_large_kmalloc(struct folio * folio,void * object)919 void free_large_kmalloc(struct folio *folio, void *object)
920 {
921 	unsigned int order = folio_order(folio);
922 
923 	if (WARN_ON_ONCE(order == 0))
924 		pr_warn_once("object pointer: 0x%p\n", object);
925 
926 	kmemleak_free(object);
927 	kasan_kfree_large(object);
928 	kmsan_kfree_large(object);
929 
930 	mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
931 			      -(PAGE_SIZE << order));
932 	__free_pages(folio_page(folio, 0), order);
933 }
934 
935 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
936 static __always_inline
__do_kmalloc_node(size_t size,gfp_t flags,int node,unsigned long caller)937 void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
938 {
939 	struct kmem_cache *s;
940 	void *ret;
941 
942 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
943 		ret = __kmalloc_large_node(size, flags, node);
944 		trace_kmalloc(caller, ret, size,
945 			      PAGE_SIZE << get_order(size), flags, node);
946 		return ret;
947 	}
948 
949 	s = kmalloc_slab(size, flags);
950 
951 	if (unlikely(ZERO_OR_NULL_PTR(s)))
952 		return s;
953 
954 	ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
955 	ret = kasan_kmalloc(s, ret, size, flags);
956 	trace_kmalloc(caller, ret, size, s->size, flags, node);
957 	return ret;
958 }
959 
__kmalloc_node(size_t size,gfp_t flags,int node)960 void *__kmalloc_node(size_t size, gfp_t flags, int node)
961 {
962 	return __do_kmalloc_node(size, flags, node, _RET_IP_);
963 }
964 EXPORT_SYMBOL(__kmalloc_node);
965 
__kmalloc(size_t size,gfp_t flags)966 void *__kmalloc(size_t size, gfp_t flags)
967 {
968 	return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
969 }
970 EXPORT_SYMBOL(__kmalloc);
971 
__kmalloc_node_track_caller(size_t size,gfp_t flags,int node,unsigned long caller)972 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
973 				  int node, unsigned long caller)
974 {
975 	return __do_kmalloc_node(size, flags, node, caller);
976 }
977 EXPORT_SYMBOL(__kmalloc_node_track_caller);
978 
979 /**
980  * kfree - free previously allocated memory
981  * @object: pointer returned by kmalloc.
982  *
983  * If @object is NULL, no operation is performed.
984  *
985  * Don't free memory not originally allocated by kmalloc()
986  * or you will run into trouble.
987  */
kfree(const void * object)988 void kfree(const void *object)
989 {
990 	struct folio *folio;
991 	struct slab *slab;
992 	struct kmem_cache *s;
993 
994 	trace_kfree(_RET_IP_, object);
995 
996 	if (unlikely(ZERO_OR_NULL_PTR(object)))
997 		return;
998 
999 	folio = virt_to_folio(object);
1000 	if (unlikely(!folio_test_slab(folio))) {
1001 		free_large_kmalloc(folio, (void *)object);
1002 		return;
1003 	}
1004 
1005 	slab = folio_slab(folio);
1006 	s = slab->slab_cache;
1007 	__kmem_cache_free(s, (void *)object, _RET_IP_);
1008 }
1009 EXPORT_SYMBOL(kfree);
1010 
1011 /**
1012  * __ksize -- Report full size of underlying allocation
1013  * @object: pointer to the object
1014  *
1015  * This should only be used internally to query the true size of allocations.
1016  * It is not meant to be a way to discover the usable size of an allocation
1017  * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1018  * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1019  * and/or FORTIFY_SOURCE.
1020  *
1021  * Return: size of the actual memory used by @object in bytes
1022  */
__ksize(const void * object)1023 size_t __ksize(const void *object)
1024 {
1025 	struct folio *folio;
1026 
1027 	if (unlikely(object == ZERO_SIZE_PTR))
1028 		return 0;
1029 
1030 	folio = virt_to_folio(object);
1031 
1032 	if (unlikely(!folio_test_slab(folio))) {
1033 		if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1034 			return 0;
1035 		if (WARN_ON(object != folio_address(folio)))
1036 			return 0;
1037 		return folio_size(folio);
1038 	}
1039 
1040 	return slab_ksize(folio_slab(folio)->slab_cache);
1041 }
1042 
kmalloc_trace(struct kmem_cache * s,gfp_t gfpflags,size_t size)1043 void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1044 {
1045 	void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1046 					    size, _RET_IP_);
1047 
1048 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1049 
1050 	ret = kasan_kmalloc(s, ret, size, gfpflags);
1051 	return ret;
1052 }
1053 EXPORT_SYMBOL(kmalloc_trace);
1054 
kmalloc_node_trace(struct kmem_cache * s,gfp_t gfpflags,int node,size_t size)1055 void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1056 			 int node, size_t size)
1057 {
1058 	void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1059 
1060 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1061 
1062 	ret = kasan_kmalloc(s, ret, size, gfpflags);
1063 	return ret;
1064 }
1065 EXPORT_SYMBOL(kmalloc_node_trace);
1066 #endif /* !CONFIG_SLOB */
1067 
kmalloc_fix_flags(gfp_t flags)1068 gfp_t kmalloc_fix_flags(gfp_t flags)
1069 {
1070 	gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1071 
1072 	flags &= ~GFP_SLAB_BUG_MASK;
1073 	pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1074 			invalid_mask, &invalid_mask, flags, &flags);
1075 	dump_stack();
1076 
1077 	return flags;
1078 }
1079 
1080 /*
1081  * To avoid unnecessary overhead, we pass through large allocation requests
1082  * directly to the page allocator. We use __GFP_COMP, because we will need to
1083  * know the allocation order to free the pages properly in kfree.
1084  */
1085 
__kmalloc_large_node(size_t size,gfp_t flags,int node)1086 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1087 {
1088 	struct page *page;
1089 	void *ptr = NULL;
1090 	unsigned int order = get_order(size);
1091 
1092 	if (unlikely(flags & GFP_SLAB_BUG_MASK))
1093 		flags = kmalloc_fix_flags(flags);
1094 
1095 	flags |= __GFP_COMP;
1096 	page = alloc_pages_node(node, flags, order);
1097 	if (page) {
1098 		ptr = page_address(page);
1099 		mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1100 				      PAGE_SIZE << order);
1101 	}
1102 
1103 	ptr = kasan_kmalloc_large(ptr, size, flags);
1104 	/* As ptr might get tagged, call kmemleak hook after KASAN. */
1105 	kmemleak_alloc(ptr, size, 1, flags);
1106 	kmsan_kmalloc_large(ptr, size, flags);
1107 
1108 	return ptr;
1109 }
1110 
kmalloc_large(size_t size,gfp_t flags)1111 void *kmalloc_large(size_t size, gfp_t flags)
1112 {
1113 	void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1114 
1115 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1116 		      flags, NUMA_NO_NODE);
1117 	return ret;
1118 }
1119 EXPORT_SYMBOL(kmalloc_large);
1120 
kmalloc_large_node(size_t size,gfp_t flags,int node)1121 void *kmalloc_large_node(size_t size, gfp_t flags, int node)
1122 {
1123 	void *ret = __kmalloc_large_node(size, flags, node);
1124 
1125 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1126 		      flags, node);
1127 	return ret;
1128 }
1129 EXPORT_SYMBOL(kmalloc_large_node);
1130 
1131 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1132 /* Randomize a generic freelist */
freelist_randomize(struct rnd_state * state,unsigned int * list,unsigned int count)1133 static void freelist_randomize(struct rnd_state *state, unsigned int *list,
1134 			       unsigned int count)
1135 {
1136 	unsigned int rand;
1137 	unsigned int i;
1138 
1139 	for (i = 0; i < count; i++)
1140 		list[i] = i;
1141 
1142 	/* Fisher-Yates shuffle */
1143 	for (i = count - 1; i > 0; i--) {
1144 		rand = prandom_u32_state(state);
1145 		rand %= (i + 1);
1146 		swap(list[i], list[rand]);
1147 	}
1148 }
1149 
1150 /* Create a random sequence per cache */
cache_random_seq_create(struct kmem_cache * cachep,unsigned int count,gfp_t gfp)1151 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1152 				    gfp_t gfp)
1153 {
1154 	struct rnd_state state;
1155 
1156 	if (count < 2 || cachep->random_seq)
1157 		return 0;
1158 
1159 	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1160 	if (!cachep->random_seq)
1161 		return -ENOMEM;
1162 
1163 	/* Get best entropy at this stage of boot */
1164 	prandom_seed_state(&state, get_random_long());
1165 
1166 	freelist_randomize(&state, cachep->random_seq, count);
1167 	return 0;
1168 }
1169 
1170 /* Destroy the per-cache random freelist sequence */
cache_random_seq_destroy(struct kmem_cache * cachep)1171 void cache_random_seq_destroy(struct kmem_cache *cachep)
1172 {
1173 	kfree(cachep->random_seq);
1174 	cachep->random_seq = NULL;
1175 }
1176 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1177 
1178 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1179 #ifdef CONFIG_SLAB
1180 #define SLABINFO_RIGHTS (0600)
1181 #else
1182 #define SLABINFO_RIGHTS (0400)
1183 #endif
1184 
print_slabinfo_header(struct seq_file * m)1185 static void print_slabinfo_header(struct seq_file *m)
1186 {
1187 	/*
1188 	 * Output format version, so at least we can change it
1189 	 * without _too_ many complaints.
1190 	 */
1191 #ifdef CONFIG_DEBUG_SLAB
1192 	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1193 #else
1194 	seq_puts(m, "slabinfo - version: 2.1\n");
1195 #endif
1196 	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1197 	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1198 	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1199 #ifdef CONFIG_DEBUG_SLAB
1200 	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1201 	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1202 #endif
1203 	seq_putc(m, '\n');
1204 }
1205 
slab_start(struct seq_file * m,loff_t * pos)1206 static void *slab_start(struct seq_file *m, loff_t *pos)
1207 {
1208 	mutex_lock(&slab_mutex);
1209 	return seq_list_start(&slab_caches, *pos);
1210 }
1211 
slab_next(struct seq_file * m,void * p,loff_t * pos)1212 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1213 {
1214 	return seq_list_next(p, &slab_caches, pos);
1215 }
1216 
slab_stop(struct seq_file * m,void * p)1217 static void slab_stop(struct seq_file *m, void *p)
1218 {
1219 	mutex_unlock(&slab_mutex);
1220 }
1221 
cache_show(struct kmem_cache * s,struct seq_file * m)1222 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1223 {
1224 	struct slabinfo sinfo;
1225 
1226 	memset(&sinfo, 0, sizeof(sinfo));
1227 	get_slabinfo(s, &sinfo);
1228 
1229 	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1230 		   s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1231 		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1232 
1233 	seq_printf(m, " : tunables %4u %4u %4u",
1234 		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1235 	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1236 		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1237 	slabinfo_show_stats(m, s);
1238 	seq_putc(m, '\n');
1239 }
1240 
slab_show(struct seq_file * m,void * p)1241 static int slab_show(struct seq_file *m, void *p)
1242 {
1243 	struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1244 
1245 	if (p == slab_caches.next)
1246 		print_slabinfo_header(m);
1247 	cache_show(s, m);
1248 	return 0;
1249 }
1250 
dump_unreclaimable_slab(void)1251 void dump_unreclaimable_slab(void)
1252 {
1253 	struct kmem_cache *s;
1254 	struct slabinfo sinfo;
1255 
1256 	/*
1257 	 * Here acquiring slab_mutex is risky since we don't prefer to get
1258 	 * sleep in oom path. But, without mutex hold, it may introduce a
1259 	 * risk of crash.
1260 	 * Use mutex_trylock to protect the list traverse, dump nothing
1261 	 * without acquiring the mutex.
1262 	 */
1263 	if (!mutex_trylock(&slab_mutex)) {
1264 		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1265 		return;
1266 	}
1267 
1268 	pr_info("Unreclaimable slab info:\n");
1269 	pr_info("Name                      Used          Total\n");
1270 
1271 	list_for_each_entry(s, &slab_caches, list) {
1272 		if (s->flags & SLAB_RECLAIM_ACCOUNT)
1273 			continue;
1274 
1275 		get_slabinfo(s, &sinfo);
1276 
1277 		if (sinfo.num_objs > 0)
1278 			pr_info("%-17s %10luKB %10luKB\n", s->name,
1279 				(sinfo.active_objs * s->size) / 1024,
1280 				(sinfo.num_objs * s->size) / 1024);
1281 	}
1282 	mutex_unlock(&slab_mutex);
1283 }
1284 
1285 /*
1286  * slabinfo_op - iterator that generates /proc/slabinfo
1287  *
1288  * Output layout:
1289  * cache-name
1290  * num-active-objs
1291  * total-objs
1292  * object size
1293  * num-active-slabs
1294  * total-slabs
1295  * num-pages-per-slab
1296  * + further values on SMP and with statistics enabled
1297  */
1298 static const struct seq_operations slabinfo_op = {
1299 	.start = slab_start,
1300 	.next = slab_next,
1301 	.stop = slab_stop,
1302 	.show = slab_show,
1303 };
1304 
slabinfo_open(struct inode * inode,struct file * file)1305 static int slabinfo_open(struct inode *inode, struct file *file)
1306 {
1307 	return seq_open(file, &slabinfo_op);
1308 }
1309 
1310 static const struct proc_ops slabinfo_proc_ops = {
1311 	.proc_flags	= PROC_ENTRY_PERMANENT,
1312 	.proc_open	= slabinfo_open,
1313 	.proc_read	= seq_read,
1314 	.proc_write	= slabinfo_write,
1315 	.proc_lseek	= seq_lseek,
1316 	.proc_release	= seq_release,
1317 };
1318 
slab_proc_init(void)1319 static int __init slab_proc_init(void)
1320 {
1321 	proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1322 	return 0;
1323 }
1324 module_init(slab_proc_init);
1325 
1326 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1327 
1328 static __always_inline __realloc_size(2) void *
__do_krealloc(const void * p,size_t new_size,gfp_t flags)1329 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
1330 {
1331 	void *ret;
1332 	size_t ks;
1333 
1334 	/* Don't use instrumented ksize to allow precise KASAN poisoning. */
1335 	if (likely(!ZERO_OR_NULL_PTR(p))) {
1336 		if (!kasan_check_byte(p))
1337 			return NULL;
1338 		ks = kfence_ksize(p) ?: __ksize(p);
1339 	} else
1340 		ks = 0;
1341 
1342 	/* If the object still fits, repoison it precisely. */
1343 	if (ks >= new_size) {
1344 		p = kasan_krealloc((void *)p, new_size, flags);
1345 		return (void *)p;
1346 	}
1347 
1348 	ret = kmalloc_track_caller(new_size, flags);
1349 	if (ret && p) {
1350 		/* Disable KASAN checks as the object's redzone is accessed. */
1351 		kasan_disable_current();
1352 		memcpy(ret, kasan_reset_tag(p), ks);
1353 		kasan_enable_current();
1354 	}
1355 
1356 	return ret;
1357 }
1358 
1359 /**
1360  * krealloc - reallocate memory. The contents will remain unchanged.
1361  * @p: object to reallocate memory for.
1362  * @new_size: how many bytes of memory are required.
1363  * @flags: the type of memory to allocate.
1364  *
1365  * The contents of the object pointed to are preserved up to the
1366  * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1367  * If @p is %NULL, krealloc() behaves exactly like kmalloc().  If @new_size
1368  * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1369  *
1370  * Return: pointer to the allocated memory or %NULL in case of error
1371  */
krealloc(const void * p,size_t new_size,gfp_t flags)1372 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1373 {
1374 	void *ret;
1375 
1376 	if (unlikely(!new_size)) {
1377 		kfree(p);
1378 		return ZERO_SIZE_PTR;
1379 	}
1380 
1381 	ret = __do_krealloc(p, new_size, flags);
1382 	if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1383 		kfree(p);
1384 
1385 	return ret;
1386 }
1387 EXPORT_SYMBOL(krealloc);
1388 
1389 /**
1390  * kfree_sensitive - Clear sensitive information in memory before freeing
1391  * @p: object to free memory of
1392  *
1393  * The memory of the object @p points to is zeroed before freed.
1394  * If @p is %NULL, kfree_sensitive() does nothing.
1395  *
1396  * Note: this function zeroes the whole allocated buffer which can be a good
1397  * deal bigger than the requested buffer size passed to kmalloc(). So be
1398  * careful when using this function in performance sensitive code.
1399  */
kfree_sensitive(const void * p)1400 void kfree_sensitive(const void *p)
1401 {
1402 	size_t ks;
1403 	void *mem = (void *)p;
1404 
1405 	ks = ksize(mem);
1406 	if (ks)
1407 		memzero_explicit(mem, ks);
1408 	kfree(mem);
1409 }
1410 EXPORT_SYMBOL(kfree_sensitive);
1411 
ksize(const void * objp)1412 size_t ksize(const void *objp)
1413 {
1414 	size_t size;
1415 
1416 	/*
1417 	 * We need to first check that the pointer to the object is valid, and
1418 	 * only then unpoison the memory. The report printed from ksize() is
1419 	 * more useful, then when it's printed later when the behaviour could
1420 	 * be undefined due to a potential use-after-free or double-free.
1421 	 *
1422 	 * We use kasan_check_byte(), which is supported for the hardware
1423 	 * tag-based KASAN mode, unlike kasan_check_read/write().
1424 	 *
1425 	 * If the pointed to memory is invalid, we return 0 to avoid users of
1426 	 * ksize() writing to and potentially corrupting the memory region.
1427 	 *
1428 	 * We want to perform the check before __ksize(), to avoid potentially
1429 	 * crashing in __ksize() due to accessing invalid metadata.
1430 	 */
1431 	if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1432 		return 0;
1433 
1434 	size = kfence_ksize(objp) ?: __ksize(objp);
1435 	/*
1436 	 * We assume that ksize callers could use whole allocated area,
1437 	 * so we need to unpoison this area.
1438 	 */
1439 	kasan_unpoison_range(objp, size);
1440 	return size;
1441 }
1442 EXPORT_SYMBOL(ksize);
1443 
1444 /* Tracepoints definitions. */
1445 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1446 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1447 EXPORT_TRACEPOINT_SYMBOL(kfree);
1448 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1449 
should_failslab(struct kmem_cache * s,gfp_t gfpflags)1450 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1451 {
1452 	if (__should_failslab(s, gfpflags))
1453 		return -ENOMEM;
1454 	return 0;
1455 }
1456 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
1457