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 
37 enum slab_state slab_state;
38 LIST_HEAD(slab_caches);
39 DEFINE_MUTEX(slab_mutex);
40 struct kmem_cache *kmem_cache;
41 
42 static LIST_HEAD(slab_caches_to_rcu_destroy);
43 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work);
44 static DECLARE_WORK(slab_caches_to_rcu_destroy_work,
45 		    slab_caches_to_rcu_destroy_workfn);
46 
47 /*
48  * Set of flags that will prevent slab merging
49  */
50 #define SLAB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
51 		SLAB_TRACE | SLAB_TYPESAFE_BY_RCU | SLAB_NOLEAKTRACE | \
52 		SLAB_FAILSLAB | SLAB_NO_MERGE | kasan_never_merge())
53 
54 #define SLAB_MERGE_SAME (SLAB_RECLAIM_ACCOUNT | SLAB_CACHE_DMA | \
55 			 SLAB_CACHE_DMA32 | SLAB_ACCOUNT)
56 
57 /*
58  * Merge control. If this is set then no merging of slab caches will occur.
59  */
60 static bool slab_nomerge = !IS_ENABLED(CONFIG_SLAB_MERGE_DEFAULT);
61 
setup_slab_nomerge(char * str)62 static int __init setup_slab_nomerge(char *str)
63 {
64 	slab_nomerge = true;
65 	return 1;
66 }
67 
setup_slab_merge(char * str)68 static int __init setup_slab_merge(char *str)
69 {
70 	slab_nomerge = false;
71 	return 1;
72 }
73 
74 #ifdef CONFIG_SLUB
75 __setup_param("slub_nomerge", slub_nomerge, setup_slab_nomerge, 0);
76 __setup_param("slub_merge", slub_merge, setup_slab_merge, 0);
77 #endif
78 
79 __setup("slab_nomerge", setup_slab_nomerge);
80 __setup("slab_merge", setup_slab_merge);
81 
82 /*
83  * Determine the size of a slab object
84  */
kmem_cache_size(struct kmem_cache * s)85 unsigned int kmem_cache_size(struct kmem_cache *s)
86 {
87 	return s->object_size;
88 }
89 EXPORT_SYMBOL(kmem_cache_size);
90 
91 #ifdef CONFIG_DEBUG_VM
kmem_cache_sanity_check(const char * name,unsigned int size)92 static int kmem_cache_sanity_check(const char *name, unsigned int size)
93 {
94 	if (!name || in_interrupt() || size > KMALLOC_MAX_SIZE) {
95 		pr_err("kmem_cache_create(%s) integrity check failed\n", name);
96 		return -EINVAL;
97 	}
98 
99 	WARN_ON(strchr(name, ' '));	/* It confuses parsers */
100 	return 0;
101 }
102 #else
kmem_cache_sanity_check(const char * name,unsigned int size)103 static inline int kmem_cache_sanity_check(const char *name, unsigned int size)
104 {
105 	return 0;
106 }
107 #endif
108 
109 /*
110  * Figure out what the alignment of the objects will be given a set of
111  * flags, a user specified alignment and the size of the objects.
112  */
calculate_alignment(slab_flags_t flags,unsigned int align,unsigned int size)113 static unsigned int calculate_alignment(slab_flags_t flags,
114 		unsigned int align, unsigned int size)
115 {
116 	/*
117 	 * If the user wants hardware cache aligned objects then follow that
118 	 * suggestion if the object is sufficiently large.
119 	 *
120 	 * The hardware cache alignment cannot override the specified
121 	 * alignment though. If that is greater then use it.
122 	 */
123 	if (flags & SLAB_HWCACHE_ALIGN) {
124 		unsigned int ralign;
125 
126 		ralign = cache_line_size();
127 		while (size <= ralign / 2)
128 			ralign /= 2;
129 		align = max(align, ralign);
130 	}
131 
132 	align = max(align, arch_slab_minalign());
133 
134 	return ALIGN(align, sizeof(void *));
135 }
136 
137 /*
138  * Find a mergeable slab cache
139  */
slab_unmergeable(struct kmem_cache * s)140 int slab_unmergeable(struct kmem_cache *s)
141 {
142 	if (slab_nomerge || (s->flags & SLAB_NEVER_MERGE))
143 		return 1;
144 
145 	if (s->ctor)
146 		return 1;
147 
148 #ifdef CONFIG_HARDENED_USERCOPY
149 	if (s->usersize)
150 		return 1;
151 #endif
152 
153 	/*
154 	 * We may have set a slab to be unmergeable during bootstrap.
155 	 */
156 	if (s->refcount < 0)
157 		return 1;
158 
159 	return 0;
160 }
161 
find_mergeable(unsigned int size,unsigned int align,slab_flags_t flags,const char * name,void (* ctor)(void *))162 struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
163 		slab_flags_t flags, const char *name, void (*ctor)(void *))
164 {
165 	struct kmem_cache *s;
166 
167 	if (slab_nomerge)
168 		return NULL;
169 
170 	if (ctor)
171 		return NULL;
172 
173 	size = ALIGN(size, sizeof(void *));
174 	align = calculate_alignment(flags, align, size);
175 	size = ALIGN(size, align);
176 	flags = kmem_cache_flags(size, flags, name);
177 
178 	if (flags & SLAB_NEVER_MERGE)
179 		return NULL;
180 
181 	list_for_each_entry_reverse(s, &slab_caches, list) {
182 		if (slab_unmergeable(s))
183 			continue;
184 
185 		if (size > s->size)
186 			continue;
187 
188 		if ((flags & SLAB_MERGE_SAME) != (s->flags & SLAB_MERGE_SAME))
189 			continue;
190 		/*
191 		 * Check if alignment is compatible.
192 		 * Courtesy of Adrian Drzewiecki
193 		 */
194 		if ((s->size & ~(align - 1)) != s->size)
195 			continue;
196 
197 		if (s->size - size >= sizeof(void *))
198 			continue;
199 
200 		if (IS_ENABLED(CONFIG_SLAB) && align &&
201 			(align > s->align || s->align % align))
202 			continue;
203 
204 		return s;
205 	}
206 	return NULL;
207 }
208 
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)209 static struct kmem_cache *create_cache(const char *name,
210 		unsigned int object_size, unsigned int align,
211 		slab_flags_t flags, unsigned int useroffset,
212 		unsigned int usersize, void (*ctor)(void *),
213 		struct kmem_cache *root_cache)
214 {
215 	struct kmem_cache *s;
216 	int err;
217 
218 	if (WARN_ON(useroffset + usersize > object_size))
219 		useroffset = usersize = 0;
220 
221 	err = -ENOMEM;
222 	s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL);
223 	if (!s)
224 		goto out;
225 
226 	s->name = name;
227 	s->size = s->object_size = object_size;
228 	s->align = align;
229 	s->ctor = ctor;
230 #ifdef CONFIG_HARDENED_USERCOPY
231 	s->useroffset = useroffset;
232 	s->usersize = usersize;
233 #endif
234 
235 	err = __kmem_cache_create(s, flags);
236 	if (err)
237 		goto out_free_cache;
238 
239 	s->refcount = 1;
240 	list_add(&s->list, &slab_caches);
241 	return s;
242 
243 out_free_cache:
244 	kmem_cache_free(kmem_cache, s);
245 out:
246 	return ERR_PTR(err);
247 }
248 
249 /**
250  * kmem_cache_create_usercopy - Create a cache with a region suitable
251  * for copying to userspace
252  * @name: A string which is used in /proc/slabinfo to identify this cache.
253  * @size: The size of objects to be created in this cache.
254  * @align: The required alignment for the objects.
255  * @flags: SLAB flags
256  * @useroffset: Usercopy region offset
257  * @usersize: Usercopy region size
258  * @ctor: A constructor for the objects.
259  *
260  * Cannot be called within a interrupt, but can be interrupted.
261  * The @ctor is run when new pages are allocated by the cache.
262  *
263  * The flags are
264  *
265  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
266  * to catch references to uninitialised memory.
267  *
268  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
269  * for buffer overruns.
270  *
271  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
272  * cacheline.  This can be beneficial if you're counting cycles as closely
273  * as davem.
274  *
275  * Return: a pointer to the cache on success, NULL on failure.
276  */
277 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 *))278 kmem_cache_create_usercopy(const char *name,
279 		  unsigned int size, unsigned int align,
280 		  slab_flags_t flags,
281 		  unsigned int useroffset, unsigned int usersize,
282 		  void (*ctor)(void *))
283 {
284 	struct kmem_cache *s = NULL;
285 	const char *cache_name;
286 	int err;
287 
288 #ifdef CONFIG_SLUB_DEBUG
289 	/*
290 	 * If no slub_debug was enabled globally, the static key is not yet
291 	 * enabled by setup_slub_debug(). Enable it if the cache is being
292 	 * created with any of the debugging flags passed explicitly.
293 	 * It's also possible that this is the first cache created with
294 	 * SLAB_STORE_USER and we should init stack_depot for it.
295 	 */
296 	if (flags & SLAB_DEBUG_FLAGS)
297 		static_branch_enable(&slub_debug_enabled);
298 	if (flags & SLAB_STORE_USER)
299 		stack_depot_init();
300 #endif
301 
302 	mutex_lock(&slab_mutex);
303 
304 	err = kmem_cache_sanity_check(name, size);
305 	if (err) {
306 		goto out_unlock;
307 	}
308 
309 	/* Refuse requests with allocator specific flags */
310 	if (flags & ~SLAB_FLAGS_PERMITTED) {
311 		err = -EINVAL;
312 		goto out_unlock;
313 	}
314 
315 	/*
316 	 * Some allocators will constraint the set of valid flags to a subset
317 	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
318 	 * case, and we'll just provide them with a sanitized version of the
319 	 * passed flags.
320 	 */
321 	flags &= CACHE_CREATE_MASK;
322 
323 	/* Fail closed on bad usersize of useroffset values. */
324 	if (!IS_ENABLED(CONFIG_HARDENED_USERCOPY) ||
325 	    WARN_ON(!usersize && useroffset) ||
326 	    WARN_ON(size < usersize || size - usersize < useroffset))
327 		usersize = useroffset = 0;
328 
329 	if (!usersize)
330 		s = __kmem_cache_alias(name, size, align, flags, ctor);
331 	if (s)
332 		goto out_unlock;
333 
334 	cache_name = kstrdup_const(name, GFP_KERNEL);
335 	if (!cache_name) {
336 		err = -ENOMEM;
337 		goto out_unlock;
338 	}
339 
340 	s = create_cache(cache_name, size,
341 			 calculate_alignment(flags, align, size),
342 			 flags, useroffset, usersize, ctor, NULL);
343 	if (IS_ERR(s)) {
344 		err = PTR_ERR(s);
345 		kfree_const(cache_name);
346 	}
347 
348 out_unlock:
349 	mutex_unlock(&slab_mutex);
350 
351 	if (err) {
352 		if (flags & SLAB_PANIC)
353 			panic("%s: Failed to create slab '%s'. Error %d\n",
354 				__func__, name, err);
355 		else {
356 			pr_warn("%s(%s) failed with error %d\n",
357 				__func__, name, err);
358 			dump_stack();
359 		}
360 		return NULL;
361 	}
362 	return s;
363 }
364 EXPORT_SYMBOL(kmem_cache_create_usercopy);
365 
366 /**
367  * kmem_cache_create - Create a cache.
368  * @name: A string which is used in /proc/slabinfo to identify this cache.
369  * @size: The size of objects to be created in this cache.
370  * @align: The required alignment for the objects.
371  * @flags: SLAB flags
372  * @ctor: A constructor for the objects.
373  *
374  * Cannot be called within a interrupt, but can be interrupted.
375  * The @ctor is run when new pages are allocated by the cache.
376  *
377  * The flags are
378  *
379  * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
380  * to catch references to uninitialised memory.
381  *
382  * %SLAB_RED_ZONE - Insert `Red` zones around the allocated memory to check
383  * for buffer overruns.
384  *
385  * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
386  * cacheline.  This can be beneficial if you're counting cycles as closely
387  * as davem.
388  *
389  * Return: a pointer to the cache on success, NULL on failure.
390  */
391 struct kmem_cache *
kmem_cache_create(const char * name,unsigned int size,unsigned int align,slab_flags_t flags,void (* ctor)(void *))392 kmem_cache_create(const char *name, unsigned int size, unsigned int align,
393 		slab_flags_t flags, void (*ctor)(void *))
394 {
395 	return kmem_cache_create_usercopy(name, size, align, flags, 0, 0,
396 					  ctor);
397 }
398 EXPORT_SYMBOL(kmem_cache_create);
399 
400 #ifdef SLAB_SUPPORTS_SYSFS
401 /*
402  * For a given kmem_cache, kmem_cache_destroy() should only be called
403  * once or there will be a use-after-free problem. The actual deletion
404  * and release of the kobject does not need slab_mutex or cpu_hotplug_lock
405  * protection. So they are now done without holding those locks.
406  *
407  * Note that there will be a slight delay in the deletion of sysfs files
408  * if kmem_cache_release() is called indrectly from a work function.
409  */
kmem_cache_release(struct kmem_cache * s)410 static void kmem_cache_release(struct kmem_cache *s)
411 {
412 	sysfs_slab_unlink(s);
413 	sysfs_slab_release(s);
414 }
415 #else
kmem_cache_release(struct kmem_cache * s)416 static void kmem_cache_release(struct kmem_cache *s)
417 {
418 	slab_kmem_cache_release(s);
419 }
420 #endif
421 
slab_caches_to_rcu_destroy_workfn(struct work_struct * work)422 static void slab_caches_to_rcu_destroy_workfn(struct work_struct *work)
423 {
424 	LIST_HEAD(to_destroy);
425 	struct kmem_cache *s, *s2;
426 
427 	/*
428 	 * On destruction, SLAB_TYPESAFE_BY_RCU kmem_caches are put on the
429 	 * @slab_caches_to_rcu_destroy list.  The slab pages are freed
430 	 * through RCU and the associated kmem_cache are dereferenced
431 	 * while freeing the pages, so the kmem_caches should be freed only
432 	 * after the pending RCU operations are finished.  As rcu_barrier()
433 	 * is a pretty slow operation, we batch all pending destructions
434 	 * asynchronously.
435 	 */
436 	mutex_lock(&slab_mutex);
437 	list_splice_init(&slab_caches_to_rcu_destroy, &to_destroy);
438 	mutex_unlock(&slab_mutex);
439 
440 	if (list_empty(&to_destroy))
441 		return;
442 
443 	rcu_barrier();
444 
445 	list_for_each_entry_safe(s, s2, &to_destroy, list) {
446 		debugfs_slab_release(s);
447 		kfence_shutdown_cache(s);
448 		kmem_cache_release(s);
449 	}
450 }
451 
shutdown_cache(struct kmem_cache * s)452 static int shutdown_cache(struct kmem_cache *s)
453 {
454 	/* free asan quarantined objects */
455 	kasan_cache_shutdown(s);
456 
457 	if (__kmem_cache_shutdown(s) != 0)
458 		return -EBUSY;
459 
460 	list_del(&s->list);
461 
462 	if (s->flags & SLAB_TYPESAFE_BY_RCU) {
463 		list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
464 		schedule_work(&slab_caches_to_rcu_destroy_work);
465 	} else {
466 		kfence_shutdown_cache(s);
467 		debugfs_slab_release(s);
468 	}
469 
470 	return 0;
471 }
472 
slab_kmem_cache_release(struct kmem_cache * s)473 void slab_kmem_cache_release(struct kmem_cache *s)
474 {
475 	__kmem_cache_release(s);
476 	kfree_const(s->name);
477 	kmem_cache_free(kmem_cache, s);
478 }
479 
kmem_cache_destroy(struct kmem_cache * s)480 void kmem_cache_destroy(struct kmem_cache *s)
481 {
482 	int err = -EBUSY;
483 	bool rcu_set;
484 
485 	if (unlikely(!s) || !kasan_check_byte(s))
486 		return;
487 
488 	cpus_read_lock();
489 	mutex_lock(&slab_mutex);
490 
491 	rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
492 
493 	s->refcount--;
494 	if (s->refcount)
495 		goto out_unlock;
496 
497 	err = shutdown_cache(s);
498 	WARN(err, "%s %s: Slab cache still has objects when called from %pS",
499 	     __func__, s->name, (void *)_RET_IP_);
500 out_unlock:
501 	mutex_unlock(&slab_mutex);
502 	cpus_read_unlock();
503 	if (!err && !rcu_set)
504 		kmem_cache_release(s);
505 }
506 EXPORT_SYMBOL(kmem_cache_destroy);
507 
508 /**
509  * kmem_cache_shrink - Shrink a cache.
510  * @cachep: The cache to shrink.
511  *
512  * Releases as many slabs as possible for a cache.
513  * To help debugging, a zero exit status indicates all slabs were released.
514  *
515  * Return: %0 if all slabs were released, non-zero otherwise
516  */
kmem_cache_shrink(struct kmem_cache * cachep)517 int kmem_cache_shrink(struct kmem_cache *cachep)
518 {
519 	kasan_cache_shrink(cachep);
520 
521 	return __kmem_cache_shrink(cachep);
522 }
523 EXPORT_SYMBOL(kmem_cache_shrink);
524 
slab_is_available(void)525 bool slab_is_available(void)
526 {
527 	return slab_state >= UP;
528 }
529 
530 #ifdef CONFIG_PRINTK
531 /**
532  * kmem_valid_obj - does the pointer reference a valid slab object?
533  * @object: pointer to query.
534  *
535  * Return: %true if the pointer is to a not-yet-freed object from
536  * kmalloc() or kmem_cache_alloc(), either %true or %false if the pointer
537  * is to an already-freed object, and %false otherwise.
538  */
kmem_valid_obj(void * object)539 bool kmem_valid_obj(void *object)
540 {
541 	struct folio *folio;
542 
543 	/* Some arches consider ZERO_SIZE_PTR to be a valid address. */
544 	if (object < (void *)PAGE_SIZE || !virt_addr_valid(object))
545 		return false;
546 	folio = virt_to_folio(object);
547 	return folio_test_slab(folio);
548 }
549 EXPORT_SYMBOL_GPL(kmem_valid_obj);
550 
kmem_obj_info(struct kmem_obj_info * kpp,void * object,struct slab * slab)551 static void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab)
552 {
553 	if (__kfence_obj_info(kpp, object, slab))
554 		return;
555 	__kmem_obj_info(kpp, object, slab);
556 }
557 
558 /**
559  * kmem_dump_obj - Print available slab provenance information
560  * @object: slab object for which to find provenance information.
561  *
562  * This function uses pr_cont(), so that the caller is expected to have
563  * printed out whatever preamble is appropriate.  The provenance information
564  * depends on the type of object and on how much debugging is enabled.
565  * For a slab-cache object, the fact that it is a slab object is printed,
566  * and, if available, the slab name, return address, and stack trace from
567  * the allocation and last free path of that object.
568  *
569  * This function will splat if passed a pointer to a non-slab object.
570  * If you are not sure what type of object you have, you should instead
571  * use mem_dump_obj().
572  */
kmem_dump_obj(void * object)573 void kmem_dump_obj(void *object)
574 {
575 	char *cp = IS_ENABLED(CONFIG_MMU) ? "" : "/vmalloc";
576 	int i;
577 	struct slab *slab;
578 	unsigned long ptroffset;
579 	struct kmem_obj_info kp = { };
580 
581 	if (WARN_ON_ONCE(!virt_addr_valid(object)))
582 		return;
583 	slab = virt_to_slab(object);
584 	if (WARN_ON_ONCE(!slab)) {
585 		pr_cont(" non-slab memory.\n");
586 		return;
587 	}
588 	kmem_obj_info(&kp, object, slab);
589 	if (kp.kp_slab_cache)
590 		pr_cont(" slab%s %s", cp, kp.kp_slab_cache->name);
591 	else
592 		pr_cont(" slab%s", cp);
593 	if (is_kfence_address(object))
594 		pr_cont(" (kfence)");
595 	if (kp.kp_objp)
596 		pr_cont(" start %px", kp.kp_objp);
597 	if (kp.kp_data_offset)
598 		pr_cont(" data offset %lu", kp.kp_data_offset);
599 	if (kp.kp_objp) {
600 		ptroffset = ((char *)object - (char *)kp.kp_objp) - kp.kp_data_offset;
601 		pr_cont(" pointer offset %lu", ptroffset);
602 	}
603 	if (kp.kp_slab_cache && kp.kp_slab_cache->object_size)
604 		pr_cont(" size %u", kp.kp_slab_cache->object_size);
605 	if (kp.kp_ret)
606 		pr_cont(" allocated at %pS\n", kp.kp_ret);
607 	else
608 		pr_cont("\n");
609 	for (i = 0; i < ARRAY_SIZE(kp.kp_stack); i++) {
610 		if (!kp.kp_stack[i])
611 			break;
612 		pr_info("    %pS\n", kp.kp_stack[i]);
613 	}
614 
615 	if (kp.kp_free_stack[0])
616 		pr_cont(" Free path:\n");
617 
618 	for (i = 0; i < ARRAY_SIZE(kp.kp_free_stack); i++) {
619 		if (!kp.kp_free_stack[i])
620 			break;
621 		pr_info("    %pS\n", kp.kp_free_stack[i]);
622 	}
623 
624 }
625 EXPORT_SYMBOL_GPL(kmem_dump_obj);
626 #endif
627 
628 /* 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)629 void __init create_boot_cache(struct kmem_cache *s, const char *name,
630 		unsigned int size, slab_flags_t flags,
631 		unsigned int useroffset, unsigned int usersize)
632 {
633 	int err;
634 	unsigned int align = ARCH_KMALLOC_MINALIGN;
635 
636 	s->name = name;
637 	s->size = s->object_size = size;
638 
639 	/*
640 	 * For power of two sizes, guarantee natural alignment for kmalloc
641 	 * caches, regardless of SL*B debugging options.
642 	 */
643 	if (is_power_of_2(size))
644 		align = max(align, size);
645 	s->align = calculate_alignment(flags, align, size);
646 
647 #ifdef CONFIG_HARDENED_USERCOPY
648 	s->useroffset = useroffset;
649 	s->usersize = usersize;
650 #endif
651 
652 	err = __kmem_cache_create(s, flags);
653 
654 	if (err)
655 		panic("Creation of kmalloc slab %s size=%u failed. Reason %d\n",
656 					name, size, err);
657 
658 	s->refcount = -1;	/* Exempt from merging for now */
659 }
660 
create_kmalloc_cache(const char * name,unsigned int size,slab_flags_t flags)661 static struct kmem_cache *__init create_kmalloc_cache(const char *name,
662 						      unsigned int size,
663 						      slab_flags_t flags)
664 {
665 	struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
666 
667 	if (!s)
668 		panic("Out of memory when creating slab %s\n", name);
669 
670 	create_boot_cache(s, name, size, flags | SLAB_KMALLOC, 0, size);
671 	list_add(&s->list, &slab_caches);
672 	s->refcount = 1;
673 	return s;
674 }
675 
676 struct kmem_cache *
677 kmalloc_caches[NR_KMALLOC_TYPES][KMALLOC_SHIFT_HIGH + 1] __ro_after_init =
678 { /* initialization for https://bugs.llvm.org/show_bug.cgi?id=42570 */ };
679 EXPORT_SYMBOL(kmalloc_caches);
680 
681 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
682 unsigned long random_kmalloc_seed __ro_after_init;
683 EXPORT_SYMBOL(random_kmalloc_seed);
684 #endif
685 
686 /*
687  * Conversion table for small slabs sizes / 8 to the index in the
688  * kmalloc array. This is necessary for slabs < 192 since we have non power
689  * of two cache sizes there. The size of larger slabs can be determined using
690  * fls.
691  */
692 static u8 size_index[24] __ro_after_init = {
693 	3,	/* 8 */
694 	4,	/* 16 */
695 	5,	/* 24 */
696 	5,	/* 32 */
697 	6,	/* 40 */
698 	6,	/* 48 */
699 	6,	/* 56 */
700 	6,	/* 64 */
701 	1,	/* 72 */
702 	1,	/* 80 */
703 	1,	/* 88 */
704 	1,	/* 96 */
705 	7,	/* 104 */
706 	7,	/* 112 */
707 	7,	/* 120 */
708 	7,	/* 128 */
709 	2,	/* 136 */
710 	2,	/* 144 */
711 	2,	/* 152 */
712 	2,	/* 160 */
713 	2,	/* 168 */
714 	2,	/* 176 */
715 	2,	/* 184 */
716 	2	/* 192 */
717 };
718 
size_index_elem(unsigned int bytes)719 static inline unsigned int size_index_elem(unsigned int bytes)
720 {
721 	return (bytes - 1) / 8;
722 }
723 
724 /*
725  * Find the kmem_cache structure that serves a given size of
726  * allocation
727  */
kmalloc_slab(size_t size,gfp_t flags,unsigned long caller)728 struct kmem_cache *kmalloc_slab(size_t size, gfp_t flags, unsigned long caller)
729 {
730 	unsigned int index;
731 
732 	if (size <= 192) {
733 		if (!size)
734 			return ZERO_SIZE_PTR;
735 
736 		index = size_index[size_index_elem(size)];
737 	} else {
738 		if (WARN_ON_ONCE(size > KMALLOC_MAX_CACHE_SIZE))
739 			return NULL;
740 		index = fls(size - 1);
741 	}
742 
743 	return kmalloc_caches[kmalloc_type(flags, caller)][index];
744 }
745 
kmalloc_size_roundup(size_t size)746 size_t kmalloc_size_roundup(size_t size)
747 {
748 	if (size && size <= KMALLOC_MAX_CACHE_SIZE) {
749 		/*
750 		 * The flags don't matter since size_index is common to all.
751 		 * Neither does the caller for just getting ->object_size.
752 		 */
753 		return kmalloc_slab(size, GFP_KERNEL, 0)->object_size;
754 	}
755 
756 	/* Above the smaller buckets, size is a multiple of page size. */
757 	if (size && size <= KMALLOC_MAX_SIZE)
758 		return PAGE_SIZE << get_order(size);
759 
760 	/*
761 	 * Return 'size' for 0 - kmalloc() returns ZERO_SIZE_PTR
762 	 * and very large size - kmalloc() may fail.
763 	 */
764 	return size;
765 
766 }
767 EXPORT_SYMBOL(kmalloc_size_roundup);
768 
769 #ifdef CONFIG_ZONE_DMA
770 #define KMALLOC_DMA_NAME(sz)	.name[KMALLOC_DMA] = "dma-kmalloc-" #sz,
771 #else
772 #define KMALLOC_DMA_NAME(sz)
773 #endif
774 
775 #ifdef CONFIG_MEMCG_KMEM
776 #define KMALLOC_CGROUP_NAME(sz)	.name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
777 #else
778 #define KMALLOC_CGROUP_NAME(sz)
779 #endif
780 
781 #ifndef CONFIG_SLUB_TINY
782 #define KMALLOC_RCL_NAME(sz)	.name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #sz,
783 #else
784 #define KMALLOC_RCL_NAME(sz)
785 #endif
786 
787 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
788 #define __KMALLOC_RANDOM_CONCAT(a, b) a ## b
789 #define KMALLOC_RANDOM_NAME(N, sz) __KMALLOC_RANDOM_CONCAT(KMA_RAND_, N)(sz)
790 #define KMA_RAND_1(sz)                  .name[KMALLOC_RANDOM_START +  1] = "kmalloc-rnd-01-" #sz,
791 #define KMA_RAND_2(sz)  KMA_RAND_1(sz)  .name[KMALLOC_RANDOM_START +  2] = "kmalloc-rnd-02-" #sz,
792 #define KMA_RAND_3(sz)  KMA_RAND_2(sz)  .name[KMALLOC_RANDOM_START +  3] = "kmalloc-rnd-03-" #sz,
793 #define KMA_RAND_4(sz)  KMA_RAND_3(sz)  .name[KMALLOC_RANDOM_START +  4] = "kmalloc-rnd-04-" #sz,
794 #define KMA_RAND_5(sz)  KMA_RAND_4(sz)  .name[KMALLOC_RANDOM_START +  5] = "kmalloc-rnd-05-" #sz,
795 #define KMA_RAND_6(sz)  KMA_RAND_5(sz)  .name[KMALLOC_RANDOM_START +  6] = "kmalloc-rnd-06-" #sz,
796 #define KMA_RAND_7(sz)  KMA_RAND_6(sz)  .name[KMALLOC_RANDOM_START +  7] = "kmalloc-rnd-07-" #sz,
797 #define KMA_RAND_8(sz)  KMA_RAND_7(sz)  .name[KMALLOC_RANDOM_START +  8] = "kmalloc-rnd-08-" #sz,
798 #define KMA_RAND_9(sz)  KMA_RAND_8(sz)  .name[KMALLOC_RANDOM_START +  9] = "kmalloc-rnd-09-" #sz,
799 #define KMA_RAND_10(sz) KMA_RAND_9(sz)  .name[KMALLOC_RANDOM_START + 10] = "kmalloc-rnd-10-" #sz,
800 #define KMA_RAND_11(sz) KMA_RAND_10(sz) .name[KMALLOC_RANDOM_START + 11] = "kmalloc-rnd-11-" #sz,
801 #define KMA_RAND_12(sz) KMA_RAND_11(sz) .name[KMALLOC_RANDOM_START + 12] = "kmalloc-rnd-12-" #sz,
802 #define KMA_RAND_13(sz) KMA_RAND_12(sz) .name[KMALLOC_RANDOM_START + 13] = "kmalloc-rnd-13-" #sz,
803 #define KMA_RAND_14(sz) KMA_RAND_13(sz) .name[KMALLOC_RANDOM_START + 14] = "kmalloc-rnd-14-" #sz,
804 #define KMA_RAND_15(sz) KMA_RAND_14(sz) .name[KMALLOC_RANDOM_START + 15] = "kmalloc-rnd-15-" #sz,
805 #else // CONFIG_RANDOM_KMALLOC_CACHES
806 #define KMALLOC_RANDOM_NAME(N, sz)
807 #endif
808 
809 #define INIT_KMALLOC_INFO(__size, __short_size)			\
810 {								\
811 	.name[KMALLOC_NORMAL]  = "kmalloc-" #__short_size,	\
812 	KMALLOC_RCL_NAME(__short_size)				\
813 	KMALLOC_CGROUP_NAME(__short_size)			\
814 	KMALLOC_DMA_NAME(__short_size)				\
815 	KMALLOC_RANDOM_NAME(RANDOM_KMALLOC_CACHES_NR, __short_size)	\
816 	.size = __size,						\
817 }
818 
819 /*
820  * kmalloc_info[] is to make slub_debug=,kmalloc-xx option work at boot time.
821  * kmalloc_index() supports up to 2^21=2MB, so the final entry of the table is
822  * kmalloc-2M.
823  */
824 const struct kmalloc_info_struct kmalloc_info[] __initconst = {
825 	INIT_KMALLOC_INFO(0, 0),
826 	INIT_KMALLOC_INFO(96, 96),
827 	INIT_KMALLOC_INFO(192, 192),
828 	INIT_KMALLOC_INFO(8, 8),
829 	INIT_KMALLOC_INFO(16, 16),
830 	INIT_KMALLOC_INFO(32, 32),
831 	INIT_KMALLOC_INFO(64, 64),
832 	INIT_KMALLOC_INFO(128, 128),
833 	INIT_KMALLOC_INFO(256, 256),
834 	INIT_KMALLOC_INFO(512, 512),
835 	INIT_KMALLOC_INFO(1024, 1k),
836 	INIT_KMALLOC_INFO(2048, 2k),
837 	INIT_KMALLOC_INFO(4096, 4k),
838 	INIT_KMALLOC_INFO(8192, 8k),
839 	INIT_KMALLOC_INFO(16384, 16k),
840 	INIT_KMALLOC_INFO(32768, 32k),
841 	INIT_KMALLOC_INFO(65536, 64k),
842 	INIT_KMALLOC_INFO(131072, 128k),
843 	INIT_KMALLOC_INFO(262144, 256k),
844 	INIT_KMALLOC_INFO(524288, 512k),
845 	INIT_KMALLOC_INFO(1048576, 1M),
846 	INIT_KMALLOC_INFO(2097152, 2M)
847 };
848 
849 /*
850  * Patch up the size_index table if we have strange large alignment
851  * requirements for the kmalloc array. This is only the case for
852  * MIPS it seems. The standard arches will not generate any code here.
853  *
854  * Largest permitted alignment is 256 bytes due to the way we
855  * handle the index determination for the smaller caches.
856  *
857  * Make sure that nothing crazy happens if someone starts tinkering
858  * around with ARCH_KMALLOC_MINALIGN
859  */
setup_kmalloc_cache_index_table(void)860 void __init setup_kmalloc_cache_index_table(void)
861 {
862 	unsigned int i;
863 
864 	BUILD_BUG_ON(KMALLOC_MIN_SIZE > 256 ||
865 		!is_power_of_2(KMALLOC_MIN_SIZE));
866 
867 	for (i = 8; i < KMALLOC_MIN_SIZE; i += 8) {
868 		unsigned int elem = size_index_elem(i);
869 
870 		if (elem >= ARRAY_SIZE(size_index))
871 			break;
872 		size_index[elem] = KMALLOC_SHIFT_LOW;
873 	}
874 
875 	if (KMALLOC_MIN_SIZE >= 64) {
876 		/*
877 		 * The 96 byte sized cache is not used if the alignment
878 		 * is 64 byte.
879 		 */
880 		for (i = 64 + 8; i <= 96; i += 8)
881 			size_index[size_index_elem(i)] = 7;
882 
883 	}
884 
885 	if (KMALLOC_MIN_SIZE >= 128) {
886 		/*
887 		 * The 192 byte sized cache is not used if the alignment
888 		 * is 128 byte. Redirect kmalloc to use the 256 byte cache
889 		 * instead.
890 		 */
891 		for (i = 128 + 8; i <= 192; i += 8)
892 			size_index[size_index_elem(i)] = 8;
893 	}
894 }
895 
__kmalloc_minalign(void)896 static unsigned int __kmalloc_minalign(void)
897 {
898 	unsigned int minalign = dma_get_cache_alignment();
899 
900 	if (IS_ENABLED(CONFIG_DMA_BOUNCE_UNALIGNED_KMALLOC) &&
901 	    is_swiotlb_allocated())
902 		minalign = ARCH_KMALLOC_MINALIGN;
903 
904 	return max(minalign, arch_slab_minalign());
905 }
906 
907 void __init
new_kmalloc_cache(int idx,enum kmalloc_cache_type type,slab_flags_t flags)908 new_kmalloc_cache(int idx, enum kmalloc_cache_type type, slab_flags_t flags)
909 {
910 	unsigned int minalign = __kmalloc_minalign();
911 	unsigned int aligned_size = kmalloc_info[idx].size;
912 	int aligned_idx = idx;
913 
914 	if ((KMALLOC_RECLAIM != KMALLOC_NORMAL) && (type == KMALLOC_RECLAIM)) {
915 		flags |= SLAB_RECLAIM_ACCOUNT;
916 	} else if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_CGROUP)) {
917 		if (mem_cgroup_kmem_disabled()) {
918 			kmalloc_caches[type][idx] = kmalloc_caches[KMALLOC_NORMAL][idx];
919 			return;
920 		}
921 		flags |= SLAB_ACCOUNT;
922 	} else if (IS_ENABLED(CONFIG_ZONE_DMA) && (type == KMALLOC_DMA)) {
923 		flags |= SLAB_CACHE_DMA;
924 	}
925 
926 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
927 	if (type >= KMALLOC_RANDOM_START && type <= KMALLOC_RANDOM_END)
928 		flags |= SLAB_NO_MERGE;
929 #endif
930 
931 	/*
932 	 * If CONFIG_MEMCG_KMEM is enabled, disable cache merging for
933 	 * KMALLOC_NORMAL caches.
934 	 */
935 	if (IS_ENABLED(CONFIG_MEMCG_KMEM) && (type == KMALLOC_NORMAL))
936 		flags |= SLAB_NO_MERGE;
937 
938 	if (minalign > ARCH_KMALLOC_MINALIGN) {
939 		aligned_size = ALIGN(aligned_size, minalign);
940 		aligned_idx = __kmalloc_index(aligned_size, false);
941 	}
942 
943 	if (!kmalloc_caches[type][aligned_idx])
944 		kmalloc_caches[type][aligned_idx] = create_kmalloc_cache(
945 					kmalloc_info[aligned_idx].name[type],
946 					aligned_size, flags);
947 	if (idx != aligned_idx)
948 		kmalloc_caches[type][idx] = kmalloc_caches[type][aligned_idx];
949 }
950 
951 /*
952  * Create the kmalloc array. Some of the regular kmalloc arrays
953  * may already have been created because they were needed to
954  * enable allocations for slab creation.
955  */
create_kmalloc_caches(slab_flags_t flags)956 void __init create_kmalloc_caches(slab_flags_t flags)
957 {
958 	int i;
959 	enum kmalloc_cache_type type;
960 
961 	/*
962 	 * Including KMALLOC_CGROUP if CONFIG_MEMCG_KMEM defined
963 	 */
964 	for (type = KMALLOC_NORMAL; type < NR_KMALLOC_TYPES; type++) {
965 		for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++) {
966 			if (!kmalloc_caches[type][i])
967 				new_kmalloc_cache(i, type, flags);
968 
969 			/*
970 			 * Caches that are not of the two-to-the-power-of size.
971 			 * These have to be created immediately after the
972 			 * earlier power of two caches
973 			 */
974 			if (KMALLOC_MIN_SIZE <= 32 && i == 6 &&
975 					!kmalloc_caches[type][1])
976 				new_kmalloc_cache(1, type, flags);
977 			if (KMALLOC_MIN_SIZE <= 64 && i == 7 &&
978 					!kmalloc_caches[type][2])
979 				new_kmalloc_cache(2, type, flags);
980 		}
981 	}
982 #ifdef CONFIG_RANDOM_KMALLOC_CACHES
983 	random_kmalloc_seed = get_random_u64();
984 #endif
985 
986 	/* Kmalloc array is now usable */
987 	slab_state = UP;
988 }
989 
free_large_kmalloc(struct folio * folio,void * object)990 void free_large_kmalloc(struct folio *folio, void *object)
991 {
992 	unsigned int order = folio_order(folio);
993 
994 	if (WARN_ON_ONCE(order == 0))
995 		pr_warn_once("object pointer: 0x%p\n", object);
996 
997 	kmemleak_free(object);
998 	kasan_kfree_large(object);
999 	kmsan_kfree_large(object);
1000 
1001 	mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B,
1002 			      -(PAGE_SIZE << order));
1003 	__free_pages(folio_page(folio, 0), order);
1004 }
1005 
1006 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node);
1007 static __always_inline
__do_kmalloc_node(size_t size,gfp_t flags,int node,unsigned long caller)1008 void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
1009 {
1010 	struct kmem_cache *s;
1011 	void *ret;
1012 
1013 	if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
1014 		ret = __kmalloc_large_node(size, flags, node);
1015 		trace_kmalloc(caller, ret, size,
1016 			      PAGE_SIZE << get_order(size), flags, node);
1017 		return ret;
1018 	}
1019 
1020 	s = kmalloc_slab(size, flags, caller);
1021 
1022 	if (unlikely(ZERO_OR_NULL_PTR(s)))
1023 		return s;
1024 
1025 	ret = __kmem_cache_alloc_node(s, flags, node, size, caller);
1026 	ret = kasan_kmalloc(s, ret, size, flags);
1027 	trace_kmalloc(caller, ret, size, s->size, flags, node);
1028 	return ret;
1029 }
1030 
__kmalloc_node(size_t size,gfp_t flags,int node)1031 void *__kmalloc_node(size_t size, gfp_t flags, int node)
1032 {
1033 	return __do_kmalloc_node(size, flags, node, _RET_IP_);
1034 }
1035 EXPORT_SYMBOL(__kmalloc_node);
1036 
__kmalloc(size_t size,gfp_t flags)1037 void *__kmalloc(size_t size, gfp_t flags)
1038 {
1039 	return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_);
1040 }
1041 EXPORT_SYMBOL(__kmalloc);
1042 
__kmalloc_node_track_caller(size_t size,gfp_t flags,int node,unsigned long caller)1043 void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
1044 				  int node, unsigned long caller)
1045 {
1046 	return __do_kmalloc_node(size, flags, node, caller);
1047 }
1048 EXPORT_SYMBOL(__kmalloc_node_track_caller);
1049 
1050 /**
1051  * kfree - free previously allocated memory
1052  * @object: pointer returned by kmalloc() or kmem_cache_alloc()
1053  *
1054  * If @object is NULL, no operation is performed.
1055  */
kfree(const void * object)1056 void kfree(const void *object)
1057 {
1058 	struct folio *folio;
1059 	struct slab *slab;
1060 	struct kmem_cache *s;
1061 
1062 	trace_kfree(_RET_IP_, object);
1063 
1064 	if (unlikely(ZERO_OR_NULL_PTR(object)))
1065 		return;
1066 
1067 	folio = virt_to_folio(object);
1068 	if (unlikely(!folio_test_slab(folio))) {
1069 		free_large_kmalloc(folio, (void *)object);
1070 		return;
1071 	}
1072 
1073 	slab = folio_slab(folio);
1074 	s = slab->slab_cache;
1075 	__kmem_cache_free(s, (void *)object, _RET_IP_);
1076 }
1077 EXPORT_SYMBOL(kfree);
1078 
1079 /**
1080  * __ksize -- Report full size of underlying allocation
1081  * @object: pointer to the object
1082  *
1083  * This should only be used internally to query the true size of allocations.
1084  * It is not meant to be a way to discover the usable size of an allocation
1085  * after the fact. Instead, use kmalloc_size_roundup(). Using memory beyond
1086  * the originally requested allocation size may trigger KASAN, UBSAN_BOUNDS,
1087  * and/or FORTIFY_SOURCE.
1088  *
1089  * Return: size of the actual memory used by @object in bytes
1090  */
__ksize(const void * object)1091 size_t __ksize(const void *object)
1092 {
1093 	struct folio *folio;
1094 
1095 	if (unlikely(object == ZERO_SIZE_PTR))
1096 		return 0;
1097 
1098 	folio = virt_to_folio(object);
1099 
1100 	if (unlikely(!folio_test_slab(folio))) {
1101 		if (WARN_ON(folio_size(folio) <= KMALLOC_MAX_CACHE_SIZE))
1102 			return 0;
1103 		if (WARN_ON(object != folio_address(folio)))
1104 			return 0;
1105 		return folio_size(folio);
1106 	}
1107 
1108 #ifdef CONFIG_SLUB_DEBUG
1109 	skip_orig_size_check(folio_slab(folio)->slab_cache, object);
1110 #endif
1111 
1112 	return slab_ksize(folio_slab(folio)->slab_cache);
1113 }
1114 
kmalloc_trace(struct kmem_cache * s,gfp_t gfpflags,size_t size)1115 void *kmalloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
1116 {
1117 	void *ret = __kmem_cache_alloc_node(s, gfpflags, NUMA_NO_NODE,
1118 					    size, _RET_IP_);
1119 
1120 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, NUMA_NO_NODE);
1121 
1122 	ret = kasan_kmalloc(s, ret, size, gfpflags);
1123 	return ret;
1124 }
1125 EXPORT_SYMBOL(kmalloc_trace);
1126 
kmalloc_node_trace(struct kmem_cache * s,gfp_t gfpflags,int node,size_t size)1127 void *kmalloc_node_trace(struct kmem_cache *s, gfp_t gfpflags,
1128 			 int node, size_t size)
1129 {
1130 	void *ret = __kmem_cache_alloc_node(s, gfpflags, node, size, _RET_IP_);
1131 
1132 	trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags, node);
1133 
1134 	ret = kasan_kmalloc(s, ret, size, gfpflags);
1135 	return ret;
1136 }
1137 EXPORT_SYMBOL(kmalloc_node_trace);
1138 
kmalloc_fix_flags(gfp_t flags)1139 gfp_t kmalloc_fix_flags(gfp_t flags)
1140 {
1141 	gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
1142 
1143 	flags &= ~GFP_SLAB_BUG_MASK;
1144 	pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1145 			invalid_mask, &invalid_mask, flags, &flags);
1146 	dump_stack();
1147 
1148 	return flags;
1149 }
1150 
1151 /*
1152  * To avoid unnecessary overhead, we pass through large allocation requests
1153  * directly to the page allocator. We use __GFP_COMP, because we will need to
1154  * know the allocation order to free the pages properly in kfree.
1155  */
1156 
__kmalloc_large_node(size_t size,gfp_t flags,int node)1157 static void *__kmalloc_large_node(size_t size, gfp_t flags, int node)
1158 {
1159 	struct page *page;
1160 	void *ptr = NULL;
1161 	unsigned int order = get_order(size);
1162 
1163 	if (unlikely(flags & GFP_SLAB_BUG_MASK))
1164 		flags = kmalloc_fix_flags(flags);
1165 
1166 	flags |= __GFP_COMP;
1167 	page = alloc_pages_node(node, flags, order);
1168 	if (page) {
1169 		ptr = page_address(page);
1170 		mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B,
1171 				      PAGE_SIZE << order);
1172 	}
1173 
1174 	ptr = kasan_kmalloc_large(ptr, size, flags);
1175 	/* As ptr might get tagged, call kmemleak hook after KASAN. */
1176 	kmemleak_alloc(ptr, size, 1, flags);
1177 	kmsan_kmalloc_large(ptr, size, flags);
1178 
1179 	return ptr;
1180 }
1181 
kmalloc_large(size_t size,gfp_t flags)1182 void *kmalloc_large(size_t size, gfp_t flags)
1183 {
1184 	void *ret = __kmalloc_large_node(size, flags, NUMA_NO_NODE);
1185 
1186 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1187 		      flags, NUMA_NO_NODE);
1188 	return ret;
1189 }
1190 EXPORT_SYMBOL(kmalloc_large);
1191 
kmalloc_large_node(size_t size,gfp_t flags,int node)1192 void *kmalloc_large_node(size_t size, gfp_t flags, int node)
1193 {
1194 	void *ret = __kmalloc_large_node(size, flags, node);
1195 
1196 	trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << get_order(size),
1197 		      flags, node);
1198 	return ret;
1199 }
1200 EXPORT_SYMBOL(kmalloc_large_node);
1201 
1202 #ifdef CONFIG_SLAB_FREELIST_RANDOM
1203 /* Randomize a generic freelist */
freelist_randomize(unsigned int * list,unsigned int count)1204 static void freelist_randomize(unsigned int *list,
1205 			       unsigned int count)
1206 {
1207 	unsigned int rand;
1208 	unsigned int i;
1209 
1210 	for (i = 0; i < count; i++)
1211 		list[i] = i;
1212 
1213 	/* Fisher-Yates shuffle */
1214 	for (i = count - 1; i > 0; i--) {
1215 		rand = get_random_u32_below(i + 1);
1216 		swap(list[i], list[rand]);
1217 	}
1218 }
1219 
1220 /* Create a random sequence per cache */
cache_random_seq_create(struct kmem_cache * cachep,unsigned int count,gfp_t gfp)1221 int cache_random_seq_create(struct kmem_cache *cachep, unsigned int count,
1222 				    gfp_t gfp)
1223 {
1224 
1225 	if (count < 2 || cachep->random_seq)
1226 		return 0;
1227 
1228 	cachep->random_seq = kcalloc(count, sizeof(unsigned int), gfp);
1229 	if (!cachep->random_seq)
1230 		return -ENOMEM;
1231 
1232 	freelist_randomize(cachep->random_seq, count);
1233 	return 0;
1234 }
1235 
1236 /* Destroy the per-cache random freelist sequence */
cache_random_seq_destroy(struct kmem_cache * cachep)1237 void cache_random_seq_destroy(struct kmem_cache *cachep)
1238 {
1239 	kfree(cachep->random_seq);
1240 	cachep->random_seq = NULL;
1241 }
1242 #endif /* CONFIG_SLAB_FREELIST_RANDOM */
1243 
1244 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
1245 #ifdef CONFIG_SLAB
1246 #define SLABINFO_RIGHTS (0600)
1247 #else
1248 #define SLABINFO_RIGHTS (0400)
1249 #endif
1250 
print_slabinfo_header(struct seq_file * m)1251 static void print_slabinfo_header(struct seq_file *m)
1252 {
1253 	/*
1254 	 * Output format version, so at least we can change it
1255 	 * without _too_ many complaints.
1256 	 */
1257 #ifdef CONFIG_DEBUG_SLAB
1258 	seq_puts(m, "slabinfo - version: 2.1 (statistics)\n");
1259 #else
1260 	seq_puts(m, "slabinfo - version: 2.1\n");
1261 #endif
1262 	seq_puts(m, "# name            <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab>");
1263 	seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>");
1264 	seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
1265 #ifdef CONFIG_DEBUG_SLAB
1266 	seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> <error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
1267 	seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
1268 #endif
1269 	seq_putc(m, '\n');
1270 }
1271 
slab_start(struct seq_file * m,loff_t * pos)1272 static void *slab_start(struct seq_file *m, loff_t *pos)
1273 {
1274 	mutex_lock(&slab_mutex);
1275 	return seq_list_start(&slab_caches, *pos);
1276 }
1277 
slab_next(struct seq_file * m,void * p,loff_t * pos)1278 static void *slab_next(struct seq_file *m, void *p, loff_t *pos)
1279 {
1280 	return seq_list_next(p, &slab_caches, pos);
1281 }
1282 
slab_stop(struct seq_file * m,void * p)1283 static void slab_stop(struct seq_file *m, void *p)
1284 {
1285 	mutex_unlock(&slab_mutex);
1286 }
1287 
cache_show(struct kmem_cache * s,struct seq_file * m)1288 static void cache_show(struct kmem_cache *s, struct seq_file *m)
1289 {
1290 	struct slabinfo sinfo;
1291 
1292 	memset(&sinfo, 0, sizeof(sinfo));
1293 	get_slabinfo(s, &sinfo);
1294 
1295 	seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d",
1296 		   s->name, sinfo.active_objs, sinfo.num_objs, s->size,
1297 		   sinfo.objects_per_slab, (1 << sinfo.cache_order));
1298 
1299 	seq_printf(m, " : tunables %4u %4u %4u",
1300 		   sinfo.limit, sinfo.batchcount, sinfo.shared);
1301 	seq_printf(m, " : slabdata %6lu %6lu %6lu",
1302 		   sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail);
1303 	slabinfo_show_stats(m, s);
1304 	seq_putc(m, '\n');
1305 }
1306 
slab_show(struct seq_file * m,void * p)1307 static int slab_show(struct seq_file *m, void *p)
1308 {
1309 	struct kmem_cache *s = list_entry(p, struct kmem_cache, list);
1310 
1311 	if (p == slab_caches.next)
1312 		print_slabinfo_header(m);
1313 	cache_show(s, m);
1314 	return 0;
1315 }
1316 
dump_unreclaimable_slab(void)1317 void dump_unreclaimable_slab(void)
1318 {
1319 	struct kmem_cache *s;
1320 	struct slabinfo sinfo;
1321 
1322 	/*
1323 	 * Here acquiring slab_mutex is risky since we don't prefer to get
1324 	 * sleep in oom path. But, without mutex hold, it may introduce a
1325 	 * risk of crash.
1326 	 * Use mutex_trylock to protect the list traverse, dump nothing
1327 	 * without acquiring the mutex.
1328 	 */
1329 	if (!mutex_trylock(&slab_mutex)) {
1330 		pr_warn("excessive unreclaimable slab but cannot dump stats\n");
1331 		return;
1332 	}
1333 
1334 	pr_info("Unreclaimable slab info:\n");
1335 	pr_info("Name                      Used          Total\n");
1336 
1337 	list_for_each_entry(s, &slab_caches, list) {
1338 		if (s->flags & SLAB_RECLAIM_ACCOUNT)
1339 			continue;
1340 
1341 		get_slabinfo(s, &sinfo);
1342 
1343 		if (sinfo.num_objs > 0)
1344 			pr_info("%-17s %10luKB %10luKB\n", s->name,
1345 				(sinfo.active_objs * s->size) / 1024,
1346 				(sinfo.num_objs * s->size) / 1024);
1347 	}
1348 	mutex_unlock(&slab_mutex);
1349 }
1350 
1351 /*
1352  * slabinfo_op - iterator that generates /proc/slabinfo
1353  *
1354  * Output layout:
1355  * cache-name
1356  * num-active-objs
1357  * total-objs
1358  * object size
1359  * num-active-slabs
1360  * total-slabs
1361  * num-pages-per-slab
1362  * + further values on SMP and with statistics enabled
1363  */
1364 static const struct seq_operations slabinfo_op = {
1365 	.start = slab_start,
1366 	.next = slab_next,
1367 	.stop = slab_stop,
1368 	.show = slab_show,
1369 };
1370 
slabinfo_open(struct inode * inode,struct file * file)1371 static int slabinfo_open(struct inode *inode, struct file *file)
1372 {
1373 	return seq_open(file, &slabinfo_op);
1374 }
1375 
1376 static const struct proc_ops slabinfo_proc_ops = {
1377 	.proc_flags	= PROC_ENTRY_PERMANENT,
1378 	.proc_open	= slabinfo_open,
1379 	.proc_read	= seq_read,
1380 	.proc_write	= slabinfo_write,
1381 	.proc_lseek	= seq_lseek,
1382 	.proc_release	= seq_release,
1383 };
1384 
slab_proc_init(void)1385 static int __init slab_proc_init(void)
1386 {
1387 	proc_create("slabinfo", SLABINFO_RIGHTS, NULL, &slabinfo_proc_ops);
1388 	return 0;
1389 }
1390 module_init(slab_proc_init);
1391 
1392 #endif /* CONFIG_SLAB || CONFIG_SLUB_DEBUG */
1393 
1394 static __always_inline __realloc_size(2) void *
__do_krealloc(const void * p,size_t new_size,gfp_t flags)1395 __do_krealloc(const void *p, size_t new_size, gfp_t flags)
1396 {
1397 	void *ret;
1398 	size_t ks;
1399 
1400 	/* Check for double-free before calling ksize. */
1401 	if (likely(!ZERO_OR_NULL_PTR(p))) {
1402 		if (!kasan_check_byte(p))
1403 			return NULL;
1404 		ks = ksize(p);
1405 	} else
1406 		ks = 0;
1407 
1408 	/* If the object still fits, repoison it precisely. */
1409 	if (ks >= new_size) {
1410 		p = kasan_krealloc((void *)p, new_size, flags);
1411 		return (void *)p;
1412 	}
1413 
1414 	ret = kmalloc_track_caller(new_size, flags);
1415 	if (ret && p) {
1416 		/* Disable KASAN checks as the object's redzone is accessed. */
1417 		kasan_disable_current();
1418 		memcpy(ret, kasan_reset_tag(p), ks);
1419 		kasan_enable_current();
1420 	}
1421 
1422 	return ret;
1423 }
1424 
1425 /**
1426  * krealloc - reallocate memory. The contents will remain unchanged.
1427  * @p: object to reallocate memory for.
1428  * @new_size: how many bytes of memory are required.
1429  * @flags: the type of memory to allocate.
1430  *
1431  * The contents of the object pointed to are preserved up to the
1432  * lesser of the new and old sizes (__GFP_ZERO flag is effectively ignored).
1433  * If @p is %NULL, krealloc() behaves exactly like kmalloc().  If @new_size
1434  * is 0 and @p is not a %NULL pointer, the object pointed to is freed.
1435  *
1436  * Return: pointer to the allocated memory or %NULL in case of error
1437  */
krealloc(const void * p,size_t new_size,gfp_t flags)1438 void *krealloc(const void *p, size_t new_size, gfp_t flags)
1439 {
1440 	void *ret;
1441 
1442 	if (unlikely(!new_size)) {
1443 		kfree(p);
1444 		return ZERO_SIZE_PTR;
1445 	}
1446 
1447 	ret = __do_krealloc(p, new_size, flags);
1448 	if (ret && kasan_reset_tag(p) != kasan_reset_tag(ret))
1449 		kfree(p);
1450 
1451 	return ret;
1452 }
1453 EXPORT_SYMBOL(krealloc);
1454 
1455 /**
1456  * kfree_sensitive - Clear sensitive information in memory before freeing
1457  * @p: object to free memory of
1458  *
1459  * The memory of the object @p points to is zeroed before freed.
1460  * If @p is %NULL, kfree_sensitive() does nothing.
1461  *
1462  * Note: this function zeroes the whole allocated buffer which can be a good
1463  * deal bigger than the requested buffer size passed to kmalloc(). So be
1464  * careful when using this function in performance sensitive code.
1465  */
kfree_sensitive(const void * p)1466 void kfree_sensitive(const void *p)
1467 {
1468 	size_t ks;
1469 	void *mem = (void *)p;
1470 
1471 	ks = ksize(mem);
1472 	if (ks) {
1473 		kasan_unpoison_range(mem, ks);
1474 		memzero_explicit(mem, ks);
1475 	}
1476 	kfree(mem);
1477 }
1478 EXPORT_SYMBOL(kfree_sensitive);
1479 
ksize(const void * objp)1480 size_t ksize(const void *objp)
1481 {
1482 	/*
1483 	 * We need to first check that the pointer to the object is valid.
1484 	 * The KASAN report printed from ksize() is more useful, then when
1485 	 * it's printed later when the behaviour could be undefined due to
1486 	 * a potential use-after-free or double-free.
1487 	 *
1488 	 * We use kasan_check_byte(), which is supported for the hardware
1489 	 * tag-based KASAN mode, unlike kasan_check_read/write().
1490 	 *
1491 	 * If the pointed to memory is invalid, we return 0 to avoid users of
1492 	 * ksize() writing to and potentially corrupting the memory region.
1493 	 *
1494 	 * We want to perform the check before __ksize(), to avoid potentially
1495 	 * crashing in __ksize() due to accessing invalid metadata.
1496 	 */
1497 	if (unlikely(ZERO_OR_NULL_PTR(objp)) || !kasan_check_byte(objp))
1498 		return 0;
1499 
1500 	return kfence_ksize(objp) ?: __ksize(objp);
1501 }
1502 EXPORT_SYMBOL(ksize);
1503 
1504 /* Tracepoints definitions. */
1505 EXPORT_TRACEPOINT_SYMBOL(kmalloc);
1506 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_alloc);
1507 EXPORT_TRACEPOINT_SYMBOL(kfree);
1508 EXPORT_TRACEPOINT_SYMBOL(kmem_cache_free);
1509 
should_failslab(struct kmem_cache * s,gfp_t gfpflags)1510 int should_failslab(struct kmem_cache *s, gfp_t gfpflags)
1511 {
1512 	if (__should_failslab(s, gfpflags))
1513 		return -ENOMEM;
1514 	return 0;
1515 }
1516 ALLOW_ERROR_INJECTION(should_failslab, ERRNO);
1517