xref: /linux/mm/kasan/common.c (revision 24ce659dcc02c21f8d6c0a7589c3320a4dfa8152)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * This file contains common generic and tag-based KASAN code.
4  *
5  * Copyright (c) 2014 Samsung Electronics Co., Ltd.
6  * Author: Andrey Ryabinin <ryabinin.a.a@gmail.com>
7  *
8  * Some code borrowed from https://github.com/xairy/kasan-prototype by
9  *        Andrey Konovalov <andreyknvl@gmail.com>
10  *
11  * This program is free software; you can redistribute it and/or modify
12  * it under the terms of the GNU General Public License version 2 as
13  * published by the Free Software Foundation.
14  *
15  */
16 
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/kasan.h>
20 #include <linux/kernel.h>
21 #include <linux/kmemleak.h>
22 #include <linux/linkage.h>
23 #include <linux/memblock.h>
24 #include <linux/memory.h>
25 #include <linux/mm.h>
26 #include <linux/module.h>
27 #include <linux/printk.h>
28 #include <linux/sched.h>
29 #include <linux/sched/task_stack.h>
30 #include <linux/slab.h>
31 #include <linux/stacktrace.h>
32 #include <linux/string.h>
33 #include <linux/types.h>
34 #include <linux/vmalloc.h>
35 #include <linux/bug.h>
36 #include <linux/uaccess.h>
37 
38 #include <asm/cacheflush.h>
39 #include <asm/tlbflush.h>
40 
41 #include "kasan.h"
42 #include "../slab.h"
43 
44 static inline depot_stack_handle_t save_stack(gfp_t flags)
45 {
46 	unsigned long entries[KASAN_STACK_DEPTH];
47 	unsigned int nr_entries;
48 
49 	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0);
50 	nr_entries = filter_irq_stacks(entries, nr_entries);
51 	return stack_depot_save(entries, nr_entries, flags);
52 }
53 
54 static inline void set_track(struct kasan_track *track, gfp_t flags)
55 {
56 	track->pid = current->pid;
57 	track->stack = save_stack(flags);
58 }
59 
60 void kasan_enable_current(void)
61 {
62 	current->kasan_depth++;
63 }
64 
65 void kasan_disable_current(void)
66 {
67 	current->kasan_depth--;
68 }
69 
70 bool __kasan_check_read(const volatile void *p, unsigned int size)
71 {
72 	return check_memory_region((unsigned long)p, size, false, _RET_IP_);
73 }
74 EXPORT_SYMBOL(__kasan_check_read);
75 
76 bool __kasan_check_write(const volatile void *p, unsigned int size)
77 {
78 	return check_memory_region((unsigned long)p, size, true, _RET_IP_);
79 }
80 EXPORT_SYMBOL(__kasan_check_write);
81 
82 #undef memset
83 void *memset(void *addr, int c, size_t len)
84 {
85 	if (!check_memory_region((unsigned long)addr, len, true, _RET_IP_))
86 		return NULL;
87 
88 	return __memset(addr, c, len);
89 }
90 
91 #ifdef __HAVE_ARCH_MEMMOVE
92 #undef memmove
93 void *memmove(void *dest, const void *src, size_t len)
94 {
95 	if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
96 	    !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
97 		return NULL;
98 
99 	return __memmove(dest, src, len);
100 }
101 #endif
102 
103 #undef memcpy
104 void *memcpy(void *dest, const void *src, size_t len)
105 {
106 	if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
107 	    !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
108 		return NULL;
109 
110 	return __memcpy(dest, src, len);
111 }
112 
113 /*
114  * Poisons the shadow memory for 'size' bytes starting from 'addr'.
115  * Memory addresses should be aligned to KASAN_SHADOW_SCALE_SIZE.
116  */
117 void kasan_poison_shadow(const void *address, size_t size, u8 value)
118 {
119 	void *shadow_start, *shadow_end;
120 
121 	/*
122 	 * Perform shadow offset calculation based on untagged address, as
123 	 * some of the callers (e.g. kasan_poison_object_data) pass tagged
124 	 * addresses to this function.
125 	 */
126 	address = reset_tag(address);
127 
128 	shadow_start = kasan_mem_to_shadow(address);
129 	shadow_end = kasan_mem_to_shadow(address + size);
130 
131 	__memset(shadow_start, value, shadow_end - shadow_start);
132 }
133 
134 void kasan_unpoison_shadow(const void *address, size_t size)
135 {
136 	u8 tag = get_tag(address);
137 
138 	/*
139 	 * Perform shadow offset calculation based on untagged address, as
140 	 * some of the callers (e.g. kasan_unpoison_object_data) pass tagged
141 	 * addresses to this function.
142 	 */
143 	address = reset_tag(address);
144 
145 	kasan_poison_shadow(address, size, tag);
146 
147 	if (size & KASAN_SHADOW_MASK) {
148 		u8 *shadow = (u8 *)kasan_mem_to_shadow(address + size);
149 
150 		if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
151 			*shadow = tag;
152 		else
153 			*shadow = size & KASAN_SHADOW_MASK;
154 	}
155 }
156 
157 static void __kasan_unpoison_stack(struct task_struct *task, const void *sp)
158 {
159 	void *base = task_stack_page(task);
160 	size_t size = sp - base;
161 
162 	kasan_unpoison_shadow(base, size);
163 }
164 
165 /* Unpoison the entire stack for a task. */
166 void kasan_unpoison_task_stack(struct task_struct *task)
167 {
168 	__kasan_unpoison_stack(task, task_stack_page(task) + THREAD_SIZE);
169 }
170 
171 /* Unpoison the stack for the current task beyond a watermark sp value. */
172 asmlinkage void kasan_unpoison_task_stack_below(const void *watermark)
173 {
174 	/*
175 	 * Calculate the task stack base address.  Avoid using 'current'
176 	 * because this function is called by early resume code which hasn't
177 	 * yet set up the percpu register (%gs).
178 	 */
179 	void *base = (void *)((unsigned long)watermark & ~(THREAD_SIZE - 1));
180 
181 	kasan_unpoison_shadow(base, watermark - base);
182 }
183 
184 /*
185  * Clear all poison for the region between the current SP and a provided
186  * watermark value, as is sometimes required prior to hand-crafted asm function
187  * returns in the middle of functions.
188  */
189 void kasan_unpoison_stack_above_sp_to(const void *watermark)
190 {
191 	const void *sp = __builtin_frame_address(0);
192 	size_t size = watermark - sp;
193 
194 	if (WARN_ON(sp > watermark))
195 		return;
196 	kasan_unpoison_shadow(sp, size);
197 }
198 
199 void kasan_alloc_pages(struct page *page, unsigned int order)
200 {
201 	u8 tag;
202 	unsigned long i;
203 
204 	if (unlikely(PageHighMem(page)))
205 		return;
206 
207 	tag = random_tag();
208 	for (i = 0; i < (1 << order); i++)
209 		page_kasan_tag_set(page + i, tag);
210 	kasan_unpoison_shadow(page_address(page), PAGE_SIZE << order);
211 }
212 
213 void kasan_free_pages(struct page *page, unsigned int order)
214 {
215 	if (likely(!PageHighMem(page)))
216 		kasan_poison_shadow(page_address(page),
217 				PAGE_SIZE << order,
218 				KASAN_FREE_PAGE);
219 }
220 
221 /*
222  * Adaptive redzone policy taken from the userspace AddressSanitizer runtime.
223  * For larger allocations larger redzones are used.
224  */
225 static inline unsigned int optimal_redzone(unsigned int object_size)
226 {
227 	if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
228 		return 0;
229 
230 	return
231 		object_size <= 64        - 16   ? 16 :
232 		object_size <= 128       - 32   ? 32 :
233 		object_size <= 512       - 64   ? 64 :
234 		object_size <= 4096      - 128  ? 128 :
235 		object_size <= (1 << 14) - 256  ? 256 :
236 		object_size <= (1 << 15) - 512  ? 512 :
237 		object_size <= (1 << 16) - 1024 ? 1024 : 2048;
238 }
239 
240 void kasan_cache_create(struct kmem_cache *cache, unsigned int *size,
241 			slab_flags_t *flags)
242 {
243 	unsigned int orig_size = *size;
244 	unsigned int redzone_size;
245 	int redzone_adjust;
246 
247 	/* Add alloc meta. */
248 	cache->kasan_info.alloc_meta_offset = *size;
249 	*size += sizeof(struct kasan_alloc_meta);
250 
251 	/* Add free meta. */
252 	if (IS_ENABLED(CONFIG_KASAN_GENERIC) &&
253 	    (cache->flags & SLAB_TYPESAFE_BY_RCU || cache->ctor ||
254 	     cache->object_size < sizeof(struct kasan_free_meta))) {
255 		cache->kasan_info.free_meta_offset = *size;
256 		*size += sizeof(struct kasan_free_meta);
257 	}
258 
259 	redzone_size = optimal_redzone(cache->object_size);
260 	redzone_adjust = redzone_size -	(*size - cache->object_size);
261 	if (redzone_adjust > 0)
262 		*size += redzone_adjust;
263 
264 	*size = min_t(unsigned int, KMALLOC_MAX_SIZE,
265 			max(*size, cache->object_size + redzone_size));
266 
267 	/*
268 	 * If the metadata doesn't fit, don't enable KASAN at all.
269 	 */
270 	if (*size <= cache->kasan_info.alloc_meta_offset ||
271 			*size <= cache->kasan_info.free_meta_offset) {
272 		cache->kasan_info.alloc_meta_offset = 0;
273 		cache->kasan_info.free_meta_offset = 0;
274 		*size = orig_size;
275 		return;
276 	}
277 
278 	*flags |= SLAB_KASAN;
279 }
280 
281 size_t kasan_metadata_size(struct kmem_cache *cache)
282 {
283 	return (cache->kasan_info.alloc_meta_offset ?
284 		sizeof(struct kasan_alloc_meta) : 0) +
285 		(cache->kasan_info.free_meta_offset ?
286 		sizeof(struct kasan_free_meta) : 0);
287 }
288 
289 struct kasan_alloc_meta *get_alloc_info(struct kmem_cache *cache,
290 					const void *object)
291 {
292 	return (void *)object + cache->kasan_info.alloc_meta_offset;
293 }
294 
295 struct kasan_free_meta *get_free_info(struct kmem_cache *cache,
296 				      const void *object)
297 {
298 	BUILD_BUG_ON(sizeof(struct kasan_free_meta) > 32);
299 	return (void *)object + cache->kasan_info.free_meta_offset;
300 }
301 
302 
303 static void kasan_set_free_info(struct kmem_cache *cache,
304 		void *object, u8 tag)
305 {
306 	struct kasan_alloc_meta *alloc_meta;
307 	u8 idx = 0;
308 
309 	alloc_meta = get_alloc_info(cache, object);
310 
311 #ifdef CONFIG_KASAN_SW_TAGS_IDENTIFY
312 	idx = alloc_meta->free_track_idx;
313 	alloc_meta->free_pointer_tag[idx] = tag;
314 	alloc_meta->free_track_idx = (idx + 1) % KASAN_NR_FREE_STACKS;
315 #endif
316 
317 	set_track(&alloc_meta->free_track[idx], GFP_NOWAIT);
318 }
319 
320 void kasan_poison_slab(struct page *page)
321 {
322 	unsigned long i;
323 
324 	for (i = 0; i < compound_nr(page); i++)
325 		page_kasan_tag_reset(page + i);
326 	kasan_poison_shadow(page_address(page), page_size(page),
327 			KASAN_KMALLOC_REDZONE);
328 }
329 
330 void kasan_unpoison_object_data(struct kmem_cache *cache, void *object)
331 {
332 	kasan_unpoison_shadow(object, cache->object_size);
333 }
334 
335 void kasan_poison_object_data(struct kmem_cache *cache, void *object)
336 {
337 	kasan_poison_shadow(object,
338 			round_up(cache->object_size, KASAN_SHADOW_SCALE_SIZE),
339 			KASAN_KMALLOC_REDZONE);
340 }
341 
342 /*
343  * This function assigns a tag to an object considering the following:
344  * 1. A cache might have a constructor, which might save a pointer to a slab
345  *    object somewhere (e.g. in the object itself). We preassign a tag for
346  *    each object in caches with constructors during slab creation and reuse
347  *    the same tag each time a particular object is allocated.
348  * 2. A cache might be SLAB_TYPESAFE_BY_RCU, which means objects can be
349  *    accessed after being freed. We preassign tags for objects in these
350  *    caches as well.
351  * 3. For SLAB allocator we can't preassign tags randomly since the freelist
352  *    is stored as an array of indexes instead of a linked list. Assign tags
353  *    based on objects indexes, so that objects that are next to each other
354  *    get different tags.
355  */
356 static u8 assign_tag(struct kmem_cache *cache, const void *object,
357 			bool init, bool keep_tag)
358 {
359 	/*
360 	 * 1. When an object is kmalloc()'ed, two hooks are called:
361 	 *    kasan_slab_alloc() and kasan_kmalloc(). We assign the
362 	 *    tag only in the first one.
363 	 * 2. We reuse the same tag for krealloc'ed objects.
364 	 */
365 	if (keep_tag)
366 		return get_tag(object);
367 
368 	/*
369 	 * If the cache neither has a constructor nor has SLAB_TYPESAFE_BY_RCU
370 	 * set, assign a tag when the object is being allocated (init == false).
371 	 */
372 	if (!cache->ctor && !(cache->flags & SLAB_TYPESAFE_BY_RCU))
373 		return init ? KASAN_TAG_KERNEL : random_tag();
374 
375 	/* For caches that either have a constructor or SLAB_TYPESAFE_BY_RCU: */
376 #ifdef CONFIG_SLAB
377 	/* For SLAB assign tags based on the object index in the freelist. */
378 	return (u8)obj_to_index(cache, virt_to_page(object), (void *)object);
379 #else
380 	/*
381 	 * For SLUB assign a random tag during slab creation, otherwise reuse
382 	 * the already assigned tag.
383 	 */
384 	return init ? random_tag() : get_tag(object);
385 #endif
386 }
387 
388 void * __must_check kasan_init_slab_obj(struct kmem_cache *cache,
389 						const void *object)
390 {
391 	struct kasan_alloc_meta *alloc_info;
392 
393 	if (!(cache->flags & SLAB_KASAN))
394 		return (void *)object;
395 
396 	alloc_info = get_alloc_info(cache, object);
397 	__memset(alloc_info, 0, sizeof(*alloc_info));
398 
399 	if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
400 		object = set_tag(object,
401 				assign_tag(cache, object, true, false));
402 
403 	return (void *)object;
404 }
405 
406 static inline bool shadow_invalid(u8 tag, s8 shadow_byte)
407 {
408 	if (IS_ENABLED(CONFIG_KASAN_GENERIC))
409 		return shadow_byte < 0 ||
410 			shadow_byte >= KASAN_SHADOW_SCALE_SIZE;
411 
412 	/* else CONFIG_KASAN_SW_TAGS: */
413 	if ((u8)shadow_byte == KASAN_TAG_INVALID)
414 		return true;
415 	if ((tag != KASAN_TAG_KERNEL) && (tag != (u8)shadow_byte))
416 		return true;
417 
418 	return false;
419 }
420 
421 static bool __kasan_slab_free(struct kmem_cache *cache, void *object,
422 			      unsigned long ip, bool quarantine)
423 {
424 	s8 shadow_byte;
425 	u8 tag;
426 	void *tagged_object;
427 	unsigned long rounded_up_size;
428 
429 	tag = get_tag(object);
430 	tagged_object = object;
431 	object = reset_tag(object);
432 
433 	if (unlikely(nearest_obj(cache, virt_to_head_page(object), object) !=
434 	    object)) {
435 		kasan_report_invalid_free(tagged_object, ip);
436 		return true;
437 	}
438 
439 	/* RCU slabs could be legally used after free within the RCU period */
440 	if (unlikely(cache->flags & SLAB_TYPESAFE_BY_RCU))
441 		return false;
442 
443 	shadow_byte = READ_ONCE(*(s8 *)kasan_mem_to_shadow(object));
444 	if (shadow_invalid(tag, shadow_byte)) {
445 		kasan_report_invalid_free(tagged_object, ip);
446 		return true;
447 	}
448 
449 	rounded_up_size = round_up(cache->object_size, KASAN_SHADOW_SCALE_SIZE);
450 	kasan_poison_shadow(object, rounded_up_size, KASAN_KMALLOC_FREE);
451 
452 	if ((IS_ENABLED(CONFIG_KASAN_GENERIC) && !quarantine) ||
453 			unlikely(!(cache->flags & SLAB_KASAN)))
454 		return false;
455 
456 	kasan_set_free_info(cache, object, tag);
457 
458 	quarantine_put(get_free_info(cache, object), cache);
459 
460 	return IS_ENABLED(CONFIG_KASAN_GENERIC);
461 }
462 
463 bool kasan_slab_free(struct kmem_cache *cache, void *object, unsigned long ip)
464 {
465 	return __kasan_slab_free(cache, object, ip, true);
466 }
467 
468 static void *__kasan_kmalloc(struct kmem_cache *cache, const void *object,
469 				size_t size, gfp_t flags, bool keep_tag)
470 {
471 	unsigned long redzone_start;
472 	unsigned long redzone_end;
473 	u8 tag = 0xff;
474 
475 	if (gfpflags_allow_blocking(flags))
476 		quarantine_reduce();
477 
478 	if (unlikely(object == NULL))
479 		return NULL;
480 
481 	redzone_start = round_up((unsigned long)(object + size),
482 				KASAN_SHADOW_SCALE_SIZE);
483 	redzone_end = round_up((unsigned long)object + cache->object_size,
484 				KASAN_SHADOW_SCALE_SIZE);
485 
486 	if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
487 		tag = assign_tag(cache, object, false, keep_tag);
488 
489 	/* Tag is ignored in set_tag without CONFIG_KASAN_SW_TAGS */
490 	kasan_unpoison_shadow(set_tag(object, tag), size);
491 	kasan_poison_shadow((void *)redzone_start, redzone_end - redzone_start,
492 		KASAN_KMALLOC_REDZONE);
493 
494 	if (cache->flags & SLAB_KASAN)
495 		set_track(&get_alloc_info(cache, object)->alloc_track, flags);
496 
497 	return set_tag(object, tag);
498 }
499 
500 void * __must_check kasan_slab_alloc(struct kmem_cache *cache, void *object,
501 					gfp_t flags)
502 {
503 	return __kasan_kmalloc(cache, object, cache->object_size, flags, false);
504 }
505 
506 void * __must_check kasan_kmalloc(struct kmem_cache *cache, const void *object,
507 				size_t size, gfp_t flags)
508 {
509 	return __kasan_kmalloc(cache, object, size, flags, true);
510 }
511 EXPORT_SYMBOL(kasan_kmalloc);
512 
513 void * __must_check kasan_kmalloc_large(const void *ptr, size_t size,
514 						gfp_t flags)
515 {
516 	struct page *page;
517 	unsigned long redzone_start;
518 	unsigned long redzone_end;
519 
520 	if (gfpflags_allow_blocking(flags))
521 		quarantine_reduce();
522 
523 	if (unlikely(ptr == NULL))
524 		return NULL;
525 
526 	page = virt_to_page(ptr);
527 	redzone_start = round_up((unsigned long)(ptr + size),
528 				KASAN_SHADOW_SCALE_SIZE);
529 	redzone_end = (unsigned long)ptr + page_size(page);
530 
531 	kasan_unpoison_shadow(ptr, size);
532 	kasan_poison_shadow((void *)redzone_start, redzone_end - redzone_start,
533 		KASAN_PAGE_REDZONE);
534 
535 	return (void *)ptr;
536 }
537 
538 void * __must_check kasan_krealloc(const void *object, size_t size, gfp_t flags)
539 {
540 	struct page *page;
541 
542 	if (unlikely(object == ZERO_SIZE_PTR))
543 		return (void *)object;
544 
545 	page = virt_to_head_page(object);
546 
547 	if (unlikely(!PageSlab(page)))
548 		return kasan_kmalloc_large(object, size, flags);
549 	else
550 		return __kasan_kmalloc(page->slab_cache, object, size,
551 						flags, true);
552 }
553 
554 void kasan_poison_kfree(void *ptr, unsigned long ip)
555 {
556 	struct page *page;
557 
558 	page = virt_to_head_page(ptr);
559 
560 	if (unlikely(!PageSlab(page))) {
561 		if (ptr != page_address(page)) {
562 			kasan_report_invalid_free(ptr, ip);
563 			return;
564 		}
565 		kasan_poison_shadow(ptr, page_size(page), KASAN_FREE_PAGE);
566 	} else {
567 		__kasan_slab_free(page->slab_cache, ptr, ip, false);
568 	}
569 }
570 
571 void kasan_kfree_large(void *ptr, unsigned long ip)
572 {
573 	if (ptr != page_address(virt_to_head_page(ptr)))
574 		kasan_report_invalid_free(ptr, ip);
575 	/* The object will be poisoned by page_alloc. */
576 }
577 
578 #ifndef CONFIG_KASAN_VMALLOC
579 int kasan_module_alloc(void *addr, size_t size)
580 {
581 	void *ret;
582 	size_t scaled_size;
583 	size_t shadow_size;
584 	unsigned long shadow_start;
585 
586 	shadow_start = (unsigned long)kasan_mem_to_shadow(addr);
587 	scaled_size = (size + KASAN_SHADOW_MASK) >> KASAN_SHADOW_SCALE_SHIFT;
588 	shadow_size = round_up(scaled_size, PAGE_SIZE);
589 
590 	if (WARN_ON(!PAGE_ALIGNED(shadow_start)))
591 		return -EINVAL;
592 
593 	ret = __vmalloc_node_range(shadow_size, 1, shadow_start,
594 			shadow_start + shadow_size,
595 			GFP_KERNEL,
596 			PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE,
597 			__builtin_return_address(0));
598 
599 	if (ret) {
600 		__memset(ret, KASAN_SHADOW_INIT, shadow_size);
601 		find_vm_area(addr)->flags |= VM_KASAN;
602 		kmemleak_ignore(ret);
603 		return 0;
604 	}
605 
606 	return -ENOMEM;
607 }
608 
609 void kasan_free_shadow(const struct vm_struct *vm)
610 {
611 	if (vm->flags & VM_KASAN)
612 		vfree(kasan_mem_to_shadow(vm->addr));
613 }
614 #endif
615 
616 extern void __kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip);
617 extern bool report_enabled(void);
618 
619 bool kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip)
620 {
621 	unsigned long flags = user_access_save();
622 	bool ret = false;
623 
624 	if (likely(report_enabled())) {
625 		__kasan_report(addr, size, is_write, ip);
626 		ret = true;
627 	}
628 
629 	user_access_restore(flags);
630 
631 	return ret;
632 }
633 
634 #ifdef CONFIG_MEMORY_HOTPLUG
635 static bool shadow_mapped(unsigned long addr)
636 {
637 	pgd_t *pgd = pgd_offset_k(addr);
638 	p4d_t *p4d;
639 	pud_t *pud;
640 	pmd_t *pmd;
641 	pte_t *pte;
642 
643 	if (pgd_none(*pgd))
644 		return false;
645 	p4d = p4d_offset(pgd, addr);
646 	if (p4d_none(*p4d))
647 		return false;
648 	pud = pud_offset(p4d, addr);
649 	if (pud_none(*pud))
650 		return false;
651 
652 	/*
653 	 * We can't use pud_large() or pud_huge(), the first one is
654 	 * arch-specific, the last one depends on HUGETLB_PAGE.  So let's abuse
655 	 * pud_bad(), if pud is bad then it's bad because it's huge.
656 	 */
657 	if (pud_bad(*pud))
658 		return true;
659 	pmd = pmd_offset(pud, addr);
660 	if (pmd_none(*pmd))
661 		return false;
662 
663 	if (pmd_bad(*pmd))
664 		return true;
665 	pte = pte_offset_kernel(pmd, addr);
666 	return !pte_none(*pte);
667 }
668 
669 static int __meminit kasan_mem_notifier(struct notifier_block *nb,
670 			unsigned long action, void *data)
671 {
672 	struct memory_notify *mem_data = data;
673 	unsigned long nr_shadow_pages, start_kaddr, shadow_start;
674 	unsigned long shadow_end, shadow_size;
675 
676 	nr_shadow_pages = mem_data->nr_pages >> KASAN_SHADOW_SCALE_SHIFT;
677 	start_kaddr = (unsigned long)pfn_to_kaddr(mem_data->start_pfn);
678 	shadow_start = (unsigned long)kasan_mem_to_shadow((void *)start_kaddr);
679 	shadow_size = nr_shadow_pages << PAGE_SHIFT;
680 	shadow_end = shadow_start + shadow_size;
681 
682 	if (WARN_ON(mem_data->nr_pages % KASAN_SHADOW_SCALE_SIZE) ||
683 		WARN_ON(start_kaddr % (KASAN_SHADOW_SCALE_SIZE << PAGE_SHIFT)))
684 		return NOTIFY_BAD;
685 
686 	switch (action) {
687 	case MEM_GOING_ONLINE: {
688 		void *ret;
689 
690 		/*
691 		 * If shadow is mapped already than it must have been mapped
692 		 * during the boot. This could happen if we onlining previously
693 		 * offlined memory.
694 		 */
695 		if (shadow_mapped(shadow_start))
696 			return NOTIFY_OK;
697 
698 		ret = __vmalloc_node_range(shadow_size, PAGE_SIZE, shadow_start,
699 					shadow_end, GFP_KERNEL,
700 					PAGE_KERNEL, VM_NO_GUARD,
701 					pfn_to_nid(mem_data->start_pfn),
702 					__builtin_return_address(0));
703 		if (!ret)
704 			return NOTIFY_BAD;
705 
706 		kmemleak_ignore(ret);
707 		return NOTIFY_OK;
708 	}
709 	case MEM_CANCEL_ONLINE:
710 	case MEM_OFFLINE: {
711 		struct vm_struct *vm;
712 
713 		/*
714 		 * shadow_start was either mapped during boot by kasan_init()
715 		 * or during memory online by __vmalloc_node_range().
716 		 * In the latter case we can use vfree() to free shadow.
717 		 * Non-NULL result of the find_vm_area() will tell us if
718 		 * that was the second case.
719 		 *
720 		 * Currently it's not possible to free shadow mapped
721 		 * during boot by kasan_init(). It's because the code
722 		 * to do that hasn't been written yet. So we'll just
723 		 * leak the memory.
724 		 */
725 		vm = find_vm_area((void *)shadow_start);
726 		if (vm)
727 			vfree((void *)shadow_start);
728 	}
729 	}
730 
731 	return NOTIFY_OK;
732 }
733 
734 static int __init kasan_memhotplug_init(void)
735 {
736 	hotplug_memory_notifier(kasan_mem_notifier, 0);
737 
738 	return 0;
739 }
740 
741 core_initcall(kasan_memhotplug_init);
742 #endif
743 
744 #ifdef CONFIG_KASAN_VMALLOC
745 static int kasan_populate_vmalloc_pte(pte_t *ptep, unsigned long addr,
746 				      void *unused)
747 {
748 	unsigned long page;
749 	pte_t pte;
750 
751 	if (likely(!pte_none(*ptep)))
752 		return 0;
753 
754 	page = __get_free_page(GFP_KERNEL);
755 	if (!page)
756 		return -ENOMEM;
757 
758 	memset((void *)page, KASAN_VMALLOC_INVALID, PAGE_SIZE);
759 	pte = pfn_pte(PFN_DOWN(__pa(page)), PAGE_KERNEL);
760 
761 	spin_lock(&init_mm.page_table_lock);
762 	if (likely(pte_none(*ptep))) {
763 		set_pte_at(&init_mm, addr, ptep, pte);
764 		page = 0;
765 	}
766 	spin_unlock(&init_mm.page_table_lock);
767 	if (page)
768 		free_page(page);
769 	return 0;
770 }
771 
772 int kasan_populate_vmalloc(unsigned long addr, unsigned long size)
773 {
774 	unsigned long shadow_start, shadow_end;
775 	int ret;
776 
777 	if (!is_vmalloc_or_module_addr((void *)addr))
778 		return 0;
779 
780 	shadow_start = (unsigned long)kasan_mem_to_shadow((void *)addr);
781 	shadow_start = ALIGN_DOWN(shadow_start, PAGE_SIZE);
782 	shadow_end = (unsigned long)kasan_mem_to_shadow((void *)addr + size);
783 	shadow_end = ALIGN(shadow_end, PAGE_SIZE);
784 
785 	ret = apply_to_page_range(&init_mm, shadow_start,
786 				  shadow_end - shadow_start,
787 				  kasan_populate_vmalloc_pte, NULL);
788 	if (ret)
789 		return ret;
790 
791 	flush_cache_vmap(shadow_start, shadow_end);
792 
793 	/*
794 	 * We need to be careful about inter-cpu effects here. Consider:
795 	 *
796 	 *   CPU#0				  CPU#1
797 	 * WRITE_ONCE(p, vmalloc(100));		while (x = READ_ONCE(p)) ;
798 	 *					p[99] = 1;
799 	 *
800 	 * With compiler instrumentation, that ends up looking like this:
801 	 *
802 	 *   CPU#0				  CPU#1
803 	 * // vmalloc() allocates memory
804 	 * // let a = area->addr
805 	 * // we reach kasan_populate_vmalloc
806 	 * // and call kasan_unpoison_shadow:
807 	 * STORE shadow(a), unpoison_val
808 	 * ...
809 	 * STORE shadow(a+99), unpoison_val	x = LOAD p
810 	 * // rest of vmalloc process		<data dependency>
811 	 * STORE p, a				LOAD shadow(x+99)
812 	 *
813 	 * If there is no barrier between the end of unpoisioning the shadow
814 	 * and the store of the result to p, the stores could be committed
815 	 * in a different order by CPU#0, and CPU#1 could erroneously observe
816 	 * poison in the shadow.
817 	 *
818 	 * We need some sort of barrier between the stores.
819 	 *
820 	 * In the vmalloc() case, this is provided by a smp_wmb() in
821 	 * clear_vm_uninitialized_flag(). In the per-cpu allocator and in
822 	 * get_vm_area() and friends, the caller gets shadow allocated but
823 	 * doesn't have any pages mapped into the virtual address space that
824 	 * has been reserved. Mapping those pages in will involve taking and
825 	 * releasing a page-table lock, which will provide the barrier.
826 	 */
827 
828 	return 0;
829 }
830 
831 /*
832  * Poison the shadow for a vmalloc region. Called as part of the
833  * freeing process at the time the region is freed.
834  */
835 void kasan_poison_vmalloc(const void *start, unsigned long size)
836 {
837 	if (!is_vmalloc_or_module_addr(start))
838 		return;
839 
840 	size = round_up(size, KASAN_SHADOW_SCALE_SIZE);
841 	kasan_poison_shadow(start, size, KASAN_VMALLOC_INVALID);
842 }
843 
844 void kasan_unpoison_vmalloc(const void *start, unsigned long size)
845 {
846 	if (!is_vmalloc_or_module_addr(start))
847 		return;
848 
849 	kasan_unpoison_shadow(start, size);
850 }
851 
852 static int kasan_depopulate_vmalloc_pte(pte_t *ptep, unsigned long addr,
853 					void *unused)
854 {
855 	unsigned long page;
856 
857 	page = (unsigned long)__va(pte_pfn(*ptep) << PAGE_SHIFT);
858 
859 	spin_lock(&init_mm.page_table_lock);
860 
861 	if (likely(!pte_none(*ptep))) {
862 		pte_clear(&init_mm, addr, ptep);
863 		free_page(page);
864 	}
865 	spin_unlock(&init_mm.page_table_lock);
866 
867 	return 0;
868 }
869 
870 /*
871  * Release the backing for the vmalloc region [start, end), which
872  * lies within the free region [free_region_start, free_region_end).
873  *
874  * This can be run lazily, long after the region was freed. It runs
875  * under vmap_area_lock, so it's not safe to interact with the vmalloc/vmap
876  * infrastructure.
877  *
878  * How does this work?
879  * -------------------
880  *
881  * We have a region that is page aligned, labelled as A.
882  * That might not map onto the shadow in a way that is page-aligned:
883  *
884  *                    start                     end
885  *                    v                         v
886  * |????????|????????|AAAAAAAA|AA....AA|AAAAAAAA|????????| < vmalloc
887  *  -------- -------- --------          -------- --------
888  *      |        |       |                 |        |
889  *      |        |       |         /-------/        |
890  *      \-------\|/------/         |/---------------/
891  *              |||                ||
892  *             |??AAAAAA|AAAAAAAA|AA??????|                < shadow
893  *                 (1)      (2)      (3)
894  *
895  * First we align the start upwards and the end downwards, so that the
896  * shadow of the region aligns with shadow page boundaries. In the
897  * example, this gives us the shadow page (2). This is the shadow entirely
898  * covered by this allocation.
899  *
900  * Then we have the tricky bits. We want to know if we can free the
901  * partially covered shadow pages - (1) and (3) in the example. For this,
902  * we are given the start and end of the free region that contains this
903  * allocation. Extending our previous example, we could have:
904  *
905  *  free_region_start                                    free_region_end
906  *  |                 start                     end      |
907  *  v                 v                         v        v
908  * |FFFFFFFF|FFFFFFFF|AAAAAAAA|AA....AA|AAAAAAAA|FFFFFFFF| < vmalloc
909  *  -------- -------- --------          -------- --------
910  *      |        |       |                 |        |
911  *      |        |       |         /-------/        |
912  *      \-------\|/------/         |/---------------/
913  *              |||                ||
914  *             |FFAAAAAA|AAAAAAAA|AAF?????|                < shadow
915  *                 (1)      (2)      (3)
916  *
917  * Once again, we align the start of the free region up, and the end of
918  * the free region down so that the shadow is page aligned. So we can free
919  * page (1) - we know no allocation currently uses anything in that page,
920  * because all of it is in the vmalloc free region. But we cannot free
921  * page (3), because we can't be sure that the rest of it is unused.
922  *
923  * We only consider pages that contain part of the original region for
924  * freeing: we don't try to free other pages from the free region or we'd
925  * end up trying to free huge chunks of virtual address space.
926  *
927  * Concurrency
928  * -----------
929  *
930  * How do we know that we're not freeing a page that is simultaneously
931  * being used for a fresh allocation in kasan_populate_vmalloc(_pte)?
932  *
933  * We _can_ have kasan_release_vmalloc and kasan_populate_vmalloc running
934  * at the same time. While we run under free_vmap_area_lock, the population
935  * code does not.
936  *
937  * free_vmap_area_lock instead operates to ensure that the larger range
938  * [free_region_start, free_region_end) is safe: because __alloc_vmap_area and
939  * the per-cpu region-finding algorithm both run under free_vmap_area_lock,
940  * no space identified as free will become used while we are running. This
941  * means that so long as we are careful with alignment and only free shadow
942  * pages entirely covered by the free region, we will not run in to any
943  * trouble - any simultaneous allocations will be for disjoint regions.
944  */
945 void kasan_release_vmalloc(unsigned long start, unsigned long end,
946 			   unsigned long free_region_start,
947 			   unsigned long free_region_end)
948 {
949 	void *shadow_start, *shadow_end;
950 	unsigned long region_start, region_end;
951 	unsigned long size;
952 
953 	region_start = ALIGN(start, PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
954 	region_end = ALIGN_DOWN(end, PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
955 
956 	free_region_start = ALIGN(free_region_start,
957 				  PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
958 
959 	if (start != region_start &&
960 	    free_region_start < region_start)
961 		region_start -= PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE;
962 
963 	free_region_end = ALIGN_DOWN(free_region_end,
964 				     PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
965 
966 	if (end != region_end &&
967 	    free_region_end > region_end)
968 		region_end += PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE;
969 
970 	shadow_start = kasan_mem_to_shadow((void *)region_start);
971 	shadow_end = kasan_mem_to_shadow((void *)region_end);
972 
973 	if (shadow_end > shadow_start) {
974 		size = shadow_end - shadow_start;
975 		apply_to_existing_page_range(&init_mm,
976 					     (unsigned long)shadow_start,
977 					     size, kasan_depopulate_vmalloc_pte,
978 					     NULL);
979 		flush_tlb_kernel_range((unsigned long)shadow_start,
980 				       (unsigned long)shadow_end);
981 	}
982 }
983 #endif
984