1 // SPDX-License-Identifier: GPL-2.0
3 * This file contains common generic and tag-based KASAN code.
5 * Copyright (c) 2014 Samsung Electronics Co., Ltd.
6 * Author: Andrey Ryabinin <ryabinin.a.a@gmail.com>
8 * Some code borrowed from https://github.com/xairy/kasan-prototype by
9 * Andrey Konovalov <andreyknvl@gmail.com>
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.
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>
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>
37 #include <asm/cacheflush.h>
38 #include <asm/tlbflush.h>
43 static inline depot_stack_handle_t save_stack(gfp_t flags)
45 unsigned long entries[KASAN_STACK_DEPTH];
46 unsigned int nr_entries;
48 nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0);
49 nr_entries = filter_irq_stacks(entries, nr_entries);
50 return stack_depot_save(entries, nr_entries, flags);
53 static inline void set_track(struct kasan_track *track, gfp_t flags)
55 track->pid = current->pid;
56 track->stack = save_stack(flags);
59 void kasan_enable_current(void)
61 current->kasan_depth++;
64 void kasan_disable_current(void)
66 current->kasan_depth--;
69 bool __kasan_check_read(const volatile void *p, unsigned int size)
71 return check_memory_region((unsigned long)p, size, false, _RET_IP_);
73 EXPORT_SYMBOL(__kasan_check_read);
75 bool __kasan_check_write(const volatile void *p, unsigned int size)
77 return check_memory_region((unsigned long)p, size, true, _RET_IP_);
79 EXPORT_SYMBOL(__kasan_check_write);
82 void *memset(void *addr, int c, size_t len)
84 if (!check_memory_region((unsigned long)addr, len, true, _RET_IP_))
87 return __memset(addr, c, len);
90 #ifdef __HAVE_ARCH_MEMMOVE
92 void *memmove(void *dest, const void *src, size_t len)
94 if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
95 !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
98 return __memmove(dest, src, len);
103 void *memcpy(void *dest, const void *src, size_t len)
105 if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
106 !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
109 return __memcpy(dest, src, len);
113 * Poisons the shadow memory for 'size' bytes starting from 'addr'.
114 * Memory addresses should be aligned to KASAN_SHADOW_SCALE_SIZE.
116 void kasan_poison_shadow(const void *address, size_t size, u8 value)
118 void *shadow_start, *shadow_end;
121 * Perform shadow offset calculation based on untagged address, as
122 * some of the callers (e.g. kasan_poison_object_data) pass tagged
123 * addresses to this function.
125 address = reset_tag(address);
127 shadow_start = kasan_mem_to_shadow(address);
128 shadow_end = kasan_mem_to_shadow(address + size);
130 __memset(shadow_start, value, shadow_end - shadow_start);
133 void kasan_unpoison_shadow(const void *address, size_t size)
135 u8 tag = get_tag(address);
138 * Perform shadow offset calculation based on untagged address, as
139 * some of the callers (e.g. kasan_unpoison_object_data) pass tagged
140 * addresses to this function.
142 address = reset_tag(address);
144 kasan_poison_shadow(address, size, tag);
146 if (size & KASAN_SHADOW_MASK) {
147 u8 *shadow = (u8 *)kasan_mem_to_shadow(address + size);
149 if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
152 *shadow = size & KASAN_SHADOW_MASK;
156 static void __kasan_unpoison_stack(struct task_struct *task, const void *sp)
158 void *base = task_stack_page(task);
159 size_t size = sp - base;
161 kasan_unpoison_shadow(base, size);
164 /* Unpoison the entire stack for a task. */
165 void kasan_unpoison_task_stack(struct task_struct *task)
167 __kasan_unpoison_stack(task, task_stack_page(task) + THREAD_SIZE);
170 /* Unpoison the stack for the current task beyond a watermark sp value. */
171 asmlinkage void kasan_unpoison_task_stack_below(const void *watermark)
174 * Calculate the task stack base address. Avoid using 'current'
175 * because this function is called by early resume code which hasn't
176 * yet set up the percpu register (%gs).
178 void *base = (void *)((unsigned long)watermark & ~(THREAD_SIZE - 1));
180 kasan_unpoison_shadow(base, watermark - base);
184 * Clear all poison for the region between the current SP and a provided
185 * watermark value, as is sometimes required prior to hand-crafted asm function
186 * returns in the middle of functions.
188 void kasan_unpoison_stack_above_sp_to(const void *watermark)
190 const void *sp = __builtin_frame_address(0);
191 size_t size = watermark - sp;
193 if (WARN_ON(sp > watermark))
195 kasan_unpoison_shadow(sp, size);
198 void kasan_alloc_pages(struct page *page, unsigned int order)
203 if (unlikely(PageHighMem(page)))
207 for (i = 0; i < (1 << order); i++)
208 page_kasan_tag_set(page + i, tag);
209 kasan_unpoison_shadow(page_address(page), PAGE_SIZE << order);
212 void kasan_free_pages(struct page *page, unsigned int order)
214 if (likely(!PageHighMem(page)))
215 kasan_poison_shadow(page_address(page),
221 * Adaptive redzone policy taken from the userspace AddressSanitizer runtime.
222 * For larger allocations larger redzones are used.
224 static inline unsigned int optimal_redzone(unsigned int object_size)
226 if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
230 object_size <= 64 - 16 ? 16 :
231 object_size <= 128 - 32 ? 32 :
232 object_size <= 512 - 64 ? 64 :
233 object_size <= 4096 - 128 ? 128 :
234 object_size <= (1 << 14) - 256 ? 256 :
235 object_size <= (1 << 15) - 512 ? 512 :
236 object_size <= (1 << 16) - 1024 ? 1024 : 2048;
239 void kasan_cache_create(struct kmem_cache *cache, unsigned int *size,
242 unsigned int orig_size = *size;
243 unsigned int redzone_size;
246 /* Add alloc meta. */
247 cache->kasan_info.alloc_meta_offset = *size;
248 *size += sizeof(struct kasan_alloc_meta);
251 if (IS_ENABLED(CONFIG_KASAN_GENERIC) &&
252 (cache->flags & SLAB_TYPESAFE_BY_RCU || cache->ctor ||
253 cache->object_size < sizeof(struct kasan_free_meta))) {
254 cache->kasan_info.free_meta_offset = *size;
255 *size += sizeof(struct kasan_free_meta);
258 redzone_size = optimal_redzone(cache->object_size);
259 redzone_adjust = redzone_size - (*size - cache->object_size);
260 if (redzone_adjust > 0)
261 *size += redzone_adjust;
263 *size = min_t(unsigned int, KMALLOC_MAX_SIZE,
264 max(*size, cache->object_size + redzone_size));
267 * If the metadata doesn't fit, don't enable KASAN at all.
269 if (*size <= cache->kasan_info.alloc_meta_offset ||
270 *size <= cache->kasan_info.free_meta_offset) {
271 cache->kasan_info.alloc_meta_offset = 0;
272 cache->kasan_info.free_meta_offset = 0;
277 *flags |= SLAB_KASAN;
280 size_t kasan_metadata_size(struct kmem_cache *cache)
282 return (cache->kasan_info.alloc_meta_offset ?
283 sizeof(struct kasan_alloc_meta) : 0) +
284 (cache->kasan_info.free_meta_offset ?
285 sizeof(struct kasan_free_meta) : 0);
288 struct kasan_alloc_meta *get_alloc_info(struct kmem_cache *cache,
291 return (void *)object + cache->kasan_info.alloc_meta_offset;
294 struct kasan_free_meta *get_free_info(struct kmem_cache *cache,
297 BUILD_BUG_ON(sizeof(struct kasan_free_meta) > 32);
298 return (void *)object + cache->kasan_info.free_meta_offset;
302 static void kasan_set_free_info(struct kmem_cache *cache,
303 void *object, u8 tag)
305 struct kasan_alloc_meta *alloc_meta;
308 alloc_meta = get_alloc_info(cache, object);
310 #ifdef CONFIG_KASAN_SW_TAGS_IDENTIFY
311 idx = alloc_meta->free_track_idx;
312 alloc_meta->free_pointer_tag[idx] = tag;
313 alloc_meta->free_track_idx = (idx + 1) % KASAN_NR_FREE_STACKS;
316 set_track(&alloc_meta->free_track[idx], GFP_NOWAIT);
319 void kasan_poison_slab(struct page *page)
323 for (i = 0; i < compound_nr(page); i++)
324 page_kasan_tag_reset(page + i);
325 kasan_poison_shadow(page_address(page), page_size(page),
326 KASAN_KMALLOC_REDZONE);
329 void kasan_unpoison_object_data(struct kmem_cache *cache, void *object)
331 kasan_unpoison_shadow(object, cache->object_size);
334 void kasan_poison_object_data(struct kmem_cache *cache, void *object)
336 kasan_poison_shadow(object,
337 round_up(cache->object_size, KASAN_SHADOW_SCALE_SIZE),
338 KASAN_KMALLOC_REDZONE);
342 * This function assigns a tag to an object considering the following:
343 * 1. A cache might have a constructor, which might save a pointer to a slab
344 * object somewhere (e.g. in the object itself). We preassign a tag for
345 * each object in caches with constructors during slab creation and reuse
346 * the same tag each time a particular object is allocated.
347 * 2. A cache might be SLAB_TYPESAFE_BY_RCU, which means objects can be
348 * accessed after being freed. We preassign tags for objects in these
350 * 3. For SLAB allocator we can't preassign tags randomly since the freelist
351 * is stored as an array of indexes instead of a linked list. Assign tags
352 * based on objects indexes, so that objects that are next to each other
353 * get different tags.
355 static u8 assign_tag(struct kmem_cache *cache, const void *object,
356 bool init, bool keep_tag)
359 * 1. When an object is kmalloc()'ed, two hooks are called:
360 * kasan_slab_alloc() and kasan_kmalloc(). We assign the
361 * tag only in the first one.
362 * 2. We reuse the same tag for krealloc'ed objects.
365 return get_tag(object);
368 * If the cache neither has a constructor nor has SLAB_TYPESAFE_BY_RCU
369 * set, assign a tag when the object is being allocated (init == false).
371 if (!cache->ctor && !(cache->flags & SLAB_TYPESAFE_BY_RCU))
372 return init ? KASAN_TAG_KERNEL : random_tag();
374 /* For caches that either have a constructor or SLAB_TYPESAFE_BY_RCU: */
376 /* For SLAB assign tags based on the object index in the freelist. */
377 return (u8)obj_to_index(cache, virt_to_page(object), (void *)object);
380 * For SLUB assign a random tag during slab creation, otherwise reuse
381 * the already assigned tag.
383 return init ? random_tag() : get_tag(object);
387 void * __must_check kasan_init_slab_obj(struct kmem_cache *cache,
390 struct kasan_alloc_meta *alloc_info;
392 if (!(cache->flags & SLAB_KASAN))
393 return (void *)object;
395 alloc_info = get_alloc_info(cache, object);
396 __memset(alloc_info, 0, sizeof(*alloc_info));
398 if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
399 object = set_tag(object,
400 assign_tag(cache, object, true, false));
402 return (void *)object;
405 static inline bool shadow_invalid(u8 tag, s8 shadow_byte)
407 if (IS_ENABLED(CONFIG_KASAN_GENERIC))
408 return shadow_byte < 0 ||
409 shadow_byte >= KASAN_SHADOW_SCALE_SIZE;
411 /* else CONFIG_KASAN_SW_TAGS: */
412 if ((u8)shadow_byte == KASAN_TAG_INVALID)
414 if ((tag != KASAN_TAG_KERNEL) && (tag != (u8)shadow_byte))
420 static bool __kasan_slab_free(struct kmem_cache *cache, void *object,
421 unsigned long ip, bool quarantine)
426 unsigned long rounded_up_size;
428 tag = get_tag(object);
429 tagged_object = object;
430 object = reset_tag(object);
432 if (unlikely(nearest_obj(cache, virt_to_head_page(object), object) !=
434 kasan_report_invalid_free(tagged_object, ip);
438 /* RCU slabs could be legally used after free within the RCU period */
439 if (unlikely(cache->flags & SLAB_TYPESAFE_BY_RCU))
442 shadow_byte = READ_ONCE(*(s8 *)kasan_mem_to_shadow(object));
443 if (shadow_invalid(tag, shadow_byte)) {
444 kasan_report_invalid_free(tagged_object, ip);
448 rounded_up_size = round_up(cache->object_size, KASAN_SHADOW_SCALE_SIZE);
449 kasan_poison_shadow(object, rounded_up_size, KASAN_KMALLOC_FREE);
451 if ((IS_ENABLED(CONFIG_KASAN_GENERIC) && !quarantine) ||
452 unlikely(!(cache->flags & SLAB_KASAN)))
455 kasan_set_free_info(cache, object, tag);
457 quarantine_put(get_free_info(cache, object), cache);
459 return IS_ENABLED(CONFIG_KASAN_GENERIC);
462 bool kasan_slab_free(struct kmem_cache *cache, void *object, unsigned long ip)
464 return __kasan_slab_free(cache, object, ip, true);
467 static void *__kasan_kmalloc(struct kmem_cache *cache, const void *object,
468 size_t size, gfp_t flags, bool keep_tag)
470 unsigned long redzone_start;
471 unsigned long redzone_end;
474 if (gfpflags_allow_blocking(flags))
477 if (unlikely(object == NULL))
480 redzone_start = round_up((unsigned long)(object + size),
481 KASAN_SHADOW_SCALE_SIZE);
482 redzone_end = round_up((unsigned long)object + cache->object_size,
483 KASAN_SHADOW_SCALE_SIZE);
485 if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
486 tag = assign_tag(cache, object, false, keep_tag);
488 /* Tag is ignored in set_tag without CONFIG_KASAN_SW_TAGS */
489 kasan_unpoison_shadow(set_tag(object, tag), size);
490 kasan_poison_shadow((void *)redzone_start, redzone_end - redzone_start,
491 KASAN_KMALLOC_REDZONE);
493 if (cache->flags & SLAB_KASAN)
494 set_track(&get_alloc_info(cache, object)->alloc_track, flags);
496 return set_tag(object, tag);
499 void * __must_check kasan_slab_alloc(struct kmem_cache *cache, void *object,
502 return __kasan_kmalloc(cache, object, cache->object_size, flags, false);
505 void * __must_check kasan_kmalloc(struct kmem_cache *cache, const void *object,
506 size_t size, gfp_t flags)
508 return __kasan_kmalloc(cache, object, size, flags, true);
510 EXPORT_SYMBOL(kasan_kmalloc);
512 void * __must_check kasan_kmalloc_large(const void *ptr, size_t size,
516 unsigned long redzone_start;
517 unsigned long redzone_end;
519 if (gfpflags_allow_blocking(flags))
522 if (unlikely(ptr == NULL))
525 page = virt_to_page(ptr);
526 redzone_start = round_up((unsigned long)(ptr + size),
527 KASAN_SHADOW_SCALE_SIZE);
528 redzone_end = (unsigned long)ptr + page_size(page);
530 kasan_unpoison_shadow(ptr, size);
531 kasan_poison_shadow((void *)redzone_start, redzone_end - redzone_start,
537 void * __must_check kasan_krealloc(const void *object, size_t size, gfp_t flags)
541 if (unlikely(object == ZERO_SIZE_PTR))
542 return (void *)object;
544 page = virt_to_head_page(object);
546 if (unlikely(!PageSlab(page)))
547 return kasan_kmalloc_large(object, size, flags);
549 return __kasan_kmalloc(page->slab_cache, object, size,
553 void kasan_poison_kfree(void *ptr, unsigned long ip)
557 page = virt_to_head_page(ptr);
559 if (unlikely(!PageSlab(page))) {
560 if (ptr != page_address(page)) {
561 kasan_report_invalid_free(ptr, ip);
564 kasan_poison_shadow(ptr, page_size(page), KASAN_FREE_PAGE);
566 __kasan_slab_free(page->slab_cache, ptr, ip, false);
570 void kasan_kfree_large(void *ptr, unsigned long ip)
572 if (ptr != page_address(virt_to_head_page(ptr)))
573 kasan_report_invalid_free(ptr, ip);
574 /* The object will be poisoned by page_alloc. */
577 #ifndef CONFIG_KASAN_VMALLOC
578 int kasan_module_alloc(void *addr, size_t size)
583 unsigned long shadow_start;
585 shadow_start = (unsigned long)kasan_mem_to_shadow(addr);
586 scaled_size = (size + KASAN_SHADOW_MASK) >> KASAN_SHADOW_SCALE_SHIFT;
587 shadow_size = round_up(scaled_size, PAGE_SIZE);
589 if (WARN_ON(!PAGE_ALIGNED(shadow_start)))
592 ret = __vmalloc_node_range(shadow_size, 1, shadow_start,
593 shadow_start + shadow_size,
595 PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE,
596 __builtin_return_address(0));
599 __memset(ret, KASAN_SHADOW_INIT, shadow_size);
600 find_vm_area(addr)->flags |= VM_KASAN;
601 kmemleak_ignore(ret);
608 void kasan_free_shadow(const struct vm_struct *vm)
610 if (vm->flags & VM_KASAN)
611 vfree(kasan_mem_to_shadow(vm->addr));
615 #ifdef CONFIG_MEMORY_HOTPLUG
616 static bool shadow_mapped(unsigned long addr)
618 pgd_t *pgd = pgd_offset_k(addr);
626 p4d = p4d_offset(pgd, addr);
629 pud = pud_offset(p4d, addr);
634 * We can't use pud_large() or pud_huge(), the first one is
635 * arch-specific, the last one depends on HUGETLB_PAGE. So let's abuse
636 * pud_bad(), if pud is bad then it's bad because it's huge.
640 pmd = pmd_offset(pud, addr);
646 pte = pte_offset_kernel(pmd, addr);
647 return !pte_none(*pte);
650 static int __meminit kasan_mem_notifier(struct notifier_block *nb,
651 unsigned long action, void *data)
653 struct memory_notify *mem_data = data;
654 unsigned long nr_shadow_pages, start_kaddr, shadow_start;
655 unsigned long shadow_end, shadow_size;
657 nr_shadow_pages = mem_data->nr_pages >> KASAN_SHADOW_SCALE_SHIFT;
658 start_kaddr = (unsigned long)pfn_to_kaddr(mem_data->start_pfn);
659 shadow_start = (unsigned long)kasan_mem_to_shadow((void *)start_kaddr);
660 shadow_size = nr_shadow_pages << PAGE_SHIFT;
661 shadow_end = shadow_start + shadow_size;
663 if (WARN_ON(mem_data->nr_pages % KASAN_SHADOW_SCALE_SIZE) ||
664 WARN_ON(start_kaddr % (KASAN_SHADOW_SCALE_SIZE << PAGE_SHIFT)))
668 case MEM_GOING_ONLINE: {
672 * If shadow is mapped already than it must have been mapped
673 * during the boot. This could happen if we onlining previously
676 if (shadow_mapped(shadow_start))
679 ret = __vmalloc_node_range(shadow_size, PAGE_SIZE, shadow_start,
680 shadow_end, GFP_KERNEL,
681 PAGE_KERNEL, VM_NO_GUARD,
682 pfn_to_nid(mem_data->start_pfn),
683 __builtin_return_address(0));
687 kmemleak_ignore(ret);
690 case MEM_CANCEL_ONLINE:
692 struct vm_struct *vm;
695 * shadow_start was either mapped during boot by kasan_init()
696 * or during memory online by __vmalloc_node_range().
697 * In the latter case we can use vfree() to free shadow.
698 * Non-NULL result of the find_vm_area() will tell us if
699 * that was the second case.
701 * Currently it's not possible to free shadow mapped
702 * during boot by kasan_init(). It's because the code
703 * to do that hasn't been written yet. So we'll just
706 vm = find_vm_area((void *)shadow_start);
708 vfree((void *)shadow_start);
715 static int __init kasan_memhotplug_init(void)
717 hotplug_memory_notifier(kasan_mem_notifier, 0);
722 core_initcall(kasan_memhotplug_init);
725 #ifdef CONFIG_KASAN_VMALLOC
726 static int kasan_populate_vmalloc_pte(pte_t *ptep, unsigned long addr,
732 if (likely(!pte_none(*ptep)))
735 page = __get_free_page(GFP_KERNEL);
739 memset((void *)page, KASAN_VMALLOC_INVALID, PAGE_SIZE);
740 pte = pfn_pte(PFN_DOWN(__pa(page)), PAGE_KERNEL);
742 spin_lock(&init_mm.page_table_lock);
743 if (likely(pte_none(*ptep))) {
744 set_pte_at(&init_mm, addr, ptep, pte);
747 spin_unlock(&init_mm.page_table_lock);
753 int kasan_populate_vmalloc(unsigned long addr, unsigned long size)
755 unsigned long shadow_start, shadow_end;
758 if (!is_vmalloc_or_module_addr((void *)addr))
761 shadow_start = (unsigned long)kasan_mem_to_shadow((void *)addr);
762 shadow_start = ALIGN_DOWN(shadow_start, PAGE_SIZE);
763 shadow_end = (unsigned long)kasan_mem_to_shadow((void *)addr + size);
764 shadow_end = ALIGN(shadow_end, PAGE_SIZE);
766 ret = apply_to_page_range(&init_mm, shadow_start,
767 shadow_end - shadow_start,
768 kasan_populate_vmalloc_pte, NULL);
772 flush_cache_vmap(shadow_start, shadow_end);
775 * We need to be careful about inter-cpu effects here. Consider:
778 * WRITE_ONCE(p, vmalloc(100)); while (x = READ_ONCE(p)) ;
781 * With compiler instrumentation, that ends up looking like this:
784 * // vmalloc() allocates memory
785 * // let a = area->addr
786 * // we reach kasan_populate_vmalloc
787 * // and call kasan_unpoison_shadow:
788 * STORE shadow(a), unpoison_val
790 * STORE shadow(a+99), unpoison_val x = LOAD p
791 * // rest of vmalloc process <data dependency>
792 * STORE p, a LOAD shadow(x+99)
794 * If there is no barrier between the end of unpoisioning the shadow
795 * and the store of the result to p, the stores could be committed
796 * in a different order by CPU#0, and CPU#1 could erroneously observe
797 * poison in the shadow.
799 * We need some sort of barrier between the stores.
801 * In the vmalloc() case, this is provided by a smp_wmb() in
802 * clear_vm_uninitialized_flag(). In the per-cpu allocator and in
803 * get_vm_area() and friends, the caller gets shadow allocated but
804 * doesn't have any pages mapped into the virtual address space that
805 * has been reserved. Mapping those pages in will involve taking and
806 * releasing a page-table lock, which will provide the barrier.
813 * Poison the shadow for a vmalloc region. Called as part of the
814 * freeing process at the time the region is freed.
816 void kasan_poison_vmalloc(const void *start, unsigned long size)
818 if (!is_vmalloc_or_module_addr(start))
821 size = round_up(size, KASAN_SHADOW_SCALE_SIZE);
822 kasan_poison_shadow(start, size, KASAN_VMALLOC_INVALID);
825 void kasan_unpoison_vmalloc(const void *start, unsigned long size)
827 if (!is_vmalloc_or_module_addr(start))
830 kasan_unpoison_shadow(start, size);
833 static int kasan_depopulate_vmalloc_pte(pte_t *ptep, unsigned long addr,
838 page = (unsigned long)__va(pte_pfn(*ptep) << PAGE_SHIFT);
840 spin_lock(&init_mm.page_table_lock);
842 if (likely(!pte_none(*ptep))) {
843 pte_clear(&init_mm, addr, ptep);
846 spin_unlock(&init_mm.page_table_lock);
852 * Release the backing for the vmalloc region [start, end), which
853 * lies within the free region [free_region_start, free_region_end).
855 * This can be run lazily, long after the region was freed. It runs
856 * under vmap_area_lock, so it's not safe to interact with the vmalloc/vmap
859 * How does this work?
860 * -------------------
862 * We have a region that is page aligned, labelled as A.
863 * That might not map onto the shadow in a way that is page-aligned:
867 * |????????|????????|AAAAAAAA|AA....AA|AAAAAAAA|????????| < vmalloc
868 * -------- -------- -------- -------- --------
871 * \-------\|/------/ |/---------------/
873 * |??AAAAAA|AAAAAAAA|AA??????| < shadow
876 * First we align the start upwards and the end downwards, so that the
877 * shadow of the region aligns with shadow page boundaries. In the
878 * example, this gives us the shadow page (2). This is the shadow entirely
879 * covered by this allocation.
881 * Then we have the tricky bits. We want to know if we can free the
882 * partially covered shadow pages - (1) and (3) in the example. For this,
883 * we are given the start and end of the free region that contains this
884 * allocation. Extending our previous example, we could have:
886 * free_region_start free_region_end
889 * |FFFFFFFF|FFFFFFFF|AAAAAAAA|AA....AA|AAAAAAAA|FFFFFFFF| < vmalloc
890 * -------- -------- -------- -------- --------
893 * \-------\|/------/ |/---------------/
895 * |FFAAAAAA|AAAAAAAA|AAF?????| < shadow
898 * Once again, we align the start of the free region up, and the end of
899 * the free region down so that the shadow is page aligned. So we can free
900 * page (1) - we know no allocation currently uses anything in that page,
901 * because all of it is in the vmalloc free region. But we cannot free
902 * page (3), because we can't be sure that the rest of it is unused.
904 * We only consider pages that contain part of the original region for
905 * freeing: we don't try to free other pages from the free region or we'd
906 * end up trying to free huge chunks of virtual address space.
911 * How do we know that we're not freeing a page that is simultaneously
912 * being used for a fresh allocation in kasan_populate_vmalloc(_pte)?
914 * We _can_ have kasan_release_vmalloc and kasan_populate_vmalloc running
915 * at the same time. While we run under free_vmap_area_lock, the population
918 * free_vmap_area_lock instead operates to ensure that the larger range
919 * [free_region_start, free_region_end) is safe: because __alloc_vmap_area and
920 * the per-cpu region-finding algorithm both run under free_vmap_area_lock,
921 * no space identified as free will become used while we are running. This
922 * means that so long as we are careful with alignment and only free shadow
923 * pages entirely covered by the free region, we will not run in to any
924 * trouble - any simultaneous allocations will be for disjoint regions.
926 void kasan_release_vmalloc(unsigned long start, unsigned long end,
927 unsigned long free_region_start,
928 unsigned long free_region_end)
930 void *shadow_start, *shadow_end;
931 unsigned long region_start, region_end;
934 region_start = ALIGN(start, PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
935 region_end = ALIGN_DOWN(end, PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
937 free_region_start = ALIGN(free_region_start,
938 PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
940 if (start != region_start &&
941 free_region_start < region_start)
942 region_start -= PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE;
944 free_region_end = ALIGN_DOWN(free_region_end,
945 PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE);
947 if (end != region_end &&
948 free_region_end > region_end)
949 region_end += PAGE_SIZE * KASAN_SHADOW_SCALE_SIZE;
951 shadow_start = kasan_mem_to_shadow((void *)region_start);
952 shadow_end = kasan_mem_to_shadow((void *)region_end);
954 if (shadow_end > shadow_start) {
955 size = shadow_end - shadow_start;
956 apply_to_existing_page_range(&init_mm,
957 (unsigned long)shadow_start,
958 size, kasan_depopulate_vmalloc_pte,
960 flush_tlb_kernel_range((unsigned long)shadow_start,
961 (unsigned long)shadow_end);