kasan: split out shadow.c from common.c
authorAndrey Konovalov <andreyknvl@google.com>
Tue, 22 Dec 2020 20:00:32 +0000 (12:00 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 22 Dec 2020 20:55:06 +0000 (12:55 -0800)
This is a preparatory commit for the upcoming addition of a new hardware
tag-based (MTE-based) KASAN mode.

The new mode won't be using shadow memory.  Move all shadow-related code
to shadow.c, which is only enabled for software KASAN modes that use
shadow memory.

No functional changes for software modes.

Link: https://lkml.kernel.org/r/17d95cfa7d5cf9c4fcd9bf415f2a8dea911668df.1606161801.git.andreyknvl@google.com
Signed-off-by: Andrey Konovalov <andreyknvl@google.com>
Signed-off-by: Vincenzo Frascino <vincenzo.frascino@arm.com>
Reviewed-by: Marco Elver <elver@google.com>
Reviewed-by: Alexander Potapenko <glider@google.com>
Tested-by: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Branislav Rankov <Branislav.Rankov@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Evgenii Stepanov <eugenis@google.com>
Cc: Kevin Brodsky <kevin.brodsky@arm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/kasan/Makefile
mm/kasan/common.c
mm/kasan/shadow.c [new file with mode: 0644]

index 7cf685bb51bd1bd80d57099c7137baeeea934b89..7cc1031e1ef8018b0c68230e36d0bdc717912114 100644 (file)
@@ -10,6 +10,7 @@ CFLAGS_REMOVE_generic_report.o = $(CC_FLAGS_FTRACE)
 CFLAGS_REMOVE_init.o = $(CC_FLAGS_FTRACE)
 CFLAGS_REMOVE_quarantine.o = $(CC_FLAGS_FTRACE)
 CFLAGS_REMOVE_report.o = $(CC_FLAGS_FTRACE)
+CFLAGS_REMOVE_shadow.o = $(CC_FLAGS_FTRACE)
 CFLAGS_REMOVE_tags.o = $(CC_FLAGS_FTRACE)
 CFLAGS_REMOVE_tags_report.o = $(CC_FLAGS_FTRACE)
 
@@ -26,9 +27,10 @@ CFLAGS_generic_report.o := $(CC_FLAGS_KASAN_RUNTIME)
 CFLAGS_init.o := $(CC_FLAGS_KASAN_RUNTIME)
 CFLAGS_quarantine.o := $(CC_FLAGS_KASAN_RUNTIME)
 CFLAGS_report.o := $(CC_FLAGS_KASAN_RUNTIME)
+CFLAGS_shadow.o := $(CC_FLAGS_KASAN_RUNTIME)
 CFLAGS_tags.o := $(CC_FLAGS_KASAN_RUNTIME)
 CFLAGS_tags_report.o := $(CC_FLAGS_KASAN_RUNTIME)
 
 obj-$(CONFIG_KASAN) := common.o report.o
-obj-$(CONFIG_KASAN_GENERIC) += init.o generic.o generic_report.o quarantine.o
-obj-$(CONFIG_KASAN_SW_TAGS) += init.o tags.o tags_report.o
+obj-$(CONFIG_KASAN_GENERIC) += init.o generic.o generic_report.o shadow.o quarantine.o
+obj-$(CONFIG_KASAN_SW_TAGS) += init.o shadow.o tags.o tags_report.o
index 166e36e0033e564b9ca6380ecaf59630a4bb5f72..88f57346c7bd3d88c8c0ad3c173b2cb9cc868d3c 100644 (file)
@@ -1,6 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
 /*
- * This file contains common generic and tag-based KASAN code.
+ * This file contains common KASAN code.
  *
  * Copyright (c) 2014 Samsung Electronics Co., Ltd.
  * Author: Andrey Ryabinin <ryabinin.a.a@gmail.com>
@@ -13,7 +13,6 @@
 #include <linux/init.h>
 #include <linux/kasan.h>
 #include <linux/kernel.h>
-#include <linux/kmemleak.h>
 #include <linux/linkage.h>
 #include <linux/memblock.h>
 #include <linux/memory.h>
 #include <linux/stacktrace.h>
 #include <linux/string.h>
 #include <linux/types.h>
-#include <linux/vmalloc.h>
 #include <linux/bug.h>
 
-#include <asm/cacheflush.h>
-#include <asm/tlbflush.h>
-
 #include "kasan.h"
 #include "../slab.h"
 
@@ -61,93 +56,6 @@ void kasan_disable_current(void)
        current->kasan_depth--;
 }
 
-bool __kasan_check_read(const volatile void *p, unsigned int size)
-{
-       return check_memory_region((unsigned long)p, size, false, _RET_IP_);
-}
-EXPORT_SYMBOL(__kasan_check_read);
-
-bool __kasan_check_write(const volatile void *p, unsigned int size)
-{
-       return check_memory_region((unsigned long)p, size, true, _RET_IP_);
-}
-EXPORT_SYMBOL(__kasan_check_write);
-
-#undef memset
-void *memset(void *addr, int c, size_t len)
-{
-       if (!check_memory_region((unsigned long)addr, len, true, _RET_IP_))
-               return NULL;
-
-       return __memset(addr, c, len);
-}
-
-#ifdef __HAVE_ARCH_MEMMOVE
-#undef memmove
-void *memmove(void *dest, const void *src, size_t len)
-{
-       if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
-           !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
-               return NULL;
-
-       return __memmove(dest, src, len);
-}
-#endif
-
-#undef memcpy
-void *memcpy(void *dest, const void *src, size_t len)
-{
-       if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
-           !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
-               return NULL;
-
-       return __memcpy(dest, src, len);
-}
-
-/*
- * Poisons the shadow memory for 'size' bytes starting from 'addr'.
- * Memory addresses should be aligned to KASAN_GRANULE_SIZE.
- */
-void poison_range(const void *address, size_t size, u8 value)
-{
-       void *shadow_start, *shadow_end;
-
-       /*
-        * Perform shadow offset calculation based on untagged address, as
-        * some of the callers (e.g. kasan_poison_object_data) pass tagged
-        * addresses to this function.
-        */
-       address = reset_tag(address);
-
-       shadow_start = kasan_mem_to_shadow(address);
-       shadow_end = kasan_mem_to_shadow(address + size);
-
-       __memset(shadow_start, value, shadow_end - shadow_start);
-}
-
-void unpoison_range(const void *address, size_t size)
-{
-       u8 tag = get_tag(address);
-
-       /*
-        * Perform shadow offset calculation based on untagged address, as
-        * some of the callers (e.g. kasan_unpoison_object_data) pass tagged
-        * addresses to this function.
-        */
-       address = reset_tag(address);
-
-       poison_range(address, size, tag);
-
-       if (size & KASAN_GRANULE_MASK) {
-               u8 *shadow = (u8 *)kasan_mem_to_shadow(address + size);
-
-               if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
-                       *shadow = tag;
-               else
-                       *shadow = size & KASAN_GRANULE_MASK;
-       }
-}
-
 void kasan_unpoison_range(const void *address, size_t size)
 {
        unpoison_range(address, size);
@@ -540,395 +448,3 @@ void kasan_kfree_large(void *ptr, unsigned long ip)
                kasan_report_invalid_free(ptr, ip);
        /* The object will be poisoned by page_alloc. */
 }
-
-#ifdef CONFIG_MEMORY_HOTPLUG
-static bool shadow_mapped(unsigned long addr)
-{
-       pgd_t *pgd = pgd_offset_k(addr);
-       p4d_t *p4d;
-       pud_t *pud;
-       pmd_t *pmd;
-       pte_t *pte;
-
-       if (pgd_none(*pgd))
-               return false;
-       p4d = p4d_offset(pgd, addr);
-       if (p4d_none(*p4d))
-               return false;
-       pud = pud_offset(p4d, addr);
-       if (pud_none(*pud))
-               return false;
-
-       /*
-        * We can't use pud_large() or pud_huge(), the first one is
-        * arch-specific, the last one depends on HUGETLB_PAGE.  So let's abuse
-        * pud_bad(), if pud is bad then it's bad because it's huge.
-        */
-       if (pud_bad(*pud))
-               return true;
-       pmd = pmd_offset(pud, addr);
-       if (pmd_none(*pmd))
-               return false;
-
-       if (pmd_bad(*pmd))
-               return true;
-       pte = pte_offset_kernel(pmd, addr);
-       return !pte_none(*pte);
-}
-
-static int __meminit kasan_mem_notifier(struct notifier_block *nb,
-                       unsigned long action, void *data)
-{
-       struct memory_notify *mem_data = data;
-       unsigned long nr_shadow_pages, start_kaddr, shadow_start;
-       unsigned long shadow_end, shadow_size;
-
-       nr_shadow_pages = mem_data->nr_pages >> KASAN_SHADOW_SCALE_SHIFT;
-       start_kaddr = (unsigned long)pfn_to_kaddr(mem_data->start_pfn);
-       shadow_start = (unsigned long)kasan_mem_to_shadow((void *)start_kaddr);
-       shadow_size = nr_shadow_pages << PAGE_SHIFT;
-       shadow_end = shadow_start + shadow_size;
-
-       if (WARN_ON(mem_data->nr_pages % KASAN_GRANULE_SIZE) ||
-               WARN_ON(start_kaddr % (KASAN_GRANULE_SIZE << PAGE_SHIFT)))
-               return NOTIFY_BAD;
-
-       switch (action) {
-       case MEM_GOING_ONLINE: {
-               void *ret;
-
-               /*
-                * If shadow is mapped already than it must have been mapped
-                * during the boot. This could happen if we onlining previously
-                * offlined memory.
-                */
-               if (shadow_mapped(shadow_start))
-                       return NOTIFY_OK;
-
-               ret = __vmalloc_node_range(shadow_size, PAGE_SIZE, shadow_start,
-                                       shadow_end, GFP_KERNEL,
-                                       PAGE_KERNEL, VM_NO_GUARD,
-                                       pfn_to_nid(mem_data->start_pfn),
-                                       __builtin_return_address(0));
-               if (!ret)
-                       return NOTIFY_BAD;
-
-               kmemleak_ignore(ret);
-               return NOTIFY_OK;
-       }
-       case MEM_CANCEL_ONLINE:
-       case MEM_OFFLINE: {
-               struct vm_struct *vm;
-
-               /*
-                * shadow_start was either mapped during boot by kasan_init()
-                * or during memory online by __vmalloc_node_range().
-                * In the latter case we can use vfree() to free shadow.
-                * Non-NULL result of the find_vm_area() will tell us if
-                * that was the second case.
-                *
-                * Currently it's not possible to free shadow mapped
-                * during boot by kasan_init(). It's because the code
-                * to do that hasn't been written yet. So we'll just
-                * leak the memory.
-                */
-               vm = find_vm_area((void *)shadow_start);
-               if (vm)
-                       vfree((void *)shadow_start);
-       }
-       }
-
-       return NOTIFY_OK;
-}
-
-static int __init kasan_memhotplug_init(void)
-{
-       hotplug_memory_notifier(kasan_mem_notifier, 0);
-
-       return 0;
-}
-
-core_initcall(kasan_memhotplug_init);
-#endif
-
-#ifdef CONFIG_KASAN_VMALLOC
-
-static int kasan_populate_vmalloc_pte(pte_t *ptep, unsigned long addr,
-                                     void *unused)
-{
-       unsigned long page;
-       pte_t pte;
-
-       if (likely(!pte_none(*ptep)))
-               return 0;
-
-       page = __get_free_page(GFP_KERNEL);
-       if (!page)
-               return -ENOMEM;
-
-       memset((void *)page, KASAN_VMALLOC_INVALID, PAGE_SIZE);
-       pte = pfn_pte(PFN_DOWN(__pa(page)), PAGE_KERNEL);
-
-       spin_lock(&init_mm.page_table_lock);
-       if (likely(pte_none(*ptep))) {
-               set_pte_at(&init_mm, addr, ptep, pte);
-               page = 0;
-       }
-       spin_unlock(&init_mm.page_table_lock);
-       if (page)
-               free_page(page);
-       return 0;
-}
-
-int kasan_populate_vmalloc(unsigned long addr, unsigned long size)
-{
-       unsigned long shadow_start, shadow_end;
-       int ret;
-
-       if (!is_vmalloc_or_module_addr((void *)addr))
-               return 0;
-
-       shadow_start = (unsigned long)kasan_mem_to_shadow((void *)addr);
-       shadow_start = ALIGN_DOWN(shadow_start, PAGE_SIZE);
-       shadow_end = (unsigned long)kasan_mem_to_shadow((void *)addr + size);
-       shadow_end = ALIGN(shadow_end, PAGE_SIZE);
-
-       ret = apply_to_page_range(&init_mm, shadow_start,
-                                 shadow_end - shadow_start,
-                                 kasan_populate_vmalloc_pte, NULL);
-       if (ret)
-               return ret;
-
-       flush_cache_vmap(shadow_start, shadow_end);
-
-       /*
-        * We need to be careful about inter-cpu effects here. Consider:
-        *
-        *   CPU#0                                CPU#1
-        * WRITE_ONCE(p, vmalloc(100));         while (x = READ_ONCE(p)) ;
-        *                                      p[99] = 1;
-        *
-        * With compiler instrumentation, that ends up looking like this:
-        *
-        *   CPU#0                                CPU#1
-        * // vmalloc() allocates memory
-        * // let a = area->addr
-        * // we reach kasan_populate_vmalloc
-        * // and call unpoison_range:
-        * STORE shadow(a), unpoison_val
-        * ...
-        * STORE shadow(a+99), unpoison_val     x = LOAD p
-        * // rest of vmalloc process           <data dependency>
-        * STORE p, a                           LOAD shadow(x+99)
-        *
-        * If there is no barrier between the end of unpoisioning the shadow
-        * and the store of the result to p, the stores could be committed
-        * in a different order by CPU#0, and CPU#1 could erroneously observe
-        * poison in the shadow.
-        *
-        * We need some sort of barrier between the stores.
-        *
-        * In the vmalloc() case, this is provided by a smp_wmb() in
-        * clear_vm_uninitialized_flag(). In the per-cpu allocator and in
-        * get_vm_area() and friends, the caller gets shadow allocated but
-        * doesn't have any pages mapped into the virtual address space that
-        * has been reserved. Mapping those pages in will involve taking and
-        * releasing a page-table lock, which will provide the barrier.
-        */
-
-       return 0;
-}
-
-/*
- * Poison the shadow for a vmalloc region. Called as part of the
- * freeing process at the time the region is freed.
- */
-void kasan_poison_vmalloc(const void *start, unsigned long size)
-{
-       if (!is_vmalloc_or_module_addr(start))
-               return;
-
-       size = round_up(size, KASAN_GRANULE_SIZE);
-       poison_range(start, size, KASAN_VMALLOC_INVALID);
-}
-
-void kasan_unpoison_vmalloc(const void *start, unsigned long size)
-{
-       if (!is_vmalloc_or_module_addr(start))
-               return;
-
-       unpoison_range(start, size);
-}
-
-static int kasan_depopulate_vmalloc_pte(pte_t *ptep, unsigned long addr,
-                                       void *unused)
-{
-       unsigned long page;
-
-       page = (unsigned long)__va(pte_pfn(*ptep) << PAGE_SHIFT);
-
-       spin_lock(&init_mm.page_table_lock);
-
-       if (likely(!pte_none(*ptep))) {
-               pte_clear(&init_mm, addr, ptep);
-               free_page(page);
-       }
-       spin_unlock(&init_mm.page_table_lock);
-
-       return 0;
-}
-
-/*
- * Release the backing for the vmalloc region [start, end), which
- * lies within the free region [free_region_start, free_region_end).
- *
- * This can be run lazily, long after the region was freed. It runs
- * under vmap_area_lock, so it's not safe to interact with the vmalloc/vmap
- * infrastructure.
- *
- * How does this work?
- * -------------------
- *
- * We have a region that is page aligned, labelled as A.
- * That might not map onto the shadow in a way that is page-aligned:
- *
- *                    start                     end
- *                    v                         v
- * |????????|????????|AAAAAAAA|AA....AA|AAAAAAAA|????????| < vmalloc
- *  -------- -------- --------          -------- --------
- *      |        |       |                 |        |
- *      |        |       |         /-------/        |
- *      \-------\|/------/         |/---------------/
- *              |||                ||
- *             |??AAAAAA|AAAAAAAA|AA??????|                < shadow
- *                 (1)      (2)      (3)
- *
- * First we align the start upwards and the end downwards, so that the
- * shadow of the region aligns with shadow page boundaries. In the
- * example, this gives us the shadow page (2). This is the shadow entirely
- * covered by this allocation.
- *
- * Then we have the tricky bits. We want to know if we can free the
- * partially covered shadow pages - (1) and (3) in the example. For this,
- * we are given the start and end of the free region that contains this
- * allocation. Extending our previous example, we could have:
- *
- *  free_region_start                                    free_region_end
- *  |                 start                     end      |
- *  v                 v                         v        v
- * |FFFFFFFF|FFFFFFFF|AAAAAAAA|AA....AA|AAAAAAAA|FFFFFFFF| < vmalloc
- *  -------- -------- --------          -------- --------
- *      |        |       |                 |        |
- *      |        |       |         /-------/        |
- *      \-------\|/------/         |/---------------/
- *              |||                ||
- *             |FFAAAAAA|AAAAAAAA|AAF?????|                < shadow
- *                 (1)      (2)      (3)
- *
- * Once again, we align the start of the free region up, and the end of
- * the free region down so that the shadow is page aligned. So we can free
- * page (1) - we know no allocation currently uses anything in that page,
- * because all of it is in the vmalloc free region. But we cannot free
- * page (3), because we can't be sure that the rest of it is unused.
- *
- * We only consider pages that contain part of the original region for
- * freeing: we don't try to free other pages from the free region or we'd
- * end up trying to free huge chunks of virtual address space.
- *
- * Concurrency
- * -----------
- *
- * How do we know that we're not freeing a page that is simultaneously
- * being used for a fresh allocation in kasan_populate_vmalloc(_pte)?
- *
- * We _can_ have kasan_release_vmalloc and kasan_populate_vmalloc running
- * at the same time. While we run under free_vmap_area_lock, the population
- * code does not.
- *
- * free_vmap_area_lock instead operates to ensure that the larger range
- * [free_region_start, free_region_end) is safe: because __alloc_vmap_area and
- * the per-cpu region-finding algorithm both run under free_vmap_area_lock,
- * no space identified as free will become used while we are running. This
- * means that so long as we are careful with alignment and only free shadow
- * pages entirely covered by the free region, we will not run in to any
- * trouble - any simultaneous allocations will be for disjoint regions.
- */
-void kasan_release_vmalloc(unsigned long start, unsigned long end,
-                          unsigned long free_region_start,
-                          unsigned long free_region_end)
-{
-       void *shadow_start, *shadow_end;
-       unsigned long region_start, region_end;
-       unsigned long size;
-
-       region_start = ALIGN(start, PAGE_SIZE * KASAN_GRANULE_SIZE);
-       region_end = ALIGN_DOWN(end, PAGE_SIZE * KASAN_GRANULE_SIZE);
-
-       free_region_start = ALIGN(free_region_start,
-                                 PAGE_SIZE * KASAN_GRANULE_SIZE);
-
-       if (start != region_start &&
-           free_region_start < region_start)
-               region_start -= PAGE_SIZE * KASAN_GRANULE_SIZE;
-
-       free_region_end = ALIGN_DOWN(free_region_end,
-                                    PAGE_SIZE * KASAN_GRANULE_SIZE);
-
-       if (end != region_end &&
-           free_region_end > region_end)
-               region_end += PAGE_SIZE * KASAN_GRANULE_SIZE;
-
-       shadow_start = kasan_mem_to_shadow((void *)region_start);
-       shadow_end = kasan_mem_to_shadow((void *)region_end);
-
-       if (shadow_end > shadow_start) {
-               size = shadow_end - shadow_start;
-               apply_to_existing_page_range(&init_mm,
-                                            (unsigned long)shadow_start,
-                                            size, kasan_depopulate_vmalloc_pte,
-                                            NULL);
-               flush_tlb_kernel_range((unsigned long)shadow_start,
-                                      (unsigned long)shadow_end);
-       }
-}
-
-#else /* CONFIG_KASAN_VMALLOC */
-
-int kasan_module_alloc(void *addr, size_t size)
-{
-       void *ret;
-       size_t scaled_size;
-       size_t shadow_size;
-       unsigned long shadow_start;
-
-       shadow_start = (unsigned long)kasan_mem_to_shadow(addr);
-       scaled_size = (size + KASAN_GRANULE_SIZE - 1) >>
-                               KASAN_SHADOW_SCALE_SHIFT;
-       shadow_size = round_up(scaled_size, PAGE_SIZE);
-
-       if (WARN_ON(!PAGE_ALIGNED(shadow_start)))
-               return -EINVAL;
-
-       ret = __vmalloc_node_range(shadow_size, 1, shadow_start,
-                       shadow_start + shadow_size,
-                       GFP_KERNEL,
-                       PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE,
-                       __builtin_return_address(0));
-
-       if (ret) {
-               __memset(ret, KASAN_SHADOW_INIT, shadow_size);
-               find_vm_area(addr)->flags |= VM_KASAN;
-               kmemleak_ignore(ret);
-               return 0;
-       }
-
-       return -ENOMEM;
-}
-
-void kasan_free_shadow(const struct vm_struct *vm)
-{
-       if (vm->flags & VM_KASAN)
-               vfree(kasan_mem_to_shadow(vm->addr));
-}
-
-#endif
diff --git a/mm/kasan/shadow.c b/mm/kasan/shadow.c
new file mode 100644 (file)
index 0000000..5faba87
--- /dev/null
@@ -0,0 +1,505 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * This file contains KASAN runtime code that manages shadow memory for
+ * generic and software tag-based KASAN modes.
+ *
+ * Copyright (c) 2014 Samsung Electronics Co., Ltd.
+ * Author: Andrey Ryabinin <ryabinin.a.a@gmail.com>
+ *
+ * Some code borrowed from https://github.com/xairy/kasan-prototype by
+ *        Andrey Konovalov <andreyknvl@gmail.com>
+ */
+
+#include <linux/init.h>
+#include <linux/kasan.h>
+#include <linux/kernel.h>
+#include <linux/kmemleak.h>
+#include <linux/memory.h>
+#include <linux/mm.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/vmalloc.h>
+
+#include <asm/cacheflush.h>
+#include <asm/tlbflush.h>
+
+#include "kasan.h"
+
+bool __kasan_check_read(const volatile void *p, unsigned int size)
+{
+       return check_memory_region((unsigned long)p, size, false, _RET_IP_);
+}
+EXPORT_SYMBOL(__kasan_check_read);
+
+bool __kasan_check_write(const volatile void *p, unsigned int size)
+{
+       return check_memory_region((unsigned long)p, size, true, _RET_IP_);
+}
+EXPORT_SYMBOL(__kasan_check_write);
+
+#undef memset
+void *memset(void *addr, int c, size_t len)
+{
+       if (!check_memory_region((unsigned long)addr, len, true, _RET_IP_))
+               return NULL;
+
+       return __memset(addr, c, len);
+}
+
+#ifdef __HAVE_ARCH_MEMMOVE
+#undef memmove
+void *memmove(void *dest, const void *src, size_t len)
+{
+       if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
+           !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
+               return NULL;
+
+       return __memmove(dest, src, len);
+}
+#endif
+
+#undef memcpy
+void *memcpy(void *dest, const void *src, size_t len)
+{
+       if (!check_memory_region((unsigned long)src, len, false, _RET_IP_) ||
+           !check_memory_region((unsigned long)dest, len, true, _RET_IP_))
+               return NULL;
+
+       return __memcpy(dest, src, len);
+}
+
+/*
+ * Poisons the shadow memory for 'size' bytes starting from 'addr'.
+ * Memory addresses should be aligned to KASAN_GRANULE_SIZE.
+ */
+void poison_range(const void *address, size_t size, u8 value)
+{
+       void *shadow_start, *shadow_end;
+
+       /*
+        * Perform shadow offset calculation based on untagged address, as
+        * some of the callers (e.g. kasan_poison_object_data) pass tagged
+        * addresses to this function.
+        */
+       address = reset_tag(address);
+
+       shadow_start = kasan_mem_to_shadow(address);
+       shadow_end = kasan_mem_to_shadow(address + size);
+
+       __memset(shadow_start, value, shadow_end - shadow_start);
+}
+
+void unpoison_range(const void *address, size_t size)
+{
+       u8 tag = get_tag(address);
+
+       /*
+        * Perform shadow offset calculation based on untagged address, as
+        * some of the callers (e.g. kasan_unpoison_object_data) pass tagged
+        * addresses to this function.
+        */
+       address = reset_tag(address);
+
+       poison_range(address, size, tag);
+
+       if (size & KASAN_GRANULE_MASK) {
+               u8 *shadow = (u8 *)kasan_mem_to_shadow(address + size);
+
+               if (IS_ENABLED(CONFIG_KASAN_SW_TAGS))
+                       *shadow = tag;
+               else
+                       *shadow = size & KASAN_GRANULE_MASK;
+       }
+}
+
+#ifdef CONFIG_MEMORY_HOTPLUG
+static bool shadow_mapped(unsigned long addr)
+{
+       pgd_t *pgd = pgd_offset_k(addr);
+       p4d_t *p4d;
+       pud_t *pud;
+       pmd_t *pmd;
+       pte_t *pte;
+
+       if (pgd_none(*pgd))
+               return false;
+       p4d = p4d_offset(pgd, addr);
+       if (p4d_none(*p4d))
+               return false;
+       pud = pud_offset(p4d, addr);
+       if (pud_none(*pud))
+               return false;
+
+       /*
+        * We can't use pud_large() or pud_huge(), the first one is
+        * arch-specific, the last one depends on HUGETLB_PAGE.  So let's abuse
+        * pud_bad(), if pud is bad then it's bad because it's huge.
+        */
+       if (pud_bad(*pud))
+               return true;
+       pmd = pmd_offset(pud, addr);
+       if (pmd_none(*pmd))
+               return false;
+
+       if (pmd_bad(*pmd))
+               return true;
+       pte = pte_offset_kernel(pmd, addr);
+       return !pte_none(*pte);
+}
+
+static int __meminit kasan_mem_notifier(struct notifier_block *nb,
+                       unsigned long action, void *data)
+{
+       struct memory_notify *mem_data = data;
+       unsigned long nr_shadow_pages, start_kaddr, shadow_start;
+       unsigned long shadow_end, shadow_size;
+
+       nr_shadow_pages = mem_data->nr_pages >> KASAN_SHADOW_SCALE_SHIFT;
+       start_kaddr = (unsigned long)pfn_to_kaddr(mem_data->start_pfn);
+       shadow_start = (unsigned long)kasan_mem_to_shadow((void *)start_kaddr);
+       shadow_size = nr_shadow_pages << PAGE_SHIFT;
+       shadow_end = shadow_start + shadow_size;
+
+       if (WARN_ON(mem_data->nr_pages % KASAN_GRANULE_SIZE) ||
+               WARN_ON(start_kaddr % (KASAN_GRANULE_SIZE << PAGE_SHIFT)))
+               return NOTIFY_BAD;
+
+       switch (action) {
+       case MEM_GOING_ONLINE: {
+               void *ret;
+
+               /*
+                * If shadow is mapped already than it must have been mapped
+                * during the boot. This could happen if we onlining previously
+                * offlined memory.
+                */
+               if (shadow_mapped(shadow_start))
+                       return NOTIFY_OK;
+
+               ret = __vmalloc_node_range(shadow_size, PAGE_SIZE, shadow_start,
+                                       shadow_end, GFP_KERNEL,
+                                       PAGE_KERNEL, VM_NO_GUARD,
+                                       pfn_to_nid(mem_data->start_pfn),
+                                       __builtin_return_address(0));
+               if (!ret)
+                       return NOTIFY_BAD;
+
+               kmemleak_ignore(ret);
+               return NOTIFY_OK;
+       }
+       case MEM_CANCEL_ONLINE:
+       case MEM_OFFLINE: {
+               struct vm_struct *vm;
+
+               /*
+                * shadow_start was either mapped during boot by kasan_init()
+                * or during memory online by __vmalloc_node_range().
+                * In the latter case we can use vfree() to free shadow.
+                * Non-NULL result of the find_vm_area() will tell us if
+                * that was the second case.
+                *
+                * Currently it's not possible to free shadow mapped
+                * during boot by kasan_init(). It's because the code
+                * to do that hasn't been written yet. So we'll just
+                * leak the memory.
+                */
+               vm = find_vm_area((void *)shadow_start);
+               if (vm)
+                       vfree((void *)shadow_start);
+       }
+       }
+
+       return NOTIFY_OK;
+}
+
+static int __init kasan_memhotplug_init(void)
+{
+       hotplug_memory_notifier(kasan_mem_notifier, 0);
+
+       return 0;
+}
+
+core_initcall(kasan_memhotplug_init);
+#endif
+
+#ifdef CONFIG_KASAN_VMALLOC
+
+static int kasan_populate_vmalloc_pte(pte_t *ptep, unsigned long addr,
+                                     void *unused)
+{
+       unsigned long page;
+       pte_t pte;
+
+       if (likely(!pte_none(*ptep)))
+               return 0;
+
+       page = __get_free_page(GFP_KERNEL);
+       if (!page)
+               return -ENOMEM;
+
+       memset((void *)page, KASAN_VMALLOC_INVALID, PAGE_SIZE);
+       pte = pfn_pte(PFN_DOWN(__pa(page)), PAGE_KERNEL);
+
+       spin_lock(&init_mm.page_table_lock);
+       if (likely(pte_none(*ptep))) {
+               set_pte_at(&init_mm, addr, ptep, pte);
+               page = 0;
+       }
+       spin_unlock(&init_mm.page_table_lock);
+       if (page)
+               free_page(page);
+       return 0;
+}
+
+int kasan_populate_vmalloc(unsigned long addr, unsigned long size)
+{
+       unsigned long shadow_start, shadow_end;
+       int ret;
+
+       if (!is_vmalloc_or_module_addr((void *)addr))
+               return 0;
+
+       shadow_start = (unsigned long)kasan_mem_to_shadow((void *)addr);
+       shadow_start = ALIGN_DOWN(shadow_start, PAGE_SIZE);
+       shadow_end = (unsigned long)kasan_mem_to_shadow((void *)addr + size);
+       shadow_end = ALIGN(shadow_end, PAGE_SIZE);
+
+       ret = apply_to_page_range(&init_mm, shadow_start,
+                                 shadow_end - shadow_start,
+                                 kasan_populate_vmalloc_pte, NULL);
+       if (ret)
+               return ret;
+
+       flush_cache_vmap(shadow_start, shadow_end);
+
+       /*
+        * We need to be careful about inter-cpu effects here. Consider:
+        *
+        *   CPU#0                                CPU#1
+        * WRITE_ONCE(p, vmalloc(100));         while (x = READ_ONCE(p)) ;
+        *                                      p[99] = 1;
+        *
+        * With compiler instrumentation, that ends up looking like this:
+        *
+        *   CPU#0                                CPU#1
+        * // vmalloc() allocates memory
+        * // let a = area->addr
+        * // we reach kasan_populate_vmalloc
+        * // and call unpoison_range:
+        * STORE shadow(a), unpoison_val
+        * ...
+        * STORE shadow(a+99), unpoison_val     x = LOAD p
+        * // rest of vmalloc process           <data dependency>
+        * STORE p, a                           LOAD shadow(x+99)
+        *
+        * If there is no barrier between the end of unpoisioning the shadow
+        * and the store of the result to p, the stores could be committed
+        * in a different order by CPU#0, and CPU#1 could erroneously observe
+        * poison in the shadow.
+        *
+        * We need some sort of barrier between the stores.
+        *
+        * In the vmalloc() case, this is provided by a smp_wmb() in
+        * clear_vm_uninitialized_flag(). In the per-cpu allocator and in
+        * get_vm_area() and friends, the caller gets shadow allocated but
+        * doesn't have any pages mapped into the virtual address space that
+        * has been reserved. Mapping those pages in will involve taking and
+        * releasing a page-table lock, which will provide the barrier.
+        */
+
+       return 0;
+}
+
+/*
+ * Poison the shadow for a vmalloc region. Called as part of the
+ * freeing process at the time the region is freed.
+ */
+void kasan_poison_vmalloc(const void *start, unsigned long size)
+{
+       if (!is_vmalloc_or_module_addr(start))
+               return;
+
+       size = round_up(size, KASAN_GRANULE_SIZE);
+       poison_range(start, size, KASAN_VMALLOC_INVALID);
+}
+
+void kasan_unpoison_vmalloc(const void *start, unsigned long size)
+{
+       if (!is_vmalloc_or_module_addr(start))
+               return;
+
+       unpoison_range(start, size);
+}
+
+static int kasan_depopulate_vmalloc_pte(pte_t *ptep, unsigned long addr,
+                                       void *unused)
+{
+       unsigned long page;
+
+       page = (unsigned long)__va(pte_pfn(*ptep) << PAGE_SHIFT);
+
+       spin_lock(&init_mm.page_table_lock);
+
+       if (likely(!pte_none(*ptep))) {
+               pte_clear(&init_mm, addr, ptep);
+               free_page(page);
+       }
+       spin_unlock(&init_mm.page_table_lock);
+
+       return 0;
+}
+
+/*
+ * Release the backing for the vmalloc region [start, end), which
+ * lies within the free region [free_region_start, free_region_end).
+ *
+ * This can be run lazily, long after the region was freed. It runs
+ * under vmap_area_lock, so it's not safe to interact with the vmalloc/vmap
+ * infrastructure.
+ *
+ * How does this work?
+ * -------------------
+ *
+ * We have a region that is page aligned, labelled as A.
+ * That might not map onto the shadow in a way that is page-aligned:
+ *
+ *                    start                     end
+ *                    v                         v
+ * |????????|????????|AAAAAAAA|AA....AA|AAAAAAAA|????????| < vmalloc
+ *  -------- -------- --------          -------- --------
+ *      |        |       |                 |        |
+ *      |        |       |         /-------/        |
+ *      \-------\|/------/         |/---------------/
+ *              |||                ||
+ *             |??AAAAAA|AAAAAAAA|AA??????|                < shadow
+ *                 (1)      (2)      (3)
+ *
+ * First we align the start upwards and the end downwards, so that the
+ * shadow of the region aligns with shadow page boundaries. In the
+ * example, this gives us the shadow page (2). This is the shadow entirely
+ * covered by this allocation.
+ *
+ * Then we have the tricky bits. We want to know if we can free the
+ * partially covered shadow pages - (1) and (3) in the example. For this,
+ * we are given the start and end of the free region that contains this
+ * allocation. Extending our previous example, we could have:
+ *
+ *  free_region_start                                    free_region_end
+ *  |                 start                     end      |
+ *  v                 v                         v        v
+ * |FFFFFFFF|FFFFFFFF|AAAAAAAA|AA....AA|AAAAAAAA|FFFFFFFF| < vmalloc
+ *  -------- -------- --------          -------- --------
+ *      |        |       |                 |        |
+ *      |        |       |         /-------/        |
+ *      \-------\|/------/         |/---------------/
+ *              |||                ||
+ *             |FFAAAAAA|AAAAAAAA|AAF?????|                < shadow
+ *                 (1)      (2)      (3)
+ *
+ * Once again, we align the start of the free region up, and the end of
+ * the free region down so that the shadow is page aligned. So we can free
+ * page (1) - we know no allocation currently uses anything in that page,
+ * because all of it is in the vmalloc free region. But we cannot free
+ * page (3), because we can't be sure that the rest of it is unused.
+ *
+ * We only consider pages that contain part of the original region for
+ * freeing: we don't try to free other pages from the free region or we'd
+ * end up trying to free huge chunks of virtual address space.
+ *
+ * Concurrency
+ * -----------
+ *
+ * How do we know that we're not freeing a page that is simultaneously
+ * being used for a fresh allocation in kasan_populate_vmalloc(_pte)?
+ *
+ * We _can_ have kasan_release_vmalloc and kasan_populate_vmalloc running
+ * at the same time. While we run under free_vmap_area_lock, the population
+ * code does not.
+ *
+ * free_vmap_area_lock instead operates to ensure that the larger range
+ * [free_region_start, free_region_end) is safe: because __alloc_vmap_area and
+ * the per-cpu region-finding algorithm both run under free_vmap_area_lock,
+ * no space identified as free will become used while we are running. This
+ * means that so long as we are careful with alignment and only free shadow
+ * pages entirely covered by the free region, we will not run in to any
+ * trouble - any simultaneous allocations will be for disjoint regions.
+ */
+void kasan_release_vmalloc(unsigned long start, unsigned long end,
+                          unsigned long free_region_start,
+                          unsigned long free_region_end)
+{
+       void *shadow_start, *shadow_end;
+       unsigned long region_start, region_end;
+       unsigned long size;
+
+       region_start = ALIGN(start, PAGE_SIZE * KASAN_GRANULE_SIZE);
+       region_end = ALIGN_DOWN(end, PAGE_SIZE * KASAN_GRANULE_SIZE);
+
+       free_region_start = ALIGN(free_region_start,
+                                 PAGE_SIZE * KASAN_GRANULE_SIZE);
+
+       if (start != region_start &&
+           free_region_start < region_start)
+               region_start -= PAGE_SIZE * KASAN_GRANULE_SIZE;
+
+       free_region_end = ALIGN_DOWN(free_region_end,
+                                    PAGE_SIZE * KASAN_GRANULE_SIZE);
+
+       if (end != region_end &&
+           free_region_end > region_end)
+               region_end += PAGE_SIZE * KASAN_GRANULE_SIZE;
+
+       shadow_start = kasan_mem_to_shadow((void *)region_start);
+       shadow_end = kasan_mem_to_shadow((void *)region_end);
+
+       if (shadow_end > shadow_start) {
+               size = shadow_end - shadow_start;
+               apply_to_existing_page_range(&init_mm,
+                                            (unsigned long)shadow_start,
+                                            size, kasan_depopulate_vmalloc_pte,
+                                            NULL);
+               flush_tlb_kernel_range((unsigned long)shadow_start,
+                                      (unsigned long)shadow_end);
+       }
+}
+
+#else /* CONFIG_KASAN_VMALLOC */
+
+int kasan_module_alloc(void *addr, size_t size)
+{
+       void *ret;
+       size_t scaled_size;
+       size_t shadow_size;
+       unsigned long shadow_start;
+
+       shadow_start = (unsigned long)kasan_mem_to_shadow(addr);
+       scaled_size = (size + KASAN_GRANULE_SIZE - 1) >>
+                               KASAN_SHADOW_SCALE_SHIFT;
+       shadow_size = round_up(scaled_size, PAGE_SIZE);
+
+       if (WARN_ON(!PAGE_ALIGNED(shadow_start)))
+               return -EINVAL;
+
+       ret = __vmalloc_node_range(shadow_size, 1, shadow_start,
+                       shadow_start + shadow_size,
+                       GFP_KERNEL,
+                       PAGE_KERNEL, VM_NO_GUARD, NUMA_NO_NODE,
+                       __builtin_return_address(0));
+
+       if (ret) {
+               __memset(ret, KASAN_SHADOW_INIT, shadow_size);
+               find_vm_area(addr)->flags |= VM_KASAN;
+               kmemleak_ignore(ret);
+               return 0;
+       }
+
+       return -ENOMEM;
+}
+
+void kasan_free_shadow(const struct vm_struct *vm)
+{
+       if (vm->flags & VM_KASAN)
+               vfree(kasan_mem_to_shadow(vm->addr));
+}
+
+#endif