linux kernel 内存踩踏之KASAN_SW

一、背景

linux kernel 内存踩踏之KASAN（一）_kasan版本跟hasan版本区别-CSDN博客

上一篇简单介绍了标准版本的KASAN使用方法和实现，这里将介绍KASAN_SW_TAGS和KASAN_HW_TAGS

的使用和背后基本原理，下图是三种方式的对比：

Overhead type	MTE	KASAN_SW_TAG(kernel)/HWASan(userspace)	KASAN(kernel)/ASan(userspace)
RAM	3%-5%	10%-35%	~2x
CPU	0%-5%	~2x	~2x
Code size	2%-4%	40%-50%	50%-2x

上表数据来源google的 userspace下MTE、HWASAN和ASAN的测试数据，内核的部分没有找到准确的对比数据，应该也差不多，套用上表。

二、KASAN_SW_TAGS使能相关配置

关键差异：CONFIG_KASAN_SW_TAGS=y

/sys/kernel/debug # zcat /proc/config.gz | grep -i kasan
CONFIG_KASAN_SHADOW_OFFSET=0xefff800000000000 //这个offset和普通版本kasan有差异
CONFIG_DRIVER_KASAN_TEST=m
CONFIG_HAVE_ARCH_KASAN=y
CONFIG_HAVE_ARCH_KASAN_SW_TAGS=y
CONFIG_HAVE_ARCH_KASAN_HW_TAGS=y
CONFIG_HAVE_ARCH_KASAN_VMALLOC=y
CONFIG_CC_HAS_KASAN_GENERIC=y
CONFIG_CC_HAS_KASAN_SW_TAGS=y
CONFIG_KASAN=y
CONFIG_CC_HAS_KASAN_MEMINTRINSIC_PREFIX=y
# CONFIG_KASAN_GENERIC is not set
CONFIG_KASAN_SW_TAGS=y     //SW_TAGS 版本kasan
# CONFIG_KASAN_HW_TAGS is not set
CONFIG_KASAN_OUTLINE=y
# CONFIG_KASAN_INLINE is not set
CONFIG_KASAN_STACK=y      //stack kasan检测，如局部变量，局部数组等操作引起的内存踩踏
CONFIG_KASAN_VMALLOC=y    //vmalloc kasan检测，使用vmalloc申请内存的内存踩踏

三、KASAN_SW_TAGS基本原理

SW_TAG shadow的原理就是利用ARM64的TBI（Top Byte Ignore）特性，在最高byte存储指针存储能访问内存区域的shadow标记，利用指针操作地址时就会检查指针的shadow和操作地址的的shadow是否一致，不一致则触发内存异常并报告原因。

sw_tag 信息

#define KASAN_PAGE_FREE		KASAN_TAG_INVALID
#define KASAN_PAGE_REDZONE	KASAN_TAG_INVALID
#define KASAN_SLAB_REDZONE	KASAN_TAG_INVALID
#define KASAN_SLAB_FREE		KASAN_TAG_INVALID
#define KASAN_VMALLOC_INVALID	KASAN_TAG_INVALID /* only used for SW_TAGS */

#define KASAN_TAG_KERNEL	0xFF /* native kernel pointers tag */
#define KASAN_TAG_INVALID	0xFE /* inaccessible memory tag */
#define KASAN_TAG_MAX		0xFD /* maximum value for random tags */

#ifdef CONFIG_KASAN_HW_TAGS
#define KASAN_TAG_MIN		0xF0 /* minimum value for random tags */
#else
#define KASAN_TAG_MIN		0x00 /* minimum value for random tags */
#endif

SW_TAG的在指针内存分配时指定，内存有效时随机生成的有效值范围：0x00 ~ 0xFD， 0xFE用来表示free或者redzone等标记；

下图是arm64 48位 pagesize 4K的内存映射图，shadow的16TB映射整个内核空间：

CONFIG_KASAN_SHADOW_OFFSET=0xefff800000000000

计算方法：

CONFIG_KASAN_SHADOW_OFFSET= KASAN_SHADOW_START - KERNEL_ADDR_START >>4

= 0xffff700000000000 - ( 0xffff000000000000 >> 4) = 0xefff800000000000

有了这个kasan_shadow_offset，后面我们需要获取一个内核地址对应的shadow 位置，只需要通过公式：

kernel_addr >> 4 + CONFIG_KASAN_SHADOW_OFFSET = kernel_addr对应的shadow_addr

四、sw_tag生成和验证流程分析

4.1 设置sw_tag

还是用kmalloc为例：

kmalloc
-->kmalloc_trace
     -->__kmem_cache_alloc_node
         -->slab_alloc_node
              -->slab_post_alloc_hook
                   -->kasan_slab_alloc

void * __must_check __kasan_slab_alloc(struct kmem_cache *cache,
					void *object, gfp_t flags, bool init)
{
	....

	/*
	 * Generate and assign random tag for tag-based modes.
	 * Tag is ignored in set_tag() for the generic mode.
	 */
	tag = assign_tag(cache, object, false);    // 1、随机数分配tag
	tagged_object = set_tag(object, tag);      // 2、设置tag 到指针 

	/*
	 * Unpoison the whole object.
	 * For kmalloc() allocations, kasan_kmalloc() will do precise poisoning.
	 */
	kasan_unpoison(tagged_object, cache->object_size, init); 
        //3、从分配地址和size确认tag是否需要更新，如果和上面新分配的tag值不同，则更新tag

	/* Save alloc info (if possible) for non-kmalloc() allocations. */
	if (kasan_stack_collection_enabled() && !is_kmalloc_cache(cache))
		kasan_save_alloc_info(cache, tagged_object, flags);
        //4、存储分配stack

	return tagged_object;
}

#if defined(CONFIG_KASAN_SW_TAGS) || defined(CONFIG_KASAN_HW_TAGS)
#define __tag_shifted(tag)  ((u64)(tag) << 56)
#define __tag_reset(addr)   __untagged_addr(addr)
#define __tag_get(addr)     (__u8)((u64)(addr) >> 56)

流程如下：

1、分配tag随机数（0x00~0xFD）

2、给指针最高byte存储新 tag

3、根据指针tag和分配的长度，检查 ptr>>4 + shadow_offset处存储的tag值是否一致，不一致则更新

4、返回指针（高byte为tag）

4.2 检查指针

检查指针即使是在kasan_check_range中进行的，

(gdb) disassemble __hwasan_store1_noabort
Dump of assembler code for function __hwasan_store1_noabort:
   0xffff8000803d6f08 <+0>:	paciasp
   0xffff8000803d6f0c <+4>:	stp	x29, x30, [sp, #-16]!
   0xffff8000803d6f10 <+8>:	xpaclri
   0xffff8000803d6f14 <+12>:	mov	w2, #0x1                   	// #1
   0xffff8000803d6f18 <+16>:	mov	x29, sp
   0xffff8000803d6f1c <+20>:	mov	x3, x30
   0xffff8000803d6f20 <+24>:	mov	x1, #0x1                   	// #1
   0xffff8000803d6f24 <+28>:	bl	0xffff8000803d6e38 <kasan_check_range>
   0xffff8000803d6f28 <+32>:	ldp	x29, x30, [sp], #16
   0xffff8000803d6f2c <+36>:	autiasp
   0xffff8000803d6f30 <+40>:	ret

bool kasan_check_range(const void *addr, size_t size, bool write,
			unsigned long ret_ip)
{
	u8 tag;
	u8 *shadow_first, *shadow_last, *shadow;
	void *untagged_addr;

	if (unlikely(size == 0))
		return true;

	if (unlikely(addr + size < addr))
		return !kasan_report(addr, size, write, ret_ip);

	tag = get_tag((const void *)addr);  //1、获取指针tag

	/*
	 * Ignore accesses for pointers tagged with 0xff (native kernel
	 * pointer tag) to suppress false positives caused by kmap.
	 *
	 * Some kernel code was written to account for archs that don't keep
	 * high memory mapped all the time, but rather map and unmap particular
	 * pages when needed. Instead of storing a pointer to the kernel memory,
	 * this code saves the address of the page structure and offset within
	 * that page for later use. Those pages are then mapped and unmapped
	 * with kmap/kunmap when necessary and virt_to_page is used to get the
	 * virtual address of the page. For arm64 (that keeps the high memory
	 * mapped all the time), kmap is turned into a page_address call.

	 * The issue is that with use of the page_address + virt_to_page
	 * sequence the top byte value of the original pointer gets lost (gets
	 * set to KASAN_TAG_KERNEL (0xFF)).
	 */
	if (tag == KASAN_TAG_KERNEL)
		return true;

	untagged_addr = kasan_reset_tag((const void *)addr); //2、将带tag指针转换成指针
	if (unlikely(!addr_has_metadata(untagged_addr)))
		return !kasan_report(addr, size, write, ret_ip);
	shadow_first = kasan_mem_to_shadow(untagged_addr);  //3、提取对应地址的sw_tag shadow值
	shadow_last = kasan_mem_to_shadow(untagged_addr + size - 1); //4、提取访问地址尾部的sw_tag shadow值
	for (shadow = shadow_first; shadow <= shadow_last; shadow++) {
		if (*shadow != tag) {                              //5、遍历检查shadow tag和指针tag是否匹配
			return !kasan_report(addr, size, write, ret_ip);
		}
	}

	return true;
}

如上面代码逻辑，检查tag的流程如下：

1、传入指针和内存操作的长度

2、获取指针tag

3、将带tag指针转换成指针

4、提取对应地址的sw_tag shadow值

5、提取访问地址尾部的sw_tag shadow值

6、遍历检查shadow tag和指针tag是否匹配

五、利用 test driver程序验证

还是上一篇的例子(linux kernel 内存踩踏之KASAN（一）_kasan版本跟hasan版本区别-CSDN博客)：

例子日志：

/test # echo 0 > /dev/kasan_test 
[  150.681333] kmalloc_oob_right d2ff000003de9c00
[  150.691414] ==================================================================
[  150.693254] BUG: KASAN: invalid-access in kmalloc_oob_right.constprop.0+0x4c/0x6c [kasan_driver]
[  150.695503] Write of size 1 at addr d2ff000003de9c81 by task sh/181
[  150.696332] Pointer tag: [d2], memory tag: [fe]
[  150.696848] 
[  150.697599] CPU: 1 PID: 181 Comm: sh Tainted: G    B            N 6.6.1-g00ad0b878692 #18
[  150.698596] Hardware name: linux,dummy-virt (DT)
[  150.699352] Call trace:
[  150.699744]  dump_backtrace+0x90/0xe8
[  150.700697]  show_stack+0x18/0x24
[  150.701221]  dump_stack_lvl+0x48/0x60
[  150.701716]  print_report+0x15c/0x54c
[  150.702204]  kasan_report+0xc4/0x108
[  150.702678]  kasan_check_range+0x80/0xa4
[  150.703198]  __hwasan_store1_noabort+0x20/0x2c
[  150.703749]  kmalloc_oob_right.constprop.0+0x4c/0x6c [kasan_driver]
[  150.704593]  kasan_test_case+0x40/0xc0 [kasan_driver]
[  150.705354]  kasan_testcase_write+0x88/0x130 [kasan_driver]
[  150.706170]  vfs_write+0x144/0x4d8
[  150.706667]  ksys_write+0xe0/0x1b0
[  150.707166]  __arm64_sys_write+0x44/0x58
[  150.707729]  invoke_syscall+0x60/0x17c
[  150.708246]  el0_svc_common.constprop.0+0x78/0x13c
[  150.708842]  do_el0_svc+0x30/0x40
[  150.709462]  el0_svc+0x40/0x100
[  150.709973]  el0t_64_sync_handler+0x120/0x12c
[  150.710410]  el0t_64_sync+0x190/0x194
[  150.710946] 
[  150.711219] The buggy address belongs to the object at ffff000003de9c80
[  150.711219]  which belongs to the cache kmalloc-128 of size 128
[  150.712055] The buggy address is located 1 bytes inside of
[  150.712055]  128-byte region [ffff000003de9c80, ffff000003de9d00)
[  150.712749] 
[  150.713093] The buggy address belongs to the physical page:
[  150.713741] page:(____ptrval____) refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x43de9
[  150.714943] flags: 0x3fffc0000000800(slab|node=0|zone=0|lastcpupid=0xffff|kasantag=0x0)
[  150.715955] page_type: 0xffffffff()
[  150.716752] raw: 03fffc0000000800 82ff000003402600 dead000000000122 0000000000000000
[  150.717349] raw: 0000000000000000 0000000080200020 00000001ffffffff 0000000000000000
[  150.717938] page dumped because: kasan: bad access detected
[  150.718358] 
[  150.718602] Memory state around the buggy address:
[  150.719208]  ffff000003de9a00: 2c 2c 2c 2c 2c 2c 2c fe 28 28 28 28 28 28 28 28
[  150.719744]  ffff000003de9b00: 66 66 66 66 66 66 66 66 f8 f8 f8 f8 f8 f8 f8 f8
[  150.720267] >ffff000003de9c00: d2 d2 d2 d2 d2 d2 d2 d2 fe fe fe fe fe fe fe fe
[  150.720886]                                            ^
[  150.721635]  ffff000003de9d00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
[  150.722291]  ffff000003de9e00: fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe fe
[  150.722978] ==================================================================
[  150.724556] kasan_test_case type 0

调试：

(gdb) disassemble 
Dump of assembler code for function kmalloc_oob_right:
0xffff80007b160300 <+0>:	paciasp
   0xffff80007b160304 <+4>:	adrp	x0, 0xffff8000822ce000 <cpu_ops+432>
   0xffff80007b160308 <+8>:	stp	x29, x30, [sp, #-32]!
   0xffff80007b16030c <+12>:	mov	x2, #0x80                  	// #128
   0xffff80007b160310 <+16>:	mov	w1, #0xcc0                 	// #3264
   0xffff80007b160314 <+20>:	mov	x29, sp
   0xffff80007b160318 <+24>:	ldr	x0, [x0, #3648]
   0xffff80007b16031c <+28>:	str	x19, [sp, #16]
   0xffff80007b160320 <+32>:	bl	0xffff80008033c920 <kmalloc_trace>  //1.指针设置sw tag 
=> 0xffff80007b160324 <+36>:	mov	x2, x0         //断点
   0xffff80007b160328 <+40>:	adrp	x1, 0xffff80007b164000
   0xffff80007b16032c <+44>:	add	x1, x1, #0x110
   0xffff80007b160330 <+48>:	add	x1, x1, #0x48
   0xffff80007b160334 <+52>:	mov	x19, x0
   0xffff80007b160338 <+56>:	adrp	x0, 0xffff80007b164000
   0xffff80007b16033c <+60>:	add	x0, x0, #0x50
   0xffff80007b160340 <+64>:	bl	0xffff80008015d280 <_printk>
   0xffff80007b160344 <+68>:	add	x0, x19, #0x81
   0xffff80007b160348 <+72>:	bl	0xffff8000803d6f08 <__hwasan_store1_noabort> 

                                                             //2.检查指针访问的内存是否合法
   0xffff80007b16034c <+76>:	mov	w1, #0x79                  	// #121
   0xffff80007b160350 <+80>:	strb	w1, [x19, #129]
   0xffff80007b160354 <+84>:	mov	x0, x19
   0xffff80007b160358 <+88>:	bl	0xffff80008033da7c <kfree>
   0xffff80007b16035c <+92>:	ldr	x19, [sp, #16]
   0xffff80007b160360 <+96>:	ldp	x29, x30, [sp], #32
   0xffff80007b160364 <+100>:	autiasp
   0xffff80007b160368 <+104>:	ret

1、在上图断点处检查kmalloc_trace分配的指针值
(gdb) p /x $x0
$7 = 0xd2ff000003de9c00

2、利用计算公式，寻找对应指针地址存储的sw_tag shadow值：
ptr >> 4 + kasan_offset = kasan sw shadow

计算时记得将指针头替换成0xff
即：0xffff000003de9c00 >> 4 + 0xefff800000000000 = 0xFFFF7000003DE9C0
(gdb) x /30b 0xFFFF7000003DE9C0
0xffff7000003de9c0: 0xd2 0xd2 0xd2 0xd2 0xd2 0xd2 0xd2 0xd2
0xffff7000003de9c8: 0xfe 0xfe 0xfe 0xfe 0xfe 0xfe 0xfe 0xfe
0xffff7000003de9d0: 0xfe 0xfe 0xfe 0xfe 0xfe 0xfe 0xfe 0xfe
0xffff7000003de9d8: 0xfe 0xfe 0xfe 0xfe 0xfe 0xfe

上面的0xd2代表指针指向有效空间的范围，8 个0xd2, 由于sw_tag是每16个byte对应一个byte, 这里表示这个指针有效的范围是8*16 =128字节，正好和测试用例 kmalloc(128) 对应；

3、kasan report原因是我们访问的指针对应地址长度为0x81, 访问到了129字节处，这里对应的tag为0xfe,最后上报异常如下：

[ 150.695503] Write of size 1 at addr d2ff000003de9c81 by task sh/181
[ 150.696332] Pointer tag: [d2], memory tag: [fe]

六、总结

从KASAN 和 KASAN_SW_TAGS的对比来看

类型	shadow内存占用	cpu占用	优缺点
KASAN	1/8	复杂，每次内存访问，需要计算对比shadow值	定位准确，8byte内的踩踏也能检测；32位/64位均能使用
KASAN_SW_TAGS	1/16	每次内存访问，需要计算对比shadow值	16 byte内的踩踏无法区分, 仅64才能使用（因为依赖arm64 TBI feature）

缺点1：16byte内的踩踏无法检测

KASAN_SW_TAGS的tag标记范围是16byte, 打一个比方：

ptr = kmalloc(129);

ptr[129] = 0; // 此时不会报错，无法检测到越界，实际上 pt[129] ~ ptr[128 + 16 -1] 内存越界操作都无法检测出来，因为这16字节的tag都是一样的，tag本身没有16byte內分配大小的记录；

缺点2：tag虽然是随机值，但是连续内存存在随机tag值一致导致漏检测可能

比如，

ptr1= kmalloc(128);

ptr2= kmalloc(128);

假如ptr1的tag是0x12, ptr1的tag也是0x12, 同时它们的内存连续，那么ptr1[128] = 0的操作就不会报错；

漏检测概率：由于0xfe和0xff两个值不会作为tag随机数, 连续内存生成重复tag的概率为1/254 * 1/254。

参考：

Android Native | 内存问题的终极武器--MTE