Bochspwn漏洞挖掘技术深究(1):Double Fetches 检测

虽然现在技术文章很少人看,大家都喜欢聊安全八卦,但技术文章输出是一种很好的学习方式。更重要的是,专业的文章是给专业的人看的,并非为了取悦所有人。

对于应用程序的代码插桩,有现成的Pin和DynamoRIO插桩框架,在Fuzzing中可以用来实现代码覆盖率的反馈驱动,这已经被应用到winafl,效果很好。除了挖洞,在逆向工程领域应用也很广泛。

上面都是针对应用层的,内核层的,上面的Pin和DynamoRIO就派不上用场了,对于这种系统内核级的指令插桩,有时就会采用虚拟化技术为实现,比如通过Qemu或Bochs虚拟机。

ProjectZero的j00ru大神就用bochs的插桩API为实现针对内核double fetches的监测,项目称为bochspwn,后来又采用污点追踪方式检测未初始化漏洞导致的内核信息泄露,叫bochspwn-reloaded。

Bochs Instrument API 文档参考:http://bochs.sourceforge.net/cgi-bin/lxr/source/instrument/instrumentation.txt,在编译bochs时指定插桩代码目录:

1
./configure [...] --enable-instrumentation="instrument/myinstrument"

下面是bochspwn中用到的API:

1
2
3
4
5
6
7
8
// Bochs初始化CPU对象时的回调函数
void bx_instr_initialize(unsigned cpu);
// Bochs析构CPU对象时的回调函数
void bx_instr_exit(unsigned cpu);
// Bochs访问线性内存时的回调函数
void bx_instr_lin_access(unsigned cpu, bx_address lin, bx_address phy,unsigned len, unsigned memtype, unsigned rw);
// Bochs执行指令前的回调函数
void bx_instr_before_execution(unsigned cpu, bxInstruction_c *i);

bx_instr_initialize用来加载配置信息,针对不同的系统环境设置不同的数据结构偏移地址,用来提供需要的进程/线程等重要信息:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
[general]
trace_log_path = memlog.bin
modules_list_path = modules.bin

os = windows
bitness = 32
version = win10_32

min_read_size = 1
max_read_size = 16
min_write_size = 1
max_write_size = 16

callstack_length = 48
write_as_text = 0

symbolize = 0
symbol_path = <symbols path>

[win7_32]
kprcb = 0x120
current_thread = 0x04
tcb = 0x0
process = 0x150
client_id = 0x22c
process_id = 0
thread_id = 4
create_time = 0x200
image_filename = 0x16c
kdversionblock = 0x34
psloadedmodulelist = 0x18
loadorder_flink = 0x0
basedllname = 0x2c
baseaddress = 0x18
sizeofimage = 0x20
us_len = 0x0
us_buffer = 0x4
teb_cid = 0x20
irql = 0x24
previous_mode = 0x13a
exception_list = 0x0
next_exception = 0x0
try_level = 0xc
......

Bochspwn的核心功能实现就在于bx_instr_lin_accessbx_instr_before_execution两个函数。先看下bx_instr_before_execution的实现逻辑:

  1. 忽略实模式real mode
  2. 忽略无关的系统调用中断指令,仅允许int 0x2eint 0x80
  3. 获取当前进程/线程ID相关的信息,当发现漏洞时方便重现
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
void bx_instr_before_execution(unsigned cpu, bxInstruction_c *i) {
static client_id thread;
BX_CPU_C *pcpu = BX_CPU(cpu);
unsigned opcode;

// We're not interested in instructions executed in real mode.
if (!pcpu->protected_mode() && !pcpu->long64_mode()) {
return;
}

// If the system needs an additional invokement from here, call it now.
if (globals::has_instr_before_execution_handler) {
invoke_system_handler(BX_OS_EVENT_INSTR_BEFORE_EXECUTION, pcpu, i);
}

// Any system-call invoking instruction is interesting - this
// is mostly due to 64-bit Linux which allows various ways
// to be used for system-call invocation.
// Note: We're not checking for int1, int3 nor into instructions.
opcode = i->getIaOpcode();
if (opcode != BX_IA_SYSCALL && opcode != BX_IA_SYSENTER && opcode != BX_IA_INT_Ib) {
return;
}

// The only two allowed interrupts are int 0x2e and int 0x80, which are legacy
// ways to invoke system calls on Windows and linux, respectively.
if (opcode == BX_IA_INT_Ib && i->Ib() != 0x2e && i->Ib() != 0x80) {
return;
}

// Obtain information about the current process/thread IDs.
if (!invoke_system_handler(BX_OS_EVENT_FILL_CID, pcpu, &thread)) {
return;
}

// Process information about a new syscall depending on the current mode.
if (!events::event_new_syscall(pcpu, &thread)) {
return;
}
}

再看下bx_instr_lin_access实现逻辑:

  1. 忽略仅读写指令
  2. 检测CPU类型(32位或64位)
  3. 判断当前指令地址pc是否为内核地址,判断访问的线性内存地址是否为用户层地址
  4. 检测读取的内存长度是否处于0~16字节之间,长度大小范围在config.txt中配置,仅处理此范围内的指令操作
  5. 通过上述条件之后,就代表可能存在内核漏洞,然后反汇编指令,然后填充日志记录信息
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
void bx_instr_lin_access(unsigned cpu, bx_address lin, bx_address phy,
unsigned len, unsigned memtype, unsigned rw) {

BX_CPU_C *pcpu = BX_CPU(cpu);
// Not going to use physical memory address.
(void)phy;

// Read-write instructions are currently not interesting.
if (rw == BX_RW)
return;

// Is the CPU in protected or long mode?
unsigned mode = 0;

// Note: DO NOT change order of these ifs. long64_mode must be called
// before protected_mode, since it will also return "true" on protected_mode
// query (well, long mode is technically protected mode).

if (pcpu->long64_mode()) {
#if BX_SUPPORT_X86_64
mode = 64;
#else
return;
#endif // BX_SUPPORT_X86_64
} else if (pcpu->protected_mode()) {
// This is either protected 32-bit mode or 32-bit compat. long mode.
mode = 32;
} else {
// Nothing interesting.
// TODO(gynvael): Well actually there is the smm_mode(), which
// might be a little interesting, even if it's just the bochs BIOS
// SMM code.
return;
}

// Is pc in kernel memory area?
// Is lin in user memory area?
bx_address pc = pcpu->prev_rip;
if (!invoke_system_handler(BX_OS_EVENT_CHECK_KERNEL_ADDR, &pc, NULL) ||
!invoke_system_handler(BX_OS_EVENT_CHECK_USER_ADDR, &lin, NULL)) {
return; /* pc not in ring-0 or lin not in ring-3 */
}

// Check if the access meets specified operand length criteria.
if (rw == BX_READ) {
if (len < globals::config.min_read_size || len > globals::config.max_read_size) {
return;
}
} else {
if (len < globals::config.min_write_size || len > globals::config.max_write_size) {
return;
}
}

// Save basic information about the access.
log_data_st::mem_access_type access_type;
switch (rw) {
case BX_READ:
access_type = log_data_st::MEM_READ;
break;
case BX_WRITE:
access_type = log_data_st::MEM_WRITE;
break;
case BX_EXECUTE:
access_type = log_data_st::MEM_EXEC;
break;
case BX_RW:
access_type = log_data_st::MEM_RW;
break;
default: abort();
}

// Disassemble current instruction.
static Bit8u ibuf[32] = {0};
static char pc_disasm[64];
if (read_lin_mem(pcpu, pc, sizeof(ibuf), ibuf)) {
disassembler bx_disassemble;
bx_disassemble.disasm(mode == 32, mode == 64, 0, pc, ibuf, pc_disasm);
}

// With basic information filled in, process the access further.
process_mem_access(pcpu, lin, len, pc, access_type, pc_disasm);
}

信息记录方式都是通过invoke_system_handler函数去处理自定义系统事件,目前主要支持4种操作系统(windows\linux\freebsd\openbsd),macOS还没搞过,原作者是说想继续实现macOS,这个值得尝试开发下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
const struct tag_kSystemEventHandlers {
const char *system;
s_event_handler_func handlers[BX_OS_EVENT_MAX];
} kSystemEventHandlers[] = {
{"windows",
{(s_event_handler_func)windows::init,
(s_event_handler_func)windows::check_kernel_addr,
(s_event_handler_func)windows::check_user_addr,
(s_event_handler_func)windows::fill_cid, // 获取线程环境块TEB,读取进程/线程ID
(s_event_handler_func)windows::fill_info, // 基于config.txt中配置的进线程结构offset去读取进线程信息,包括进程文件名、创建时间、栈回溯等信息
(s_event_handler_func)NULL}
},
{"linux",
{(s_event_handler_func)linux::init,
(s_event_handler_func)linux::check_kernel_addr,
(s_event_handler_func)linux::check_user_addr,
(s_event_handler_func)linux::fill_cid,
(s_event_handler_func)linux::fill_info,
(s_event_handler_func)NULL}
},
{"freebsd",
{(s_event_handler_func)freebsd::init,
(s_event_handler_func)freebsd::check_kernel_addr,
(s_event_handler_func)freebsd::check_user_addr,
(s_event_handler_func)freebsd::fill_cid,
(s_event_handler_func)freebsd::fill_info,
(s_event_handler_func)freebsd::instr_before_execution}
},
{"openbsd",
{(s_event_handler_func)openbsd::init,
(s_event_handler_func)openbsd::check_kernel_addr,
(s_event_handler_func)openbsd::check_user_addr,
(s_event_handler_func)openbsd::fill_cid,
(s_event_handler_func)openbsd::fill_info,
(s_event_handler_func)openbsd::instr_before_execution}
},
{NULL, {NULL, NULL, NULL, NULL, NULL}}
};

最后就是输出记录的信息,比如作者发现的CVE-2018-0894漏洞信息:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
------------------------------ found uninit-copy of address fffff8a000a63010

[pid/tid: 000001a0/000001a4] { wininit.exe}
COPY of fffff8a000a63010 ---> 1afab8 (64 bytes), pc = fffff80002698600
[ mov r11, rcx ]
Allocation origin: 0xfffff80002a11101
(ntoskrnl.exe!IopQueryNameInternal+00000071)
--- Shadow memory:
00000000: 00 00 00 00 ff ff ff ff 00 00 00 00 00 00 00 00 ................
00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
00000030: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
--- Actual memory:
00000000: 2e 00 30 00 aa aa aa aa 20 30 a6 00 a0 f8 ff ff ..0..... 0......
00000010: 5c 00 44 00 65 00 76 00 69 00 63 00 65 00 5c 00 \.D.e.v.i.c.e.\.
00000020: 48 00 61 00 72 00 64 00 64 00 69 00 73 00 6b 00 H.a.r.d.d.i.s.k.
00000030: 56 00 6f 00 6c 00 75 00 6d 00 65 00 32 00 00 00 V.o.l.u.m.e.2...
--- Stack trace:
#0 0xfffff80002698600 (ntoskrnl.exe!memmove+00000000)
#1 0xfffff80002a11319 (ntoskrnl.exe!IopQueryNameInternal+00000289)
#2 0xfffff800028d4426 (ntoskrnl.exe!IopQueryName+00000026)
#3 0xfffff800028e8fa8 (ntoskrnl.exe!ObpQueryNameString+000000b0)
#4 0xfffff8000291313b (ntoskrnl.exe!NtQueryVirtualMemory+000005fb)
#5 0xfffff800026b9283 (ntoskrnl.exe!KiSystemServiceCopyEnd+00000013)