最近在搞Iot的时候接触到Qiling框架,用了一段时间后感觉确实模拟功能挺强大的,还支持Fuzz,于是开始学习对Iot webserver这样的程序进行Fuzz。
官方给出了类似的例子如Tenda AC15 的httpd的fuzz脚本,但是也就光秃秃一个脚本还是需要自己来一遍才能学到一些东西;因为面向的是Iot webserver的Fuzz因此需要对嵌入式设备中常用web开源框架有一些了解,这里是对于Boa框架的fuzz案例
环境准备:
• qiling-dev branch:这里并没有选择直接pip安装,方便修改源码
• AFL++:在python中可以import unicornafl
就行
• git clone https://github.com/AFLplusplus/AFLplusplus.git
make -C AFLplusplus
cd AFLplusplus/unicorn_mode ; ./build_unicorn_support.sh
• 一个坑:最好获取版本高于3.15的cmake
,要不然编译的时候有些cmake参数识别有问题,我遇到的就是:cmake -S unicorn/ -B unicorn/build -D BUILD_SHARED_LIBS=no
• 需要对Qiling、AFL有些了解
Fuzz思路:
Iot设备就连环境模拟都比较棘手就就更别说Fuzz了,但是Qiling提供的进程快照(snapshot)
功能给了我们一个不错的思路,这也是Qiling官方Fuzz案例的一个思路:即对某函数部分Fuzz(Partial Fuzz)
Qiling使用4个脚本来实现对该款路由器上httpd程序的Fuzz
image-20221213114209793
首先是saver_tendaac15_httpd.py
用于保存fuzz的起始状态快照,主要代码如下:
def save_context(ql, *args, **kw):
ql.save(cpu_context=False, snapshot="snapshot.bin")
def check_pc(ql):
print("=" * 50)
print("Hit fuzz point, stop at PC = 0x%x" % ql.arch.regs.arch_pc)
print("=" * 50)
ql.emu_stop()
def my_sandbox(path, rootfs):
ql = Qiling(path, rootfs, verbose=QL_VERBOSE.DEBUG)
ql.add_fs_mapper("/dev/urandom","/dev/urandom")
ql.hook_address(save_context, 0x10930) #<=======
ql.hook_address(patcher, ql.loader.elf_entry)
ql.hook_address(check_pc, 0x7a0cc) #<=======
ql.run()
ql.hook_address(save_context, 0x10930):表示当程序跑到0x10930地址时调用save_context函数将保存此刻模拟状态
但需要输入来触发程序按照预想的跑到0x10930位置,带上面脚本跑起来后使用addressNat_overflow.sh
触发
#!/bin/sh
curl -v -H "X-Requested-With: XMLHttpRequest" -b "password=1234" -e http://localhost:8080/samba.html -H "Content-Type:application/x-www-form-urlencoded" --data "entrys=sync" --data "page=CCCCAAAA" http://localhost:8080/goform/addressNat
那么我们获得了模拟进程快照snapshot.bin
之后fuzz就重复利用该文件启动就行,对应fuzz_tendaac15_httpd.py
def main(input_file, enable_trace=False):
ql = Qiling(["rootfs/bin/httpd"], "rootfs", verbose=QL_VERBOSE.DEBUG, console = True if enable_trace else False)
# save current emulated status
ql.restore(snapshot="snapshot.bin")
# return should be 0x7ff3ca64
fuzz_mem=ql.mem.search(b"CCCCAAAA")
target_address = fuzz_mem[0]
def place_input_callback(_ql: Qiling, input: bytes, _):
_ql.mem.write(target_address, input)
def start_afl(_ql: Qiling):
"""
Callback from inside
"""
ql_afl_fuzz(_ql, input_file=input_file, place_input_callback=place_input_callback, exits=[ql.os.exit_point])
ql.hook_address(callback=start_afl, address=0x10930+8)
try:
ql.run(begin = 0x10930+4, end = 0x7a0cc+4)
os._exit(0)
except:
if enable_trace:
print("\nFuzzer Went Shit")
os._exit(0)
if __name__ == "__main__":
if len(sys.argv) == 1:
raise ValueError("No input file provided.")
if len(sys.argv) > 2 and sys.argv[1] == "-t":
main(sys.argv[2], enable_trace=True)
else:
main(sys.argv[1])
• 恢复快照:ql.restore(snapshot="snapshot.bin")
• 变异数据缓存定位:fuzz_mem=ql.mem.search(b"CCCCAAAA")
• 以hook方式从起始地址附近的开始fuzz:ql.hook_address(callback=start_afl, address=0x10930+8)
最后开始Fuzz
#!/usr/bin/sh
AFL_DEBUG_CHILD_OUTPUT=1 AFL_AUTORESUME=1 AFL_PATH="$(realpath ./AFLplusplus)" PATH="$AFL_PATH:$PATH" ./AFLplusplus/afl-fuzz -i afl_inputs -o afl_outputs -U -- python3 ./fuzz_tendaac15_httpd.py @@
说实话这样连最关键的fuzz范围0x10930
,0x7a0cc
怎么来的都不知道当时逆向定位这两个地址也是一头雾水毫无特征,还是得自己实操
因此选定了Boa框架(之前了解过源码)从零开始对其进行Fuzz
选择一个网上有许多漏洞分析的设备:vivetok 摄像头,链接见参考,而且webservre为Boa框架
Poc:
echo -en "POST /cgi-bin/admin/upgrade.cgi HTTP/1.0\nContent-Length:AAAAAAAAAAAAAAAAAAAABBBBCCCCDDDDEEEEFFFFGGGGHHHHIIIIXXXX\n\r\n\r\n" | ncat -v 192.168.57.20 80
Boa框架:
主要处理逻辑在process_requests
函数中:
/*获取就绪队列并处理*/
current = request_ready;
while (current) {
time(¤t_time);
if (current->buffer_end && /* there is data in the buffer */
current->status != DEAD && current->status != DONE) {
retval = req_flush(current);
/*
* retval can be -2=error, -1=blocked, or bytes left
*/
if (retval == -2) { /* error */
current->status = DEAD;
retval = 0;
} else if (retval >= 0) {
/* notice the >= which is different from below?
Here, we may just be flushing headers.
We don't want to return 0 because we are not DONE
or DEAD */
retval = 1;
}
} else {/*主要处理请求部分在这里*/
switch (current->status) {
case READ_HEADER:
case ONE_CR:
case ONE_LF:
case TWO_CR:
retval = read_header(current); //解析request头部,该函数类似与FILE_IO
break; //函数request内部有8192+1字节的buffer,data的头尾指针等,最终调用
case BODY_READ: //bytes = read(req->fd, buffer + req->client_stream_pos, buf_bytes_left);读取
retval = read_body(current);
break;
case BODY_WRITE:
retval = write_body(current);
break;
case WRITE:
retval = process_get(current);
break;
case PIPE_READ:
retval = read_from_pipe(current);
break;
case PIPE_WRITE:
retval = write_from_pipe(current);
break;
case DONE:
/* a non-status that will terminate the request */
retval = req_flush(current);
/*
* retval can be -2=error, -1=blocked, or bytes left
*/
if (retval == -2) { /* error */
current->status = DEAD;
retval = 0;
} else if (retval > 0) {
retval = 1;
}
break;
case DEAD:
retval = 0;
current->buffer_end = 0;
SQUASH_KA(current);
break;
default:
retval = 0;
fprintf(stderr, "Unknown status (%d), "
"closing!\n", current->status);
current->status = DEAD;
break;
}
}
主要看中间的Switch case:
• read_header:解析request头部,该函数类似FILE_IO函数
• request内部有8192+1字节的buffer,data的头尾指针等,最终调用bytes = read(req->fd, buffer + req->client_stream_pos, buf_bytes_left);读取client发送的请求
• 会提取并解析头部信息
• 对于GET传参,主要使用read_header, read_from_pipe, write_from_pipe完成cgi的调用
• 对于POST传参,主要调用read_header, read_body, write_body完成cgi调用
就拿read_header函数来说,厂商应该会在里面增加一些url过虑以及响应处理,在这个摄像头中漏洞也确实出在这个函数:
image-20221213133117933
没有对Content-Length
成员做限制;根据源码中提示字符串Unknown status (%d), closing
可以轻松定位到这几个函数:
image-20221213133545416
那么接下来就尝试利用Qiling 启动这个程序并且Partial Fuzz函数"read_header"
模拟启动的宗旨(我的)是遇到啥错误修最后一个报错点
启动模板:
import os, sys
sys.path.append('/home/iot/workspace/Emulator/qiling-dev')
from qiling import Qiling
from qiling.const import QL_INTERCEPT, QL_VERBOSE
def boa_run(path: list, rootfs: str, profile: str = 'default'):
ql = Qiling(path, rootfs, profile=profile, verbose=QL_VERBOSE.OFF, multithread=False)
"""setup files"""
ql.add_fs_mapper('/dev/null', '/dev/null')
"""hooks"""
ql.run()
if __name__ == '__main__':
os.chdir('/home/iot/workspace/Emulator/qiling-dev/vivetok')
path = ['./rootfs/usr/sbin/httpd', "-c", "/etc/conf.d/boa", "-d"]
rootfs = './rootfs'
profile = './boa_arm.ql'
boa_run(path=path, rootfs=rootfs, profile=profile)
尝试启动
首先遇到的是:gethostbyname:: Success
在IDA中定位到:
image-20221213134138571
函数原型:
struct hostent *gethostbyname(const char *hostname);
struct hostent{
char *h_name; //official name
char **h_aliases; //alias list
int h_addrtype; //host address type
int h_length; //address lenght
char **h_addr_list; //address list
}
获取返回的结构体还挺复杂的,问题的原因是 在调用gethostname
将获得ql_vm作为主机名所以当以此调用gethostbyname
无法获得主机信息,所以hook这个函数,并提前开辟空间存放伪造信息:
"""
struct hostent{
char *h_name; //official name
char **h_aliases; //alias list
int h_addrtype; //host address type
int h_length; //address lenght
char **h_addr_list; //address list
}
"""
def hook_memSpace(ql: Qiling):
ql.mem.map(0x1000, 0x1000, info='my_hook')
data = struct.pack('<IIIII', 0x1100, 0x1100, AF_INET, 4, 0x1100)
ql.mem.write(0x1000, data)
ql.mem.write(0x1100, b'qiling')
def lib_gethostbyname(ql: Qiling):
args = ql.os.resolve_fcall_params({'name':STRING})
print('[gethostbyname]: ' + args['name'])
ql.arch.regs.write('r0', 0x1000)
还有一个严重问题就是模拟过程中程序自动采用ipv6
协议,这就很烦因为qiling的ipv6协议支持的不是很好
AttributeError: 'sockaddr_in' object has no attribute 'sin6_addr'
问题处在对ipv6的系统调用bind:
elif sa_family == AF_INET6 and ql.os.ipv6:
sockaddr_in6 = make_sockaddr_in(abits, endian)
sockaddr_obj = sockaddr_in6.from_buffer(data)
port = ntohs(ql, sockaddr_obj.sin_port)
host = inet6_ntoa(sockaddr_obj.sin6_addr.s6_addr)
if ql.os.bindtolocalhost:
host = '::1'
if not ql.os.root and port <= 1024:
port = port + 8000
def make_sockaddr_in(archbits: int, endian: QL_ENDIAN):
Struct = struct.get_aligned_struct(archbits, endian)
class in_addr(Struct):
_fields_ = (
('s_addr', ctypes.c_uint32),
)
class sockaddr_in(Struct):
_fields_ = (
('sin_family', ctypes.c_int16),
('sin_port', ctypes.c_uint16),
('sin_addr', in_addr),
('sin_zero', ctypes.c_byte * 8)
)
return sockaddr_in
def make_sockaddr_in6(archbits: int, endian: QL_ENDIAN):
Struct = struct.get_aligned_struct(archbits, endian)
class in6_addr(Struct):
_fields_ = (
('s6_addr', ctypes.c_uint8 * 16),
)
class sockaddr_in6(Struct):
_fields_ = (
('sin6_family', ctypes.c_int16),
('sin6_port', ctypes.c_uint16),
('sin6_flowinfo', ctypes.c_uint32),
('sin6_addr', in6_addr),
('sin6_scope_id', ctypes.c_uint32)
)
return sockaddr_in6
make_sockaddr_in, make_sockaddr_in6基于ctypes
构造严格的sockaddr结构体,因为是ipv6所以得用make_sockaddr_in6
还有就是函数(function) inet6_ntoa: (addr: bytes) -> str
需要bytes对象,而sockaddr_obj.sin6_addr.s6_addr是cbytes类型所以得bytes转
sockaddr_in6 = make_sockaddr_in6(abits, endian)
sockaddr_obj = sockaddr_in6.from_buffer(data)
port = ntohs(ql, sockaddr_obj.sin6_port)
host = inet6_ntoa(bytes(sockaddr_obj.sin6_addr.s6_addr))
OSError: [Errno 98] Address already in use
还是在调用bind时候,因为qiling会对低于1024的端口bind进行修改:
if not ql.os.root and port <= 1024:
port = port + 8000
而后面还对8080端口进行一次bind,所以这里得改,然后其实就能进入核心处理逻辑了 :
image-20221213134113202
当然还得看看链接有没有问题:尝试访问又出现问题
$ echo -en "GET /index.html HTTP/1.0\n\rContent-Length:20\n\r\n\r" | nc -v ::1 9080
Connection to ::1 9080 port [tcp/*] succeeded!
File "/home/iot/workspace/Emulator/qiling-dev-stb/qiling/os/posix/syscall/socket.py", line 669, in ql_syscall_accept
host, port = address
ValueError: too many values to unpack (expected 2)
ValueError: too many values to unpack (expected 2)
经调试原来在python中accept ipv6的连接后会返回一个长度为4的元组的address:
image-20221213134207632
同样的问题还发生在ql_syscall_getsockname:sockname = sock.getsockname()
TypeError: expected c_ubyte_Array_16 instance, got int
[x] Syscall ERROR: ql_syscall_accept DEBUG: expected c_ubyte_Array_16 instance, got int
Traceback (most recent call last):
File "/home/iot/workspace/Emulator/qiling-dev-stb/qiling/os/posix/posix.py", line 280, in load_syscall
retval = syscall_hook(self.ql, *params)
File "/home/iot/workspace/Emulator/qiling-dev-stb/qiling/os/posix/syscall/socket.py", line 674, in ql_syscall_accept
obj.sin6_addr.s6_addr = inet6_aton(str(host))
TypeError: expected c_ubyte_Array_16 instance, got int
解决:bytes转cbyts类
obj.sin6_addr.s6_addr = (ctypes.c_ubyte * 16).from_buffer_copy(inet6_aton(str(host)).to_bytes(16, 'big'))
主要问题就这些(修了挺久的),然后就可以对一些函数进行fuzz了
确定Fuzz范围,这个范围主要是给到ql_afl_fuzz函数,这里是打算Fuzz read_header函数(sub_17F80),那么从数据入口下手:
image-20221213135606979
读取POST或者GET方法的http包那么肯定要解析处理的,处理完成返回一个状态(源码中retval)来指示下一步处理,找到退出点:
image-20221213135843221
因此要从0x180F8
附近开始Fuzz,然后0x18398
表示函数正常退出将执行下一轮fuzz
脚本模板:
import os, sys
sys.path.append('/home/iot/workspace/Emulator/qiling-dev')
from qiling.const import QL_INTERCEPT, QL_VERBOSE
from qiling import Qiling
from qiling.extensions.afl import ql_afl_fuzz
def main(input_file: str, trace: bool = False):
ql = Qiling(['./rootfs/usr/sbin/httpd', "-c", "/etc/conf.d/boa", "-d"], rootfs='./rootfs', profile='./boa_arm.ql', verbose=QL_VERBOSE.OFF, console = True if trace else False)
ql.restore(snapshot='./context.bin')
def place_input_callback(_ql: Qiling, input: bytes, _):
# print(b"**************** " + input)
_ql.mem.write(target_addr, input)
def start_afl(_ql: Qiling):
"""
Callback from inside
"""
ql_afl_fuzz(_ql, input_file=input_file, place_input_callback=place_input_callback, exits=[0x018398])
ql.hook_address(callback=start_afl, address=0x180F8)
try:
# ql.debugger = True
ql.run(begin=0x180F8)
os._exit(0)
except:
if trace:
print("\nFuzzer Went Shit")
os._exit(0)
if __name__ == "__main__":
if len(sys.argv) == 1:
raise ValueError("No input file provided.")
os.chdir('/home/iot/workspace/Emulator/qiling-dev/vivetok')
if len(sys.argv) > 2 and sys.argv[1] == "-t":
main(sys.argv[2], trace=True)
else:
main(sys.argv[1])
• ql.hook_address(callback=start_afl, address=0x180F8):在执行到0x180F8
这个位置时调用start_afl函数
• ql.run(begin=0x180F8):从0x180F8
开始执行
• ql_afl_fuzz:就是unicornafl提供的fuzz接口uc_afl_fuzz_custom
的一个wrapper
• place_input_callback:ql_afl_fuzz会调用的回调函数,负责写入fuzz数据
根据网上的漏洞分析比对源码框架,利用:
cho -en "POST /cgi-bin/admin/upgrade.cgi HTTP/1.0nContent-Length:AAAAAAAAAAAAAAAAAAAABBBBCCCCDDDDEEEEFFFFGGGGHHHHIIIIXXXXnrnrn" | nc -v ::1 9080
可以触发漏洞,具体位于框架中http头部解析函数:read_header,位于httpd中17F80位置
那么该如何fuzz呢,根据网上unicorn-afl官方用例和qiling官方用例:buf-fuzz,即定位代码中读取数据位置,然后读取完后劫持搜索特定字符串定位fuzz的buff_addr,当然需要状态保存(当然这个方法肯定不是很严谨,因此后面还会介绍劫持read函数方法)
import os, sys, struct
from socket import AF_INET
sys.path.append('/home/iot/workspace/Emulator/qiling-dev')
from qiling import Qiling
from qiling.const import QL_INTERCEPT, QL_VERBOSE
from qiling.os.const import STRING
from unicorn.unicorn import UcError
"""
struct hostent{
char *h_name; //official name
char **h_aliases; //alias list
int h_addrtype; //host address type
int h_length; //address lenght
char **h_addr_list; //address list
}
"""
def hook_memSpace(ql: Qiling):
ql.mem.map(0x1000, 0x1000, info='my_hook')
data = struct.pack('<IIIII', 0x1100, 0x1100, AF_INET, 4, 0x1100)
ql.mem.write(0x1000, data)
ql.mem.write(0x1100, b'qiling')
def lib_gethostbyname(ql: Qiling):
args = ql.os.resolve_fcall_params({'name':STRING})
print('[gethostbyname]: ' + args['name'])
ql.arch.regs.write('r0', 0x1000)
def saver(ql: Qiling):
print('[!] Hit Saver 0x%X'%(ql.arch.regs.arch_pc))
ql.save(cpu_context=False, snapshot='./context.bin')
print(ql.mem.search(b'fuck'))
#[read(5, 0x4edca, 0x2000)] locate buf
def read_syscall(ql: Qiling, fd: int, buf: int, size: int, *args) -> None:
print(f'[read({fd}, {buf: #x}, {size: #x})]')
def boa_run(path: list, rootfs: str, profile: str = 'default'):
ql = Qiling(path, rootfs, profile=profile, verbose=QL_VERBOSE.OFF, multithread=False)
"""setup files"""
ql.add_fs_mapper('/dev/null', '/dev/null')
"""set ram"""
hook_memSpace(ql)
"""hooks"""
ql.os.set_api('gethostbyname', lib_gethostbyname, QL_INTERCEPT.CALL)
ql.os.set_syscall('read', read_syscall, QL_INTERCEPT.ENTER)
"""setup saver"""
ql.hook_address(saver, 0x0180FC) #read finish
ql.run()
if __name__ == '__main__':
os.chdir('/home/iot/workspace/Emulator/qiling-dev/vivetok')
path = ['./rootfs/usr/sbin/httpd', "-c", "/etc/conf.d/boa", "-d"]
rootfs = './rootfs'
profile = './boa_arm.ql'
boa_run(path=path, rootfs=rootfs, profile=profile)
然后使用poc触发就行
import os, sys, struct
import capstone as Cs
sys.path.append('/home/iot/workspace/Emulator/qiling-dev')
from qiling.const import QL_INTERCEPT, QL_VERBOSE
from qiling import Qiling
from qiling.extensions.afl import ql_afl_fuzz
def simple_diassembler(ql: Qiling, address: int, size: int, md: Cs) -> None:
buf = ql.mem.read(address, size)
for insn in md.disasm(buf, address):
ql.log.debug(f':: {insn.address:#x} : {insn.mnemonic:24s} {insn.op_str}')
def main(input_file: str, trace: bool = False):
ql = Qiling(['./rootfs/usr/sbin/httpd', "-c", "/etc/conf.d/boa", "-d"], rootfs='./rootfs', profile='./boa_arm.ql', verbose=QL_VERBOSE.OFF, console = True if trace else False)
ql.restore(snapshot='./context.bin')
fuzz_mem = ql.mem.search(b'fuck')
target_addr = fuzz_mem[0]
def place_input_callback(_ql: Qiling, input: bytes, _):
# print(b"**************** " + input)
_ql.mem.write(target_addr, input)
def start_afl(_ql: Qiling):
"""
Callback from inside
"""
ql_afl_fuzz(_ql, input_file=input_file, place_input_callback=place_input_callback, exits=[0x018398])
ql.hook_address(callback=start_afl, address=0x0180FC+4)
# ql.hook_code(simple_diassembler, begin=0x0180FC, end=0x018600, user_data=ql.arch.disassembler)
try:
# ql.debugger = True
ql.run(begin=0x0180FC+4, end=0x018600) #注意arm函数返回地址比较奇怪,不一定在函数末尾
os._exit(0)
except:
if trace:
print("\nFuzzer Went Shit")
os._exit(0)
if __name__ == "__main__":
if len(sys.argv) == 1:
raise ValueError("No input file provided.")
os.chdir('/home/iot/workspace/Emulator/qiling-dev/vivetok')
if len(sys.argv) > 2 and sys.argv[1] == "-t":
main(sys.argv[2], trace=True)
else:
main(sys.argv[1])
这里很坑的一点是,在漏洞中因为Content-Length
成员不以\n结尾时就会让v31等于0会让strncpy报错但是不一定是pc指针错误,而是某些指令地址操作数问题
v30 = strstr(haystack, "Content-Length");
v31 = strchr(v30, '\n');
v32 = strchr(v30, ':');
strncpy(dest, v32 + 1, v31 - (v32 + 1));
在源码中AFL模块调用以下函数完成fuzz执行:
def _dummy_fuzz_callback(_ql: "Qiling"):
if isinstance(_ql.arch, QlArchARM):
pc = _ql.arch.effective_pc
else:
pc = _ql.arch.regs.arch_pc
try:
_ql.uc.emu_start(pc, 0, 0, 0)
except UcError as e:
os.abort() #添加部分
return e.errno
因此添加os.abort
通知AFL程序崩溃
image-20221213140214049
上面直接对buf写入Fuzz数据肯定不是一个很理想的办法(比如Fuzz数据超出读取长度),当然人家给的例子就是这么Fuzz的也不失一种方法;之后
就尝试利用Qiling的系统调用劫持功能让Fuzz效果更好。
从read函数调用处开始执行,在这之前劫持read函数调用让程序直接读取文件输入:
def read_syscall(ql: Qiling, fd: int, buf: int, size: int, *args) -> int:
# print(fd, buf, size)
data = ql.os.stdin.read(size)
# print(data)
ql.mem.write(buf, data)
return len(data)
def place_input_callback(_ql: Qiling, input: bytes, _):
# print(b"**************** " + input)
ql.os.stdin.write(input)
return True
def start_afl(_ql: Qiling):
"""
Callback from inside
"""
ql_afl_fuzz(_ql, input_file=input_file, place_input_callback=place_input_callback, exits=[0x018398])
同样写个脚本把服务并且设置debugger等待gdb连接:
image-20221213143927097
然后将crash中的数据发送:
image-20221213144007558
也确实触发到了漏洞:
0x900a5d74 in strncpy () from target:/lib/libc.so.0
gef➤ backtrace
#0 0x900a5d74 in strncpy () from target:/lib/libc.so.0
#1 0x0001853c in ?? ()
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
gef➤
fuzz过程中不好调试连写的harness有没有效果都不知道,可以使用capstone同步解析执行汇编情况:
def simple_diassembler(ql: Qiling, address: int, size: int, md: Cs) -> None:
buf = ql.mem.read(address, size)
for insn in md.disasm(buf, address):
ql.log.debug(f':: {insn.address:#x} : {insn.mnemonic:24s} {insn.op_str}')
• 固件链接
• Demo - Qiling Framework Documentation
• IOT Fuzz 两种思路
• vivetok 摄像头远程栈溢出漏洞分析-安全客 - 安全资讯平台 (anquanke.com)
• Vivotek远程栈溢出漏洞分析与复现 - 先知社区 (aliyun.com)
• 基于Unicorn和LibFuzzer的模拟执行fuzzing
• 基于 unicorn 的单个函数模拟执行和 fuzzer 实现