We need to crack a DH-based protocol that allows for so called "contributory behaviour". Thus we can use one of the 32-byte values for Curve25519 that will make MITM-keys equal. The protocol was actually discussed in Thai Duong's blog several times and shows, that in some cases public keys for Curve25519 must be validated.
The organizers provided us with a network endpoint and an archive with code. The app can work as either server or client depending on the first network message.
server_or_client = ReadLine(reader)
is_server = server_or_client[0] in b'sS'
is_client = server_or_client[0] in b'cC'Client and server run a Diffie-Hellman based protocol to get a shared key and perform mutual authentication using a pre shared secret (PSS) like it's depicted on the following scheme:
And how it looks in the code:
def Handshake(password, reader, writer):
myPrivateKey = Private()
myNonce = os.urandom(32)
WriteBin(writer, myPrivateKey.get_public().serialize())
WriteBin(writer, myNonce)
theirPublicKey = ReadBin(reader)
theirNonce = ReadBin(reader)
if myNonce == theirNonce:
return None
if theirPublicKey in (b'\x00'*32, b'\x01' + (b'\x00' * 31)):
return None
theirPublicKey = Public(theirPublicKey)
sharedKey = myPrivateKey.get_shared_key(theirPublicKey)
myProof = ComputeProof(sharedKey, theirNonce + password)
WriteBin(writer, myProof)
theirProof = ReadBin(reader)
if not VerifyProof(sharedKey, myNonce + password, theirProof):
return None
return sharedKeyProof is computed as follows:
def ComputeProof(key, data):
return hmac.new(key, data, digestmod=hashlib.sha256).digest()def VerifyProof(key, data, proof):
return hmac.compare_digest(ComputeProof(key, data), proof)where key is a shared key computed during the DH-dance and the data parameter is nonce + password.
To get the flag we need to send an encrypted message b'getflag' to the server.
As we can see from the handshake code:
-
There's no way to leak the password (PSS) as it's used only as a parameter of HMAC-SHA256.
-
Server checks whether his nonce equals to the client's nonce, so we can't impersonate the client and replay the server's nonce and server's proof back to the server.
-
The public key is checked for being not in
(b'\x00'*32, b'\x01' + (b'\x00' * 31))
So as we can't leak the PSS, probably we don't need it and we can try to subvert the protocol in another way. If we perform a MITM attack and make the keys equal for the both MITM sides (from the client to the attacker and from the attacker to the server), then we don't need to know the PSS and we can pass the client's proof to the server and pass the authentication step. Thus:
- We need to have the same key for the both sides during MITM
- We need to know the MITM key
The public key check looks "weak" so we can look for some more values which the public key isn't checked against. We can find some on djb's web site, and I picked the following one:
325606250916557431795983626356110631294008115727848805560023387167927233504and carried out some experiments:
from curve25519 import Private, Public
from binascii import hexlify
keys = set()
for i in range(30):
myPrivate = Private()
val = int.to_bytes(325606250916557431795983626356110631294008115727848805560023387167927233504, 32, 'little')
theirPublic = Public(val)
shared = myPrivate.get_shared_key(theirPublic)
print(hexlify(shared))
keys.add(shared)
print("[~] Num of different keys: {}".format(len(keys)))The test produced 30 equal shared keys, so we can carry out the attack.
The complete code of the attack:
#!/usr/bin/env python3
from curve25519 import Public
import nacl.secret
import socket
import binascii
socket_server = ("mitm.ctfcompetition.com", 1337)
public_key_val = Public(int.to_bytes(325606250916557431795983626356110631294008115727848805560023387167927233504, 32, 'little'))
shared_key = binascii.unhexlify(b'68b59f127c671255346e099c3b9ea067a5595ba2adf26daa5e69d6a8a29d191a')
def make_sockets():
server = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
server.connect(socket_server)
client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
client.connect(socket_server)
return server, client
def attack():
s, c = make_sockets()
s.send(b's\n')
c.send(b'c\n')
server_pub = s.recv(4096)[:-1]
print(server_pub.decode())
server_nonce = s.recv(4096)[:-1]
print(server_nonce.decode())
client_pub = c.recv(4096)[:-1]
print(client_pub.decode())
client_nonce = c.recv(4096)[:-1]
print(client_nonce.decode())
s.send(binascii.hexlify(public_key_val.serialize()) + b'\n')
s.send(client_nonce + b'\n')
c.send(binascii.hexlify(public_key_val.serialize()) + b'\n')
c.send(server_nonce + b'\n')
server_proof = s.recv(4096)[:-1]
client_proof = c.recv(4096)[:-1]
# send server's proof to the client
c.send(server_proof + b'\n')
# send client's proof to the server
s.send(client_proof + b'\n')
# close client socket
c.close()
# read auth string
auth_string = s.recv(4096)[:-1]
box = nacl.secret.SecretBox(shared_key)
data = box.decrypt(binascii.unhexlify(auth_string))
if data == b'AUTHENTICATED':
print("[~] Got AUTHENTICATED")
get_flag_cmd = b"getflag"
encrypted_cmd = box.encrypt(get_flag_cmd)
s.send(binascii.hexlify(encrypted_cmd) + b'\n')
encrypted_flag = s.recv(4096)[:-1]
data = box.decrypt(binascii.unhexlify(encrypted_flag))
print("[~] The flag is {}".format(data.decode()))
if __name__ == "__main__":
attack()In the code above I used the shared key I got from the experiment. The code just uses the 32-byte value that produces the mentioned shared key and passes client's nonce and proof to the server. Then it encrypts b'getflag' using the shared key and reads the frag from the server.
The code produces the following output:
python solution.py
9e5a8f40824b36adad9505050cfb89eccd94c3c19226b735dd8f862736387e77
26c206bebc1fbb6ebf018588476854c401f1bb594399f3aff661bc65f631cadf
2dc4548ab69d7906023d1dd3db080da88304b08fd8c989e5ead9f39a93802e0a
87a32b8304af3c1225b7c1b78ec0b0759710c98a3790b75c01312b8b3a5ad140
[~] Got AUTHENTICATED
b'CTF{kae3eebav8Ac7Mi0RKgh6eeLisuut9oP}'-
If the protocol allows for so called "contributory behavior", Curve25519 public keys must either be validated or another mitigation must be applied.
-
There's an interesting discussion whether Curve25519 public keys should be validated.
-
The Thai Duong's The Internet of Broken Protocols series of posts are worth reading :)