mirror of
https://github.com/SystemRage/py-kms.git
synced 2024-11-29 19:54:16 +01:00
126 lines
3.8 KiB
Python
126 lines
3.8 KiB
Python
|
#!/usr/bin/env python
|
||
|
|
||
|
import binascii
|
||
|
import struct
|
||
|
import time
|
||
|
import logging
|
||
|
|
||
|
from kmsBase import kmsBase
|
||
|
from structure import Structure
|
||
|
from aes import AES
|
||
|
from formatText import shell_message, justify
|
||
|
|
||
|
# v4 AES Key
|
||
|
key = bytearray([0x05, 0x3D, 0x83, 0x07, 0xF9, 0xE5, 0xF0, 0x88, 0xEB, 0x5E, 0xA6, 0x68, 0x6C, 0xF0, 0x37, 0xC7, 0xE4, 0xEF, 0xD2, 0xD6])
|
||
|
|
||
|
# Xor Buffer
|
||
|
def xorBuffer(source, offset, destination, size):
|
||
|
for i in range(0, size):
|
||
|
destination[i] ^= source[i + offset]
|
||
|
|
||
|
class kmsRequestV4(kmsBase):
|
||
|
class RequestV4(Structure):
|
||
|
commonHdr = ()
|
||
|
structure = (
|
||
|
('bodyLength1', '<I'),
|
||
|
('bodyLength2', '<I'),
|
||
|
('request', ':', kmsBase.kmsRequestStruct),
|
||
|
('hash', '16s'),
|
||
|
('padding', ':'),
|
||
|
)
|
||
|
|
||
|
class ResponseV4(Structure):
|
||
|
commonHdr = ()
|
||
|
structure = (
|
||
|
('bodyLength1', '<I=len(response) + len(hash)'),
|
||
|
('unknown', '!I=0x00000200'),
|
||
|
('bodyLength2', '<I=len(response) + len(hash)'),
|
||
|
('response', ':', kmsBase.kmsResponseStruct),
|
||
|
('hash', '16s'),
|
||
|
('padding', ':'),
|
||
|
)
|
||
|
|
||
|
def executeRequestLogic(self):
|
||
|
requestData = self.RequestV4(self.data)
|
||
|
|
||
|
response = self.serverLogic(requestData['request'])
|
||
|
hash = self.generateHash(bytearray(str(response)))
|
||
|
|
||
|
self.responseData = self.generateResponse(response, hash)
|
||
|
|
||
|
time.sleep(1) # request sent back too quick for Windows 2008 R2, slow it down.
|
||
|
|
||
|
def generateHash(self, message):
|
||
|
"""
|
||
|
The KMS v4 hash is a variant of CMAC-AES-128. There are two key differences:
|
||
|
* The 'AES' used is modified in particular ways:
|
||
|
* The basic algorithm is Rjindael with a conceptual 160bit key and 128bit blocks.
|
||
|
This isn't part of the AES standard, but it works the way you'd expect.
|
||
|
Accordingly, the algorithm uses 11 rounds and a 192 byte expanded key.
|
||
|
* The trailing block is not XORed with a generated subkey, as defined in CMAC.
|
||
|
This is probably because the subkey generation algorithm is only defined for
|
||
|
situations where block and key size are the same.
|
||
|
"""
|
||
|
aes = AES()
|
||
|
|
||
|
messageSize = len(message)
|
||
|
lastBlock = bytearray(16)
|
||
|
hashBuffer = bytearray(16)
|
||
|
|
||
|
# MessageSize / Blocksize
|
||
|
j = messageSize >> 4
|
||
|
|
||
|
# Remainding bytes
|
||
|
k = messageSize & 0xf
|
||
|
|
||
|
# Hash
|
||
|
for i in range(0, j):
|
||
|
xorBuffer(message, i << 4, hashBuffer, 16)
|
||
|
hashBuffer = bytearray(aes.encrypt(hashBuffer, key, len(key)))
|
||
|
|
||
|
# Bit Padding
|
||
|
ii = 0
|
||
|
for i in range(j << 4, k + (j << 4)):
|
||
|
lastBlock[ii] = message[i]
|
||
|
ii += 1
|
||
|
lastBlock[k] = 0x80
|
||
|
|
||
|
xorBuffer(lastBlock, 0, hashBuffer, 16)
|
||
|
hashBuffer = bytearray(aes.encrypt(hashBuffer, key, len(key)))
|
||
|
|
||
|
return str(hashBuffer)
|
||
|
|
||
|
def generateResponse(self, responseBuffer, hash):
|
||
|
bodyLength = len(responseBuffer) + len(hash)
|
||
|
response = self.ResponseV4()
|
||
|
response['response'] = responseBuffer
|
||
|
response['hash'] = hash
|
||
|
response['padding'] = self.getResponsePadding(bodyLength)
|
||
|
|
||
|
shell_message(nshell = 16)
|
||
|
logging.debug("KMS V4 Response: %s" % justify(response.dump(print_to_stdout = False)))
|
||
|
logging.debug("KMS V4 Response Bytes: %s" % justify(binascii.b2a_hex(str(response))))
|
||
|
|
||
|
return str(response)
|
||
|
|
||
|
def getResponse(self):
|
||
|
return self.responseData
|
||
|
|
||
|
def generateRequest(self, requestBase):
|
||
|
hash = str(self.generateHash(bytearray(str(requestBase))))
|
||
|
|
||
|
bodyLength = len(requestBase) + len(hash)
|
||
|
|
||
|
request = kmsRequestV4.RequestV4()
|
||
|
request['bodyLength1'] = bodyLength
|
||
|
request['bodyLength2'] = bodyLength
|
||
|
request['request'] = requestBase
|
||
|
request['hash'] = hash
|
||
|
request['padding'] = self.getResponsePadding(bodyLength)
|
||
|
|
||
|
shell_message(nshell = 10)
|
||
|
logging.debug("Request V4 Data: %s" % justify(request.dump(print_to_stdout = False)))
|
||
|
logging.debug("Request V4: %s" % justify(binascii.b2a_hex(str(request))))
|
||
|
|
||
|
return request
|