mirror of
https://github.com/hashcat/hashcat.git
synced 2024-12-22 22:58:30 +00:00
347 lines
9.1 KiB
Python
Executable File
347 lines
9.1 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
|
|
#
|
|
# Author......: See docs/credits.txt
|
|
# License.....: MIT
|
|
#
|
|
|
|
from argparse import ArgumentParser
|
|
from collections import namedtuple
|
|
from dataclasses import dataclass
|
|
from os import SEEK_SET
|
|
from struct import Struct
|
|
from typing import List
|
|
|
|
try:
|
|
from enum import auto, IntEnum, StrEnum
|
|
except ImportError:
|
|
from enum import auto, Enum, IntEnum
|
|
|
|
class StrEnum(str, Enum):
|
|
def _generate_next_value_(name, start, count, last_values):
|
|
return name.lower()
|
|
|
|
__str__ = str.__str__
|
|
|
|
__format__ = str.__format__
|
|
|
|
|
|
# consts
|
|
|
|
|
|
SIGNATURE = "$luks$"
|
|
SECTOR_SIZE = 512
|
|
|
|
|
|
# utils
|
|
|
|
|
|
def bytes_to_str(value):
|
|
"""
|
|
Convert encoded padded bytes string into str.
|
|
"""
|
|
return value.rstrip(b"\0").decode()
|
|
|
|
|
|
# pre-header
|
|
|
|
|
|
TmpHeaderPre = namedtuple(
|
|
"TmpHeaderPre",
|
|
(
|
|
"magic",
|
|
"version",
|
|
),
|
|
)
|
|
|
|
|
|
# version 1
|
|
|
|
|
|
TmpKeyVersion1 = namedtuple(
|
|
"TmpKeyVersion1",
|
|
(
|
|
"active",
|
|
"iterations",
|
|
"salt",
|
|
"material_offset",
|
|
"stripes",
|
|
),
|
|
)
|
|
|
|
|
|
@dataclass(init=False)
|
|
class KeyVersion1:
|
|
class Active(IntEnum):
|
|
ENABLED = 0x00AC71F3
|
|
DISABLED = 0x0000DEAD
|
|
ENABLED_OLD = 0xCAFE
|
|
DISABLED_OLD = 0x0000
|
|
|
|
active: Active
|
|
iterations: int
|
|
salt: bytes
|
|
af: bytes
|
|
|
|
def __init__(self, active, iterations, salt, af):
|
|
self.active = self.Active(active)
|
|
if (self.active in [self.Active.ENABLED, self.Active.ENABLED_OLD]) and (iterations <= 0):
|
|
raise ValueError("key iterations cannot be less than zero")
|
|
self.iterations = iterations
|
|
self.salt = salt
|
|
self.af = af
|
|
|
|
|
|
TmpHeaderVersion1 = namedtuple(
|
|
"TmpHeaderVersion1",
|
|
(
|
|
"magic",
|
|
"version",
|
|
"cipher",
|
|
"mode",
|
|
"hash",
|
|
"payload_offset",
|
|
"key_bytes",
|
|
"digest",
|
|
"salt",
|
|
"iterations",
|
|
"uuid",
|
|
"keys",
|
|
),
|
|
)
|
|
|
|
|
|
@dataclass(init=False)
|
|
class HeaderVersion1:
|
|
MAGIC = b"LUKS\xba\xbe"
|
|
VERSION = 0x0001
|
|
|
|
class Cipher(StrEnum):
|
|
AES = auto()
|
|
TWOFISH = auto()
|
|
SERPENT = auto()
|
|
|
|
class Mode(StrEnum):
|
|
CBC_ESSIV_SHA256 = "cbc-essiv:sha256"
|
|
CBC_PLAIN = "cbc-plain"
|
|
CBC_PLAIN64 = "cbc-plain64"
|
|
XTS_PLAIN = "xts-plain"
|
|
XTS_PLAIN64 = "xts-plain64"
|
|
|
|
class Hash(StrEnum):
|
|
RIPEMD160 = auto()
|
|
SHA1 = auto()
|
|
SHA256 = auto()
|
|
SHA512 = auto()
|
|
WHIRLPOOL = auto()
|
|
|
|
class KeySize(IntEnum):
|
|
SIZE_128 = 128
|
|
SIZE_256 = 256
|
|
SIZE_512 = 512
|
|
|
|
magic: bytes
|
|
version: int
|
|
cipher: Cipher
|
|
mode: Mode
|
|
hash: Hash
|
|
payload: bytes
|
|
key_size: KeySize
|
|
digest: bytes
|
|
salt: bytes
|
|
iterations: int
|
|
uuid: str
|
|
keys: List[KeyVersion1]
|
|
|
|
def __init__(self, magic, version, cipher, mode, hash, payload, key_size, digest, salt, iterations, uuid, keys):
|
|
if magic != self.MAGIC:
|
|
raise ValueError("invalid magic bytes")
|
|
self.magic = magic
|
|
if version != self.VERSION:
|
|
raise ValueError("invalid version")
|
|
self.version = version
|
|
if isinstance(cipher, bytes):
|
|
try:
|
|
cipher = bytes_to_str(cipher)
|
|
self.cipher = self.Cipher(cipher)
|
|
except UnicodeDecodeError as e:
|
|
raise ValueError("cannot decode cipher") from e
|
|
except ValueError as e:
|
|
raise ValueError("invalid cipher value") from e
|
|
if isinstance(mode, bytes):
|
|
try:
|
|
mode = bytes_to_str(mode)
|
|
self.mode = self.Mode(mode)
|
|
except UnicodeDecodeError as e:
|
|
raise ValueError("cannot decode mode") from e
|
|
except ValueError as e:
|
|
raise ValueError("invalid mode value") from e
|
|
if isinstance(hash, bytes):
|
|
try:
|
|
hash = bytes_to_str(hash)
|
|
self.hash = self.Hash(hash)
|
|
except UnicodeDecodeError as e:
|
|
raise ValueError("cannot decode hash") from e
|
|
except ValueError as e:
|
|
raise ValueError("invalid hash value") from e
|
|
self.payload = payload
|
|
try:
|
|
self.key_size = self.KeySize(key_size)
|
|
except ValueError as e:
|
|
raise ValueError("invalid key size provided") from e
|
|
self.digest = digest
|
|
self.salt = salt
|
|
try:
|
|
iterations = int(iterations)
|
|
except ValueError as e:
|
|
raise ValueError("iterations is not a number") from e
|
|
if iterations <= 0:
|
|
raise ValueError("iterations cannot be less or equal to zero")
|
|
self.iterations = iterations
|
|
if isinstance(uuid, bytes):
|
|
try:
|
|
uuid = bytes_to_str(uuid)
|
|
except UnicodeDecodeError as e:
|
|
raise ValueError("cannot decode UUID") from e
|
|
self.uuid = uuid
|
|
if all(isinstance(key, tuple) for key in keys):
|
|
keys = [KeyVersion1(*key) for key in keys]
|
|
elif all(isinstance(key, dict) for key in keys):
|
|
keys = [KeyVersion1(**key) for key in keys]
|
|
if any(not isinstance(key, KeyVersion1) for key in keys):
|
|
raise ValueError("not a key object provided")
|
|
self.keys = keys
|
|
|
|
|
|
def extract_version1(file):
|
|
# consts
|
|
KEYS_COUNT = 8
|
|
PADDING_LENGTH = 432
|
|
PAYLOAD_SIZE = 512 # sizeof (u32) * 128
|
|
|
|
# prepare structs
|
|
key_struct = Struct(">LL32sLL")
|
|
header_struct = Struct(
|
|
">6sH32s32s32sLL20s32sL40s" + str(key_struct.size * KEYS_COUNT) + "s" + str(PADDING_LENGTH) + "x"
|
|
)
|
|
|
|
# read header
|
|
header = file.read(header_struct.size)
|
|
if len(header) < header_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
# convert bytes into temporary header
|
|
header = header_struct.unpack(header)
|
|
header = TmpHeaderVersion1(*header)
|
|
|
|
# convert bytes into temporary keys
|
|
tmp_keys = [TmpKeyVersion1(*key) for key in key_struct.iter_unpack(header.keys)]
|
|
|
|
# read keys' af
|
|
keys = []
|
|
for key in tmp_keys:
|
|
file.seek(key.material_offset * SECTOR_SIZE, SEEK_SET)
|
|
af = file.read(header.key_bytes * key.stripes)
|
|
if len(af) < (header.key_bytes * key.stripes):
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
key = KeyVersion1(key.active, key.iterations, key.salt, af)
|
|
keys.append(key)
|
|
|
|
# read payload
|
|
file.seek(header.payload_offset * SECTOR_SIZE, SEEK_SET)
|
|
payload = file.read(PAYLOAD_SIZE)
|
|
if len(payload) < PAYLOAD_SIZE:
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
# convert into header
|
|
header = HeaderVersion1(
|
|
header.magic,
|
|
header.version,
|
|
header.cipher,
|
|
header.mode,
|
|
header.hash,
|
|
payload,
|
|
header.key_bytes * 8,
|
|
header.digest,
|
|
header.salt,
|
|
header.iterations,
|
|
header.uuid,
|
|
keys,
|
|
)
|
|
|
|
# check for any active key
|
|
for key in header.keys:
|
|
if key.active not in [KeyVersion1.Active.ENABLED, KeyVersion1.Active.ENABLED_OLD]:
|
|
continue
|
|
|
|
hash = SIGNATURE + "$".join(
|
|
map(
|
|
str,
|
|
[
|
|
header.version,
|
|
header.hash,
|
|
header.cipher,
|
|
header.mode,
|
|
int(header.key_size),
|
|
key.iterations,
|
|
key.salt.hex(),
|
|
key.af.hex(),
|
|
header.payload.hex(),
|
|
],
|
|
)
|
|
)
|
|
print(hash)
|
|
break
|
|
else:
|
|
# all keys are disabled
|
|
raise ValueError("all keys are disabled")
|
|
|
|
|
|
# main
|
|
|
|
|
|
if __name__ == "__main__":
|
|
# prepare parser and parse args
|
|
parser = ArgumentParser(description="luks2hashcat extraction tool")
|
|
parser.add_argument("path", type=str, help="path to LUKS container")
|
|
args = parser.parse_args()
|
|
|
|
# prepare struct
|
|
header_struct = Struct(">6sH")
|
|
|
|
try:
|
|
with open(args.path, "rb") as file:
|
|
# read pre header
|
|
header = file.read(header_struct.size)
|
|
if len(header) < header_struct.size:
|
|
parser.error("file contains less data than needed")
|
|
|
|
# convert bytes into temporary pre header
|
|
header = header_struct.unpack(header)
|
|
header = TmpHeaderPre(*header)
|
|
|
|
# check magic bytes
|
|
magic_bytes = {
|
|
HeaderVersion1.MAGIC,
|
|
}
|
|
if header.magic not in magic_bytes:
|
|
parser.error("improper magic bytes")
|
|
|
|
# back to start of the file
|
|
file.seek(0, SEEK_SET)
|
|
|
|
# extract with proper function
|
|
try:
|
|
mapping = {
|
|
HeaderVersion1.VERSION: extract_version1,
|
|
}
|
|
extract = mapping[header.version]
|
|
extract(file)
|
|
except KeyError as e:
|
|
raise ValueError("unsupported version") from e
|
|
except IOError as e:
|
|
parser.error(e.strerror.lower())
|
|
except ValueError as e:
|
|
parser.error(str(e))
|