mirror of
https://github.com/hashcat/hashcat.git
synced 2024-11-18 06:08:16 +00:00
336 lines
8.7 KiB
Python
Executable File
336 lines
8.7 KiB
Python
Executable File
#!/usr/bin/env python3
|
|
|
|
# Author......: See docs/credits.txt
|
|
# License.....: MIT
|
|
# Target......: Exodus wallet extractor
|
|
# Example.....: exodus2hashcat.py <path to exodus seed seco file>
|
|
|
|
from __future__ import annotations
|
|
|
|
import hashlib
|
|
import struct
|
|
from argparse import ArgumentParser
|
|
from base64 import b64encode
|
|
from dataclasses import dataclass
|
|
from enum import Enum
|
|
from io import BytesIO
|
|
from struct import Struct
|
|
|
|
|
|
# consts
|
|
|
|
|
|
SIGNATURE = "EXODUS"
|
|
|
|
HEADER_MAGIC = b"SECO"
|
|
HEADER_VERSION = 0
|
|
HEADER_VERSION_TAG = b"seco-v0-scrypt-aes"
|
|
HEADER_SIZE = 224
|
|
|
|
CHECKSUM_SIZE = 256 // 8
|
|
|
|
METADATA_SALT_SIZE = 32
|
|
METADATA_CIPHER_SIZE = 32
|
|
METADATA_BLOB_KEY_IV_SIZE = 12
|
|
METADATA_BLOB_KEY_AUTH_TAG_SIZE = 16
|
|
METADATA_BLOB_KEY_KEY_SIZE = 32
|
|
METADATA_BLOB_IV_SIZE = 12
|
|
METADATA_BLOB_AUTH_TAG_SIZE = 16
|
|
METADATA_SIZE = 256
|
|
|
|
|
|
# structs
|
|
|
|
|
|
@dataclass
|
|
class File:
|
|
header: Header
|
|
checksum: bytes
|
|
metadata: Metadata
|
|
blob: bytes
|
|
|
|
|
|
@dataclass
|
|
class Header:
|
|
magic: bytes
|
|
version: int
|
|
version_tag: bytes
|
|
app_name: bytes
|
|
app_version: bytes
|
|
|
|
|
|
@dataclass(init=False)
|
|
class Metadata:
|
|
class Cipher(str, Enum):
|
|
AES_256_GCM = "aes-256-gcm"
|
|
|
|
@dataclass
|
|
class BlobKey:
|
|
iv: bytes
|
|
auth_tag: bytes
|
|
key: bytes
|
|
|
|
@dataclass
|
|
class Blob:
|
|
iv: bytes
|
|
auth_tag: bytes
|
|
|
|
salt: bytes
|
|
n: int
|
|
r: int
|
|
p: int
|
|
cipher: Cipher
|
|
blob_key: BlobKey
|
|
blob: Blob
|
|
|
|
def __init__(self, salt, n, r, p, cipher, blob_key, blob):
|
|
self.salt = salt
|
|
self.n = n
|
|
self.r = r
|
|
self.p = p
|
|
if isinstance(cipher, bytes):
|
|
cipher = cipher.rstrip(b"\x00")
|
|
cipher = cipher.decode()
|
|
cipher = self.Cipher(cipher)
|
|
self.cipher = cipher
|
|
self.blob_key = blob_key
|
|
self.blob = blob
|
|
|
|
|
|
# header
|
|
|
|
|
|
def read_header(file):
|
|
# prepare structs
|
|
partial_header_struct = Struct(">4sL4x")
|
|
byte_struct = Struct(">B")
|
|
|
|
# read whole header space
|
|
file = BytesIO(file.read(HEADER_SIZE))
|
|
|
|
# read partial header
|
|
partial_header = file.read(partial_header_struct.size)
|
|
if len(partial_header) < partial_header_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
partial_header = partial_header_struct.unpack(partial_header)
|
|
|
|
# read header version tag
|
|
header_version_tag_length = file.read(byte_struct.size)
|
|
if len(header_version_tag_length) < byte_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
(header_version_tag_length,) = byte_struct.unpack(header_version_tag_length)
|
|
header_version_tag = file.read(header_version_tag_length)
|
|
if len(header_version_tag) < header_version_tag_length:
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
# read header app name
|
|
header_app_name_length = file.read(byte_struct.size)
|
|
if len(header_app_name_length) < byte_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
(header_app_name_length,) = byte_struct.unpack(header_app_name_length)
|
|
header_app_name = file.read(header_app_name_length)
|
|
if len(header_app_name) < header_app_name_length:
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
# read header app version
|
|
header_app_version_length = file.read(byte_struct.size)
|
|
if len(header_app_version_length) < byte_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
(header_app_version_length,) = byte_struct.unpack(header_app_version_length)
|
|
header_app_version = file.read(header_app_version_length)
|
|
if len(header_app_version) < header_app_version_length:
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
# make header
|
|
header = Header(*partial_header, header_version_tag, header_app_name, header_app_version)
|
|
|
|
return header
|
|
|
|
|
|
# checksum
|
|
|
|
|
|
def read_checksum(file):
|
|
# read checksum
|
|
checksum = file.read(CHECKSUM_SIZE)
|
|
if len(checksum) < CHECKSUM_SIZE:
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
return checksum
|
|
|
|
|
|
def validate_checksum(checksum, metadata, blob):
|
|
# prepare hash
|
|
sha256 = hashlib.sha256()
|
|
|
|
# update with metadata
|
|
sha256.update(metadata.salt)
|
|
sha256.update(struct.pack(">LLL", metadata.n, metadata.r, metadata.p))
|
|
sha256.update(metadata.cipher.value.encode().ljust(METADATA_CIPHER_SIZE, b"\x00"))
|
|
|
|
# update with metadata blob key
|
|
sha256.update(metadata.blob_key.iv)
|
|
sha256.update(metadata.blob_key.auth_tag)
|
|
sha256.update(metadata.blob_key.key)
|
|
|
|
# update with metadata metadata.blob
|
|
sha256.update(metadata.blob.iv)
|
|
sha256.update(metadata.blob.auth_tag)
|
|
|
|
# update with metadata padding
|
|
metadata_size = (
|
|
METADATA_SALT_SIZE
|
|
+ struct.calcsize(">LLL")
|
|
+ METADATA_CIPHER_SIZE
|
|
+ METADATA_BLOB_KEY_IV_SIZE
|
|
+ METADATA_BLOB_KEY_AUTH_TAG_SIZE
|
|
+ METADATA_BLOB_KEY_KEY_SIZE
|
|
+ METADATA_BLOB_IV_SIZE
|
|
+ METADATA_BLOB_AUTH_TAG_SIZE
|
|
)
|
|
sha256.update(bytes(METADATA_SIZE - metadata_size))
|
|
|
|
# update with blob
|
|
sha256.update(struct.pack(">L", len(blob)))
|
|
sha256.update(blob)
|
|
|
|
# make digest
|
|
digest = sha256.digest()
|
|
|
|
# compare
|
|
if checksum != digest:
|
|
raise ValueError("file corrupted - checksum validation failed")
|
|
|
|
|
|
# metadata
|
|
|
|
|
|
def read_metadata(file):
|
|
# prepare structs
|
|
partial_metadata_struct = Struct(">" + str(METADATA_SALT_SIZE) + "sLLL" + str(METADATA_CIPHER_SIZE) + "s")
|
|
blob_key_struct = Struct(
|
|
">"
|
|
+ str(METADATA_BLOB_KEY_IV_SIZE)
|
|
+ "s"
|
|
+ str(METADATA_BLOB_KEY_AUTH_TAG_SIZE)
|
|
+ "s"
|
|
+ str(METADATA_BLOB_KEY_KEY_SIZE)
|
|
+ "s"
|
|
)
|
|
blob_struct = Struct(">" + str(METADATA_BLOB_IV_SIZE) + "s" + str(METADATA_BLOB_AUTH_TAG_SIZE) + "s")
|
|
|
|
# read whole metadata space
|
|
file = BytesIO(file.read(METADATA_SIZE))
|
|
|
|
# read partial metadata
|
|
partial_metadata = file.read(partial_metadata_struct.size)
|
|
if len(partial_metadata) < partial_metadata_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
partial_metadata = partial_metadata_struct.unpack(partial_metadata)
|
|
|
|
# read blob key
|
|
blob_key = file.read(blob_key_struct.size)
|
|
if len(blob_key) < blob_key_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
blob_key = blob_key_struct.unpack(blob_key)
|
|
blob_key = Metadata.BlobKey(*blob_key)
|
|
|
|
# read blob
|
|
blob = file.read(blob_struct.size)
|
|
if len(blob) < blob_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
blob = blob_struct.unpack(blob)
|
|
blob = Metadata.Blob(*blob)
|
|
|
|
# make metadata
|
|
metadata = Metadata(*partial_metadata, blob_key, blob)
|
|
|
|
return metadata
|
|
|
|
|
|
# blob
|
|
|
|
|
|
def read_blob(file):
|
|
# prepare structs
|
|
size_struct = Struct(">L")
|
|
|
|
# read size
|
|
size = file.read(size_struct.size)
|
|
if len(size) < size_struct.size:
|
|
raise ValueError("file contains less data than needed")
|
|
(size,) = size_struct.unpack(size)
|
|
|
|
# read blob
|
|
blob = file.read(size)
|
|
if len(blob) < size:
|
|
raise ValueError("file contains less data than needed")
|
|
|
|
return blob
|
|
|
|
|
|
# file
|
|
|
|
|
|
def read_file(file):
|
|
# read header
|
|
header = read_header(file)
|
|
|
|
# validate header values
|
|
if header.magic != HEADER_MAGIC:
|
|
raise ValueError("not a SECO file")
|
|
if (header.version != HEADER_VERSION) or (header.version_tag != HEADER_VERSION_TAG):
|
|
raise ValueError("unsupported version")
|
|
|
|
# read checksum
|
|
checksum = read_checksum(file)
|
|
|
|
# read metadata
|
|
metadata = read_metadata(file)
|
|
|
|
# read blob
|
|
blob = read_blob(file)
|
|
|
|
# validate digest
|
|
validate_checksum(checksum, metadata, blob)
|
|
|
|
# make file
|
|
file = File(header, checksum, metadata, blob)
|
|
|
|
return file
|
|
|
|
|
|
# main
|
|
|
|
|
|
if __name__ == "__main__":
|
|
# prepare parser and parse args
|
|
parser = ArgumentParser(description="exodus2hashcat extraction tool")
|
|
parser.add_argument("path", type=str, help="path to SECO file")
|
|
args = parser.parse_args()
|
|
|
|
try:
|
|
with open(args.path, "rb") as file:
|
|
file = read_file(file)
|
|
|
|
hash = ":".join(
|
|
map(
|
|
str,
|
|
[
|
|
SIGNATURE,
|
|
file.metadata.n,
|
|
file.metadata.r,
|
|
file.metadata.p,
|
|
b64encode(file.metadata.salt).decode(),
|
|
b64encode(file.metadata.blob_key.iv).decode(),
|
|
b64encode(file.metadata.blob_key.key).decode(),
|
|
b64encode(file.metadata.blob_key.auth_tag).decode(),
|
|
],
|
|
)
|
|
)
|
|
print(hash)
|
|
except IOError as e:
|
|
parser.error(e.strerror.lower())
|
|
except ValueError as e:
|
|
parser.error(str(e))
|