lbmk/util/me7_update_parser/me7_update_parser.py

617 lines
21 KiB
Python
Raw Normal View History

#!/usr/bin/env python3
"""ME7 Update binary parser."""
# Copyright (C) 2020 Tom Hiller <thrilleratplay@gmail.com>
# Copyright (C) 2016-2018 Nicola Corna <nicola@corna.info>
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# Based on the amazing me_cleaner, https://github.com/corna/me_cleaner, parses
# the required signed partition from an ME update file to generate a valid
# flashable ME binary.
#
# This was written for Heads ROM, https://github.com/osresearch/heads
# to allow continuous integration reproducible builds for Lenovo xx20 models
# (X220, T420, T520, etc).
#
# A full model list can be found:
# https://download.lenovo.com/ibmdl/pub/pc/pccbbs/mobiles/83rf46ww.txt
from struct import pack, unpack
from typing import List
import argparse
import sys
import hashlib
import binascii
import os.path
#############################################################################
FTPR_END = 0x76000
MINIFIED_FTPR_OFFSET = 0x400 # offset start of Factory Partition (FTPR)
ORIG_FTPR_OFFSET = 0xCC000
PARTITION_HEADER_OFFSET = 0x30 # size of partition header
DEFAULT_OUTPUT_FILE_NAME = "flashregion_2_intel_me.bin"
#############################################################################
class EntryFlags:
"""EntryFlag bitmap values."""
ExclBlockUse = 8192
WOPDisable = 4096
Logical = 2048
Execute = 1024
Write = 512
Read = 256
DirectAccess = 128
Type = 64
def generateHeader() -> bytes:
"""Generate Header."""
ROM_BYPASS_INSTR_0 = binascii.unhexlify("2020800F")
ROM_BYPASS_INSTR_1 = binascii.unhexlify("40000010")
ROM_BYPASS_INSTR_2 = pack("<I", 0)
ROM_BYPASS_INSTR_3 = pack("<I", 0)
# $FPT Partition table header
HEADER_TAG = "$FPT".encode()
HEADER_NUM_PARTITIONS = pack("<I", 1)
HEADER_VERSION = b"\x20" # version 2.0
HEADER_ENTRY_TYPE = b"\x10"
HEADER_LENGTH = b"\x30"
HEADER_CHECKSUM = pack("<B", 0)
HEADER_FLASH_CYCLE_LIFE = pack("<H", 7)
HEADER_FLASH_CYCLE_LIMIT = pack("<H", 100)
HEADER_UMA_SIZE = pack("<H", 32)
HEADER_FLAGS = binascii.unhexlify("000000FCFFFF")
HEADER_FITMAJOR = pack("<H", 0)
HEADER_FITMINOR = pack("<H", 0)
HEADER_FITHOTFIX = pack("<H", 0)
HEADER_FITBUILD = pack("<H", 0)
FTPR_header_layout = bytearray(
ROM_BYPASS_INSTR_0
+ ROM_BYPASS_INSTR_1
+ ROM_BYPASS_INSTR_2
+ ROM_BYPASS_INSTR_3
+ HEADER_TAG
+ HEADER_NUM_PARTITIONS
+ HEADER_VERSION
+ HEADER_ENTRY_TYPE
+ HEADER_LENGTH
+ HEADER_CHECKSUM
+ HEADER_FLASH_CYCLE_LIFE
+ HEADER_FLASH_CYCLE_LIMIT
+ HEADER_UMA_SIZE
+ HEADER_FLAGS
+ HEADER_FITMAJOR
+ HEADER_FITMINOR
+ HEADER_FITHOTFIX
+ HEADER_FITBUILD
)
# Update checksum
FTPR_header_layout[27] = (0x100 - sum(FTPR_header_layout) & 0xFF) & 0xFF
return FTPR_header_layout
def generateFtpPartition() -> bytes:
"""Partition table entry."""
ENTRY_NAME = binascii.unhexlify("46545052")
ENTRY_OWNER = binascii.unhexlify("FFFFFFFF") # "None"
ENTRY_OFFSET = binascii.unhexlify("00040000")
ENTRY_LENGTH = binascii.unhexlify("00600700")
ENTRY_START_TOKENS = pack("<I", 1)
ENTRY_MAX_TOKENS = pack("<I", 1)
ENTRY_SCRATCH_SECTORS = pack("<I", 0)
ENTRY_FLAGS = pack(
"<I",
(
EntryFlags.ExclBlockUse
+ EntryFlags.Execute
+ EntryFlags.Write
+ EntryFlags.Read
+ EntryFlags.DirectAccess
),
)
partition = (
ENTRY_NAME
+ ENTRY_OWNER
+ ENTRY_OFFSET
+ ENTRY_LENGTH
+ ENTRY_START_TOKENS
+ ENTRY_MAX_TOKENS
+ ENTRY_SCRATCH_SECTORS
+ ENTRY_FLAGS
)
# offset of the partition - length of partition entry -length of header
pad_len = MINIFIED_FTPR_OFFSET - (len(partition) + PARTITION_HEADER_OFFSET)
padding = b""
for i in range(0, pad_len):
padding += b"\xFF"
return partition + padding
############################################################################
class OutOfRegionException(Exception):
"""Out of Region Exception."""
pass
class clean_ftpr:
"""Clean Factory Parition (FTPR)."""
UNREMOVABLE_MODULES = ("ROMP", "BUP")
COMPRESSION_TYPE_NAME = ("uncomp.", "Huffman", "LZMA")
def __init__(self, ftpr: bytes):
"""Init."""
self.orig_ftpr = ftpr
self.ftpr = ftpr
self.mod_headers: List[bytes] = []
self.check_and_clean_ftpr()
#####################################################################
# tilities
#####################################################################
def slice(self, offset: int, size: int) -> bytes:
"""Copy data of a given size from FTPR starting from offset."""
offset_end = offset + size
return self.ftpr[offset:offset_end]
def unpack_next_int(self, offset: int) -> int:
"""Sugar syntax for unpacking a little-endian UINT at offset."""
return self.unpack_val(self.slice(offset, 4))
def unpack_val(self, data: bytes) -> int:
"""Sugar syntax for unpacking a little-endian unsigned integer."""
return unpack("<I", data)[0]
def bytes_to_ascii(self, data: bytes) -> str:
"""Decode bytes into ASCII."""
return data.rstrip(b"\x00").decode("ascii")
def clear_ftpr_data(self, start: int, end: int) -> None:
"""Replace values in range with 0xFF."""
empty_data = bytes()
for i in range(0, end - start):
empty_data += b"\xff"
self.write_ftpr_data(start, empty_data)
def write_ftpr_data(self, start: int, data: bytes) -> None:
"""Replace data in FTPR starting at a given offset."""
end = len(data) + start
new_partition = self.ftpr[:start]
new_partition += data
if end != FTPR_END:
new_partition += self.ftpr[end:]
self.ftpr = new_partition
######################################################################
# FTPR cleanig/checking functions
######################################################################
def get_chunks_offsets(self, llut: bytes):
"""Calculate Chunk offsets from LLUT."""
chunk_count = self.unpack_val(llut[0x04:0x08])
huffman_stream_end = sum(unpack("<II", llut[0x10:0x18]))
nonzero_offsets = [huffman_stream_end]
offsets = []
for i in range(0, chunk_count):
llut_start = 0x40 + (i * 4)
llut_end = 0x44 + (i * 4)
chunk = llut[llut_start:llut_end]
offset = 0
if chunk[3] != 0x80:
offset = self.unpack_val(chunk[0:3] + b"\x00")
offsets.append([offset, 0])
if offset != 0:
nonzero_offsets.append(offset)
nonzero_offsets.sort()
for i in offsets:
if i[0] != 0:
i[1] = nonzero_offsets[nonzero_offsets.index(i[0]) + 1]
return offsets
def relocate_partition(self) -> int:
"""Relocate partition."""
new_offset = MINIFIED_FTPR_OFFSET
name = self.bytes_to_ascii(self.slice(PARTITION_HEADER_OFFSET, 4))
old_offset, partition_size = unpack(
"<II", self.slice(PARTITION_HEADER_OFFSET + 0x8, 0x8)
)
llut_start = 0
for mod_header in self.mod_headers:
if (self.unpack_val(mod_header[0x50:0x54]) >> 4) & 7 == 0x01:
llut_start = self.unpack_val(mod_header[0x38:0x3C])
llut_start += old_offset
break
if self.mod_headers and llut_start != 0:
# Bytes 0x9:0xb of the LLUT (bytes 0x1:0x3 of the AddrBase) are
# added to the SpiBase (bytes 0xc:0x10 of the LLUT) to compute the
# final start of the LLUT. Since AddrBase is not modifiable, we can
# act only on SpiBase and here we compute the minimum allowed
# new_offset.
llut_start_corr = unpack("<H", self.slice(llut_start + 0x9, 2))[0]
new_offset = max(
new_offset, llut_start_corr - llut_start - 0x40 + old_offset
)
new_offset = ((new_offset + 0x1F) // 0x20) * 0x20
offset_diff = new_offset - old_offset
print(
"Relocating {} from {:#x} - {:#x} to {:#x} - {:#x}...".format(
name,
old_offset,
old_offset + partition_size,
new_offset,
new_offset + partition_size,
)
)
print(" Adjusting FPT entry...")
self.write_ftpr_data(
PARTITION_HEADER_OFFSET + 0x08,
pack("<I", new_offset),
)
if self.mod_headers:
if llut_start != 0:
if self.slice(llut_start, 4) == b"LLUT":
print(" Adjusting LUT start offset...")
llut_offset = pack(
"<I", llut_start + offset_diff + 0x40 - llut_start_corr
)
self.write_ftpr_data(llut_start + 0x0C, llut_offset)
print(" Adjusting Huffman start offset...")
old_huff_offset = self.unpack_next_int(llut_start + 0x14)
ftpr_offset_diff = MINIFIED_FTPR_OFFSET - ORIG_FTPR_OFFSET
self.write_ftpr_data(
llut_start + 0x14,
pack("<I", old_huff_offset + ftpr_offset_diff),
)
print(" Adjusting chunks offsets...")
chunk_count = self.unpack_next_int(llut_start + 0x4)
offset = llut_start + 0x40
offset_end = chunk_count * 4
chunks = bytearray(self.slice(offset, offset_end))
for i in range(0, offset_end, 4):
i_plus_3 = i + 3
if chunks[i_plus_3] != 0x80:
chunks[i:i_plus_3] = pack(
"<I",
self.unpack_val(chunks[i:i_plus_3] + b"\x00")
+ (MINIFIED_FTPR_OFFSET - ORIG_FTPR_OFFSET),
)[0:3]
self.write_ftpr_data(offset, bytes(chunks))
else:
sys.exit("Huffman modules present but no LLUT found!")
else:
print(" No Huffman modules found")
print(" Moving data...")
partition_size = min(partition_size, FTPR_END - old_offset)
if (
old_offset + partition_size <= FTPR_END
and new_offset + partition_size <= FTPR_END
):
for i in range(0, partition_size, 4096):
block_length = min(partition_size - i, 4096)
block = self.slice(old_offset + i, block_length)
self.clear_ftpr_data(old_offset + i, len(block))
self.write_ftpr_data(new_offset + i, block)
else:
raise OutOfRegionException()
return new_offset
def remove_modules(self) -> int:
"""Remove modules."""
unremovable_huff_chunks = []
chunks_offsets = []
base = 0
chunk_size = 0
end_addr = 0
for mod_header in self.mod_headers:
name = self.bytes_to_ascii(mod_header[0x04:0x14])
offset = self.unpack_val(mod_header[0x38:0x3C])
size = self.unpack_val(mod_header[0x40:0x44])
flags = self.unpack_val(mod_header[0x50:0x54])
comp_type = (flags >> 4) & 7
comp_type_name = self.COMPRESSION_TYPE_NAME[comp_type]
print(" {:<16} ({:<7}, ".format(name, comp_type_name), end="")
# If compresion type uncompressed or LZMA
if comp_type == 0x00 or comp_type == 0x02:
offset_end = offset + size
range_msg = "0x{:06x} - 0x{:06x} ): "
print(range_msg.format(offset, offset_end), end="")
if name in self.UNREMOVABLE_MODULES:
end_addr = max(end_addr, offset + size)
print("NOT removed, essential")
else:
offset_end = min(offset + size, FTPR_END)
self.clear_ftpr_data(offset, offset_end)
print("removed")
# Else if compression type huffman
elif comp_type == 0x01:
if not chunks_offsets:
# Check if Local Look Up Table (LLUT) is present
if self.slice(offset, 4) == b"LLUT":
llut = self.slice(offset, 0x40)
chunk_count = self.unpack_val(llut[0x4:0x8])
base = self.unpack_val(llut[0x8:0xC]) + 0x10000000
chunk_size = self.unpack_val(llut[0x30:0x34])
llut = self.slice(offset, (chunk_count * 4) + 0x40)
# calculate offsets of chunks from LLUT
chunks_offsets = self.get_chunks_offsets(llut)
else:
no_llut_msg = "Huffman modules found,"
no_llut_msg += "but LLUT is not present."
sys.exit(no_llut_msg)
module_base = self.unpack_val(mod_header[0x34:0x38])
module_size = self.unpack_val(mod_header[0x3C:0x40])
first_chunk_num = (module_base - base) // chunk_size
last_chunk_num = first_chunk_num + module_size // chunk_size
huff_size = 0
chunk_length = last_chunk_num + 1
for chunk in chunks_offsets[first_chunk_num:chunk_length]:
huff_size += chunk[1] - chunk[0]
size_in_kiB = "~" + str(int(round(huff_size / 1024))) + " KiB"
print(
"fragmented data, {:<9}): ".format(size_in_kiB),
end="",
)
# Check if module is in the unremovable list
if name in self.UNREMOVABLE_MODULES:
print("NOT removed, essential")
# add to list of unremovable chunks
for x in chunks_offsets[first_chunk_num:chunk_length]:
if x[0] != 0:
unremovable_huff_chunks.append(x)
else:
print("removed")
# Else unknown compression type
else:
unkwn_comp_msg = " 0x{:06x} - 0x{:06x}): "
unkwn_comp_msg += "unknown compression, skipping"
print(unkwn_comp_msg.format(offset, offset + size), end="")
if chunks_offsets:
removable_huff_chunks = []
for chunk in chunks_offsets:
# if chunk is not in a unremovable chunk, it must be removable
if all(
not (
unremovable_chk[0] <= chunk[0] < unremovable_chk[1]
or unremovable_chk[0] < chunk[1] <= unremovable_chk[1]
)
for unremovable_chk in unremovable_huff_chunks
):
removable_huff_chunks.append(chunk)
for removable_chunk in removable_huff_chunks:
if removable_chunk[1] > removable_chunk[0]:
chunk_start = removable_chunk[0] - ORIG_FTPR_OFFSET
chunk_end = removable_chunk[1] - ORIG_FTPR_OFFSET
self.clear_ftpr_data(chunk_start, chunk_end)
end_addr = max(
end_addr, max(unremovable_huff_chunks, key=lambda x: x[1])[1]
)
end_addr -= ORIG_FTPR_OFFSET
return end_addr
def find_mod_header_size(self) -> None:
"""Find module header size."""
self.mod_header_size = 0
data = self.slice(0x290, 0x84)
# check header size
if data[0x0:0x4] == b"$MME":
if data[0x60:0x64] == b"$MME" or self.num_modules == 1:
self.mod_header_size = 0x60
elif data[0x80:0x84] == b"$MME":
self.mod_header_size = 0x80
def find_mod_headers(self) -> None:
"""Find module headers."""
data = self.slice(0x290, self.mod_header_size * self.num_modules)
for i in range(0, self.num_modules):
header_start = i * self.mod_header_size
header_end = (i + 1) * self.mod_header_size
self.mod_headers.append(data[header_start:header_end])
def resize_partition(self, end_addr: int) -> None:
"""Resize partition."""
spared_blocks = 4
if end_addr > 0:
end_addr = (end_addr // 0x1000 + 1) * 0x1000
end_addr += spared_blocks * 0x1000
# partition header not added yet
# remove trailing data the same size as the header.
end_addr -= MINIFIED_FTPR_OFFSET
me_size_msg = "The ME minimum size should be {0} "
me_size_msg += "bytes ({0:#x} bytes)"
print(me_size_msg.format(end_addr))
print("Truncating file at {:#x}...".format(end_addr))
self.ftpr = self.ftpr[:end_addr]
def check_and_clean_ftpr(self) -> None:
"""Check and clean FTPR (factory partition)."""
self.num_modules = self.unpack_next_int(0x20)
self.find_mod_header_size()
if self.mod_header_size != 0:
self.find_mod_headers()
# ensure all of the headers begin with b'$MME'
if all(hdr.startswith(b"$MME") for hdr in self.mod_headers):
end_addr = self.remove_modules()
new_offset = self.relocate_partition()
end_addr += new_offset
self.resize_partition(end_addr)
# flip bit
# XXX: I have no idea why this works and passes RSA signiture
self.write_ftpr_data(0x39, b"\x00")
else:
sys.exit(
"Found less modules than expected in the FTPR "
"partition; skipping modules removal and exiting."
)
else:
sys.exit(
"Can't find the module header size; skipping modules"
"removal and exiting."
)
##########################################################################
def check_partition_signature(f, offset) -> bool:
"""check_partition_signature copied/shamelessly stolen from me_cleaner."""
f.seek(offset)
header = f.read(0x80)
modulus = int(binascii.hexlify(f.read(0x100)[::-1]), 16)
public_exponent = unpack("<I", f.read(4))[0]
signature = int(binascii.hexlify(f.read(0x100)[::-1]), 16)
header_len = unpack("<I", header[0x4:0x8])[0] * 4
manifest_len = unpack("<I", header[0x18:0x1C])[0] * 4
f.seek(offset + header_len)
sha256 = hashlib.sha256()
sha256.update(header)
tmp = f.read(manifest_len - header_len)
sha256.update(tmp)
decrypted_sig = pow(signature, public_exponent, modulus)
return "{:#x}".format(decrypted_sig).endswith(sha256.hexdigest()) # FIXME
##########################################################################
def generate_me_blob(input_file: str, output_file: str) -> None:
"""Generate ME blob."""
print("Starting ME 7.x Update parser.")
orig_f = open(input_file, "rb")
cleaned_ftpr = clean_ftpr(orig_f.read(FTPR_END))
orig_f.close()
fo = open(output_file, "wb")
fo.write(generateHeader())
fo.write(generateFtpPartition())
fo.write(cleaned_ftpr.ftpr)
fo.close()
def verify_output(output_file: str) -> None:
"""Verify Generated ME file."""
file_verifiy = open(output_file, "rb")
if check_partition_signature(file_verifiy, MINIFIED_FTPR_OFFSET):
print(output_file + " is VALID")
file_verifiy.close()
else:
print(output_file + " is INVALID!!")
file_verifiy.close()
sys.exit("The FTPR partition signature is not valid.")
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description="Tool to remove as much code "
"as possible from Intel ME/TXE 7.x firmware "
"update and create paratition for a flashable ME parition."
)
parser.add_argument("file", help="ME/TXE image or full dump")
parser.add_argument(
"-O",
"--output",
metavar="output_file",
help="save "
"save file name other than the default '" + DEFAULT_OUTPUT_FILE_NAME + "'",
)
args = parser.parse_args()
output_file_name = DEFAULT_OUTPUT_FILE_NAME if not args.output else args.output
# Check if output file exists, ask to overwrite or exit
if os.path.isfile(output_file_name):
input_msg = output_file_name
input_msg += " exists. Do you want to overwrite? [y/N]: "
if not str(input(input_msg)).lower().startswith("y"):
sys.exit("Not overwriting file. Exiting.")
generate_me_blob(args.file, output_file_name)
verify_output(output_file_name)