mirror of
https://gitlab.com/Nanolx/qwad.git
synced 2024-11-25 03:56:55 +01:00
358 lines
12 KiB
Python
358 lines
12 KiB
Python
from common import *
|
|
from title import *
|
|
|
|
class WOD: #WiiOpticalDisc
|
|
class discHeader(Struct):
|
|
__endian__ = Struct.BE
|
|
def __format__(self):
|
|
self.discId = Struct.string(1)
|
|
self.gameCode = Struct.string(2)
|
|
self.region = Struct.string(1)
|
|
self.makerCode = Struct.uint8[2]
|
|
self.h = Struct.uint8
|
|
self.version = Struct.uint8
|
|
self.audioStreaming = Struct.uint8
|
|
self.streamingBufSize = Struct.uint8
|
|
self.unused = Struct.uint8[14]
|
|
self.magic = Struct.uint32
|
|
self.title = Struct.string(64)
|
|
self.hashVerify = Struct.uint8
|
|
self.h3verify = Struct.uint8
|
|
def __str__(self):
|
|
ret = ''
|
|
ret += '%s [%s%s%s]\n' % (self.title, self.discId, self.gameCode, self.region)
|
|
if self.region == 'P':
|
|
ret += 'Region : PAL\n'
|
|
elif self.region == 'E':
|
|
ret += 'Region : NTSC\n'
|
|
elif self.region == 'J':
|
|
ret += 'Region : JPN\n'
|
|
ret += 'Version 0x%x Maker %i%i Audio streaming %x\n' % (self.version, self.makerCode[0], self.makerCode[1], self.audioStreaming)
|
|
ret += 'Hash verify flag 0x%x H3 verify flag : 0x%x\n' % (self.hashVerify, self.h3verify)
|
|
|
|
return ret
|
|
|
|
# Many many thanks to Wiipower
|
|
class Apploader(Struct):
|
|
__endian__ = Struct.BE
|
|
def __format__(self):
|
|
self.buildDate = Struct.string(16)
|
|
self.entryPoint = Struct.uint32
|
|
self.size = Struct.uint32
|
|
self.trailingSize = Struct.uint32
|
|
self.padding = Struct.uint8[4]
|
|
def __str__(self):
|
|
ret = ''
|
|
ret += 'Apploader built on %s\n' % self.buildDate
|
|
ret += 'Entry point 0x%x\n' % self.entryPoint
|
|
ret += 'Size %i (%i of them are trailing)\n' % (self.size, self.trailingSize)
|
|
|
|
return ret
|
|
|
|
def __str__(self):
|
|
ret = ''
|
|
ret += '%s\n' % self.discHdr
|
|
ret += 'Found %i partitions (table at 0x%x)\n' % (self.partitionCount, self.partsTableOffset)
|
|
ret += 'Found %i channels (table at 0x%x)\n' % (self.channelsCount, self.chansTableOffset)
|
|
ret += '\n'
|
|
ret += 'Partition %i opened (type 0x%x) at 0x%x\n' % (self.partitionOpen, self.partitionType, self.partitionOffset)
|
|
ret += 'Partition name : %s' % self.partitionHdr
|
|
ret += 'Partition key : %s\n' % hexdump(self.partitionKey)
|
|
ret += 'Partition IOS : IOS%i\n' % self.partitionIos
|
|
ret += 'Partition tmd : 0x%x (%x)\n' % (self.tmdOffset, self.tmdSize)
|
|
ret += 'Partition main.dol : 0x%x (%x)\n' % (self.dolOffset, self.dolSize)
|
|
ret += 'Partition FST : 0x%x (%x)\n' % (self.fstSize, self.fstOffset)
|
|
ret += '%s\n' % (self.appLdr)
|
|
|
|
return ret
|
|
|
|
def __init__(self, f):
|
|
self.f = f
|
|
self.fp = open(f, 'rb')
|
|
|
|
self.discHdr = self.discHeader().unpack(self.fp.read(0x400))
|
|
if self.discHdr.magic != 0x5D1C9EA3:
|
|
raise Exception('Wrong disc magic')
|
|
|
|
self.fp.seek(0x40000)
|
|
|
|
self.partitionCount = 1 + struct.unpack(">I", self.fp.read(4))[0]
|
|
self.partsTableOffset = struct.unpack(">I", self.fp.read(4))[0] << 2
|
|
|
|
self.channelsCount = struct.unpack(">I", self.fp.read(4))[0]
|
|
self.chansTableOffset = struct.unpack(">I", self.fp.read(4))[0] << 2
|
|
|
|
self.markedBlocks = []
|
|
|
|
self.partitionOpen = -1
|
|
self.partitionOffset = -1
|
|
self.partitionType = -1
|
|
|
|
def markContent(self, offset, size):
|
|
blockStart = offset / 0x7C00
|
|
blockLen = (align(size, 0x7C00)) / 0x7C00
|
|
|
|
for x in range(blockStart, blockStart + blockLen):
|
|
try:
|
|
self.markedBlocks.index(blockStart + x)
|
|
except:
|
|
self.markedBlocks.append(blockStart + x)
|
|
|
|
def decryptBlock(self, block):
|
|
if len(block) != 0x8000:
|
|
raise Exception('Block size too big/small')
|
|
|
|
blockIV = block[0x3d0:0x3e0]
|
|
#print 'IV %s (len %i)\n' % (hexdump(blockIV), len(blockIV))
|
|
blockData = block[0x0400:0x8000]
|
|
|
|
return Crypto().decryptData(self.partitionKey, blockIV, blockData, True)
|
|
|
|
def readBlock(self, blockNumber):
|
|
self.fp.seek(self.partitionOffset + 0x20000 + (0x8000 * blockNumber))
|
|
return self.decryptBlock(self.fp.read(0x8000))
|
|
|
|
def readPartition(self, offset, size):
|
|
|
|
readStart = offset / 0x7C00
|
|
readLen = (align(size, 0x7C00)) / 0x7C00
|
|
blob = ''
|
|
|
|
#print 'Read at 0x%x (Start on %i block, ends at %i block) for %i bytes' % (offset, readStart, readStart + readLen, size)
|
|
|
|
self.fp.seek(self.partitionOffset + 0x20000 + (0x8000 * readStart))
|
|
|
|
for x in range(readLen + 1):
|
|
blob += self.decryptBlock(self.fp.read(0x8000))
|
|
|
|
self.markContent(offset, size)
|
|
|
|
#print 'Read from 0x%x to 0x%x' % (offset, offset + size)
|
|
offset -= readStart * 0x7C00
|
|
return blob[offset:offset + size]
|
|
|
|
def readUnencrypted(self, offset, size):
|
|
if offset + size > 0x20000:
|
|
raise Exception('This read is on encrypted data')
|
|
|
|
# FIXMII : Needs testing, extracting the tmd cause to have 10 null bytes in the end instead of 10 useful bytes at start :|
|
|
self.fp.seek(self.partitionOffset + 0x2A4 + offset)
|
|
return self.fp.read(size)
|
|
class fstObject(object):
|
|
#TODO: add ability to extract file by path
|
|
def __init__(self, name, iso=None):
|
|
''' do init stuff here '''
|
|
self.parent = None
|
|
self.type = 1 #directory: 1, file:0
|
|
self.name = name
|
|
self.nameOff = 0
|
|
self.fileOffset = 0
|
|
self.size = 0
|
|
self.children = []
|
|
self.iso = iso
|
|
def addChild(self, child):
|
|
if self.type == 0:
|
|
raise Exception('I am not a directory.')
|
|
child.parent = self
|
|
self.children.append(child)
|
|
def getISO(self):
|
|
if(self.parent == None):
|
|
return self.iso
|
|
return self.parent.getISO()
|
|
def getList(self, pad=0):
|
|
if self.type == 0:
|
|
return ("\t" * pad) + self.getPath() + "\n"
|
|
str = "%s[%s]\n" % ("\t" * (pad), self.getPath())
|
|
for child in self.children:
|
|
str += child.getList(pad+1)
|
|
return str
|
|
def count(self):
|
|
if self.type == 0:
|
|
return 1
|
|
i = 0
|
|
for child in self.children:
|
|
i += child.count()
|
|
return i
|
|
def getPath(self):
|
|
if(self.parent == None):
|
|
return "/"
|
|
if(self.type == 1):
|
|
return self.parent.getPath() + self.name + "/"
|
|
return self.parent.getPath() + self.name
|
|
def write(self, cwd):
|
|
if(self.type==0):
|
|
print cwd + self.getPath()
|
|
#print self.nameOff
|
|
open(cwd + self.getPath(), 'w+b').write(self.getISO().readPartition(self.fileOffset, self.size))
|
|
if(self.type==1):
|
|
if(self.parent != None):
|
|
try:
|
|
os.makedirs(cwd + self.getPath())
|
|
except:
|
|
j = None
|
|
for child in self.children:
|
|
child.write(cwd)
|
|
def parseFst(self, fst, names, i, fstDir):
|
|
size = struct.unpack(">I", fst[(12*i + 8):(12*i + 8) + 4])[0]
|
|
nameOff = struct.unpack(">I", fst[(12*i):(12*i) + 4])[0] & 0x00ffffff
|
|
fileName = names[nameOff:]
|
|
fileName = fileName[:fileName.find('\0')]
|
|
|
|
if i == 0:
|
|
j = 1
|
|
while(j<size):
|
|
j = self.parseFst(fst, names, j, fstDir)
|
|
return size
|
|
if fst[12 * i] == '\x01':
|
|
newDir = self.fstObject(fileName)
|
|
j = i+1
|
|
while(j<size):
|
|
j = self.parseFst(fst, names, j, newDir)
|
|
fstDir.addChild(newDir)
|
|
return size
|
|
else:
|
|
fileOffset = 4 * struct.unpack(">I", fst[(12*i + 4):(12*i + 4) + 4])[0]
|
|
newFile = self.fstObject(fileName)
|
|
newFile.type = 0
|
|
newFile.fileOffset = fileOffset
|
|
newFile.size = size
|
|
newFile.nameOff = nameOff
|
|
fstDir.addChild(newFile)
|
|
self.markContent(fileOffset, size)
|
|
return i+1
|
|
|
|
def openPartition(self, index):
|
|
if index+1 > self.partitionCount:
|
|
raise ValueError('Partition index too big')
|
|
|
|
self.partitionOpen = index
|
|
|
|
self.partitionOffset = self.partsTableOffset + (8 * self.partitionOpen)
|
|
|
|
self.fp.seek(self.partsTableOffset + (8 * self.partitionOpen))
|
|
|
|
self.partitionOffset = struct.unpack(">I", self.fp.read(4))[0] << 2
|
|
self.partitionType = struct.unpack(">I", self.fp.read(4))[0]
|
|
|
|
self.fp.seek(self.partitionOffset)
|
|
|
|
self.tikData = self.fp.read(0x2A4)
|
|
self.partitionKey = Ticket(self.tikData).getTitleKey()
|
|
|
|
self.tmdSize = struct.unpack(">I", self.fp.read(4))[0]
|
|
self.tmdOffset = struct.unpack(">I", self.fp.read(4))[0] >> 2
|
|
|
|
self.certsSize = struct.unpack(">I", self.fp.read(4))[0]
|
|
self.certsOffset = struct.unpack(">I", self.fp.read(4))[0] >> 2
|
|
|
|
self.H3TableOffset = struct.unpack(">I", self.fp.read(4))[0] >> 2
|
|
|
|
self.dataOffset = struct.unpack(">I", self.fp.read(4))[0] >> 2
|
|
self.dataSize = struct.unpack(">I", self.fp.read(4))[0] >> 2
|
|
|
|
self.fstOffset = 4 * struct.unpack(">I", self.readPartition (0x424, 4))[0]
|
|
self.fstSize = 4 * struct.unpack(">I", self.readPartition (0x428, 4))[0]
|
|
|
|
self.dolOffset = 4 * struct.unpack(">I", self.readPartition (0x420, 4))[0]
|
|
self.dolSize = self.fstOffset - self.dolOffset
|
|
|
|
self.appLdr = self.Apploader().unpack(self.readPartition (0x2440, 32))
|
|
self.partitionHdr = self.discHeader().unpack(self.readPartition (0x0, 0x400))
|
|
|
|
self.partitionIos = TMD(self.getPartitionTmd()).getIOSVersion() & 0x0fffffff
|
|
|
|
def getFst(self):
|
|
fstBuf = self.readPartition(self.fstOffset, self.fstSize)
|
|
return fstBuf
|
|
|
|
def getIsoBootmode(self):
|
|
if self.discHdr.discId == 'R' or self.discHdr.discId == '_':
|
|
return 2
|
|
elif self.discHdr.discId == '0':
|
|
return 1
|
|
|
|
def getOpenedPartition(self):
|
|
return self.partitionOpen
|
|
|
|
def getOpenedPartitionOffset(self):
|
|
return self.partitionOffset
|
|
|
|
def getOpenedPartitionType(self):
|
|
return self.partitionType
|
|
|
|
def getPartitionsCount(self):
|
|
return self.partitionCount
|
|
|
|
def getChannelsCount(self):
|
|
return self.channelsCount
|
|
|
|
def getPartitionCerts(self):
|
|
return self.readUnencrypted(self.certsOffset, self.certsSize)
|
|
|
|
def getPartitionH3Table(self):
|
|
return self.readUnencrypted(self.H3TableOffset, 0x18000)
|
|
|
|
def getPartitionTmd(self):
|
|
return self.readUnencrypted(self.tmdOffset, self.tmdSize)
|
|
|
|
def getPartitionTik(self):
|
|
self.fp.seek(self.partitionOffset)
|
|
return self.fp.read(0x2A4)
|
|
|
|
def getPartitionApploader(self):
|
|
return self.readPartition (0x2440, self.appLdr.size + self.appLdr.trailingSize + 32)
|
|
|
|
def getPartitionMainDol(self):
|
|
return self.readPartition (self.dolOffset, self.dolSize)
|
|
|
|
def dumpPartition(self, fn):
|
|
rawPartition = open(fn, 'w+b')
|
|
|
|
print 'Partition useful data %i Mb' % (align(len(self.markedBlocks) * 0x7C00, 1024) / 1024 / 1024)
|
|
|
|
self.fp.seek(self.partitionOffset)
|
|
rawPartition.write(self.fp.read(0x2A4)) # Write teh TIK
|
|
rawPartition.write(self.readUnencrypted(0, 0x20000 - 0x2A4)) # Write the TMD and other stuff
|
|
|
|
for x in range(len(self.markedBlocks)):
|
|
rawPartition.write(self.readBlock(self.markedBlocks[x])) # Write each decrypted block
|
|
|
|
class updateInf():
|
|
def __init__(self, f):
|
|
self.buffer = open(f, 'r+b').read()
|
|
def __str__(self):
|
|
out = ''
|
|
|
|
self.buildDate = self.buffer[:0x10]
|
|
self.fileCount = struct.unpack('>L', self.buffer[0x10:0x14])[0]
|
|
|
|
out += 'This update partition was built on %s and has %i files\n\n' % (self.buildDate, self.fileCount)
|
|
|
|
for x in range(self.fileCount):
|
|
updateEntry = self.buffer[0x20 + x * 0x200:0x20 + (x + 1) * 0x200]
|
|
titleType = struct.unpack('>L', updateEntry[:0x4])[0]
|
|
titleAttr = struct.unpack('>L', updateEntry[0x4:0x8])[0]
|
|
titleUnk1 = struct.unpack('>L', updateEntry[0x8:0xC])[0]
|
|
titleType2 = struct.unpack('>L', updateEntry[0xC:0x10])[0]
|
|
titleFile = updateEntry[0x10:0x50]
|
|
titleFile = titleFile[:titleFile.find('\x00')]
|
|
titleID = struct.unpack('>Q', updateEntry[0x50:0x58])[0]
|
|
titleMajor = struct.unpack('>B', updateEntry[0x58:0x59])[0]
|
|
titleMinor = struct.unpack('>B', updateEntry[0x59:0x5A])[0]
|
|
titleName = updateEntry[0x60:0xA0]
|
|
titleName = titleName[:titleName.find('\x00')]
|
|
titleInfo = updateEntry[0xA0:0xE0]
|
|
titleInfo = titleInfo[:titleInfo.find('\x00')]
|
|
out += 'Update type : 0x%x\n' % titleType
|
|
out += 'Update flag : %i (0 means critical, 1 means need reboot)\n' % titleAttr
|
|
out += 'Update file : %s\n' % titleFile
|
|
out += 'Update ID : %lu\n' % titleID
|
|
out += 'Update version : %i.%i\n' % (titleMajor, titleMinor)
|
|
out += 'Update name : %s\n' % titleName
|
|
out += 'Update info : %s\n' % titleInfo
|
|
out += '\n'
|
|
|
|
return out
|
|
|
|
|