DeDRM_tools/Calibre_Plugins/eReaderPDB2PML_plugin/erdr2pml.py

528 lines
21 KiB
Python
Raw Normal View History

#!/usr/bin/env python
# vim:ts=4:sw=4:softtabstop=4:smarttab:expandtab
#
# erdr2pml.py
2009-12-17 22:54:15 +01:00
#
# This is a python script. You need a Python interpreter to run it.
# For example, ActiveState Python, which exists for windows.
# Changelog
2009-12-17 22:54:15 +01:00
#
# Based on ereader2html version 0.08 plus some later small fixes
#
# 0.01 - Initial version
# 0.02 - Support more eReader files. Support bold text and links. Fix PML decoder parsing bug.
# 0.03 - Fix incorrect variable usage at one place.
# 0.03b - enhancement by DeBockle (version 259 support)
# Custom version 0.03 - no change to eReader support, only usability changes
# - start of pep-8 indentation (spaces not tab), fix trailing blanks
# - version variable, only one place to change
# - added main routine, now callable as a library/module,
# means tools can add optional support for ereader2html
# - outdir is no longer a mandatory parameter (defaults based on input name if missing)
# - time taken output to stdout
# - Psyco support - reduces runtime by a factor of (over) 3!
# E.g. (~600Kb file) 90 secs down to 24 secs
# - newstyle classes
# - changed map call to list comprehension
# may not work with python 2.3
# without Psyco this reduces runtime to 90%
# E.g. 90 secs down to 77 secs
# Psyco with map calls takes longer, do not run with map in Psyco JIT!
# - izip calls used instead of zip (if available), further reduction
# in run time (factor of 4.5).
# E.g. (~600Kb file) 90 secs down to 20 secs
# - Python 2.6+ support, avoid DeprecationWarning with sha/sha1
# 0.04 - Footnote support, PML output, correct charset in html, support more PML tags
# - Feature change, dump out PML file
# - Added supprt for footnote tags. NOTE footnote ids appear to be bad (not usable)
# in some pdb files :-( due to the same id being used multiple times
# - Added correct charset encoding (pml is based on cp1252)
# - Added logging support.
# 0.05 - Improved type 272 support for sidebars, links, chapters, metainfo, etc
# 0.06 - Merge of 0.04 and 0.05. Improved HTML output
# Placed images in subfolder, so that it's possible to just
# drop the book.pml file onto DropBook to make an unencrypted
# copy of the eReader file.
# Using that with Calibre works a lot better than the HTML
# conversion in this code.
# 0.07 - Further Improved type 272 support for sidebars with all earlier fixes
# 0.08 - fixed typos, removed extraneous things
2009-12-17 22:54:15 +01:00
# 0.09 - fixed typos in first_pages to first_page to again support older formats
# 0.10 - minor cleanups
# 0.11 - fixups for using correct xml for footnotes and sidebars for use with Dropbook
# 0.12 - Fix added to prevent lowercasing of image names when the pml code itself uses a different case in the link name.
# 0.13 - change to unbuffered stdout for use with gui front ends
2010-02-24 15:09:01 +01:00
# 0.14 - contributed enhancement to support --make-pmlz switch
2010-03-02 13:46:56 +01:00
# 0.15 - enabled high-ascii to pml character encoding. DropBook now works on Mac.
2010-11-11 23:11:36 +01:00
# 0.16 - convert to use openssl DES (very very fast) or pure python DES if openssl's libcrypto is not available
# 0.17 - added support for pycrypto's DES as well
2011-01-06 08:10:38 +01:00
# 0.18 - on Windows try PyCrypto first and OpenSSL next
2011-02-02 15:41:15 +01:00
# 0.19 - Modify the interface to allow use of import
# 0.20 - modify to allow use inside new interface for calibre plugins
2011-10-28 08:24:15 +02:00
# 0.21 - Support eReader (drm) version 11.
# - Don't reject dictionary format.
# - Ignore sidebars for dictionaries (different format?)
2011-10-28 08:24:15 +02:00
__version__='0.21'
2011-01-06 08:10:38 +01:00
class Unbuffered:
def __init__(self, stream):
self.stream = stream
def write(self, data):
self.stream.write(data)
self.stream.flush()
def __getattr__(self, attr):
return getattr(self.stream, attr)
import sys
import struct, binascii, getopt, zlib, os, os.path, urllib, tempfile
2010-11-11 23:11:36 +01:00
if 'calibre' in sys.modules:
inCalibre = True
else:
inCalibre = False
2011-01-06 08:10:38 +01:00
Des = None
if sys.platform.startswith('win'):
# first try with pycrypto
if inCalibre:
from calibre_plugins.erdrpdb2pml import pycrypto_des
else:
import pycrypto_des
Des = pycrypto_des.load_pycrypto()
2011-01-06 08:10:38 +01:00
if Des == None:
# they try with openssl
if inCalibre:
from calibre_plugins.erdrpdb2pml import openssl_des
else:
import openssl_des
2011-01-06 08:10:38 +01:00
Des = openssl_des.load_libcrypto()
else:
# first try with openssl
if inCalibre:
from calibre_plugins.erdrpdb2pml import openssl_des
else:
import openssl_des
2011-01-06 08:10:38 +01:00
Des = openssl_des.load_libcrypto()
if Des == None:
# then try with pycrypto
if inCalibre:
from calibre_plugins.erdrpdb2pml import pycrypto_des
else:
import pycrypto_des
2011-01-06 08:10:38 +01:00
Des = pycrypto_des.load_pycrypto()
2010-11-11 23:11:36 +01:00
# if that did not work then use pure python implementation
# of DES and try to speed it up with Psycho
if Des == None:
if inCalibre:
from calibre_plugins.erdrpdb2pml import python_des
else:
import python_des
2010-11-11 23:11:36 +01:00
Des = python_des.Des
# Import Psyco if available
try:
# http://psyco.sourceforge.net
import psyco
psyco.full()
except ImportError:
pass
try:
from hashlib import sha1
except ImportError:
# older Python release
import sha
sha1 = lambda s: sha.new(s)
2011-02-02 15:41:15 +01:00
import cgi
import logging
logging.basicConfig()
#logging.basicConfig(level=logging.DEBUG)
2011-02-02 15:41:15 +01:00
class Sectionizer(object):
2011-10-28 08:24:15 +02:00
bkType = "Book"
def __init__(self, filename, ident):
self.contents = file(filename, 'rb').read()
self.header = self.contents[0:72]
self.num_sections, = struct.unpack('>H', self.contents[76:78])
2011-10-28 08:24:15 +02:00
# Dictionary or normal content (TODO: Not hard-coded)
if self.header[0x3C:0x3C+8] != ident:
2011-10-28 08:24:15 +02:00
if self.header[0x3C:0x3C+8] == "PDctPPrs":
self.bkType = "Dict"
else:
raise ValueError('Invalid file format')
self.sections = []
for i in xrange(self.num_sections):
offset, a1,a2,a3,a4 = struct.unpack('>LBBBB', self.contents[78+i*8:78+i*8+8])
flags, val = a1, a2<<16|a3<<8|a4
self.sections.append( (offset, flags, val) )
def loadSection(self, section):
if section + 1 == self.num_sections:
end_off = len(self.contents)
else:
end_off = self.sections[section + 1][0]
off = self.sections[section][0]
return self.contents[off:end_off]
def sanitizeFileName(s):
r = ''
2010-01-06 11:35:15 +01:00
for c in s:
if c in "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789_.-":
r += c
return r
def fixKey(key):
def fixByte(b):
return b ^ ((b ^ (b<<1) ^ (b<<2) ^ (b<<3) ^ (b<<4) ^ (b<<5) ^ (b<<6) ^ (b<<7) ^ 0x80) & 0x80)
return "".join([chr(fixByte(ord(a))) for a in key])
def deXOR(text, sp, table):
r=''
j = sp
for i in xrange(len(text)):
r += chr(ord(table[j]) ^ ord(text[i]))
j = j + 1
if j == len(table):
j = 0
return r
class EreaderProcessor(object):
2011-10-28 08:24:15 +02:00
def __init__(self, sect, username, creditcard):
self.section_reader = sect.loadSection
data = self.section_reader(0)
version, = struct.unpack('>H', data[0:2])
self.version = version
logging.info('eReader file format version %s', version)
if version != 272 and version != 260 and version != 259:
raise ValueError('incorrect eReader version %d (error 1)' % version)
2011-10-28 08:24:15 +02:00
data = self.section_reader(1)
self.data = data
des = Des(fixKey(data[0:8]))
cookie_shuf, cookie_size = struct.unpack('>LL', des.decrypt(data[-8:]))
if cookie_shuf < 3 or cookie_shuf > 0x14 or cookie_size < 0xf0 or cookie_size > 0x200:
raise ValueError('incorrect eReader version (error 2)')
input = des.decrypt(data[-cookie_size:])
def unshuff(data, shuf):
r = [''] * len(data)
j = 0
for i in xrange(len(data)):
j = (j + shuf) % len(data)
r[j] = data[i]
assert len("".join(r)) == len(data)
return "".join(r)
r = unshuff(input[0:-8], cookie_shuf)
def fixUsername(s):
r = ''
for c in s.lower():
if (c >= 'a' and c <= 'z' or c >= '0' and c <= '9'):
r += c
return r
user_key = struct.pack('>LL', binascii.crc32(fixUsername(username)) & 0xffffffff, binascii.crc32(creditcard[-8:])& 0xffffffff)
drm_sub_version = struct.unpack('>H', r[0:2])[0]
self.num_text_pages = struct.unpack('>H', r[2:4])[0] - 1
self.num_image_pages = struct.unpack('>H', r[26:26+2])[0]
self.first_image_page = struct.unpack('>H', r[24:24+2])[0]
2011-10-28 08:24:15 +02:00
# Default values
self.num_footnote_pages = 0
self.num_sidebar_pages = 0
self.first_footnote_page = -1
self.first_sidebar_page = -1
if self.version == 272:
self.num_footnote_pages = struct.unpack('>H', r[46:46+2])[0]
self.first_footnote_page = struct.unpack('>H', r[44:44+2])[0]
2011-10-28 08:24:15 +02:00
if (sect.bkType == "Book"):
self.num_sidebar_pages = struct.unpack('>H', r[38:38+2])[0]
self.first_sidebar_page = struct.unpack('>H', r[36:36+2])[0]
2009-12-17 22:54:15 +01:00
# self.num_bookinfo_pages = struct.unpack('>H', r[34:34+2])[0]
# self.first_bookinfo_page = struct.unpack('>H', r[32:32+2])[0]
# self.num_chapter_pages = struct.unpack('>H', r[22:22+2])[0]
# self.first_chapter_page = struct.unpack('>H', r[20:20+2])[0]
# self.num_link_pages = struct.unpack('>H', r[30:30+2])[0]
# self.first_link_page = struct.unpack('>H', r[28:28+2])[0]
# self.num_xtextsize_pages = struct.unpack('>H', r[54:54+2])[0]
# self.first_xtextsize_page = struct.unpack('>H', r[52:52+2])[0]
# **before** data record 1 was decrypted and unshuffled, it contained data
# to create an XOR table and which is used to fix footnote record 0, link records, chapter records, etc
self.xortable_offset = struct.unpack('>H', r[40:40+2])[0]
self.xortable_size = struct.unpack('>H', r[42:42+2])[0]
self.xortable = self.data[self.xortable_offset:self.xortable_offset + self.xortable_size]
else:
2011-10-28 08:24:15 +02:00
# Nothing needs to be done
pass
2009-12-17 22:54:15 +01:00
# self.num_bookinfo_pages = 0
# self.num_chapter_pages = 0
# self.num_link_pages = 0
# self.num_xtextsize_pages = 0
# self.first_bookinfo_page = -1
# self.first_chapter_page = -1
# self.first_link_page = -1
# self.first_xtextsize_page = -1
logging.debug('self.num_text_pages %d', self.num_text_pages)
logging.debug('self.num_footnote_pages %d, self.first_footnote_page %d', self.num_footnote_pages , self.first_footnote_page)
logging.debug('self.num_sidebar_pages %d, self.first_sidebar_page %d', self.num_sidebar_pages , self.first_sidebar_page)
self.flags = struct.unpack('>L', r[4:8])[0]
reqd_flags = (1<<9) | (1<<7) | (1<<10)
if (self.flags & reqd_flags) != reqd_flags:
print "Flags: 0x%X" % self.flags
raise ValueError('incompatible eReader file')
des = Des(fixKey(user_key))
if version == 259:
if drm_sub_version != 7:
raise ValueError('incorrect eReader version %d (error 3)' % drm_sub_version)
encrypted_key_sha = r[44:44+20]
encrypted_key = r[64:64+8]
elif version == 260:
2011-10-28 08:24:15 +02:00
if drm_sub_version != 13 and drm_sub_version != 11:
raise ValueError('incorrect eReader version %d (error 3)' % drm_sub_version)
2011-10-28 08:24:15 +02:00
if drm_sub_version == 13:
encrypted_key = r[44:44+8]
encrypted_key_sha = r[52:52+20]
else:
encrypted_key = r[64:64+8]
encrypted_key_sha = r[44:44+20]
elif version == 272:
encrypted_key = r[172:172+8]
encrypted_key_sha = r[56:56+20]
self.content_key = des.decrypt(encrypted_key)
if sha1(self.content_key).digest() != encrypted_key_sha:
raise ValueError('Incorrect Name and/or Credit Card')
def getNumImages(self):
return self.num_image_pages
def getImage(self, i):
sect = self.section_reader(self.first_image_page + i)
name = sect[4:4+32].strip('\0')
data = sect[62:]
return sanitizeFileName(name), data
2009-12-17 22:54:15 +01:00
# def getChapterNamePMLOffsetData(self):
# cv = ''
# if self.num_chapter_pages > 0:
# for i in xrange(self.num_chapter_pages):
# chaps = self.section_reader(self.first_chapter_page + i)
# j = i % self.xortable_size
# offname = deXOR(chaps, j, self.xortable)
# offset = struct.unpack('>L', offname[0:4])[0]
# name = offname[4:].strip('\0')
# cv += '%d|%s\n' % (offset, name)
# return cv
# def getLinkNamePMLOffsetData(self):
# lv = ''
# if self.num_link_pages > 0:
# for i in xrange(self.num_link_pages):
# links = self.section_reader(self.first_link_page + i)
# j = i % self.xortable_size
# offname = deXOR(links, j, self.xortable)
# offset = struct.unpack('>L', offname[0:4])[0]
# name = offname[4:].strip('\0')
# lv += '%d|%s\n' % (offset, name)
# return lv
# def getExpandedTextSizesData(self):
# ts = ''
# if self.num_xtextsize_pages > 0:
# tsize = deXOR(self.section_reader(self.first_xtextsize_page), 0, self.xortable)
# for i in xrange(self.num_text_pages):
# xsize = struct.unpack('>H', tsize[0:2])[0]
# ts += "%d\n" % xsize
# tsize = tsize[2:]
# return ts
# def getBookInfo(self):
# bkinfo = ''
# if self.num_bookinfo_pages > 0:
# info = self.section_reader(self.first_bookinfo_page)
# bkinfo = deXOR(info, 0, self.xortable)
# bkinfo = bkinfo.replace('\0','|')
# bkinfo += '\n'
# return bkinfo
def getText(self):
des = Des(fixKey(self.content_key))
r = ''
for i in xrange(self.num_text_pages):
logging.debug('get page %d', i)
r += zlib.decompress(des.decrypt(self.section_reader(1 + i)))
# now handle footnotes pages
if self.num_footnote_pages > 0:
2009-12-17 22:54:15 +01:00
r += '\n'
# the record 0 of the footnote section must pass through the Xor Table to make it useful
sect = self.section_reader(self.first_footnote_page)
fnote_ids = deXOR(sect, 0, self.xortable)
# the remaining records of the footnote sections need to be decoded with the content_key and zlib inflated
des = Des(fixKey(self.content_key))
for i in xrange(1,self.num_footnote_pages):
logging.debug('get footnotepage %d', i)
id_len = ord(fnote_ids[2])
id = fnote_ids[3:3+id_len]
2009-12-17 22:54:15 +01:00
fmarker = '<footnote id="%s">\n' % id
fmarker += zlib.decompress(des.decrypt(self.section_reader(self.first_footnote_page + i)))
fmarker += '\n</footnote>\n'
r += fmarker
fnote_ids = fnote_ids[id_len+4:]
2011-10-28 08:24:15 +02:00
# TODO: Handle dictionary index (?) pages - which are also marked as
# sidebar_pages (?). For now dictionary sidebars are ignored
# For dictionaries - record 0 is null terminated strings, followed by
# blocks of around 62000 bytes and a final block. Not sure of the
# encoding
# now handle sidebar pages
if self.num_sidebar_pages > 0:
2009-12-17 22:54:15 +01:00
r += '\n'
# the record 0 of the sidebar section must pass through the Xor Table to make it useful
sect = self.section_reader(self.first_sidebar_page)
sbar_ids = deXOR(sect, 0, self.xortable)
# the remaining records of the sidebar sections need to be decoded with the content_key and zlib inflated
des = Des(fixKey(self.content_key))
for i in xrange(1,self.num_sidebar_pages):
id_len = ord(sbar_ids[2])
id = sbar_ids[3:3+id_len]
2009-12-17 22:54:15 +01:00
smarker = '<sidebar id="%s">\n' % id
2011-10-28 08:24:15 +02:00
smarker += zlib.decompress(des.decrypt(self.section_reader(self.first_sidebar_page + i)))
2009-12-17 22:54:15 +01:00
smarker += '\n</sidebar>\n'
r += smarker
sbar_ids = sbar_ids[id_len+4:]
return r
2010-03-02 13:46:56 +01:00
def cleanPML(pml):
# Convert special characters to proper PML code. High ASCII start at (\x80, \a128) and go up to (\xff, \a255)
pml2 = pml
for k in xrange(128,256):
badChar = chr(k)
pml2 = pml2.replace(badChar, '\\a%03d' % k)
return pml2
2009-12-17 22:54:15 +01:00
def convertEreaderToPml(infile, name, cc, outdir):
if not os.path.exists(outdir):
os.makedirs(outdir)
2011-02-02 15:41:15 +01:00
bookname = os.path.splitext(os.path.basename(infile))[0]
2009-12-17 22:54:15 +01:00
print " Decoding File"
sect = Sectionizer(infile, 'PNRdPPrs')
2011-10-28 08:24:15 +02:00
er = EreaderProcessor(sect, name, cc)
if er.getNumImages() > 0:
2009-12-17 22:54:15 +01:00
print " Extracting images"
imagedir = bookname + '_img/'
imagedirpath = os.path.join(outdir,imagedir)
if not os.path.exists(imagedirpath):
os.makedirs(imagedirpath)
2009-12-17 22:54:15 +01:00
for i in xrange(er.getNumImages()):
name, contents = er.getImage(i)
file(os.path.join(imagedirpath, name), 'wb').write(contents)
2009-12-17 22:54:15 +01:00
print " Extracting pml"
pml_string = er.getText()
pmlfilename = bookname + ".pml"
2010-03-02 13:46:56 +01:00
file(os.path.join(outdir, pmlfilename),'wb').write(cleanPML(pml_string))
2009-12-17 22:54:15 +01:00
# bkinfo = er.getBookInfo()
# if bkinfo != '':
# print " Extracting book meta information"
# file(os.path.join(outdir, 'bookinfo.txt'),'wb').write(bkinfo)
2011-02-02 15:41:15 +01:00
def decryptBook(infile, outdir, name, cc, make_pmlz):
if make_pmlz :
# ignore specified outdir, use tempdir instead
outdir = tempfile.mkdtemp()
try:
print "Processing..."
convertEreaderToPml(infile, name, cc, outdir)
if make_pmlz :
import zipfile
import shutil
print " Creating PMLZ file"
zipname = infile[:-4] + '.pmlz'
myZipFile = zipfile.ZipFile(zipname,'w',zipfile.ZIP_STORED, False)
list = os.listdir(outdir)
for file in list:
localname = file
filePath = os.path.join(outdir,file)
if os.path.isfile(filePath):
myZipFile.write(filePath, localname)
elif os.path.isdir(filePath):
imageList = os.listdir(filePath)
localimgdir = os.path.basename(filePath)
for image in imageList:
localname = os.path.join(localimgdir,image)
imagePath = os.path.join(filePath,image)
if os.path.isfile(imagePath):
myZipFile.write(imagePath, localname)
myZipFile.close()
# remove temporary directory
shutil.rmtree(outdir, True)
print 'output is %s' % zipname
else :
print 'output in %s' % outdir
print "done"
except ValueError, e:
print "Error: %s" % e
return 1
return 0
2010-02-24 15:09:01 +01:00
def usage():
print "Converts DRMed eReader books to PML Source"
print "Usage:"
print " erdr2pml [options] infile.pdb [outdir] \"your name\" credit_card_number "
print " "
print "Options: "
print " -h prints this message"
print " --make-pmlz create PMLZ instead of using output directory"
print " "
print "Note:"
print " if ommitted, outdir defaults based on 'infile.pdb'"
print " It's enough to enter the last 8 digits of the credit card number"
return
2011-02-02 15:41:15 +01:00
def main(argv=None):
2010-02-24 15:09:01 +01:00
try:
opts, args = getopt.getopt(sys.argv[1:], "h", ["make-pmlz"])
except getopt.GetoptError, err:
print str(err)
usage()
return 1
make_pmlz = False
for o, a in opts:
if o == "-h":
usage()
return 0
elif o == "--make-pmlz":
make_pmlz = True
2009-12-17 22:54:15 +01:00
print "eRdr2Pml v%s. Copyright (c) 2009 The Dark Reverser" % __version__
2010-02-24 15:09:01 +01:00
if len(args)!=3 and len(args)!=4:
usage()
return 1
2010-02-24 15:09:01 +01:00
2011-02-02 15:41:15 +01:00
if len(args)==3:
infile, name, cc = args[0], args[1], args[2]
outdir = infile[:-4] + '_Source'
elif len(args)==4:
infile, outdir, name, cc = args[0], args[1], args[2], args[3]
return decryptBook(infile, outdir, name, cc, make_pmlz)
if __name__ == "__main__":
sys.stdout=Unbuffered(sys.stdout)
2009-12-17 22:54:15 +01:00
sys.exit(main())
2010-11-11 23:11:36 +01:00