Skip to content
Create a gist now

Instantly share code, notes, and snippets., "undelete" version
# Copyright 2007 by Tobia Conforto <>
# 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 2 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.
# You should have received a copy of the GNU General Public License along with this program.
# If not, see
# Use this program to extract files from a disk image from a Creative Zen Xtra or Zen Vision M player.
# Unlike the this is based on (, this also finds deleted files
# and can extract most files after the player has been formatted. However, it doesn't distinguish between
# "songs" and "archives" areas on the player, and also extracts a few player system files.
# NOTE: Make sure it's set for the right player before you run it. Otherwise it won't find anything.
# See the "ZENVISION=..." line just after the comments.
# Versions: 0.1 2007-08-13 Initial release
# 0.2 2008-05-12 Small fixes for Zen Xtra models
# 0.3 2009-02-23 Zen Vision M compatible version (Leho Kraav <>)
# 0.2a 2010-07-20 Undid most of 0.3's changes to make it Zen Xtra-compatible again
# (because Tobia Conforto's site no longer has it and 0.3 is the only version of this
# script I can find anywhere)
# Modified it to use LRUCache from
# because I couldn't find the module LRU that it used before.
# Added code to reencode path and filenames to UTF-8 since certain "weird" characters in
# filenames appearing on the device caused an unhandled exception.
# (Joe McKenzie <>)
# 0.2b 2010-07-21 Fixed problem when the filename field contains a path.
# Fixed problem with files added by XNJB for Max OS X
# Fixed a serious bug in the previous revision of this on Gist that created insane dirs
# Combined Xtra and Vision M tag checks to make it a bit easier to convert the script
# Removed use of LRUCache because:
# a. Most sectors are accessed exactly once, making the cache pointless.
# b. On my Slackware 13.1 AMD64 box with Python 2.6,
# with LRUCache: 2.5 MB/s
# without cache: 80 MB/s (d'oh...)
# (AMD Sempron 140, 2 GB RAM, 40 GB Zen Xtra disk image stored on one WD20EARS
# [2 TB WD Green drive], recovering to a directory on another WD20EARS)
# **** Cleaned up code and comments a bit, for instructions to make it work with a ****
# **** Zen Vision M, search this file for "visionm" ****
# (Joe McKenzie <>)
# 0.4 2011-02-07 Changed it to scan for anything that looks like an inode and recover it.
# Basically works like an "undelete" program now. It'll find some files even after you
# format your Zen. NOTE: May not recover ALL files after a format!!
# (Joe McKenzie <>)
# 0.5 2011-08-02 Removed parsing of directories in CFSInode.__init__ so it ignores dirs
# with errors in them. Now it should work better on damaged filesystems.
from __future__ import division
import sys, os, codecs, array, time, operator, getopt, re
# visionm: Set this to True for Zen Vision M, False for Zen Xtra
class CFS:
clusterSize = 0x8000
clusterSize = 0x2000
def __init__(self, filename, offset = 0):
'''Filename and optional offset where the CFS filesystem begins
(offset of cluster -1, the one filled with 0xff)'''
self.image = file(filename)
self.offset = offset
def __getitem__(self, key):
'''Get the nth CFS cluster from the image and cache it for later usage.
Accepts simple slices of clusters, but doesn't process negative indices.
In any case it returns the requested data as a byte string.'''
if isinstance(key, slice):
cstart, cstop = key.start, key.stop
cstart, cstop = key, key + 1
data = ''
for cluster in range(cstart, cstop): + (cluster + 1) * self.clusterSize)
data +=
return data
def get_byteswapped_data(self, cluster):
'''Get the nth CFS cluster from the image, without caching it.
Swap the position of every two bytes and return it as an array object.
This method is designed for bulk file retrieving.'''
a = array.array('H') + (cluster + 1) * self.clusterSize)
a.fromfile(self.image, self.clusterSize // 2)
return a
def inode(self, cluster):
return CFSInode(self, cluster)
def pdp_uint32_xtra(data, offset = 0):
o2, o1, o4, o3 = map(ord, data[offset : offset + 4])
return (o1 << 24) | (o2 << 16) | (o3 << 8) | o4
def pdp_uint32_vision(data, offset = 0):
o4, o3, o2, o1 = map(ord, data[offset : offset + 4])
return (o1 << 24) | (o2 << 16) | (o3 << 8) | o4
def pdp_uint32(data, offset = 0):
return pdp_uint32_vision(data, offset)
return pdp_uint32_xtra(data, offset)
def pdp_uint16(data, offset = 0):
o2, o1 = map(ord, data[offset : offset + 2])
return (o1 << 8) | o2
def ucs2string(data, offset, length): # length in bytes
return codecs.utf_16_le_decode(data[offset : offset + length])[0]
def pdp_getbit(bitmap, bit_no):
return (pdp_uint32(bitmap, bit_no // 32 * 4) >> (bit_no % 32)) & 1
class CFSInode:
filename = '(no filename)'
filesize = 0
path = []
def __init__(self, cfs, cluster):
self.filename = '(no filename)'
self.filesize = 0
self.path = []
self.cluster = cluster
self.cfs = cfs
inode = cfs[cluster]
# reading misc flags and values
# print "pdp_uint: %x" % pdp_uint32(inode[4:8])
# print "cluster: %x" % cluster
assert pdp_uint32(inode[4:8]) == cluster # self-reference
self.serial = pdp_uint32(inode, 0x78)
# reading metadata
count_metadata = pdp_uint32(inode, 0x7c)
offset = 0x80
self.metadata = {}
for i in range(count_metadata):
assert pdp_uint16(inode, offset) == 3
length = pdp_uint16(inode, offset + 2)
tag = ucs2string(inode, offset + 4, 4)
self.metadata[tag] = inode[offset + 10 : offset + 10 + length]
# byte reordering issue, 07 -> 70, 0= -> =0, 0> -> >0
# but we cannot figure out where to get path info, tag '51' doesnt work
#...putting both the Xtra and Vision M tags here
# First one listed is for Zen Xtra, second/third are for Zen Vision M
# Remove the ones for the Nomad you don't have if the extra checks cause problems
if tag == '07' or tag == '70':
self.filename = unicode(ucs2string(inode, offset + 10, length - 2)).strip('\\/').encode("utf-8")
#I don't have a Vision M, so I can't test this... 51 or =0 might work...
elif tag == '0=' or tag == '51' or tag == '=0':
# handle UTF-8 properly
self.path = unicode(ucs2string(inode, offset + 10, length - 2)).encode("utf-8")
# split on both \ and /
self.path = re.split(r"[\\/]+", self.path.strip('\\/'))
elif tag == '0>' or tag == '>0':
self.filesize = pdp_uint32(inode, offset + 10)
offset += 10 + length
#if filename has / or \ in it, split and append those to the path
# (XNJB puts whole path in filename and leaves path blank)
if '/' in self.filename or '\\' in self.filename:
self.path.extend(re.split(r"[\\/]+", self.filename))
self.filename = self.path.pop()
print 'adjusted filename: %s' % self.filename
print 'adjusted path: %s' % repr(self.path)
# collecting flat list of data clusters
self.dataclusters = []
pointerclusters = []
for off in range(0x20, 0x4c + 1, 4):
c = pdp_uint32(inode, off)
if c != 0xFFFFFFFFL:
second_class_chain = pdp_uint32(inode, 0x58)
if second_class_chain != 0xFFFFFFFFL:
third_class_chain = pdp_uint32(inode, 0x64)
if third_class_chain != 0xFFFFFFFFL:
for off in range(0, clusterSize, 4):
c = pdp_uint32(cfs[third_class_chain], off)
if c == 0xFFFFFFFFL:
for pnt in pointerclusters:
for off in range(0, clusterSize, 4):
c = pdp_uint32(cfs[pnt], off)
if c == 0xFFFFFFFFL:
def __getitem__(self, key):
'''Returns the given byte (or byte slice) from the file contents.'''
if isinstance(key, slice):
bstart, bstop = key.start, key.stop
bstart, bstop = key, key + 1
cs = self.cfs.clusterSize
cstart = bstart // cs
cstop = (bstop - 1) // cs + 1
data = ''.join([ self.cfs[x] for x in self.dataclusters[cstart : cstop] ])
return data[bstart - cs * cstart : bstop - cs * cstart]
class CFSDirEntry:
def __init__(self, cfs, entrydata):
self.cluster = pdp_uint32(entrydata) # cluster no. of the inode
# length of full filename
self.len_filename = pdp_uint16(entrydata, 4)
# first 15 chars of filename
self.shortname = ucs2string(entrydata, 8, min(30, self.len_filename * 2))
if __name__ == '__main__':
# commandline arguments
optlist, args = getopt.gnu_getopt(sys.argv[1:], 'o:')
opts = dict(optlist)
offset = int(opts.get('-o', 20 * 2**20))
if len(args) != 2:
print 'DISK_OR_IMAGE is the disk containing the filesystem, or an image thereof'
print 'OFFSET is the offset at which the filesystem starts (in bytes, default 20M)'
print 'OUTPUT_DIR is the directory in which to place the recovered files'
print '***NOTICE***: This version of zenrecover extracts everything it can find,'
print 'no matter what section it\'s in. This can and will extract songs, archives,'
print 'Zen system files like "sg00.log", and probably even some deleted stuff!'
cfs = CFS(args[0], offset)
outdir = args[1]
except OSError:
print "Can't create output directory \""+outdir+"\" -- make sure it doesn't already exist"
lastfiles = [(1,1)] # timing of latest few files recovered (size in bytes, time in secs)
while True:
if pdp_uint32(cfs[clust][:4]) == 0x3bbe0ad9:
print "\n\nFound inode at cluster 0x%x" % clust
except ValueError:
print "Hit end of disk image, DONE."
print "cluster # "+str(clust)
if isAnInode:
# recover this inode if it's a single file
inode = cfs.inode(clust)
if (not inode.metadata) or inode.filesize == 0:
print " ignoring directory inode or empty file"
t0 = time.time()
print repr(m)
for j in m:
if len(m[j])==4:
print repr(j), pdp_uint32(m[j])
print repr(j), repr(''.join([m[j][x] for x in range(0,len(m[j]),2)]))
print '\r%.1fMB/s "%s" (%.1fMB)\033[K' % (
operator.truediv(*map(sum, zip(*lastfiles))) / 2**20,
inode.filesize / 2**20),
path = os.path.join(outdir, *inode.path)
f = file(os.path.join(path, inode.filename), 'w')
remaining = inode.filesize
for c in inode.dataclusters:
if remaining >= cfs.clusterSize:
remaining -= min(cfs.clusterSize, remaining)
assert remaining == 0
if len(lastfiles) >= 32: #transfer speed is calculated on latest 32 files
lastfiles.append((inode.filesize, time.time() - t0))
# end of recover code

Do you have any suggestions on imaging the jukebox 3 disk? I have tried this with a basic software I bought and this script runs and recovers nothing. I suspect my image may not have been created properly. Any tips would be greatly appreciated.


btw, here's the output:

breathmint@ubuntu:/media/DATA2/JB3_images$ sudo python ./jb3_20120414.img ./recover

Found inode at cluster 0x29
ignoring directory inode or empty file

Found inode at cluster 0x34
ignoring directory inode or empty file

Found inode at cluster 0x3f
ignoring directory inode or empty file

Found inode at cluster 0x4a
ignoring directory inode or empty file

Found inode at cluster 0x55
ignoring directory inode or empty file

Found inode at cluster 0x60
ignoring directory inode or empty file

Found inode at cluster 0x1edff
ignoring directory inode or empty file

Found inode at cluster 0x1ee03
ignoring directory inode or empty file

Found inode at cluster 0x1ee07
ignoring directory inode or empty file

Found inode at cluster 0x149350
ignoring directory inode or empty file

Found inode at cluster 0x195adf
ignoring directory inode or empty file

Found inode at cluster 0x195ae3
ignoring directory inode or empty file

Found inode at cluster 0x1b8095
ignoring directory inode or empty file

Found inode at cluster 0x1b8099
ignoring directory inode or empty file
Hit end of disk image, DONE.
cluster # 2439319


Firstly, thanks for building/sharing this code, it looks like exactly the thing I need. I had a hard-drive failure with my music collection and the only backup I have is on my Creative Zen Nomad 30Gb player.

I need to run this on Windows 7. Which version of Python do you recommend?
Also I am new to Python, thus what would be the command line including directory structure eg ($ python e:\zenrecover\ e:\music i:) ?




I have come back to it and see that my disk is "not Initialized" in windows. What should I do? I'm hesitant to initialize it as I imagine I may be able to recover less of my music. Is this the case?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.