safecopy Logo

safecopy is a data recovery tool which tries to extract as much data as possible from a problematic (i.e. damaged sectors) source - like floppy drives, hard disk partitions, CDs, tape devices, ..., where other tools like dd would fail due to I/O errors.

Safecopy includes a low level IO layer to read CDROM disks in raw mode, and issue device resets and other helpful low level operations on a number of other device classes.

The project also includes a device simulator which can be used to simulate bad media for testing and benchmarking safecopy as well as other data rescue tools. This article (Statistical analysis of damaged data storage media read timings) (pdf version)covers the theoretical background and methods for gathering statistics on to be simulated media.


2012-03-10: safecopy-1.7
Added new feature to force continue a previous safecopy run at arbitrary position, and to keep trying to open source files even when they went away. This allows copying from devices that vanish temporarily in case of errors, like USB drives that renumerate in case of device resets.

2009-07-28: safecopy-1.5
Safecopy now supports generic direct hardware access with O_DIRECT, as well as reading from /dev/raw devices. Direct IO is enabled with the --sync option. For this it was necessary to align the main read buffer with the blocksize address boundary, so DMA transfers to the read buffer are possible, bypassing CPU-based in-memory copying by the kernel.
The --debug switch, present since about safecopy 1.3, has now been documented, and a new switch -T <timingfile> will generate a statistics file with the read timing information in microseconds for each read block. This is useful for statistical analysis of bad hardware behaviour and finding strategies to speed up recovery tools like safecopy, ddrescue and the like.
Read Statistical analysis of damaged data storage media read timings for more information about what this is good for.

The test suite of safecopy has been rewritten to be more portable.

2009-05-13: Safecopy has been internally redesigned to improve reliability and predictability of behaviour. It now ships with a detailed textual definition of this behaviour along with set of automated tests that attempt to proof conformity with this specification under various test conditions. A separate libc wrapper library is used to simulate IO errors on virtual files in a predictable way, providing test cases.
Safecopys command line option parameters have been extended. This now allows specifying block and skip sizes as a percentage of the file/disk size, or as multiples of the hardware block size. Along with better auto detection of hardware sector and disk sizes, this increases safecopys usability significantly.
New sensible option presets for incremental data rescuing make safecopy almost as intuitive and quick to use as ddrescue, while still allowing full control over all parameters and full predictability of behaviour.

Fixes in low level data recovery from CDROM and DVD code.
Raw sector reading is not possible on DVDs, but DVDs had been incorrectly detected as CDs leading to failures on low level access. DVDs are now detected successfully and offer a drive reset as the only low level operation.
The type of a CD (Audio, Mode1, Mode2, etc...) is now successfully detected, and sector alignments are calculated correctly. Since its not possible with reasonable overhead to calculate sector addressing on Mixed-mode CDs (since the sector sizes can change with each track) safecopy does not attempt to raw read from these anymore. Device reset will still be attempted if necessary.
Device resets will now only be issued in case of previous read errors.
The "twaddle" system call is now supported on floppy drives to potentially aid in data resurrection.
Safecopy has now a manpage, which includes basically all the information of the README file in more standardized format.

Sourceforge Project page

Freshmeat Project page

safecopy download page (SF)

Project by Corvus Corax (corvuscorax<at> - distributed under the GPL (v2 or higher)

Safe copying of files and partitions.


A main problem with damaged storage hardware is, that once you get an
unrecoverable IO error further reading from the file / device often fails until
the file has been closed and re-opened.

The normal copy tools like cat, cp or dd do not allow creation of an image file
from a disk or CD-ROM once reading of a sector failed.

Safecopy tries to get as much data from the source as possible, even resorting
to device specific low level operations if applicable.

This is achieved by identifying problematic or damaged areas, skipping over
them and continuing reading afterwards. The corresponding area in the
destination file is either skipped (on initial creation that means padded with
zeros) or deliberately filled with a recognizable pattern to later find
affected files on a corrupted device.
Safecopy uses an incremental algorithm to identify the exact beginning and end
of bad areas, allowing the user to trade minimum accesses to bad areas for
thorough data resurrection.
Multiple passes over the same file are possible, to first retrieve as much data
from a device as possible with minimum harm, and then trying to retrieve some
of the remaining data with increasingly aggressive read attempts.

For this to work, the source device or file has to be seekable. For unseekable
devices (like tapes) you can try to use an external script to execute a
controlled skip over the damaged part for you.
(For example by using "mt seek" and "mt tell" on an SCSI tape device)
See the "-S <seekscript>" parameter for details.

Performance and success of this tool depend extremely on the device driver,
firmware and underlying hardware.

Currently safecopy supports RAW access to CDROM drives to read data directly of
a CD, bypassing some driver dependant error correction. This can speed up data
retrieval from CDs and reduce system load during recovery, as well as increase
the success rate. Safecopy uses the disc status syscall to determine sector
size and addressing of CDs. This fails on mixed-mode or multi-session CDs,
since the sector layout can change within the disk, but would still work on the
the big majority of disks. Other disks can still be recovered using normal high
level data access. Safecopy auto-detects the disk type involved during the scan
for disk and block size.

Some CD/DVD drives are known to cause the ATAPI bus to crash on errors, causing
the device driver to freeze for times up to and beyond a minute per error. Try
to avoid using such drives for media recovery. Using safecopys low level access
features might help under some circumstances.

Some drives can read bad media better than others. Be sure to attempt data
recovery of CDs and DVDs on several different drives and computers.
You can use safecopys incremental recovery feature to read previously
unreadable sectors only.


Data recovery from damaged media is a delicate task, in the worst case its
success or failure can safe or ruin whole companies and seal affected peoples
personal fate. It is paramount that any tools written for that purpose are
reliable and trustworthy.

A user needs to know what exactly the software is doing to his hardware and
data. The outcome of any operation needs to be both understandable and

An "intelligent data resurrection wizard" with unknown complex internal
behaviour may be a nifty tool, but does not meet the above requirements of
predictable outcome, nor will the user know in advance what is done to his

The operation sequence of safecopy has been kept relatively simple to assure
this predictability. Unfortunately feature additions have risen the complexity
and lead to undefined outcome in the past when include and exclude lists had
been mixed, especially when mixing different block sizes. In the worst case
this could have lead to overwritten data in the destination file on a later
incremental run with the mark (-M) option.

From version 1.3 on, safecopy ships with a test suite that can be used to verify
safecopys behaviour in a set of test cases, simulating the combination of bad
blocks in input with different include and exclude lists, both with and without
marking. Releases are only made if safecopy passes those test cases according
to the specification.

This textual specification of behaviour of safecopy can be found in the file
specification.txt shipped with safecopy.

Different use cases:

	How do I...

- resurrect a file from a mounted but damaged media, that copy will fail on:
	safecopy /path/to/problemfile ~/saved-file

- create an filesystem image of a damaged disk/cdrom:
	safecopy /dev/device ~/diskimage

- resurrect data as thoroughly as possible?
	safecopy source dest -f 1* -R 8 -Z 2
	(assuming logical misalignment of blocks to sectors)
	safecopy source dest -f 1* -r 1 -R 8 -Z 2

- resurrect data as fast as possible, or
- resurrect data with low risk of damaging the media further:
	safecopy source dest -f 10% -R 0 -Z 0

- resurrect some data fast, then read more data thoroughly later:
	safecopy source dest -f 10% -R 0 -Z 0 -o badblockfile
	safecopy source dest -f 1* -R 8 -Z 2 -I badblockfile
	Alternate approach using the new preset features:
	safecopy source dest --stage1
	safecopy source dest --stage2
	safecopy source dest --stage3

- utilize some friends CD-ROM drives to complete the data from my damaged CD:
	safecopy /dev/mydrive imagefile <someoptions> -b <myblocksize> \
		-o myblockfile;
	safecopy /dev/otherdrive imagefile <someoptions> -b <otherblocksize> \
		-I myblockfile -i <myblocksize> -o otherblockfile;
	safecopy /dev/anotherdrive imagefile <someoptions> \
		-b <anotherblocksize> -I otherblockfile -i <otherblocksize>

- interrupt and later resume a data rescue operation:
	safecopy source dest
	<CTRL+C> (safecopy aborts)
	safecopy source dest -I /dev/null

- interrupt and later resume a data rescue operation with correct
  badblocks output:
	safecopy source dest <options> -o badblockfile
	<CTRL+C> (safecopy aborts)
	mv badblockfile savedbadblockfile
	safecopy source dest -I /dev/null -o badblockfile
	cat badblockfile >>savedbadblockfile

- interrupt and resume in incremental mode:
        (this needs a bit of bash scripting to get the correct badblock lists)
	safecopy source dest <options> -o badblockfile1
	safecopy source dest <options> -I badblockfile1 -o badblockfile2
	<CTRL+C> (safecopy aborts)
	latest=$( tail -n 1 badblockfile2 )
	if [ -z $latest ]; then latest=-1; fi;
	cat badblockfile1 | while read block; do
		[ $block -gt $latest ] && echo $block >>badblockfile2;
	safecopy source dest <options> -I badblockfile2 -o badblockfile3

- find the corrupted files on a partially successful rescued file system:
	safecopy /dev/filesystem image -M CoRrUpTeD
	fsck image
	mount -o loop image /mnt/mountpoint
	grep -R /mnt/mountpoint "CoRrUpTeD"
  (hint: this might not find all affected files if the unreadable
   parts are smaller in size than your marker string)

- exclude the previously known badblocks list of a filesystem from
  filesystem image creation:
  	dumpe2fs -b /dev/filesystem >badblocklist
	safecopy /dev/filesystem image \
		-X badblocklist -x <blocksize of your fs>

- create an image of a device that starts at X and is Y in size:
	safecopy /dev/filesystem -b <bsize> -s <X/bsize> -l <Y/bsize>

- combine two partial images of rescued data without access to the actual
  (damaged) source data:
	(This is a bit tricky. You need to get badblocks lists for both files
	somehow to make safecopy know where the missing data is. If you used
	the -M (mark) feature you might be able to automatically compute these,
	however this feature is not provided by safecopy. Lets assume you have
	two badblocks files.
	you have:	image1.dat
			image1.badblocks (blocksize1)
			image2.badblocks (blocksize2)
	The file size of image1 needs to be greater or equal to that of image2.
	(If not, swap them)

	cp image2.dat combined.dat
	safecopy image1.dat combined.dat -I image2.badblocks -i blocksize2 \
		-X image1.badblocks -x blocksize1

	(This gets you the combined data, but no output badblocklist.
	The resulting badblocks list would be the badblocks that are
		a: in both badblocks lists, or
		b: in image1.badblocks and beyond the file size of image2
	It should be reasonably easy to solve this logic in a short shell
	script. One day this might be shipped with safecopy, until then
	consider this your chance to contribute to a random open source

- rescue data of a tape device:
	If the tape device driver supports lseek(), treat it as any file,
	otherwise utilize the "-S" option of safecopy with a to be
	self-written script to skip over the bad blocks.
	(for example using "mt seek")
	Make sure your tape device doesn't auto-rewind on close.
	Send me feedback if you had any luck doing so, so I can update
	this documentation.

    Q:	Why create this tool if there already is something like dd-rescue and
	other tools for that purpose?
    A:  Because I didn't know of dd(-)rescue when I started, and I felt like
	it. Also I think safecopy suits the needs of a user in data loss peril
	better due to more readable output and more understandable options than
	some of the other tools. (Then again I am biased. Compare them
	yourself.) Meanwhile safecopy supports low level features other tools

    Q:	What exactly does the -Z option do?
    A:	Remember back in MS-DOS times when a floppy would make a "neek nark"
    	sound 3 times every time when running into a read error?
	This happened when the BIOS or DOS disk driver moved the IO head
	to its boundaries to possibly correct small cylinder misalignment,
	before it tried again.
	Linux doesn't do that by default, neither do common CDROM drives or
	drivers.  Nevertheless forcing this behaviour can increase your
	chance of reading bad sectors from a CD __BIG__ time.
	(Unlike floppies where it usually has little effect)

    Q:	Whats my best chance to resurrect a CD that has become unreadable?
    A:	Try making a backup image on many different computers and drives.
	The abilities to read from bad media vary extremely. I have a 6 year
	old Lite On CDRW drive, that even reads deeply and purposely scratched
	CDs (as in with my key, to make it unreadable) flawlessly. A CDRW drive
	of the same age at work doesn't read any data from that part of the CD
	at all, while most DVD and combo drives have bad blocks every couple
	hundred bytes.
	Make full use of safecopys RAW access features if applicable.
	(-L 2 option)
	As a general guideline:
	   -CDRW drives usually do better than read-only CD drives.
	   -CD only drives sometimes do better on CDs than DVD drives.
	   -PC drives are sometimes better than laptop ones.
	   -A drive with a clean lens does better than a dirtball.
	   -Cleaning up CDs helps.
	   -Unless you use chemicals.
	   -Using sticky tape on a CD will rip of the reflective layer
	    permanently rendering the disk unreadable.

    Q:	Whats my best chance to resurrect a floppy that became unreadable?
    A:	Again try different floppy drives. Keep in mind that it might be
	easier to further damage data on a bad floppy than on a CD.
	(Don't overdo read attempts)

    Q:	What about BlueRay/HDDVD disks?
    A:	Hell if I knew, but generally they should be similar to DVDs.
	It probably depends how the drives firmware acts up.

    Q:  My hard drive suddenly has many bad sectors, what should I do?
    A:  Speed is an essential factor when rescuing data from a bad hard drive.
	Accesses to bad areas or even just time running can further damage the
	drive and make formerly readable areas unreadable, be it due to
	temperature rise, damaged heads scratching still good parts of the
	surface, bearings degrading due to vibration, etc.  Its advisable to
	shut the system down and remove the hard drive from the computer as soon
	as errors occur and as fast as possible without causing further damage.
	(Don't pull the plug! Press reset to force reboot and then power down
	via power button/ACPI)
	Set up a rescue system with enough disk space to store all the data
	from the damaged drive (and possibly multiple copies of it).  If you
	have an external hard drive case that connects via USB SATA or SCSI,
	allowing a hot plug of the drive into a running system, use it.  This
	allows you to prepare everything without the need for the drive to power
	up and possibly BIOS or operating system involuntarily accessing it.
	You also get easier access to the drive to check temperature or noise
	during operation.
	When you rescue data, rescue good data first before attempting to
	access bad sectors. Safecopy allows you to skip known problem sectors
	using a badblock exclude file (-X) which you might be able to retrieve
	from system logs or from the drive internal logs, via smartmontools or
	similar software. Be aware that you might possibly need to convert
	physical sector numbers into logical block numbers depending on your
	Also you should tell safecopy to jump out of any problematic areas ASAP
	and continue somewhere else. Parameters "-f 10% -r 10% -R 0 -Z 0" would
	do the trick by making safecopy skip 10% of the device content and
	continue there without backtracking. You can always attempt to get the
	data in between later, first get the supposedly good data on the rest
	of the drive. Depending on the method of data recovery you plan on
	using, it may make sense to mark the bad data with the "-M" option.
	This allows you to later find files affected by data corruption more
	easily.  Use the "-o" option to make safecopy write a badblock list
	with all blocks skipped or unreadable.
	When safecopy is done with this first run, you can attempt a second go
	trying to get more data. Using smaller values for "-f" and allowing
	safecopy to backtrack for the end of the affected area "-r 1*".
	Be sure to use incremental mode "-I" to only read the blocks skipped
	in the first run.
	It may make sense to let the drive cool down between runs.
	Once you got all the data from the "good" areas of the drive you can
	risk more "thorough" access to it. Increase the numbers of retries
	"-R" in case of bad blocks, maybe together with a head realignment
	"-Z 1". "-f 1*" would make safecopy try to read on every single block,
	not skipping over bad areas at all.
	If your drive stops responding, power it down, let it cool down for a
	while, then try again.
	(I heard from people who brought dead drives back to live for a short
	time by cooling them to low temperatures with ice-packs.)

	!!! If the data is really important, go to a professional data recovery
	!!! specialist right away, before doing further damage to the drive.

Safecopy 1.7 by CorvusCorax
Usage: safecopy [options] <source> <target>
	--stage1 : Preset to rescue most of the data fast,
	           using no retries and avoiding bad areas.
	           Presets: -f 10% -r 10% -R 1 -Z 0 -L 2 -M BaDbLoCk
	                    -o stage1.badblocks
	--stage2 : Preset to rescue more data, using no retries
	           but searching for exact ends of bad areas.
	           Presets: -f 128* -r 1* -R 1 -Z 0 -L 2
	                    -I stage1.badblocks
	                    -o stage2.badblocks
	--stage3 : Preset to rescue everything that can be rescued
	           using maximum retries, head realignment tricks
	           and low level access.
	           Presets: -f 1* -r 1* -R 4 -Z 1 -L 2
	                    -I stage2.badblocks
	                    -o stage3.badblocks
	All stage presets can be overridden by individual options.
	-b <size> : Blocksize for default read operations.
	            Set this to the physical sectorsize of your media.
	            Default: 1*
	            Hardware block size if reported by OS, otherwise 4096
	-f <size> : Blocksize when skipping over badblocks.
	            Higher settings put less strain on your hardware,
	            but you might miss good areas in between two bad ones.
	            Default: 16*
	-r <size> : Resolution in bytes when searching for the exact
	            beginning or end of a bad area.
	            If you read data directly from a device there is no
	            need to set this lower than the hardware blocksize.
	            On mounted filesystems however, read blocks
	            and physical blocks could be misaligned.
	            Smaller values lead to very thorough attempts to read
	            data at the edge of damaged areas,
	            but increase the strain on the damaged media.
	            Default: 1*
	-R <number> : At least that many read attempts are made on the first
	              bad block of a damaged area with minimum resolution.
	              More retries can sometimes recover a weak sector,
	              but at the cost of additional strain.
	              Default: 3
	-Z <number> : On each error, force seek the read head from start to
	              end of the source device as often as specified.
	              That takes time, creates additional strain and might
	              not be supported by all devices or drivers.
	              Default: 1
	-L <mode> : Use low level device calls as specified:
	                   0  Do not use low level device calls
	                   1  Attempt low level device calls
	                      for error recovery only
	                   2  Always use low level device calls
	                      if available
	            Supported low level features in this version are:
	                SYSTEM  DEVICE TYPE   FEATURE
	                Linux   cdrom/dvd     bus/device reset
	                Linux   cdrom         read sector in raw mode
	                Linux   floppy        controller reset, twaddle
	            Default: 1
	--sync : Use synchronized read calls (disable driver buffering).
	         Safecopy will use O_DIRECT if supported by the OS
	         and O_SYNC otherwise.
	         Default: Asynchronous read buffering by the OS is allowed
	--forceopen : Keep trying to reopen the source after a read errer
	              useful for USB drives that go away temporarily.
	              Warning: This can cause safecopy to hang
	                       until aborted manually!
	              Default: Abort on fopen() error
	-s <blocks> : Start position where to start reading.
	              Will correspond to position 0 in the destination file.
	              Default: block 0
	-l <blocks> : Maximum length of data to be read.
	              Default: Entire size of input file
	-I <badblockfile> : Incremental mode. Assume the target file already
	                    exists and has holes specified in the badblockfile.
	                    It will be attempted to retrieve more data from
	                    the listed blocks or from beyond the file size
	                    of the target file only.
	                    Warning: Without this option, the destination file
	                    will be emptied prior to writing.
	                    Use -I /dev/null if you want to continue a previous
	                    run of safecopy without a badblock list.
	                    Implies: -c 0 if -c is not specified
	                    Default: none ( /dev/null if -c is given )
	-i <bytes> : Blocksize to interpret the badblockfile given with -I.
	             Default: Blocksize as specified by -b
	-c <blocks> : Continue copying at this position.
	              This allows continuing if the output is a block device
	              with a fixed size as opposed to a growable file,
	              where safecopy cannot determine how far it already got.
	              The blocksize used is the same as for the -I option.
	              -c 0 will continue at the current destination size.
	              Implies: -I /dev/null if -I is not specified
	              Default: none, 0 if -I is specified
	-X <badblockfile> : Exclusion mode. If used together with -I,
	                    excluded blocks override included blocks.
	                    Safecopy will not read or write any data from
	                    areas covered by exclude blocks.
	                    Default: none ( 0 if -I is given ) 
	-x <bytes> : Blocksize to interpret the badblockfile given with -X.
	             Default: Blocksize as specified by -b
	-o <badblockfile> : Write a badblocks/e2fsck compatible bad block file.
	                    Default: none
	-S <seekscript> : Use external script for seeking in input file.
	                  (Might be useful for tape devices and similar).
	                  Seekscript must be an executable that takes the
	                  number of blocks to be skipped as argv1 (1-64)
	                  the blocksize in bytes as argv2
	                  and the current position (in bytes) as argv3.
	                  Return value needs to be the number of blocks
	                  successfully skipped, or 0 to indicate seek failure.
	                  The external seekscript will only be used
	                  if lseek() fails and we need to skip over data.
	                  Default: none
	-M <string> : Mark unrecovered data with this string instead of
	              skipping it. This helps in later finding corrupted
	              files on rescued file system images.
	              The default is to zero unreadable data on creation
	              of output files, and leaving the data as it is
	              on any later run.
	              Warning: When used in combination with
	              incremental mode (-I) this may overwrite data
	              in any block that occurs in the -I file.
	              Blocks not in the -I file, or covered by the file
	              specified with -X are save from being overwritten.
	              Default: none
	--debug <level> : Enable debug output. Level is a bit field,
	                  add values together for more information:
	                     program flow:     1
	                     IO control:       2
	                     badblock marking: 4
	                     seeking:          8
	                     incremental mode: 16
	                     exclude mode:     32
	                  or for all debug output: 255
	                  Default: 0
	-T <timingfile> : Write sector read timing information into
	                  this file for later analysis.
	                  Default: none
	-h | --help : Show this text

Valid parameters for -f -r -b <size> options are:
	<integer>	Amount in bytes - i.e. 1024
	<percentage>%	Percentage of whole file/device size - e.g. 10%
	<number>*	-b only, number times blocksize reported by OS
	<number>*	-f and -r only, number times the value of -b

Description of output:
	. : Between 1 and 1024 blocks successfully read.
	_ : Read of block was incomplete. (possibly end of file)
	    The blocksize is now reduced to read the rest.
	|/| : Seek failed, source can only be read sequentially.
	> : Read failed, reducing blocksize to read partial data.
	! : A low level error on read attempt of smallest allowed size
	    leads to a retry attempt.
	[xx](+yy){ : Current block and number of bytes continuously
	             read successfully up to this point.
	X : Read failed on a block with minimum blocksize and is skipped.
	    Unrecoverable error, destination file is padded with zeros.
	    Data is now skipped until end of the unreadable area is reached.
	< : Successful read after the end of a bad area causes
	    backtracking with smaller blocksizes to search for the first
	    readable data.
	}[xx](+yy) : current block and number of bytes of recent
	             continuous unreadable data.

Copyright 2012 CorvusCorax
This is free software. You may redistribute copies of it under
the terms of the GNU General Public License version 2 or above.
There is NO WARRANTY, to the extent permitted by law.