myrescue is a program to rescue the still-readable data from a
damaged harddisk. It is similiar in purpose to dd_rescue, but it
tries to quickly get out of damaged areas to first handle the not
yet damaged part of the disk and return later.
The program tries to copy the device blockwise to a file and
keeps a table ("block bitmap") noting whether a block has been
successfully copied, not yet handled or has had errors. This block
bitmap can be used in successive runs to concentrate on the not yet
The program has a special skip mode to handle read errors. Usually
harddisk surface defects cover more than just one block and continuous
reading in defect areas can damage the surface, the heads and (by
permanent recalibration) the drive mechanics. If this happens, the
chances of rescuing the remaining undamaged data drop dramatically. So
in skip mode, myrescue tries to get out of damaged areas quickly by
exponentially increasing the stepsize. The skipped blocks are marked
as unhandled in the block bitmap and can be retried later.
Finally, the program has an option to multiply try to read a
block before considering it damaged.
This tools is no replacement for a professional data recovery service!
If you do have the latter option, don't even think of using
as it may further damage your disk. This tool is
provided only for the case that you are absolutely desperate and
definitely cannot afford a professional data recovery. Or in case
you know what you are doing, e.g. if you know that it is the aging
of the magnetisation layer that is causing your problem.
In any case do not expect too much. While complete restores have been
witnessed, you should not take them for granted. A better attitude is
to consider your data lost and be glad for any survivors that turn up.
The usual GPL disclaimer applies. Especially the NON-WARRANTY OF
FITNESS FOR A PARTICULAR PURPOSE. Don't blame (or sue) me if it
fails to recover or further damages your data.
And a final word you probably don't want to hear in this situation:
For the future consider a routinely backup to avoid a "next time".
The size of the blocks (in bytes). Set this to your
harddiscs error detection/correction unit size. Usually
this is 4096, which happens to be the default.
The file containing the status table of all blocks. Nice (or
frightening...) to view with hexdump. 01 means OK; 00 means not yet
done; negative values mean the number of failed read attempts. If not
given, defaults to
Abort when encountering errors.
Activate skip mode: When encountering errors increase the stepsize
exponentially until a readable block is found.
Skip blocks that have already had
failures. Useful to avoid scratching the same block over and over again.
The number of times to read a block before it is declared
bad for this run. (You can still retry it on the next run.) Default: 1
The number of the block to start with. Default: 0
The number of the block, where reading stops (not included!).
Default: size of
Reverse reading direction, i.e. from
Only try to read blocks within
blocks from an already successfully read block.
Extends -f to also skip any block within
blocks of a block to be skipped as specified by -f.
Randomly jump across the disc after reading
blocks. This might be useful to scan discs with scattered defects.
In jump mode -S causes myrescue to jump to a new block upon the first
failed sector or upon hitting a sector to be skipped as specified by
-f, -G or -F.
Display usage information.
Make sure you have sufficient disk space to copy the whole
partition (not just the used ammount of data) to plus some
space for the block bitmap (1 byte per block).
Determine the hardware block size (CRC/ECC unit) of your
harddisk. This may be found out from hdparm, some entries in
/proc/ide/hd? or on the web. I have not yet checked whether
this is possible with an ioctl. If you have, please let me
Start a skip mode run with one retry per block to first copy
the undamaged area.
Start a normal run with one retry per block to copy the
remaining skipped blocks. You may try to use -f 1 to skip the
damaged blocks from the first run.
Repeat until the number of errors seems to have converged.
Try waiting a couple of hours between the retries.
Repeat this with higher retry counts and wait for convergence.
Make a copy of the rescued data and run fsck on it.
Mount the filesystem (if copied to a file: via loopback) and check
your data. If directory information has been destroyed, fsck moves
unidentifiable file fragments to lost+found, so you should also check
It may help to try reading non-defect areas in between to allow the
drive to recalibrate.
The developers are glad to hear about your experiences. Please post
them to the
forum on the Sourceforge Project page. Thank you!
The handling of the
currently relies on the filesystem semantics, that when
ing beyond the end of file and then writing, the space in between is
filled with zero-bytes.
The block bitmap overflows after 128 failed read attempts.