dd_rescue

The dd_rescue project is specially designed to copy data from one file or block device to another.

It is intended for error recovery. It doesn't abort on errors and doesn't truncate the output file.

dd_rescue uses large block sizes to quicken the copying, but falls back to small blocks upon encountering errors.

What is new in this release:

• The -p/--preserve option now also copies extended attributes (including ACLs) if there are any.
• Many checks and test cases have been added to make check.
• Half-empty blocks are now detected now in sparse detection.
• If copying with an odd file offset, the odd-sized write will be done first so that further accesses should be better aligned, which should help with performance (and optionally sparse detection).

What is new in version 1.39:

• The main change is a fix for a bug where the last block could have appended zeros if hardbs==softbs (bnc #833765).
• Beyond that, ARM sparse detection has been sped up a bit (~15%), and the man page has a clarification w.r.t. the meaning of -y.
• autoconf is now used to detect platform features to help portability.

What is new in version 1.38:

• This version brings some further optimization of the SSE2 sparse block detection (adding 40% performance).
• There's also code for AVX2, but it's disabled, as it couldn't be tested.
• There's a test case for the 1.35/1.36 bug now that can be run by make check.
• ETA and curr.rate have been improved a bit (floating averages), and the --force/-f switch gained the capability to override a non-zero output position for non-seekable ouput files.

What is new in version 1.37:

• The SSE2 optimized sparse block detection had a bug that slipped through the test cases in 1.35/1.36.
• Fortunately, --sparse / -a is not on by default.
• The bug is fixed, and that's the reason for a quick release.
• Apart from that, there are some fixes for this routine for big-endian machines, though that's not relevant for dd_rescue itself.

What is new in version 1.36:

• This version fixes an issue with an overflow when displaying avg.load after a while.
• It also enables faster detection of zero-filled blocks (sparse mode, option -a) on x86 (32-bit) with runtime detection (unlike x86-64 where you always have SSE2-capable CPUs).
• There's also some asm code to achieve a 3x speedup of ARM CPUs on this task.

What is new in version 1.34:

• Minor bugs were fixed.
• A bunch of minor code adjustments were made to improve compilation on many compilers (including C++) and more *nix systems.
• Specifically, there were fixes for FreeBSD. dd_rescue can now load libfallocate at runtime (with libdl), if you want to build it that way.
• Some more write errors are now treated as fatal to avoid pointless repeats.

What is new in version 1.33:

• dd_rescue 1.33 brings a new double-overwrite (random and zero) mode.
• More importantly, dd_rescue now supports long options and has a man page that properly documents all the options and modes.

What is new in version 1.32:

• This version adds two new options: -x facilitates appending to an output file, and -Y OUTFILE (can be specified multiple times) allows the user to specify secondary output files which receive the same data (at the same position) as the primary output file.

What is new in version 1.31:

• This version brings a few minor cleanups (messages) and one feature: it now has a mode where the output file/partition/disk is overwritten two (-3) or three (-4) times with random data and finally with zeros.
• This supports secure data destruction according to German data privacy standards from BSI.