Enough of that; let's talk about what makes GNU ddrescue my favorite. It performs block-level copies of the failing media, and so it doesn't matter what filesystem is on the media. You're probably thinking it sounds like the venerable dd command, and it is similar, with some significant improvements. dd works fine on healthy disks, but when it encounters a read error it stops, and you have to manually restart it. It reads the media sequentially, which is very slow, and if there are a lot of bad blocks it may never complete a full pass.

GNU ddrescue is fully automatic and fast for a block-level copy program, and you want speed when a drive full of important data is dying. It seeks out good blocks to copy and skips over the bad blocks. It optionally records all activity in a logfile, so you can resume where you left off if the copying is interrupted for any reason. It is best to always generate a logfile, because every time you power up the failing drive the more likely it is to die completely. Using a logfile ensures that ddrescue will not repeat operations, but will move on and look for new good blocks to copy.

When you are rescuing a failing drive, the first step is to copy it with ddrescue. Then take the original offline, and perform any additional recovery operations on the copy. Don't touch the original any more than you have to. You can copy the copy as many times as you need for insurance.

You need a healthy drive to copy your rescued data to. I prefer USB-attached media such as a USB hard drive, USB thumb drive, Compact Flash, or SD cards. Of course a second internal hard drive is a good option, or this might be your chance to finally use that eSATA port that always looked like it should be cool and useful, but you never found a reason to use it. Your second drive should be at least 50% larger than the drive you're recovering. The troubled drive must not be mounted. The simplest invocation looks like this:

# ddrescue /dev/sda1 /dev/sdb1 logfile

Here, /dev/sda1 is a partition on the failing drive. Everything on /dev/sdb1 will be overwritten, and the logfile will be written to /dev/sdb1. You can name the logfile anything you want. You can rescue an entire drive if you prefer, like this:

# ddrescue /dev/sda /dev/sdb logfile

Note that if there is more than one partition on the failing drive and the partition table is damaged, you will have to re-create it on the rescue drive. I copy one partition at a time to avoid this sort of drama.

You can have ddrescue make multiple passes with the -r option; sometimes you can make a more complete recovery this way. You can go as high as you want; I use 3-5:

# ddrescue -r5 /dev/sda2 /dev/sdb1 logfile

Sometimes ddrescue is nearly magical for rescuing scratched CDs and DVDs. The first command copies the disk, and the second command copies it to a blank disk:

# ddrescue -n -b2048 /media/cdrom image logfile
# ddrescue -d -b2048 /media/cdrom image logfile

You can give the image file whatever name you like. While I've never needed to go beyond the basics in this article, ddrescue has a whole lot of other capabilities that you can learn about in the GNU ddrescue manual.