June 20, 2010
The documentation for the very good SysRescCD has changed, so I can’t find the standard “here’s how to back up a disk” using sfdisk and partimage anymore.
Luckily, it got saved here:
For future reference, this is a good way to back up a disk.
- Boot the machine with the disk using SysRescCD.
- Mount a big NFS/CIFS drive.
- Copy the MBR:
dd if=/dev/sda of=/cifs/sda.mbr count=1 bs=512
- Copy the partition table from the drive:
sfdisk -d /dev/sda > /cifs/sda.sfdisk
- Run partimage on each partition on the drive
- Finally, for good measure,
dd if=/dev/sda | gzip > /cifs/sda.dd.gz
You can use kill -USR1 pid on the dd process id (not the gzip process id) to see progress.
Most of the restore is straightforward. sfdisk can take its output as input; partimage has restore built in. To decompress the gzip dd image, use:
gzip -dc /cifs/sda.dd.gz | dd of=/dev/sda
Incidentally, a good site for dd information is here:
June 5, 2010
Back in December, I wrote about setting UUIDs so multiple disks could be mounted in the same place using /etc/fstab.
I’ve since decided that was dumb. Disks should have unique UUIDs – Linux may use that for more than just mounting the file system. Luckily, there is a better option: the disk label.
Mounting a disk using the disk label is very similar to mounting using UUIDs. First, you need to give the disk a label. I think you can do that when you mkfs, but it’s pretty cheap to do it after you’ve already bulit the file system using:
e2label /dev/sdb1 mylabel
(substituting whatever device you want for /dev/sdb1 and whatever label you want for mylabel. This will work for ext2, ext3 and ext4 – if you’re using another file system, it should have its own label utility.)
Then, rather than specifying UUID= in /etc/fstab, specify:
Your disk will mount automatically, but still have a unique ID. Sounds better to me.
Sun has a useful tutorial at: http://wikis.sun.com/display/BigAdmin/Using+Disk+Labels+on+Linux+File+Systems