Documentation about Backing-Up under Linux

At another page, I have described how I backup Windows machines. Under Windows, I prefer image backups because the integration between user data and the OS is tight, especially for user settings. Back-up under Windows is simplified by the fact that volume-shadow-copy technology is built-in.

When installing Debian Linux, I had decided to keep things simple and not use LVM (Logical Volume Manager). LVM offers similar functionality to Windows shadow copy. E.g. LVM allows to back-up a system while it is running.

Not using LVM has the consequence that I need to boot the PC from an external medium in order to back-up the system, because only then you can be sure that there are no files/data bases, that are in an intermediate state.

However, I used the flexibility Linux offers in the following way:

• I opted to use the "root" user to administer the system rather than using sudo, because root's home directory is /root and this directory is not part of the /home directory
• All data of other users and data multiple users share live under the /home directory
• I put /home on a separate disk partition
  • This allows to mount /home read-only before back-up to make sure that no write access takes place while back-up is running
  • As /home is quite large, I put it on a separate hard disk drive

Hence I do two different back-op processes:

1. I run a file based back-up of /home as root, while the PC is booted normally, but only root is logged in
2. To back up the other disk (=the system disk), I boot a live Debian-system from an USB-stick. Then I back-up the complete system-disk partition.

1. File-Based Back-up

For 1. I adopted tar for file backups, because it

• is proven
• is available for all distributions
• is easy to understand
• supports incremental and differential back-ups

2. Image Back-up

For 2. I generated a live Debian Linux system using https://fai-project.org/FAIme/live/ and use the partclone tool (Debian package partclone) to do an image back-up of the system-disk. This image back-up does not take long, because the system disk only uses about 25 GB.

About tar

tar, whose name comes from "tape archive", is an old and proven utility. It is available for all Unix-like operating systems.

There seem to be two major tar variants:

• GNU tar. This is available on most Linux systems. However, there is also a port available for Windows. For example, I got GNU tar version 1.30 from 2017 as part of git for Windows.
• BSD tar. This is available on systems like FreeBSD, but is also standard on Mac OS. Since Windows 10 1903, it is also part of the standard Windows command line tools (it had been introduced the same time as curl)

Comparison of Functionality of the Two Variants

tar is mostly used to do backups. It used to be used to do backups on physical tapes. The functionality still reflects limitations of tapes. For example the tar "update" mode will write the updated file(s) to the end of the archive, while leaving the older information in the archive untouched. This is because "seeking" on a tape was slow and costly.

Incremental backups using "update" mode

To do incremental backups, you might consider using the "update" mode, to add new or updated files to the archive. This mode is supported by both BSD tar and GNU tar.

To restore, you would use the "extract" mode together with option -k "Keep (don't overwrite) existing files". However, this approach does not cater for files that were deleted in the source between the full backup and subsequent updates. Also files that were renamed would appear twice after extract:

1. under their old name, and
2. under their new name.

Similar for files that were moved: they appear

1. at their old location, and
2. at their new location.

Another drawback of "update" is the fact, that it works only on uncompressed archives. This is due to the fact that tar compresses after creating the archive and before writing to tape/file. The approach seems to be due to legacy constraints:

• it is slow and costly to "seek" on tapes
• compression used to be costly when CPU power was limited

Incremental backups using the "listed-incremental" option

GNU tar has additional functionality for incremental backups. This is not available in BSD tar. GNU tar creates and maintains an additional file with meta data along side the archive file. The standard extension for this file is snar. Example syntax:

tar -czvg mybackup_full.snar -f mybackup_full.tgz dir-to-backup

The options

• c stands for "create"
• z for gzip compression
• v for verbose messages
• g is short hand for --listed-incremental.

An additional option p ‑ to preserve meta data like access rights ‑ is only relevant for extract.

There is no mention of the "differential backup" strategy in the on-line documentation. However, the GNU tar manual (see section 5.) calls a full backup a "level 0" backup and the first incremental backup "level 1". The manual explicitly states that you might want to do more level 1 backups by creating a "working copy" of the snar file created at level 0. I tend to call this approach a differential backup. The approach would be:

cp mybackup_full.snar mybackup_diff-1.snar
tar -czvg mybackup_diff-1.snar -f mybackup_diff-1.tgz dir-to-backup

Extract from Incremental Backups

To restore or "extract" from incremental backups, the GNU tar manual stipulates to also use option --listed-incremental with =/dev/null as argument (it could be any argument). So with just a full and a level 1 backup:

tar -xzpf mybackup_full.tgz
tar -xzpg /dev/null -f mybackup_diff-1.tgz

The following alternative also works:

tar --incremental -xzpf mybackup_diff-1.tgz

To list what is going on, you might do:

tar --incremental -tvvzpf mybackup_diff-1.tgz

Caveats

tar stores "device numbers" in the snapshot file and also uses these to check if a file has changed since the last backup. You might want to use --no-check-device option to avoid a full backup if device numbers change for some reason. There is a tar-snapshot-edit (tar-edit-snapshot ?) utility to deal with such snapshot file issues.

References

1. https://unix.stackexchange.com/questions/13093/how-to-add-update-a-file-to-an-existing-tar-gz-archive
2. Introduction to the tar command line syntax https://www.computernetworkingnotes.com/linux-tutorials/tar-command-examples-in-linux.html
3. Introduction to the syntax of incremental backups https://www.computernetworkingnotes.com/linux-tutorials/create-and-restore-incremental-backups-in-linux-with-tar.html
4. Explains level 1 (=differential) and multi-level backups http://paulwhippconsulting.com/blog/using-tar-for-full-and-incremental-backups/
5. The tar manual https://www.gnu.org/savannah-checkouts/gnu/tar/manual/tar.html#Backups
6. Also explains why to copy the snar file https://etutorials.org/Linux+systems/how+linux+works/Chapter+13+Backups/13.4+Using+tar+for+Backups+and+Restores/
7. Difference between --listed-incremental and --newer https://unix.stackexchange.com/questions/307530/what-is-the-difference-between-tars-newer-and-listed-incremental-options
8. More details on incremental backups https://floatingoctothorpe.uk/2018/incremental-tar-backups.html
9. Torture-testing Unix Backup and Archive Programs https://web.archive.org/web/20010622155048id_/http://reality.sgi.com/zwicky_neu/testdump.doc.html

Last change: 2025-12-23
© 2002-2025 Dr. Thomas Redelberger redetho(a‍t)gmx.de

• Home
  • Projects
  • Documentation
    • SW Development
    • Computing
      • Windows
      • Linux
        • WSL
        • Back-Up
      • VM
    • Content
  • Site Map
  • Data Privacy