Archives and compression: tar, gzip, bzip2 and xz

Archive versus compress

An archive joins many files and folders into one file, keeping names, owners, permissions and times; tar (tape archive) does this. Compression makes a file smaller by encoding repeated patterns; gzip, bzip2 and xz do this, one file at a time. That is why you so often see .tar.gz: first tar joins, then gzip shrinks.

The tar options you need

# tar -czf /backup/etc-2026-10-19.tar.gz /etc
tar: Removing leading `/' from member names
# tar -tzf /backup/etc-2026-10-19.tar.gz | head -3
etc/
etc/mtab
etc/fstab
# mkdir /tmp/restore
# tar -xzf /backup/etc-2026-10-19.tar.gz -C /tmp/restore etc/fstab   # one file only
# ls /tmp/restore/etc
fstab
# tar -cJf home.tar.xz -C /home asha ravi                             # paths relative to /home
# tar -xf home.tar.xz -C /srv/restore                                 # modern tar detects the compression

Choosing a compressor

$ gzip access.log          # becomes access.log.gz (the original is replaced)
$ gunzip access.log.gz     # back
$ gzip -k access.log       # -k keeps the original
$ xz -9 big.sql            # -9 = best compression, slow
$ zcat access.log.gz | grep 503     # read a .gz without unpacking (also xzcat, bzcat)
$ ls -lh
-rw-r--r--. 1 root root  86M Oct 19 16:01 big.sql.xz

Text such as logs and SQL dumps often compresses to a tenth of its size or less. Photos, videos and already-compressed files hardly shrink at all; do not waste CPU compressing them again.

Copying archives between servers

Archives are usually copied to another machine for safety. scp file user@host:/path copies over SSH; rsync -av copies only what changed, which is ideal for repeated backups. A backup kept only on the same disk is not a backup.

Common mistakes

• tar -cfz archive.tar.gz ...: the word after -f is taken as the archive name, so this creates a file named z. Keep f last: -czf.
• Extracting in the wrong directory and spreading files everywhere, or over live files. List with -t first and use -C to a fresh directory.
• Backing up /var/www without --selinux or --acls, then restoring and getting permission errors. For system files, add --selinux --acls --xattrs or run restorecon after restore.
• Never testing a restore. A backup you have not restored once is a hope, not a backup.