tar: creating archives

Copying a project, sending a set of logs or taking a backup all have the same problem: many files, many directories, permissions and dates that should survive the trip. tar (from *tape archive*, because it was born to write to tape drives) solves it by gluing a whole tree into one file, the archive or *tarball*. Inside, every file keeps its name, its path, its owner, its permissions and its modification time.

The three letters: c, f, v

tar always needs one operation: c to create, t to list, x to extract (the next lesson covers the last two). Then almost always f NAME, the archive file to write; without it tar would try the old tape device. v (verbose) prints each name as it is added. The letters can be written in a bundle without a dash, the traditional way, or with one: tar cvf and tar -cvf are the same. The f must come last in the bundle because the next word is its file name.

~% cd lab/l71
~% tar cvf docs.tar docs
docs/
docs/notes.txt
docs/report.txt
~% ls -l docs.tar
-rw-r--r--    1 root     root          3584 Oct  4 23:16 docs.tar
~% tar cf both.tar docs logs        # several sources in one archive

Directories are added recursively: name docs and everything under it goes in. Notice the size: two tiny text files became 3584 bytes. tar stores each file in 512-byte blocks plus a header block per entry, and pads the end. That is the key fact of this lesson: tar bundles, it does not compress. Compression is a separate step you will meet in lesson 73.

Paths inside the archive: -C

tar stores each name exactly as you typed it. Archive docs from inside lab/l71 and the members are docs/.... Archive /root/lab/l71/docs and they become root/lab/l71/docs/...: tar drops the leading / (it says so with *removing leading '/' from member names*) so that extracting can never overwrite system files, but the long path stays. Whoever unpacks it gets a root/lab/l71 tree they did not want. The fix is -C DIR, which makes tar change into DIR before reading the names that follow, without you having to cd:

~% tar cf /root/lab/l71/abs.tar /root/lab/l71/docs
tar: removing leading '/' from member names
~% tar tf /root/lab/l71/abs.tar | head -1
root/lab/l71/docs/
~% tar cf /root/lab/l71/clean.tar -C /root/lab/l71 docs
~% tar tf /root/lab/l71/clean.tar | head -1
docs/

Leaving things out

--exclude PATTERN skips every name that matches a shell-style pattern, and can be repeated. Quote the pattern so the shell does not expand it first. For long lists, put one pattern per line in a file and pass it with -X FILE; the opposite, -T FILE, archives only the names listed in a file.

~% tar cvf backup.tar --exclude '*.log' docs logs
docs/
docs/notes.txt
docs/report.txt
logs/
logs/README