If you encounter Unix data on the Internet, it is probably provided as a tar archive. This is due to several reasons:

• tar is so easy to use that even many novice Unix users know how to use it.
• every Unix and linux system has tar, and it is truly a universal format.
• most globs of data that are provided on the Internet are directory trees that are transported as a single file. This is exactly what tar does easily.

Although many modern Unix (and expecially linux) systems have versions of tar that are fully functional, with new options to govern how it performs, few users use them. We will cover tar in its most basic form and use only the options that everyone knows about. tar was intended to be simple: let it do what it does well! As we will see, we can use our other archive utility cpio as a good complement to round out our archiving capabilities.

tar is one of the oldest Unix programs. It is so old, in fact, that it is customarily used with the original Unix commandline format, where the first word on the commandline consisted of a sequence of single-letter options, without a preceding dash. Hence, tar cv dir1 is often used instead of the current tar -c -v dir1 or tar -cv dir1. tar, along with some other older Unix utilities, will accept either format, but adding the dashes may be taken as an indication that you are a less-seasoned Unix user.

tar syntax

tar { -c | -t | -x } [ -v ] [ -f pathtotarball ]  [ files and directories ]

tar, whose name stands for tape archiver, was meant to write to magnetic tapes. Thus, by default, it places its output on the default tape device. Today, most users choose to place the archive in a file (lovingly called a tarball), which can be specified on the commandline as -f path, where path is where to place the tar archive. (As with all Unix files, a .tar extension is not necessary, but is customary, and serves as a clue to the contents.)

one of the options -c (create new archive), -t (output a table of contents of an existing archive), or -x (extract from archive) must be given. There is also a less-used -r option to replace files in an existing archive. (-r actually appends new data to the end of an archive, essentially replacing the older version with the newer version when the data is extracted due to the way tar works)

-v tells tar to be verbose. Otherwise, tar silently does its work when using -c, -x, or -r. If -v is used with -t, a long-style listing (including dates and permissions as with ls -l) of the archive contents is produced.

Note: by default, tar is recursive.

Creating archives

tar cvf xxx.tar dir1 dir2 file1 file2 file3

creates a new archive named xxx.tar in the current directory and places copies of dir1 and dir2 (recursively) and the three files in it. Since the v option is used, tar will output a list of the paths archived. 

It is highly recommended that paths given to tar be relative. Old versions of tar accepted absolute paths, but most current versions will delete the leading / of an absolute path. In order to restore those files the user must extract the archive when connected to the root directory /.

Listing a table of contents

tar tf xxx.tar [ paths ]

list the paths of each item found in the archive. If paths are found on the commandline, only paths matching those are listed. If the v option is added, an ls -l style listing is produced.

Extracting data from an archive

tar xf xxx.tar [ paths ]

Extract the entire archive [the default] or only the pieces indicated by paths from the archive. The data is extracted unconditionally, creating missing directories if necessary, and the attributes contained on the archive are restored if possible. Thus, if you ask tar to extract xyz from the archive and there is an xyz already present (in the way), tar will simply overwrite it without asking and irrespective of whether the copy on the archive is newer than the existing one. tar will set the modification date of the data extracted to the one on the archive. It will try to restore the permissions, owner and group from the archive, but file creation rules for regular users cause the owner and group to be reset. (For root, the owner and group are restored from the archive). In addition, umask will alter the permissions while extracting if the user has not reset it.

tar can read/write an archive to standard input or output by using - in the position of the archive path.

Linux versions include a -z option to tar to essentially run gunzip before extracting the archive.

Examples (as a normal user):

• tar cvf myproject.tar myproject

• tar tf myproject.tar 

• tar xfv myproject.tar myproject/Makefile

Note: remember that this syntax is shorthand for tar -x -v -f myproject.tar myproject/Makefile In our example, we have failed to place the f last in the options string so that it is followed immediately by the name of the archive. Some versions of tar will accept this and some will not.

• tar xf myproject.tar 

• tar cf xxx.tar .

Don't forget, archives are not compressed! Remember to compress your archive by running it through a compression utility after creating it. tar archives are almost as large as the source data!

Using tar to restore a file from a simple backup

Jamal should have specified the single file to restore on the tar command

tar xvf ~/project.tar project/foo.java 

Further, perhaps Jamal should add the v option in future so that he can try to limit the damage by quickly using control-C.

Using tar to copy a directory tree with attributes (as root)

You have discovered that your /usr partition is too small. Since you used hard partitions when you were configuring your system, this is a problem. However, one solution is to 'patch' a new partition into the /usr tree at a reasonable point. Examining the sizes of various directories beneath /usr, you discover that the /usr/src directory (which is being used to build linux installed executables from source for customization) is huge. So, you create a new partition large enough to hold /usr/src, create a filesystem on it, and mount it at /mnt/tmp temporarily. The goal is to copy all of /usr/src to the new partition without disturbing it, delete the original, then mount our new parition at /usr/src. This frees up space on /usr. The modern way to do the copy operation is with cp -p. You can, however, do it with tar. Both require you to be root.

Example:

You have downloaded a new version of firefox, as /tmp/firefox-2.0.23.tar.gz which is needed to fix a bug in the version installed on your system.  You notice that the archive contains a single directory named firefox, which contains an executable named firefox. You know that if you execute the firefox executable it runs the new version, but, since this is an archive rather than an RPM file, the firefox files do not replace the default ones on the system, such as /usr/bin/firefox. Thus, when someone (or some GUI launcher) runs firefox, it will run the old installed version rather than your new version. Without figuring out where to drop each of the 1000 new files to replace those currently on the system, how can you upgrade your version?

Solution:

Un-archive the replacement firefox in /usr/local/src. Then create a symbolic link as /usr/local/bin/firefox that points to the firefox executable in /usr/local/src/firefox. Since most linux systems are configured to look for binaries in /usr/local/bin before the system directories, your replacement firefox will be found before the installed firefox.

Copyright 2008 Greg Boyd - All Rights Reserved.