Jump to content

tar (computing)

From Wikipedia, the free encyclopedia

This is an old revision of this page, as edited by 164.67.235.98 (talk) at 05:36, 16 September 2009 (Random access). The present address (URL) is a permanent link to this revision, which may differ significantly from the current revision.

tar
GNU tar 1.16 showing three common types of Tarballs (shown in red).
Filename extension
.tar
Internet media type
application/x-tar
Uniform Type Identifier (UTI)public.tar-archive
Magic numberustar at byte 257
Type of formatFile archiver
Contained bygzip, bzip2, lzma, xz, lzop, compress

In computing, tar (derived from tape archive) is both a file format (in the form of a type of archive bitstream) and the name of a program used to handle such files. The format was created in the early days of Unix and standardized by POSIX.1-1988 and later POSIX.1-2001.

Initially developed to be written directly to sequential I/O devices for tape backup purposes, it is now commonly used to collect many files into one larger file, for distribution or archiving, while preserving file system information such as user and group permissions, dates, and directory structures.

Compression and naming

Conventionally, uncompressed tar archive files have names ending in ".tar". Unlike more recent archive formats, tar files (somefile.tar) are commonly compressed as a whole rather than piecemeal. Applying a compression utility such as gzip, bzip2, lzma or compress to a tar file produces a compressed tar file, typically named with an extension indicating the type of compression (e.g.: somefile.tar.gz).

Popular tar programs like the BSD and GNU versions of tar support the command line options -z (gzip), and -j (bzip2) to automatically compress or decompress the archive file it is currently working with. GNU tar from version 1.20 onwards also supports --lzma (LZMA). 1.21 also supports lzop via --lzop, and 1.22 adds support for xz via --xz or -J. Both will automatically extract compressed gzip and bzip2 archives with or without these options.

MS-DOS's 8.3 filename limitations, and a desire for brevity, resulted in additional popular conventions for naming compressed tar archives, though this practice has declined with FAT offering long filenames.

  • “.tgz” is equivalent to “.tar.gz”
  • “.tbz” (or “.tb2”) is equivalent to “.tar.bz2”
  • “.taz” is equivalent to “.tar.Z”
  • “.tlz” is equivalent to “.tar.lzma”
  • “.txz” is equivalent to “.tar.xz”

A tar file or compressed tar file is commonly referred to as a tarball.

Format details

A tar file is the concatenation of one or more files. Each file is preceded by a 512-byte header record. The file data is written unaltered except that its length is rounded up to a multiple of 512 bytes and the extra space is zero filled. The end of an archive is marked by at least two consecutive zero-filled records. (The origin of tar's record size appears to be the 512-byte disk sectors used in the Version 7 Unix file system.)

Many historic tape drives read and write variable-length data blocks, leaving significant wasted space on the tape between blocks (for the tape to physically start and stop moving). Some tape drives (and raw disks) only support fixed-length data blocks. Also, when writing to any medium such as a filesystem or network, it takes less time to write one large block than many small blocks. Therefore, the tar command writes data in blocks of many 512 byte records. The user can specify a blocking factor, which is the number of records per block; the default is 20, producing 10 kilobyte blocks (which was large when UNIX was invented, but now seems rather small). The final block of an archive is padded out to full length with zero bytes.

File header

The file header block contains metadata about a file. To ensure portability across different architectures with different byte orderings, the information in the header block is encoded in ASCII. Thus if all the files in an archive are text files, and have ASCII names, then the archive is essentially an ASCII text file (containing many NUL or all-zero characters). Subsequent extensions have diluted this original attribute (which was never particularly valuable, due to the presence of NULs, which are often poorly handled by text-manipulation programs).

The fields defined by the original Unix tar format are listed in the table below. The link indicator/file type table includes some modern extensions. When a field is unused it is filled with NUL bytes. The header is padded with NUL bytes to make it fill a 512 byte block.

Field Offset Field Size Field
0 100 File name
100 8 File mode
108 8 Owner's numeric user ID
116 8 Group's numeric user ID
124 12 File size in bytes
136 12 Last modification time in numeric Unix time format
148 8 Checksum for header block
156 1 Link indicator (file type)
157 100 Name of linked file
Link indicator field
Value Meaning
'0' Normal file
(ASCII NUL) Normal file (now obsolete)
'1' Hard link
'2' Symbolic link
'3' Character special
'4' Block special
'5' Directory
'6' FIFO

A directory is also indicated by having a trailing slash (/) in the name.

Numeric values are encoded in octal numbers using ASCII digits, with leading zeroes. For historical reasons, a final null or space character should be used. Thus although there are 12 bytes reserved for storing the file size, only 11 octal digits can be stored. This gives a maximum file size of 8 gigabytes on archived files. To overcome this limitation, some versions of tar, including the GNU implementation, support an extension in which the high-order bit of the leftmost byte of a numeric field indicates that the number is encoded in big-endian binary. Additionally, versions of tar from 1999 and before pad the values with spaces instead of zeroes.

The checksum is calculated by taking the sum of the unsigned byte values of the header block with the eight checksum bytes taken to be ascii spaces (decimal value 32). It is stored as a six digit octal number with leading zeroes followed by a null and then a space. Various implementations do not adhere to this, so relying on the first white space trimmed six digits for checksum yields better compatibility. In addition, some historic tar implementations treated bytes as signed. Readers must calculate the checksum both ways, and treat it as good if either the signed or unsigned sum matches the included checksum.

Unix filesystems support multiple links (names) for the same file. If several such files appear in a tar archive, only the first one is archived as a normal file; the rest are archived as hard links, with the "name of linked file" field set to the first one's name. On extraction, such hard links should be recreated in the file system.

UStar format

Most modern tar programs read and write archives in the new UStar (Uniform Standard Tape Archive) format, which has an extended header definition as defined by the POSIX (IEEE P1003.1) standards group. Older tar programs will ignore the extra information, while newer programs will test for the presence of the "ustar" string to determine if the new format is in use. The UStar format allows for longer file names and stores extra information about each file.

Field Offset Field Size Field
0 156 (as in old format)
156 1 Type flag
157 100 (as in old format)
257 6 UStar indicator "ustar"
263 2 UStar version "00"
265 32 Owner user name
297 32 Owner group name
329 8 Device major number
337 8 Device minor number
345 155 Filename prefix

Uses

Tarpipe

A tarpipe is the process of creating a tar archive on stdout and then, in another directory, extracting the tar file from the piped stdin. This is one way to copy directories and subdirectories, even if the directories contain special files, such as symlinks, and character or block devices. Example Tarpipe:

 tar -cf - "$srcdir" | tar -C "$destdir" -xvf -

Many tar implementations (including GNU tar) do not record extended attributes (xattrs) or ACLs, so a more precise copy of a directory can be created with (GNU) `cp -a` or rsync. Tarpipes are inefficient because of the overhead of the second tar invocation and the pipe (which is especially expensive on DOS/Windows, as it is buffered to a file). Also, pipes may cause problems in automated scripts, as the success or failure of each command cannot be verified easily.

Software distribution

The tar format however continues to be used extensively, as it is part of the GNU project and comes as part of every distribution of GNU/Linux. Its use features prominently in software distribution in the Linux world, with most software source code made available in .tar.gz (compressed tar archives), and its use featuring as a basis of other types of containment and distribution, such as in Debian's .DEB archives, which are also used in Ubuntu.

The most common publication format for the released source code of free software is a tarball. For example, the GNU Project published a version of Emacs as "emacs-22.3.tar.gz" in September 2008. This use of the common tar format (and publication of the tarball on the Internet via http or ftp) masks the varying details of each project's source code revision control system.

Problems and limitations

The tar format was created in the early days of UNIX, and despite current widespread use, many of its design features are considered dated.[1] A new replacement format has been proposed by the people responsible for the rdiff-backup utility, which aims to address these limitations of the tar format, a discussion of which can be found at the Duplicity branch of the Savannah software site.

A tarbomb is derogatory hacker slang used to refer to a tarball that does not follow the usual conventions, i.e. it contains many files that extract into the working directory. Such a tarball can create problems by overwriting files of the same name in the working directory, or mixing one project's files into another. It is almost always an inconvenience to the user, who is obliged to identify and delete a number of files scattered throughout the directory's contents. Such behavior is considered bad etiquette on the part of the archive's creator.

A related problem is the use of absolute paths or parent directory references when creating tarballs. Files extracted from such tarballs will often be created in unusual locations outside the working directory, and like a tarbomb have the potential to overwrite existing files. GNU tar by default refuses to create or extract absolute paths, but is still vulnerable to parent-directory references.

A command line user can avoid both of these problems by first examining a tar file with a command like tar -tf archive.tar. This does not extract any files, but displays the names of all files in the archive. If any are problematic, the user can create a new empty directory and extract the tarball into it -- or avoid the tarball entirely. Users of modern graphical tools can just view the contents of the archive before extracting them.

Random access

Another weakness of the tar format compared to other archive formats is that there is no centralized location for the information about the contents of the file (a "table of contents" of sorts). So to list the names of the files that are in the archive, one must read through the entire archive and look for places where files start. Also, to extract one small file from the archive, instead of being able to lookup the offset in a table and go directly to that location, like other archive formats, with tar, one has to read through the entire archive, looking for the place where the desired file starts. For large tar archives, this causes a big performance penalty, making tar archives unsuitable for situations that often require random access of individual files. The possible reason for not using a centralized location of information is rooted in the fact that tar was originally meant for tapes (the name tar is derived from tape archive), which are bad at random access anyway: if the TOC were at the start of the archive, expanding the TOC would mean to copy the data following the TOC elsewhere — which may itself be a lengthy operation. On the other hand, if the TOC was at the end-of-file (as is the case with ZIP files, for example), reading the TOC would require that the tape be winded to the end, also taking up time and degrading the tape by excessive wear and tear.

A more modern reason for not upgrading the tar format with the inclusion of a TOC is that large TAR archives (those that would benefit most from a TOC) are typically compressed through gzip, bzip2 or similar as a whole. Extracting only a few files from the archive would then necessitate reading through the whole compressed archive, even if one knew the offset inside the (uncompressed) TAR where the file begins. The reason for this is that it is not generally possible to request from conventional compression programs that only a portion of a single compressed file be decompressed: From the point of view of the compression program, the original TAR archive is a single file.

See also

References

  • "tar(5): format of tape archive files". FreeBSD Manual Pages. FreeBSD Project. Includes documentation on how different implementations store various types of information and specialize headers.
  • "The tar Command". The Linux Information Project (LINFO).
  • "Detailed information on Tar and USTAR file headers".
  • "tar(1) man page". OpenBSD Manual Pages. OpenBSD Project.
  • "Tar — GNU Project". Free Software Foundation (FSF).