path: root/doc/tarlz.info

This is tarlz.info, produced by makeinfo version 4.13+ from tarlz.texi.

INFO-DIR-SECTION Archiving
START-INFO-DIR-ENTRY
* Tarlz: (tarlz).               Archiver with multimember lzip compression
END-INFO-DIR-ENTRY


File: tarlz.info,  Node: Top,  Next: Introduction,  Up: (dir)

Tarlz Manual
************

This manual is for Tarlz (version 0.25, 3 January 2024).

* Menu:

* Introduction::              Purpose and features of tarlz
* Invoking tarlz::            Command-line interface
* Portable character set::    POSIX portable filename character set
* File format::               Detailed format of the compressed archive
* Amendments to pax format::  The reasons for the differences with pax
* Program design::            Internal structure of tarlz
* Multi-threaded decoding::   Limitations of parallel tar decoding
* Minimum archive sizes::     Sizes required for full multi-threaded speed
* Examples::                  A small tutorial with examples
* Problems::                  Reporting bugs
* Concept index::             Index of concepts


   Copyright (C) 2013-2024 Antonio Diaz Diaz.

   This manual is free documentation: you have unlimited permission to copy,
distribute, and modify it.


File: tarlz.info,  Node: Introduction,  Next: Invoking tarlz,  Prev: Top,  Up: Top

1 Introduction
**************

Tarlz is a massively parallel (multi-threaded) combined implementation of
the tar archiver and the lzip compressor. Tarlz uses the compression
library lzlib.

   Tarlz creates tar archives using a simplified and safer variant of the
POSIX pax format compressed in lzip format, keeping the alignment between
tar members and lzip members. The resulting multimember tar.lz archive is
backward compatible with standard tar tools like GNU tar, which treat it
like any other tar.lz archive. Tarlz can append files to the end of such
compressed archives.

   Keeping the alignment between tar members and lzip members has two
advantages. It adds an indexed lzip layer on top of the tar archive, making
it possible to decode the archive safely in parallel. It also minimizes the
amount of data lost in case of corruption. Compressing a tar archive with
plzip may even double the amount of files lost for each lzip member damaged
because it does not keep the members aligned.

   Tarlz can create tar archives with five levels of compression
granularity: per file ('--no-solid'), per block ('--bsolid', default), per
directory ('--dsolid'), appendable solid ('--asolid'), and solid
('--solid'). It can also create uncompressed tar archives.

Of course, compressing each file (or each directory) individually can't
achieve a compression ratio as high as compressing solidly the whole tar
archive, but it has the following advantages:

   * The resulting multimember tar.lz archive can be decompressed in
     parallel, multiplying the decompression speed.

   * New members can be appended to the archive (by removing the
     end-of-archive member), and unwanted members can be deleted from the
     archive. Just like an uncompressed tar archive.

   * It is a safe POSIX-style backup format. In case of corruption, tarlz
     can extract all the undamaged members from the tar.lz archive,
     skipping over the damaged members, just like the standard
     (uncompressed) tar. Moreover, the option '--keep-damaged' can be used
     to recover as much data as possible from each damaged member, and
     lziprecover can be used to recover some of the damaged members.

   * A multimember tar.lz archive is usually smaller than the corresponding
     solidly compressed tar.gz archive, except when individually
     compressing files smaller than about 32 KiB.

   Tarlz protects the extended records with a Cyclic Redundancy Check (CRC)
in a way compatible with standard tar tools. *Note crc32::.

   Tarlz does not understand other tar formats like 'gnu', 'oldgnu',
'star', or 'v7'. The command 'tarlz -t -f archive.tar.lz > /dev/null' can
be used to check that the format of the archive is compatible with tarlz.


File: tarlz.info,  Node: Invoking tarlz,  Next: Portable character set,  Prev: Introduction,  Up: Top

2 Invoking tarlz
****************

The format for running tarlz is:

     tarlz OPERATION [OPTIONS] [FILES]

All operations except '--concatenate' and '--compress' operate on whole
trees if any FILE is a directory. All operations except '--compress'
overwrite output files without warning. If no archive is specified, tarlz
tries to read it from standard input or write it to standard output. Tarlz
refuses to read archive data from a terminal or write archive data to a
terminal. Tarlz detects when the archive being created or enlarged is among
the files to be archived, appended, or concatenated, and skips it.

   Tarlz does not use absolute file names nor file names above the current
working directory (perhaps changed by option '-C'). On archive creation or
appending tarlz archives the files specified, but removes from member names
any leading and trailing slashes and any file name prefixes containing a
'..' component. On extraction, leading and trailing slashes are also
removed from member names, and archive members containing a '..' component
in the file name are skipped. Tarlz does not follow symbolic links during
extraction; not even symbolic links replacing intermediate directories.

   On extraction and listing, tarlz removes leading './' strings from
member names in the archive or given in the command line, so that
'tarlz -xf foo ./bar baz' extracts members 'bar' and './baz' from archive
'foo'.

   If several compression levels or '--*solid' options are given, the last
setting is used. For example '-9 --solid --uncompressed -1' is equivalent
to '-1 --solid'.

   tarlz supports the following operations:

'--help'
     Print an informative help message describing the options and exit.

'-V'
'--version'
     Print the version number of tarlz on the standard output and exit.
     This version number should be included in all bug reports.

'-A'
'--concatenate'
     Append one or more archives to the end of an archive. If no archive is
     specified with the option '-f', concatenate the input archives to
     standard output. All the archives involved must be regular (seekable)
     files, and must be either all compressed or all uncompressed.
     Compressed and uncompressed archives can't be mixed. Compressed
     archives must be multimember lzip files with the two end-of-archive
     blocks plus any zero padding contained in the last lzip member of each
     archive. The intermediate end-of-archive blocks are removed as each
     new archive is concatenated. If the archive is uncompressed, tarlz
     parses tar headers until it finds the end-of-archive blocks. Exit with
     status 0 without modifying the archive if no FILES have been specified.

     Concatenating archives containing files in common results in two or
     more tar members with the same name in the resulting archive, which
     may produce nondeterministic behavior during multi-threaded extraction.
     *Note mt-extraction::.

'-c'
'--create'
     Create a new archive from FILES.

'-d'
'--diff'
     Compare and report differences between archive and file system. For
     each tar member in the archive, check that the corresponding file in
     the file system exists and is of the same type (regular file,
     directory, etc). Report on standard output the differences found in
     type, mode (permissions), owner and group IDs, modification time, file
     size, file contents (of regular files), target (of symlinks) and
     device number (of block/character special files).

     As tarlz removes leading slashes from member names, the option '-C' may
     be used in combination with '--diff' when absolute file names were used
     on archive creation: 'tarlz -C / -d'. Alternatively, tarlz may be run
     from the root directory to perform the comparison.

'--delete'
     Delete files and directories from an archive in place. It currently can
     delete only from uncompressed archives and from archives with files
     compressed individually ('--no-solid' archives). Note that files of
     about '--data-size' or larger are compressed individually even if
     '--bsolid' is used, and can therefore be deleted. Tarlz takes care to
     not delete a tar member unless it is possible to do so. For example it
     won't try to delete a tar member that is not compressed individually.
     Even in the case of finding a corrupt member after having deleted some
     member(s), tarlz stops and copies the rest of the file as soon as
     corruption is found, leaving it just as corrupt as it was, but not
     worse.

     To delete a directory without deleting the files under it, use
     'tarlz --delete -f foo --exclude='dir/*' dir'. Deleting in place may
     be dangerous. A corrupt archive, a power cut, or an I/O error may cause
     data loss.

'-r'
'--append'
     Append files to the end of an archive. The archive must be a regular
     (seekable) file either compressed or uncompressed. Compressed members
     can't be appended to an uncompressed archive, nor vice versa. If the
     archive is compressed, it must be a multimember lzip file with the two
     end-of-archive blocks plus any zero padding contained in the last lzip
     member of the archive. It is possible to append files to an archive
     with a different compression granularity. Appending works as follows;
     first the end-of-archive blocks are removed, then the new members are
     appended, and finally two new end-of-archive blocks are appended to
     the archive. If the archive is uncompressed, tarlz parses and skips
     tar headers until it finds the end-of-archive blocks. Exit with status
     0 without modifying the archive if no FILES have been specified.

     Appending files already present in the archive results in two or more
     tar members with the same name, which may produce nondeterministic
     behavior during multi-threaded extraction. *Note mt-extraction::.

'-t'
'--list'
     List the contents of an archive. If FILES are given, list only the
     FILES given.

'-x'
'--extract'
     Extract files from an archive. If FILES are given, extract only the
     FILES given. Else extract all the files in the archive. To extract a
     directory without extracting the files under it, use
     'tarlz -xf foo --exclude='dir/*' dir'. Tarlz removes files and empty
     directories unconditionally before extracting over them. Other than
     that, it does not make any special effort to extract a file over an
     incompatible type of file. For example, extracting a file over a
     non-empty directory usually fails.

'-z'
'--compress'
     Compress existing POSIX tar archives aligning the lzip members to the
     tar members with choice of granularity ('--bsolid' by default,
     '--dsolid' works like '--asolid'). Exit with error status 2 if any
     input archive is an empty file. The input archives are kept unchanged.
     Existing compressed archives are not overwritten. A hyphen '-' used as
     the name of an input archive reads from standard input and writes to
     standard output (unless the option '--output' is used). Tarlz can be
     used as compressor for GNU tar by using a command like
     'tar -c -Hustar foo | tarlz -z -o foo.tar.lz'. Tarlz can be used as
     compressor for zupdate (zutils) by using a command like
     'zupdate --lz="tarlz -z" foo.tar.gz'. Note that tarlz only works
     reliably on archives without global headers, or with global headers
     whose content can be ignored.

     The compression is reversible, including any garbage present after the
     end-of-archive blocks. Tarlz stops parsing after the first
     end-of-archive block is found, and then compresses the rest of the
     archive. Unless solid compression is requested, the end-of-archive
     blocks are compressed in a lzip member separated from the preceding
     members and from any non-zero garbage following the end-of-archive
     blocks. '--compress' implies plzip argument style, not tar style. Each
     input archive is compressed to a file with the extension '.lz' added
     unless the option '--output' is used. When '--output' is used, only
     one input archive can be specified. '-f' can't be used with
     '--compress'.

'--check-lib'
     Compare the version of lzlib used to compile tarlz with the version
     actually being used at run time and exit. Report any differences
     found. Exit with error status 1 if differences are found. A mismatch
     may indicate that lzlib is not correctly installed or that a different
     version of lzlib has been installed after compiling tarlz. Exit with
     error status 2 if LZ_API_VERSION and LZ_version_string don't match.
     'tarlz -v --check-lib' shows the version of lzlib being used and the
     value of LZ_API_VERSION (if defined). *Note Library version:
     (lzlib)Library version.


   tarlz supports the following options: *Note Argument syntax:
(arg_parser)Argument syntax.

'-B BYTES'
'--data-size=BYTES'
     Set target size of input data blocks for the option '--bsolid'. *Note
     --bsolid::. Valid values range from 8 KiB to 1 GiB. Default value is
     two times the dictionary size, except for option '-0' where it
     defaults to 1 MiB. *Note Minimum archive sizes::.

'-C DIR'
'--directory=DIR'
     Change to directory DIR. When creating, appending, comparing, or
     extracting, the position of each '-C' option in the command line is
     significant; it changes the current working directory for the following
     FILES until a new '-C' option appears in the command line. '--list'
     and '--delete' ignore any '-C' options specified. DIR is relative to
     the then current working directory, perhaps changed by a previous '-C'
     option.

     Note that a process can only have one current working directory (CWD).
     Therefore multi-threading can't be used to create or decode an archive
     if a '-C' option appears after a (relative) file name in the command
     line. (All file names are made relative when decoding).

'-f ARCHIVE'
'--file=ARCHIVE'
     Use archive file ARCHIVE. A hyphen '-' used as an ARCHIVE argument
     reads from standard input or writes to standard output.

'-h'
'--dereference'
     Follow symbolic links during archive creation, appending or comparison.
     Archive or compare the files they point to instead of the links
     themselves.

'-n N'
'--threads=N'
     Set the number of (de)compression threads, overriding the system's
     default. Valid values range from 0 to "as many as your system can
     support". A value of 0 disables threads entirely. If this option is
     not used, tarlz tries to detect the number of processors in the system
     and use it as default value. 'tarlz --help' shows the system's default
     value. See the note about multi-threading in the option '-C' above.

     Note that the number of usable threads is limited during compression to
     ceil( uncompressed_size / data_size ) (*note Minimum archive sizes::),
     and during decompression to the number of lzip members in the tar.lz
     archive, which you can find by running 'lzip -lv archive.tar.lz'.

'-o FILE'
'--output=FILE'
     Write the compressed output to FILE. '-o -' writes the compressed
     output to standard output. Currently '--output' only works with
     '--compress'.

'-p'
'--preserve-permissions'
     On extraction, set file permissions as they appear in the archive.
     This is the default behavior when tarlz is run by the superuser. The
     default for other users is to subtract the umask of the user running
     tarlz from the permissions specified in the archive.

'-q'
'--quiet'
     Quiet operation. Suppress all messages.

'-v'
'--verbose'
     Verbosely list files processed. Further -v's (up to 4) increase the
     verbosity level.

'-0 .. -9'
     Set the compression level for '--create', '--append', and
     '--compress'. The default compression level is '-6'. Like lzip, tarlz
     also minimizes the dictionary size of the lzip members it creates,
     reducing the amount of memory required for decompression.

     Level   Dictionary size   Match length limit
     -0      64 KiB            16 bytes
     -1      1 MiB             5 bytes
     -2      1.5 MiB           6 bytes
     -3      2 MiB             8 bytes
     -4      3 MiB             12 bytes
     -5      4 MiB             20 bytes
     -6      8 MiB             36 bytes
     -7      16 MiB            68 bytes
     -8      24 MiB            132 bytes
     -9      32 MiB            273 bytes

'--uncompressed'
     With '--create', don't compress the tar archive created. Create an
     uncompressed tar archive instead. With '--append', don't compress the
     new members appended to the tar archive. Compressed members can't be
     appended to an uncompressed archive, nor vice versa. '--uncompressed'
     can be omitted if it can be deduced from the archive name. (An
     uncompressed archive name lacks a '.lz' or '.tlz' extension).

'--asolid'
     When creating or appending to a compressed archive, use appendable
     solid compression. All the files being added to the archive are
     compressed into a single lzip member, but the end-of-archive blocks
     are compressed into a separate lzip member. This creates a solidly
     compressed appendable archive. Solid archives can't be created nor
     decoded in parallel.

'--bsolid'
     When creating or appending to a compressed archive, use block
     compression. Tar members are compressed together in a lzip member
     until they approximate a target uncompressed size. The size can't be
     exact because each solidly compressed data block must contain an
     integer number of tar members. Block compression is the default
     because it improves compression ratio for archives with many files
     smaller than the block size. This option allows tarlz revert to
     default behavior if, for example, it is invoked through an alias like
     'tar='tarlz --solid''. *Note --data-size::, to set the target block
     size.

'--dsolid'
     When creating or appending to a compressed archive, compress each file
     specified in the command line separately in its own lzip member, and
     use solid compression for each directory specified in the command
     line. The end-of-archive blocks are compressed into a separate lzip
     member. This creates a compressed appendable archive with a separate
     lzip member for each file or top-level directory specified.

'--no-solid'
     When creating or appending to a compressed archive, compress each file
     separately in its own lzip member. The end-of-archive blocks are
     compressed into a separate lzip member. This creates a compressed
     appendable archive with a lzip member for each file.

'--solid'
     When creating or appending to a compressed archive, use solid
     compression. The files being added to the archive, along with the
     end-of-archive blocks, are compressed into a single lzip member. The
     resulting archive is not appendable. No more files can be later
     appended to the archive. Solid archives can't be created nor decoded
     in parallel.

'--anonymous'
     Equivalent to '--owner=root --group=root'.

'--owner=OWNER'
     When creating or appending, use OWNER for files added to the archive.
     If OWNER is not a valid user name, it is decoded as a decimal numeric
     user ID.

'--group=GROUP'
     When creating or appending, use GROUP for files added to the archive.
     If GROUP is not a valid group name, it is decoded as a decimal numeric
     group ID.

'--exclude=PATTERN'
     Exclude files matching a shell pattern like '*.o'. A file is considered
     to match if any component of the file name matches. For example, '*.o'
     matches 'foo.o', 'foo.o/bar' and 'foo/bar.o'. If PATTERN contains a
     '/', it matches a corresponding '/' in the file name. For example,
     'foo/*.o' matches 'foo/bar.o'. Multiple '--exclude' options can be
     specified.

'--ignore-ids'
     Make '--diff' ignore differences in owner and group IDs. This option is
     useful when comparing an '--anonymous' archive.

'--ignore-metadata'
     Make '--diff' ignore any differences in metadata (file permissions,
     owner and group IDs, modification time). Compare only file type, file
     size, and file content. This option is useful when file permissions
     have not been fully restored because uid/gid changed on extraction.

'--ignore-overflow'
     Make '--diff' ignore differences in mtime caused by overflow on 32-bit
     systems with a 32-bit time_t.

'--keep-damaged'
     Don't delete partially extracted files. If a decompression error
     happens while extracting a file, keep the partial data extracted. Use
     this option to recover as much data as possible from each damaged
     member. It is recommended to run tarlz in single-threaded mode
     ('--threads=0') when using this option.

'--missing-crc'
     Exit with error status 2 if the CRC of the extended records is
     missing. When this option is used, tarlz detects any corruption in the
     extended records (only limited by CRC collisions). But note that a
     corrupt 'GNU.crc32' keyword, for example 'GNU.crc30', is reported as a
     missing CRC instead of as a corrupt record. This misleading
     'Missing CRC' message is the consequence of a flaw in the POSIX pax
     format; i.e., the lack of a mandatory check sequence of the extended
     records. *Note crc32::.

'--mtime=DATE'
     When creating or appending, use DATE as the modification time for
     files added to the archive instead of their actual modification times.
     The value of DATE may be either '@' followed by the number of seconds
     since (or before) the epoch, or a date in format
     '[-]YYYY-MM-DD HH:MM:SS' or '[-]YYYY-MM-DDTHH:MM:SS', or the name of
     an existing reference file starting with '.' or '/' whose modification
     time is used. The time of day 'HH:MM:SS' in the date format is
     optional and defaults to '00:00:00'. The epoch is
     '1970-01-01 00:00:00 UTC'. Negative seconds or years define a
     modification time before the epoch.

'--out-slots=N'
     Number of 1 MiB output packets buffered per worker thread during
     multi-threaded creation or appending to compressed archives.
     Increasing the number of packets may increase compression speed if the
     files being archived are larger than 64 MiB compressed, but requires
     more memory. Valid values range from 1 to 1024. The default value is
     64.

'--warn-newer'
     During archive creation, warn if any file being archived has a
     modification time newer than the archive creation time. This option
     may slow archive creation somewhat because it makes an extra call to
     'stat' after archiving each file, but it guarantees that file contents
     were not modified during the creation of the archive. Note that the
     file must be at least one second newer than the archive for it to be
     detected as newer.


   Exit status: 0 for a normal exit, 1 for environmental problems (file not
found, files differ, invalid command-line options, I/O errors, etc), 2 to
indicate a corrupt or invalid input file, 3 for an internal consistency
error (e.g., bug) which caused tarlz to panic.


File: tarlz.info,  Node: Portable character set,  Next: File format,  Prev: Invoking tarlz,  Up: Top

3 POSIX portable filename character set
***************************************

The set of characters from which portable file names are constructed.

     A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
     a b c d e f g h i j k l m n o p q r s t u v w x y z
     0 1 2 3 4 5 6 7 8 9 . _ -

   The last three characters are the period, underscore, and hyphen-minus
characters, respectively.

   File names are identifiers. Therefore, archiving works better when file
names use only the portable character set without spaces added.


File: tarlz.info,  Node: File format,  Next: Amendments to pax format,  Prev: Portable character set,  Up: Top

4 File format
*************

In the diagram below, a box like this:

+---+
|   | <-- the vertical bars might be missing
+---+

   represents one byte; a box like this:

+==============+
|              |
+==============+

   represents a variable number of bytes or a fixed but large number of
bytes (for example 512).


   A tar.lz file consists of one or more lzip members (compressed data
sets). The members simply appear one after another in the file, with no
additional information before, between, or after them.

   Each lzip member contains one or more tar members in a simplified POSIX
pax interchange format. The only pax typeflag value supported by tarlz (in
addition to the typeflag values defined by the ustar format) is 'x'. The
pax format is an extension on top of the ustar format that removes the size
limitations of the ustar format.

   Each tar member contains one file archived, and is represented by the
following sequence:

   * An optional extended header block followed by one or more blocks that
     contain the extended header records as if they were the contents of a
     file; i.e., the extended header records are included as the data for
     this header block. This header block is of the form described in pax
     header block, with a typeflag value of 'x'.

   * A header block in ustar format that describes the file. Any fields
     defined in the preceding optional extended header records override the
     associated fields in this header block for this file.

   * Zero or more blocks that contain the contents of the file.

   Each tar member must be contiguously stored in a lzip member for the
parallel decoding operations like '--list' to work. If any tar member is
split over two or more lzip members, the archive must be decoded
sequentially. *Note Multi-threaded decoding::.

   At the end of the archive file there are two 512-byte blocks filled with
binary zeros, interpreted as an end-of-archive indicator. These EOA blocks
are either compressed in a separate lzip member or compressed along with the
tar members contained in the last lzip member. For a compressed archive to
be recognized by tarlz as appendable, the last lzip member must contain
between 512 and 32256 zeros alone (without any non-zero bytes).

   The diagram below shows the correspondence between each tar member
(formed by one or two headers plus optional data) in the tar archive and
each lzip member in the resulting multimember tar.lz archive, when per file
compression is used: *Note File format: (lzip)File format.

tar
+========+======+=================+===============+========+======+========+
| header | data | extended header | extended data | header | data |   EOA  |
+========+======+=================+===============+========+======+========+

tar.lz
+===============+=================================================+========+
|     member    |                      member                     | member |
+===============+=================================================+========+


4.1 Pax header block
====================

The pax header block is identical to the ustar header block described below
except that the typeflag has the value 'x' (extended). The field 'size' is
the size of the extended header data in bytes. Most other fields in the pax
header block are zeroed on archive creation to prevent trouble if the
archive is read by an ustar tool, and are ignored by tarlz on archive
extraction. *Note flawed-compat::.

   Tarlz limits the size of the pax extended header data so that the whole
header set (extended header + extended data + ustar header) can be read and
decoded in a buffer of size INT_MAX.

   The pax extended header data consists of one or more records, each of
them constructed as follows:
'"%d %s=%s\n", <length>, <keyword>, <value>'

   The fields <length> and <keyword> in the record must be limited to the
portable character set (*note Portable character set::). The field <length>
contains the decimal length of the record in bytes, including the trailing
newline. The field <value> is stored as-is, without conversion to UTF-8 nor
any other transformation. The fields are separated by the ASCII characters
space, equal-sign, and newline.

   These are the <keyword> values currently supported by tarlz:

'atime'
     The signed decimal representation of the access time of the following
     file in seconds since (or before) the epoch, obtained from the function
     'stat'. The atime record is created only for files with a modification
     time outside of the ustar range. *Note ustar-mtime::.

'gid'
     The unsigned decimal representation of the group ID of the group that
     owns the following file. The gid record is created only for files with
     a group ID greater than 2_097_151 (octal 7_777_777). *Note
     ustar-uid-gid::.

'linkpath'
     The file name of a link being created to another file, of any type,
     previously archived. This record overrides the field 'linkname' in the
     following ustar header block. The following ustar header block
     determines the type of link created. If typeflag of the following
     header block is 1, a hard link is created. If typeflag is 2, a
     symbolic link is created and the linkpath value is used as the
     contents of the symbolic link. The linkpath record is created only for
     links with a link name that does not fit in the space provided by the
     ustar header.

'mtime'
     The signed decimal representation of the modification time of the
     following file in seconds since (or before) the epoch, obtained from
     the function 'stat'. This record overrides the field 'mtime' in the
     following ustar header block. The mtime record is created only for
     files with a modification time outside of the ustar range. *Note
     ustar-mtime::.

'path'
     The file name of the following file. This record overrides the fields
     'name' and 'prefix' in the following ustar header block. The path
     record is created for files with a name that does not fit in the space
     provided by the ustar header, but is also created for files that
     require any other extended record so that the fields 'name' and
     'prefix' in the following ustar header block can be zeroed.

'size'
     The size of the file in bytes, expressed as a decimal number using
     digits from the ISO/IEC 646:1991 (ASCII) standard. This record
     overrides the field 'size' in the following ustar header block. The
     size record is created only for files with a size value greater than
     8_589_934_591 (octal 77_777_777_777); that is, 8 GiB (2^33 bytes) or
     larger.

'uid'
     The unsigned decimal representation of the user ID of the file owner
     of the following file. The uid record is created only for files with a
     user ID greater than 2_097_151 (octal 7_777_777). *Note
     ustar-uid-gid::.

'GNU.crc32'
     CRC32-C (Castagnoli) of the extended header data excluding the 8 bytes
     representing the CRC <value> itself. The <value> is represented as 8
     hexadecimal digits in big endian order, '22 GNU.crc32=00000000\n'. The
     keyword of the CRC record is protected by the CRC to guarantee that
     corruption is always detected when using '--missing-crc' (except in
     case of CRC collision). A CRC was chosen because a checksum is too
     weak for a potentially large list of variable sized records. A
     checksum can't detect simple errors like the swapping of two bytes.


   At verbosity level 1 or higher tarlz prints a diagnostic for each unknown
extended header keyword found in an archive, once per keyword.


4.2 Ustar header block
======================

The ustar header block has a length of 512 bytes and is structured as shown
in the following table. All lengths and offsets are in decimal.

Field Name   Offset   Length (in bytes)
name         0        100
mode         100      8
uid          108      8
gid          116      8
size         124      12
mtime        136      12
chksum       148      8
typeflag     156      1
linkname     157      100
magic        257      6
version      263      2
uname        265      32
gname        297      32
devmajor     329      8
devminor     337      8
prefix       345      155

   All characters in the header block are coded using the ISO/IEC 646:1991
(ASCII) standard, except in fields storing names for files, users, and
groups. For maximum portability between implementations, names should only
contain characters from the portable character set (*note Portable
character set::), but if an implementation supports the use of characters
outside of '/' and the portable character set in names for files, users,
and groups, tarlz will use the byte values in these names unmodified.

   The fields 'name', 'linkname', and 'prefix' are null-terminated
character strings except when all characters in the array contain non-null
characters including the last character.

   The fields 'name' and 'prefix' produce the file name. A new file name is
formed, if prefix is not an empty string (its first character is not null),
by concatenating prefix (up to the first null character), a slash
character, and name; otherwise, name is used alone. In either case, name is
terminated at the first null character. If prefix begins with a null
character, it is ignored. In this manner, file names of at most 256
characters can be supported. If a file name does not fit in the space
provided, an extended record is used to store the file name.

   The field 'linkname' does not use the prefix to produce a file name. If
the link name does not fit in the 100 characters provided, an extended
record is used to store the link name.

   The field 'mode' provides 12 access permission bits. The following table
shows the symbolic name of each bit and its octal value:

Bit Name   Value   Bit Name   Value   Bit Name   Value
---------------------------------------------------
S_ISUID    04000   S_ISGID    02000   S_ISVTX    01000
S_IRUSR    00400   S_IWUSR    00200   S_IXUSR    00100
S_IRGRP    00040   S_IWGRP    00020   S_IXGRP    00010
S_IROTH    00004   S_IWOTH    00002   S_IXOTH    00001

   The fields 'uid' and 'gid' are the user and group IDs of the owner and
group of the file, respectively. If the file uid or gid are greater than
2_097_151 (octal 7_777_777), an extended record is used to store the uid or
gid.

   The field 'size' contains the octal representation of the size of the
file in bytes. If the field 'typeflag' specifies a file of type '0'
(regular file) or '7' (high performance regular file), the number of logical
records following the header is (size / 512) rounded to the next integer.
For all other values of typeflag, tarlz either sets the size field to 0 or
ignores it, and does not store or expect any logical records following the
header. If the file size is larger than 8_589_934_591 bytes
(octal 77_777_777_777), an extended record is used to store the file size.

   The field 'mtime' contains the octal representation of the modification
time of the file at the time it was archived, obtained from the function
'stat'. If the modification time is negative or larger than 8_589_934_591
(octal 77_777_777_777) seconds since the epoch, an extended record is used
to store the modification time. The ustar range of mtime goes from
'1970-01-01 00:00:00 UTC' to '2242-03-16 12:56:31 UTC'.

   The field 'chksum' contains the octal representation of the value of the
simple sum of all bytes in the header logical record. Each byte in the
header is treated as an unsigned value. When calculating the checksum, the
chksum field is treated as if it were all space characters.

   The field 'typeflag' contains a single character specifying the type of
file archived:

''0''
     Regular file.

''1''
     Hard link to another file, of any type, previously archived. Hard
     links must not contain file data.

''2''
     Symbolic link.

''3', '4''
     Character special file and block special file respectively. In this
     case the fields 'devmajor' and 'devminor' contain information defining
     the device in unspecified format.

''5''
     Directory.

''6''
     FIFO special file.

''7''
     Reserved to represent a file to which an implementation has associated
     some high-performance attribute (contiguous file). Tarlz treats this
     type of file as a regular file (type 0).


   The field 'magic' contains the ASCII null-terminated string "ustar". The
field 'version' contains the characters "00" (0x30,0x30). The fields
'uname' and 'gname' are null-terminated character strings except when all
characters in the array contain non-null characters including the last
character. Each numeric field contains a leading space- or zero-filled,
optionally null-terminated octal number using digits from the ISO/IEC
646:1991 (ASCII) standard. Tarlz is able to decode numeric fields 1 byte
longer than standard ustar by not requiring a terminating null character.


File: tarlz.info,  Node: Amendments to pax format,  Next: Program design,  Prev: File format,  Up: Top

5 The reasons for the differences with pax
******************************************

Tarlz creates safe archives that allow the reliable detection of invalid or
corrupt metadata during decoding even when the integrity checking of lzip
can't be used because the lzip members are only decompressed partially, as
it happens in parallel '--diff', '--list', and '--extract'. In order to
achieve this goal and avoid some other flaws in the pax format, tarlz makes
some changes to the variant of the pax format that it uses. This chapter
describes these changes and the concrete reasons to implement them.


5.1 Add a CRC of the extended records
=====================================

The POSIX pax format has a serious flaw. The metadata stored in pax extended
records are not protected by any kind of check sequence. Corruption in a
long file name may cause the extraction of the file in the wrong place
without warning. Corruption in a large file size may cause the truncation of
the file or the appending of garbage to the file, both followed by a
spurious warning about a corrupt header far from the place of the undetected
corruption.

   Metadata like file name and file size must be always protected in an
archive format because of the adverse effects of undetected corruption in
them, potentially much worse that undetected corruption in the data. Even
more so in the case of pax because the amount of metadata it stores is
potentially large, making undetected corruption and archiver misbehavior
more probable.

   Headers and metadata must be protected separately from data because the
integrity checking of lzip may not be able to detect the corruption before
the metadata have been used, for example, to create a new file in the wrong
place.

   Because of the above, tarlz protects the extended records with a Cyclic
Redundancy Check (CRC) in a way compatible with standard tar tools. *Note
key_crc32::.


5.2 Remove flawed backward compatibility
========================================

In order to allow the extraction of pax archives by a tar utility conforming
to the POSIX-2:1993 standard, POSIX.1-2008 recommends selecting extended
header field values that allow such tar to create a regular file containing
the extended header records as data. This approach is broken because if the
extended header is needed because of a long file name, the fields 'name'
and 'prefix' are unable to contain the full file name. (Some tar
implementations store the truncated name in the field 'name' alone,
truncating the name to only 100 bytes instead of 256). Therefore the files
corresponding to both the extended header and the overridden ustar header
are extracted using truncated file names, perhaps overwriting existing
files or directories. It may be a security risk to extract a file with a
truncated file name.

   To avoid this problem, tarlz writes extended headers with all fields
zeroed except 'size' (which contains the size of the extended records),
'chksum', 'typeflag', 'magic', and 'version'. In particular, tarlz sets the
fields 'name' and 'prefix' to zero. This prevents old tar programs from
extracting the extended records as a file in the wrong place. Tarlz also
sets to zero those fields of the ustar header overridden by extended
records. Finally, tarlz skips members with zeroed 'name' and 'prefix' when
decoding, except when listing. This is needed to detect certain format
violations during parallel extraction.

   If an extended header is required for any reason (for example a file
size of 8 GiB or larger, or a link name longer than 100 bytes), tarlz also
moves the file name to the extended records to prevent an ustar tool from
trying to extract the file or link. This also makes easier during parallel
decoding the detection of a tar member split between two lzip members at
the boundary between the extended header and the ustar header.


5.3 As simple as possible (but not simpler)
===========================================

The tarlz format is mainly ustar. Extended pax headers are used only when
needed because the length of a file name or link name, or the size or other
attribute of a file exceed the limits of the ustar format. Adding 1 KiB of
extended header and records to each member just to save subsecond
timestamps seems wasteful for a backup format. Moreover, minimizing the
overhead may help recovering the archive with lziprecover in case of
corruption.

   Global pax headers are tolerated, but not supported; they are parsed and
ignored. Some operations may not behave as expected if the archive contains
global headers.


5.4 Improve reproducibility
===========================

Pax includes by default the process ID of the pax process in the ustar name
of the extended headers, making the archive not reproducible. Tarlz stores
the true name of the file just once, either in the ustar header or in the
extended records, making it easier to produce reproducible archives.

   Pax allows an extended record to have length x-1 or x if x is a power of
ten; '99<97_bytes>' or '100<97_bytes>'. Tarlz minimizes the length of the
record and always produces a length of x-1 in these cases.


5.5 No data in hard links
=========================

Tarlz does not allow data in hard link members. The data (if any) must be in
the member determining the type of the file (which can't be a link). If all
the names of a file are stored as hard links, the type of the file is lost.
Not allowing data in hard links also prevents invalid actions like
extracting file data for a hard link to a symbolic link or to a directory.


5.6 Avoid misconversions to/from UTF-8
======================================

There is no portable way to tell what charset a text string is coded into.
Therefore, tarlz stores all fields representing text strings unmodified,
without conversion to UTF-8 nor any other transformation. This prevents
accidental double UTF-8 conversions. If the need arises this behavior will
be adjusted with a command-line option in the future.


File: tarlz.info,  Node: Program design,  Next: Multi-threaded decoding,  Prev: Amendments to pax format,  Up: Top

6 Internal structure of tarlz
*****************************

The parts of tarlz related to sequential processing of the archive are more
or less similar to any other tar and won't be described here. The
interesting parts described here are those related to Multi-threaded
processing.

   The structure of the part of tarlz performing Multi-threaded archive
creation is somewhat similar to that of plzip with the added complication
of the solidity levels. *Note Program design: (plzip)Program design. A
grouper thread and several worker threads are created, acting the main
thread as muxer (multiplexer) thread. A "packet courier" takes care of data
transfers among threads and limits the maximum number of data blocks
(packets) being processed simultaneously.

   The grouper traverses the directory tree, groups together the metadata of
the files to be archived in each lzip member, and distributes them to the
workers. The workers compress the metadata received from the grouper along
with the file data read from the file system. The muxer collects processed
packets from the workers, and writes them to the archive.

.--------.
|    data|---> to each worker below
|        |                    .------------.
| file   |                ,-->| worker   0 |--,
| system |                |   `------------'  |
|        |    .---------. |   .------------.  |   .-------.   .---------.
|metadata|--->| grouper |-+-->| worker   1 |--+-->| muxer |-->| archive |
`--------'    `---------' |   `------------'  |   `-------'   `---------'
                          |        ...        |
                          |   .------------.  |
                          `-->| worker N-1 |--'
                              `------------'

   Decoding an archive is somewhat similar to how plzip decompresses a
regular file to standard output, with the differences that it is not the
data but only messages what is written to stdout/stderr, and that each
worker may access files in the file system either to read them (diff) or
write them (extract). As in plzip, each worker reads members directly from
the archive.

.--------.
| file   |<---> data to/from each worker below
| system |
`--------'      .------------.
            ,-->| worker   0 |--,
            |   `------------'  |
.---------. |   .------------.  |   .-------.   .--------.
| archive |-+-->| worker   1 |--+-->| muxer |-->| stdout |
`---------' |   `------------'  |   `-------'   | stderr |
            |        ...        |               `--------'
            |   .------------.  |
            `-->| worker N-1 |--'
                `------------'

   As misaligned tar.lz archives can't be decoded in parallel, and the
misalignment can't be detected until after decoding has started, a
"mastership request" mechanism has been designed that allows the decoding to
continue instead of signalling an error.

   During parallel decoding, if a worker finds a misalignment, it requests
mastership to decode the rest of the archive. When mastership is requested,
an error_member_id is set, and all subsequently received packets with
member_id > error_member_id are rejected. All workers requesting mastership
are blocked at the request_mastership call until mastership is granted.
Mastership is granted to the delivering worker when its queue is empty to
make sure that all preceding packets have been processed. When mastership is
granted, all packets are deleted and all subsequently received packets not
coming from the master are rejected.

   If a worker can't continue decoding for any cause (for example lack of
memory or finding a split tar member at the beginning of a lzip member), it
requests mastership to print an error and terminate the program. Only if
some other worker requests mastership in a previous lzip member can this
error be avoided.


File: tarlz.info,  Node: Multi-threaded decoding,  Next: Minimum archive sizes,  Prev: Program design,  Up: Top

7 Limitations of parallel tar decoding
**************************************

Safely decoding an arbitrary tar archive in parallel is only possible if one
decodes the headers sequentially first. For example, if a tar archive
containing another tar archive is decoded starting from some position other
than the beginning, there is no way to know if the first header found there
belongs to the outer tar archive or to the inner tar archive. Tar is a
format inherently serial; it was designed for tapes.

   The pax format is even more serial than the ustar format. Two headers
need to be decoded sequentially for each file. The extended header may even
need parsing to reveal something as basic as file size. If a thread decodes
the ustar header skipping the preceding extended header, it may extract a
file of incorrect size at the wrong place. Moreover, a pax archive with
global headers can't be decoded in parallel because each thread can't know
about the global headers decoded by other threads.

   In the case of compressed tar archives, the start of each compressed
block determines one point through which the tar archive can be decoded in
parallel. Therefore, in tar.lz archives the decoding operations can't be
parallelized if the tar members are not aligned with the lzip members. Tar
archives compressed with plzip can't be decoded in parallel because tar and
plzip do not have a way to align both sets of members. Certainly one can
decompress one such archive with a multi-threaded tool like plzip, but the
increase in speed is not as large as it could be because plzip must
serialize the decompressed data and pass them to tar, which decodes them
sequentially, one tar member at a time.

   On the other hand, if the tar.lz archive is created with a tool like
tarlz, which can guarantee the alignment between tar members and lzip
members because it controls both archiving and compression, then the lzip
format becomes an indexed layer on top of the tar archive which makes
possible decoding it safely in parallel.

   Tarlz is able to automatically decode aligned and unaligned multimember
tar.lz archives, keeping backwards compatibility. If tarlz finds a member
misalignment during multi-threaded decoding, it switches to single-threaded
mode and continues decoding the archive.

   If the files in the archive are large, multi-threaded '--list' on a
regular (seekable) tar.lz archive can be hundreds of times faster than
sequential '--list' because, in addition to using several processors, it
only needs to decompress part of each lzip member. See the following
example listing the Silesia corpus on a dual core machine:

     tarlz -9 --no-solid -cf silesia.tar.lz silesia
     time lzip -cd silesia.tar.lz | tar -tf -            (5.032s)
     time plzip -cd silesia.tar.lz | tar -tf -           (3.256s)
     time tarlz -tf silesia.tar.lz                       (0.020s)

   On the other hand, multi-threaded '--list' won't detect corruption in
the tar member data because it only decodes the part of each lzip member
corresponding to the tar member header. This is another reason why the tar
headers must provide their own integrity checking.


7.1 Limitations of multi-threaded extraction
============================================

Multi-threaded extraction may produce different output than single-threaded
extraction in some cases:

   During multi-threaded extraction, several independent threads are
simultaneously reading the archive and creating files in the file system.
The archive is not read sequentially. As a consequence, any error or
weirdness in the archive (like a corrupt member or an end-of-archive block
in the middle of the archive) won't be usually detected until part of the
archive beyond that point has been processed.

   If the archive contains two or more tar members with the same name,
single-threaded extraction extracts the members in the order they appear in
the archive and leaves in the file system the last version of the file. But
multi-threaded extraction may extract the members in any order and leave in
the file system any version of the file nondeterministically. It is
unspecified which of the tar members is extracted.

   If the same file is extracted through several paths (different member
names resolve to the same file in the file system), the result is undefined.
(Probably the resulting file will be mangled).

   Extraction of a hard link may fail if it is extracted before the file it
links to.


File: tarlz.info,  Node: Minimum archive sizes,  Next: Examples,  Prev: Multi-threaded decoding,  Up: Top

8 Minimum archive sizes required for multi-threaded block compression
*********************************************************************

When creating or appending to a compressed archive using multi-threaded
block compression, tarlz puts tar members together in blocks and compresses
as many blocks simultaneously as worker threads are chosen, creating a
multimember compressed archive.

   For this to work as expected (and roughly multiply the compression speed
by the number of available processors), the uncompressed archive must be at
least as large as the number of worker threads times the block size (*note
--data-size::). Else some processors do not get any data to compress, and
compression is proportionally slower. The maximum speed increase achievable
on a given archive is limited by the ratio (uncompressed_size / data_size).
For example, a tarball the size of gcc or linux scales up to 10 or 14
processors at level -9.

   The following table shows the minimum uncompressed archive size needed
for full use of N processors at a given compression level, using the default
data size for each level:

Processors   2         4         8         16        64        256
------------------------------------------------------------------
Level                                                          
-0           2 MiB     4 MiB     8 MiB     16 MiB    64 MiB    256 MiB
-1           4 MiB     8 MiB     16 MiB    32 MiB    128 MiB   512 MiB
-2           6 MiB     12 MiB    24 MiB    48 MiB    192 MiB   768 MiB
-3           8 MiB     16 MiB    32 MiB    64 MiB    256 MiB   1 GiB
-4           12 MiB    24 MiB    48 MiB    96 MiB    384 MiB   1.5 GiB
-5           16 MiB    32 MiB    64 MiB    128 MiB   512 MiB   2 GiB
-6           32 MiB    64 MiB    128 MiB   256 MiB   1 GiB     4 GiB
-7           64 MiB    128 MiB   256 MiB   512 MiB   2 GiB     8 GiB
-8           96 MiB    192 MiB   384 MiB   768 MiB   3 GiB     12 GiB
-9           128 MiB   256 MiB   512 MiB   1 GiB     4 GiB     16 GiB


File: tarlz.info,  Node: Examples,  Next: Problems,  Prev: Minimum archive sizes,  Up: Top

9 A small tutorial with examples
********************************

Example 1: Create a multimember compressed archive 'archive.tar.lz'
containing files 'a', 'b' and 'c'.

     tarlz -cf archive.tar.lz a b c


Example 2: Append files 'd' and 'e' to the multimember compressed archive
'archive.tar.lz'.

     tarlz -rf archive.tar.lz d e


Example 3: Create a solidly compressed appendable archive 'archive.tar.lz'
containing files 'a', 'b' and 'c'. Then append files 'd' and 'e' to the
archive.

     tarlz --asolid -cf archive.tar.lz a b c
     tarlz --asolid -rf archive.tar.lz d e


Example 4: Create a compressed appendable archive containing directories
'dir1', 'dir2' and 'dir3' with a separate lzip member per directory. Then
append files 'a', 'b', 'c', 'd' and 'e' to the archive, all of them
contained in a single lzip member. The resulting archive 'archive.tar.lz'
contains 5 lzip members (including the end-of-archive member).

     tarlz --dsolid -cf archive.tar.lz dir1 dir2 dir3
     tarlz --asolid -rf archive.tar.lz a b c d e


Example 5: Create a solidly compressed archive 'archive.tar.lz' containing
files 'a', 'b' and 'c'. Note that no more files can be later appended to
the archive.

     tarlz --solid -cf archive.tar.lz a b c


Example 6: Extract all files from archive 'archive.tar.lz'.

     tarlz -xf archive.tar.lz


Example 7: Extract files 'a' and 'c', and the whole tree under directory
'dir1' from archive 'archive.tar.lz'.

     tarlz -xf archive.tar.lz a c dir1


Example 8: Copy the contents of directory 'sourcedir' to the directory
'destdir'.

     tarlz -C sourcedir --uncompressed -cf - . | tarlz -C destdir -xf -


Example 9: Compress the existing POSIX archive 'archive.tar' and write the
output to 'archive.tar.lz'. Compress each member individually for maximum
availability. (If one member in the compressed archive gets damaged, the
other members can still be extracted).

     tarlz -z --no-solid archive.tar


Example 10: Compress the archive 'archive.tar' and write the output to
'foo.tar.lz'.

     tarlz -z -o foo.tar.lz archive.tar


Example 11: Concatenate and compress two archives 'archive1.tar' and
'archive2.tar', and write the output to 'foo.tar.lz'.

     tarlz -A archive1.tar archive2.tar | tarlz -z -o foo.tar.lz


File: tarlz.info,  Node: Problems,  Next: Concept index,  Prev: Examples,  Up: Top

10 Reporting bugs
*****************

There are probably bugs in tarlz. There are certainly errors and omissions
in this manual. If you report them, they will get fixed. If you don't, no
one will ever know about them and they will remain unfixed for all
eternity, if not longer.

   If you find a bug in tarlz, please send electronic mail to
<lzip-bug@nongnu.org>. Include the version number, which you can find by
running 'tarlz --version' and 'tarlz -v --check-lib'.


File: tarlz.info,  Node: Concept index,  Prev: Problems,  Up: Top

Concept index
*************

[index]
* Menu:

* Amendments to pax format:              Amendments to pax format.  (line 6)
* bugs:                                  Problems.                  (line 6)
* examples:                              Examples.                  (line 6)
* file format:                           File format.               (line 6)
* getting help:                          Problems.                  (line 6)
* introduction:                          Introduction.              (line 6)
* invoking:                              Invoking tarlz.            (line 6)
* minimum archive sizes:                 Minimum archive sizes.     (line 6)
* options:                               Invoking tarlz.            (line 6)
* parallel tar decoding:                 Multi-threaded decoding.   (line 6)
* portable character set:                Portable character set.    (line 6)
* program design:                        Program design.            (line 6)
* usage:                                 Invoking tarlz.            (line 6)
* version:                               Invoking tarlz.            (line 6)



Tag Table:
Node: Top216
Node: Introduction1207
Node: Invoking tarlz4032
Ref: --data-size13076
Ref: --bsolid17512
Node: Portable character set23425
Node: File format24068
Ref: key_crc3231050
Ref: ustar-uid-gid34315
Ref: ustar-mtime35122
Node: Amendments to pax format37125
Ref: crc3237834
Ref: flawed-compat39146
Node: Program design43228
Node: Multi-threaded decoding47153
Ref: mt-extraction50434
Node: Minimum archive sizes51740
Node: Examples53867
Node: Problems56234
Node: Concept index56789

End Tag Table


Local Variables:
coding: iso-8859-15
End: