LIBARCHIVE(3) BSD Library Functions Manual LIBARCHIVE(3)NAME
libarchive — functions for reading and writing streaming archives
The libarchive library provides a flexible interface for reading and
writing streaming archive files such as tar and cpio. The library is
inherently stream-oriented; readers serially iterate through the archive,
writers serially add things to the archive. In particular, note that
there is no built-in support for random access nor for in-place modifica‐
When reading an archive, the library automatically detects the format and
the compression. The library currently has read support for:
· old-style tar archives,
· most variants of the POSIX “ustar” format,
· the POSIX “pax interchange” format,
· GNU-format tar archives,
· most common cpio archive formats,
· ISO9660 CD images (with or without RockRidge extensions),
· Zip archives.
The library automatically detects archives compressed with gzip(1),
bzip2(1), or compress(1) and decompresses them transparently.
When writing an archive, you can specify the compression to be used and
the format to use. The library can write
· POSIX-standard “ustar” archives,
· POSIX “pax interchange format” archives,
· POSIX octet-oriented cpio archives,
· two different variants of shar archives.
Pax interchange format is an extension of the tar archive format that
eliminates essentially all of the limitations of historic tar formats in
a standard fashion that is supported by POSIX-compliant pax(1) implemen‐
tations on many systems as well as several newer implementations of
tar(1). Note that the default write format will suppress the pax
extended attributes for most entries; explicitly requesting pax format
will enable those attributes for all entries.
The read and write APIs are accessed through the archive_read_XXX() func‐
tions and the archive_write_XXX() functions, respectively, and either can
be used independently of the other.
The rest of this manual page provides an overview of the library opera‐
tion. More detailed information can be found in the individual manual
pages for each API or utility function.
READING AN ARCHIVE
To read an archive, you must first obtain an initialized struct archive
object from archive_read_new(). You can then modify this object for the
desired operations with the various archive_read_set_XXX() and
archive_read_support_XXX() functions. In particular, you will need to
invoke appropriate archive_read_support_XXX() functions to enable the
corresponding compression and format support. Note that these latter
functions perform two distinct operations: they cause the corresponding
support code to be linked into your program, and they enable the corre‐
sponding auto-detect code. Unless you have specific constraints, you
will generally want to invoke archive_read_support_compression_all() and
archive_read_support_format_all() to enable auto-detect for all formats
and compression types currently supported by the library.
Once you have prepared the struct archive object, you call
archive_read_open() to actually open the archive and prepare it for read‐
ing. There are several variants of this function; the most basic expects
you to provide pointers to several functions that can provide blocks of
bytes from the archive. There are convenience forms that allow you to
specify a filename, file descriptor, FILE * object, or a block of memory
from which to read the archive data. Note that the core library makes no
assumptions about the size of the blocks read; callback functions are
free to read whatever block size is most appropriate for the medium.
Each archive entry consists of a header followed by a certain amount of
data. You can obtain the next header with archive_read_next_header(),
which returns a pointer to a struct archive_entry structure with informa‐
tion about the current archive element. If the entry is a regular file,
then the header will be followed by the file data. You can use
archive_read_data() (which works much like the read(2) system call) to
read this data from the archive. You may prefer to use the higher-level
archive_read_data_skip(), which reads and discards the data for this
entry, archive_read_data_to_buffer(), which reads the data into an in-
memory buffer, archive_read_data_to_file(), which copies the data to the
provided file descriptor, or archive_read_extract(), which recreates the
specified entry on disk and copies data from the archive. In particular,
note that archive_read_extract() uses the struct archive_entry structure
that you provide it, which may differ from the entry just read from the
archive. In particular, many applications will want to override the
pathname, file permissions, or ownership.
Once you have finished reading data from the archive, you should call
archive_read_close() to close the archive, then call
archive_read_finish() to release all resources, including all memory
allocated by the library.
The archive_read(3) manual page provides more detailed calling informa‐
tion for this API.
WRITING AN ARCHIVE
You use a similar process to write an archive. The archive_write_new()
function creates an archive object useful for writing, the various
archive_write_set_XXX() functions are used to set parameters for writing
the archive, and archive_write_open() completes the setup and opens the
archive for writing.
Individual archive entries are written in a three-step process: You first
initialize a struct archive_entry structure with information about the
new entry. At a minimum, you should set the pathname of the entry and
provide a struct stat with a valid st_mode field, which specifies the
type of object and st_size field, which specifies the size of the data
portion of the object. The archive_write_header() function actually
writes the header data to the archive. You can then use
archive_write_data() to write the actual data.
After all entries have been written, use the archive_write_finish() func‐
tion to release all resources.
The archive_write(3) manual page provides more detailed calling informa‐
tion for this API.
Detailed descriptions of each function are provided by the corresponding
All of the functions utilize an opaque struct archive datatype that pro‐
vides access to the archive contents.
The struct archive_entry structure contains a complete description of a
single archive entry. It uses an opaque interface that is fully docu‐
mented in archive_entry(3).
Users familiar with historic formats should be aware that the newer vari‐
ants have eliminated most restrictions on the length of textual fields.
Clients should not assume that filenames, link names, user names, or
group names are limited in length. In particular, pax interchange format
can easily accommodate pathnames in arbitrary character sets that exceed
Most functions return zero on success, non-zero on error. The return
value indicates the general severity of the error, ranging from
ARCHIVE_WARN, which indicates a minor problem that should probably be
reported to the user, to ARCHIVE_FATAL, which indicates a serious problem
that will prevent any further operations on this archive. On error, the
archive_errno() function can be used to retrieve a numeric error code
(see errno(2)). The archive_error_string() returns a textual error mes‐
sage suitable for display.
archive_read_new() and archive_write_new() return pointers to an allo‐
cated and initialized struct archive object.
archive_read_data() and archive_write_data() return a count of the number
of bytes actually read or written. A value of zero indicates the end of
the data for this entry. A negative value indicates an error, in which
case the archive_errno() and archive_error_string() functions can be used
to obtain more information.
There are character set conversions within the archive_entry(3) functions
that are impacted by the currently-selected locale.
SEE ALSOtar(1), archive_entry(3), archive_read(3), archive_util(3),
The libarchive library first appeared in FreeBSD 5.3.
The libarchive library was written by Tim Kientzle ⟨firstname.lastname@example.org⟩.
Some archive formats support information that is not supported by struct
archive_entry. Such information cannot be fully archived or restored
using this library. This includes, for example, comments, character
sets, or the arbitrary key/value pairs that can appear in pax interchange
Conversely, of course, not all of the information that can be stored in a
struct archive_entry is supported by all formats. For example, cpio for‐
mats do not support nanosecond timestamps; old tar formats do not support
large device numbers.
BSD July 17, 2010 BSD