PAX(1)

🌈 pax(1) — General Commands Manual 🌈

okay so bestie, welcome to the pax page 💖 this is the main character of archive tools and honestly? she does it all. reading, writing, listing, copying whole directory trees — the range. grab a snack, we're going through the whole thing together, no skimming, no skipping 🫶

---

✨ NAME ✨

pax — read and write file archives and copy directory hierarchies 💅

---

✨ SYNOPSIS ✨

pax [-0cdJjnOvz] [-E limit] [-f archive] [-G group] [-s replstr]
    [-T range] [-U user] [pattern ...]
pax -r [-0cDdiJjknOuvYZz] [-E limit] [-f archive] [-G group] [-M flag]
    [-o options] [-p string] [-s replstr] [-T range] [-U user]
    [pattern ...]
pax -w [-0adHiJjLOPtuvXz] [-B bytes] [-b blocksize] [-f archive]
    [-G group] [-M flag] [-o options] [-s replstr] [-T range] [-U user]
    [-x format] [file ...]
pax -rw [-0DdHiJjkLlnOPtuvXYZ] [-G group] [-p string] [-s replstr]
    [-T range] [-U user] [file ...] directory

like, look at her. four whole moods and each one is serving 💅

---

🌸 DESCRIPTION 🌸

okay so here's the tea: pax will read, write, and list the members of an archive file, AND it will copy directory hierarchies. and the iconic part? her whole operation is independent of the specific archive format</> — she supports a wide variety of different archive formats. the full list of what she can handle lives down under the -x option, so hold that thought 💖

the way pax knows which mood she's in is by checking whether -r and -w are there. those two decide which of the following functional modes she runs: list, read, write, and copy.

• <none> — List mode. pax writes a table of contents of the members of the archive file read from standard input, straight to standard output. she only does this for members whose pathnames match the pattern arguments you gave her. the table of contents has one filename per line and it's written using single line buffering. very tidy, very demure ✨
• -r — Read mode. pax extracts the members of the archive file read from standard input, and again, only the ones whose pathnames match the pattern arguments. the archive format and blocking get automatically determined on input (she's psychic like that 🔮). when an extracted file is a directory, the entire file hierarchy rooted at that directory gets extracted too. all extracted files are created relative to the current file hierarchy. and the whole situation with ownership, access and modification times, and file mode of the extracted files? that's discussed in more detail under the -p option. patience bestie 💜
• -w — Write mode. pax writes an archive containing the file operands to standard output using the specified archive format. when you hand her no file operands, she reads a list of files to copy from standard input, one per line. and when a file operand is also a directory, the entire file hierarchy rooted at that directory will be included 😌
• -rw — Copy mode. pax copies the file operands to the destination directory. when no file operands are specified, she reads a list of files to copy from standard input, one per line. when a file operand is also a directory, the entire file hierarchy rooted at that directory will be included. and the effect of the copy is as if the copied files were written to an archive file and then subsequently extracted — except that there may be hard links between the original and the copied files (see the -l option below 👀).

⚠️ Warning: the destination directory must not be one of the file operands, and it must not be a member of a file hierarchy rooted at one of the file operands. the result of a copy under these conditions is unpredictable. do not do this. fr fr.

while processing a damaged archive during a read or list operation, pax will attempt to recover from media defects and will search through the archive to locate and process the largest number of archive members possible (see the -E option for more details on error handling). she's a fighter 💪

the directory operand specifies a destination directory pathname. if the directory operand does not exist, or it is not writable by the user, or it is not of type directory, pax will exit with a non-zero exit status. no negotiations 😤

the pattern operand is used to select one or more pathnames of archive members. archive members get selected using the pattern matching notation described by glob(7). when the pattern operand is not supplied, all members of the archive will be selected. when a pattern matches a directory, the entire file hierarchy rooted at that directory will be selected. and when a pattern operand does not select at least one archive member, pax will write those pattern operands in a diagnostic message to standard error and then exit with a non-zero exit status 💀

the file operand specifies the pathname of a file to be copied or archived. when a file operand does not select at least one archive member, pax will write those file operand pathnames in a diagnostic message to standard error and then exit with a non-zero exit status.

the options are as follows:

• -0 — use the NUL ('\0') character as a pathname terminator instead of newline ('\n'). this applies only to the pathnames read from standard input in the write and copy modes, and to the pathnames written to standard output in list mode. this option is expected to be used in concert with the -print0 function in find(1), the -d '' option to the read built-in utility of mksh(1), or the -0 flag in xargs(1). trio of icons honestly 🎀
• -a — append the given file operands to the end of an archive that was previously written. if an archive format is not specified with a -x option, the format currently being used in the archive will be selected. any attempt to append to an archive in a format different from the format already used in the archive will cause pax to exit immediately with a non-zero exit status. the blocking size used in the archive volume where writing starts will continue to be used for the remainder of that archive volume.

⚠️ Warning: many storage devices are not able to support the operations necessary to perform an append operation. any attempt to append to an archive stored on such a device may damage the archive or have other unpredictable results. tape drives in particular are more likely to not support an append operation. an archive stored in a regular filesystem file or on a disk device will usually support an append operation. so like… check before you wreck 🫣

• -B bytes — limit the number of bytes written to a single archive volume to bytes. the bytes limit can end with 'm', 'k', or 'b' to specify multiplication by 1048576 (1M), 1024 (1K) or 512, respectively. a pair of bytes limits can be separated by 'x' to indicate a product.

⚠️ Warning: only use this option when writing an archive to a device which supports an end of file read condition based on last (or largest) write offset (such as a regular file or a tape drive). the use of this option with a floppy or hard disk is not recommended. noted 💀

• -b blocksize — when writing an archive, block the output at a positive decimal integer number of bytes per write to the archive file. the blocksize must be a multiple of 512 bytes with a maximum of 64512 bytes. archive block sizes larger than 32256 bytes violate the POSIX standard and will not be portable to all systems. a blocksize can end with 'k' or 'b' to specify multiplication by 1024 (1K) or 512, respectively. a pair of blocksizes can be separated by 'x' to indicate a product. a specific archive device may impose additional restrictions on the size of blocking it will support. when blocking is not specified, the default blocksize is dependent on the specific archive format being used (see the -x option).
• -c — match all file or archive members except those specified by the pattern and file operands. it's giving inverse energy 🙈
• -D — this option is the same as the -u option, except that the file inode change time is checked instead of the file modification time. the file inode change time can be used to select files whose inode information (e.g., UID, GID, etc.) is newer than a copy of the file in the destination directory.
• -d — cause files of type directory being copied or archived, or archive members of type directory being extracted, to match only the directory file or archive member and not the file hierarchy rooted at the directory. just the one, thanks ✨
• -E limit — limit the number of consecutive read faults while trying to read a flawed archive to limit. with a positive limit, pax will attempt to recover from an archive read error and will continue processing starting with the next file stored in the archive. a limit of 0 will cause pax to stop operation after the first read error is detected on an archive volume. the default limit is a small positive number of retries. she's persistent but not that persistent 💅
• -f archive — specify archive as the pathname of the input or output archive, overriding the default standard input (for list and read) or standard output (for write). a single archive may span multiple files and different archive devices. when required, pax will prompt for the pathname of the file or device of the next volume in the archive. multi-volume queen 👑
• -G group — select a file based on its group name, or when starting with a #, a numeric GID. a '\' can be used to escape the #. multiple -G options may be supplied and checking stops with the first match.
• -H — follow only command-line symbolic links while performing a physical file system traversal.
• -i — interactively rename files or archive members. for each archive member matching a pattern operand or each file matching a file operand, pax will prompt to /dev/tty giving the name of the file, its file mode, and its modification time. pax will then read a line from /dev/tty. if this line is blank, the file or archive member is skipped. if this line consists of a single period, the file or archive member is processed with no modification to its name. otherwise, its name is replaced with the contents of the line. pax will immediately exit with a non-zero exit status if EOF is encountered when reading a response, or if /dev/tty cannot be opened for reading and writing. very hands-on, very collaborative 🫶
• -J — use the xz utility to compress (decompress) the archive while writing (reading). incompatible with -a. 🚫
• -j — use the bzip2 utility to compress (decompress) the archive while writing (reading). incompatible with -a. 🚫
• -k — do not overwrite existing files. simple as that, bestie 💅
• -L — follow all symbolic links to perform a logical filesystem traversal.
• -l — (yes bestie, that's a lowercase "ell" 💅) Link files. When you're in copy mode (-r -w), hard links get made between the source and destination file hierarchies whenever it's possible. slay.
• -M flag — This one configures the archive normaliser ✨. The flag is either a numeric value that plays nice with strtonum(3) (it gets stored straight into the flags word, no cap), or one of these iconic little values below — and you can slap a no- in front of any of them to turn them off:
  • inodes 0x0001: Serialise inodes, zero device info. (cpio, sv4cpio, sv4crc)
  • links 0x0002: Store content of hard links only once. (cpio, sv4cpio, sv4crc)
  • mtime 0x0004: Zero out the file modification time. (ar, cpio, sv4cpio, sv4crc, ustar)
  • uidgid 0x0008: Set owner to 0:0 (root:wheel). (ar, cpio, sv4cpio, sv4crc, ustar)
  • verb 0x0010: Debug this option.
  • debug 0x0020: Debug file header storage.
  • lncp 0x0040: Extract hard links by copy if link fails.
  • numid 0x0080: Use only numeric uid and gid values. (ustar)
  • gslash 0x0100: Append a slash after directory names. (ustar)
  • set 0x0003: Keep ownership and mtime intact.
  • dist 0x008B: Clean everything except mtime.
  • norm 0x008F: Clean everything.
  • root 0x0089: Clean owner and device information.

Okay so when you're creating an archive and verbosely listing output, these normalisation operations do not show up in that output — because they only happen after the output has already been shown. iykyk 🫶

Heads up: this option is only implemented for the ar, cpio, sv4cpio, sv4crc, and ustar file format writing routines.

TODO: The pax frontend should be using the -o option for handling this feature instead. (not me calling out upstream like this 😭)

• -n — Select the first archive member that matches each pattern operand. No more than one archive member gets matched per pattern, periodt. And when members of type directory match, the whole file hierarchy rooted at that directory matches too (unless -d is also specified, bestie 🎀).
• -O — Force the archive to be one volume. If a volume ends prematurely, pax will not prompt for a new volume. This is lowkey perfect for automated tasks where a human can't be there to do error recovery. 🤖
• -o options — Info to modify the algorithm for extracting or writing archive files, specific to the archive format picked by -x. In general, options take the form:

name=value

Here are the options available for the ustar and old BSD tar formats:

• write_opt=nodir — When writing archives, omit the storage of directories.

• -P — Do not follow symbolic links, do a physical filesystem traversal instead. This is the default mode, bestie. 🦋
• -p string — Specify one or more file characteristic options (privileges). The string option-argument is a string saying which file characteristics get retained or discarded on extraction. That string is built from the specification characters a, e, m, o, and p. You can concatenate multiple characteristics inside the same string, and you can specify multiple -p options too. Here's what each character means, babes:
  • a — Do not preserve file access times. By default, file access times are preserved whenever possible.
  • e — "Preserve everything": the user ID, group ID, file mode bits, file access time, and file modification time. This one is intended for root — someone with all the appropriate privileges — so every aspect of the files gets preserved exactly as recorded in the archive. Fun fact: the e flag is literally the sum of the o and p flags. ate. 👑
  • m — Do not preserve file modification times. By default, file modification times are preserved whenever possible.
  • o — Preserve the user ID and group ID.
  • p — "Preserve" the file mode bits. This one's for a user with regular privileges who wants to preserve everything about the file except ownership. The file times are preserved by default, but there are two other flags offered to disable that and use the time of extraction instead.

In that list above, "preserve" means an attribute stored in the archive gets given to the extracted file — subject to the permissions of the invoking process, obviously ✨. Otherwise, the attribute of the extracted file is determined as part of the normal file creation action. If neither the e nor the o specification character is specified, or the user ID and group ID aren't preserved for any reason, pax will not set the S_ISUID (setuid) and S_ISGID (setgid) bits of the file mode. And if preserving any of these items fails for any reason, pax writes a diagnostic message to standard error. Failure to preserve these items will affect the final exit status, but it will not cause the extracted file to be deleted. If the file characteristic letters in any of the string option-arguments are duplicated or conflict with each other, the one(s) given last win. For example: if you specify -p eme, file modification times are still preserved. (last one spoke, periodt 💅)

• -r — Read an archive file from standard input and extract the specified file operands. If any intermediate directories are needed in order to extract an archive member, those directories get created as if mkdir(2) was called with the bitwise OR of S_IRWXU, S_IRWXG, and S_IRWXO as the mode argument. And when the selected archive format supports specifying linked files and those files can't be linked while the archive is being extracted, pax will write a diagnostic message to standard error and exit with a non-zero exit status at the completion of the operation. 😭
• -s replstr — Modify the archive member names according to the substitution expression replstr, using the syntax of the ed(1) utility regular expressions. You can give file or pattern arguments to restrict the list of archive members to those specified.

The format of these regular expressions is:

/old/new/[gp]

Just like in ed(1), old is a basic regular expression (peep re_format(7)) and new can contain an ampersand (&), \n (where n is a digit) back-references, or subexpression matching. The old string may also contain newline characters. Any non-null character can be used as a delimiter (/ is shown here). Multiple -s expressions can be specified. The expressions are applied in the order they're specified on the command line, terminating with the first successful substitution. 🌈

That optional trailing g keeps applying the substitution expression to the pathname substring, which starts with the first character following the end of the last successful substitution. The first unsuccessful substitution stops the operation of the g option. And the optional trailing p will cause the final result of a successful substitution to be written to standard error in this format:

original-pathname >> new-pathname

File or archive member names that substitute to the empty string are not selected and will be skipped. bye bestie 👋

• -T range — Allow files to be selected based on a file modification or inode change time falling within the specified time range. The range has this format:

[from_date][,to_date][/[c][m]]

The dates specified by from_date to to_date are inclusive. If only a from_date is supplied, all files with a modification or inode change time equal to or younger are selected. If only a to_date is supplied, all files with a modification or inode change time equal to or older will be selected. When from_date equals to_date, only files with a modification or inode change time of exactly that time will be selected.

When pax is in write or copy mode, that optional trailing field [c][m] can be used to determine which file time (inode change, file modification, or both) gets used in the comparison. If neither is specified, the default is to use file modification time only. The m specifies the comparison of file modification time (the time when the file was last written). The c specifies the comparison of inode change time (the time when the file inode was last changed; e.g., a change of owner, group, mode, etc). When c and m are both specified, then the modification and inode change times are both compared. 🕰️

The inode change time comparison is useful for selecting files whose attributes were recently changed, or selecting files which were recently created and had their modification time reset to an older time (as what happens when a file is extracted from an archive and the modification time is preserved). Time comparisons using both file times is useful when pax is used to create a time based incremental archive — only files that were changed during a specified time range get archived. iconic behaviour fr fr ✨

A time range is made up of six different fields and every field must contain two digits. The format is:

[[[[[cc]yy]mm]dd]HH]MM[.SS]

Where cc is the first two digits of the year (the century), yy is the last two digits of the year, the first mm is the month (from 01 to 12), dd is the day of the month (from 01 to 31), HH is the hour of the day (from 00 to 23), MM is the minute (from 00 to 59), and SS is the seconds (from 00 to 59). The minute field MM is required, while the other fields are optional and must be added in this order: HH, dd, mm, yy, cc.

The SS field may be added independently of the other fields. Time ranges are relative to the current time, so -T 1234/cm would select all files with a modification or inode change time of 12:34 PM today or later. Multiple -T time ranges can be supplied, and checking stops with the first match. 🎀

• -t — resets the access times of every file or directory that pax read or accessed, putting them back to exactly what they were before pax touched them. like, no trace left behind 💅
• -U user — picks a file by its user name. or, if the argument starts with a #, by numeric UID instead. you can escape that # with a \ if you're being cheeky. you can stack multiple -U options and checking stops the second one matches. first match wins, no cap.
• -u — ignore files that are older than a pre-existing file or archive member with the same name (older = less recent file modification time). breaking it down by mode: while reading, an archive member with the same name as a filesystem file gets extracted only if the archive member is newer than the file. while writing, a filesystem member with the same name as an archive member gets written into the archive only if it's newer than the archive member. while copying, the file in the destination hierarchy gets replaced by the source file (or by a link to the source file) — but only if the source one is newer. newer stays, older gets skipped, periodt.
• -v — during a list operation, gives you a verbose table of contents in the exact format ls(1) uses with the -l option. for pathnames that are a hard link to a previous member of the archive, the output looks like:

ls -l listing == link-name

and for pathnames that are a symbolic link, it looks like:

ls -l listing -> link-name

where ls -l listing is literally the output format ls(1) gives you with -l. for every other operational mode (read, write, and copy), pathnames get written and flushed to standard error with no trailing newline, the moment processing starts on that file or archive member. the trailing newline is not buffered and only gets written once the file has actually been read or written. chaotic but that's the spec 🫶

• -w — writes files to standard output in the archive format you specified. when you give no file operands, standard input is read as a list of pathnames, one per line, with no leading or trailing <blanks>.
• -X — when it's walking the file hierarchy for a pathname, don't descend into directories that have a different device ID. peep the st_dev field as described in stat(2) if you wanna know more about device IDs.
• -x format — sets the output archive format. default is ustar. pax currently supports these formats:
  • ar — the Unix Archiver library format. this one matches APT repositories and the BSD ar(1) spec, not GNU binutils (though binutils can read them) and not SYSV systems. check ar(5) on some operating systems for more ✨
  • bcpio — the old binary cpio format. default blocksize is 5120 bytes. this format is not very portable and should not be used when other formats are available, fr fr. inode and device info about a file (which this format uses to detect file hard links) may get truncated by this format — pax detects that and repairs it. 🔧
  • cpio — the extended cpio interchange format specified in the IEEE Std 1003.2 (“POSIX.2”) standard. default blocksize is 5120 bytes. inode and device info about a file (used for detecting file hard links by this format), which this format may truncate, is detected by pax and repaired.
  • sv4cpio — the System V release 4 cpio. default blocksize is 5120 bytes. same inode/device truncation situation: pax detects it and repairs it.
  • sv4crc — the System V release 4 cpio with file CRC checksums. default blocksize is 5120 bytes. and yep, inode and device info may be truncated by this format, pax detects it and repairs it.
  • tar — the old BSD tar format as found in 4.3BSD. default blocksize is 10240 bytes. pathnames stored by this format must be 100 characters or less in length. only regular files, hard links, symlinks and directories will be archived — other filesystem types are not supported. for backwards compatibility with even older tar formats, you can use a -o option when writing an archive to omit the storage of directories. it takes this form:

-o write_opt=nodir

• ustar — the extended tar interchange format specified in the IEEE Std 1003.2 (“POSIX.2”) standard. default blocksize is 10240 bytes. filenames stored by this format must be 100 characters or less in length; the total pathname must be 256 characters or less.

pax will detect and report any file it can't store or extract because of any specific archive format restrictions. each individual archive format may impose extra restrictions too. typical restrictions include (but are not limited to): file pathname length, file size, link pathname length, and the type of the file. 👑

• -Y — same as the -D option, except the inode change time is checked using the pathname created after all the file name modifications have completed.
• -Z — same as the -u option, except the modification time is checked using the pathname created after all the file name modifications have completed.
• -z — uses the gzip(1) utility to compress (decompress) the archive while writing (reading). incompatible with -a. 🚫

---

The options that operate on the names of files or archive members (-c, -i, -n, -s, -u, -v, -D, -G, -T, -U, -Y, and -Z) interact like this:

When extracting files during a read operation, archive members are ‘selected’ based only on the user specified pattern operands as modified by the -c, -n, -u, -D, -G, -T, -U options. then any -s and -i options will modify the names of those selected files, in that order. then the -Y and -Z options get applied based on the final pathname. finally, the -v option writes out the names resulting from all those modifications.

When archiving files during a write operation, or copying files during a copy operation, archive members are ‘selected’ based only on the user specified pathnames as modified by the -n, -u, -D, -G, -T, and -U options (and heads up, -D only applies during a copy operation). then any -s and -i options modify the names of those selected files, in that order. then during a copy operation the -Y and -Z options get applied based on the final pathname. and finally, -v writes the names resulting from these modifications.

When one or both of the -u or -D options are specified along with the -n option, a file is not considered selected unless it's newer than the file it's being compared to. no shortcuts 🫡

🌸 ENVIRONMENT 🌸

• TMPDIR — path where temporary files get stored.

✨ EXIT STATUS ✨

The pax utility exits 0 on success, and >0 if an error occurs.

💖 EXAMPLES 💖

copy the contents of the current directory to the device /dev/rst0:

$ pax -w -f /dev/rst0 .

give the verbose table of contents for an archive stored in filename:

$ pax -v -f filename

this sequence of commands copies the entire olddir directory hierarchy to newdir:

$ mkdir newdir
$ cd olddir
$ pax -rw . ../newdir

extract files from the archive a.pax. files rooted in /usr get extracted relative to the current working directory; all other files get extracted to their unmodified path:

$ pax -r -s ',^/usr/,,' -f a.pax

this one lets you interactively select which files to copy from the current directory to dest_dir:

$ pax -rw -i . dest_dir

extract all files from the archive a.pax that are owned by root with group bin, and preserve all file permissions:

$ pax -r -pe -U root -G bin -f a.pax

update (and list) only those files in the destination directory /backup that are older (less recent inode change or file modification times) than files with the same name found in the source file tree home:

$ pax -r -w -v -Y -Z home /backup

😭 DIAGNOSTICS 😭

Whenever pax cannot create a file or a link while reading an archive, or cannot find a file while writing an archive, or cannot preserve the user ID, group ID, or file mode when the -p option is specified, a diagnostic message gets written to standard error and a non-zero exit status will be returned — but processing continues. and in the case where pax cannot create a link to a file, unless -M lncp is given, pax will not create a second copy of the file. noted. 📝

If the extraction of a file from an archive is prematurely terminated by a signal or error, pax may have only partially extracted a file you wanted. additionally, the file modes of extracted files and directories may have incorrect file bits, and the modification and access times may be wrong. yeah. 😬

If the creation of an archive is prematurely terminated by a signal or error, pax may have only partially created the archive, which may violate the specific archive format specification.

If while doing a copy, pax detects that a file is about to overwrite itself, the file is not copied, a diagnostic message gets written to standard error, and when pax completes it will exit with a non-zero exit status.

🦋 SEE ALSO 🦋

ar(1), cpio(1), deb(5), paxcpio(1), paxtar(1), tar(1)

👑 STANDARDS 👑

The pax utility is mostly compliant with an older version of the IEEE Std 1003.1 (“POSIX”) specification, except for the known “BUGS” listed below, and that the pax archive format and the listopt keyword are unsupported.

The flags -0BDEGHJjLMOPTUYZz, the archive formats ar, bcpio, sv4cpio, sv4crc and tar, the b, k, and x additions to the -b flag and the flawed archive handling during list and read operations are extensions to that specification. slay, they went above and beyond 💜

💖 HISTORY 💖

okay so story time bestie 🫶 a pax utility appeared in 4.4BSD. that's it, that's the origin story, our girl just showed up one day and never left 👑

---

👑 AUTHORS 👑

the originators, the blueprint, the main characters:

• Keith Muller at the University of California, San Diego 💜
• MirBSD extensions by mirabilos <m@mirbsd.org> ✨

ngl the MirBSD extensions ate fr fr 🌈

---

💀 BUGS 💀

okay so here's the tea and it's lowkey messy 😭

The pattern matching does not match either POSIX or this documentation completely.

that's the whole bug. it just… doesn't fully do the thing it's supposed to do 🙈 so if you're out here trusting the pattern matching on vibes alone, don't — it's giving "close enough" energy. no cap.

See also “STANDARDS” above — go read it, it's up there, it matters 💅

---

and that's a wrap on PAX(1) bestie, you made it all the way through 🎀 this page comes to us from MirBSD and it's dated January 5, 2024 ✨ archive responsibly, and remember: it's not hoarding if it's a well-formed archive 🌸💖🦋