Scroll to navigation

RDIFF-BACKUP(1) Manual 2.2.6 RDIFF-BACKUP(1)

NAME

rdiff-backup - local/remote mirror and incremental backup

SYNOPSIS

rdiff-backup [options...] action [sub-options...] [locations...]
rdiff-backup [--new] [-h|--help|-V|--version]

DESCRIPTION

rdiff-backup is a script, written in python(1) that backs up one directory to another. The target directory ends up a copy (mirror) of the source directory, but extra reverse diffs are stored in a special sub-directory of that target directory, so you can still recover files lost some time ago. The idea is to combine the best features of a mirror and an incremental backup. rdiff-backup also preserves symlinks, special files, hardlinks, permissions, uid/gid ownership, and modification times.

rdiff-backup can also operate in a bandwidth efficient manner over a pipe, like rsync(1). Thus you can use ssh and rdiff-backup to securely back a hard drive up to a remote location, and only the differences will be transmitted. Using the default settings, rdiff-backup requires that the remote system accept ssh connections, and that rdiff-backup is installed in the user’s PATH on the remote system. See the REMOTE OPERATION section for details.

Note that you should not write to the mirror directory except with rdiff-backup. Many of the increments are stored as reverse diffs, so if you delete or modify a file, you may lose the ability to restore previous versions of that file.

Finally, this man page is intended more as a precise description of the behavior and syntax of rdiff-backup. New users may want to check out the examples file included in the rdiff-backup distribution.

The rdiff-backup commands knows four types of parameters

1.generic options valid for all actions,

2.one action out of backup, calculate, complete, compare, info, list, regress, remove, restore, server, test, verify,

3.sub-options applicable to each action specifically, even though some are common to multiple actions,

4.zero, one, two or more location paths, either local or remote.

Note that this documents the new command line interface of rdiff-backup since 2.1+; for the traditional one, check rdiff-backup-old(1) but consider that it is deprecated and will disappear.

Generic options

-h, --help

Prints brief usage information and exits. Add --new to be sure to get this CLI description, and not the old one. Placed after the action, outputs the action’s specific help message.

-V, --version

Prints the current version number and exits.

--api-version apiversion

Sets the API version to the given integer between minimum and maximum versions as given by the info action. It is the responsibility of the user to make sure that this version is also supported by any server started by this client.

--chars-to-quote, --override-chars-to-quote chars

If the filesystem to which we are backing up is not case-sensitive, automatic "quoting" of characters occurs. For example, a file 'Developer.doc' will be converted into ';068eveloper.doc'. To quote other characters or force quoting, e.g. in case rdiff-backup doesn’t recognize a case-insensitive file system, you need to specify this option. chars is a string of characters fit to be used in regexp square brackets (e.g. 'A-Z' as in '[A-Z]').


Caution

do NOT change the chars to quote within the same repository!
Actually, you only need to set this parameter when creating a new backup repository. Do also NOT quote any character used by rdiff-backup in rdiff-backup-data (any of 'a-z0-9._-')!

--current-time currenttime

This option is useful mainly for testing. If set, rdiff-backup will use it for the current time instead of consulting the clock. The argument is the number of seconds since the epoch.

--force

Authorize a more drastic modification of a directory than usual (for instance, when overwriting of a destination path, or when removing multiple sessions with remove). rdiff-backup will generally tell you if it needs this.


Caution

You can cause data loss if you mis-use this option.
Furthermore, do NOT use this option when doing a restore, as it will DELETE files, unless you absolutely know what you are doing.

--fsync, --no-fsync

This will enable/disable issuing fsync from rdiff-backup altogether. This option is designed to optimize performance on busy backup systems.


Caution

This may render your backup unusable in case of filesystem failure.
Default is hence for fsync to be enabled.

--new, --no-new

enforce (or not) the usage of the new parameters. The default currently is to show the old usage, but this will change in the near future.

--null-separator

Use nulls (\0) instead of newlines (\n) as line separators, which may help when dealing with filenames containing newlines. This affects the expected format of the files specified by the --{include|exclude}-filelist[-stdin] switches as well as the format of the files statistics.

--parsable-output

If set, rdiff-backup’s output will be tailored for easy parsing by computers, instead of convenience for humans. Currently this only applies when listing increments using the list increments action, where the time will be given in seconds since the epoch.

--remote-schema remoteschema

Specify an alternate method of connecting to a remote computer. This is necessary to get rdiff-backup not to use ssh for remote backups, or if, for instance, rdiff-backup is not in the PATH on the remote side. See the REMOTE OPERATION section for details.

--remote-tempdir dirpath

use path as temporary directory on the remote side of the connection. If set explicitly, remember that "no space left" error messages might apply to this directory.

--ssh-compression, --no-ssh-compression

use SSH with or without compression with default remote-schema. This option is ignored when using --remote-schema. Compression is on by default.

--tempdir dirpath

Sets the directory that rdiff-backup uses for temporary files to the given path. The environment variables TMPDIR, TEMP, and TMP can also be used to set the temporary files directory. See the documentation of the Python tempfile module for more information. If set explicitly, remember that "no space left" error messages might apply to this directory.

--terminal-verbosity {0,1,2,3,4,5,6,7,8,9}

select which verbosity to use for messages on the terminal, the default is given by --verbosity.

--use-compatible-timestamps

Create timestamps in which the hour/minute/second separator is a - (hyphen) instead of a : (colon). It is safe to use this option on one backup, and then not use it on another; rdiff-backup supports the intermingling of different timestamp formats. This option is enabled by default on platforms which require that the colon be escaped.

-v, --verbosity {0,1,2,3,4,5,6,7,8,9}

Specify verbosity level (0 is totally silent, 3 is the default, and 9 is noisiest). This determines how much is written to the log file, and without using --terminal-verbosity to the terminal..

Actions

backup [CREATION OPTIONS] [COMPRESSION OPTIONS] [SELECTION OPTIONS] [FILESYSTEM OPTIONS] [USER GROUP OPTIONS] [STATISTICS OPTIONS] sourcedir targetdir

back-up a source directory to a target backup repository.

calculate [--method average] statfile1 statfile2 [...]

calculate average across multiple statistics files

--method average

there is currently only one method and it is the default, but it might change in the future.

complete [--cword index] [--unique|--no-unique] -- words [...]

outputs a list of fitting options given already entered parameters. This option is of no relevance to backup tasks, it is only to be used to support programmatic completion like in bash. See the documentation for more details if you plan to write your own completion code, e.g. for an alternative shell.

--cword index

index where the cursor currently is within the list of words.

--unique,--no-unique

should parameters already entered by the user be offered again, or not?

compare [SELECTION OPTIONS] [--method method] [--at time] sourcedir targetdir

Compare a directory with the backup set at the given time. This can be useful to see how archived data differs from current data, or to check that a backup is current.

--method method

method used to compare can be either meta, full or hash, where the default is meta, which is also how rdiff-backup decides which file needs to be backed-up. Note that with full, the repository data will be copied in its entirety to the source side and compared byte by byte. This is the slowest but most complete compare method. With hash only the SHA1 checksum of regular files will be compared. With meta only the metadata of files will be compared (name, size, date, type, etc).

--at time

at which time of the back-up directory should the comparaison take place. The default is now, meaning the latest version. See TIME FORMATS for details.

info

outputs information about the current system in YAML format, so that it can be used in a bug report, and exits.

list files [--changed-since time|--at time] repository

list modified or existing files in a given back-up repository.

--changed-since time

List the files that have changed in the destination directory since the given time. See TIME FORMATS for the format of time. If a directory in the archive is specified, list only the files under that directory. This option does not read the source directory; it is used to compare the contents of two different rdiff-backup sessions. See TIME FORMATS for details.

--at time

List the files in the archive that were present at the given time. If a directory in the archive is specified, list only the files under that directory. See TIME FORMATS for details.

list increments [--no-size|--size] repository

list increments with date in a given back-up repository.

--no-size,--size

Show or not the size of each increment in the repository. The default is to not show sizes. When showing sizes, it becomes allowable to specify a directory within a repository, then only the cumulated sizes of that directory will be shown.

regress [COMPRESSION OPTIONS] [USER GROUP OPTIONS] [TIMESTAMP OPTIONS] repository

If an rdiff-backup session fails, this action will undo the failed directory. This happens automatically if you attempt to back-up to a directory and the last backup failed. You can use the --force option to undo the last backup even if it wasn’t failed (starting with API 201, use --api-version if necessary).

remove increments --older-than time [--size] repository

Remove the incremental backup information in the destination directory that has been around longer than the given time, or the oldest one if no time is provided.

By default, rdiff-backup will only delete information from one session at a time. To remove two or more sessions at the same time, supply the --force option (rdiff-backup will tell you if it is required).

Note that snapshots of deleted files are covered by this operation. Thus if you deleted a file two weeks ago, backed up immediately afterwards, and then ran rdiff-backup with 'remove increments --older-than 10D' today, no trace of that file would remain.

--older-than time

all the increments older than the given time will be deleted. See TIME FORMATS for details.

--size

Show the size of each increment being removed. The default is to not show sizes.

restore [CREATION OPTIONS] [COMPRESSION OPTIONS] [SELECTION OPTIONS] [FILESYSTEM OPTIONS] [USER GROUP OPTIONS] [--at time|--increment] source targetdir

restore a source backup repository at a specific time or a specific source increment to a target directory. See RESTORING for details.

--at time

the source parameter is interpreted as a back-up directory, and the content is restored from the given time. See TIME FORMATS for details.

--increment

the source parameter is expected to be an increment within a back-up repository, to be restored into the given target directory.

server [RESTRICT OPTIONS] [--debug]

Enter server mode (not to be invoked directly, but instead used by another rdiff-backup process on a remote computer).

--debug

Start the server in debug mode so that it stops on an early breakpoint and can be remotely debugged using rpdb <https://github.com/tamentis/rpdb>. See the developer’s documentation <https://github.com/rdiff-backup/rdiff-backup/blob/master/docs/DEVELOP.adoc#debug-client-server-mode> for details.

test remote_location_1 [remote_location_2 ...]

Test for the presence of a compatible rdiff-backup server as specified in the remote location argument(s) (of which the filename section will be checked for existence). See the REMOTE OPERATION section for details.

verify [--at time] location

Check all the data in the repository at the given time by computing the SHA1 hash of all the regular files and comparing them with the hashes stored in the metadata file.

--at time

the time of the data which needs to be verified. See TIME FORMATS for details.

COMPRESSION OPTIONS

--compression, --no-compression

Enable or disable the default gzip compression of most of the .snapshot and .diff increment files stored in the rdiff-backup-data directory. A backup volume can contain compressed and uncompressed increments, so using this option inconsistently is fine. Default is to compress all files, except those excluded as noted below.

--not-compressed-regexp regexp

Do not compress increments based on files whose filenames match regexp. The default includes many common audiovisual and archive files, and may be found from the help.

CREATION OPTIONS

--create-full-path

Normally only the final directory of the destination path will be created if it does not exist. With this option, all missing directories on the destination path will be created. Use this option with care: if there is a typo in the remote path, the remote filesystem could fill up very quickly (by creating a duplicate backup tree). For this reason this option is primarily aimed at scripts which automate backups.

FILESYSTEM OPTIONS

--acls, --no-acls

enable/disable back-up of Access Control Lists.

--carbonfile, --no-carbonfile

enable/disable back-up of carbon files (MacOS X).

--eas, --no-eas

enable/disable back-up of Extended Attributes.

--resource-forks, --no-resource-forks

enable/disable back-up of resource forks (MacOS X).

--hard-links, --no-hard-links

do (or not) keep hard-link relationships between files. Disabling hard-links generally increases the disk space usage but decreases memory usage. Hard-links are disabled by default if the backup source or restore destination is running on native Windows.

--compare-inode, --no-compare-inode

This option prevents rdiff-backup from flagging a hardlinked file as changed when its device number and/or inode changes. This option is useful in situations where the source filesystem lacks persistent device and/or inode numbering. For example, network filesystems may have mount-to-mount differences in their device number (but possibly stable inode numbers); USB/1394 devices may come up at different device numbers each remount (but would generally have same inode number); and there are filesystems which don’t even have the same inode numbers from use to use. Without the option rdiff-backup may generate unnecessary numbers of tiny diff files.

--never-drop-acls

Exit with error instead of dropping ACLs or ACL entries. Normally this may happen (with a warning) because the destination does not support them or because the relevant user/group names do not exist on the destination side.

RESTRICT OPTIONS

--restrict-path dirpath

Require that all file access be inside the given path. This switch, and --restrict-mode, are intended to be used with the server action to provide a bit more protection when doing automated remote backups.


Caution

Those options are not intended as your only line of defense so please don’t do something silly like allow public access to an rdiff-backup server run with --restrict-mode read-only.

--restrict-mode {read-write,read-only,update-only}

restriction mode for the directory given by --restrict-path, either full access (aka read-write), read-only, or only to update incrementally an already existing back-up (default is read-write).

SELECTION OPTIONS

This section only quickly lists the existing options, the section FILE SELECTION explains those more in details.

Globs, Regex, File lists selection

--include,--exclude glob

Include/exclude the file or files matched by glob (also known as shell pattern). If a directory is excluded, then files under that directory will also be excluded.

--include-globbing-filelist,--exclude-globbing-filelist globsfile

Include/exclude according to the listed globs, similar to --include or --exclude.

--include-globbing-filelist-stdin,--exclude-globbing-filelist-stdin

Like the previous option but the list of globs is coming from standard input.

--include-regexp,--exclude-regexp regexp

Include/exclude files matching the given regexp (according to Python rules).

--include-filelist,--exclude-filelist listfile

Include/exclude the files listed in filelist. This is a best fit for an automatically generated list of files, else use globbing.

--include-filelist-stdin,--exclude-filelist-stdin

Like the previous but the filelist is coming from standard input.

Special files selection


Note

All special files are included by default, so that including them explicitly isn’t generally required.
Exceptions are described.

--include-device-files,--exclude-device-files

Include/exclude all device files. This can be useful for security/permissions reasons or if rdiff-backup is not handling device files correctly.

--include-fifos,--exclude-fifos

Include/exclude all fifo files.

--include-sockets,--exclude-sockets

Include/exclude all socket files.

--include-symbolic-links,--exclude-symbolic-links

Include/exclude all symbolic links. Contrary to the general rule, symlinks are excluded by default under Windows so that NTFS reparse points aren’t backed-up.

--include-special-files,--exclude-special-files

Include/exclude all the special files listed above.

Other selections

--include-other-filesystems,--exclude-other-filesystems

Include/exclude files on file systems (identified by device number) other than the file system the root of the source directory is on. The default is to include other filesystems.

--include-if-present,--exclude-if-present filename

Include/exclude directories if they contain the given filename. Files directly in an included directory are also considered included. This doesn’t apply recursively though so that the filename must be present in all directories of a hierarchy for it to be fully included.

--max-file-size sizeinbytes

Exclude files that are larger than the given size in bytes.

--min-file-size sizeinbytes

Exclude files that are smaller than the given size in bytes.

STATISTICS OPTIONS

--file-statistics, --no-file-statistics

Enable/disable writing to the 'file_statistics' file in the rdiff-backup-data directory. rdiff-backup will run slightly quicker and take up a bit less space. Default is to write the statistics file(s).

See the FILES section for more information about statistics files.

--no-print-statistics, --print-statistics

Summary statistics will be printed (or not) after a successful backup. Even if disabled (the default), this information will still be available from the session statistics file.

TIMESTAMP OPTIONS

--allow-duplicate-timestamps

This option is only to be used if you encounter the issue of metadata mirrors with the same timestamp. In such cases, you may use this flag to first recover from the failed backup with something like

rdiff-backup regress --allow-duplicate-timestamps {targetdir}

after which you will need to remove those old duplicate entries using the remove increments action.

USER GROUP OPTIONS

See the USERS AND GROUPS section for more information.

--group-mapping-file mapfile

Map group names and IDs according to the group mapping file mapfile.

--user-mapping-file mapfile

Map user names and IDs according to the user mapping file mapfile.

--preserve-numerical-ids

If set, rdiff-backup will preserve uids/gids instead of trying to preserve unames and gnames.

RESTORING

There are two ways to tell rdiff-backup to restore a file or directory:

1.you can run rdiff-backup restore on a mirror file and define a time from which to restore (by default the latest one).

2.you can run the restore action on an increment file with the sub-option --increment.

For example, suppose in the past you have run:

rdiff-backup backup /usr /usr.backup

to back up the '/usr' directory into the '/usr.backup' directory, and now want a copy of the '/usr/local' directory the way it was 3 days ago placed at '/usr/local.old'.

One way to do this is to run:

rdiff-backup restore --at 3D /usr.backup/local /usr/local.old

here above the '3D' means 3 days (for other ways to specify the time, see the TIME FORMATS section). The '/usr.backup/local' directory was selected, because that is the directory containing the current version of 'usr/local'.

Note that the parameter of --at always specifies an exact time. (So '3D' refers to the moment 72 hours before the present). If there was no backup made at that time, rdiff-backup restores the state recorded for the previous backup. For instance, in the above case, if '3D' is used, and there are only backups from 2 days and 4 days ago, '/usr/local' as it was 4 days ago will be restored.

The second way to restore files involves finding the corresponding increment file. It would be in the '/backup/rdiff-backup-data/increments/usr' directory, and its name would be something like 'local.2002-11-09T12:43:53-04:00.dir' where the time indicates it is from 3 days ago. Note that the increment files all end in '.diff', '.snapshot', '.dir', or '.missing', where '.missing' just means that the file didn’t exist at that time (finally, some of these may be gzip-compressed, and have an extra '.gz' to indicate this). Then running:

rdiff-backup restore --increment \

/backup/rdiff-backup-data/increments/usr/local.{time}.dir \
/usr/local.old

would also restore the file as desired.

If you are not sure exactly which version of a file you need, it is probably easiest to either restore from the increments files as described immediately above, or to see which increments are available with 'list increments', and then specify an exact time with --at.

TIME FORMATS

rdiff-backup uses time strings in two places.

Firstly, all of the increment files rdiff-backup creates will have the time in their filenames in the w3 datetime format as described in a w3 note at <https://www.w3.org/TR/NOTE-datetime>. Basically they look like '2001-07-15T04:09:38-07:00', which is basically "{Year}-{Month}-{Day}T{Hours}:{Minutes}:{Seconds}{Timezone}", the time zone being 7 hours behind UTC in this example (hence the minus).

Secondly, the --at, --changed-since, --older-than options take a time string, which can be given in any of several formats:

1.the string 'now' (refers to the current time)

2.a sequences of digits, like '123456890' (indicating the time in seconds after the epoch)

3.A string like '2002-01-25T07:00:00+02:00' in datetime format

4.An interval, which is a number followed by one of the characters s, m, h, D, W, M, or Y (indicating seconds, minutes, hours, days, weeks, months, or years respectively), or a series of such pairs. In this case the string refers to the time that preceded the current time by the length of the interval. For instance, '1h78m"' indicates the time that was one hour and 78 minutes ago. The calendar here is unsophisticated: a month is always 30 days, a year is always 365 days, and a day is always 86400 seconds.

5.A date format of the form "YYYY/MM/DD", "YYYY-MM-DD", "MM/DD/YYYY", or "MM-DD-YYYY", which indicates midnight on the day in question, relative to the current timezone settings. For instance, '2002/3/5', '03-05-2002', and '2002-3-05' all mean March 5th, 2002 (needless to say that starting with the year is less confusing for non-Americans).

6.A backup session specification which is a non-negative integer followed by 'B'. For instance, '0B' specifies the time of the current mirror, and '3B' specifies the time of the 3rd newest increment.

REMOTE OPERATION

In order to access remote files, rdiff-backup opens up a pipe to a copy of rdiff-backup running on the remote machine. Thus rdiff-backup must be installed on both ends. To open this pipe, rdiff-backup first splits the location into 'host_info::pathname'. It then substitutes 'host_info' into the remote schema, and runs the resulting command, reading its input and output.

The 'host_info' can be anything understood as a destination by your version of SSH. Assuming it is the standard OpenSSH, it can be:

•either '[user@]hostname'

•or a URI of the form 'ssh://[user@]hostname[:port]'.

The default remote schema is 'ssh -C {h} rdiff-backup --server' where 'host_info' is substituted for '{h}'. So if the 'host_info' is 'user@host.net', then rdiff-backup runs 'ssh <user@host.net> rdiff-backup --server'. Using --remote-schema, rdiff-backup can invoke an arbitrary command in order to open up a remote pipe. For instance,

rdiff-backup --remote-schema 'cd /usr; {h}' backup \

foo 'rdiff-backup server'::bar

is basically equivalent to (but slower than)

rdiff-backup backup foo /usr/bar

Concerning quoting, if for some reason you need to put two consecutive colons in the 'host_info' section of a 'host_info::pathname' argument, or in the pathname of a local file, you can quote one of them by prepending a backslash. So in 'a\::b::c', 'host_info' is 'a::b' and the pathname is 'c'. Similarly, if you want to refer to a local file whose filename contains two consecutive colons, like 'strange::file', you’ll have to quote one of the colons as in 'strange\::file'. Because the backslash is a quote character in these circumstances, it too must be quoted to get a literal backslash, so 'foo\::\\bar' evaluates to 'foo::\bar'. To make things more complicated, because the backslash is also a common shell quoting character, you may need to type in '\\\\' at the shell prompt to get a literal backslash.

You may also use the placehoders '{Vx}', '{Vy}' and '{Vz}' for the 'x.y.z' version of rdiff-backup, so that you can have multiple versions of rdiff-backup installed on the server, and automatically targeted from the client.

For example, if you have rdiff-backup 2.1.5 and 2.2.1 installed in virtual environments on the server, respectively under '/usr/local/lib/rdiff-backup-2.0' and '/usr/local/lib/rdiff-backup-2.1' (we assume that the z-Version isn’t relevant to any kind of compatibility), then the client may be called with the following remote schema:

ssh -C {h} /usr/local/lib/rdiff-backup-{Vx}.{Vy} --server

The client will then use the correct version of rdiff-backup based on its own version 'x.y.z'. You’ll find more explanations in the migration file in the documentation.

If you need to include a literal '%' in the string specified by --remote-schema, quote it with another '%', as in '%%' (this is due to the compatibility with the deprecated host placeholder '%s', which you shouldn’t use anymore).

And finally, if you need to include literal '{ }' (curly braces) in the the string specified by --remote-schema, quote them (both) by doubling each of them up, as in '{{ foo=0; }}'.

Although ssh itself may be secure, using rdiff-backup in the default way presents some security risks. For instance if the server is run as root, then an attacker who compromised the client could then use rdiff-backup to overwrite arbitrary server files by "backing up" over them. Such a setup can be made more secure by using the sshd configuration option 'command="rdiff-backup server"' possibly along with the --restrict-path and --restrict-mode options to rdiff-backup. For more information, see the web page, the wiki, and the entries for those options on this man page.

FILE SELECTION

rdiff-backup has a number of file selection options. When rdiff-backup is run, it searches through the given source directory and backs up all the files matching the specified options. This selection system may appear complicated, but it is supposed to be flexible and easy-to-use. If you just want to learn the basics, first look at the selection examples in the examples file included in the package, or on the web at <https://rdiff-backup.net/examples.html>.

rdiff-backup’s selection system was originally inspired by rsync(1), but there are many differences. For instance, trailing backslashes have no special significance.


Important

include and exclude patterns under Windows solely support slashes '/' as file separators, given that backslashes '\' have a special meaning in regex/glob patterns.

All the available file selection conditions are listed under SELECTION OPTIONS.

Two principles need to be understood before really starting:

1.pattern matching is stupid about paths, it just does pattern matching and can’t interpret patterns like path, especially it can’t resolve absolute into relative paths and vice-versa (compare with the '-path' option of find).

2.pattern matching is done on the complete path of each found file (no partial matching and no file name matching). Beware that complete path does not mean full path, it can be a complete relative path.

For example, the pattern 'bar' matches the path 'bar', but doesn’t match the path 'foo/bar' and neither the path './bar'. Both are matched by the pattern '*/bar', as well as by '**/bar'. This last pattern would match any path containing the file 'bar', e.g. 'foo/boz/bar'.

Each file selection condition either matches or doesn’t match a given file. A given file is excluded by the file selection system exactly when the first matching file selection condition specifies that the file be excluded; otherwise the file is included. When backing up, if a file is excluded, rdiff-backup acts as if that file does not exist in the source directory. When restoring, an excluded file is considered not to exist in either the source or target directories.

For instance,

rdiff-backup backup --include /usr \

--exclude /usr /usr /backup

is exactly the same as

rdiff-backup backup /usr /backup

because the include and exclude directives match exactly the same files, and the --include comes first, giving it precedence. Similarly,

rdiff-backup backup --include /usr/local/bin \

--exclude /usr/local /usr /backup

would backup the '/usr/local/bin' directory (and its contents), but not '/usr/local/doc'.

The include, exclude, include-globbing-filelist, and exclude-globbing-filelist options accept extended shell globbing patterns. These patterns can contain the special patterns '*', '**', '?', and '[...]'. As in a normal shell, '*' can be expanded to any string of characters not containing '/', '?' expands to any character except '/', and '[...]' expands to a single character of those characters specified (ranges are acceptable). The new special pattern, '**', expands to any string of characters whether or not it contains '/'. Furthermore, if the pattern starts with 'ignorecase:' (case insensitive), then this prefix will be removed and any character in the string can be replaced with an upper or lowercase version of itself.

If you need to match filenames which contain the above globbing characters, they may be escaped using a backslash '\'. The backslash will only escape the character following it so for '**' you will need to use '\*\*' to avoid escaping it to the '*' globbing character.

Remember that you may need to quote these characters when typing them into a shell, so the shell does not interpret the globbing patterns before rdiff-backup sees them.

The --exclude pattern option matches a file if and only if:

1.pattern can be expanded into the file’s filename, or

2.the file is inside a directory matched by the option.

Conversely, --include pattern matches a file if and only if:

1.pattern can be expanded into the file’s filename,

2.the file is inside a directory matched by the option, or

3.the file is a directory which contains a file matched by the option.

For example,

--exclude /usr/local

matches '/usr/local', '/usr/local/lib', and '/usr/local/lib/netscape'. It is the same as

--exclude /usr/local --exclude '/usr/local/**'

And similarly:

--include /usr/local

specifies that '/usr', '/usr/local', '/usr/local/lib', and '/usr/local/lib/netscape' (but not '/usr/doc') all be backed up. Thus you don’t have to worry about including parent directories to make sure that included subdirectories have somewhere to go. Finally,

--include ignorecase:'/usr/[a-z0-9]foo/*/**.py'

would match a file like '/usr/5fOO/hello/there/world.py'. If it did match anything, it would also match '/usr'. If there is no existing file that the given pattern can be expanded into, the option will not match '/usr'.

The --include-filelist, --exclude-filelist, --include-filelist-stdin, and --exclude-filelist-stdin options also introduce file selection conditions. They direct rdiff-backup to read in a file, each line of which is a file specification, and to include or exclude the matching files. Lines are separated by newlines or nulls, depending on whether the --null-separator switch was given. Each line in a filelist is interpreted similarly to the way extended shell patterns are, with a few exceptions:

1.Globbing patterns like '*', '**', '?', and '[...]' are not expanded.

2.Include patterns do not match files in a directory that is included. So '/usr/local' in an include file will not match '/usr/local/doc'.

3.Lines starting with '+ [...]' (plus followed by a space) are interpreted as include directives, even if found in a filelist referenced by --exclude-filelist. Similarly, lines starting with '- [...]' (minus followed by a space) exclude files even if they are found within an include filelist.

For example, if the file 'list.txt' contains the lines:

/usr/local
- /usr/local/doc
/usr/local/bin
+ /var
- /var

then '--include-filelist list.txt' would include '/usr', '/usr/local', and '/usr/local/bin'. It would exclude '/usr/local/doc', '/usr/local/doc/python', etc. It neither excludes nor includes '/usr/local/man', leaving the fate of this directory to the next specification condition. Finally, it is undefined what happens with '/var'. A single file list should not contain conflicting file specifications.

The --include-globbing-filelist and --exclude-globbing-filelist options also specify filelists, but each line in the filelist will be interpreted as a globbing pattern the way --include and --exclude options are interpreted (although '+ ' and '- ' prefixing is still allowed). For instance, if the file 'globbing-list.txt' contains the lines:

dir/foo

Then '--include-globbing-filelist globbing-list.txt' would be exactly the same as specifying on the command line:

--include dir/foo --include dir/bar --exclude **

Finally, the --include-regexp and --exclude-regexp allow files to be included and excluded if their filenames match a python regular expression. Regular expression syntax is too complicated to explain here, but is covered in Python’s library reference. Unlike the --include and --exclude options, the regular expression options don’t match files containing or contained in matched files. So for instance

--include '[0-9]{7}(?!foo)'

matches any files whose full pathnames contain 7 consecutive digits which aren’t followed by 'foo'. However, it wouldn’t match '/home' even if '/home/ben/1234567' existed.

USERS AND GROUPS

There can be complications preserving ownership across systems. For instance the username that owns a file on the source system may not exist on the destination. Here is how rdiff-backup maps ownership on the source to the destination (or vice-versa, in the case of restoring):

1.If the --preserve-numerical-ids option is given, the remote files will always have the same uid and gid, both for ownership and ACL entries. This may cause unames and gnames to change.

2.Otherwise, attempt to preserve the user and group names for ownership and in ACLs. This may result in files having different uids and gids across systems.

3.If a name cannot be preserved (e.g. because the username does not exist), preserve the original id, but only in cases of user and group ownership. For ACLs, omit any entry that has a bad user or group name.

4.The --user-mapping-file and --group-mapping-file options override this behavior. If either of these options is given, the policy described in 2 and 3 above will be followed, but with the mapped user and group instead of the original. If you specify both --preserve-numerical-ids and one of the mapping options, the behavior is undefined.

The user and group mapping files both have the same format:

old_name_or_id1:new_name_or_id1
old_name_or_id2:new_name_or_id2
[...etc...]

Each line should contain a name or id, followed by a colon ':', followed by another name or id. If a name or id is not listed, they are treated in the default way described above.

When restoring, the above behavior is also followed, but note that the original source user/group information will be the input, not the already mapped user/group information present in the backup repository. For instance, suppose you have mapped all the files owned by alice in the source so that they are owned by ben in the repository, and now you want to restore, making sure the files owned originally by alice are still owned by alice. In this case there is no need to use any of the mapping options. However, if you wanted to restore the files so that the files originally owned by alice on the source are now owned by ben, you would have to use the mapping options, even though you just want the unames of the repository’s files preserved in the restored files.

See USER GROUP OPTIONS for a list and description of related options.

FILES

any-config-file

you can create a file with one option/action/sub-option per line and use it on the command line with an at sign prefix like @any-config-file and its content will be interpreted as if given on the command line.

For example, creating a file 'mybackup' with following content:

--verbosity
5
backup
source_dir
target_dir

and calling 'rdiff-backup @mybackup' will be the same as calling 'rdiff-backup --verbosity 5 backup source_dir target_dir'.

session_statistics, file_statistics

Every session rdiff-backup saves various statistics into two files, the session statistics file at 'rdiff-backup-data/session_statistics.{datetime}.data' and the files statistics at 'rdiff-backup-data/directory_statistics.{datetime}.data'. They are both text files and contain similar information: how many files changed, how many were deleted, the total size of increment files created, etc. However, the session statistics file is intended to be very readable and only describes the session as a whole. The files statistics file is more compact (and slightly less readable) but describes every directory backed up. It also may be compressed to save space.

See also STATISTICS OPTIONS and the --null-separator option.

backup.log, restore.log, error_log

rdiff-backup will save various messages to the log file, which is 'rdiff-backup-data/backup.log' for backup sessions and 'rdiff-backup-data/restore.log' for restore sessions. Generally what is written to this file will coincide with the messages displayed to stdout or stderr, although this can be changed with the --terminal-verbosity option.

Errors during backup are also written to a file 'rdiff-backup-data/error_log.{datetime}.data'.

The log files are not compressed and can become quite large if rdiff-backup is run with high verbosity.

ENVIRONMENT

RDIFF_BACKUP_VERBOSITY=[0-9]

the default verbosity for log file and terminal, can be overwritten by the corresponding options -v/--verbosity and --terminal-verbosity.

RDIFF_BACKUP_DEBUG=[address][:port]

set a non-default listening address and/or port (default is 127.0.0.1:4444) for rpdb. Valid values are address, address:port or :port.

RDIFF_BACKUP_API_VERSION={[dictionary]}

Overwrite the 'actual', 'default', 'max' and/or 'min' value of the API VERSION using a YAML object, e.g. {actual: 201, default: 201}. This environment variable is rather meant for development and test purposes to change the values without modifying the code, so use with care.

RETURN CODES

The following return codes have not been fully implemented so test before you rely on them. Also note that they can be combined, so that for example a return code 3 might be returned if a warning was found, then an error.

0 - OK

the action was completely successful

1 - ERROR

something fatal happened, the whole action failed

2 - WARNING

any kind of unexpected behavior without complete failure

4 - FILE ERROR

the action failed on a single file (or more), but it wasn’t the reason for a complete failure

8 - FILE WARNING

the action stumbled on a single file (or more), or detected differences in a comparaison


Tip

any other error code can and should be reported as a bug.

BUGS

See GitHub issues

In doubt subscribe to and ask the mailing list

AUTHORS

•Ben Escoto <ben@emerose.org> (retired)

•Eric Lavarde <ewl+rdiffbackup@lavar.de> (active)

SEE ALSO

rdiff-backup-old(1), python(1), rdiff(1), rsync(1), ssh(1).

The main rdiff-backup web page is at <https://rdiff-backup.net/>. It has more documentation, links to the mailing list and source code.

July 2024 rdiff-backup