There is no formal specification, but here is a rough description (settings are always cumulative, so 5 displays everything 4 does):

In general, rdiff-backup does not strive to make newer clients compatible with older servers (or vice versa). However, there is no intention to purposefully make different versions incompatible across the network — changes are introduced primarily to fix bugs or introduce new features that cannot be implemented without breaking the network protocol. Furthermore, rdiff-backup does try to make it possible to read older archives.

When running as a client, rdiff-backup checks the version of rdiff-backup running on the server, and prints a warning message if the two versions are different. If you have any problems with your backup, it is strongly recommended that you upgrade the older version before reporting any issues.

Yes, although it is not a heavily tested configuration. Rdiff-backup can be run as a native Windows application or under Cygwin. To run as a native Windows application, simply download the provided .exe binary. To setup remote operation, you will also need an SSH client, such as Putty or SSH Secure Shell.

If you wish to run rdiff-backup under Cygwin, use at least version 1.1.12. The setup under Cygwin is the same as under other Unix-like operating systems. From the Cygwin installer you will need Python 3.6 or higher (under Interpreters), autoconf, automake, binutils, gcc, make, and patchutils (all under Devel). Then you will need to compile and install librsync, which can be downloaded from Sourceforge. Finally, you can compile and install rdiff-backup using the usual instructions.

Although some Windows filesystems lack features like FIFOs, case sensitive filenames, or files with colons (":") in them, all of these situations should be autodetected and compensated for by rdiff-backup.

If you would like more detailed instructions for compiling and installing rdiff-backup on Cygwin, you can read this blog entry: https://katastrophos.net/andre/blog/2005/11/02/rdiff-backup-on-windows/. Note: The patch that the blog suggests that you download is no longer necessary starting with version 1.1.8 of rdiff-backup.

Yes, quite a few people seem to be using rdiff-backup under Mac OS X. rdiff-backup can also backup resource forks and other Mac OS X metadata to a traditional unix filesystem, which is can be a handy feature for Mac users. When rdiff-backup is used to do the restore, all of the metadata is recovered from rdiff-backup’s storage.

The easiest option is probably to use Fink http://fink.sourceforge.net/, which can install rdiff-backup automatically for you. With Fink, you should install the librsync, librsync-shlibs, python25, python25-shlibs, xattr-py25, and rdiff-backup packages. Another option is DarwinPorts https://www.macports.org/, for which you should install the py-xattr and rdiff-backup packages.

If you want to build rdiff-backup yourself, you should be able to build librsync and rdiff-backup using the standard Unix instructions. You can also see this message from Gerd Knops:

An important note if you use the Apple-provided version of Python: Apple’s version of Python will install rdiff-backup in something like /System/Library/Frameworks/Python.framework/Versions/Current/bin/rdiff-backup and rdiff-backup will not be in your $PATH. You can copy rdiff-backup out of this folder and into someplace reasonable like /usr/bin or another directory in your $PATH to use it. For a full explanation of why this happens see this post to the mailing list: https://lists.nongnu.org/archive/html/rdiff-backup-users/2007-06/msg00107.html.

No, and it doesn’t plan to. It leaves this exercise to external utilities like (we’ll list only open source tools here):

You can certainly try! Using a CIFS or smbfs mount as the mirror directory has been troublesome for some users because of the wide variety of Samba configurations. If possible, the best solution is always to use rdiff-backup over SSH in the default configuration. Using rdiff-backup in the default configuration is also guaranteed to be faster because there is lower network utilization. Rdiff-backup uses the rsync algorithm to minimize the amount of bandwidth consumed. By using smbfs or CIFS, the complete file is transferred over the network.

Under both Linux and Mac OS X, smbfs seems to be working quite well. However, it has a 2 GB file limit and is deprecated on Linux. CIFS users sometimes experience one of these common errors:

If you’re still having trouble backing up to a CIFS or smbfs mount, try searching the mailing-list archives and then sending further questions to the list.

When backing up from a case-sensitive filesystem to a case-insensitive filesystem (such as Mac’s HFS+ or Windows’s FAT32 or NTFS), rdiff-backup escapes uppercase characters in filenames to make sure that no files are accidentally overwritten. When a filesystem is case-preserving but case-insensitive, it means that it remembers that a file is named "Foo" but doesn’t distinguish between "Foo", "foo", "foO", "fOo", etc. However, filesystems such as Linux’s ext3 do treat these names as separate files.

Imagine you have a Linux directory with two files, "bar" and "BAR", and you copy them to a Mac system. You will wind up with only one file (!) since HFS+ doesn’t distinguish between the names, and the second file copied will overwrite the first. Therefore, when rdiff-backup copies files from case-sensitive to case-insensitive filesystems, it escapes the uppercase characters (eg, "M" is replaced with ";077", and "F" with ";070") so that no filename conflicts occur. Upon restore (from the Mac backup server to the Linux system), the filenames are unquoted and you will get "MyFile" back.

The only official way to remove files from an rdiff-backup repository is by letting them expire using the --remove-older-than option. Deleting increments from the rdiff-backup-data directory will prevent you from recovering those files, but shouldn’t prevent the rest of the repository from being restored.

There may be a problem with rdiff-backup and Solaris' libthread. Adding "ulimit -n unlimited" may fix the problem though. Here is a post by Kevin Spicer on the subject:

rdiff-backup can be limited by the CPU, disk IO, or available bandwidth, and the length of a session can be affected by the amount of data, how much the data changed, and how many files are present. That said, in the typical case the number/size of changed files is relatively small compared to that of unchanged files, and rdiff-backup is often either CPU or bandwidth bound, and takes time proportional to the total number of files. Initial mirrorings will usually be bandwidth or disk bound, and will take much longer than subsequent updates.

To give one arbitrary data point, when I back up my personal HD locally (about 36GB, 530000 files, maybe 500 MB turnover, Athlon 2000, 7200 IDE disks, version 0.12.2) rdiff-backup takes about 15 minutes and is usually CPU bound.

Let’s examine an example session statistics file:

StartTime and EndTime are measured in seconds since the epoch. ElapsedTime is just EndTime - StartTime, the length of the rdiff-backup session.

SourceFiles are the number of files found in the source directory, and SourceFileSize is the total size of those files. MirrorFiles are the number of files found in the mirror directory (not including the rdiff-backup-data directory) and MirrorFileSize is the total size of those files. All sizes are in bytes. If the source directory hasn’t changed since the last backup, MirrorFiles == SourceFiles and SourceFileSize == MirrorFileSize.

NewFiles and NewFileSize are the total number and size of the files found in the source directory but not in the mirror directory. They are new as of the last backup.

DeletedFiles and DeletedFileSize are the total number and size of the files found in the mirror directory but not the source directory. They have been deleted since the last backup.

ChangedFiles are the number of files that exist both on the mirror and on the source directories and have changed since the previous backup. ChangedSourceSize is their total size on the source directory, and ChangedMirrorSize is their total size on the mirror directory.

IncrementFiles is the number of increment files written to the rdiff-backup-data directory, and IncrementFileSize is their total size. Generally one increment file will be written for every new, deleted, and changed file.

TotalDestinationSizeChange is the number of bytes the destination directory as a whole (mirror portion and rdiff-backup-data directory) has grown during the given rdiff-backup session. This is usually close to IncrementFileSize + NewFileSize - DeletedFileSize + ChangedSourceSize - ChangedMirrorSize, but it also includes the space taken up by the hardlink_data file to record hard links.

There is no internal rdiff-backup option to do this. However, external utilities such as cstream can be used to monitor bandwidth explicitly. trevor\@tecnopolis.ca writes:

To only limit bandwidth in one directory, simply remove one of the cstream commands. Two cstream caveats may be worth mentioning:

Another option is to limit bandwidth at a lower (and perhaps more appropriate) level. Adam Lazur mentions The Wonder Shaper.

The amount of memory rdiff-backup uses should not depend much on the size of directories being processed. Keeping track of hard links may use up memory, so if you have, say, hundreds of thousands of files hard linked together, rdiff-backup may need tens of MB.

If rdiff-backup seems to be leaking memory, it is probably because it is using an early version of librsync. librsync 0.9.5 leaks lots of memory. Later versions should not leak and are available from the librsync homepage.

Several users have reported seeing errors that contain lines like this:

All of these users were backing up onto NFS (Network File System). I think this is probably a bug in NFS, although tell me if you know how to make rdiff-backup more NFS-friendly. To avoid this problem, run rdiff-backup locally on both ends instead of over NFS. This should be faster anyway.

Firstly, this shouldn’t happen. If it does, it indicates a corrupted destination directory, a bug in rdiff-backup, or some other serious recurring problem.

However, here is a workaround that you might want to use, even though it probably won’t solve the underlying problem: In the destination’s rdiff-backup-data directory, there should be two "current_mirror" files, for instance:

Delete the one with the earlier date. Also move the mirror_metadata file with the later date out of the way, because it probably didn’t get written correctly because that session was aborted:

The next time rdiff-backup runs it won’t try regressing the destination. Metadata will be read from the file system, which may result in some extra files being backed up, but there shouldn’t be any data loss.

When backing up, rdiff-backup needs free space in the mirror directory. The amount of free space required is usually a bit more than the size of the file getting backed up, but can be as much as twice the size of the current file. For instance, suppose you ran rdiff-backup foo bar and the largest file, foo/largefile, was 1GB. Then rdiff-backup would need 1+GB of free space in the bar directory.

When restoring or regressing, rdiff-backup needs free space in the default temp directory. Under unix systems this is usually the /tmp directory. The temp directory that rdiff-backup uses can be set using the --tempdir and --remote-tempdir options available in versions 1.1.13 and newer. See the entry for tempfile.tempdir in the Python tempfile docs for more information on the default temp directory. The amount of free space required can vary, but it usually about the size of the largest file being restored.

Usually free space errors are intelligible, like IOError: [Errno 28] No space left on device or similar. However, due to a gzip quirk they may look like ValueError: Incorrect length of data produced.

This error happens due to a bug in librsync that prevents it from handling files greater than 4 GB in some situations, such as when transferring between a 32-bit host and a 64-bit host. A patch is available from the librsync project page on Sourceforge. The CVS version of librsync also contains the patch. More information is also available in Debian bug report #355178.

If you get an error like sh: line1: rdiff-backup: command not found, but rdiff-backup is in your $PATH when you login to the remote host, it is happening because the value of bash’s $PATH is set differently when you login to an interactive shell than when you run a command remotely via SSH. For more information, read the bash manpage and look at your .bashrc and .bash_profile files.

In particular, this can happen if rdiff-backup was installed via Fink on a remote Mac OS X system. /sw/bin is magically added to your $PATH by the script /sw/bin/init.sh when you login with an interactive shell. Fink did this behind the scenes when you set it up. Simply add /sw/bin to your path manually, or copy rdiff-backup to a directory that is in your $PATH.

If you see the error “tuple index out of range” after running a command like:\ \ $ rdiff-backup -l /path/to/backup/rdiff-backup-data/\ \ then the solution is to simply remove the extra "rdiff-backup-data" from the end of the path. The list increments option, and others like it, take the path to the repository, not the path to the rdiff-backup-data directory. In the above example, you should run again with:\ \ $ rdiff-backup -l /path/to/backup\ \ If you get this error message for an unrelated reason, try contacting the mailing list.

This error message means that a Cyclic Redundancy Check failed during some operation, most likely while gzip’ing or un-gzip’ing a file. Possible causes of this error include an incomplete gzip operation, and hardware failure. A brute-force way to recover from this error is to remove the rdiff-backup-data directory. However, this will remove all of your past increments. A better approach may be to delete the particular file that is causing the problem. A command like:\ \ $ find rdiff-backup-data -type f -name \*.gz -print0 | xargs -0r gzip --test\ \ will find the failing file. For more information on this approach, see this mailing list post: https://lists.nongnu.org/archive/html/rdiff-backup-users/2007-11/msg00008.html.

If rdiff-backup fails with the message "`AssertionError: Bad index order`," it could be because the files in a directory have changed while rdiff-backup is running. Possible ways of dealing with this situation include implementing filesystem snapshots using the volume manager, excluding the offending directory, or suspending the process that is changing the directory. After the text "Bad index order", the error message will indicate which files have caused the problem.

If you get this message for an unrelated reason, try contacting the mailing list.

Like other Unix and Python programs, rdiff-backup respects the TZ environment variable, which can be used to temporarily change the timezone. On Unix, for UTC, simply set TZ=UTC in your shell, or prepend TZ=UTC to the command line used to run rdiff-backup. On Windows, set the TZ environment variable with the set TZ=UTC command in the Cmd.exe command interpreter (or in a batch script), or with $env:TZ='UTC' in PowerShell. If you want to use a different timezone than UTC, you can refer to the _tzset CRT documentation which describes in detail the format Windows expects for the value of the TZ variable.

If you’ve done something wrong in your last back-up, you have potentially two solutions to get rid of it. If you’ve backed-up a file or directory you shouldn’t have backed-up, you can remove it again using rdiff-backup-delete <repo>/<file-or-dir>; beware that all files, including all earlier versions, will be removed without any question back! If the situation is more complicated, you might want to have a look at rdiff-backup-regress which completely removes the last made backup, but beware that regression takes a long time. Note that we made a copy of this nice utility, placed under tools/misc in our Git repo, just to be sure it doesn’t get lost.

rdiff-backup expects certain qualities from a file system and checks for them. Experience shows that some file systems do fail these tests (or even at runtime) and can’t be supported: how could you rely on your backup if rdiff-backup can’t rely on the file system?

The list might grow but issues are currently known with especially slow combinations like:

File systems like VFAT or NTFS are what I call case aware file systems: they are "officially" case insensitive but they store file names in a case sensitive manner.

This becomes a problem for rdiff-backup if a file is renamed in a case sensitive manner (e.g. from MyFile to mYfIlE) because it doesn’t recognize them as the same file, but the target file system doesn’t accept both files next to each other.

You can either:

Since rdiff-backup 2.1+, a command like rdiff-backup restore --include myrestore/subdir/somefile --exclude '' myrepo/subdir myrestore/subdir isn’t possible anymore, because it could lead to data loss. Such calls are anyway equivalent to something like rdiff-backup restore --include myrestore/subdir/somefile --exclude '' myrepo myrestore. This means that the new limitation doesn’t imply a loss in feature, it only enforces a new approach without risk of losing data.

It can be as easy as importing the rdiffbackup.run module and starting the function rdiffbackup.run.main_run with the usual command line parameters as list parameter.

The rdiff-backup script itself doesn’t do it much differently and looks in a simplified manner like this:

Two open source projects exist providing a user interface for rdiff-backup.

Rdiffweb is a web interface for rdiff-backup. It can be used to browse and restore your data from the convenience of your web browser.

Minarca is a centralized backup solution. It provides an agent to automate the backup process and a server to browse and restore the data to be installed on your centralized backup server. The agent is cross-platform allowing you to seamlessly use a single solution.

If your SSH server is running on a custom port, say 2222, and you are using the OpenSSh client, you may use the URI form ssh://[user@]hostname[:port] of the hostname to connect to it, e.g.

Calling for example rdiff-backup backup loc1 server2::loc2 leads to a message WARNING: this command line interface is deprecated and will disappear, start using the new one as described with '--new --help'.

You must be using a remote location in your call, as in our example. In order to make sure that the other side understands the call, rdiff-backup uses the CLI form fitting the default API version. For example, at time of writing, rdiff-backup 2.2 uses API 200 by default and hence calls rdiff-backup --server so that the CLI can be understood by rdiff-backup 2.0. If the server side is v2.2+, then the warning message will appear.

Call rdiff-backup with the higher API and the message will disappear.

Calling for example rdiff-backup --api-version 201 backup loc1 server2::loc2 will make sure that rdiff-backup server is being called, getting rid of the warning message.

If you’re doing a backup and get an exception [Errno 28] No space left on device: …​ (or something similar under Windows), it means exactly what it states, there is not enough space to create your backup. You need to first fix the issue, then recover, then make sure that it doesn’t happen again.

In the following lines, we’ll assume you’ve called something like rdiff-backup backup /from /mnt/bak, which failed with a "No space left" error.