This tutorial shows you how to backup your file system (partially or totally) on USB key (thing works the same with harddisks or cloud storage), but we will keep USB keys for simplicity. Most important, we will also see how to restore your system from scratch in case of hard disk failure (or other cataclysms).
In the following, for each feature we will use, you will find the description of
what it does followed by the way to activate it both using its
the long options and its
the short option.
Of course, that's up to you to use either the short or the
long opton (but not both at the same time for a particular feature).
Short option begin by a single dash (-) and have only a single letter to identify them
-s. Long option begins with two dashes (--) and usually
have a descriptive word to identify them:
Short and long option may have no argument (
may have a mandatory argument which is the word following the option (
and some rare ones may have an optional argument, leading the option to either
-z or sticked with its optional argument
for long option is done by mean of the equal signe (=):
We need first to make a full backup, let's go:
Let's assume the size of the usb keys is 100 MiB, we ask dar to
split the backup in many files (also known as slices) of 100 MiB:
--slice 100M or
On your first usb key drive we want to copy the dar binary outside the backup to be able to restore it in case of hard disk failure, for example.
You can also add man pages or a copy of this tutorial, if you are scared
not to be able to
remember all the many feature of dar ;-) while find the
--help option too
sparse. Note that all the dar documentation is available on the web. OK
you need an Internet access to read it.
This make the free space on the first usb key floppy a bit smaller, I let
you make the substraction because this is subject to change from system
to system, but let's assume dar_static is less than 5 MiB, thus the
initial slice should not exceed 95 MB:
--first-slice 95M or
(Note that '-s' is lowercase for all the slices, and '-S' is UPPERCASE
meaning the initial slice only).
We need to pause between slices to change the usb key when it is
As we don't want to stick in front of the screen during the backup,
we ask dar to to ring the terminal bell when user action is needed:
We will compress data inside the backup:
-z option uses gzip compression algorithm
(gzip, bzip2, lzo, xz, lz4, zstd, and some others are
also available). Optionally, if speed is more important than archive
size, you can degrade compression specifying the compression level:
-z1 for example for gzip, or
compression level 5 with xz algorithm.
By default the maximum compression is used (
is equivalent to
Now, we want to backup the whole file system.
--fs-root / or
This tells dar that no files out of the provided directory tree will be saved. Here, it means that no files will be excluded from the backup, if no filter is specified, see below)
There are some files you probably don't want to save, like backup files
generated by emacs
--exclude "*~" --exclude ".~*" or
-X "*~" -X ".*~"
Note that you have to quote the mask for it not to be interpreted by
the shell, the
-X options do not apply to directories,
nor to path, they just apply to filenames. See also the opposite
-I option (
--include) in man page for more information.
Among these files are several sub-trees you must not save: the
file system for example, as well as the
These are virtual filesystems, saving them would only make your backup bigger filled
with useless stuff:
--prune dev --prune proc --prune sys or
-P dev -P proc -P sys
Note that path must be relative to
-R option (thus no leading '/' must
be used) Unlike the -X/-I options, the -P option applies to full file path+names. If
a directory matches -P option, all its subdirectory will also be
excluded. note also that -P can receive wildcards, and they must be
quoted not to be interpreted by the shell:
for example. Lastly, -P can also be used to exclude a plain file (if
you don't want to exclude all files of a given name using -X option):
-P home/joe/.bashrc for example would only exclude joe's .bashrc file
not any other file, while
-X .bashrc would exclude any file of that name
including joe's file. See also
options in man page for more, as well as the "file selection in brief" paragraph
More importantly we must not save the backup itself:
--prune mnt/usr or
assuming that your usb key is mounted under /mnt/usb. We could also have
excluded all files of extension "dar" which are backup generated by
-X "*.*.dar", but this would have also exclude
other dar archive from the backup, which may not always fit your need.
Now, as we previously excluded the /dev/pts /proc and /mnt/usb directories, we
would have to create these directory mount-points by hand at recovery
time to be able to mount the corresponding filesystems. But we can
better use the -D option: it changes dar's behavior by not totally
ignoring excluded directories (whatever is the feature used to exclude them)
but rather storing them as empty directory in the backup:
Thus at recovery time excluded directories will be generated automatically as an empty directories
Last, we have to give a name to this full backup. Let's call it
"linux_full" and as it is supposed to take place on the usb key, its path
will be /mnt/usb/linux_full:
--create /mnt/usb/linux_full or
Note that linux_full is not a complete filename, it is a "basename", on which dar will add a number and the ".dar" extension, this way the first slice will be a file of name linux_full.1.dar located in /mnt/usb
Now, as we will have to mount and umount the /mnt/usb file system, we must not have any process using it, in particular, dar current directory must no be /mnt/usb so we change to / for example.
All together we follow this procedure for our example:
Plug an empty usb key and mount it according to your /etc/fstab file.
Copy the dar binary to the first usb key (to be able to restore in case of big problem, like a hard disk failure) and eventually man pages and/or this tutorial.
cp `which dar_static` /mnt/usb
then, type the following:
dar -c /mnt/usb/linux_full -s 100M -S 95M -p -b -z -R / -X "*~" -X ".*~" -P dev/pts -P sys -P proc -P mnt/usb -D
Note that option order has no importance. Some options may be used several times (-X, -I, -P) some others cannot (see man page for more).
When the first slice will be done, DAR will pause, ring the terminal bell and display a message. You will have to unmount the usb key:
and replace it by an empty new one and mount it:
To be able to do that, you can swap to another virtual console pressing ALT+F? keys (if under Linux), or open another xterm if under X-Windows, or suspend dar by typing CTRL-Z and reactivating it after mounting/unmounting by typing `fg' (without the quotes).
Then proceed with dar for the next slice, pressing the <enter> key. Dar will label slices this way:
That's it! We have finished the first step: the backup, it may take a long time depending on the size of the data to backup. The following step (differential backup) however can be done often, and it will stay fast every time (OK, except if a big part of your system has changed, in that case you can consider making another full backup).
There is so many reason a backup can be useless, it may be human error, saturated disk, lack of permission, and so on. The best test is to restore the data at least once. But there are some more quick way (though less exhaustive) to test a backup:
This one is usually quick, you know the backup is readable but have to verify that all expected files are present in the output:
dar -l /mnt/usb key/linux_full
One step further you can let dar try to restore everything without effectively
restoring anything, (this mimics the
cat > /dev/null paradigm).
Doing so you validate the data and metadata of all files is not corrupted.
This is usually a good thing to add in your backup script (or more generally your
dar -t /mnt/usb key/linux_full
If using removable media of poor quality, it is
recommended to first unmount and remount removable disk, this to flush
the system cache. Else you may read data from cache (in memory) and do not detect
an error on you disk.
dar -t cannot check a single slice, it checks
all the archive. If you need to check a single slice, (for example after
burning it on DVD-RW, you can use the diff command: for example, you
have burnt the last completed slices on DVD-RW, but have just enough free space to
store one slice on disk. You can thus check the slice typing something
diff /mnt/cdrom/linux_full.132.dar /tmp/linux_full.132.dar
You can also add the
--hash command when you create the
backup (for example
it will produce for each slice a small hash file named after the slice name
"linux_full.1.dar.md5", "linux_full.2.dar.md5", etc. Then using the
unix standard command "md5sum" you can check the integrity of the slice:
md5sum -c linux_full.1.dar.md5
If all is ok for the slice on target medium (diff does not complain or md5sum returns "OK"), you can continue for dar to proceed with the next slice.
instead of testing the whole archive you could also compare it with the just saved system:
dar -d /mnt/usb key/linux_full -R /
This will compare the archive with filesystem tree located at / . Same remark as previously, it is recommended to first unmount and mount the removable media to flush the system cache.
If you backup a live filesystem, you may prefer 'testing' an archive as it will not issue errors about files that changed since the backup was made, but if you are archiving files, diffing is probably a better idea as you really compare the content of the files and you should not experiment file changes on data you are archiving as most of the time such data about to be archived is old steady data that is not likely to change.
The only thing to add is the base name of the backup we take as reference:
--ref /mnt/usb/linux_full or
Of course, we have to choose another name for that new backup, let's call
--create /mnt/usb/linux_diff1 or
Last point: if you want to put the new backup at the end of the full
backup, you will have to change the
-S option according to the
remaining space on the last usb key. suppose the last slice of linux_full takes 34MB
you have 76MB available for the first slice of the differential backup
(and always 100MB for the following ones):
--first-slice 76M or
but if you want to put the backup on a new usb key, just forget the -S option.
here we also want to produce a hash file to test each slice integrity before
removing it from hard disk (md5, sha1, sh512 are the available hash algorithm today):
--hash md5 or
All together we get:
dar -c /mnt/usb/linux_diff1 -A /mnt/usb key/linux_full -s 100M -S 76M -p -b -z -R / -X "*~" -X ".*~" -P dev/pts -P proc -P mnt/usb key -P sys -D --hash md5
The only new point is that, just before effectively starting to backup, dar will ask for the last slice of the archive of reference (linux_full), then dar will pause (thanks to the -p option) for you to change the disk if necessary and put the one where you want to write the new backup's first slice, then pause again for you to change the disk for the second slice and so on.
You can make another differential backup, taking linux_diff1 as reference
(which is called an incremental backup, while a differential backup
has always the a full backup as reference).
In this case you would change only the following:
-c /mnt/usb/linux_diff2 -A /mnt/usb key/linux_diff1
You could also decide to change of device, taking 4,4 GiB DVD-RAM... or maybe rather something more recent and bigger if you want, this would not cause any problem at all.
After some time when you get many incremental backups for a single full backup, you will have to make a new full backup, depending on your available time for doing it, or on your patient if one day you have to recover the whole data after a disk crash: You would then have to restore the full backup, then all the following incremental backup up to the most recent one. This requires more user intervention than restoring a single full backup, all is a matter of balance, between the time it takes to backup and the time it takes to restore.
Note, that starting with release 1.2.0 a new command appeared that helps restoring a few files from a lot a differential backup. Its name is dar_manager. See at the end of this tutorial and the associated man page for more.
Another solution, is when you have too much incremental backup, is to make the next backup a differential backup taking the last full_backup as reference, instead of the last differential backup done. This way, it will take less time than doing a full backup, and you will not have to restore all intermediate differential backup.
For dar, there is not difference in structure between a differential backup (having a full backup as reference) and an incremental backup (having a differential or another incremental backup as reference). This is just the way you chose the backup of reference that let you use two different words naming differently what dar considers of the the kind.
Of course, a given backup can be used as reference for several differential backup, there is no limitation in number nor in nature (the reference can be a full of differential backup).
Yet another solution is to setup decremental backups, this is let you have the full backup as the most recent one and the older ones as difference from the backup done just after them... but nothing is perfect, doing so takes much more time than doing full backup at each step but as less storage space as doing incremental backups and restoration time is as simple as restoring a full backup. here too all is a matter of choice, taste and use case.
Sorry, it arrived, your old disk has crashed. OK, you are happy because you have now a good argument to buy the very fast and very enormous very lastest hard disk available. Usually, you also cry because you have lost data and you will have to reinstall all your system, that was working so well and for so long!
If however the last backup you made is recent, then keep smiling! OK, you have installed your new hard disk and configured you BIOS to it (well at ancient time it was necessary to manually setup the BIOS with the new disk, today you can forget it).
You first need to boot your new computer with the empty disk in order to restore your data onto it. For that I would advise using Knoppix or better system rescue CD that let you boot from CD or USB key. You don't need to install something on your brand-new disk, just make partitions and format them as you want (we will detail that below). You may even change the partition layout add new ones or merge several ones into a single one: what is important is that you setup each one with enough space to hold the data to be restored in them: We suppose your new disk is /dev/sda and /dev/sdb is your removable media drive (USB key, DVD device, ...) For clarity, in the following we will keep assuming it to be a set of USB keys, it could be CD, DVD, or other disk you would do slightly the same.
Create the partition table as you wish, using
fdisk /dev/sda or
gdisk /dev/sda for a
more versatil and modern partition table.
Format the partition which will receive your data, dar is filesystem
independent, you can use ext2 (as here in the example), ext3, ext4,
ReiserFS, Minix, UFS, HFS Plus, XFS, whatever is the Unix-like
filesystem you want, even if the backed up data did not reside on such
filesystem at backup time!
copy and record in a temporary file the UUID of the generated filesystem if
the /etc/fstab we will restore in the next steps rely in that instead
of fixed path (like /dev/sda1 or /dev/mapper/...). You can also
retrieve the UUID calling
Additionally if you have created it, format the swap partition
and also record the generated UUID if necessary:
mkswap -c /dev/sda2
If you have a lot of file to restore, you can activate the swap
on the partition of your new hard drive:
Now we must mount the hard disk
mount -t ext2 /dev/hda1 /disk
As an alternative, If you want to restore your system over several partitions like /usr /var /home and / , you must create the partitions, format them as seen above and then create the directories that will be used as mounting point an mount the partitions on these directories. For example if you have / , /usr , /var and /home partitions this would look like this:
mkdir /disk/usr /disk/var /disk/home
mount /dev/sda2 /disk/usr
mount /dev/sda3 /disk/var
mount /dev/sda4 /disk/home
If the boot system used does not already include dar/libdar (unlike system rescue CD and Knoppix for example) we need to copy the dar binary from a removable medium to your disk: insert the USB key containing the dar_static binary to be able to freely change of key later on:
mount /dev/sdb /usb_key
cp /usb_key/dar_static /disk
/dev/sdb points to your usb_key drive
(run "dmesg" just after plugging the key to know which device to
use in place of the fancy /dev/sdb). We
will remove dar_static from your new hard drive at the end of
All the restored data has to go in /disk subdirectory:
The process may be long, thus it might be useful to be noticed when
a user action is required by dar:
note that -p option is not required here because if a slice is missing
dar will pause and ask you its number (If slice "0" is requested by dar, it
means the "last" slice of the backup is requested).
OK, now we have seen all the options, let's go restoring!
/disk/dar_static -x /usb_key/linux_full -R /disk -b
...and when the next USB key is needed:
...then unplug the key, plug the next one and mount it:
mount /dev/sdb /usb_key
As previously, to do that either use an second xterm virtual console or suspend dar by CTRL-Z and awake it back by the 'fg' command. Then press <enter> to proceed with dar
Once finished with the restoration of linux_full, we have to do the
same with any following differential/incremental backup. However, doing
so will warn you any time dar restores a more recent file (file
overwriting) or any time a file
that has been removed since the backup of reference, has to be removed
from file system (suppression). If you don't want to press the
<enter> key several thousand times:
(don't warn). All file will be overwritten without warning, and this is not
an issue as be restore more recent data over older one.
All together for each potential differential backups, we have to call:
/disk/dar_static -x /usb_key/linux_diff1 -R /disk -b -w
/disk/dar_static -x /usb_key/linux_diff2 -R /disk -b -w
/disk/dar_static -x /usb_key/linux...... -R /disk -b -w
Finally, remove the dar binary from the disk:
and we have to modify the /etc/fstab with the new UUID you have recorded
blkid command to get them listed and modify /etc/fstab
Last, reinstall you original boot loader from the restored data:
If you still use lilo type:
lilo -r /disk
If your boot loader is grub/grub2 type:
You can reboot you machine and be happy with you brand-new hard disk with your old precious data on it:
shutdown -r now
In this operation dar in particular restored sparse files and hard linked inodes, thus you will have no drawback and even possibly better space usage than the original filesystem as dar can even transparently convert big plain files into smaller sparse files without any impact
The Flexibly Restoring a whole system with dar document goes one step further in this direction by illustrating many use cases like, the use of LVM, LUKS encrypted filesystems, even the full restoration of a Proxmox Virtual Environment system with all its Virtual Machines
Gosh, you have remove a important file by error. Thus, you just need to restore it, not the rest of the full and differential backups.
We could as previously, try all archive starting from the full backup up to the most recent differential backup, and restore just the file if it is present in the archive:
dar -R / -x /usb/linux_full -g home/denis/my_precious_file
This would restore only the file /home/denis/my_precious_file from the full backup.
OK, now we would also have to restore from all differential backup the same way we did. Of course, this file may have changed since the full backup.
dar -R / -x /usb/linux_diff1 -g home/denis/my_precious_file
and so on, up to the last differential archive.
dar -R / -x /usb/linux_diff29 -g home/denis/my_precious_file
We will restore our lost file, starting from the most recent differential backup and *maybe* up to the full backup. Our file may or may not be present in the a differential archive as it may have changed or not since the previous version, thus we have to check if our file is restored, using the -v option (verbose):
dar -R / -x /usb/linux_diff29 -v -g home/denis/my_precious_file
If we can see a line like this:
restoring file: /home/denis/my_precious_file
Then we are good. We can stops here, because we got the most recent backup version of our lost file. Otherwise we have to continue with the previous differential backup, up to the full backup if necessary. This method has an advantage over the first one, which is not to have *in all case* the need to use all the backup done since the full backup.
OK, now you have two files to restore. No problem, just do the second method but add -r option not to override any more recent file already restored in a previous step:
dar -x /usb key/linux_diff29 -R / -r -v -g home/denis/my_precious_file -g etc/fstab
Check the output to see if one or both of your files got restored. If not, continue with the previous backup, up to the time you have seen for each file a line indicating it has been restored. Note that the most recent version of each files may not be located in the same archive, thus you might get /etc/fstab restored from linux_diff28, and /home/denis/my_precious_file restored at linux_diff27. In the case /etc/fstab is also present in linux_diff27 it would not have been overwritten by an older version, thanks to the -r option.
This option is very important when restoring more than one file using
the second method. Instead, in the first method is used (restoring first from
the full backup, then from all the following differential backups), -r
option is not so important because if overwriting occurs when you restore
lost files, you would only overwrite an older version by a newer.
If you are lazy (as I am) have a look at dar_manager (at the end of the tutorial), it relies on a database that compile the content of all of your backups. You can then ask dar_manager a particular file, files or even directories, it will look in which backup to fetch them from and will invoke dar for you on the correct backup and file set.
We have seen previously how to do differential backups. Doing so, dar asks the last slice of the archive of reference. This operation is required to read the table of contents (also known as "catalogue" [this is a French word that means "catalog" in English, I will keep this French word in the following because it is also the name of the C++ class used in libdar]) which is located at the end of the archive (thus on the last slice(s)). You have the possibility to isolate (that's it to extract) a copy of this table of content to a small file. This small file is quite exactly the same as a differential archive that holds no data in it. Let's take an example with the full backup we did previously to see how to extract a catalogue:
dar -C /root/CAT_linux_full -A /mnt/usb/linux_full -z
Note here that we used the UPPERCASE 'C' letter, by opposition the the lowercase 'c' which is used for archive creation, here we just created an isolated catalogue, which is usually a small archive. In addition, you can use -z option to have it compressed, -s and -S option to have it split in slices, -p option, -b option, but for an isolated catalogue this is not often necessary as it is usually rather small. The only thing we have seen for backup that you will not be able to do for isolation is to filter files (-X, -I, -g, -P, -[ and -] option are not available for that operation).
So what, now we have our extracted catalogue, what can we do with it? Two things:
we can use the extracted catalogue in place of the archive, as reference for a differential backup. No need to manipulate the old usb key, you can store the last's backup isolated catalogue on your hard disk instead and use it as reference for the next backup. If we had used an isolated catalogue in the previous examples, we would have built our first differential backup this way (note that here we have chose to use the CAT_ prefix to indicate that the archive is an isolated catalogue, but the choice is yours to label isolated catalogue the way you want):
dar -c linux_diff1 -A /root/CAT_linux_full ... (other options seen above stay the same)
we can use the isolated catalogue as backup of the internal catalogue if it get corrupted. Well to face to data corruption the best solution ever invented is Parchive, an autonomous program that builds parity file (same mechanism as the one used for RAID disks) for a given file. Here we can use Parchive to create a parity file for each slice. So, assuming you lack Parchive, and that you failed reading the full backup because the usb key is corrupted in the part used to store the internal catalogue, you can use an isolated catalogue as rescue:
dar -x linux_full -A /root/CAT_linux_full ...
dar -d linux_full -A /root/CAT_linux_full ...
dar -t linux_full -A /root/CAT_linux_full ...
dar -l /root/CAT_linux_full
An isolated catalogue can be built for any type of archive (full, differential or incremental archive, even for an already isolated catalogue, which I admit is rather useless). You can also create an isolated catalogue at the same time you do a backup, thanks to the -@ option:
dar -c linux_diff1 -A /mnt/usb key/linux_full -@ CAT_linux_diff1 ... (other options...)
dar -c linux_full -@ CAT_linux_full ... (other options see above stay the same for backup)
This is know as "on-fly" isolation.
dar_manager builds a database of all your archive contents, to automatically restore the latest versions of a given set of files. Dar_manager is not targeted to the restoration a whole filesystem, the best ways to restore a whole filesystem has been described above and does not rely on dar_manager. So let's use dar_manager to restore a set of files or a whole directory. First, we have to create a "database" file:
dar_manager -C my_base.dmd
This created a file "my_base.dmd" where dmd stands for Dar Manager Database, but you are free to use any other extension.
This database is created empty. Each time you make a backup, may it be full or differential, you will have to add its table of contents (aka "catalogue") to this database using the following command:
dar_manager -B my_base.dmd -A /mnt/usb/linux_full
This will add ("A" stands for "add") the archive contents to the base. In some cases you may not have the archive available but its extracted catalogue instead. Of course, you can use the extracted catalogue in place of the archive!
dar_manager -B my_base.dmd -A ~/Catalogues/CAT_linux_full
The problem however is that when dar_manager will need to recover a file located in this archive it will try to open the archive ~/Catalogue/CAT_linux_full for restoration, which does not contain any data because it is just the catalogue of the archive.
No problem in that case, thanks to the -b option we can change afterward the basename of the archive, and thanks to the -p option you can change afterward the path at any time. Let's now list the database contents:
dar_manager -B my_base.dmd -l
It shows the following:
dar path :
dar options :
archive # | path | basename
1 /home/denis/Catalogues CAT_linux_full
We should change the path of archive number 1 for dar_manager looks on the usb key drive:
dar_manager -B my_base.dmd -p 1 /mnt/usb
...and also replace the name of the extracted catalogue by the real archive name
dar_manager -B my_base.dmd -b 1 linux_full
Now we have exactly the same database as if we had use the real archive instead of its catalogue:
dar_manager -B my_base.dmd -l
dar path :
dar options :
archive # | path | basename
1 /mnt/usb linux_full
In place of using -b and -p options, you can also tell the path and the name of the real archive to use at restoration time, when you add the catalogue to the database:
dar_manager -B my_base.dmd -A ~/Catalogues/CAT_linux_full /mnt/usb/linux_full
This is done adding an optional argument. The first ~/Catalogue... is the archive where to read the catalogue from, and the second /mnt/usb/... is the name to keep for it. No access is done to this second archive at the time of the addition, thus it may stay unavailable at the time the command is typed.
You can add up to 65534 archives to a given database, and have as much base as you want.
Note that we did not yet gave important options in the database to be passed to dar. For example, you will likely restore from the root of your filesystem, therefore when called from dar_manager, dar must get the "-R /" option. This is done with:
dar_manager -B my_base.dmd -o -R /
All that follows -o is passed to dar as-is. You can see the options passed to dar when listing the database contents (-l option).
Let's now suppose that after each backup you took the time to update your database, and you now just have removed an important file by mistake.
We can restore our /home/denis/my/precious/file using dar_manager that way:
dar_manager -B my_base.dmd -r home/denis/my/precious/file
dar_manager will find the proper archive to use, and call dar with the
dar -x archive -R / -g home/denis/my/precious/file
which in turn will ask you the corresponding slices. If you want to restore
more files at a time or even a directory tree, you can add several
arguments after -r option of dar_manager:
dar_manager -B my_base.dmd -r home/denis/my/precious/file etc/fstab home/joe
Once an archive become obsolete you can delete it from the database thanks to the -D option, you can also change archive order (-m option), get a list in which is located a given file (-f option), get the list of most recent files in a given archive (-u option), and get overall statistics per archive (-s option). Lastly you can specify which dar command to use given its path (-d option), by default, dar_manager uses the PATH shell variable to choose the dar command.
A new feature for those that are really very lazy (still as I am myself): dar_manager has an interactive mode, so you don't have to remeber all these command-line switches except one:
dar_manager -B my_base.dmd
Interactive mode allow you to do all operation except restoration which can be done as previously explained.
Well, we have reached the end of this tutorial, but dar/libdar has still a lot of features to be discovered:
all this is described in much details in the following documents:
You can also find document starting from the feature point of view using the feature description page. However if you find something unclear, feel free to report or ask for help ondar-support mailing-list.