btrbk FAQ

How can I auto-mount btrfs filesystems used by btrbk?

Given that the "volume" lines in the btrbk configuration file are valid mount-points, you can loop through the configuration and mount the volumes like this:

#!/bin/sh
btrbk list volume --format=raw | while read line; do
    eval $line
    $volume_rsh mount $volume_path
done

Note that the btrbk list command accepts filters (see btrbk(1), FILTER STATEMENTS), which means you can e.g. add "group automount" tags in your configuration and dump only the volumes of this group: btrbk list volume automount.

How can I setup a debian pre-install hook?

Create a file /etc/apt/apt.conf.d/70btrbk, e.g.:

// create a btrfs snapshot before (un)installing packages
Dpkg::Pre-Invoke {"/usr/sbin/btrbk run /mnt/btr_pool/rootfs";};

In order to make sure that the snapshots are always generated and nothing is deleted, add the btrbk command line options --preserve --override=snapshot_create=always.

Why is it not possible to backup '/' (btrfs root) ?

or in other words: why does this config not work:

/etc/btrbk/btrbk.conf:

volume /
  subvolume /
    snapshot_name rootfs

ERROR: Only relative files allowed for option "subvolume".

Answer

btrbk is designed to never alter your source subvolume. In the config above, the btrbk snapshots would be created inside the source subvolume, altering it.

The same applies to any "btrfs root" mount point (subvolid=5). In the example below, you will not be able to backup /mnt/data using btrbk:

/etc/fstab:

/dev/sda1  /mnt/data  btrfs  subvolid=5 [...]

btrbk is designed to operate on the subvolumes within /mnt/data. The recommended way is to split your data into subvolumes, e.g.:

 # btrfs subvolume create /mnt/data/www
 # btrfs subvolume create /mnt/data/mysql
 # btrfs subvolume create /mnt/data/projectx

This way you make full advantage of the btrfs filesystem, as all your data now has a name, which helps organizing things a lot. This gets even more important as soon as you start snapshotting and send/receiving.

The btrbk configuration for this would be:

volume /mnt/data
  subvolume www
    [...]
  subvolume mysql
    [...]
  subvolume projectx
    [...]

Tech Answer

While btrfs root (subvolid=5) is a regular subvolume, it is still special: being the root node, it does not have a "name" inside the subvolume tree.

Here, /mnt/btr_pool is mounted with subvolid=5:

# btrfs sub show /mnt/btr_pool/
/mnt/btr_data is toplevel subvolume

# btrfs sub show /mnt/btr_pool/rootfs
/mnt/btr_pool/rootfs
        Name:    rootfs
        uuid:    [...]

How should I organize my btrfs filesystem?

There's lots of ways to do this, and each one of them has its reason to exist. Make sure to read the btrfs SysadminGuide on kernel.org as a good entry point.

btrfs root

If your linux root filesystem is btrfs, I recommend booting linux from a btrfs subvolume, and use the btrfs root only as a container for subvolumes (i.e. NOT booting from "subvolid=5"). This has the big advantage that you can choose the subvolume in which to boot by simply switching the rootflags=subvol=<subvolume> kernel boot option.

Example (/boot/grub/grub.cfg):

menuentry 'Linux' {
  linux /boot/vmlinuz root=/dev/sdb3 ro rootflags=subvol=rootfs quiet
}
menuentry 'Linux (testing)' {
  linux /boot/vmlinuz root=/dev/sdb3 ro rootflags=subvol=rootfs_testing
}

Note that btrbk snapshots and backups are read-only, this means you have to create a run-time (rw) snapshot before booting into it:

# btrfs subvolume snapshot /mnt/btr_pool/backup/btrbk/rootfs-20150101 /mnt/btr_pool/rootfs_testing

How do I convert '/' (subvolid=5) into a subvolume?

There's several ways to achieve this, the solution described below is that it guarantees not to create new files (extents) on disk.

Step 1: make a snapshot of your root filesystem

Assuming that '/' is mounted with subvolid=5:

# btrfs subvolume snapshot / /rootfs

Note that this command does NOT make any physical copy of the files of your subvolumes within "/", it will only add some metadata.

Step 2: (optional) add the toplevel subvolume to fstab

Add mount point for subvolid=5 to fstab, something like this:

/rootfs/etc/fstab:

/dev/sda1  /mnt/btr_pool  btrfs  subvolid=5,noatime  0 0

This step is not critical for a proper root change, but will save your time by preventing further configurations/reboots and manually mounting the toplevel subvolume.

Step 3: boot from the new subvolume "rootfs".

Either add rootflags=subvol=rootfs to grub.cfg, or set subvolume "rootfs" as default:

# btrfs subvolume set-default <subvolid> /

You can obtain <subvolid> via btrfs subvolume show /rootfs | grep "Subvolume ID"

Editing grub.cfg manually may lead you some troubles if you perform some actions that will fire grub-mkconfig.

Step 4: after reboot, check if everything went fine:

First check your system log for btrfs errors:

cat /var/log/messages | grep -i btrfs | grep -i error

then check if current / is our new subvolume:

# btrfs subvolume show /
        Name:                   rootfs
        ...

Great, this tells us that we just booted into our new snapshot!

# btrfs subvolume show /mnt/btr_pool
/mnt/btr_pool is toplevel subvolume

This means that the root volume (subvolid=5) is correctly mounted.

Step 5: delete old (duplicate) files

Carefully delete all old files from /mnt/btr_pool, except "rootfs" and any other subvolumes within "/mnt/btr_pool". In other words, delete any folders that are NOT LISTED by btrfs subvolume list -a /mnt/btr_pool:

# cd /mnt/btr_pool
# mkdir TO_BE_REMOVED
# mv bin sbin usr lib var ... TO_BE_REMOVED

Then reboot. If everything went fine, remove the directory:

# cd /mnt/btr_pool
# rm -rf TO_BE_REMOVED

What is the most efficient way to clone btrfs storage?

It is very common (and avisable!) to keep backups on a separate location. In some situations, is is also required to transport the data physically, either to the datacenter or to your safe in the basement.

Answer 1: Use "btrbk archive"

A robust approach is to use external disks as archives (secondary backups), and regularly run "btrbk archive" on them. As a nice side effect, this also detects possible read-errors on your backup targets (Note that a "btrfs scrub" is still more effective for that purpose).

See btrbk archive command in btrbk(1) for more details.

Note that kernels >=4.1 and <4.4 have a bug when re-sending subvolumes, make sure you run a recent/patched kernel or step 3 will fail. Read this thread on gmane (the patch provided is confirmed working on kernels 4.2.x and 4.3.x).

Answer 2: Use external storage as "stream-fifo"

This example uses a USB disk as "stream-fifo" for transferring (cloning) of btrfs subvolumes:

1. For all source subvolumes (in order of generation):

   btrfs send /source/subvolX -p PARENT > /usbdisk/streamN

2. At the target location, restore the streams (in order of generation):

   cat /usbdisk/streamN | btrfs receive /target

This approach has the advantage that you don't need to reformat your USB disk. This works fine, but be aware that you may run into trouble if a single stream gets corrupted, making all subsequent streams unusable.

Warning: Avoid using "dd" on btrfs filesystems!

If you use dd (e.g. in order to clone a partition), make sure you don't mount the cloned filesystem at the same time as the original one. You will end up having multiple filesystems sharing identical UUID, which will break things. If you really want to do this, make sure to run:

btrfstune -u /dev/sdaX

which changes the UUID of the given device. Note that the btrfs subvolumes still share identical UUID's, but at least the kernel can cope with it (see this post on stackexchange ).

Btrbk on the other hand relies on subvolume UUID's being universally unique, and uses them as hash keys for identifying and caching filesystem and subvolume trees, which leads to undefined behavior if multiple identical UUID's are processed.

I'm getting an error: Aborted: "Received UUID" is set

You probably restored a backup with send-receive, and made it read/write using btrfs property set. This is bad, as all snapshots and backups will inherit this identical "Received UUID", which results in all these subvolumes will be treated as "containing same data".

To fix this, create a "proper" snapshot:

# cd /mnt/btr_pool
# mv mysubvolume mysubvolume.broken
# btrfs subvolume snapshot mysubvolume.broken mysubvolume

Now, mysubvolume should have an empty "Received UUID". Note that in order to have a clean environment, you also need to fix all subvolumes (snapshots as well as backups) that you created with the broken subvolume.

Check if there are more broken subvolumes:

# btrfs subvolume show mysubvolume.broken
# btrfs subvolume list -a -R /mnt/btr_pool | grep <"Received UUID" from above>
# btrfs subvolume list -a -R /mnt/btr_backup | grep <"Received UUID" from above>

Either delete them (they won't be used for incremental send-receive anyways), or clean them as follows:

# btrfs subvolume snapshot listed_ro_subvol listed_ro_subvol.rw
# btrfs subvolume delete listed_ro_subvol
# btrfs subvolume snapshot -r listed_ro_subvol.rw listed_ro_subvol
# btrfs subvolume delete listed_ro_subvol.rw

Finally, don't forget to delete the broken source subvolume:

# btrfs subvolume delete mysubvolume.broken

You should now have a clean environment, and btrbk will not complain any more.