Root on DRBD


I recently started using DRBD (Distributed Replicated Block Device) on Debian Linux to have a setup in which there are two servers: One server which hosts some virtual machines, and its hot-standby companion which holds exactly the same data and can take over if the hardware of the master server dies.

Now, it is clear how to setup DRBD for the RAID array which holds all the data — DRBD’s documentation is really good. What remained unclear to me, though, is how I can also use DRBD for the root file system. Otherwise, I’d need to put in some extra effort to remember to replicate all root filesystem changes, which makes the whole setup much more complex to use.

I suspect people are deploying machines like this with root filesystems that are centrally managed by puppet or similar.

Instead, I decided to also use DRBD for the root device. While that setup is largely undocumented and not recommended on the DRBD mailing list, for experienced Linux administrators, it is not THAT complex. Essentially, you need to shrink the existing root filesystem, create the DRBD metadata and then change the initramfs so that it will start DRBD before mounting the root filesystem.

Shrinking the root filesystem

To calculate the size to which you have to shrink the existing filesystem, you can use the following script which performs the calculation documented in the DRBD manual:


which bc >/dev/null 2>&1
if [ ! $? -eq 0 ]; then
    echo "Error: bc is not installed"
    exit 1

if [ $# -lt 1 ]; then
    echo "Error: Please supply block device path"
    echo "Eg. /dev/vg1/backups"
    exit 1


SECTOR_SIZE=$( blockdev --getss $DEVICE )
SECTORS=$( blockdev --getsz $DEVICE )
MD_SIZE=$( echo "((($SECTORS + (2^18)-1) / 262144 * 8) + 72)" | bc )
FS_SIZE=$( echo "$SECTORS - $MD_SIZE" | bc )

MD_SIZE_MB=$( echo "($MD_SIZE / 4 / $SECTOR_SIZE) + 1" | bc )
FS_SIZE_MB=$( echo "($FS_SIZE / 4 / $SECTOR_SIZE)" | bc )

echo "Filesystem: $FS_SIZE_MB MiB"
echo "Filesystem: $FS_SIZE Sectors"
echo "Meta Data:  $MD_SIZE_MB MiB"
echo "Meta Data:  $MD_SIZE Sectors"
echo "--"
echo "Resize commands: resize2fs -p "$DEVICE $FS_SIZE_MB"M"

You might need to boot the system using a live system so that you can shrink the filesystem. ext4 for example does not support online shrinking.

Configure the DRBD resource

After rebooting into the system with the shrinked root filesystem, you need to configure the DRBD resource itself. Here is what I use:

cat > /etc/drbd.d/root.res <<'EOT'
resource root {
       handlers {
               pri-on-incon-degr "/usr/lib/drbd/";
               pri-lost-after-sb "/usr/lib/drbd/";
               local-io-error "/usr/lib/drbd/";
       startup {
               become-primary-on master;
               # Wait 10 seconds on boot until the peer connects.
               wfc-timeout 10;
       net {
           # While DRBD uses TCP, it might not detect all errors when
           # checksum offloading is enabled. CRC32 is computationally
           # cheap enough to just turn it on.
               data-integrity-alg crc32c;
       syncer {
               rate 100M;
               verify-alg crc32c;
       on master {
               device /dev/drbd0;
               disk /dev/vda2;
               meta-disk internal;
       on slave {
               device /dev/drbd0;
               disk /dev/vda2;
               meta-disk internal;

Configuring the initramfs hook

The first script we create is the one which will be placed in the initramfs itself. It needs to set the correct hostname, setup the ethernet interface, possibly start mdadm, then create the DRBD devices and finally mount the root filesystem:

cat > /usr/share/initramfs-tools/scripts/drbd <<'EOT'
# vim:ts=4:sw=4:noet
# DRBD mounting




    [ "$quiet" != "y" ] && log_begin_msg "Running /scripts/drbd-premount"
    run_scripts /scripts/drbd-premount
    [ "$quiet" != "y" ] && log_end_msg

    ifconfig eth0 up
    ifconfig eth0 netmask

    hostname master

    # In case you are using mdraid:
    #mdadm --assemble --scan

    /sbin/drbdadm up all
    /sbin/drbdadm sh-b-pri all

    for x in $(cat /proc/cmdline); do
        case $x in

    mount -t ext4 ${DRBDROOT} ${rootmnt}

    [ "$quiet" != "y" ] && log_begin_msg "Running /scripts/drbd-top"
    run_scripts /scripts/drbd-top
    [ "$quiet" != "y" ] && log_end_msg

    modprobe drbd
    # For DHCP
    modprobe af_packet

    wait_for_udev 10

    # Default delay is around 180s

    # loop until nfsmount succeeds
    while [ ${retry_nr} -lt ${delay} ] && [ ! -e ${rootmnt}${init} ]; do
        [ "$quiet" != "y" ] && log_begin_msg "Retrying drbd mount"
        /bin/sleep 1
        retry_nr=$(( ${retry_nr} + 1 ))
        [ "$quiet" != "y" ] && log_end_msg

    [ "$quiet" != "y" ] && log_begin_msg "Running /scripts/drbd-bottom"
    run_scripts /scripts/drbd-bottom
    [ "$quiet" != "y" ] && log_end_msg

After reading the script, it should be clear to you why such a script is not normally included in distributions nor recommended: The dependencies are hard to set up in a generic way (e.g. configuring the network, starting RAID arrays, etc.).

The second script will run every time we generate a new initramfs and include the necessary tools and files.

cat > /usr/share/initramfs-tools/hooks/drbd <<'EOT'


       echo "$PREREQ"

case $1 in
       exit 0

. /usr/share/initramfs-tools/hook-functions

copy_exec /sbin/drbdadm
copy_exec /sbin/drbdmeta
copy_exec /sbin/drbdsetup

cp -R /etc/drbd.* "${DESTDIR}/etc/"
mkdir -p "${DESTDIR}/var/lib/drbd"
cp -p /var/lib/drbd/node_id "${DESTDIR}/var/lib/drbd/node_id"

exit 0

Afterwards, use update-initramfs -u to generate a new initramfs. You can verify that the new files are included in the initramfs by using lsinitramfs /boot/initrd.img-$(uname -r).

Without any further changes, nothing will change when you reboot.

Creating the metadata (once)

An easy way to create the metadata is to stop booting in the initramfs and use the provided shell. Reboot the machine, then, in grub, add the parameters break=premount boot=drbd drbdroot=/dev/drbd0, then run the following commands in the resulting shell:

modprobe drbd
ifconfig eth0 up
ifconfig eth0 netmask
hostname master
drbdadm up root
drbdadm -- --overwrite-data-of-peer primary root
mount -t ext4 /dev/drbd0 /root

Afterwards, your system should boot normally.

Boot parameters

To make the changes persistent, modify GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub to include boot=drbd drbdroot=/dev/drbd0. Afterwards, run update-grub.

That’s it. Enjoy your root on DRBD :-).