Full root filesystem encryption on an Armbian system
(new, fully rewritten, replaces my earlier tutorial on this topic)
MMGen (https://github.com/mmgen)
This tutorial provides detailed, step-by-step instructions for setting up full root filesystem encryption on an Armbian system. The disk can be unlocked remotely via SSH or the serial console, permitting unattended bootup.
An automated script that performs the same steps, saving you much time and effort, can be found at https://github.com/mmgen/mmgen-geek-tools
Note that unlike my earlier tutorial all steps are performed within a running Armbian system.
The tutorial is known to work with the following board/image combinations:
Orange Pi PC2
Debian Buster mainline / Ubuntu Bionic and Focal legacy
RockPi 4
Debian Buster mainline / Ubuntu Bionic and Focal legacy
RockPro 64
Ubuntu Focal mainline
Odroid HC4
Debian Buster mainline / Ubuntu Focal mainline
You may have success with other boards/images too. If so, please post the details below (or open an issue in the mmgen-geek-tools Github repository), and I’ll add your board to the list.
Requirements:
A SoC with a running, upgradeable and Internet-connected Armbian system
A blank Micro-SD card and USB card reader, or, alternatively, a blank eMMC installed on the board
The ability to edit text files and do simple administrative tasks on the Linux command line
Step 1 - Preliminaries
All steps in this tutorial are performed as root user on a running Armbian system (the “host”).
The encrypted system (the “target”) will be created on a blank micro-SD card. If your board has an eMMC not currently in use, the system can be created on it instead.
Architecture of host and target (e.g. 64-bit or 32-bit ARM) must be the same.
For best results, the host and target hardware should also be identical or similar. Building on a host with more memory than the target, for example, may lead to disk unlocking failure on the target.
If you’re building the target system for the currently running board and with the currently running image, which is the recommended approach, the two preceding points will be a non-issue.
Packages will be installed using APT, so the host machine must be Internet-connected and its clock correctly set.
Step 2 - Upgrade your system and install the cryptsetup-bin package
# apt update && apt upgrade
# apt install cryptsetup-bin
Step 3 - Get and unpack the latest Armbian image for your board
Create your build directory:
# mkdir armbenc-build && cd armbenc-build
Download the Armbian image of your choice for your board, place it in this directory and unpack:
# xz -dv *.img.xz
Step 4 - Create mount directories and set up the loop mount
Create the mount directories:
# mkdir -p mnt boot root
Determine your first free loop device:
# losetup -f
Associate the image file with the loop device name displayed by the previous command. This will be '/dev/loop0' in most cases, but if your output was different, substitute that for '/dev/loop0' in the following steps.
# losetup -P /dev/loop0 *.img
Examine the disk image using fdisk on the loop device:
# fdisk -l /dev/loop0
The output should look something like this:
Device Boot Start End Sectors Size Id Type
/dev/loop0p1 32768 3489791 3457024 1.7G 83 Linux
Make a note of the start sector (32768 in this case). You’ll need this value in the steps below.
Now mount the loop device:
# mount /dev/loop0p1 mnt
Step 5 - Copy the boot loader to the SD card
Insert the blank micro-SD card and card reader into a USB port.
Determine the SD card’s device name using 'dmesg' or 'lsblk'. We’ll assume it to be '/dev/sda', since that’s the most likely case. If your device name is different, substitute it for '/dev/sda' in the the following steps. For an eMMC, the device name will probably be '/dev/mmcblk1'.
WARNING: if '/dev/sda' refers to some other storage device, running the following commands unchanged will destroy data on that device, so always remember to substitute the correct device name!!! The best way to eliminate this danger is to disconnect all unused storage devices on the board before proceeding further.
Copy the image’s boot loader to the SD card, using the Start sector value from Step 4 as the argument for 'count':
# dd if=$(echo *.img) of=/dev/sda bs=512 count=32768
Step 6 - Partition the SD card
# fdisk /dev/sda
At the fdisk prompt, create a new DOS disk label with the 'o' command. Use the 'n' command to create a primary partition of size +200M beginning at the same Start sector as the disk image. Type 'p' to view the partition table, which should now look something like this:
Device Boot Start End Sectors Size Id Type
/dev/sda1 32768 442367 409600 200M 83 Linux
Use 'n' again to create another primary partition beginning one sector after the first partition’s end sector and filling the remainder of the card. Type 'p' once more to view the partition table:
Device Boot Start End Sectors Size Id Type
/dev/sda1 32768 442367 409600 200M 83 Linux
/dev/sda2 442368 30636031 30193664 14.4G 83 Linux
Ensure that the first partition’s Start sector matches that of the disk image (32768 in this example) and that the second partition’s Start sector is one greater than the End sector of the first (442368 and 442367, respectively, in this example). If you’ve made a mistake, use 'd' to delete a partition and start again.
Once everything looks correct, type 'w' to write the partition table to disk.
Step 7 - Copy the system to the SD card
The following commands will create a filesystem on the SD card’s boot partition and copy the boot partition data from the image file to it. Don’t forget to substitute the correct device name if necessary. If you’re building the system on an eMMC, the boot partition device is likely to be '/dev/mmcblk1p1' instead of '/dev/sda1'.
# mkfs.ext4 /dev/sda1 # or '/dev/mmcblk1p1', for an eMMC target
# e2label /dev/sda1 CRYPTO_BOOT
# mount /dev/sda1 boot
# cp -av mnt/boot/* boot
# (cd boot; ln -s . boot)
Create the encrypted root partition. When prompted for a passphrase, it’s advisable to choose an easy one like 'abc' for now. The passphrase can be changed later with the 'cryptsetup luksChangeKey' command (type 'man cryptsetup' for details) once your encrypted system is up and running.
# cryptsetup luksFormat /dev/sda2 # or '/dev/mmcblk1p2', for an eMMC target
Activate the encrypted root partition and create a filesystem on it:
# cryptsetup luksOpen /dev/sda2 rootfs # enter your passphrase from above
# mkfs.ext4 /dev/mapper/rootfs
Mount the encrypted root partition and copy the system to it:
# mount /dev/mapper/rootfs root
# (cd mnt && rsync -a --info=progress2 --exclude=boot * ../root)
# sync # be patient, this could take a while
# mkdir root/boot
# touch root/root/.no_rootfs_resize
Unmount the boot partition and image and free the loop device:
# umount mnt boot
# losetup -d /dev/loop0
Step 8 - Prepare the target system chroot
# BOOT_PART=($(lsblk -l -o NAME,LABEL | grep CRYPTO_BOOT))
# ROOT_PART=${BOOT_PART%1}2
# ROOT_UUID="$(lsblk --nodeps --noheadings --output=UUID /dev/$ROOT_PART)"
# BOOT_UUID="$(lsblk --noheadings --output=UUID /dev/$BOOT_PART)"
# cd root
# mount /dev/$BOOT_PART boot
# mount -o rbind /dev dev
# mount -t proc proc proc
# mount -t sysfs sys sys
Copy '/etc/resolv.conf' and '/etc/hosts' so you’ll have a working Internet connection within the chroot:
# cat /etc/resolv.conf > etc/resolv.conf
# cat /etc/hosts > etc/hosts
If you’re using non-default APT repositories, you may need to copy their configuration files as well so that 'apt update' and 'apt install' will use them inside the chroot. Note that you can only do this if the host and target systems have the same distro/version. If that’s not the case, you’ll have to edit the target files by hand.
# cat /etc/apt/sources.list > etc/apt/sources.list
# cat /etc/apt/sources.list.d/armbian.list > etc/apt/sources.list.d/armbian.list
If you’re using an apt proxy, then copy its configuration file too:
# cp /etc/apt/apt.conf.d/*proxy etc/apt/apt.conf.d/
Step 9 - Edit or create required configuration files in the target system
Perform the editing steps below using a text editor of your choice:
Edit 'boot/armbianEnv.txt' so that the 'rootdev', 'console' and 'bootlogo' lines read as follows. If you’ll be unlocking the disk via the serial console, then use 'console=serial' instead of 'console=display'. Note that enabling the serial console will make it impossible to unlock the disk from the keyboard and monitor, though unlocking via SSH will still work:
rootdev=/dev/mapper/rootfs
console=display
bootlogo=false
Edit 'etc/initramfs-tools/initramfs.conf'. If your board will have a statically configured IP, add the following line to the end of the file, substituting the correct IP in place of 192.168.0.88:
IP=192.168.0.88:::255.255.255.0::eth0:off
If the board will be configured via DHCP, then edit the DEVICE line as follows:
DEVICE=eth0
If host and target systems are both Debian buster, you may wish add some key modules to the initramfs to avoid a blank display at bootup time. The easiest way to do this is to add all currently loaded modules as follows:
# lsmod | cut -d ' ' -f1 | tail -n+2 > etc/initramfs-tools/modules
Retrieve the SSH public key from the remote unlocking host and copy it to the target:
# mkdir -p etc/dropbear-initramfs
# rsync yourusername@remote_machine:.ssh/id_*.pub etc/dropbear-initramfs/authorized_keys
If you want to unlock the disk from more than one host, then edit the authorized_keys file by hand, adding the required additional keys.
Create 'etc/crypttab':
# echo "rootfs UUID=$ROOT_UUID none initramfs,luks" > etc/crypttab
Create 'etc/fstab':
# echo '/dev/mapper/rootfs / ext4 defaults,noatime,nodiratime,commit=600,errors=remount-ro 0 1' > etc/fstab
# echo "UUID=$BOOT_UUID /boot ext4 defaults,noatime,nodiratime,commit=600,errors=remount-ro 0 2" >> etc/fstab
# echo 'tmpfs /tmp tmpfs defaults,nosuid 0 0' >> etc/fstab
Create the dropbear configuration file:
# echo 'DROPBEAR_OPTIONS="-p 2222"' > etc/dropbear-initramfs/config
# echo 'DROPBEAR=y' >> etc/dropbear-initramfs/config
If the target is Ubuntu bionic, then a deprecated environment variable must be set as follows:
# echo 'export CRYPTSETUP=y' > etc/initramfs-tools/conf.d/cryptsetup
Set up automatic disk unlock prompt. Performing this optional step will cause the disk password prompt to appear automatically when you log in remotely via SSH to unlock the disk. Using your text editor, create the file 'etc/initramfs-tools/hooks/cryptroot-unlock.sh' with the following contents:
#!/bin/sh
if [ "$1" = 'prereqs' ]; then echo 'dropbear-initramfs'; exit 0; fi
. /usr/share/initramfs-tools/hook-functions
source='/tmp/cryptroot-unlock-profile'
root_home=$(echo $DESTDIR/root-*)
root_home=${root_home#$DESTDIR}
echo 'if [ "$SSH_CLIENT" ]; then /usr/bin/cryptroot-unlock; fi' > $source
copy_file ssh_login_profile $source $root_home/.profile
exit 0
Save the file and execute the command:
chmod 755 'etc/initramfs-tools/hooks/cryptroot-unlock.sh'
Step 10 - Chroot into the target system, install packages and configure
Now chroot into the encrypted system. All remaining steps will be performed inside the chroot:
# chroot .
Install the cryptsetup package and the dropbear SSH server:
# apt update
# echo 'force-confdef' > /root/.dpkg.cfg
# apt --yes install cryptsetup-initramfs dropbear-initramfs # for a buster or focal image
# apt --yes install cryptsetup dropbear-initramfs # for a bionic image
# rm /root/.dpkg.cfg
Make sure everything was included in the initramfs (all three commands should produce output):
# lsinitramfs /boot/initrd.img* | grep 'usr.*cryptsetup'
# lsinitramfs /boot/initrd.img* | grep dropbear
# lsinitramfs /boot/initrd.img* | grep authorized_keys
Your work is finished! Exit the chroot and shut down the board:
# exit
# halt -p
Insert your freshly written SD card into the board’s main SD slot (or, if the target is an eMMC, just remove the SD card from that slot) and reboot.
Unlock the disk by executing the following command on your remote unlocking machine, substituting the correct IP address if necessary:
$ ssh -p 2222 root@192.168.0.88
If you performed step 9.10 above, the disk password prompt should appear automatically after login. If not, you must enter the command 'cryptroot-unlock'.
You may also unlock the disk from the target board’s console if you wish. Note, however, that certain disk images (RockPi 4 buster mainline, for example) might give you a blank display at startup, so you’ll have to enter your disk password “blindly”. This bug will hopefully be fixed in the future.
If all went well, your root-filesystem encrypted Armbian system is now up and running!
Glad everything worked! However, the bind mount wasn't necessary. Since you're still booting from the SD/eMMC, the old fstab would have worked unmodified. Can't say why you couldn't log in via SSH initially, but in any case this is a minor issue.