FreeBSD: Full Disk Encryption
Note: The system installation process would be the same for either server or a desktop. Obviously, the 3rd party software will make the difference.
Important note: Naturally, installation of the FDE carries its obvious burden in relation to the fact that the encryption/decryption key needs to be provided each time the server is restarted. Reader discretion advised!
The FreeBSD’s installer’s “Welcome screen” (tested with version 11.1) enables administrator to jump to shell and return back later by pressing
exit. However, the disks can be manually configured later.
<Install> to continue with the installation and its further options selections: keymap, hostname, system components, and partitioning.
<Shell> to escape to shell.
Available disks can be checked as follows (usually,
ada0 represents the first disk):
$ sysctl kern.disks
Disks can be checked for current partition layout using
gpart. Note: An empty disk scheme (lacking MBR or GPT) results in:
gpart: No such geom: ada0.
$ gpart show ada0
More info about disks can be found using
diskinfo as follows:
$ diskinfo -v ada0
Disk Format and Partitioning
In order to retain compatibility with other operating systems that only understood MBR (MS Windows), it was a common practice in FreeBSD to encapsulate a bsdlabel inside an MBR slice resulting in e.g.:
ada0s1a, i.e. the first label (
a) in the first slice (
1) on the first disk (
Running FreeBSD exclusively on a system, the MBR slices are not needed anymore and the disk can utilise
bsdlabels directly (sometimes called dangerously dedicated mode) e.g.:
ada0a, i.e. the first label (
a) on the first disk (
ada0p1a, i.e. first label (
a) in the first slice (
1) on the first disk (
Optionally, disk partitions can be deleted and partition tables destroyed as follows:
$ gpart delete -i PARTITIONNUMBER adaDISKID $ gpart destroy -F adaDISKID
Disk format is also performed using
gpart and WILL DELETE DATA. An MBR table format (
-s mbr) of the first SATA disk (
ada0) can be achieved running the following command:
$ gpart create -s mbr /dev/ada0
freebsd slices (aka partitions) the first of 768MB for
/boot (large enough to hold two kernels) and the rest for an encrypted
swap) can be created on the disk (
ada0) as follows:
$ gpart add -t freebsd -a 4k -s 768m ada0 $ gpart add -t freebsd -a 4k ada0
FreeBSD administrators have two main file system options available. The newest is called the Z File System (ZFS) and requires ECC RAMs, but such approach was not aligned with the goals of this project. Obviously, other members of the community do not agree with that statement. Either way, there is another (older) options called the Unix File System (
A UFS BSD partition scheme can be created on the first slice as follows (this time, the partition type is
$ gpart create -s bsd ada0s1 $ gpart add -t freebsd-ufs -a 4k ada0s1
Partition labelling can be utilised for better reference (survives computer restarts, addition of new disk etc.) as follows:
$ glabel label -v bootfs /dev/ada0s1a $ glabel label -v sysfs /dev/ada0s2
Now, a simple MBR boot loader needs to be installed into the first disk (
ada0) and a boot manager into the first active slice (
ada0s1) as follows:
$ gpart bootcode -b /boot/mbr ada0 $ gpart bootcode -b /boot/boot ada0s1 $ gpart set -a active -i 1 ada0
$ kldload aesni $ geli init -b -s 4096 -l 256 /dev/label/sysfs
Metadata backup can be found in
/var/backups/label_sysfs.eli. This file can be copied on a backup medium in order to help restore a corrupted encrypted partition should it occur in the future. Reader discretion advised!
The encrypted partition now needs to be attached using:
$ geli attach /dev/label/sysfs GEOM_ELI: Device label/sysfs.eli created.
Being attached as
/dev/label/sysfs.eli, the decrypted partition can now be modified in order to contain the
root partitions as follows (using the BSD partition scheme again):
$ gpart create -s bsd /dev/label/sysfs.eli $ gpart add -t freebsd-swap -s 2048m /dev/label/sysfs.eli $ gpart add -t freebsd-ufs /dev/label/sysfs.eli
And again, it is useful to label the partitions accordingly:
$ glabel label -v swapfs label/sysfs.elia $ glabel label -v rootfs label/sysfs.elib
Now, it is time to format the partitions as follows (the
-t flag can be used with
newfs to enable TRIM support):
$ newfs -j /dev/label/bootfs $ newfs -j /dev/label/rootfs
Having each partition formatted, the
root partitions be mounted as follows:
$ swapon /dev/label/swapfs $ mount /dev/label/rootfs /mnt/
Since the FreeBSD boot loader expects its data to reside in the
/boot directory of the boot partition, it needs to be mounted under a different directory (e.g.
/bootfs) and the
/boot directory needs to be “symlinked” to its
boot sub-directory as follows:
$ mkdir /mnt/bootfs $ cd /mnt $ ln -s bootfs/boot boot $ mount /dev/label/bootfs /mnt/bootfs $ mkdir bootfs/boot
exit command returns back to the installer.
Optionally, the whole process can be performed for the GPT scheme as follows:
$ ### GTP scheme format $ gpart create -s gpt /dev/ada0 $ ### protective MBR installation (backward compatibility) + 2nd stage boot loader $ gpart add -t freebsd-boot -a 4k -s 64k ada0 $ gpart bootcode -b /boot/pmbr -p /boot/gptboot -i 1 ada0 $ ### unencrypted /bootfs directory to hold kernel(s) $ gpart add -t freebsd-ufs -a 4k -s 768m ada0 $ ### encrypted system directory (for root and swap slices) $ gpart add -t freebsd-ufs -a 4k ada0 $ ### labelling for better reference $ glabel label -v bootld /dev/ada0p1 $ glabel label -v bootfs /dev/ada0p2 $ glabel label -v sysfs /dev/ada0p3 $ ### encryption (loading AESNI if available) $ kldload aesni $ geli init -b -s 4096 -l 256 label/sysfs $ geli attach label/sysfs $ ### bsdlabels for swap & root partitions $ gpart create -s bsd label/sysfs.eli $ gpart add -t freebsd-swap -s 2048m label/sysfs.eli $ gpart add -t freebsd-ufs label/sysfs.eli $ ### labelling for better reference $ glabel label -v swapfs label/sysfs.elia $ glabel label -v rootfs label/sysfs.elib $ ### formatting (-t flag to newfs to enable TRIM support) $ newfs -j /dev/label/bootfs $ newfs -j /dev/label/rootfs $ ### mounting $ swapon /dev/label/swapfs $ mount /dev/label/rootfs /mnt/ $ mkdir /mnt/bootfs $ cd /mnt $ ln -s bootfs/boot boot $ mount /dev/label/bootfs /mnt/bootfs $ mkdir bootfs/boot $ exit # back to the installer
The trickiest part is luckily over. Since the installer expects the root of the filesystem to be mounted in
/mnt, which has been just done, it installs the necessary files (base, kernel, libs, ports etc.) there. Consequently, it prompts for further details necessary to finish the installation process (such as root password, network interface setup, DNS setup, services setup, user setup, time zones).
Unfortunately, the installer does not make sure the proper setup is in the
/boot/loader.conf files. Therefore, before system restart, a manual configuration needs to be performed.
/dev directory is suddenly empty (
null) and so is the
/mnt directory like the following examples:
$ ls -lA /dev total 8 -rw-r--r-- 1 root wheel 8 Jan 1 00:00 null $ ls -lA /mnt total 0
the installer needs to be completed by the
exit command. And on the very last screen instead of “Reboot”, the “Live CD” option needs to be selected.
root can now log in (with empty password) in order to manually configure the mount points in
/etc/fstab as follows:
$ echo "# Device Mountpoint FStype Options Dump Pass#" > /mnt/etc/fstab $ echo "/dev/label/bootfs /bootfs ufs rw,noatime 1 1" >> /mnt/etc/fstab $ echo "/dev/label/swapfs none swap sw 0 0" >> /mnt/etc/fstab $ echo "/dev/label/rootfs / ufs rw,noatime 1 1" >> /mnt/etc/fstab
and boot loader parameters as follows:
$ echo 'aesni_load="YES"' > /mnt/boot/loader.conf $ echo 'geom_eli_load="YES"' >> /mnt/boot/loader.conf $ echo 'geom_eli_passphrase_prompt="YES"' >> /mnt/boot/loader.conf $ echo 'vfs.root.mountfrom="ufs:/dev/label/rootfs"' >> /mnt/boot/loader.conf
Finally, the system can be restarted now.
It may be useful to visit the FreeBSD: Post Installation Steps.
There is also a brief manual for FreeBSD: Full Disk Encryption with UEFI.