Installing Void on a ZFS Root

Because the Void installer does not support ZFS, it is necessary to install via chroot. Aside from a few caveats regarding bootloader and initramfs support, installing Void on a ZFS root filesystem is not significantly different from any other advanced installation. ZFSBootMenu is a bootloader designed from the ground up to support booting Linux distributions directly from a ZFS pool. However, it is also possible to use traditional bootloaders with a ZFS root.

ZFSBootMenu

Although it will boot (and can be run atop) a wide variety of distributions, ZFSBootMenu officially considers Void a first-class distribution. ZFSBootMenu supports native ZFS encryption, offers a convenient recovery environment that can be used to clone prior snapshots or perform advanced manipulation in a pre-boot environment, and will support booting from any pool that is importable by modern ZFS drivers. The ZFSBootMenu wiki offers, among other content, several step-by-step guides for installing a Void system from scratch. The UEFI guide describes the procedure of bootstrapping a Void system for modern systems. For legacy BIOS systems, the syslinux guide provides comparable instructions.

Traditional bootloaders

For those that wish to forego ZFSBootMenu, it is possible to bootstrap a Void system with another bootloader. To avoid unnecessary complexity, systems that use bootloaders other than ZFSBootMenu should plan to use a separate /boot that is located on an ext4 or xfs filesystem.

Installation media

Installing Void to a ZFS root requires an installation medium with ZFS drivers. It is possible to build a custom image from the official void-mklive repository by providing the command-line option -p zfs to the mklive.sh script. However, for x86_64 systems, it may be more convenient to fetch a pre-built hrmpf image. These images, maintained by a Void team member, are extensions of the standard Void live images that include pre-compiled ZFS modules in addition to other useful tools.

Partition disks

After booting a live image with ZFS support, partition your disks. The considerations in the partitioning guide apply to ZFS installations as well, except that

  • The boot partition should be considered necessary unless you intend to use gummiboot, which expects that your EFI system partition will be mounted at /boot. (This alternative configuration will not be discussed here.)
  • Aside from any EFI system partition, GRUB BIOS boot partition, swap or boot partitions, the remainder of the disk should typically be a single partition with type code BF00 that will be dedicated to a single ZFS pool. There is no benefit to creating separate ZFS pools on a single disk.

As needed, format the EFI system partition using mkfs.vfat(8) and the the boot partition using mke2fs(8) or mkfs.xfs(8). Initialize any swap space using mkswap(8).

It is possible to put Linux swap space on a ZFS zvol, although there may be a risk of deadlocking the kernel when under high memory pressure. This guide takes no position on the matter of swap space on a zvol. However, if you wish to use suspension-to-disk (hibernation), note that the kernel is not capable of resuming from memory images stored on a zvol. You will need a dedicated swap partition to use hibernation. Apart from this caveat, there are no special considerations required to resume a suspended image when using a ZFS root.

Create a ZFS pool

Create a ZFS pool on the partition created for it using zpool(8). For example, to create a pool on /dev/disk/by-id/wwn-0x5000c500deadbeef-part3:

# zpool create -f -o ashift=12 \
    -O compression=lz4 \
    -O acltype=posixacl \
    -O xattr=sa \
    -O relatime=on \
    -o autotrim=on \
    -m none zroot /dev/disk/by-id/wwn-0x5000c500deadbeef-part3

Adjust the pool (-o) and filesystem (-O) options as desired, and replace the partition identifier wwn-0x5000c500deadbeef-part3 with that of the actual partition to be used.

When adding disks or partitions to ZFS pools, it is generally advisable to refer to them by the symbolic links created in /dev/disk/by-id or (on UEFI systems) /dev/disk/by-partuuid so that ZFS will identify the right partitions even if disk naming should change at some point. Using traditional device nodes like /dev/sda3 may cause intermittent import failures.

Next, export and re-import the pool with a temporary, alternate root path:

# zpool export zroot
# zpool import -N -R /mnt zroot

Create initial filesystems

The filesystem layout on your ZFS pool is flexible. However, it is customary to put operating system root filesystems ("boot environments") under a ROOT parent:

# zfs create -o mountpoint=none zroot/ROOT
# zfs create -o mountpoint=/ -o canmount=noauto zroot/ROOT/void

Setting canmount=noauto on filesystems with mountpoint=/ is useful because it permits the creation of multiple boot environments (which may be clones of a common Void installation or contain completely separate distributions) without fear that ZFS auto-mounting will attempt to mount one over another.

To separate user data from the operating system, create a filesystem to store home directories:

# zfs create -o mountpoint=/home zroot/home

Other filesystems may be created as desired.

Mount the ZFS hierarchy

All ZFS filesystems should be mounted under the /mnt alternate root established by the earlier re-import. Mount the manual-only root filesystem before allowing ZFS to automatically mount everything else:

# zfs mount zroot/ROOT/void
# zfs mount -a

At this point, the entire ZFS hierarchy should be mounted and ready for installation. To improve boot-time import speed, it is useful to record the current pool configuration in a cache file that Void will use to avoid walking the entire device hierarchy to identify importable pools:

# mkdir -p /mnt/etc/zfs
# zpool set cachefile=/mnt/etc/zfs/zpool.cache zroot

Mount non-ZFS filesystems at the appropriate places. For example, if /dev/sda2 holds an ext4 filesystem that should be mounted at /boot and /dev/sda1 is the EFI system partition:

# mkdir -p /mnt/boot
# mount /dev/sda2 /mnt/boot
# mkdir -p /mnt/boot/efi
# mount /dev/sda1 /mnnt/boot/efi

Installation

At this point, ordinary installation can proceed from the "Base Installation" section. of the standard chroot installation guide. However, before following the "Finalization" instructions, make sure that the zfs package has been installed and dracut is configured to identify a ZFS root filesystem:

(chroot) # mkdir -p /etc/dracut.conf.d
(chroot) # cat > /etc/dracut.conf.d/zol.conf <<EOF
nofsck="yes"
add_dracutmodules+=" zfs "
omit_dracutmodules+=" btrfs resume "
EOF
(chroot) # xbps-install zfs

Finally, follow the "Finalization" instructions and reboot into your new system.