Ubuntu Desktop 20.04 supports a single ZFS boot drive out of the box. I wanted a ZFS mirror, without going through an entirely manual setup of Ubuntu as described by OpenZFS in their instructions for Ubuntu 20.04 and instructions for Ubuntu 22.04
This adds a mirror to an existing Ubuntu ZFS boot drive after the fact. It's been tested on Ubuntu 20.04 by me and all the way up to Ubuntu 22.10 by users in comments.
ZFS requires native encryption to be added at pool / dataset creation. Ubuntu 22.04 supports this during installation. Whether these instructions are suitable for mirroring such a setup has not been tested. For Ubuntu 20.04, these instructions are not suitable for creating an encrypted ZFS boot disk, please use the full instructions linked above for that. You can, however, add an encrypted dataset after the fact: You could encrypt just the portion of your file system that holds secrets.
If your use case is storage with some containers and VMs, and not a full-fledged Ubuntu install, then take a look at TrueNAS SCALE, which will manage the ZFS parts for you.
The ZFS Boot Menu project aims to provide a cleaner, FreeBSD-ish boot experience complete with boot environments and full support for native ZFS encryption. Instructions for Ubuntu 22.04 exist.
You could also boot from a regular ext4 disk, whether single or mirrored, and then use a ZFS mirror pool for just /home and /var.
ZFS has a few advantages that are good to have
- It uses checksums, which means that hardware failure and disk corruption will be detected and flagged during regular "scrub" operations
- It supports mirrors, which means that even with a failed drive, data is not lost
- It is a Copy-on-Write file system, which means that snapshots are fast to create and fast to roll back to (seconds), and only take as much space as what was written after their creation. They can be created on a per-dataset basis.
- It has the concept of datasets, making it easy to take snapshots of specific portions of the file system, as desired. Automated ZFS snapshots with a rotation lifetime make a lot of sense.
- It can expand the size of a vdev by replacing first one, then the other drive with a larger one.
ZFS functions unlike traditional file systems such as ext4. Ars Technica has a good introduction to ZFS.
- All drives will be formatted. These instructions are not suitable for dual-boot
- No hardware or software RAID is to be used, these would keep ZFS from detecting disk errors and correcting them. In UEFI settings, set controller mode to AHCI, not RAID
- These instructions are specific to UEFI systems and GPT. If you have an older BIOS/MBR system, please use the full instructions linked above
- Install from an Ubuntu Desktop 20.04 or later install USB. Ubuntu Server does not offer ZFS boot disk
- For the "Erase disk and install Ubuntu" option, click "Advanced Features" and choose "Experimental ZFS"
- Continue install as normal and boot into Ubuntu
Note: @benitogf created a handy script that automates these steps. Use his gist for feedback on that script.
All work will be done from CLI. Open a Terminal. In the following, use copy & paste extensively, it'll help avoid typos. Right-click in Terminal pastes.
- Update Ubuntu:
sudo apt update && sudo apt dist-upgrade - Find the names of your two disks:
ls -l /dev/disk/by-id. The first disk will have four partitions, the second none. - Let's set variables for those disk paths so we can refer to them in the following
DISK1=/dev/disk/by-id/scsi-disk1
DISK2=/dev/disk/by-id/scsi-disk2
- Install tools:
sudo apt install -y gdisk mdadm grub-efi-amd64
- List partitions:
sudo sgdisk -p $DISK1, you expect to see four of them - Change swap partition type:
sudo sgdisk -t2:FD00 $DISK1 - Copy partition table from disk 1 to disk 2:
sudo sgdisk -R$DISK2 $DISK1 - Change GUID of second disk:
sudo sgdisk -G $DISK2
Ubuntu 21.04 required a reboot at this point in my testing, so that
/dev/disk/by-partuuidwas correct.If you need to do that, recreateDISK1andDISK2after the reboot.
- Confirm that disk 1 partition 3 is the device in the bpool by comparing "Partition unique GUID" to the device id shown in zpool status:
sudo sgdisk -i3 $DISK1andzpool status bpool - Get GUID of partition 3 on disk 2:
sudo sgdisk -i3 $DISK2 - Add that partition to the pool:
sudo zpool attach bpool EXISTING-UID /dev/disk/by-partuuid/DISK2-PART3-GUID, for examplesudo zpool attach bpool ac78ee0c-2d8d-3641-97dc-eb8b50abd492 /dev/disk/by-partuuid/8e1830b3-4e59-459c-9c02-a09c80428052 - Verify with
zpool status bpool. You expect to see mirror-0 now, which has been resilvered
- Confirm that disk 1 partition 4 is the device in the rpool by comparing "Partition unique GUID" to the device id shown in zpool status:
sudo sgdisk -i4 $DISK1andzpool status rpool - Get GUID of partition 4 on disk 2:
sudo sgdisk -i4 $DISK2 - Add that partition to the pool:
sudo zpool attach rpool EXISTING-UID /dev/disk/by-partuuid/DISK2-PART4-GUID, for examplesudo zpool attach rpool d9844f27-a1f8-3049-9831-77b51318d9a7 /dev/disk/by-partuuid/d9844f27-a1f8-3049-9831-77b51318d9a7 - Verify with
zpool status rpool. You expect to see mirror-0 now, which either is resilvering or has been resilvered
- Remove existing swap:
sudo swapoff -a - Remove the swap mount line in /etc/fstab:
sudo nano /etc/fstab, find the swap line at the end of the file and delete it, then save with Ctrl-x - Create software mirror drive for swap:
sudo mdadm --create /dev/md0 --metadata=1.2 --level=mirror --raid-devices=2 ${DISK1}-part2 ${DISK2}-part2 - Configure it for swap:
sudo mkswap -f /dev/md0 - Place it into fstab:
sudo sh -c "echo UUID=$(sudo blkid -s UUID -o value /dev/md0) none swap discard 0 0 >> /etc/fstab" - Verify that line is in fstab:
cat /etc/fstab - Use the new swap:
sudo swapon -a - And verify:
swapon -s
- Verify grub can "see" the ZFS boot pool:
sudo grub-probe /boot - Create EFI file system on second disk:
sudo mkdosfs -F 32 -s 1 -n EFI ${DISK2}-part1 - Remove /boot/grub from fstab:
sudo nano /etc/fstab, find the line for /boot/grub and remove it. Leave the line for /boot/efi in place. Save with Ctrl-x. - Unmount /boot/grub:
sudo umount /boot/grub - Verify with
df -h,/bootshould be mounted onbpool/BOOT/ubuntu_UID,/boot/efion/dev/sda1or similar depending on device name of your first disk, and no/boot/grub - Remove /boot/grub:
sudo rm -rf /boot/grub - And create a ZFS dataset for it:
sudo zfs create -o com.ubuntu.zsys:bootfs=no bpool/grub - Refresh initrd files:
sudo update-initramfs -c -k all - Disable memory zeroing to address a performance regression of ZFS on Linux:
sudo nano /etc/default/gruband addinit_on_alloc=0toGRUB_CMDLINE_LINUX_DEFAULT, it'll likely look like this:GRUB_CMDLINE_LINUX_DEFAULT="quiet splash init_on_alloc=0". Save with Ctrl-x - Update the boot config:
sudo update-gruband ignore any errors you may see from osprober - Make sure systemd services are up to date:
sudo systemctl daemon-reload. As per @jonkiszp in comments, this is required for Ubuntu 22.10 to still boot after these changes. - Install GRUB to the ESP:
sudo grub-install --target=x86_64-efi --efi-directory=/boot/efi --bootloader-id=ubuntu --recheck --no-floppy - Disable grub-initrd-fallback.service:
sudo systemctl mask grub-initrd-fallback.service. This is the service for /boot/grub/grubenv which does not work on mirrored or raidz topologies. Disabling this keeps it from blocking subsequent mounts of /boot/grub if that mount ever fails.
- Cross fingers and reboot!
sudo reboot - Once back up, open a Terminal again and install GRUB to second disk:
sudo dpkg-reconfigure grub-efi-amd64, keep defaults and when it comes to system partitions, use space bar to select first partition on both drives, e.g. /dev/sda1 and /dev/sdb1 - And for good measure signed efi:
sudo dpkg-reconfigure grub-efi-amd64-signed, this should finish without prompting you - If you like, you can remove the primary drive and reboot. You expect reboot to take a little longer, and to be successful.
zpool statusshould show degraded pools without error
If a mirrored drive fails, you can replace it by following a similar method as adding a second drive in the first place.
First, find the id of the replacement drive with ls -l /dev/disk/by-id and create a variable for it:
NEWDISK=/dev/disk/by-id/NEWDRIVEID
The new drive may already contain ZFS or mdadm signatures. Check using sudo wipefs $NEWDISK. If that output is not empty, run sudo wipefs -a $NEWDISK.
- Copy partition table from existing disk to replacement disk:
sudo sgdisk -R$NEWDISK /dev/disk/by-id/ID-OF-EXISTING-DRIVE - Change GUID of replacement disk:
sudo sgdisk -G $NEWDISK
- Get the ID of the "UNAVAIL" disk on bpool with
zpool status bpool - Get GUID of partition 3 on the replacement disk:
sudo sgdisk -i3 $NEWDISK - Replace the failed member with that partition:
sudo zpool replace bpool EXISTING-UID /dev/disk/by-partuuid/NEWDISK-PART4-GUID, for examplesudo zpool replace bpool 6681469899058372901 /dev/disk/by-partuuid/06f5ef6d-cb69-45e8-ad3b-c69cad5c216a - Verify with
zpool status bpool. You expect to see state "ONLINE" for the pool and both devices in mirror-0.
- Get the ID of the "UNAVAIL" disk on rpool with
zpool status rpool - Get GUID of partition 4 on the replacement disk:
sudo sgdisk -i4 $NEWDISK - Replace the failed member with that partition:
sudo zpool replace rpool EXISTING-UID /dev/disk/by-partuuid/NEWDISK-PART4-GUID, for examplesudo zpool replace rpool 8712274632631823759 /dev/disk/by-partuuid/8c4ec74f-cd4d-4048-bfca-b4a58756563d - Verify with
zpool status rpool. You expect to see state "ONLINE" for the pool and both devices in mirror-0, or state "DEGRADED" for the pool with "resilver in progress" and a "replacing-0" entry under "mirror-0"
- Verify that the failed disk shows as "removed":
sudo mdadm -D /dev/md0 - Add partition 2 of the replacement disk:
sudo mdadm /dev/md0 --add ${NEWDISK}-part2 - And verify that you can see "spare rebuilding" or "active sync":
sudo mdadm -D /dev/md0
- Create EFI file system on replacement disk:
sudo mkdosfs -F 32 -s 1 -n EFI ${NEWDISK}-part1 - Install GRUB to replacement disk:
sudo dpkg-reconfigure grub-efi-amd64, keep defaults and when it comes to system partitions, use space bar to select first partition on both drives, e.g. /dev/sda1 and /dev/sdb1 - And for good measure signed efi:
sudo dpkg-reconfigure grub-efi-amd64-signed, this should finish without prompting you
If you like, test by rebooting: sudo reboot, and confirm that pools are healthy after reboot with zpool status
Similar to replacing a failed drive, just that partition 4, the rpool partition, will be bigger. Wait for resilver after replacement, then replace the second drive. Once both drives have been replaced, rpool has the new capacity.
First, find the id of the replacement drive with ls -l /dev/disk/by-id and create a variable for it:
NEWDISK=/dev/disk/by-id/NEWDRIVEID
The new drive may already contain ZFS or mdadm signatures. Check using sudo wipefs $NEWDISK. If that output is not empty, run sudo wipefs -a $NEWDISK.
- Copy partition table from existing disk to replacement disk:
sudo sgdisk -R$NEWDISK /dev/disk/by-id/ID-OF-EXISTING-DRIVE - Change GUID of replacement disk:
sudo sgdisk -G $NEWDISK - Remove partition 4:
sudo sgdisk -d4 $NEWDISK - Recreate partition 4 with maximum size:
sudo sgdisk -n4:0:0 -t4:BF00 $NEWDISK - Tell the kernel to use the new partition table:
sudo partprobe - Tell ZFS to use expanded space automatically:
sudo zpool set autoexpand=on rpool
Follow the instructions under "Replacing a failed drive", starting from "Repair boot pool". Wait for resilver to complete afterwards. Then, run through these instructions again, replacing the second drive. Once resilver is done a second time, you will have the new capacity on the rpool.
I did a walkthrough of these instructions.
I can't confirm this at the moment, but as far as I know, fully removing the disk with swap isn't even an issue.
One way to test would be to add a swap entry to fstab that points to a non-existing node. For example by doing this:
Place it into fstab:
sudo sh -c "echo UUID=FAKE-UID-HERE none swap discard 0 0 >> /etc/fstab"Note that it might need to look like a real one, no idea how fstab would handle it having a different format. Alternatively for testing use the path syntax and the path
/dev/non-existent-diskor something like that.