hi, i'm rob
gregory-lee-bartholomew/fedora-on-zfs
Fedora-on-ZFS
Fedora-on-ZFS is a script for automating the installation of Fedora Linux on a ZFS filesystem.
This script requires at least two physical drives. It will create a mirrored ZFS filesystem pool on the supplied drives and then install Fedora Linux on the new pool (i.e., root on ZFS).
It is possible to use ZFS with a single drive. However, ZFS will suspend all I/O to the drive if an error is detected and your system will freeze. With multiple drives, ZFS will "self heal" errors by fetching data from other drives and continue functioning normally.
Consider using Btrfs if you do not need multi-drive redundancy or support for degraded booting (i.e., booting your system when a drive has failed).
See YouTube: OpenZFS Basics if you are new to ZFS.
Installation
The following commands, if run from a Fedora Linux live image, will download this script and use it to install a minimal Fedora Linux OS with a mirrored-disk configuration on /dev/sdX and /dev/sdY. All existing data on /dev/sdX and /dev/sdY will be erased. You might need to install the git-core package if the git command is unavailable. TIP: Try running lsblk -o +model if you are unsure what device paths correspond to the disks on which you want to install Fedora Linux.
$ sudo -i
# git clone https://github.com/gregory-lee-bartholomew/fedora-on-zfs.git
# cd fedora-on-zfs
# ./prereqs
# ./install fedora-disk-minimal.ks /dev/sdX /dev/sdY
The script will fetch a fresh copy of the Fedora kickstarts from https://pagure.io/fedora-kickstarts/ if the fedora-kickstarts repo has not already been cloned. Other fedora-disk-*.ks kickstart scripts should also work, but they will take longer to complete. An alternative to using the default Fedora Workstation/KDE/etc kickstarts would be to first get your root on ZFS Fedora Linux install working with the minimal kickstart and then use a command like sudo dnf --exclude=kernel-core install @kde-desktop-environment; systemctl set-default graphical.target to upgrade it to one of the larger desktop enviornments (use dnf environment list to list available enviroments).
This script does not have to be run from a live image, but that might be the easiest option since the Fedora Workstation live image has the prerequisite software (curl, rpm2cpio, gpg, qemu, etc.).
This script will configure Fedora Linux with the systemd-boot bootloader by default. It will also create separate ESPs at the start of each physical drive and install bootsync to keep them syncronized. With this configuration, it will be possible to boot your system in case one of the drives fails (regardless of which drive failed).
If you select the fedora-disk-minimal.ks kickstart, you will be prompted if you would like to encrypt your home directory and install the homelock script to automate the locking and unlocking of your home directory when you sign in or sign out. If you answer y, the homelock script will be added to the PAM stack and it will use your sign-in credentials to unlock your home directory. When/if you change your account password, you will need to manually change the password on your ZFS home-directory filesystem to match. Homelock only works when signing in on the console, which is why the option to install the homelock script is only presented if you use the fedora-disk-minimal.ks kickstart. If you need the homelock script and a graphical interface, you can launch the Sway window manager from your console (TTY). You will be asked if you would like to add the Sway window manager with additional scripting to autostart it after signing in on TTY1 if you select the fedora-disk-minimal.ks kickstart.
This script will also configure the installed system to exclude ZFS and the Linux kernel from the normal sudo dnf update command. This is done because it is not uncommon for ZFS updates to fail on Fedora Linux when newer Linux kernels are availabe before ZFS is compatible with them. Instead, the user should use zfs-update to update ZFS and kernel-update to update the Linux kernel. These overrides and scripts are installed under /usr/local/bin. You can remove those scripts to revert this behavior if you so choose.
Note: This installation script will install the N-1 release of Fedora Linux by default. This is done intentionally because it can happen that the newest release will come with a kernel that is not yet supported by ZFS. All security patches are backported to the N-1 release. If you want to try the newest release, you can specify the release version number of Fedora Linux you want to install as the first parameter to the install script. However, my recommendation would be to install the N-1 release and then run dnf upgrade --releasever=<N> instead. (And take a snapshot of root/0 before upgrading so that you can roll back your root filesystem in case it doesn't work. 😉)
Note: This installation script configures UEFI booting by default. If you need legacy BIOS booting, uncomment the BIOSBOOT='biosboot.ks' line in config/evariables.conf. Most PCs less than a decade old support the newer UEFI boot protocol.
Note: Due to the nature of how the Fedora Linux live images work, and because this script uses QEMU to run another instance of the Fedora Linux OS, this script requires a great deal of RAM to run to completion. If your system does not have sufficiant RAM for this script, you can work around the problem by attaching an extra USB drive and using it to create extra "virtual" (swap) memory. Use mkswap --force --label swap /dev/sdZ to format the USB device as swap memory. Then reboot the Fedora Linux live image and add the kernel parameter systemd.swap-extra=/dev/disk/by-label/swap to activate it. (You might want to add 3 to the list of kernel parameters while you are at it to avoid loading the full graphical environment. Recent Fedora Linux live images will let you sign in with root and a blank password.) The additional memory should be visible in the output of the free -m command. Warning: Adding swap memory as a workaround for insufficient RAM will make the installation very slow! If, after adding extra swap memory, QEMU errors out with something along the lines of phys-bits too low, you probably added too much swap memory. Try partitioning the USB drive first to limit the size of the swap image (e.g. sgdisk -z -n 0:0:+8GiB -t 0:8200 -c 0:swap /dev/sdZ, then mkswap --force --label swap /dev/sdZ1). Also, you might need to add rd.live.overlay.overlayfs to the kernel command line and run mount -o remount,size=3G /run if the install script (curl) runs out of space when downloading the boot images. Using these parameters, I have succeeded in running this script on a 2008 PC with 4GiB of RAM. (Don't forget to uncomment BIOSBOOT in config/evariables.conf if necessary.)
Note: Network connectivity problems have been reported when attempting to run this script from within VirtualBox with virtualbox-guest-additions installed. The open issue report can be found here and if anyone knows how to resolve the problem, I welcome your feedback. 🙂
Helper Scripts (oscp and osrm)
This installer now provides two additional helper scripts to simplify replicating your root filesystem (i.e. your operating system). The scripts expect the entire OS (not counting user data like /home/linus) to be under one dataset which is the default configuration for this installer script. These helper scripts cannot handle replicating the OS if you have customized the installation to split subdirectories like /var out into separate datasets.
Invoking the oscp and osrm scripts is straight forward.
Use the following (substituting your desired source and destination filesystems) to duplicate an OS installation.
$ sudo -i
# oscp root/0 root/1
# exit
Use the following (substituting the desired filesystem name) to delete an OS installation.
$ sudo -i
# osrm root/1
# exit
These helper scripts are written in Bash and they are saved under /usr/local/bin.
I recommend creating separate user accounts for the separate OS installations if you will be using any of the large desktop environments such as GNOME or KDE. The following commands demonstrate how to add a user to a Fedora-on-ZFS installation.
$ sudo -i
# USERNAME='douglas'
# FULLNAME='Douglas Adams'
# zfs create -o mountpoint=legacy root/$USERNAME
# sed -i "$ a root/$USERNAME /home/$USERNAME zfs nofail 0 0" /etc/fstab
# systemctl daemon-reload
# mkdir /home/$USERNAME
# mount /home/$USERNAME
# chmod 0700 /home/$USERNAME
# shopt -s dotglob
# cp -v -a /etc/skel/* /home/$USERNAME
# shopt -u dotglob
# useradd --home-dir /home/$USERNAME --no-create-home --comment "$FULLNAME" $USERNAME
# chown -R $USERNAME: /home/$USERNAME
# restorecon -r /home/$USERNAME
# passwd $USERNAME
# exit
If you chose to install homelock and you want the new user's home directory encrypted, you can replace -o mountpoint=legacy with -o mountpoint=/home/$USERNAME, -o encryption=on, -o keylocation=prompt, and -o keyformat=passphrase when running the zfs create command. Be sure to use the same password for the ZFS filesystem and the user account. Then update /etc/security/homelock.conf and add the new username to the USERS list. You will also need to skip the four commands following zfs create if you set the non-legacy mountpoint. (Do not skip the chmod command.)
Caution: Legacy mounts (-o mountpoint=legacy) are recommended for unencrypted home directories when multiple operating systems are installed on one ZFS pool. The legacy option allows the mountpoint to be tied to a specific OS instance by being listed in its /etc/fstab configuration file. Home directories configured with non-legacy mountpoints (and canmount=on) will automount on all operating system instances. Because the UID → user mapping might not be unique accross all operating system instances, it is possible that a home directory configured to use non-legacy mounting (and without encryption) could allow access by the wrong user.
Operating System Recover (osrc)
The osrc script is a variant of the oscp script that is designed for use when transferring a Fedora-on-ZFS installation from another PC or when restoring a backup for which the current system no longer has valid boot menu entries. The osrc script will reset the machine-id of the provided Fedora-on-ZFS installation and generate new boot menu entries for it.
Transferring a Fedora-on-ZFS installation from one PC to another
Here is an example of how you would use osrc to restore an OS snapshot that was transferred from one PC to another.
[root@MACHINE-A ~]# zfs snapshot "root/0@$(date +%F)"
[root@MACHINE-A ~]# zfs send "root/0@$(date +%F)" | ssh MACHINE-B "zfs receive -v -u -o canmount=noauto -o mountpoint=/ root/1"
[root@MACHINE-B ~]# osrc root/1
The osrc script will work when transferring a Fedora-on-ZFS installation from a legacy BIOS system to a newer UEFI system or vice versa, but a few extra steps are required to fix the bootsync service after booting the restored system for the first time.
When transferring an OS instance from a legacy BIOS system to a UEFI system, the following commands will be required to fix-up the bootsync service.
[root@MACHINE-B ~]# sed -i '/\s\/boot\s/ d' /etc/fstab
[root@MACHINE-B ~]# systemctl enable bootbind.service
When transferring an OS instance from a UEFI system to a legacy BIOS system, the following commands will be required to fix-up the bootsync service.
[root@MACHINE-B ~]# systemctl disable bootbind.service
[root@MACHINE-B ~]# echo '/boot@a /boot none bind,x-systemd.before=bootsync.service,nofail 0 0' >> /etc/fstab
Restoring a backup Fedora-on-ZFS installation
The osrc script can also be used to restore a Fedora-on-ZFS backup filesystem image to a new computer with blank SSDs. You would need to boot a Fedora Linux Live image, install ZFS, manually partition the SSDs, manually create a ZFS root pool, transfer your backup to the new root pool, mount the restored filesystem and grab a copy of the osrc script, then call the osrc script and provide it the name of the to-be-recovered Fedora-on-ZFS root filesystem and the device node paths to the blank EFI System Partitions created earlier.
The osrc script will label and format the EFI System Partitions and then it will install systemd-boot on them.
The osrc script is not capable of restoring a Syslinux boot loader for legacy boot. To restore a legacy boot system, you would have to run the full Fedora-on-ZFS installation script, boot the newly-installed system and then use osrc to restore the backup.
Below is an example that demonstrates how to run osrc from a Fedora Live image and use it to restore a Fedora-on-ZFS backup filesystem image on a new PC with blank SSDs.
- Once you've booted the Live image, follow the instructions at OpenZFS -- Getting Started -- Fedora to install ZFS. The instructions are repeated below for convenience, but the version below might be out of date.
# mount -o remount,size=3G /run
# rpm -e --nodeps zfs-fuse
# dnf install -y https://zfsonlinux.org/fedora/zfs-release-2-6$(rpm --eval "%{dist}").noarch.rpm
# dnf install -y kernel-devel-$(uname -r | awk -F'-' '{print $1}')
# dnf install -y zfs
# modprobe zfs
- Partition the new SSDs.
WARNING: The wipefs command below will instantly destroy all data on the drives.
TIP: Connect your backup drive after you have run the following commands. 😉
NOTE: You'll need to substitute the X and Y in the below DRIVES=... line with the paths to the drives you want to erase as listed in the output of the lsblk command.
# lsblk --nodeps --paths -o name,size,model
# DRIVES=('/dev/sdX' '/dev/sdY')
# wipefs --all ${DRIVES[@]/*/&*} 1> /dev/null
# dnf install -y gdisk
# POOL='root'
# for i in $(seq 0 $((${#DRIVES[@]}-1))); do sgdisk -Z -n 0:0:+4GiB -t 0:ef00 -n 0:0:0 -t 0:8304 -c 0:"$(printf "$POOL@\x$(printf '%x' $((97+$i)))")" -p "${DRIVES[$i]}"; echo; done
- Create a new ZFS root pool on the root@a and root@b partitions that you created in the previous step.
# zgenhostid
# zpool create -f -m none -R /mnt -o ashift=12 -O acltype=posix -O canmount=off -O compression=on -O dnodesize=auto -O relatime=on -O xattr=sa "$POOL" mirror /dev/disk/by-partlabel/"$POOL"@[a-z]
# zpool list -v
- Transfer your backup to the new root pool.
# mkdir /tmp/backup
# mount /dev/disk/by-partlabel/backup /tmp/backup
# ls /tmp/backup
[email protected]
# cat /tmp/backup/[email protected] | zfs receive -v -u -o canmount=noauto -o mountpoint=/ "$POOL/0"
receiving full stream of root/1@2025-04-09 into root/0@2025-04-09
received 4.26G stream in 28.80 seconds (152M/sec)
* The above example assumes your backup is stored on a partition that is labeled "backup". You'll likely need to adjust the mount command to match the path to your personal backup device.
- Mount the restored ZFS filesystem and grab a copy of the osrc script. Alternatively, you could clone this repo and grab a copy of the
osrcscript from thesupplementssubdirectory.
# zfs mount "$POOL/0"
# cp /mnt/usr/local/bin/osrc ~
- Call the
osrcscript.
# zfs unmount "$POOL/0"
# ~/osrc "$POOL/0" "${DRIVES[@]/*/&1}"
WARNING: This script will generate a new machine id for root/0 and create new boot menu entries on /dev/sdb1 /dev/sdc1.
Do you wish to continue? [y/n]: y
No filesystem detected on /dev/sdb1.
Format /dev/sdb1 with a VFAT filesystem? [y/n]: y
mkfs.fat 4.2 (2021-01-31)
A BLS Type 1 boot loader was not detected on /dev/sdb1
Install systemd-boot on /dev/sdb1? [y/n]: y
The systemd-boot-unsigned package is not available.
Install the systemd-boot-unsigned package now? [y/n]: y
Package Arch Version Repository Size
Installing:
systemd-boot-unsigned x86_64 256.12-1.fc41 updates 190.6 KiB
Transaction Summary:
Installing: 1 package
[1/3] Verify package files 100% | 250.0 B/s | 1.0 B | 00m00s
[2/3] Prepare transaction 100% | 3.0 B/s | 1.0 B | 00m00s
[3/3] Installing systemd-boot-unsigned-0:256.12-1.fc41.x86_64 100% | 433.9 KiB/s | 191.8 KiB | 00m00s
No filesystem detected on /dev/sdc1.
Format /dev/sdc1 with a VFAT filesystem? [y/n]: y
mkfs.fat 4.2 (2021-01-31)
A BLS Type 1 boot loader was not detected on /dev/sdc1
Install systemd-boot on /dev/sdc1? [y/n]: y
Creating a new boot menu entry and initramfs image on /dev/sdb1 ...
Entry created. Reboot to select and use Fedora Linux 40 (Forty) (root/0).
Creating a new boot menu entry and initramfs image on /dev/sdc1 ...
Entry created. Reboot to select and use Fedora Linux 40 (Forty) (root/0).
Current BIOS boot device list:
BootCurrent:
Timeout: 2 seconds
BootOrder:
TIP: You can run efibootmgr -b XXXX -B immediately after this script terminates to remove old or duplicate entries from your BIOS boot device list.
Add SD-BOOT B f9569c05-7138-4568-9baf-e3fa15597d60 (/dev/sdc1) /efi/systemd/systemd-bootx64.efi to the list? [y/n]: y
Current BIOS boot device list:
BootCurrent: 0009
Timeout: 2 seconds
BootOrder: 0000
Boot0000* SD-BOOT B HD(1,GPT,f9569c05-7138-4568-9baf-e3fa15597d60,0x800,0x800000)/\efi\systemd\systemd-bootx64.efi
Add SD-BOOT A 8c86655e-cf0e-4dce-8f84-a449d7d3ae6f (/dev/sdb1) /efi/systemd/systemd-bootx64.efi to the list? [y/n]: y
Current BIOS boot device list:
BootCurrent: 0009
Timeout: 2 seconds
BootOrder: 0001,0000
Boot0000* SD-BOOT B HD(1,GPT,f9569c05-7138-4568-9baf-e3fa15597d60,0x800,0x800000)/\efi\systemd\systemd-bootx64.efi
Boot0001* SD-BOOT A HD(1,GPT,8c86655e-cf0e-4dce-8f84-a449d7d3ae6f,0x800,0x800000)/\efi\systemd\systemd-bootx64.efi
Set root/0 as the default systemd-boot menu option? [y/n]: y
The default boot menu entry is now root/0.
# zpool export "$POOL"
# reboot
Operating System Create (oscr)
The oscr script will initialize a new minimal Fedora Linux installation and create new boot menu entries. It will install the Fedora Linux release version that is listed in the /etc/os-release file of the current system.
Below is an example that shows how the oscr script could be run to create a new root/1 filesystem containing a new minimal Fedora Linux installation.
$ sudo -i
# oscr root/1
# exit
As an example, after you have initialized a new Fedora Linux installation with oscr, you could reboot your computer and select it from the boot menu. Then you could sign-in as root on the console and use a command such as dnf --exclude=kernel-core install @kde-desktop-environment; systemctl set-default graphical.target to add the packages for a more complete operating system. You would also need to create a new user account for use with the new desktop environment.
The oscr script can also be run from a Fedora Live image. To do so, first complete the steps 1 (Install ZFS), 2 (Partition the new SSDs), and 3 (Create a new ZFS root pool) from the Restoring a backup Fedora-on-ZFS installation section above. Then use to the following commands to download and run the oscr script.
# dnf install -y git-core
# git clone https://github.com/gregory-lee-bartholomew/fedora-on-zfs.git
# cp fedora-on-zfs/supplements/oscr ~
# setenforce permissive
# ~/oscr "$POOL/0" "${DRIVES[@]/*/&1}"
# zpool export "$POOL"
# reboot
Tips & Tricks
Below are some optional enhancements that you might find useful on your Fedora-on-ZFS systems.
Enable automated snapshots of your OS
You can run the following command to enable automatic snapshoting of your root dataset whenever the kernel is updated.
$ sudo sed -i '1 { s/$/ bootfs.snapshot/; }' /etc/kernel/cmdline
$ cat /etc/kernel/cmdline
root=zfs:root/0 quiet rhgb bootfs.snapshot
After you have made the above change, a snapshot of your root filesystem will be created whenever you boot your system with a new kernel. The kernel version will be used for the snapshot name.
$ sudo zfs list -t snapshot
NAME USED AVAIL REFER MOUNTPOINT
root/[email protected]_64 2.43G - 12.0G -
root/[email protected]_64 57.6M - 11.5G -
Beware that keeping the snapshots will gradually use more and more disk space. You will need to manually delete old snapshots with a command similar to the following to prevent your system from (eventually) running out of space (substitute the snapshot name according to what is listed in your output from the above command).*
$ sudo zfs destroy root/[email protected]_64
Also, beware that this mechanism will not automatically snapshot decendant filesystems. To create snapshots of your home director(y/ies), consider creating a cron job or perhaps a systemd service that runs when you sign out.
To roll back your OS, highlight the kernel that you want to roll back to from your boot menu, press E (or Tab if using Syslinux), and change bootfs.snapshot to bootfs.rollback. If the version you roll back to is not the lastest snapshot/kernel, you will need to manually remove any newer kernels, initramfs images, and boot configuration snippets from your ESPs since their corresponding modules will no longer exist under /lib/modules on the root filesystem.
* With version 2.9 of this installer script, a /etc/kernel/install.d/89-snapshot-remove.install script has been added, but it is not enabled by default. Use sudo chmod +x /etc/kernel/install.d/89-snapshot-remove.install to enable it. It will automatically remove the OS snapshot when the corresponding kernel is removed.
Make /etc a git repo that is backed by a separate ZFS dataset
If you make your /etc directory a git repo as shown below, you can push your customizations to a local repo when you make them and then you will be able to recover them later if you roll back your OS.
Substitute POOL='root' and DSET='0' to match your ZFS pool and root dataset, respectively.
$ sudo -i
# POOL='root'
# DSET='0'
# REPO="$POOL/${DSET//\//-}-etc.git"
# dnf install git-core rpmconf
# cat <<- 'END' > ~/.gitconfig
[user]
name = Super User
email = root@localhost
END
# zfs create -o mountpoint=legacy $REPO
# sed -i "$ a $REPO /srv/etc.git zfs nofail 0 0" /etc/fstab
# systemctl daemon-reload
# mkdir /srv/etc.git
# mount /srv/etc.git
# git init -b main --bare /srv/etc.git
# git init -b main /etc
# bash -ec "cd /etc; git add .; git commit -m '$(printf '%(%FT%T)T')'; git push -u -f /srv/etc.git main;"
# dnf install libdnf5-plugin-actions
# cat <<- 'END' > /etc/dnf/libdnf5-plugins/actions.d/git.actions
post_transaction:::enabled=host-only:/usr/bin/sh -ec cd\ /etc;\ git\ add\ .;\ git\ commit\ -m\ $(printf\ '%(%FT%T)T');\ git\ push\ -f;
END
# cat <<- 'END' > /usr/local/bin/rpmconf
#!/usr/bin/bash
command -p rpmconf "$@"
set -e; cd /etc; git commit -a --amend --no-edit; git push -f;
END
# chmod +x /usr/local/bin/rpmconf
# exit
To demo the system configured above, suppose you made a change to some file under /etc as shown below.
$ sudo -i
# cd /etc
# cat <<- 'END' >> inputrc
# do not encourage careless pasting of commands
set enable-bracketed-paste off
END
# git add ./inputrc
# git commit -m 'disable bracketed paste'
# git push -f
# exit
If at some future point in time you roll back your root filesystem, you will be able to retrieve and re-apply your change from your git repo with commands similar to the ones shown in the following code block.
Set DATESTAMP='YYYY-MM-DD' to point to the archive you created right after you rolled back your root dataset. (See the next paragraph for details about how such an archive might be created.)
$ sudo -i
# DATESTAMP='YYYY-MM-DD'
# cd /etc
# git -C /srv/etc.git-$DATESTAMP log --oneline
# git -C /srv/etc.git-$DATESTAMP show -p :/"disable bracketed paste" | patch -p 1
patching file inputrc
# git add ./inputrc
# git commit -m 'disable bracketed paste'
# git push -f
# exit
Note that you should archive your current etc.git repo and create a new one using commands similar to the ones shown in the following code block immediately after rolling back your root filesystem and before retrieving older changes to avoid confusion.
Substitute POOL='root' and DSET='0' to match your ZFS pool and root dataset, respectively.
$ sudo -i
# POOL='root'
# DSET='0'
# DATESTAMP="$(printf '%(%F)T')"
# REPO="$POOL/${DSET//\//-}-etc.git"
# ARCHIVE_REPO="$REPO-$DATESTAMP"
# ARCHIVE_PATH="/srv/etc.git-$DATESTAMP"
# umount /srv/etc.git
# zfs rename "$REPO" "$ARCHIVE_REPO"
# sed -i "$ a $ARCHIVE_REPO $ARCHIVE_PATH zfs nofail 0 0" /etc/fstab
# systemctl daemon-reload
# mkdir "$ARCHIVE_PATH"
# mount "$ARCHIVE_PATH"
# zfs create -o mountpoint=legacy $REPO
# mount /srv/etc.git
# git init -b main --bare /srv/etc.git
# rm -rf /etc/.git
# git init -b main /etc
# bash -ec "cd /etc; git add .; git commit -m '$(printf '%(%FT%T)T')'; git push -u -f /srv/etc.git main;"
# exit
Turn an oscr instance into a Systemd container
Substitue root/1 and root-1 to match the name of the OS instance you want to convert.
Note that the ZFS commands (zpool and zfs) will not work from within the container. Many system calls will also be restricted.
$ sudo -i
# dnf install systemd-container
# zfs set mountpoint=/var/lib/machines/root-1 root/1
# zfs mount root/1
# mkdir /etc/systemd/nspawn
# cat <<- END > /etc/systemd/nspawn/root-1.nspawn
[Exec]
PrivateUsers=identity
[Files]
Bind=/boot
BindReadOnly=/etc/resolv.conf
BindReadOnly=/dev/null:/etc/systemd/system/auditd.service
BindReadOnly=/dev/null:/etc/systemd/system/audit-rules.service
BindReadOnly=/dev/null:/etc/systemd/system/bootbind.service
BindReadOnly=/dev/null:/etc/systemd/system/bootsync.service
BindReadOnly=/dev/null:/etc/systemd/system/sshd.service
[Network]
VirtualEthernet=off
END
# machinectl start root-1; sleep 5;
# machienctl login root-1
To revert the above changes and turn the container back into a bootable OS, use the following commands.
$ sudo -i
# machinectl stop root-1; sleep 5;
# touch /var/lib/machines/root-1/.autorelabel
# zfs unmount root/1
# zfs set mountpoint=/ root/1
# exit
It is possible to run graphical applications in Systemd containers. I have not attempted it, but below is a link to a guide for running GUI apps under Systemd containers.
Known Bugs
In rare cases, I've seen systemd get confused about which is the correct machine-id for the running system after running the os?? scripts. A reboot will clear the problem, but one of the side-effects of the problem is that the normal shutdown -r now command does not work. As a workaround, systemctl --force --force reboot can be used to forcibly reboot your system.
Demo (installation)
If the above animated SVG doesn't play after clicking the link, try refreshing the page. Also, the file is somewhat large. You might need to wait a bit for it to render. Also, the animations seem to play a little smoother in Firefox than they do in Chrome.
Testing (optional)
There is a boot script in the root of this repository that I used to test the installations during development of this script.
Demo (booting and updating the kernel)
Disclaimer
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
2 activities