Skip to content

Latest commit

 

History

History
564 lines (378 loc) · 26.4 KB

README.md

File metadata and controls

564 lines (378 loc) · 26.4 KB
Raw

uCore

stable testing

uCore is an OCI image of Fedora CoreOS with "batteries included". More specifically, it's an opinionated, custom CoreOS image, built daily with some common tools added in. The idea is to make a lightweight server image including commonly used services or the building blocks to host them.

Please take a look at the included modifications, and help us improve uCore if the project interests you.

Table of Contents

Announcements

2024.11.12 - uCore has updated to Fedora 41

As of today our upstream Fedora CoreOS stable image updated to Fedora 41 under the hood, so expect a lot of package updates.

2024.11.12 - uCore stable has pinned to kernel version 6.11.3

Kernel version 6.11.3 was the previous stable update's kernel, and despite the update to Fedora 41, we've stuck with 6.11.3 rather than updating to 6.11.5 from upstream.

This is due to a kernel bug in versions 6.11.4/6.11.5 which breaks tailscale status reporting. As many users of uCore do use tailscale, we've decided to be extra cautious and hold back the kernel, even though the rest of stable updated as usual.

We expect the next update of Fedora CoreOS to be on 6.11.6 per the current state of the testing stream. So uCore will follow when that update occurs.

Features

The uCore project builds four images, each with different tags for different features.

The image names are:

The tag matrix includes combinations of the following:

  • stable - for an image based on the Fedora CoreOS stable stream
  • testing - for an image based on the Fedora CoreOS testing stream
  • nvidia - for an image which includes nvidia driver and container runtime
  • zfs - for an image which includes zfs driver and tools

Images

fedora-coreos

Important

This was previously named fedora-coreos-zfs, but that version of the image did not offer the nvidia option. If on the previous image name, please rebase with rpm-ostree rebase.

A generic Fedora CoreOS image image with choice of add-on kernel modules:

Note

zincati fails to start on all systems with OCI based deployments (like uCore). Upstream efforts are active to develop an alternative.

ucore-minimal

Suitable for running containerized workloads on either bare metal or virtual machines, this image tries to stay lightweight but functional.

  • Starts with a Fedora CoreOS image
  • Adds the following:
  • Optional nvidia versions add:
    • nvidia driver - latest driver built from negativo17's akmod package
    • nvidia-container-toolkit - latest toolkit which supports both root and rootless podman containers and CDI
    • nvidia container selinux policy - allows using --security-opt label=type:nvidia_container_t for some jobs (some will still need --security-opt label=disable as suggested by nvidia)
  • Optional ZFS versions add:
    • ZFS driver - latest driver (currently pinned to 2.2.x series) - see below for details
    • pv is installed with zfs as a complementary tool
  • Disables Zincati auto upgrade/reboot service
  • Enables staging of automatic system updates via rpm-ostreed
  • Enables password based SSH auth (required for locally running cockpit web interface)
  • Provides public key allowing SecureBoot (for ucore signed nvidia or zfs drivers)

Important

Per cockpit's instructions the cockpit-ws RPM is not installed, rather it is provided as a pre-defined systemd service which runs a podman container.

ucore

This image builds on ucore-minimal but adds drivers, storage tools and utilities making it more useful on bare metal or as a storage server (NAS).

  • Starts with a ucore-minimal image providing everything above, plus:
  • Adds the following:
    • cockpit-storaged (udisks2 based storage management)
    • distrobox - a toolbox alternative
    • duperemove
    • all wireless (wifi) card firmwares (CoreOS does not include them) - hardware enablement FTW
    • mergerfs
    • nfs-utils - nfs utils including daemon for kernel NFS server
    • pcp Performance Co-pilot monitoring
    • rclone - file synchronization and mounting of cloud storage
    • samba and samba-usershares to provide SMB sevices
    • snapraid
    • usbutils(and pciutils) - technically pciutils is pulled in by open-vm-tools in ucore-minimal
  • Optional ZFS versions add:

ucore-hci

Hyper-Coverged Infrastructure(HCI) refers to storage and hypervisor in one place... This image primarily adds libvirt tools for virtualization.

  • Starts with a ucore image providing everything above, plus:
  • Adds the following:
    • cockpit-machines: Cockpit GUI for managing virtual machines
    • libvirt-client: virsh command-line utility for managing virtual machines
    • libvirt-daemon-kvm: libvirt KVM hypervisor management
    • virt-install: command-line utility for installing virtual machines

Note

Fedora uses DefaultTimeoutStop=45s for systemd services which could cause libvirtd to quit before shutting down slow VMs. Consider adding TimeoutStopSec=120s as an override for libvirtd.service if needed.

Tag Matrix

IMAGE TAG
fedora-coreos - stable stable-nvidia, stable-zfs,stable-nvidia-zfs
fedora-coreos - testing testing-nvidia, testing-zfs, testing-nvidia-zfs
ucore-minimal - stable stable, stable-nvidia, stable-zfs,stable-nvidia-zfs
ucore-minimal - testing testing, testing-nvidia, testing-zfs, testing-nvidia-zfs
ucore - stable stable, stable-nvidia, stable-zfs,stable-nvidia-zfs
ucore - testing testing, testing-nvidia, testing-zfs, testing-nvidia-zfs
ucore-hci - stable stable, stable-nvidia, stable-zfs,stable-nvidia-zfs
ucore-hci - testing testing, testing-nvidia, testing-zfs, testing-nvidia-zfs

Installation

Important

Read the CoreOS installation guide before attempting installation. uCore extends Fedora CoreOS; it does not provide it's own custom or GUI installer.

There are varying methods of installation for bare metal, cloud providers, and virtualization platforms.

All CoreOS installation methods require the user to produce an Ignition file. This Ignition file should, at mimimum, set a password and SSH key for the default user (default username is core).

Tip

For bare metal installs, first test your ignition configuration by installing in a VM (or other test hardware) using the bare metal process.

Image Verification

These images are signed with sigstore's cosign. You can verify the signature by running the following command:

cosign verify --key https://github.com/ublue-os/ucore/raw/main/cosign.pub ghcr.io/ublue-os/IMAGE:TAG

Auto-Rebase Install

One of the fastest paths to running uCore is using examples/ucore-autorebase.butane as a template for your CoreOS butane file.

  1. As usual, you'll need to follow the docs to setup a password. Substitute your password hash for YOUR_GOOD_PASSWORD_HASH_HERE in the ucore-autorebase.butane file, and add your ssh pub key while you are at it.
  2. Generate an ignition file from your new ucore-autorebase.butane using the butane utility.
  3. Now install CoreOS for hypervisor, cloud provider or bare-metal. Your ignition file should work for any platform, auto-rebasing to the ucore:stable (or other IMAGE:TAG combo), rebooting and leaving your install ready to use.

Manual Install/Rebase

Once a machine is running any Fedora CoreOS version, you can easily rebase to uCore. Installing CoreOS itself can be done through a number of provisioning methods.

Warning

Rebasing from Fedora IoT or Atomic Desktops is not supported! If ignition doesn't provide a desired feature, then Fedora CoreOS doesn't support that feature. Rebasing from another system to gain a filesystem feature or GUI installation is very likely to cause problems later on.

To rebase an existing CoreOS machine to the latest uCore:

  1. Execute the rpm-ostree rebase command (below) with desired IMAGE and TAG.
  2. Reboot, as instructed.
  3. After rebooting, you should pin the working deployment which allows you to rollback if required.
sudo rpm-ostree rebase ostree-unverified-registry:ghcr.io/ublue-os/IMAGE:TAG

Verified Image Updates

The ucore* images include container policies to support image verification for improved trust of upgrades. Once running one of the ucore* images, the following command will rebase to the verified image reference:

sudo rpm-ostree rebase ostree-image-signed:docker://ghcr.io/ublue-os/IMAGE:TAG

Note

This policy is not included with fedora-coreos:* as those images are kept very stock.*

Tips and Tricks

CoreOS and ostree Docs

It's a good idea to become familar with the Fedora CoreOS Documentation as well as the CoreOS rpm-ostree docs. Note especially, this image is only possible due to ostree native containers.

Podman

Immutability and Podman

A CoreOS root filesystem system is immutable at runtime, and it is not recommended to install packages like in a mutable "normal" distribution.

Fedora CoreOS expects the user to run services using podman. moby-engine, the free Docker implementation, is also installed for those who desire docker instead of podman.

Docker/Moby and Podman

Important

CoreOS cautions against running podman and docker containers at the same time. Thus, docker.socket is disabled by default to prevent accidental activation of the docker daemon, given podman is the default.

Only run both simultaneously if you understand the risk.

Podman and FirewallD

Podman and firewalld can sometimes conflict such that a firewall-cmd --reload removes firewall rules generated by podman.

As of netavark v1.9.0 a service is provided to handle re-adding netavark (Podman) firewall rules after a firewalld reload occurs. If needed, enable like so: systemctl enable netavark-firewalld-reload.service

Automatically start containers on boot

By default, UCore does not automatically start restart: always containers on system boot, however this can be easily enabled:

For containers running under the core user
# Copy the system's podman-restart service to the user location
mkdir -p /var/home/core/.config/systemd/user
cp /lib/systemd/system/podman-restart.service /var/home/core/.config/systemd/user

# Enable the user service
systemctl --user enable podman-restart.service

# Check that it's running
systemctl --user list-unit-files | grep podman

When you next reboot the system, your restart: always containers will automatically start.

You may also need to enable “linger” mode on your user session, to prevent containers exiting which you have started interactively. To do that, run:

loginctl enable-linger $UID

You can find more information regarding this on the Podman troubleshooting page.

For containers running under the root user (rootful containers)

You just need to enable the built-in service:

sudo systemctl enable podman-restart.service

Default Services

To maintain this image's suitability as a minimal container host, most add-on services are not auto-enabled.

To activate pre-installed services (cockpit, docker, tailscaled, etc):

sudo systemctl enable --now SERVICENAME.service

Note

The libvirtd is enabled by default, but only starts when triggerd by it's socket (eg, using virsh or other clients).

SELinux Troubleshooting

SELinux is an integral part of the Fedora Atomic system design. Due to a few interelated issues, if SELinux is disabled, it's difficult to re-enable.

Warning

We STRONGLY recommend: DO NOT DISABLE SELinux!

Should you suspect that SELinux is causing a problem, it is easy to enable permissive mode at runtime, which will keep SELinux functioning, provide reporting of problems, but not enforce restrictions.

# setenforce 0
$ getenforce
Permissive

After the problem is resolved, don't forget to re-enable:

# setenforce 1
$ getenforce
Enforcing

Fedora provides useful docs on SELinux troubleshooting.

Distrobox

Users may use distrobox to run images of mutable distributions where applications can be installed with traditional package managers. This may be useful for installing interactive utilities such has htop, nmap, etc. As stated above, however, services should run as containers.

NAS - Storage

ucore includes a few packages geared towards a storage server which will require individual research for configuration:

But two others are included, which though common, warrant some explanation:

  • nfs-utils - replaces a "light" version typically in CoreOS to provide kernel NFS server
  • samba and samba-usershares - to provide SMB sevices

NFS

It's suggested to read Fedora's NFS Server docs plus other documentation to understand how to setup this service. But here's a few quick tips...

Firewall - NFS

Unless you've disabled firewalld, you'll need to do this:

sudo firewall-cmd --permanent --zone=FedoraServer --add-service=nfs
sudo firewall-cmd --reload
SELinux - NFS

By default, nfs-server is blocked from sharing directories unless the context is set. So, generically to enable NFS sharing in SELinux run:

For read-only NFS shares:

sudo semanage fcontext --add --type "public_content_t" "/path/to/share/ro(/.*)?
sudo restorecon -R /path/to/share/ro

For read-write NFS shares:

sudo semanage fcontext --add --type "public_content_rw_t" "/path/to/share/rw(/.*)?
sudo restorecon -R /path/to/share/rw

Say you wanted to share all home directories:

sudo semanage fcontext --add --type "public_content_rw_t" "/var/home(/.*)?
sudo restorecon -R /var/home

The least secure but simplest way to let NFS share anything configured, is...

For read-only:

sudo setsebool -P nfs_export_all_ro 1

For read-write:

sudo setsebool -P nfs_export_all_rw 1

There is more to read on this topic.

Shares - NFS

NFS shares are configured in /etc/exports or /etc/exports.d/* (see docs).

Run It - NFS

Like all services, NFS needs to be enabled and started:

sudo systemctl enable --now nfs-server.service
sudo systemctl status nfs-server.service

Samba

It's suggested to read Fedora's Samba docs plus other documentation to understand how to setup this service. But here's a few quick tips...

Firewall - Samba

Unless you've disabled firewalld, you'll need to do this:

sudo firewall-cmd --permanent --zone=FedoraServer --add-service=samba
sudo firewall-cmd --reload
SELinux - Samba

By default, samba is blocked from sharing directories unless the context is set. So, generically to enable samba sharing in SELinux run:

sudo semanage fcontext --add --type "samba_share_t" "/path/to/share(/.*)?
sudo restorecon -R /path/to/share

Say you wanted to share all home directories:

sudo semanage fcontext --add --type "samba_share_t" "/var/home(/.*)?
sudo restorecon -R /var/home

The least secure but simplest way to let samba share anything configured, is this:

sudo setsebool -P samba_export_all_rw 1

There is much to read on this topic.

Shares - Samba

Samba shares can be manually configured in /etc/samba/smb.conf (see docs), but user shares are also a good option.

An example follows, but you'll probably want to read some docs on this, too:

net usershare add sharename /path/to/share [comment] [user:{R|D|F}] [guest_ok={y|n}]
Run It - Samba

Like all services, Samba needs to be enabled and started:

sudo systemctl enable --now smb.service
sudo systemctl status smb.service

SecureBoot w/ kmods

For those wishing to use nvidia or zfs images with pre-built kmods AND run SecureBoot, the kernel will not load those kmods until the public signing key has been imported as a MOK (Machine-Owner Key).

Do so like this:

sudo mokutil --import /etc/pki/akmods/certs/akmods-ublue.der

The utility will prompt for a password. The password will be used to verify this key is the one you meant to import, after rebooting and entering the UEFI MOK import utility.

NVIDIA

Included Drivers

If you installed an image with -nvidia in the tag, the nvidia kernel module, basic CUDA libraries, and the nvidia-container-toolkit are all are pre-installed.

Note, this does NOT add desktop graphics services to your images, but it DOES enable your compatible nvidia GPU to be used for nvdec, nvenc, CUDA, etc. Since this is CoreOS and it's primarily intended for container workloads the nvidia container toolkit should be well understood.

The included driver is the latest nvidia driver as bundled by negativo17. This package was chosen over rpmfusion's due to it's granular packages which allow us to install just the minimal nvidia-driver-cuda packages.

Other Drivers

If you need an older (or different) driver, consider looking at the container-toolkit-fcos driver. It provides pre-bundled container images with nvidia drivers for FCOS, allowing auto-build/loading of the nvidia driver IN podman, at boot, via a systemd service.

If going this path, you likely won't want to use the ucore -nvidia image, but would use the suggested systemd service. The nvidia container toolkit will still be required but can by layered easily.

ZFS

If you installed an image with -zfs in the tag (or fedora-coreos-zfs), the ZFS kernel module and tools are pre-installed, but like other services, ZFS is not pre-configured to load on default.

Load it with the command modprobe zfs and use zfs and zpool commands as desired.

Per the OpenZFS Fedora documentation:

By default ZFS kernel modules are loaded upon detecting a pool. To always load the modules at boot:

echo zfs > /etc/modules-load.d/zfs.conf

ZFS and immutable root filesystem

The default mountpoint for any newly created zpool tank is /tank. This is a problem in CoreOS as the root filesystem (/) is immutable, which means a directory cannot be created as a mountpoint for the zpool. An example of the problem looks like this:

# zpool create tank /dev/sdb
cannot mount '/tank': failed to create mountpoint: Operation not permitted

To avoid this problem, always create new zpools with a specified mountpoint:

# zpool create -m /var/tank tank /dev/sdb

If you do forget to specify the mountpoint, or you need to change the mountpoint on an existing zpool:

# zfs set mountpoint=/var/tank tank

ZFS scrub timers

It's good practice to run a zpool scrub periodically on ZFS pools to check and repair the integrity of data. This can be easily configured with ucore by enabling the timer. There are two timers available: weekly and monthly.

# Substitute <pool> with the name of the zpool
systemctl enable --now zfs-scrub-weekly@<pool>.timer

# Or to run it monthly:
systemctl enable --now zfs-scrub-monthly@<pool>.timer

This can be enabled for multiple storage pools by enabling and starting a timer for each.

Sanoid/Syncoid

sanoid/syncoid is a great tool for manual and automated snapshot/transfer of ZFS datasets. However, there is not a current stable RPM, rather they provide instructions on installing via git.

ucore has pre-install all the (lightweight) required dependencies (perl-Config-IniFiles perl-Data-Dumper perl-Capture-Tiny perl-Getopt-Long lzop mbuffer mhash pv), such that a user wishing to use sanoid/syncoid only need install the "sbin" files and create configuration/systemd units for it.

DIY

Is all this too easy, leaving you with the desire to create a custom uCore image?

Then create an image FROM ucore using our image template!

Metrics

Alt