LXD

Linux Containers

LXD はコンテナ(LXC経由)および仮想マシン(QEMU経由)のマネージャー/ハイパーバイザーです。

セットアップ

必要なソフトウェア

lxd パッケージをインストールして、lxd.service を有効にします。

他のインストール方法

snapd^AUR パッケージをインストールしてから以下のコマンドを実行することで snapd を使って LXD をインストールできます:

# snap install lxd

Setup for unpriviledged containers

It is recommended to use unpriviledged containers (See Linux_Containers#Privileged_containers_or_unprivileged_containers for an explanation of the difference).

In order to use them, you need to enable support to run unprivileged containers.

Once enabled, every container will be started unpriviledged by default.

For the alternative see howto set up priviledged containers.

LXD の設定

LXD を使うにはストレージプールと (インターネットを使いたい場合) ネットワークを設定する必要があります。以下のコマンドを root で実行してください:

# lxd init

非特権ユーザーとして LXD にアクセス

デフォルトでは LXD デーモンは lxd グループのユーザーからのアクセスを許可するので、ユーザーをグループに追加してください:

# usermod -a -G lxd <user>

使用方法

LXD consists of two parts:

the daemon (the lxd binary)
the client (the lxc binary)

ノート: lxc is not LXC; the naming is a bit confusing, you can read the forum post on comparing LXD vs LXC regarding the difference.

The client is used to control one or multiple daemon(s).

The client can also be used to control remote LXD servers.

Overview of commands

You can get an overview of all available commands by typing:

$ lxc

Create a container

You can create a container with lxc launch, for example:

$ lxc launch ubuntu:20.04

Container are based on images, that are downloaded from image servers or remote LXD servers.
You can see the list of already added servers with:

$ lxc remote list

You can list all images on a server with lxc image list, for example:

$ lxc image list images:

This will show you all images on one of the default servers: images.linuxcontainers.org

You can also search for images by adding terms like the distribution name:

$ lxc image list images:debian

Launch a container with an image from a specific server with:

$ lxc launch servername:imagename

For example:

$ lxc launch images:centos/8/amd64 centos

To create an amd64 Arch container:

$ lxc launch images:archlinux/current/amd64 arch

Create a virtual machine

Just add the flag --vm to lxc launch:

$ lxc launch ubuntu:20.04 --vm

ノート:

For now virtual machines support less features than containers (see Difference between containers and virtual machines for example).
Only cloud variants of the official images enable the lxd-agent out-of-the-box (which is needed for the usual lxc commands like lxc exec).
You can search for cloud images with lxc image list images: cloud or lxc image list images: distribution-name cloud.
If you use other images or encounter problems take a look at #lxd-agent_inside_a_virtual_machine.

Use and manage a container or VM

See Instance managament in the official Getting Started Guide of LXD.

Container/VM configuration (optional)

You can add various options to instances (containers and VMs).
See Configuration of instances in the official Advanced Guide of LXD for details.

ヒントとテクニック

Access the containers by name on the host

This assumes that you are using the default bridge, that it is named lxdbr0 and that you are using systemd-resolved.

 # systemd-resolve --interface lxdbr0 --set-domain '~lxd' --set-dns $(lxc network get lxdbr0 ipv4.address | cut -d / -f 1)

You can now access the containers by name:

 $ ping containername.lxd

Use Wayland and Xorg applications

ノート: Always consider security implications, as some of the described methods may weaken the seperation between container and host.

There are multiple methods to use GUI applications inside containers.

You can find an overview in the official Forum of LXD: https://discuss.linuxcontainers.org/t/overview-gui-inside-containers/8767

Method 1: Use the host's Wayland or Xorg Server

ノート: Using Xorg might weaken the seperation between container and host, because Xorg allows applications to access other applications windows. So container applications might have access to host applications windows.
Use Wayland instead (but be aware that Xorgs downsides also apply to XWayland).

Summary: In this method we grant containers access to the host's sockets of Wayland (+XWayland) or Xorg.

1. Add the following devices to a containers profile.

General device for the GPU:

mygpu:
   type: gpu

ノート: The path under "listen" is different, because /run and /tmp folders might be overridden, see: https://github.com/lxc/lxd/issues/4540

Device for the Wayland Socket:
Notes:

Adjust the Display (wayland-0) accordingly.
Add the folders in /mnt and /tmp inside the container, if they don't already exist.

Waylandsocket:
    bind: container
    connect: unix:/run/user/1000/wayland-0
    listen: unix:/mnt/wayland1/wayland-0
    uid: "1000"
    gid: "1000"
    security.gid: "1000"
    security.uid: "1000"
    mode: "0777"
    type: proxy

Device for the Xorg (or XWayland) Socket:
Note: Adjust the Display Number accordingly (for example X1 instead of X0).

Xsocket:
    bind: container
    connect: unix:/tmp/.X11-unix/X0
    listen: unix:/mnt/xorg1/X0
    uid: "1000"
    gid: "1000"
    security.gid: "1000"
    security.uid: "1000"
    mode: "0777"
    type: proxy

2. Link the sockets to the right location inside the container.

Note: These Scripts need to be run after each start of the container; you can automate this with systemd for example.

Shell-Script to link the Wayland socket:

#!/bin/sh
mkdir /run/user/1000
ln -s /mnt/wayland1/wayland-0 /run/user/1000/wayland-0

Link the Xorg (or XWayland) socket:

#!/bin/sh
ln -s /mnt/xorg1/X0 /tmp/.X11-unix/X0

3. Add Environment variables to the users config inside the container.

Note: Adjust the Display Numbers and/or the filename (.profile) accordingly.

For Wayland:

$ echo "export XDG_RUNTIME_DIR=/run/user/1000" >> ~/.profile
$ echo "export WAYLAND_DISPLAY=wayland-0" >> ~/.profile
$ echo "export QT_QPA_PLATFORM=wayland" >> ~/.profile

For Xorg (or XWayland):

$ echo "export DISPLAY=:0" >> .profile

Reload the .profile:

$ . .profile

トラブルシューティング

lxd-agent inside a virtual machine

Inside some virtual machine images the lxd-agent is not enabled by default.
In this case you have to enable it manually, for example by mounting a 9p network share. This requires console access with a valid user.

1. Login with lxc console:
Replace virtualmachine-name accordingly.

$ lxc console virtualmachine-name

Login as root:

ノート: On some systems you have to setup a root password first to be able to login as root.
You can use cloud-init for this for example.

$ su root

Mount the network share:

$ mount -t 9p config /mnt/

Go into the folder and run the install script (this will enable the lxd-agent inside the VM):

$ cd /mnt/
$ ./install.sh

After sucessful install, reboot with:

$ reboot

Afterwards the lxd-agent is available and lxc exec should work.

カーネルコンフィグの確認

デフォルトで Arch Linux のカーネルは Linux Containers とフロントエンドの LXD が動くようにコンパイルされています。カスタムカーネルを使っている場合やカーネルオプションを変更している場合、LXD が動かない場合があります。コンテナを実行できるようにカーネルが設定されているか確認してください:

$ lxc-checkconfig

Resource limits are not applied when viewed from inside a container

Install lxcfs and start lxcfs.service.

lxd will need to be restarted. Enable lxcfs.service for the service to be started at boot time.

Starting a virtual machine fails

If you see the error: Error: Required EFI firmware settings file missing: /usr/share/ovmf/x64/OVMF_VARS.ms.fd

Arch Linux does not distribute secure boot signed ovmf firmware, to boot virtual machines you need to disable secure boot for the time being.

$ lxc launch ubuntu:18.04 test-vm --vm -c security.secureboot=false

This can also be added to the default profile by doing:

$ lxc profile set default security.secureboot=false

No IPv4 with systemd-networkd

Starting with version version 244.1, systemd detects if /sys is writable by containers. If it is, udev is automatically started and breaks IPv4 in unprivileged containers. See commit bf331d8 and discussion on linuxcontainers.

On containers created past 2020, there should already be a systemd.networkd.service override to work around this issue, create it if it is not:

/etc/systemd/system/systemd-networkd.service.d/lxc.conf

[Service]
BindReadOnlyPaths=/sys

You could also work around this issue by setting raw.lxc: lxc.mount.auto = proc:rw sys:ro in the profile of the container to ensure /sys is read-only for the entire container, although this may be problematic, as per the linked discussion above.

参照

LXD

目次