Following up to yesterday's post on LXC: Linux Containers, I had a quick play with 2 ULA subnets (aka RFC4193 addresses - dont forget that site-local was depreciated) - one subnet was dedicated to the LXC containers, one for my normal LAN. Perhaps unsurprisingly IPv6 appears to work perfectly well in this setup.
I also altered the setup and bridged a container directly to eth0 on the host node, and watched the container assign itself a stateless address based on my prefix, and again everything appeared to work perfectly well out onto the public v6 network (courtesy of Hurricane Electric's Tunnel Broker service).
So all in all I'd say that LXC is looking pretty good so far. There are a few other things I'd like to test, like how effective iptables are in the context of containers, and whether or not it is secure enough. Unfortunately I'm not going to have time to play with these things this weekend really. Answers on a postcard to the usual address if you already know though!
I've been toying with migrating my server into a containerized system, and almost bought a new server in preparation to migrate everything across. I'd chosen and tested my solution, OpenVz. All was good with the world. Until I saw that OpenVz was effectively being dropped from Ubuntu 10.04 (Lucid Lynx) and most likely Debian 6.0 (Squeeze).
The reason for the drop is simply because the OpenVz patches haven't been forward ported into the current kernel. Now thats obviously a bit of a problem, since I prefer Debian no straight dist-upgrade from Lenny to Squeeze would be slightly annoying. To put it mildly.
It turns out that a few people that realised this is going to be equally annoying, should such a thing occur, and they've been working LXC: Linux Containers and got the implementation into the main kernel tree. However there are a few issues with LXC;
- Documentation: It's thin on the ground, but if you're capable there is enough to get you going, and if absolutely necessary you can dive into the source (although I will admit it can does take a bit of time if you're not familiar with the basics of linux kernel development [like me]).
- It's not as mature as OpenVz (userspace tools, for instance can do odd things, and there are data leaks [for instance in Squeeze at the moment you can see the host mounts via /proc/mounts] although many seem to be resolved in git).
The OpenVz team seem to be updating the OpenVz patches for a more recent kernel, however the decision by a few distros to migrate to LXC might provoke the same sort of effect that migrating to KVM had on Xen's popularity.
Anyway, here's my quick start guide:
- You need a relatively recent kernel with control groups (cgroups) and capabilities support. If you're not planning on rolling your own kernel the current kernel in Squeeze or Lucid should be fine.
- Use the package manager of your choice to install lxc, bridgeutils and debootstrap. That should pull in everything you need for a basic install (if you want to run Debian based container).
- Run lxc-checkconfig and you should see everything is enabled and ready to go, with exception of cgroup memory controller, which LXC doesnt appear to require.
- So now we need the control groups pseudo-file system setup
mkdir /cgroups and mounted (add the following to /etc/fstab cgroup /cgroups cgroup defaults 0 0 and run mount). - Next choose how you're going to network your containers. In both these examples br0 is the device that the containers will be attached to. Aside from these 2 options there are other ways you could do it, but honestly I can envisage these being the most common options.
- You could attach them straight onto your network by creating a bridge including eth0 (for example), which your containers will attach to. In this example your /etc/network/interfaces might look something like
auto lo
iface lo inet loopback
iface br0 inet static
max_wait 0
bridge_ports eth0
address 192.168.0.3
netmask 255.255.255.0
gateway 192.168.0.1
- Alternatively you could create a dummy network device, add that to the bridge and then attach your containers to that, and either NAT (using iptables) or route to your containers. Or just keep them on their own. It really depends what do you want to do. I chose to add a dummy network and route between the 2. The host node interface in this example could look like
auto lo
iface lo inet loopback
allow-hotplug eth0
iface eth0 inet static
address 192.168.0.3
netmask 255.255.255.0
gateway 192.168.0.1
iface br0 inet static
max_wait 0
bridge_ports dummy0
address 10.10.10.1
netmask 255.255.255.0
- Now you need to create the files for the container. debootstrap or febootstrap, etc. are your friends. I chose to keep my containers in /var/lxc/guests. So for my first container (called "one", which is a Lenny basic install)
debootstrap lenny /var/lxc/guests/one You'll need to ensure that the /etc/resolv.conf is setup correctly under /var/lxc/guests/one, as is /etc/network/interfaces. For resolv.conf you can most likely copy from the host node. Your interfaces will just need eth0 and lo setup with the correct IPs. - Next you need to create a LXC config. This file is pretty much full of voodoo and dragons. I've saved this one under /var/lxc/guests/one.conf
# Hostname
lxc.utsname = one
# Number of TTYs to allocate to the container
# Relies on some lxc.cgroup.devices settings
lxc.tty = 4
# Networking type
lxc.network.type = veth
# State of networking at boot
lxc.network.flags = up
# Bridge you want to attach to
lxc.network.link = br0
# Internal container network interface name
lxc.network.name = eth0
lxc.network.mtu = 1500
# Address you intend to add the container to
# Doesn't seem to care too much as far as I can tell
lxc.network.ipv4 = 192.168.0.3/24
# Location of the root, from within the host node
lxc.rootfs = /var/lxc/guests/one
# Lots of stuff I've not fully yet looked into, but attempted to make
# some intelligent guesses about
lxc.cgroup.devices.deny = a
# /dev/null, /dev/zero
lxc.cgroup.devices.allow = c 1:3 rwm
lxc.cgroup.devices.allow = c 1:5 rwm
# TTYs
lxc.cgroup.devices.allow = c 5:1 rwm
lxc.cgroup.devices.allow = c 5:0 rwm
lxc.cgroup.devices.allow = c 4:0 rwm
lxc.cgroup.devices.allow = c 4:1 rwm
# /dev/random, /dev/urandom
lxc.cgroup.devices.allow = c 1:9 rwm
lxc.cgroup.devices.allow = c 1:8 rwm
# /dev/pts/* - Seems to be unused at this point?
lxc.cgroup.devices.allow = c 136:* rwm
lxc.cgroup.devices.allow = c 5:2 rwm
# No idea
lxc.cgroup.devices.allow = c 254:0 rwm
- Next create the container
lxc-create -n one -f /var/lxc/guests/one.conf. - And now you can start it. I've discovered that you need to set the container to be daemonised otherwise lxc-start will never return:
lxc-start -n one -d To attach to your container just run lxc-console -n one To stop lxc-stop - You will notice at this point that lxc-ls returns 2 sets of lists. The top list is the list of available containers, and the second list of the currently running containers.
This is obviously by no means a definitive guide, but it is just what I've done this evening to get stuff up and running. I've not yet tried getting IPv6 working into the containers, mostly because I wanted to try the v4 networking in a few different ways, and it's now bed time. However looking at the docs, it shouldn't be all that tricky working straight out of the box - something that OpenVz doesn't do in all circumstances (the same can be said for Linux Vservers).
As for whether or not it's worth it.. Lets just say it's not been unpleasant.
One point twenty one jiggawatts! Yesterday (March 8, 2010) the OpenSSH project released version 5.4 and naturally will start hitting the various distributions and platforms soon, and again there are some great things to be interested in:
- Although many distributions of OpenSSH have SSH1 disabled, the project is now shipping with SSH1 disabled by default.
- There is the ability to revoke keys (host and user) in both sshd and ssh.
- Netcat mode connects stdio on the client to a single port forward on the server. For example the following would connect to smtp.server.example.org on port 25, and redirect the output to stdio on my client side. Useful if you need to test connectivity to a mail server, but can't from your direct location, but can from your SSH server (my.ssh.server.example.org).
ssh -W smtp.server.example.org:25 my.ssh.server.example.org That has pretty much bags of possibilities, ranging from simple connection tests to piping a file to a remote server that you can't get to directly. - sftp-server has gained a read only mode!
- Passphrase-protected SSH2 private keys are now protected with AES-128 instead of 3DES. This counts if you reencrypt your key or create a new one.
