Linux Containers

From ArchWiki
Revision as of 13:57, 20 September 2011 by Eigrad (Talk | contribs)

Jump to: navigation, search


Current state of this HowTo

Delerious010 21:35, 1 December 2009 (EST)

  • Currently just a rough draft... I think I'll need to restructure this a bit and I've also noticed I've become a bit too verbose -_-;; I'll be along shortly to complete this as well as clean it up.



Linux Containers (LXC) are an operating system-level virtualization method for running multiple isolated server installs (containers) on a single control host. LXC does not provide a virtual machine, but rather provides a virtual environment that has its own process and network space. It is similar to a chroot, but offers much more isolation.

About this HowTo

This document is intended as an overview on setting up and deploying containers, and is not an in depth detailed instruction by instruction guide. A certain amount of prerequisite knowledge and skills are assumed (running commands as root, kernel configuration, mounting filesystems, shell scripting, chroot type environments, networking setup, etc).

Much of this was taken verbatim from Dwight Schauer, Tuxce and Ulhume. It has been copied here both to enable to community to share their collective wisdom and to expand on a few points.

Less verbose tutorial

Delerious010 21:43, 1 December 2009 (EST) I've come to realize I've added a lot of text to this HowTo. If you'd like something more streamlined, please head on over to for Dwight's excellent guide.

Kernel configuration

As of atleast the default Arch kernel 2.6.35, most of the required LXC options are built in. The only exception is CONFIG_DEVPTS_MULTIPLE_INSTANCES. Default Arch Kernel 3.0 includes all required config options.

Through the GUI

General Setup

  • [*] Group CPU scheduler
    • [*] Group scheduling for SCHED_OTHER
    • [*] Group scheduling for SCHED_RR/FIFO
    • Basis for grouping tasks (Control groups)
      • [*] Control groups
  • [*] Control Group support
    • [*] Namespace cgroup subsystem
    • [*] Freezer cgroup subsystem
    • [*] Device controller for cgroups
    • [*] Cpuset support
      • [*] Include legacy /proc/ /cpuset file
    • [*] Simple CPU accounting cgroup subsystem
    • [*] Resource counters
      • [*] Memory Resource Controller for Control Groups
        • [*] Memory Resource Controller Swap Extension(EXPERIMENTAL)
  • [*] Namespace support
    • [*] UTS namespace
    • [*] IPC namespace
    • [*] User namespace (EXPERIMENTAL)
    • [*] PID Namespaces (EXPERIMENTAL)
    • [*] Network namespace

Networking support

  • Networking options
    • [*] QoS and/or fair queueing
      • [*] Control Group Classifier

Device drivers

  • Character devices
    • [*] Unix98 pty support
      • [*] Support multiple instances of devpts

Security options

  • [*] File POSIX Capabilities

Through the .config


Testing capabilities

Once the lxc package is installed, running lxc-checkconfig will print out a list of your system's capabilities

Host configuration

Control group filesystem

LXC depends on the control group filesystem being mounted. At present, there exists no standard location for it. As such, you're free to create it where ever you see fit.

Mounting manually

mkdir /cgroup
mount -t cgroup none /cgroup

In /etc/fstab

none /cgroup cgroup defaults 0 0

Userspace tools

The packages can be found in AUR: lxc, lxc-git

Bridge device setup

The package, bridge-utils, will be required in order to use the brctl command below.

# pacman -S bridge-utils


config_br0="brctl setfd br0 0"


MODULES=(... bridge ...) # YMMV, but this was not required
eth0="eth0 up" # I had to do the, "eth0 up" was not sufficient.
br0="dhcp" # or however you set your address
INTERFACES=(eth0 br0)

Bridge forward delay

In order for br0 to dhcp quickly (and for container network devices to be available quickly) one must set the forward delay of the bridge device to zero.

brctl setfd br0 0

Patch for /etc/rc.d/network

This is required to use the above mentioned config_br0 statement as of initscripts 2010.07-1.

--- network.0     2010-07-11 05:48:47.000000000 -0400
+++ network     2010-09-03 15:01:25.000000000 -0400
@@ -181,6 +181,15 @@
                                        /usr/sbin/brctl addif $br $brif || error=1
+                            eval brconfig="\$config_${br}"
+                            if [ -n "${brconfig}" ]; then
+                              if ${brconfig}; then
+                                true
+                              else
+                                echo config_${br}=\"${brconfig}\" \<-- invalid  configuration statement
+                                error=1
+                              fi
+                            fi

See also: FS#16625

Container setup

There are various different means to do this

Creating the filesystem


Bootstrap an install ( mkarchroot, debootstrap, rinse, Install From Existing Linux ). You can also just copy/use an existing installation’s complete root filesystem.

Download existing

You can download a base install tar ball. OpenVZ templates work just fine.

Using the lxc tools

/usr/bin/lxc-debian {create|destroy|purge|help}
/usr/bin/lxc-fedora {create|destroy|purge|help}

Creating the device nodes

Since udev does not work within the container, you'll want to make sure that a certain minimum amount of devices is created for it. This may be done with the following script :

mv ${DEV} ${DEV}.old
mkdir -p ${DEV}
mknod -m 666 ${DEV}/null c 1 3
mknod -m 666 ${DEV}/zero c 1 5
mknod -m 666 ${DEV}/random c 1 8
mknod -m 666 ${DEV}/urandom c 1 9
mkdir -m 755 ${DEV}/pts
mkdir -m 1777 ${DEV}/shm
mknod -m 666 ${DEV}/tty c 5 0
mknod -m 600 ${DEV}/console c 5 1
mknod -m 666 ${DEV}/tty0 c 4 0
mknod -m 666 ${DEV}/full c 1 7
mknod -m 600 ${DEV}/initctl p
mknod -m 666 ${DEV}/ptmx c 5 2

Container configuration

Configuration file

The main configuration files are used to describe how to originally create a container. Though these files may be located anywhere, /etc/lxc is probably a good place.

23/Aug/2010: Be aware that the kernel may not handle additional whitespace in the configuration file. This has been experienced on "lxc.cgroup.devices.allow" settings but may also be true on other settings. If in doubt use only one space wherever whitespace is required.

Basic settings

lxc.utsname = $CONTAINER_NAME

Basic settings explained

lxc.utsname : This will be the name of the cgroup for the container. Once the container is started, you should be able to see a new folder named /cgroup/$CONTAINER_NAME.

Furthermore, this will also be the value returned by hostname from within the container. Assuming you've not removed access, the container may overwrite this with it's init script.

lxc.mount : This points to an fstab formatted file that is a listing of the mount points used when lxc-start is called. This file is further explained further

Terminal settings

The following configuration is optional. You may add them to your main configuration file if you wish to login via lxc-console, or through a terminal ( ex.: Ctrl+Alt+F1 ).

The container can be configured with virtual consoles (tty devices). These may be devices from the host that the container is given permission to use (by its configuration file) or they may be devices created locally within the container.

The host's virtual consoles are accessed using the key sequence ALT+Fn (or CTRL+ALT+Fn from within an X11 session). The left ALT key reaches consoles 1 through 12 and the right ALT key reaches consoles 13 through 24. Further virtual consoles may be reached by the ALT+Right Arrow key sequence which steps to the next virtual console.

The container's local virtual consoles may be accessed using the "lxc-console" command.

Host Virtual Consoles

The container may access the host's virtual consoles if the host is not using them and the container's configuration allows it. Typical container configuration would deny access to all devices and then allow access to specific devices like this:

 lxc.cgroup.devices.deny = a          # Deny all access to devices
 lxc.cgroup.devices.allow = c 4:0 rwm # /dev/tty0
 lxc.cgroup.devices.allow = c 4:1 rwm # /dev/tty1
 lxc.cgroup.devices.allow = c 4:2 rwm # /dev/tty2

For a container to be able to use a host's virtual console it must not be in use by the host. This will most likely require the host's /etc/inittab to be modified to ensure no getty or other process runs on any virtual console that is to be used by the container.

After editing the host's /etc/inittab file, issung a "killall -HUP init" will terminate any getty processes that are no longer configured and this will free up the virtual conosole for use by the container.

Note that local virtual consoles take precedence over host virtual consoles. This is described in the next section.

Local Virtual Consoles

The number of local virtual consoles that the container has is defined in the container's configuration file (normally on the host in /etc/lxc). It is defined thus:

 lxc.tty = n

where n is the number of local virtual consoles required.

The local virtual consoles are numbered starting at tty1 and take precedence over any of the host's virtual consoles that the container might be entitled to use. This means that, for example, if n = 2 then the container will not be able to use the host's tty1 and tty2 devices even entitled to do so by its configuration file. Setting n to 0 will prevent local virtual consoles from being created thus allowing full access to any of host's virtual consoles that the container might be entitled to use.

/dev/tty Device Files

The container must have a tty device file (e.g. /dev/tty1) for each virtual console (host or local). These can be created thus:

   # mknod -m 666 /dev/tty1 c 4 1
   # mknod -m 666 /dev/tty2 c 4 2

and so on...

In the above, c means character device, 4 is the major device number (tty devices) and 1,2,3, etc, is the minor device number (specific tty device). Note that tty0 is special and always refers to the current virtual console.

For further info on tty devices:

If a virtual console's device file does not exist in the container then the container can not use the virtual console.

Configuring Log-In Ability

The container's virtual consoles may be used for login sessions if the container runs "getty" services on their tty devices. This is normally done by the container's "init" process and is configured in the container's "/etc/inittab" file using lines like this:

 c1:2345:respawn:/sbin/agetty -8 38400 tty1 linux

There is one line per device. The first part "c1" is just a unique label, the second part defines applicable run levels, the third part tells init to start a new getty when the current one terminates and the last part gives the command line for the getty. For further information refer to "man init"

If there is no getty process on a virtual console it will not be possible to log in via that virtual console. A getty is not required on a virtual console unless it is to be used to log in.

If a virtual console is to allow root logins it also needs to be listed in the container's /etc/securetty file.

Troubleshooting virtual consoles

If lxc.tty is set to a number, n, then no host devices numbered n or below will be accessible even if the above configuration is present because they will be replaced with local virtual consoles instead.

A tty device file's major number will change from 4 to 136 if it is a local virtual console. This change is visible within the container but not when viewing the container's devices from the host's filesystem. This information is useful when troubleshooting.

This can be checked from within a container thus:

 # ls -Al /dev/tty*
 crw------- 1 root root 136, 10 Aug 21 21:28 /dev/tty1
 crw------- 1 root root   4, 2  Aug 21 21:28 /dev/tty2

Pseudo Terminals

 lxc.pseudo = 1024

Maximum amount of pseudo terminals that are may be created in /dev/pts. Delerious010 18:57, 3 December 2009 (EST) Currently, assuming the kernel was compiled with CONFIG_DEVPTS_MULTIPLE_INSTANCES, this tells lxc-start to mount the devpts filesystem with the newinstance flag.

Host device access settings

lxc.cgroup.devices.deny = a # Deny all access to devices
lxc.cgroup.devices.allow = c 1:3 rwm # dev/null lxc.cgroup.devices.allow = c 1:5 rwm # dev/zero
lxc.cgroup.devices.allow = c 5:1 rwm # dev/console lxc.cgroup.devices.allow = c 5:0 rwm # dev/tty lxc.cgroup.devices.allow = c 4:0 rwm # dev/tty0
lxc.cgroup.devices.allow = c 1:9 rwm # dev/urandom lxc.cgroup.devices.allow = c 1:8 rwm # dev/random lxc.cgroup.devices.allow = c 136:* rwm # dev/pts/* lxc.cgroup.devices.allow = c 5:2 rwm # dev/pts/ptmx
# No idea what this is .. dev/bsg/0:0:0:0 ??? lxc.cgroup.devices.allow = c 254:0 rwm

Host device access settings explained

lxc.cgroup.devices.deny : By settings this to a, we're stating that the container has access to no devices unless explicitely defined within the configuration file.

Configuration file notes

At runtime /dev/ttyX devices are recreated

If you've enabled multiple DevPTS instances in your kernel, lxc-start will recreate lxc.tty amount of /dev/ttyX devices when it is executed.

This means that you will have lxc.tty amount of pseudo ttys. If you're planning on accessing the container via a "real" terminal ( Ctrl+Alt+FX ), make sure that it's a number that is inferior to lxc.tty.

To tell whether it's been re-created, just log into the container via either lxc-console or ssh and perform a ls -Al command on the tty. Devices with a major number of 4 are "real" tty devices where as a major number of 136 indicates a pts.

Be aware that this is only visible from within the container itself and not from the host.

Containers have access to host's TTY nodes

If you do not properly restrict the container's access to the /dev/tty nodes, the container may have access to the host's.

Taking into consideration that, as previously mentioned, lxc-start recreates lxc.tty amount of /dev/tty devices, any tty nodes present in the container that are of a greater minor number than lxc.tty will be linked to the host's.

To access the container from a host TTY

  1. On the host, verify no getty is started for that tty by checking /etc/inittab.
  2. In the container, start a getty for that tty.

To prevent access to the host TTY

Please have a look at the configuration statements found in host device access settings.

Via the lxc.cgroup.devices.deny = a we're preventing access to all host level devices. And then, throuh lxc.cgroup.devices.allow = c 4:1 rwm we're allowing access to the host's /dev/tty1. In the above example, simply removing all allow statements for major number 4 and minor > 1 should be sufficient.

To test this access

I may be off here, but looking at the output of the ls command below should show you both the major and minor device numbers. These are located after the user and group and represented as : 4, 2

  1. Set lxc.tty to 1
  2. Make there that the container has dev/tty1 and /dev/tty2
  3. lxc-start the container
  4. lxc-console into the container
  5. ls -Al /dev/tty
    crw------- 1 root root 4, 2 Dec 2 00:20 /dev/tty2
  6. echo "test output" > /dev/tty2
  7. Ctrl+Alt+F2 to view the host's second terminal
  8. You should see "test output" printed on the screen

Configuration troubleshooting

console access denied: Permission denied

If, when executing lxc-console, you receive the error lxc-console: console access denied: Permission denied you've most likely either omitted lxc.tty or set it to 0.

lxc-console does not provide a login prompt

Though you're reaching a tty on the container, it most likely is not running a getty. You'll want to double check that you have a getty defined in the container's /etc/inittab for the specific tty.

Configuring fstab

none $CONTAINER_ROOTFS/dev/pts devpts defaults 0 0
none $CONTAINER_ROOTFS/proc    proc   defaults 0 0
none $CONTAINER_ROOTFS/sys     sysfs  defaults 0 0
none $CONTAINER_ROOTFS/dev/shm tmpfs  defaults 0 0

This fstab is used by lxc-start when mounting the container. As such, you can define any mount that would be possible on the host such as bind mounting to the host's own filesystem. However, please be aware of any and all security implications that this may have.

Warning : You certainly do not want to bind mount the host's /dev to the container as this would allow it to, amongst other things, reboot the host.

Container Creation and Destruction



lxc-create will create /var/lib/lxc/$CONTAINER_NAME with a new copy of the container configuration file found in $CONTAINER_CONFIGPATH.

As such, if you need to make modifications to the container's configuration file, it's advisable to modify only the original file and then perform lxc-destroy and lxc-create operations afterwards. No data will be lost by doing this.

Note : When copying the file over, lxc-create will strip all comments from the file.

Note : As of lxc-git from atleast 2009-12-01, performing lxc-create no longer splits the config file into multiple files and folders. Therefore, we only have the configuration file to worry about.


lxc-destroy -n $CONTAINER_NAME

This will delete /var/lib/lxc/$CONTAINER_NAME which only contains configuration files. No data will be lost.

Readying the host for virtualization


  1. Comment out any getty that are not required

/etc/rc.sysinit replacement

Since we're running in a virtual environment, a number of steps undertaken by rc.sysinit are superfluous and may even flat out fail or stall. As such, until the initscripts are made virtualization aware, this'll take some hack and slash.

For now, simply replace the file :

# Whatever is needed to clean out old daemon/service pids from your container
rm -f $(find /var/run -name '*pid')
rm -f /var/lock/subsys/*
# Configure network settings ## You can either use dhcp here, manually configure your ## interfaces or try to get the rc.d/network script working. ## There have been reports that network failed in this ## environment. route add default gw echo > /etc/resolv.conf search your-domain echo >> /etc/resolv.conf nameserver
# Initally we don't have any container originated mounts rm -f /etc/mtab touch /etc/mtab

/etc/rc.conf cleanup

You may want to remove any and all hardware related daemons from the DAEMONS line. Furthermore, depending on your situation, you may also want to remove the network daemon.