- Xen overview
- Domain 0 (host) installation
- DomU (guest) installation
- Older releases
- Package maintenance
- Common errors
- PV drivers on HVM guest
Xen is an open-source (GPL) type-1 or baremetal hypervisor, which makes it possible to run many instances of an operating system or indeed different operating systems in parallel on a single machine (or host)
Some of Xen's key features are:
- Small footprint and interface (is around 1MB in size). Because Xen uses a microkernel design, with a small memory footprint and limited interface to the guest, it is more robust and secure than other hypervisors.
Operating system agnostic: Most installations run with Linux as the main control stack (aka "domain 0"). But a number of other operating systems can be used instead, including NetBSD and ?OpenSolaris.
- Driver Isolation: Xen has the capability to allow the main device driver for a system to run inside of a virtual machine. If the driver crashes, or is compromised, the VM containing the driver can be rebooted and the driver restarted without affecting the rest of the system.
- Paravirtualization: Fully paravirtualized guests have been optimized to run as a virtual machine. This allows the guests to run much faster than with hardware extensions (HVM). Additionally, Xen can run on hardware that doesn't support virtualization extensions.
See the Xen Overview on the Xen wiki for more information.
Xen supports running two different types of guests: Paravirtualization (PV) and Full or Hardware assisted Virtualization (HVM). Both guest types can be used at the same time on a single Xen system. It is also possible to use techniques used for Paravirtualization in an HVM guest: essentially creating a continuum between PV and HVM. This approach is called PV on HVM. Again see the Xen Overview on the Xen wiki for more information.
Xen has a special domain called domain 0 which contains drivers for the hardware, as well as the toolstack to control VMs. Domain 0 is often referred to as dom0.
Domain 0 (host) installation
Before installing Xen you should install Debian on the host machine. This installation will form the basis of Domain 0.
In order to install Xen you will either a 32-bit PC (i386) or 64-bit PC (amd64) installation of Debian. Although it is recommended to always run a 64-bit hypervisor note that this does not mean one has to run a 64-bit domain 0. It is quite common to run a 32-bit domain 0 on a 64-bit hypervisor (a so-called "32on64" configuration).
In general you can install your Domain 0 Debian as you would any other Debian install. The main thing to consider is the partition layout since this will have an impact on the disk configurations available to the guests. The Xen wiki has some Host OS Installation Considerations which may be of interest. To paraphrase that source: if your Domain 0 Debian system will be primarily used to run guests, a good rule of thumb is to set aside 4GB for the domain 0 root filesystem (/) and some swap space (swap=RAM if RAM<=2GB; swap=2GB if RAM>2GB). The swap space should be determined by the amount of RAM provided to Dom0, see Configure Domain 0 Memory
Use the rest of the disk space for a LVM physical volume.
If you have one disk, the following is a reasonable setup: create 3 physical partitions: sda1, sda2, sda3. The root (ext4) and swap will be on the first two and the remainder will be under Logical Volume Management (lvm). With the LVM setup, create 1 physical volume and then one volume group. Give the volume group a name, such as `vg0'.
Installing Xen packages
The Xen and debootstrap software in Squeeze (Debian 6.0) are very much newer than that in Lenny. Because of that, working with Xen becomes a lot easier.
The setup described here is tested for Debian Squeeze and Ubuntu Maverick virtual machines, but should work for a lot more.
First install the hypervisor, xen aware kernel and xen tools. This can be done by a metapackage:
apt-get install xen-linux-system
Since Debian Wheezy, it's better to install this metapackage :
apt-get install xen-system
Checking for hardware HVM support
Hardware-assisted virtualization, requires CPU support for either the extension: AMD Secure Virtual Machine (AMD Virtualisation; AMD-V); or Intel Virtual Machine Extensions (VT-x).
On your intended host system, you can run this command:
egrep '(vmx|svm)' /proc/cpuinfo
On squeeze (but not on wheezy), providing the necessary emulation infrastructure for an HVM guest, the qemu device model package is also required:
apt-get install xen-qemu-dm-4.0
Prioritise booting Xen over native Linux
Debian Wheezy and Squeeze use Grub 2 whose default is to list normal kernels first, and only then list the Xen hypervisor and its kernels.
You can change this to cause Grub to prefer to boot Xen by changing the priority of Grub's Xen configuration script (20_linux_xen) to be higher than the standard Linux config (10_linux). This is most easily done using dpkg-divert:
dpkg-divert --divert /etc/grub.d/08_linux_xen --rename /etc/grub.d/20_linux_xen
to undo this:
dpkg-divert --rename --remove /etc/grub.d/20_linux_xen
After any update to the Grub configuration you must apply the configuration by running:
In order to give network access to guest domains it is necessary to configure the domain 0 network appropriately. The most common configuration is to use a software bridge.
It is recommended that you manage your own network bridge using the Debian network bridge. The Xen wiki page Host Configuration/Networking also has some useful information. The Xen supplied network scripts are not always reliable and will be removed from a later version. They are disabled by default in Debian's packages.
If you have a router that assigns ip addresses through dhcp, the following is a working example of the /etc/network/interfaces file using bridge-utils software.
#The loopback network interface auto lo iface lo inet loopback iface eth0 inet manual auto xenbr0 iface xenbr0 inet dhcp bridge_ports eth0 #other possibly useful options in a virtualized environment #bridge_stp off # disable Spanning Tree Protocol #bridge_waitport 0 # no delay before a port becomes available #bridge_fd 0 # no forwarding delay ## configure a (separate) bridge for the DomUs without giving Dom0 an IP on it #auto xenbr1 #iface xenbr1 inet manual # bridge_ports eth1
Other configuration tweaks
Domain 0 memory
By default on a Xen system the majority of the hosts memory is assigned to dom0 on boot and dom0's size is dynamically modified ("ballooned") automatically in order to accommodate new guests which are started.
However on a system which is dedicated to running Xen guests it is better to instead give dom0 some static amount of RAM and to disable ballooning.
The following examples use 1024M.
In order to do this you must first add the dom0_mem option to your hypervisor command line. This is done by editing /etc/default/grub and adding
# Xen boot parameters for all Xen boots GRUB_CMDLINE_XEN="dom0_mem=1024M,max:1024M"
at the bottom of the file.
Note : On servers with huge memory, Xen kernel crash. You must set a dom0 memory limit. Take care on Wheezy, 1024M is not enough and cause kernel crash at boot with out-of-memory message.
Remember to apply the change to the grub configuration by running update-grub!
Then edit /etc/xen/xend-config.sxp to configure the toolstack to match by changing the following settings:
(dom0-min-mem 1024) (enable-dom0-ballooning no)
With the new xl toolstack, edit /etc/xen/xl.conf and disable autoballoon with autoballoon="0"
At this point you should reboot so that these changes take effect.
Domain 0 CPUs
There are some useful tweaks of dom0 cpu utilization.
By default all CPUs are shared among dom0 and all domU (guests). It may broke dom0 responsibility if guests consume too much CPU time. To avoid this, it is possible to grant one (or more) processor core to dom0 and also pin it to dom0.
Add following options to /etc/default/grub to allocate one cpu core to dom0:
Make such changes in /etc/xen/xend-config.sxp:
Guest behaviour on host reboot
By default, when Xen dom0 shuts down or reboots, it tries to save (i.e. hibernate) the state of the domUs. Sometimes there are problems with that - it could fail because of a lack of disk space in /var, or because of random software bugs. Because it is also clean to just have the VMs shutdown upon host shutdown, if you want you can make sure they get shut down normally by setting these parameters in /etc/default/xendomains:
You may also want to pass some boot parameters to Xen when starting up in normal or recovery mode. Add these variables to /etc/default/grub to achieve this:
# Xen boot parameters for all Xen boots GRUB_CMDLINE_XEN="something" # Xen boot parameters for non-recovery Xen boots (in addition to GRUB_CMDLINE_XEN) GRUB_CMDLINE_XEN_DEFAULT="something else"
Remember to apply the change to the grub configuration by running update-grub!
More information on the available hypervisor command line options can be found in the upstream documentation.
PCI pass-through parameters
This information is incomplete for Squeeze and needs to be updated for Wheezy
To enable PCI pass-through, you need to know the BDF (Bus, Device, Function) id of the device. This is obtained through the lspci command, with the output containing the BDF in the format: (BB:DD.F) at the start of the line. To hide a device from Dom0 you will need to pass these boot parameters to Xen when starting. For example if using a Dom0 with 512M of memory and two devices at 01:08.1 and 01:09.2, add these variables to /etc/default/grub to achieve this:
# Xen boot parameters for all Xen boots GRUB_CMDLINE_XEN="dom0_mem=512M pciback.hide=(01:08.1)(01:09.2)" # Xen boot parameters for non-recovery Xen boots (in addition to GRUB_CMDLINE_XEN) GRUB_CMDLINE_XEN_DEFAULT="something else"
for Squeeze use "pciback.hide" (kernels < 18.104.22.168), for Wheezy (I have not tested this yet) use "xen-pciback.hide"
for Squeeze you need to pass all of the devices on the bus, eg to pass any device on the 01:DD.F bus, you have pass all of them: (01:08.1)(01:09.2)(01:09.3)etc.
Remember to apply the change to the grub configuration by running update-grub!
At least in Wheezy (not tested in Squeeze) the xen-pciback module needs to be configured through modprobe.conf and added to the initramfs additionally.
Configure the xen-pciback module by adding a modprobe include file (e.g. /etc/modprobe.d/xen-pciback.conf) with the following content (given that the PCI device would be assigned to module e1000e normally):
install e1000e /sbin/modprobe xen-pciback; /sbin/modprobe --first-time --ignore-install e1000e options xen-pciback hide=(0000:03:00.0)
Add the xen-pciback module to initramfs by adding it to /etc/initramfs/modules and running update-initramfs -u afterwards.
Please note that pci-passthrough is broken when MSI are enabled (default) in Linux kernels < 3.14. Use Linux kernel >= 3.14 in DomU/VM or set pci=nomsi for DomU/VM kernel as workaround. See the following thread for detailed information: http://thread.gmane.org/gmane.comp.emulators.xen.user/81944/focus=191437
To get output from GRUB, the Xen hypervisor, the kernel and getty (login prompt) via both VGA and serial console to work, here's an example of the right settings on squeeze:
Edit /etc/default/grub and add:
GRUB_SERIAL_COMMAND="serial --unit=0 --speed=9600 --word=8 --parity=no --stop=1" GRUB_TERMINAL="console serial" GRUB_TIMEOUT=5 GRUB_CMDLINE_XEN="com1=9600,8n1 console=com1,vga" GRUB_CMDLINE_LINUX="console=tty0 console=hvc0"
Here's what I used to configure the serial console (for a Supermicro X8STi-F motherboard with IPMI and SOL):
GRUB_CMDLINE_XEN="loglvl=all guest_loglvl=all com1=115200,8n1,0x3e8,5 console=com1,vga" GRUB_CMDLINE_LINUX="console=hvc0 earlyprintk=xen"
In /etc/inittab you need at least these lines:
1:2345:respawn:/sbin/getty 38400 hvc0 2:23:respawn:/sbin/getty 38400 tty1 # NO getty on ttyS0!
With systemd, you do not have an /etc/inittab any more. systemd will spawn a getty on /dev/hvc0 if you specify console=hvc0 on the kernel command line.
This way, tty1 will show up at the VGA output, and the hvc0 will show up at the serial console.
To keep both Xen and dom0 kernel output on the same tty, just omit the "vga"-related settings from the above setup.
If you need to debug Xen and see a crash dump of the kernel, you can do it using IPMITool if your server has SOL:
ipmitool -I lanplus -H server-ip-address -U your-username sol activate | tee my-log-file.txt
DomU (guest) installation
xen-tools is a set of scripts which can easily create fully configured Xen guest domains.
Once you have installed dom0 you can install xen-tools on your host with:
apt-get install xen-tools
To configure xen-tools, you can edit /etc/xen-tools/xen-tools.conf which contains default values that the xen-create-image script will use. The xen-create-image(8) manual page contains information on the available options.
To give a different path where the domU images being saved and enable the superuser password in the initial build, we will edit the /etc/xen-tools/xen-tools.conf file and uncomment this lines:
dir = /home/xen/ passwd = 1
Then you can create virtual machines with this command:
xen-create-image --hostname <hostname> --ip <ip> --vcpus 2 --pygrub --dist <lenny|squeeze|maverick|whatever>
To start the created VM run the command:
xl create /etc/xen/<hostname>.cfg
To erase a VM image (even the main directory) run the command:
Possible problems and bugs
If your domU kernel happens to miss support for the xvda* disk devices (the xen-blkfront driver), use the --scsi option that makes the VM use normal SCSI HD names like sda*. You can also set scsi=1 in /etc/xen-tools/xen-tools.conf to make this the default.
When using xen-tools with --role on Squeeze be aware of #588783: 'Should mount /dev/pts when creating image'. This is fixed in Wheezy.
Using Debian Installer
The Xen wiki page PvGrub2 contains instructions on how to use PV Grub2 with PV guests from Jessie onwards.
Upgrading domU kernels
A trivial way to boot a domU is to use a kernel in /boot of dom0. Unfortunately, this means that to upgrade the kernel used by a domU, one needs to synchronize with the dom0 kernel. I.e. if the dom0 kernel is upgraded, then the domU will not be able to load its kernel modules since the version will be different. Conversely, if the kernel is upgraded inside the domU, i.e. its kernel modules get upgraded, then the dom0 kernel will not fit either.
One way is to upgrade the kernel inside the domU, and propagate by hand the kernel image into a separate directory of dom0, where the configuration for the domU will look for it.
A better way is to use PyGrub or PvGrub: dom0 will just hold a version of grub, and use it to load the kernel image from the domU image itself, and thus avoids from having to propagate the kernel upgrade, domU can simply upgrade its kernel and reboot, without even having to destroy the domain. PvGrub is more secure than PyGrub, because PvGrub runs inside domU itself, thus having the same isolation as the target domU, while PyGrub runs inside dom0 and could an eventual compromise would be very concerning.
See also: Debian Release Notes
Upgrading a server to Squeeze that uses both Lenny Dom0 and DomU's is fairly straightforward. There are a few catches that one needs to be aware of however: Reference
- The Xen packages will not upgrade themselves. They must be manually removed and the latest Xen packages must be installed from the Debian Squeeze repository through apt.
xen.independent_wallclock sysctl setting is not available for newer squeeze kernels supporting pvops. If you have relied on it, you would have to run ntpd in dom0 and domUs. source
- A Squeeze DomU will not be able to boot on the Xen-3.2 package supplied by Lenny because this older version will not support grub2. A Lenny DomU can be upgraded to Squeeze while running on a Lenny Dom0 but it will not be able to be booted until the Dom0 has been upgraded to the Xen-4.0 packages.
- The entries added to chain load grub1 to grub2 will not allow pygrub to find the correct partition. Before rebooting a freshly upgraded Squeeze DomU, make sure to rename or remove /boot/grub/menu.lst. This will force pygrub to look for the /boot/grub/grub.cfg file which will be in the correct format.
- A qcow image created with qcow-create and the BACKING_FILENAME option on Lenny will not be able to boot on Squeeze because the ability to use qcow images as backing files has been removed in Xen versions after 3.2. Also, if you try to boot such an image on Squeeze, Xen will silently convert the qcow images L1 table to big endian (you'll find "Converting image to big endian L1 table" in the logfiles), effectively rendering the image unusable on both Squeeze and Lenny!
Kernel version compatibility
In Debian 6.0 (squeeze) the Linux '686-bigmem' and 'amd64' kernel images have paravirt_ops-based Xen domU support. From Debian 7 (wheezy) onward, the '686-pae' and 'amd64' kernel images support running as either dom0 or domU, and no Xen-specific kernel is provided.
When you create an image for a modern Debian or Ubuntu domU machine, it will include a kernel that has pv_ops domU support, it will therefore not use a Xen kernel, but the "stock" one, as it is capable of running on Xen's hypervisor.
?Xen Installation on Debian 5.0 ( Lenny )
?Xen Installation on Debian 4.0 ( Etch )
The Debian Developer's Package Overview page lists source packages that are maintained by the team.
dom0 automatic reboots
Note: if Xen is crashing and reboot automatically, you may want to use noreboot xen option, to prevent it from rebooting automatically.
Edit /etc/default/grub and add the "noreboot" option to GRUB_CMDLINE_XEN, for example:
Error "Device ... (vif) could not be connected"
You need to configure some basic networking between dom0 and domU.
Note: The use of the 'network-script' option in /etc/xen/xend-config.sxp is no longer recommended.
'clocksource/0: Time went backwards'
If a domU crashes or freezes while uttering the famous lasts words 'clocksource/0: Time went backwards' see Xen/Clocksource.
Error "unknown compression format"
The Xen system may fail to load a domU kernel, reporting the error:
ERROR Invalid kernel: xc_dom_probe_bzimage_kernel: unknown compression format xc_dom_bzimageloader.c:394: panic: xc_dom_probe_bzimage_kernel: unknown compression format
This indicates that your toolstack is not able to cope with the compression scheme used by the kernel binary. Most commonly this occurs with newer kernels which use xz compression when booting on older Xen installations. Debian switched to xz compression from package version 3.6.8-1~experimental.1 onwards.
Xz is supported by the version of Xen in Debian 7 (Wheezy) onwards. If you are running Debian guests on a non-Debian host then you will need to consult the non-Debian host's provider.
Packetloss / txqueuelen 32
If you experience a high packet loss check your vif interfaces for the qlen (in ifconfig output it's called txqueuelen). The default is 32 but physical interfaces and the interface you'll have in your domU will usually use something much greater like 1000. You can change the setting with
ip link set qlen 1000 dev vif-mydomU
It's documented in several places but the default is still unchanged:
PV drivers on HVM guest
It may be possible to build the PV drivers for use on HVM guests. These drivers are called unmodified_drivers and are part of the xen-unstable.hg repository. You can fetch the repository using mercurial thus:
hg clone http://xenbits.xen.org/xen-unstable.hg
The drivers reside under xen-unstable.hg/unmodified_drivers/linux-2.6. The README in this directory gives compilation instructions.
A somewhat dated, detailed set of instructions for building these drivers can be found here:
Xen Homepage: http://www.xen.org
Basic (and low-level) upstream Documentation is here. Includes:
German Wiki on Xen: http://www.xen-info.de/wiki
- Additional information required:
- Compiling a custom Xen DomU kernel. (e.g. adding tun device)
Two-way migration between Xen and KVM is described here: HowToMigrateBackAndForthBetweenXenAndKvm