Xen

From Gentoo Wiki
Jump to: navigation, search
Resources

Xen is a native, or bare-metal, hypervisor that allows multiple distinct virtual machines (referred to as domains) to share a single physical machine. As the highest privilege process on the system, Xen is responsible for the distribution of processor and memory resources between guest domains on the host. Other hardware resources such as network interfaces, disks, or direct PCI/USB devices are controlled by a privileged domain known as domain-0 (dom0).

From it's inception Xen has focused on the para-virtualization approach to hypervisor design. As a result, Xen guests or unprivileged domains (domU) are typically aware of the hypervisor and their status as guests. The base system, Domain-0, must have inherent Xen support, however, unmodified domU guests are supported on hardware which implements Intel (VT-x) or AMD (SVM) virtualization technology.

Host Configuration (domain-0)

Preparation

BIOS / UEFI Firmware

While paravirtualized (PV) guests can run on a host without any form of hardware accelerated virtualization extensions this severely restricts the set of available guests and prevents the use of new and exciting modes such as PVH.

The following command should provide a highlighted row for each processor if the corresponding feature is present:

user $grep --color -E "vmx|svm" /proc/cpuinfo

Unfortunately motherboard manufacturers often explicitly disable these virtualization extensions via BIOS or UEFI settings. If you don't see support listed check your BIOS/UEFI configuration for flags related to virtualization.

Rebuilding the Gentoo Installation?

A dramatic change that might be necessary on 32-bit systems is to rebuild the entire Gentoo installation with a different CFLAGS setting. Guest operating systems running under Xen might otherwise see major performance degradation. If you, however, are planning on checking out Xen rather than installing it for production use and are not terribly fond of rebuilding all programs, you can skip this step. In this case you will notice performance degradation but you will still be able to use Xen.

Important
It is advised that, if you change your CFLAGS and build your system with a gcc lower than version 4, you do not have -Os set as it has been reported to produce broken code.

Add -mno-tls-direct-seg-refs ONLY if you have a 32-bit dom0. You don't need this flag if you have a 64-bit dom0.

FILE /etc/portage/make.confCFLAGS change for mno-tls-direct-seg-refs
CFLAGS="... -mno-tls-direct-seg-refs"
root #emerge -e world

If you boot your system using an initial ramdisk (initrd) you need to rebuild the initrd as well (which is best done by running all steps you would do when you rebuild your kernel).

Kernel

Next we'll build the Linux kernel with Xen support. This kernel, whose sources are available at /usr/src/linux , will be our main running kernel (i.e. the one running domain 0). In the XEN section you'll find drivers for all kinds of input/output, each driver having a backend and frontend implementation available. For the domain 0 kernel you need to select the backend implementation: these are used by the other domains (who use the frontend drivers) to communicate directly with the hardware. However, you should be able to configure the kernel to provide support for both frontend (guest) and backend (host) drivers.

If you're wondering about networking: each interface in a domain has a point-to-point link to an interface on domain 0 (called vifX.Y where X is the domain number and Y the Yth interface of that domain), so you can configure your network the way you want (bridging, NAT, etc.)

Enable general Xen support:

KERNEL Xen Base
Processor type and features  --->
     [*] Linux guest support  --->
          [*]   Enable paravirtualization code
          [*]     Xen guest support
          [*]       Support for running as a PVH guest
          [*]     Paravirtualization layer for spinlocks

Add support for paravirtualized console connections:

KERNEL PV Console
Device Drivers  --->
     Character devices  --->
          [*] Xen Hypervisor Console support
          [*]   Xen Hypervisor Multiple Consoles support

Facilitates guest access to block and network devices via dom0:

KERNEL Disk and Network
Device Drivers  --->
     [*] Block devices  --->
          <*>   Xen virtual block device support
     [*] Network device support  --->
          <*>   Xen network device frontend driver

In some configurations it can be desirable to provide a guest with direct access to a PCI device. This is known as Xen PCI Passthrough :

KERNEL Guest PCI Passthrough
Bus options (PCI etc.)  --->
     <*> Xen PCI Frontend

Keyboard, mouse, and display support via dom0 backend:

KERNEL Guest Human Interface
Device Drivers  --->
     Input device support  --->
          [*]   Miscellaneous devices  --->
               <*>   Xen virtual keyboard and mouse support
     Graphics support  --->
          Frame buffer Devices  --->
               <*> Xen virtual frame buffer support

Xen dom0 support depends on APCI; without it dom0 related options will be hidden:

KERNEL APCI Support
Power management and ACPI options  --->
     [*] ACPI (Advanced Configuration and Power Interface) Support  --->

Typical network configurations depend on linux bridge functionality:

KERNEL Linux Bridge
[*] Networking support --->
     Networking options  --->
          <*> 802.1d Ethernet Bridging
	  [*] Network packet filtering framework (Netfilter) --->
	       [*] Advanced netfilter configuration
	       [*]   Bridged IP/ARP packets filtering

Fully virtualised (HVM) guests depend on Universal TUN/TAP device driver support for their network interfaces. This option is required if you plan to create fully Emulated Network Devices within Dom0/DomU configuration.

KERNEL TUN and TAP virtual network kernel devices
[*] Device Drivers  --->
     Network device support --->
     <M> Universal TUN/TAP device driver support

The remaining drivers flesh out memory management, domain-to-domain communication, and communication to Xen via sysfs interfaces:

KERNEL Xen Drivers
Device Drivers  --->
     [*] Block devices  --->
          <*>   Xen block-device backend driver
     [*] Network device support --->
          <*>   Xen backend network device
     Xen driver support  --->
          [*] Xen memory balloon driver
          [*]   Scrub pages before returning them to system
          <*> Xen /dev/xen/evtchn device
          [*] Backend driver support
          <*> Xen filesystem
          [*]   Create compatibility mount point /proc/xen
          [*] Create xen entries under /sys/hypervisor
          <*> userspace grant access device driver
          <*> User-space grant reference allocator driver
          <M> Xen PCI-device backend driver
          <*> Xen ACPI processor
          [*] Xen platform mcelog

With all of the above configuration enabled, this kernel image should be able to boot as the dom0 host or as another domU guest. Note that the domU kernel can be slimmed down significantly if desired.

USE Flags

The first set of use flags correspond directly to the Xen hypervisor. Of these, ensure that EFI support is selected if you plan to configure EFI boot.

USE flags for app-emulation/xen The Xen virtual machine monitor

custom-cflags Build with user-specified CFLAGS (unsupported) global
debug Enable extra debug codepaths, like asserts and extra output. If you want to get meaningful backtraces see https://wiki.gentoo.org/wiki/Project:Quality_Assurance/Backtraces global
efi Adds efi boot support, requires LDFLAG -melf_x86_64 for amd64 local
flask Enable the Flask XSM module from NSA local

In addition to the core hypervisor Xen depends on a set of supporting libraries and management tools.

USE flags for app-emulation/xen-tools Xen tools including QEMU and xl

api Build the C libxenapi bindings local
custom-cflags Build with user-specified CFLAGS (unsupported) global
debug Enable extra debug codepaths, like asserts and extra output. If you want to get meaningful backtraces see https://wiki.gentoo.org/wiki/Project:Quality_Assurance/Backtraces global
doc Add extra documentation (API, Javadoc, etc). It is recommended to enable per package instead of globally global
flask Enable the Flask XSM module from NSA local
hvm Enable support for hardware based virtualization (VT-x,AMD-v) local
ocaml Enable support for the ocaml language local
ovmf Enable support to boot UEFI guest vm, needed by hvm local
pam Enable pam support local
pygrub Install the pygrub boot loader local
python Add optional support/bindings for the Python language global
qemu Enable IOEMU support via the use of qemu-dm local
screen Enable support for running domain U console in an app-misc/screen session local
sdl Add support for Simple Direct Layer (media library) global
static-libs Build static versions of dynamic libraries as well global
system-qemu Using app-emulation/qemu instead of the bundled one local
system-seabios Using sys-firmware/seabios instead of the bundled one local

Note
In case you need hardware virtualization, the 'hvm' USE flag must be enabled for the xen-tools package. This flag can not be enabled on no-multilib installations. You need a multilib installation (and have a multilib profile selected) to enable the 'hvm' USE flag. (You also need a CPU which supports hardware virtualization.)

In case 'equery uses xen-tools' does not list the 'hvm' flag at all, or when 'emerge -a xen-tools' shows this flag in brackets, the flag cannot be enabled (unless you figure out how to unmask it). Simply selecting a multilib profile on hosts that do not have a multilib installation is likely to give you error messages like about a missing stub-32.h when the xen-tools package is compiled, and emerging the package fails. You may need to re-install the host to a multilib installation then. (32bit support in the kernel is not required.)

Installation

Emerge

Xen will automatically deploy the compiled, boot-able hypervisor in /boot. If you have a separate boot file system ensure that it is mounted.

root #mount /boot

After you have reviewed and updated the relevant USE flags, emerge xen and xen-tools:

root #emerge --ask app-emulation/xen app-emulation/xen-tools

Once the installation process completes /boot/xen.gz should be present.

Performance Tuning

When configuring a host with a significant amount of memory it is beneficial to fix the allocated to dom0 and disable ballooning functionality. This ensures Xen and the dom0 operating system are able to build working structures based on the memory available at boot-time and have their assumptions hold throughout operation. Additionally, in the event that a large number of guests are created, dom0 is explicitly protected from external memory pressure.

In a GRUB2 environment the following directive assigns 1 GiB of memory to the dom0 guest.

FILE /etc/default/grubGRUB2 Limit dom0 Memory
GRUB_CMDLINE_XEN_DEFAULT="dom0_mem=1024M,max:1024M"

Ballooning should also be disabled by setting autoballoon to either "off" or "auto" in /etx/xen/xl.conf.

Further detail on this topic and other related tuning suggestions can be found on the Tuning Xen for Performance article on the Xen wiki.

Runlevel

In order for your new Xen environment to function correctly a number of daemons must be started. The following table provides a brief overview of each daemon and its function.

Init Script Required Description
xencommons yes This daemon ensures all Xen related kernel modules are loaded in a dom0 system. Additionally, in dom0 and domU pv_ops systems it mounts /proc/xen as type xenfs.
xenconsoled yes Manages xl console sessions between dom0 and domU guests.
xendomains no Automatically starts all domU configuration files found in /etc/xen/auto/.
xenstored yes Manages both disk and network I/O configuration and operations between the dom0 environment and other domU guests.
xen-watchdog  ? Probably kills the dom0 environment or Xen in general if it becomes unresponsive for too long. It looks like this may have been removed
xenqemudev  ? Facilitates QEMU based disks for Xen guests. This comment is incomplete.
Note
The xen-watchdog init script currently has not been migrated from sysvinit to the runscript/openRC format.

To add these daemons directly to the default runlevel perform the following:

root #rc-update add xencommons default
root #rc-update add xenconsoled default
root #rc-update add xendomains default
root #rc-update add xenstored default
root #rc-update add xen-watchdog xen

These daemons should only be started when the system boots into Xen. As this may not always be the case, it may be desirable to keep these services out of the default runlevel. One possible alternative is to utilize an alternate default runlevel via the softlevel kernel parameter. The following example, based on this blog post, creates a new runlevel named xen.

First create the new runlevel:

root #install -d /etc/runlevels/xen

Populate it with everything currently present in the existing default runlevel:

CODE
for SERVICE in $(ls /etc/runlevels/default)
do
    rc-update add $SERVICE xen
done

Then instead of adding the xen daemons to the default runlevel add them to the xen level:

root #rc-update add xencommons xen
root #rc-update add xenconsoled xen
root #rc-update add xendomains xen
root #rc-update add xenstored xen
root #rc-update add xen-watchdog xen

Once all of these steps have been completed add the following statement to the grub default file:

FILE /etc/default/grubSetting the softlevel
# Replace the default runlevel with "xen"
GRUB_CMDLINE_LINUX_XEN_REPLACE_DEFAULT="softlevel=xen"

Bootloader

In order for Xen to be the highest privilege process it must be the initial boot target, however, it must also subsequently boot the dom0 kernel.

Legacy GRUB

The following configuration handles basic GRUB settings:

FILE /boot/grub/grub.cfgGRUB Configuration for Xen
title Xen Gentoo Linux 4.0
root (hd0,0)
kernel /boot/xen.gz
module /boot/kernel-4.0.x.y-xen0 root=/dev/sda3

In the above configuration GRUB will load xen.gz as the kernel with the actual dom0 kernel provided as a module.

GRUB2

GRUB2 provides auto-configuration scripts through grub2-mkconfig which will generate the appropriate settings based on your kernel .config. If the scripts detect Xen dom0 options they will append Xen hypervisor boot lines to the grub menu. Note that for this to function correctly the kernel config file must be located in one of the following directories with a suffix matching the desired kernel:

Directory File Prefix Example
/etc/kernels/ kernel-config-* /etc/kernels/kernel-config-4.0.5-gentoo
/boot config-* /boot/config-4.0.5-gentoo

The example column above assumes a kernel named /boot/kernel-4.0.5-gentoo.

The following variables in /etc/default/grub are the most common ones to set to control how GRUB2 will function:

Variable Explanation
GRUB_CMDLINE_XEN Extra parameters to be passed on the Xen command line for all Xen menu entries. For instance, to have a timestamped Xen dmesg, users will need to add GRUB_CMDLINE_XEN="console_timestamps=boot".
GRUB_CMDLINE_XEN_DEFAULT Parameters to be passed on the Xen command line for non-recovery Xen menu entries.

For a more complete list, please refer to the GRUB2 configuration variables sub-page.

GRUB2 support is also included in genkernel and can be triggered during the build process:

root #genkernel --oldconfig --menuconfig --install --bootloader=grub --symlink --disklabel --lvm --mdadm --makeopts=-j9 all

Booting Xen

Once you are satisfied with your configuration, reboot the system into Xen. Assuming everything worked correctly Xen should transfer display control to the dom0 environment, which will operate just like your original bare-metal system.

Running xl list should result in output similar to the following if Xen is functioning correctly:

root #xl list
Name                                        ID   Mem VCPUs      State   Time(s)
Domain-0                                     0  1024    24     r-----      22.0

At this point you may want to configure your system to boot into Xen by default. If you are utilizing GRUB2, add the following lines to the grub defaults file:

FILE /etc/default/grubEnabling Saved Defaults
# Default menu entry
GRUB_DEFAULT="saved"
GRUB_SAVEDEFAULT="true"

Then, copy the name or ID of the Xen boot config and set it via grub2-set-default:

root #grub2-set-default 'Gentoo GNU/Linux, with Xen hypervisor'

Creating an Unpriviledged Domain (domU)

Building the Kernel

Go to the Xen-powered Linux kernel source and, if necessary, update the configuration. It is wise to keep as many topics as possible similar to the main kernel. Then build the kernel and place the resulting vmlinuz file where you want (we assume this is /mnt/data/xen/kernel ):

root #make O=~/build/domU
root #cp ~/build/domU/vmlinuz /mnt/data/xen/kernel/kernel-3.5.x.y-xen
Note
On modern systems (e.g. xen-4.3.2, kernel-3.12) it is possible to use only one kernel for both: dom0 and domU. Just configure all options needed for a guest system in the main kernel. Don't forget to copy the modules /lib/modules/<your kernel> to the guest system.

If you'd like to trim down the domU kernel the following flags are necessary.

Enable general Xen support:

KERNEL Xen Base
Processor type and features  --->
     [*] Linux guest support  --->
          [*]   Enable paravirtualization code
          [*]     Xen guest support
          [*]       Support for running as a PVH guest
          [*]     Paravirtualization layer for spinlocks

Facilitates guest access to block and network devices via dom0:

KERNEL Disk and Network
Device Drivers  --->
     [*] Block devices  --->
          <*>   Xen virtual block device support
     [*] Network device support  --->
          <*>   Xen network device frontend driver

In some configurations it can be desirable to provide a guest with direct access to a PCI device. This is known as Xen PCI Passthrough :

KERNEL Guest PCI Passthrough
Bus options (PCI etc.)  --->
     [*] Xen PCI Frontend

Keyboard, mouse, and display support via dom0 backend:

KERNEL Guest Human Interface
Device Drivers  --->
     Input device support  --->
          [*]   Miscellaneous devices  --->
               <*>   Xen virtual keyboard and mouse support
     Graphics support  --->
          Frame buffer Devices  --->
               <*> Xen virtual frame buffer support

The remaining drivers flesh out memory management, domain-to-domain communication, and communication to Xen via sysfs interfaces:

KERNEL Xen Drivers
Device Drivers  --->
     Xen driver support  --->
          [*] Xen memory balloon driver
          [*]   Scrub pages before returning them to system
          <*> Xen /dev/xen/evtchn device
          <*> Xen filesystem
          [*]   Create compatibility mount point /proc/xen
          [*] Create xen entries under /sys/hypervisor
          <*> userspace grant access device driver
          <*> User-space grant reference allocator driver
          <*> Xen ACPI processor
          [*] Xen platform mcelog

Creating the Domain Disks

For best performance, it is best to dedicate a partition (or logical volume) to a domain rather than a file based filesystem. However, if you are going to use Xen primarily for tests using a file based filesystem does have its advantages (especially regarding maintenance).

You can create a file based filesystem using dd and mke2fs (or any other file system creation tool). For instance, to create a 4 Gbyte ext4 filesystem:

root #dd if=/dev/zero of=/mnt/data/xen/disks/ext4root.img bs=1M count=4096
root #mkfs.ext4 /mnt/data/xen/disks/ext4root.img

Configuring a Domain

Next we create a Xen configuration file for a domain. You can store these configuration files where you want, for instance at /mnt/data/xen/configs . As an example, we create a configuration file for a small Gentoo environment which uses the disk image we created previously:

FILE /mnt/data/xen/configs/gentoo
kernel = "/mnt/data/xen/kernel/kernel-3.5.x.y-xen"
memory = 512
name   = "gentoo"
(Map the disk image to the virtual /dev/sda1)
disk   = ['file:/mnt/data/xen/disks/ext4root.img,sda1,w']
root   = "/dev/sda1 ro"

If you are using a block device (such as an lvm volume or partition) for the disk use 'phy:' instead of 'file:' and leave off /dev. For example:

CODE Using a block device
(LVM Volume)
disk = [ 'phy:lvm/xen-guest-root,sda1,w' ]
  
(Physical Partition)
disk = [ 'phy:sdb6,sda1,w' ]

You can find example configuration files in /etc/xen, which is also the default location for domU config files.

Launching the New Domain

Now we're all set and we can launch the new domain. If the disk image contained an operating system, we could just create and attach the domain using the xl command:

root #xl create /mnt/data/xen/gentoo -c

The domain would be booted inside the terminal in which you executed the command. However, in our case, the disk image is empty so the domain won't boot up in anything useful. To fix this, you can loop-mount the image and install Gentoo as you're used to.

If you want to disconnect from the domain, press Ctrl+] . You can always reconnect to the domains' console using xl console gentoo . However, there is only one console per domain, so only use it when you can't access the domain otherwise (for instance, through SSH).

If you are missing login prompt on the console, make sure you have entries like this in your inittab files on dom0 and domU pointing to /dev/hvc0:

FILE dom0:/etc/inittab
c0:2345:respawn:/sbin/agetty 38400 hvc0 linux
FILE domU:/etc/inittab
c0:2345:respawn:/sbin/agetty 38400 hvc0 linux

To apply the changes in /etc/inittab without a reboot issue the following command:

root #telinit q

If it still does not work, check the kernel config for the entries: DEVTMPFS and DEVTMPFS_MOUNT. They should be set.

KERNEL Kernel Config DEVTMPFS
Device Drivers --->
    Generic Driver Options  --->
        -*- Maintain a devtmpfs filesystem to mount at /dev
        [*]   Automount devtmpfs at /dev, after the kernel mounted the rootfs

Networking on Domains

Introduction

Xen works best when using a bridged mode network configuration. This means that your default network interface on the administrative domain becomes a bridge which accepts connections to the virtual domains as well as to the IP address your administrative domain has.

Bridged Interfaces

Create a bridge interface by creating a new link to the networking init script as provided by Gentoo:

root #cd /etc/init.d
root #ln -s net.lo net.br0

Next, edit /etc/conf.d/net and setup the bridge:

FILE dom0:/etc/conf.d/net
# eth0 should NOT have an ip configured
config_eth0="null"

# configure bridge to replace eth0 on dom0. Make sure the netmask for the bridge includes ip addresses of all your domU's!
bridge_br0="eth0"
config_br0="192.168.XX.XX netmask 255.255.0.0 brd 192.168.255.255"
routes_br0="default via 192.168.XX.XX"
mac_br0="00:16:3e:5b:XX:XX"

# bridge options to make interface coming up immediately
brctl_br0="stp off
        setfd 0
        sethello 10"

rc_net_br0_need="net.eth0"
rc_net_br0_provide="!net"
FILE domU:/etc/conf.d/net
config_eth0="192.168.1.200 netmask 255.255.255.0 brd 192.168.1.255"
# route all trafic through dom0 bridge address
routes_eth0="default via 192.168.XX.XX"
# make sure all your domU's have different mac addresses! Set them if needed in xen domU 
# config files in /etc/xen/<domU_name> with "vif = [ "ip=192.68.XX.XX,mac=XX:XX:XX:XX:XX:XX,bridge=br0" ];" !

Finally, install the net-misc/bridge-utils package, and make sure the net.br0 init script is loaded at boot.

root #emerge net-misc/bridge-utils
root #rc-update add net.br0 default

If you use bridged networks with real internet IP's in hosted environments, it may be necessary to add one or all of the following lines (depending on your environment) in your /etc/sysctl.conf file to prevent redirects, that can cause intermittent network interruptions:

FILE /etc/sysctl.conf
net.ipv4.conf.all.send_redirects=0
net.ipv4.conf.eth0.send_redirects=0
net.ipv4.conf.br0.send_redirects=0
net.ipv4.conf.default.send_redirects=0

To get the changes in /etc/sysctl.conf to work, use:

root #sysctl -p /etc/sysctl.conf

if you encounter a poor network performance or if your domU network permanently stops working under heavy load (backup jobs, etc) (from outside it looks, like the instance would crash, but deactivating and activating the interface e.g. form the xl console <domU name> with /etc/init.d/net.eth0 stop/start, restores normal operation) , use ethtool to improve/prevent it on all interfaces connected to the bridge (don't forget the bridge himself):

root #ethtool --offload <network device> gso off tso off sg off gro off
Note
You have to do it after each reboot, so use e.g. /etc/crontab to make it permanent.

Further Resources

Xen Documentation

Xen Tools

Xen Tuning

  • Xen network tuning
    This article is based on a document formerly found on our main website gentoo.org.
    The following people contributed to the original document: Sven Vermeulen, nightmorph
    They are listed here as the Wiki history does not allow for any external attribution. If you edit the Wiki article, please do not add yourself here; your contributions are recorded on the history page.