README.md

Vagrant Libvirt Provider

This is a Vagrant plugin that adds a Libvirt provider to Vagrant, allowing Vagrant to control and provision machines via Libvirt toolkit.

Note: Actual version is still a development one. Feedback is welcome and can help a lot :-)

Index

• Features
• Future work
• Using the container image
  • Using Docker
  • Using Podman
  • Extending the Docker image with additional vagrant plugins
• Installation
  • Possible problems with plugin installation on Linux
  • Additional Notes for Fedora and Similar Linux Distributions
• Vagrant Project Preparation
  • Add Box
  • Create Vagrantfile
  • Start VM
  • How Project Is Created
  • Libvirt Configuration
  • Provider Options
  • Domain Specific Options
    • Reload behavior
• Networks
  • Private Network Options
  • Public Network Options
  • Management Network
• Additional Disks
  • Reload behavior
• CDROMs
• Input
• PCI device passthrough
• Using USB Devices
  • USB Controller Configuration
  • USB Device Passthrough
  • USB Redirector Devices
    • Filter for USB Redirector Devices
• Random number generator passthrough
• Serial Console Devices
• Watchdog device
• Smartcard device
• Hypervisor Features
• Clock
• CPU features
• Memory Backing
• No box and PXE boot
• SSH Access To VM
• Forwarded Ports
  • Forwarding the ssh-port
• Synced Folders
• QEMU Session Support
• Customized Graphics
• TPM Devices
• Memory balloon
• Libvirt communication channels
• Custom command line arguments and environment variables
• Box Formats
  • Version 1
  • Version 2 (Experimental)
• Create Box
• Package Box from VM
• Troubleshooting VMs
• Development
• Contributing

Features

• Control local Libvirt hypervisors.
• Vagrant up, destroy, suspend, resume, halt, ssh, reload, package and provision commands.
• Upload box image (qcow2 format) to Libvirt storage pool.
• Create volume as COW diff image for domains.
• Create private networks.
• Create and boot Libvirt domains.
• SSH into domains.
• Setup hostname and network interfaces.
• Provision domains with any built-in Vagrant provisioner.
• Synced folder support via rsync, nfs, 9p or virtiofs.
• Snapshots via sahara.
• Package caching via vagrant-cachier.
• Use boxes from other Vagrant providers via vagrant-mutate.
• Support VMs with no box for PXE boot purposes (Vagrant 1.6 and up)

Future work

• Take a look at open issues.

Using the container image

Due to the number of issues encountered around compatibility between the ruby runtime environment that is part of the upstream vagrant installation and the library dependencies of libvirt that this project requires to communicate with libvirt, there is a docker image build and published.

This should allow users to execute vagrant with vagrant-libvirt without needing to deal with the compatibility issues, though you may need to extend the image for your own needs should you make use of additional plugins.

Note the default image contains the full toolchain required to build and install vagrant-libvirt and it's dependencies. There is also a smaller image published with the -slim suffix if you just need vagrant-libvirt and don't need to install any additional plugins for your environment.

If you are connecting to a remote system libvirt, you may omit the -v /var/run/libvirt/:/var/run/libvirt/ mount bind. Some distributions patch the local vagrant environment to ensure vagrant-libvirt uses qemu:///session, which means you may need to set the environment variable LIBVIRT_DEFAULT_URI to the same value if looking to use this in place of your distribution provided installation.

Using Docker

To get the image with the most recent release:

docker pull vagrantlibvirt/vagrant-libvirt:latest

---

Note If you want the very latest code you can use the edge tag instead.

docker pull vagrantlibvirt/vagrant-libvirt:edge

---

Running the image:

docker run -it --rm \
  -e LIBVIRT_DEFAULT_URI \
  -v /var/run/libvirt/:/var/run/libvirt/ \
  -v ~/.vagrant.d:/.vagrant.d \
  -v $(realpath "${PWD}"):${PWD} \
  -w $(realpath "${PWD}") \
  --network host \
  vagrantlibvirt/vagrant-libvirt:latest \
    vagrant status

It's possible to define a function in ~/.bashrc, for example:

vagrant(){
  docker run -it --rm \
    -e LIBVIRT_DEFAULT_URI \
    -v /var/run/libvirt/:/var/run/libvirt/ \
    -v ~/.vagrant.d:/.vagrant.d \
    -v $(realpath "${PWD}"):${PWD} \
    -w $(realpath "${PWD}") \
    --network host \
    vagrantlibvirt/vagrant-libvirt:latest \
      vagrant $@
}

Using Podman

Preparing the podman run, only once:

mkdir -p ~/.vagrant.d/{boxes,data,tmp}

N.B. This is needed until the entrypoint works for podman to only mount the ~/.vagrant.d directory

To run with Podman you need to include

  --entrypoint /bin/bash \
  --security-opt label=disable \
  -v ~/.vagrant.d/boxes:/vagrant/boxes \
  -v ~/.vagrant.d/data:/vagrant/data \
  -v ~/.vagrant.d/data:/vagrant/tmp \

for example:

vagrant(){
  podman run -it --rm \
    -e LIBVIRT_DEFAULT_URI \
    -v /var/run/libvirt/:/var/run/libvirt/ \
    -v ~/.vagrant.d/boxes:/vagrant/boxes \
    -v ~/.vagrant.d/data:/vagrant/data \
    -v ~/.vagrant.d/data:/vagrant/tmp \
    -v $(realpath "${PWD}"):${PWD} \
    -w $(realpath "${PWD}") \
    --network host \
    --entrypoint /bin/bash \
    --security-opt label=disable \
    docker.io/vagrantlibvirt/vagrant-libvirt:latest \
      vagrant $@
}

Running Podman in rootless mode maps the root user inside the container to your host user so we need to bypass entrypoint.sh and mount persistent storage directly to /vagrant.

Extending the Docker image with additional vagrant plugins

By default the image published and used contains the entire tool chain required to reinstall the vagrant-libvirt plugin and it's dependencies, as this is the default behaviour of vagrant anytime a new plugin is installed. This means it should be possible to use a simple FROM statement and ask vagrant to install additional plugins.

FROM vagrantlibvirt/vagrant-libvirt:latest

RUN vagrant plugin install <plugin>

Installation

First, you should have both QEMU and Libvirt installed if you plan to run VMs on your local system. For instructions, refer to your Linux distribution's documentation.

NOTE: Before you start using vagrant-libvirt, please make sure your Libvirt and QEMU installation is working correctly and you are able to create QEMU or KVM type virtual machines with virsh or virt-manager.

Next, you must have Vagrant installed. Vagrant-libvirt supports Vagrant 2.0, 2.1 & 2.2. It should also work with earlier releases from 1.5 onwards but they are not actively tested.

Check the unit tests for the current list of tested versions.

We only test with the upstream version! If you decide to install your distro's version and you run into problems, as a first step you should switch to upstream.

Now you need to make sure your have all the build dependencies installed for vagrant-libvirt. This depends on your distro. An overview:

• Ubuntu 18.10, Debian 9 and up:

apt-get build-dep vagrant ruby-libvirt
apt-get install qemu libvirt-daemon-system libvirt-clients ebtables dnsmasq-base
apt-get install libxslt-dev libxml2-dev libvirt-dev zlib1g-dev ruby-dev
apt-get install libguestfs-tools

• Ubuntu 18.04, Debian 8 and older:

apt-get build-dep vagrant ruby-libvirt
apt-get install qemu libvirt-bin ebtables dnsmasq-base
apt-get install libxslt-dev libxml2-dev libvirt-dev zlib1g-dev ruby-dev
apt-get install libguestfs-tools

(It is possible some users will already have libraries from the third line installed, but this is the way to make it work OOTB.)

• CentOS 6, 7, Fedora 21:

yum install qemu libvirt libvirt-devel ruby-devel gcc qemu-kvm libguestfs-tools

• Fedora 22 and up:

dnf install -y gcc libvirt libvirt-devel libxml2-devel make ruby-devel libguestfs-tools

• OpenSUSE leap 15.1:

zypper install qemu libvirt libvirt-devel ruby-devel gcc qemu-kvm libguestfs

• Arch Linux: please read the related ArchWiki page.

pacman -S vagrant

Now you're ready to install vagrant-libvirt using standard Vagrant plugin installation methods.

For some distributions you will need to specify CONFIGURE_ARGS variable before running vagrant plugin install:

• Fedora 32 + upstream Vagrant:

  export CONFIGURE_ARGS="with-libvirt-include=/usr/include/libvirt with-libvirt-lib=/usr/lib64"

vagrant plugin install vagrant-libvirt

Possible problems with plugin installation on Linux

In case of problems with building nokogiri and ruby-libvirt gem, install missing development libraries for libxslt, libxml2 and libvirt.

On Ubuntu, Debian, make sure you are running all three of the apt commands above with sudo.

On RedHat, Centos, Fedora, ...

$ sudo dnf install libxslt-devel libxml2-devel libvirt-devel ruby-devel gcc

On Arch Linux it is recommended to follow steps from ArchWiki.

If have problem with installation - check your linker. It should be ld.gold:

sudo alternatives --set ld /usr/bin/ld.gold
# OR
sudo ln -fs /usr/bin/ld.gold /usr/bin/ld

If you have issues building ruby-libvirt, try the following:

CONFIGURE_ARGS='with-ldflags=-L/opt/vagrant/embedded/lib with-libvirt-include=/usr/include/libvirt with-libvirt-lib=/usr/lib' GEM_HOME=~/.vagrant.d/gems GEM_PATH=$GEM_HOME:/opt/vagrant/embedded/gems PATH=/opt/vagrant/embedded/bin:$PATH vagrant plugin install vagrant-libvirt

Additional Notes for Fedora and Similar Linux Distributions

If you encounter the following load error when using the vagrant-libvirt plugin (note the required by libssh):

/opt/vagrant/embedded/lib/ruby/2.4.0/rubygems/core_ext/kernel_require.rb:55:in `require': /opt/vagrant/embedded/lib64/libcrypto.so.1.1: version `OPENSSL_1_1_1b' not found (required by /lib64/libssh.so.4) - /home/xxx/.vagrant.d/gems/2.4.6/gems/ruby-libvirt-0.7.1/lib/_libvirt.so (LoadError)

then the following steps have been found to resolve the problem. Thanks to James Reynolds (see hashicorp/vagrant#11020 (comment)). The specific version of libssh will change over time so references to the rpm in the commands below will need to be adjusted accordingly.

# Fedora
dnf download --source libssh

# centos 8 stream, doesn't provide source RPMs, so you need to download like so
git clone https://git.centos.org/centos-git-common
# centos-git-common needs its tools in PATH
export PATH=$(readlink -f ./centos-git-common):$PATH
git clone https://git.centos.org/rpms/libssh
cd libssh
git checkout imports/c8s/libssh-0.9.4-1.el8
into_srpm.sh -d c8s
cd SRPMS

# common commands (make sure to adjust verison accordingly)
rpm2cpio libssh-0.9.0-5.fc30.src.rpm | cpio -imdV
tar xf libssh-0.9.0.tar.xz
mkdir build
cd build
cmake ../libssh-0.9.0 -DOPENSSL_ROOT_DIR=/opt/vagrant/embedded/
make
sudo cp lib/libssh* /opt/vagrant/embedded/lib64

If you encounter the following load error when using the vagrant-libvirt plugin (note the required by libk5crypto):

/opt/vagrant/embedded/lib/ruby/2.4.0/rubygems/core_ext/kernel_require.rb:55:in `require': /usr/lib64/libk5crypto.so.3: undefined symbol: EVP_KDF_ctrl, version OPENSSL_1_1_1b - /home/rbelgrave/.vagrant.d/gems/2.4.9/gems/ruby-libvirt-0.7.1/lib/_libvirt.so (LoadError)

then the following steps have been found to resolve the problem. After the steps below are complete, then reinstall the vagrant-libvirt plugin without setting the CONFIGURE_ARGS. Thanks to Marco Bevc (see hashicorp/vagrant#11020 (comment)):

# Fedora
dnf download --source krb5-libs

# centos 8 stream, doesn't provide source RPMs, so you need to download like so
git clone https://git.centos.org/centos-git-common
# centos-git-common needs its tools in PATH
export PATH=$(readlink -f ./centos-git-common):$PATH
git clone https://git.centos.org/rpms/krb5
cd krb5
git checkout imports/c8s/krb5-1.18.2-8.el8
into_srpm.sh -d c8s
cd SRPMS

# common commands (make sure to adjust verison accordingly)
rpm2cpio krb5-1.18-1.fc32.src.rpm | cpio -imdV
tar xf krb5-1.18.tar.gz
cd krb5-1.18/src
./configure
make
sudo cp -P lib/crypto/libk5crypto.* /opt/vagrant/embedded/lib64/

Vagrant Project Preparation

Add Box

After installing the plugin (instructions above), the quickest way to get started is to add Libvirt box and specify all the details manually within a config.vm.provider block. So first, add Libvirt box using any name you want. You can find more Libvirt-ready boxes at Vagrant Cloud. For example:

vagrant init fedora/32-cloud-base

Create Vagrantfile

And then make a Vagrantfile that looks like the following, filling in your information where necessary. For example:

Vagrant.configure("2") do |config|
  config.vm.define :test_vm do |test_vm|
    test_vm.vm.box = "fedora/32-cloud-base"
  end
end

Start VM

In prepared project directory, run following command:

$ vagrant up --provider=libvirt

Vagrant needs to know that we want to use Libvirt and not default VirtualBox. That's why there is --provider=libvirt option specified. Other way to tell Vagrant to use Libvirt provider is to setup environment variable

export VAGRANT_DEFAULT_PROVIDER=libvirt

How Project Is Created

Vagrant goes through steps below when creating new project:

1. Connect to Libvirt locally or remotely via SSH.
2. Check if box image is available in Libvirt storage pool. If not, upload it to remote Libvirt storage pool as new volume.
3. Create COW diff image of base box image for new Libvirt domain.
4. Create and start new domain on Libvirt host.
5. Check for DHCP lease from dnsmasq server.
6. Wait till SSH is available.
7. Sync folders and run Vagrant provisioner on new domain if setup in Vagrantfile.

Libvirt Configuration

Provider Options

Although it should work without any configuration for most people, this provider exposes quite a few provider-specific configuration options. The following options allow you to configure how vagrant-libvirt connects to Libvirt, and are used to generate the Libvirt connection URI:

• driver - A hypervisor name to access. For now only KVM and QEMU are supported
• host - The name of the server, where Libvirtd is running
• connect_via_ssh - If use ssh tunnel to connect to Libvirt. Absolutely needed to access Libvirt on remote host. It will not be able to get the IP address of a started VM otherwise.
• username - Username and password to access Libvirt
• password - Password to access Libvirt
• id_ssh_key_file - If not nil, uses this ssh private key to access Libvirt. Default is $HOME/.ssh/id_rsa. Prepends $HOME/.ssh/ if no directory
• socket - Path to the Libvirt unix socket (e.g. /var/run/libvirt/libvirt-sock)
• proxy_command - For advanced usage. When connecting to remote libvirt instances, if the default constructed proxy_command which uses -W %h:%p does not work, set this as needed. It performs interpolation using {key} and supports only {host}, {username}, and {id_ssh_key_file}. This is to try and avoid issues with escaping % and $ which might be necessary to the ssh command itself. e.g.: libvirt.proxy_command = "ssh {host} -l {username} -i {id_ssh_key_file} nc %h %p"
• uri - For advanced usage. Directly specifies what Libvirt connection URI vagrant-libvirt should use. Overrides all other connection configuration options

In the event that none of these are set (excluding the driver option) the provider will attempt to retrieve the uri from the environment variable LIBVIRT_DEFAULT_URI similar to how virsh works. If any of them are set, it will ignore the environment variable. The reason the driver option is ignored is that it is not uncommon for this to be explicitly set on the box itself and there is no easily to determine whether it is being set by the user or the box packager.

Connection-independent options:

• storage_pool_name - Libvirt storage pool name, where box image and instance snapshots (if snapshot_pool_name is not set) will be stored.
• snapshot_pool_name - Libvirt storage pool name. If set, the created snapshot of the instance will be stored at this location instead of storage_pool_name.

For example:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.host = "example.com"
  end
end

Domain Specific Options

• title - A short description of the domain.

• description - A human readable description of the virtual machine.

• disk_bus - The type of disk device to emulate. Defaults to virtio if not set. Possible values are documented in Libvirt's description for target. NOTE: this option applies only to disks associated with a box image. To set the bus type on additional disks, see the Additional Disks section.

• disk_device - The disk device to emulate. Defaults to vda if not set, which should be fine for paravirtualized guests, but some fully virtualized guests may require hda. NOTE: this option also applies only to disks associated with a box image.

• disk_driver - Extra options for the main disk driver (see Libvirt documentation). NOTE: this option also applies only to disks associated with a box image. In all cases, the value nil can be used to force the hypervisor default behaviour (e.g. to override settings defined in top-level Vagrantfiles). Supported options include:

  • :cache - Controls the cache mechanism. Possible values are "default", "none", "writethrough", "writeback", "directsync" and "unsafe".
  • :io - Controls specific policies on I/O. Possible values are "threads" and "native".
  • :copy_on_read - Controls whether to copy read backing file into the image file. The value can be either "on" or "off".
  • :discard - Controls whether discard requests (also known as "trim" or "unmap") are ignored or passed to the filesystem. Possible values are "unmap" or "ignore". Note: for discard to work, you will likely also need to set disk_bus = 'scsi'
  • :detect_zeroes - Controls whether to detect zero write requests. The value can be "off", "on" or "unmap".

• nic_model_type - parameter specifies the model of the network adapter when you create a domain value by default virtio KVM believe possible values, see the documentation for Libvirt.

• shares - Proportional weighted share for the domain relative to others. For more details see documentation.

• memory - Amount of memory in MBytes. Defaults to 512 if not set.

• cpus - Number of virtual cpus. Defaults to 1 if not set.

• cpuset - Physical cpus to which the vcpus can be pinned. For more details see documentation.

• cputopology - Number of CPU sockets, cores and threads running per core. All fields of :sockets, :cores and :threads are mandatory, cpus domain option must be present and must be equal to total count of sockets * cores * threads. For more details see documentation.

• nodeset - Physical NUMA nodes where virtual memory can be pinned. For more details see documentation.

  Vagrant.configure("2") do |config|
    config.vm.provider :libvirt do |libvirt|
      libvirt.cpus = 4
      libvirt.cpuset = '1-4,^3,6'
      libvirt.cputopology :sockets => '2', :cores => '2', :threads => '1'
    end
  end

• nested - Enable nested virtualization. Default is false.

• cpu_mode - CPU emulation mode. Defaults to 'host-model' if not set. Allowed values: host-model, host-passthrough, custom.

• cpu_model - CPU Model. Defaults to 'qemu64' if not set and cpu_mode is custom and to '' otherwise. This can really only be used when setting cpu_mode to custom.

• cpu_fallback - Whether to allow Libvirt to fall back to a CPU model close to the specified model if features in the guest CPU are not supported on the host. Defaults to 'allow' if not set. Allowed values: allow, forbid.

• numa_nodes - Specify an array of NUMA nodes for the guest. The syntax is similar to what would be set in the domain XML. memory must be in MB. Symmetrical and asymmetrical topologies are supported but make sure your total count of defined CPUs adds up to v.cpus.

  The sum of all the memory defined here will act as your total memory for your guest VM. This sum will override what is set in v.memory

  v.cpus = 4
  v.numa_nodes = [
    {:cpus => "0-1", :memory => "1024"},
    {:cpus => "2-3", :memory => "4096"}
  ]
  

• loader - Sets path to custom UEFI loader.

• kernel - To launch the guest with a kernel residing on host filesystems. Equivalent to qemu -kernel.

• initrd - To specify the initramfs/initrd to use for the guest. Equivalent to qemu -initrd.

• random_hostname - To create a domain name with extra information on the end to prevent hostname conflicts.

• default_prefix - The default Libvirt guest name becomes a concatenation of the <current_directory>_<guest_name>. The current working directory is the default prefix to the guest name. The default_prefix options allow you to set the guest name prefix.

• cmd_line - Arguments passed on to the guest kernel initramfs or initrd to use. Equivalent to qemu -append, only possible to use in combination with initrd and kernel.

• graphics_type - Sets the protocol used to expose the guest display. Defaults to vnc. Possible values are "sdl", "curses", "none", "gtk", "vnc" or "spice".

• graphics_port - Sets the port for the display protocol to bind to. Defaults to 5900.

• graphics_ip - Sets the IP for the display protocol to bind to. Defaults to "127.0.0.1".

• graphics_passwd - Sets the password for the display protocol. Working for vnc and Spice. by default working without passsword.

• graphics_autoport - Sets autoport for graphics, Libvirt in this case ignores graphics_port value, Defaults to 'yes'. Possible value are "yes" and "no"

• graphics_gl - Set to true to enable OpenGL. Defaults to true if video_accel3d is true.

• keymap - Set keymap for vm. default: en-us

• kvm_hidden - Hide the hypervisor from the guest. Useful for GPU passthrough on stubborn drivers. Default is false.

• video_type - Sets the graphics card type exposed to the guest. Defaults to "cirrus". Possible values are "vga", "cirrus", "vmvga", "xen", "vbox", or "qxl".

• video_vram - Used by some graphics card types to vary the amount of RAM dedicated to video. Defaults to 9216.

• video_accel3d - Set to true to enable 3D acceleration. Defaults to false.

• sound_type - Set the virtual sound card Defaults to "ich6".

• machine_type - Sets machine type. Equivalent to qemu -machine. Use qemu-system-x86_64 -machine help to get a list of supported machines.

• machine_arch - Sets machine architecture. This helps Libvirt to determine the correct emulator type. Possible values depend on your version of QEMU. For possible values, see which emulator executable qemu-system-* your system provides. Common examples are aarch64, alpha, arm, cris, i386, lm32, m68k, microblaze, microblazeel, mips, mips64, mips64el, mipsel, moxie, or32, ppc, ppc64, ppcemb, s390x, sh4, sh4eb, sparc, sparc64, tricore, unicore32, x86_64, xtensa, xtensaeb.

• machine_virtual_size - Sets the disk size in GB for the machine overriding the default specified in the box. Allows boxes to defined with a minimal size disk by default and to be grown to a larger size at creation time. Will ignore sizes smaller than the size specified by the box metadata. Note that currently there is no support for automatically resizing the filesystem to take advantage of the larger disk.

• emulator_path - Explicitly select which device model emulator to use by providing the path, e.g. /usr/bin/qemu-system-x86_64. This is especially useful on systems that fail to select it automatically based on machine_arch which then results in a capability error.

• boot - Change the boot order and enables the boot menu. Possible options are "hd", "network", "cdrom". Defaults to "hd" with boot menu disabled. When "network" is set without "hd", only all NICs will be tried; see below for more detail.

• nic_adapter_count - Defaults to '8'. Only use case for increasing this count is for VMs that virtualize switches such as Cumulus Linux. Max value for Cumulus Linux VMs is 33.

• uuid - Force a domain UUID. Defaults to autogenerated value by Libvirt if not set.

• suspend_mode - What is done on vagrant suspend. Possible values: 'pause', 'managedsave'. Pause mode executes a la virsh suspend, which just pauses execution of a VM, not freeing resources. Managed save mode does a la virsh managedsave which frees resources suspending a domain.

• tpm_model - The model of the TPM to which you wish to connect.

• tpm_type - The type of TPM device to which you are connecting.

• tpm_path - The path to the TPM device on the host system.

• tpm_version - The TPM version to use.

• dtb - The device tree blob file, mostly used for non-x86 platforms. In case the device tree isn't added in-line to the kernel, it can be manually specified here.

• autostart - Automatically start the domain when the host boots. Defaults to 'false'.

• channel - Libvirt channels. Configure a private communication channel between the host and guest, e.g. for use by the QEMU guest agent and the Spice/QXL graphics type.

• mgmt_attach - Decide if VM has interface in mgmt network. If set to 'false' it is not possible to communicate with VM through vagrant ssh or run provisioning. Setting to 'false' is only possible when VM doesn't use box. Defaults set to 'true'.

• serial - libvirt serial devices. Configure a serial/console port to communicate with the guest. Can be used to log to file boot time messages sent to ttyS0 console by the guest.

Specific domain settings can be set for each domain separately in multi-VM environment. Example below shows a part of Vagrantfile, where specific options are set for dbserver domain.

Vagrant.configure("2") do |config|
  config.vm.define :dbserver do |dbserver|
    dbserver.vm.box = "centos64"
    dbserver.vm.provider :libvirt do |domain|
      domain.memory = 2048
      domain.cpus = 2
      domain.nested = true
      domain.disk_driver :cache => 'none'
    end
  end

  # ...

The following example shows part of a Vagrantfile that enables the VM to boot from a network interface first and a hard disk second. This could be used to run VMs that are meant to be a PXE booted machines. Be aware that if hd is not specified as a boot option, it will never be tried.

Vagrant.configure("2") do |config|
  config.vm.define :pxeclient do |pxeclient|
    pxeclient.vm.box = "centos64"
    pxeclient.vm.provider :libvirt do |domain|
      domain.boot 'network'
      domain.boot 'hd'
    end
  end

  # ...

Reload behavior

On vagrant reload the following domain specific attributes are updated in defined domain:

• disk_bus - Is updated only on disks. It skips CDROMs
• nic_model_type - Updated
• memory - Updated
• cpus - Updated
• nested - Updated
• cpu_mode - Updated. Pay attention that custom mode is not supported
• graphics_type - Updated
• graphics_port - Updated
• graphics_ip - Updated
• graphics_passwd - Updated
• graphics_autoport - Updated
• keymap - Updated
• video_type - Updated
• video_vram - Updated
• tpm_model - Updated
• tpm_type - Updated
• tpm_path - Updated
• tpm_version - Updated

Networks

Networking features in the form of config.vm.network support private networks concept. It supports both the virtual network switch routing types and the point to point Guest OS to Guest OS setting using UDP/Mcast/TCP tunnel interfaces.

http://wiki.libvirt.org/page/VirtualNetworking

https://libvirt.org/formatdomain.html#elementsNICSTCP

http://libvirt.org/formatdomain.html#elementsNICSMulticast

http://libvirt.org/formatdomain.html#elementsNICSUDP (in Libvirt v1.2.20 and higher)

Public Network interfaces are currently implemented using the macvtap driver. The macvtap driver is only available with the Linux Kernel version >= 2.6.24. See the following Libvirt documentation for the details of the macvtap usage.

http://www.libvirt.org/formatdomain.html#elementsNICSDirect

An examples of network interface definitions:

  # Private network using virtual network switching
  config.vm.define :test_vm1 do |test_vm1|
    test_vm1.vm.network :private_network, :ip => "10.20.30.40"
  end

  # Private network using DHCP and a custom network
  config.vm.define :test_vm1 do |test_vm1|
    test_vm1.vm.network :private_network,
      :type => "dhcp",
      :libvirt__network_address => '10.20.30.0'
  end

  # Private network (as above) using a domain name
  config.vm.define :test_vm1 do |test_vm1|
    test_vm1.vm.network :private_network,
      :ip => "10.20.30.40",
      :libvirt__domain_name => "test.local"
  end

  # Private network. Point to Point between 2 Guest OS using a TCP tunnel
  # Guest 1
  config.vm.define :test_vm1 do |test_vm1|
    test_vm1.vm.network :private_network,
      :libvirt__tunnel_type => 'server',
      # default is 127.0.0.1 if omitted
      # :libvirt__tunnel_ip => '127.0.0.1',
      :libvirt__tunnel_port => '11111'
    # network with ipv6 support
    test_vm1.vm.network :private_network,
      :ip => "10.20.5.42",
      :libvirt__guest_ipv6 => "yes",
      :libvirt__ipv6_address => "2001:db8:ca2:6::1",
      :libvirt__ipv6_prefix => "64"

  # Guest 2
  config.vm.define :test_vm2 do |test_vm2|
    test_vm2.vm.network :private_network,
      :libvirt__tunnel_type => 'client',
      # default is 127.0.0.1 if omitted
      # :libvirt__tunnel_ip => '127.0.0.1',
      :libvirt__tunnel_port => '11111'
    # network with ipv6 support
    test_vm2.vm.network :private_network,
      :ip => "10.20.5.45",
      :libvirt__guest_ipv6 => "yes",
      :libvirt__ipv6_address => "2001:db8:ca2:6::1",
      :libvirt__ipv6_prefix => "64"


  # Public Network
  config.vm.define :test_vm1 do |test_vm1|
    test_vm1.vm.network :public_network,
      :dev => "virbr0",
      :mode => "bridge",
      :type => "bridge"
  end

In example below, one network interface is configured for VM test_vm1. After you run vagrant up, VM will be accessible on IP address 10.20.30.40. So if you install a web server via provisioner, you will be able to access your testing server on http://10.20.30.40 URL. But beware that this address is private to Libvirt host only. It's not visible outside of the hypervisor box.

If network 10.20.30.0/24 doesn't exist, provider will create it. By default created networks are NATed to outside world, so your VM will be able to connect to the internet (if hypervisor can). And by default, DHCP is offering addresses on newly created networks.

The second interface is created and bridged into the physical device eth0. This mechanism uses the macvtap Kernel driver and therefore does not require an existing bridge device. This configuration assumes that DHCP and DNS services are being provided by the public network. This public interface should be reachable by anyone with access to the public network.

Private Network Options

Note: These options are not applicable to public network interfaces.

There is a way to pass specific options for Libvirt provider when using config.vm.network to configure new network interface. Each parameter name starts with libvirt__ string. Here is a list of those options:

• :libvirt__network_name - Name of Libvirt network to connect to. By default, network 'default' is used.
• :libvirt__netmask - Used only together with :ip option. Default is '255.255.255.0'.
• :libvirt__network_address - Used only when :type is set to dhcp. Only /24 subnet is supported. Default is 172.28.128.0.
• :libvirt__host_ip - Address to use for the host (not guest). Default is first possible address (after network address).
• :libvirt__domain_name - DNS domain of the DHCP server. Used only when creating new network.
• :libvirt__dhcp_enabled - If DHCP will offer addresses, or not. Used only when creating new network. Default is true.
• :libvirt__dhcp_start - First address given out via DHCP. Default is third address in range (after network name and gateway).
• :libvirt__dhcp_stop - Last address given out via DHCP. Default is last possible address in range (before broadcast address).
• :libvirt__dhcp_bootp_file - The file to be used for the boot image. Used only when dhcp is enabled.
• :libvirt__dhcp_bootp_server - The server that runs the DHCP server. Used only when dhcp is enabled.By default is the same host that runs the DHCP server.
• :libvirt__adapter - Number specifiyng sequence number of interface.
• :libvirt__forward_mode - Specify one of veryisolated, none, open, nat or route options. This option is used only when creating new network. Mode none will create isolated network without NATing or routing outside. You will want to use NATed forwarding typically to reach networks outside of hypervisor. Routed forwarding is typically useful to reach other networks within hypervisor. veryisolated described here. By default, option nat is used.
• :libvirt__forward_device - Name of interface/device, where network should be forwarded (NATed or routed). Used only when creating new network. By default, all physical interfaces are used.
• :libvirt__tunnel_type - Set to 'udp' if using UDP unicast tunnel mode (libvirt v1.2.20 or higher). Set this to either "server" or "client" for tcp tunneling. Set this to 'mcast' if using multicast tunneling. This configuration type uses tunnels to generate point to point connections between Guests. Useful for Switch VMs like Cumulus Linux. No virtual switch setting like libvirt__network_name applies with tunnel interfaces and will be ignored if configured.
• :libvirt__tunnel_ip - Sets the source IP of the Libvirt tunnel interface. By default this is 127.0.0.1 for TCP and UDP tunnels and 239.255.1.1 for Multicast tunnels. It populates the address field in the <source address="XXX"> of the interface xml configuration.
• :libvirt__tunnel_port - Sets the source port the tcp/udp/mcast tunnel with use. This port information is placed in the <source port=XXX/> section of interface xml configuration.
• :libvirt__tunnel_local_port - Sets the local port used by the udp tunnel interface type. It populates the port field in the <local port=XXX"> section of the interface xml configuration. (This feature only works in Libvirt 1.2.20 and higher)
• :libvirt__tunnel_local_ip - Sets the local IP used by the udp tunnel interface type. It populates the ip entry of the <local address=XXX"> section of the interface xml configuration. (This feature only works in Libvirt 1.2.20 and higher)
• :libvirt__guest_ipv6 - Enable or disable guest-to-guest IPv6 communication. See here, and here for for more information. Note: takes either 'yes' or 'no' for value
• :libvirt__ipv6_address - Define ipv6 address, require also prefix.
• :libvirt__ipv6_prefix - Define ipv6 prefix. generate string <ip family="ipv6" address="address" prefix="prefix" >
• :libvirt__iface_name - Define a name for the private network interface. With this feature one can simulate physical link failures
• :mac - MAC address for the interface. Note: specify this in lowercase since Vagrant network scripts assume it will be!
• :libvirt__mtu - MTU size for the Libvirt network, if not defined, the created network will use the Libvirt default (1500). VMs still need to set the MTU accordingly.
• :model_type - parameter specifies the model of the network adapter when you create a domain value by default virtio KVM believe possible values, see the documentation for Libvirt
• :libvirt__driver_name - Define which network driver to use. More info
• :libvirt__driver_queues - Define a number of queues to be used for network interface. Set equal to numer of vCPUs for best performance. More info
• :autostart - Automatic startup of network by the Libvirt daemon. If not specified the default is 'false'.
• :bus - The bus of the PCI device. Both :bus and :slot have to be defined.
• :slot - The slot of the PCI device. Both :bus and :slot have to be defined.
• :libvirt__always_destroy - Allow domains that use but did not create a network to destroy it when the domain is destroyed (default: true). Set to false to only allow the domain that created the network to destroy it.

When the option :libvirt__dhcp_enabled is to to 'false' it shouldn't matter whether the virtual network contains a DHCP server or not and vagrant-libvirt should not fail on it. The only situation where vagrant-libvirt should fail is when DHCP is requested but isn't configured on a matching already existing virtual network.

Public Network Options

• :dev - Physical device that the public interface should use. Default is 'eth0'.
• :mode - The mode in which the public interface should operate in. Supported modes are available from the libvirt documentation. Default mode is 'bridge'.
• :type - is type of interface.(<interface type="#{@type}">)
• :mac - MAC address for the interface.
• :network_name - Name of Libvirt network to connect to.
• :portgroup - Name of Libvirt portgroup to connect to.
• :ovs - Support to connect to an Open vSwitch bridge device. Default is 'false'.
• :ovs_interfaceid - Add Open vSwitch 'interfaceid' parameter.
• :trust_guest_rx_filters - Support trustGuestRxFilters attribute. Details are listed here. Default is 'false'.

Management Network

vagrant-libvirt uses a private network to perform some management operations on VMs. All VMs will have an interface connected to this network and an IP address dynamically assigned by Libvirt unless you set :mgmt_attach to 'false'. This is in addition to any networks you configure. The name and address used by this network are configurable at the provider level.

• management_network_name - Name of Libvirt network to which all VMs will be connected. If not specified the default is 'vagrant-libvirt'.
• management_network_address - Address of network to which all VMs will be connected. Must include the address and subnet mask. If not specified the default is '192.168.121.0/24'.
• management_network_mode - Network mode for the Libvirt management network. Specify one of veryisolated, none, open, nat or route options. Further documented under Private Networks
• management_network_guest_ipv6 - Enable or disable guest-to-guest IPv6 communication. See here, and here for for more information.
• management_network_autostart - Automatic startup of mgmt network, if not specified the default is 'false'.
• management_network_pci_bus - The bus of the PCI device.
• management_network_pci_slot - The slot of the PCI device.
• management_network_mac - MAC address of management network interface.
• management_network_domain - Domain name assigned to the management network.
• management_network_mtu - MTU size of management network. If not specified, the Libvirt default (1500) will be used.

You may wonder how vagrant-libvirt knows the IP address a VM received. Libvirt doesn't provide a standard way to find out the IP address of a running domain. But we do know the MAC address of the virtual machine's interface on the management network. Libvirt is closely connected with dnsmasq, which acts as a DHCP server. dnsmasq writes lease information in the /var/lib/libvirt/dnsmasq directory. Vagrant-libvirt looks for the MAC address in this file and extracts the corresponding IP address.

It is also possible to use the Qemu Agent to extract the management interface configuration from the booted virtual machine. This is helpful in libvirt environments where no local dnsmasq is used for automatic address assigment, but external dhcp services via bridged libvirt networks.

Prerequisite is to enable the qemu agent channel via (Libvirt communication channels) and the virtual machine image must have the agent pre-installed before deploy. The agent will start automatically if it detects an attached channel during boot.

• qemu_use_agent - false by default, if set to true, attempt to extract configured ip address via qemu agent.

To use the management network interface with an external dhcp service you need to setup a bridged host network manually and define it via management_network_name in your Vagrantfile.

Additional Disks

You can create and attach additional disks to a VM via libvirt.storage :file. It has a number of options:

• path - Location of the disk image. If unspecified, a path is automtically chosen in the same storage pool as the VMs primary disk.
• device - Name of the device node the disk image will have in the VM, e.g. vdb. If unspecified, the next available device is chosen.
• size - Size of the disk image. If unspecified, defaults to 10G.
• type - Type of disk image to create. Defaults to qcow2.
• bus - Type of bus to connect device to. Defaults to virtio.
• allow_existing - Set to true if you want to allow the VM to use a pre-existing disk. If the disk doesn't exist it will be created. Disks with this option set to true need to be removed manually.
• shareable - Set to true if you want to simulate shared SAN storage.
• serial - Serial number of the disk device.
• wwn - WWN number of the disk device.

The following disk performance options can also be configured (see the libvirt documentation for possible values or here for a fuller explanation). In all cases, the options use the hypervisor default if not specified, or if set to nil.

• cache - Cache mode to use. Value may be default, none, writeback, writethrough, directsync or unsafe.
• io - Controls specific policies on I/O. Value may be threads or native.
• copy_on_read - Controls whether to copy read backing file into the image file. Value may be on or off.
• discard - Controls whether discard requests (also known as "trim" or "unmap") are ignored or passed to the filesystem. Value may be unmap or ignore. Note: for discard to work, you will likely also need to set :bus => 'scsi'
• detect_zeroes - Controls whether to detect zero write requests. Value may be off, on or unmap.

The following example creates two additional disks.

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.storage :file, :size => '20G'
    libvirt.storage :file, :size => '40G', :bus => 'scsi', :type => 'raw', :discard => 'unmap', :detect_zeroes => 'on'
  end
end

For shared SAN storage to work the following example can be used:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.storage :file, :size => '20G', :path => 'my_shared_disk.img', :allow_existing => true, :shareable => true, :type => 'raw'
  end
end

Reload behavior

On vagrant reload the following additional disk attributes are updated in defined domain:

• bus - Updated. Uses device as a search marker. It is not required to define device, but it's recommended. If device is defined then the order of addtitional disk definition becomes irrelevant.

CDROMs

You can attach up to four CDROMs to a VM via libvirt.storage :file, :device => :cdrom. Available options are:

• path - The path to the iso to be used for the CDROM drive.
• dev - The device to use (hda, hdb, hdc, or hdd). This will be automatically determined if unspecified.
• bus - The bus to use for the CDROM drive. Defaults to ide

The following example creates three CDROM drives in the VM:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.storage :file, :device => :cdrom, :path => '/path/to/iso1.iso'
    libvirt.storage :file, :device => :cdrom, :path => '/path/to/iso2.iso'
    libvirt.storage :file, :device => :cdrom, :path => '/path/to/iso3.iso'
  end
end

Input

You can specify multiple inputs to the VM via libvirt.input. Available options are listed below. Note that both options are required:

• type - The type of the input
• bus - The bus of the input

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # this is the default
    # libvirt.input :type => "mouse", :bus => "ps2"

    # very useful when having mouse issues when viewing VM via VNC
    libvirt.input :type => "tablet", :bus => "usb"
  end
end

PCI device passthrough

You can specify multiple PCI devices to passthrough to the VM via libvirt.pci. Available options are listed below. Note that all options are required, except domain, which defaults to 0x0000:

• domain - The domain of the PCI device
• bus - The bus of the PCI device
• slot - The slot of the PCI device
• function - The function of the PCI device

You can extract that information from output of lspci command. First characters of each line are in format [<domain>]:[<bus>]:[<slot>].[<func>]. For example:

$ lspci| grep NVIDIA
0000:03:00.0 VGA compatible controller: NVIDIA Corporation GK110B [GeForce GTX TITAN Black] (rev a1)

In that case domain is 0x0000, bus is 0x03, slot is 0x00 and function is 0x0.

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.pci :domain => '0x0000', :bus => '0x06', :slot => '0x12', :function => '0x5'

    # Add another one if it is neccessary
    libvirt.pci :domain => '0x0000', :bus => '0x03', :slot => '0x00', :function => '0x0'
  end
end

Note! Above options affect configuration only at domain creation. It won't change VM behaviour on vagrant reload after domain was created.

Don't forget to set kvm_hidden option to true especially if you are passthroughing NVIDIA GPUs. Otherwise GPU is visible from VM but cannot be operated.

Using USB Devices

There are several ways to pass a USB device through to a running instance:

• Use libvirt.usb to attach a USB device at boot, with the device ID specified in the Vagrantfile
• Use a client (such as virt-viewer or virt-manager) to attach the device at runtime via USB redirectors
• Use virsh attach-device once the VM is running (however, this is outside the scope of this readme)

In all cases, if you wish to use a high-speed USB device, you will need to use libvirt.usb_controller to specify a USB2 or USB3 controller, as the default configuration only exposes a USB1.1 controller.

USB Controller Configuration

The USB controller can be configured using libvirt.usb_controller, with the following options:

• model - The USB controller device model to emulate. (mandatory)
• ports - The number of devices that can be connected to the controller.

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Set up a USB3 controller
    libvirt.usb_controller :model => "qemu-xhci"
  end
end

See the libvirt documentation for a list of valid models.

If any USB devices are passed through by setting libvirt.usb or libvirt.redirdev, a default controller will be added using the model qemu-xhci in the absence of a user specified one. This should help ensure more devices work out of the box as the default configured by libvirt is pii3-uhci, which appears to only work for USB 1 devices and does not work as expected when connected via a USB 2 controller, while the xhci stack should work for all versions of USB.

USB Device Passthrough

You can specify multiple USB devices to passthrough to the VM via libvirt.usb. The device can be specified by the following options:

• bus - The USB bus ID, e.g. "1"
• device - The USB device ID, e.g. "2"
• vendor - The USB devices vendor ID (VID), e.g. "0x1234"
• product - The USB devices product ID (PID), e.g. "0xabcd"

At least one of these has to be specified, and bus and device may only be used together.

The example values above match the device from the following output of lsusb:

Bus 001 Device 002: ID 1234:abcd Example device

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # pass through specific device based on identifying it
    libvirt.usbdev :vendor => '0x1234', :product => '0xabcd'
    # pass through a host device where multiple of the same vendor/product exist
    libvirt.usbdev :bus => '1', :device => '1'
  end
end

Additionally, the following options can be used:

• startupPolicy - Is passed through to Libvirt and controls if the device has to exist. Libvirt currently allows the following values: "mandatory", "requisite", "optional".

USB Redirector Devices

You can specify multiple redirect devices via libvirt.redirdev. There are two types, tcp and spicevmc supported, for forwarding USB-devices to the guest. Available options are listed below.

• type - The type of the USB redirector device. (tcp or spicevmc)
• host - The host where the device is attached to. (mandatory for type tcp)
• port - The port where the device is listening. (mandatory for type tcp)

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # add two devices using spicevmc channel
    (1..2).each do
      libvirt.redirdev :type => "spicevmc"
    end
    # add device, provided by localhost:4000
    libvirt.redirdev :type => "tcp", :host => "localhost", :port => "4000"
  end
end

Note that in order to enable USB redirection with Spice clients, you may need to also set libvirt.graphics_type = "spice"

Filter for USB Redirector Devices

You can define filter for redirected devices. These filters can be positiv or negative, by setting the mandatory option allow=yes or allow=no. All available options are listed below. Note the option allow is mandatory.

• class - The device class of the USB device. A list of device classes is available on Wikipedia.
• vendor - The vendor of the USB device.
• product - The product id of the USB device.
• version - The version of the USB device. Note that this is the version of bcdDevice
• allow - allow or disallow redirecting this device. (mandatory)

You can extract that information from output of lsusb command. Every line contains the information in format Bus [<bus>] Device [<device>]: ID [<vendor>:[<product>]. The version can be extracted from the detailed output of the device using lsusb -D /dev/usb/[<bus>]/[<device>]. For example:

# get bcdDevice from
$: lsusb
Bus 001 Device 009: ID 08e6:3437 Gemalto (was Gemplus) GemPC Twin SmartCard Reader

$: lsusb -D /dev/bus/usb/001/009 | grep bcdDevice
  bcdDevice            2.00

In this case, the USB device with class 0x0b, vendor 0x08e6, product 0x3437 and bcdDevice version 2.00 is allowed to be redirected to the guest. All other devices will be refused.

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.redirdev :type => "spicevmc"
    libvirt.redirfilter :class => "0x0b", :vendor => "0x08e6", :product => "0x3437", :version => "2.00", :allow => "yes"
    libvirt.redirfilter :allow => "no"
  end
end

Serial Console Devices

You can define settings to redirect output from the serial console of any VM brought up with libvirt to a file or other devices that are listening. See libvirt documentation.

Currently only redirecting to a file is supported.

• type - only value that has an effect is file, in the future support may be added for virtual console, pty, dev, pipe, tcp, udp, unix socket, spiceport & nmdm.
• source - options pertaining to how the connection attaches to the host, contains sub-settings dependent on type. source options for type file
  • path - file on host to connect to the serial port to record all output. May be created by qemu system user causing some permissions issues.

Vagrant.configure("2") do |config|
  config.vm.define :test do |test|
    test.vm.provider :libvirt do |domain|
      domain.serial :type => "file", :source => {:path => "/var/log/vm_consoles/test.log}
    end
  end
end

Random number generator passthrough

You can pass through /dev/random to your VM by configuring the domain like this:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Pass through /dev/random from the host to the VM
    libvirt.random :model => 'random'
  end
end

At the moment only the random backend is supported.

Watchdog device

A virtual hardware watchdog device can be added to the guest via the libvirt.watchdog element. The option model is mandatory and could have on of the following values.

• i6300esb - the recommended device, emulating a PCI Intel 6300ESB
• 'ib700` - emulating an ISA iBase IB700
• diag288 - emulating an S390 DIAG288 device

The optional action attribute describes what action to take when the watchdog expires. Valid values are specific to the underlying hypervisor. The default behavior is reset.

• reset - default, forcefully reset the guest
• shutdown - gracefully shutdown the guest (not recommended)
• poweroff - forcefully power off the guest
• pause - pause the guest
• none - do nothing
• dump - automatically dump the guest
• inject-nmi - inject a non-maskable interrupt into the guest

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Add Libvirt watchdog device model i6300esb
    libvirt.watchdog :model => 'i6300esb', :action => 'reset'
  end
end

Smartcard device

A virtual smartcard device can be supplied to the guest via the libvirt.smartcard element. The option mode is mandatory and currently only value passthrough is supported. The value spicevmc for option type is default value and can be supressed. On using type = tcp, the options source_mode, source_host and source_service are mandatory.

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Add smartcard device with type 'spicevmc'
    libvirt.smartcard :mode => 'passthrough', :type => 'spicevmc'
  end
end

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Add smartcard device with type 'tcp'
    domain.smartcard :mode => 'passthrough', :type => 'tcp', :source_mode => 'bind', :source_host => '127.0.0.1', :source_service => '2001'
  end
end

Hypervisor Features

Hypervisor features can be specified via libvirt.features as a list. The default options that are enabled are acpi, apic and pae. If you define libvirt.features you overwrite the defaults, so keep that in mind.

An example:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Specify the default hypervisor features
    libvirt.features = ['acpi', 'apic', 'pae' ]
  end
end

A different example for ARM boards:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Specify the default hypervisor features
    libvirt.features = ["apic", "gic version='2'" ]
  end
end

You can also specify a special set of features that help improve the behavior of guests running Microsoft Windows.

You can specify HyperV features via libvirt.hyperv_feature. Available options are listed below. Note that both options are required:

• name - The name of the feature Hypervisor feature (see Libvirt doc)
• state - The state for this feature which can be either on or off.

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Relax constraints on timers
    libvirt.hyperv_feature :name => 'relaxed', :state => 'on'
    # Enable virtual APIC
    libvirt.hyperv_feature :name => 'vapic', :state => 'on'
    # Enable spinlocks (requires retries to be specified)
    libvirt.hyperv_feature :name => 'spinlocks', :state => 'on', :retries => '8191'
  end
end

Clock

Clock offset can be specified via libvirt.clock_offset. (Default is utc)

Additionally timers can be specified via libvirt.clock_timer. Available options for timers are: name, track, tickpolicy, frequency, mode, present

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Set clock offset to localtime
    libvirt.clock_offset = 'localtime'
    # Timers ...
    libvirt.clock_timer :name => 'rtc', :tickpolicy => 'catchup'
    libvirt.clock_timer :name => 'pit', :tickpolicy => 'delay'
    libvirt.clock_timer :name => 'hpet', :present => 'no'
    libvirt.clock_timer :name => 'hypervclock', :present => 'yes'
  end
end

CPU features

You can specify CPU feature policies via libvirt.cpu_feature. Available options are listed below. Note that both options are required:

• name - The name of the feature for the chosen CPU (see Libvirt's cpu_map.xml)
• policy - The policy for this feature (one of force, require, optional, disable and forbid - see Libvirt documentation)

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # The feature will not be supported by virtual CPU.
    libvirt.cpu_feature :name => 'hypervisor', :policy => 'disable'
    # Guest creation will fail unless the feature is supported by host CPU.
    libvirt.cpu_feature :name => 'vmx', :policy => 'require'
    # The virtual CPU will claim the feature is supported regardless of it being supported by host CPU.
    libvirt.cpu_feature :name => 'pdpe1gb', :policy => 'force'
  end
end

Memory Backing

You can specify memoryBacking options via libvirt.memorybacking. Available options are shown below. Full documentation is available at the libvirt memoryBacking section.

NOTE: The hugepages <page> element is not yet supported

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.memorybacking :hugepages
    libvirt.memorybacking :nosharepages
    libvirt.memorybacking :locked
    libvirt.memorybacking :source, :type => 'file'
    libvirt.memorybacking :access, :mode => 'shared'
    libvirt.memorybacking :allocation, :mode => 'immediate'
  end
end

No box and PXE boot

There is support for PXE booting VMs with no disks as well as PXE booting VMs with blank disks. There are some limitations:

• Requires Vagrant 1.6.0 or newer
• No provisioning scripts are ran
• No network configuration is being applied to the VM
• No SSH connection can be made
• vagrant halt will only work cleanly if the VM handles ACPI shutdown signals

In short, VMs without a box can be created, halted and destroyed but all other functionality cannot be used.

An example for a PXE booted VM with no disks whatsoever:

Vagrant.configure("2") do |config|
  config.vm.define :pxeclient do |pxeclient|
    pxeclient.vm.provider :libvirt do |domain|
      domain.boot 'network'
    end
  end
end

And an example for a PXE booted VM with no box but a blank disk which will boot from this HD if the NICs fail to PXE boot:

Vagrant.configure("2") do |config|
  config.vm.define :pxeclient do |pxeclient|
    pxeclient.vm.provider :libvirt do |domain|
      domain.storage :file, :size => '100G', :type => 'qcow2'
      domain.boot 'network'
      domain.boot 'hd'
    end
  end
end

Example for vm with 2 networks and only 1 is bootable and has dhcp server in this subnet, for example foreman with dhcp server Name of network "foreman_managed" is key for define boot order

    config.vm.define :pxeclient do |pxeclient|
      pxeclient.vm.network :private_network,ip: '10.0.0.5',
            libvirt__network_name: "foreman_managed",
            libvirt__dhcp_enabled: false,
            libvirt__host_ip: '10.0.0.1'

       pxeclient.vm.provider :libvirt do |domain|
          domain.memory = 1000
          boot_network = {'network' => 'foreman_managed'}
          domain.storage :file, :size => '100G', :type => 'qcow2'
          domain.boot boot_network
          domain.boot 'hd'
        end
      end

An example VM that is PXE booted from the br1 device (which must already be configured in the host machine), and if that fails, is booted from the disk:

Vagrant.configure("2") do |config|
  config.vm.define :pxeclient do |pxeclient|
    pxeclient.vm.network :public_network,
      dev: 'br1',
      auto_config: false
    pxeclient.vm.provider :libvirt do |domain|
      boot_network = {'dev' => 'br1'}
      domain.storage :file, :size => '100G'
      domain.boot boot_network
      domain.boot 'hd'
    end
  end
end

SSH Access To VM

vagrant-libvirt supports vagrant's standard ssh settings.

Forwarded Ports

vagrant-libvirt supports Forwarded Ports via ssh port forwarding. Please note that due to a well known limitation only the TCP protocol is supported. For each forwarded_port directive you specify in your Vagrantfile, vagrant-libvirt will maintain an active ssh process for the lifetime of the VM. If your VM should happen to be rebooted, the SSH session will need to be restablished by halting the VM and bringing it back up.

vagrant-libvirt supports an additional forwarded_port option gateway_ports which defaults to false, but can be set to true if you want the forwarded port to be accessible from outside the Vagrant host. In this case you should also set the host_ip option to '*' since it defaults to 'localhost'.

You can also provide a custom adapter to forward from by 'adapter' option. Default is eth0.

Internally Accessible Port Forward

config.vm.network :forwarded_port, guest: 80, host: 2000

Externally Accessible Port Forward

config.vm.network :forwarded_port, guest: 80, host: 2000, host_ip: "0.0.0.0"

Forwarding the ssh-port

Vagrant-libvirt now supports forwarding the standard ssh-port on port 2222 from the localhost to allow for consistent provisioning steps/ports to be used when defining across multiple providers.

To enable, set the following:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Enable forwarding of forwarded_port with id 'ssh'.
    libvirt.forward_ssh_port = true
  end
end

Previously by default libvirt skipped the forwarding of the ssh-port because you can access the machine directly. In the future it is expected that this will be enabled by default once autocorrect support is added to handle port collisions for multi machine environments gracefully.

Synced Folders

Vagrant automatically syncs the project folder on the host to /vagrant in the guest. You can also configure additional synced folders.

SECURITY NOTE: for remote Libvirt, nfs synced folders requires a bridged public network interface and you must connect to Libvirt via ssh.

NFS

vagrant-libvirt supports NFS as default with bidirectional synced folders.

Example with NFS:

Vagrant.configure("2") do |config|
  config.vm.synced_folder "./", "/vagrant"
end

RSync

vagrant-libvirt supports rsync with unidirectional synced folders.

Example with rsync:

Vagrant.configure("2") do |config|
  config.vm.synced_folder "./", "/vagrant", type: "rsync"
end

9P

vagrant-libvirt supports VirtFS (9p or Plan 9) with bidirectional synced folders.

Difference between NFS and 9p is explained here.

For 9p shares, a mount: false option allows to define synced folders without mounting them at boot.

Example for accessmode: "squash" with 9p:

Vagrant.configure("2") do |config|
  config.vm.synced_folder "./", "/vagrant", type: "9p", disabled: false, accessmode: "squash", owner: "1000"
end

Example for accessmode: "mapped" with 9p:

Vagrant.configure("2") do |config|
  config.vm.synced_folder "./", "/vagrant", type: "9p", disabled: false, accessmode: "mapped", mount: false
end

Further documentation on using 9p can be found in kernel docs and in QEMU wiki.

Please do note that 9p depends on support in the guest and not all distros come with the 9p module by default.

Virtio-fs

vagrant-libvirt supports Virtio-fs with bidirectional synced folders.

For virtiofs shares, a mount: false option allows to define synced folders without mounting them at boot.

So far, passthrough is the only supported access mode and it requires running the virtiofsd daemon as root.

QEMU needs to allocate the backing memory for all the guest RAM as shared memory, e.g. Use file-backed memory by enable memory_backing_dir option in /etc/libvirt/qemu.conf:

memory_backing_dir = "/dev/shm"

Example for Libvirt >= 6.2.0 (e.g. Ubuntu 20.10 with Linux 5.8.0 + QEMU 5.0 + Libvirt 6.6.0, i.e. NUMA nodes required) with virtiofs:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.cpus = 2
    libvirt.numa_nodes = [{ :cpus => "0-1", :memory => 8192, :memAccess => "shared" }]
    libvirt.memorybacking :access, :mode => "shared"
  end
  config.vm.synced_folder "./", "/vagrant", type: "virtiofs"
end

Example for Libvirt >= 6.9.0 (e.g. Ubuntu 21.04 with Linux 5.11.0 + QEMU 5.2 + Libvirt 7.0.0, or Ubuntu 20.04 + PPA enabled) with virtiofs:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.cpus = 2
    libvirt.memory = 8192
    libvirt.memorybacking :access, :mode => "shared"
  end
  config.vm.synced_folder "./", "/vagrant", type: "virtiofs"
end

Further documentation on using virtiofs can be found in official HowTo and in Libvirt KB.

Please do note that virtiofs depends on:

• Host: Linux >= 5.4, QEMU >= 4.2 and Libvirt >= 6.2 (e.g. Ubuntu 20.10)
• Guest: Linux >= 5.4 (e.g. Ubuntu 20.04)

QEMU Session Support

vagrant-libvirt supports using QEMU user sessions to maintain Vagrant VMs. As the session connection does not have root access to the system features which require root will not work. Access to networks created by the system QEMU connection can be granted by using the QEMU bridge helper. The bridge helper is enabled by default on some distros but may need to be enabled/installed on others.

There must be a virbr network defined in the QEMU system session. The libvirt default network which comes by default, the vagrant vagrant-libvirt network which is generated if you run a Vagrantfile using the System session, or a manually defined network can be used. These networks can be set to autostart with sudo virsh net-autostart <net-name>, which'll mean no further root access is required even after reboots.

The QEMU bridge helper is configured via /etc/qemu/bridge.conf. This file must include the virbr you wish to use (e.g. virbr0, virbr1, etc). You can find this out via sudo virsh net-dumpxml <net-name>.

allow virbr0

An example configuration of a machine using the QEMU session connection:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    # Use QEMU session instead of system connection
    libvirt.qemu_use_session = true
    # URI of QEMU session connection, default is as below
    libvirt.uri = 'qemu:///session'
    # URI of QEMU system connection, use to obtain IP address for management, default is below
    libvirt.system_uri = 'qemu:///system'
    # Path to store Libvirt images for the virtual machine, default is as ~/.local/share/libvirt/images
    libvirt.storage_pool_path = '/home/user/.local/share/libvirt/images'
    # Management network device, default is below
    libvirt.management_network_device = 'virbr0'
  end

  # Public network configuration using existing network device
  # Note: Private networks do not work with QEMU session enabled as root access is required to create new network devices
  config.vm.network :public_network, :dev => "virbr1",
      :mode => "bridge",
      :type => "bridge"
end

Customized Graphics

vagrant-libvirt supports customizing the display and video settings of the managed guest. This is probably most useful for VNC-type displays with multiple guests. It lets you specify the exact port for each guest to use deterministically.

Here is an example of using custom display options:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.graphics_port = 5901
    libvirt.graphics_ip = '0.0.0.0'
    libvirt.video_type = 'qxl'
  end
end

TPM Devices

Modern versions of Libvirt support connecting to TPM devices on the host system. This allows you to enable Trusted Boot Extensions, among other features, on your guest VMs.

To passthrough a hardware TPM, you will generally only need to modify the tpm_path variable in your guest configuration. However, advanced usage, such as the application of a Software TPM, may require modifying the tpm_model, tpm_type and tpm_version variables.

The TPM options will only be used if you specify a TPM path or version. Declarations of any TPM options without specifying a path or version will result in those options being ignored.

Here is an example of using the TPM options:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.tpm_model = 'tpm-tis'
    libvirt.tpm_type = 'passthrough'
    libvirt.tpm_path = '/dev/tpm0'
  end
end

It's also possible for Libvirt to start an emulated TPM device on the host. Requires swtpm and swtpm-tools

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.tpm_model = "tpm-crb"
    libvirt.tpm_type = "emulator"
    libvirt.tpm_version = "2.0"
  end
end

Memory balloon

The configuration of the memory balloon device can be overridden. By default, libvirt will automatically attach a memory balloon; this behavior is preserved by not configuring any memballoon-related options. The memory balloon can be explicitly disabled by setting memballoon_enabled to false. Setting memballoon_enabled to true will allow additional configuration of memballoon-related options.

Here is an example of using the memballoon options:

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.memballoon_enabled = true
    libvirt.memballoon_model = 'virtio'
    libvirt.memballoon_pci_bus = '0x00'
    libvirt.memballoon_pci_slot = '0x0f'
  end
end

Libvirt communication channels

For certain functionality to be available within a guest, a private communication channel must be established with the host. Two notable examples of this are the QEMU guest agent, and the Spice/QXL graphics type.

Below is a simple example which exposes a virtio serial channel to the guest. Note: in a multi-VM environment, the channel would be created for all VMs.

vagrant.configure(2) do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.channel :type => 'unix', :target_name => 'org.qemu.guest_agent.0', :target_type => 'virtio'
  end
end

Below is the syntax for creating a spicevmc channel for use by a qxl graphics card.

vagrant.configure(2) do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.channel :type => 'spicevmc', :target_name => 'com.redhat.spice.0', :target_type => 'virtio'
  end
end

These settings can be specified on a per-VM basis, however the per-guest settings will OVERRIDE any global 'config' setting. In the following example, we create 3 VMs with the following configuration:

• master: No channel settings specified, so we default to the provider setting of a single virtio guest agent channel.
• node1: Override the channel setting, setting both the guest agent channel, and a spicevmc channel
• node2: Override the channel setting, setting both the guest agent channel, and a 'guestfwd' channel. TCP traffic sent by the guest to the given IP address and port is forwarded to the host socket /tmp/foo. Note: this device must be unique for each VM.

For example:

Vagrant.configure(2) do |config|
  config.vm.box = "fedora/32-cloud-base"
  config.vm.provider :libvirt do |libvirt|
    libvirt.channel :type => 'unix', :target_name => 'org.qemu.guest_agent.0', :target_type => 'virtio'
  end

  config.vm.define "master" do |master|
    master.vm.provider :libvirt do |domain|
        domain.memory = 1024
    end
  end
  config.vm.define "node1" do |node1|
    node1.vm.provider :libvirt do |domain|
      domain.channel :type => 'unix', :target_name => 'org.qemu.guest_agent.0', :target_type => 'virtio'
      domain.channel :type => 'spicevmc', :target_name => 'com.redhat.spice.0', :target_type => 'virtio'
    end
  end
  config.vm.define "node2" do |node2|
    node2.vm.provider :libvirt do |domain|
      domain.channel :type => 'unix', :target_name => 'org.qemu.guest_agent.0', :target_type => 'virtio'
      domain.channel :type => 'unix', :target_type => 'guestfwd', :target_address => '192.0.2.42', :target_port => '4242',
                     :source_path => '/tmp/foo'
    end
  end
end

Custom command line arguments and environment variables

You can also specify multiple qemuargs arguments or qemuenv environment variables for qemu-system

• value - Value

Vagrant.configure("2") do |config|
  config.vm.provider :libvirt do |libvirt|
    libvirt.qemuargs :value => "-device"
    libvirt.qemuargs :value => "intel-iommu"
    libvirt.qemuenv QEMU_AUDIO_DRV: 'pa'
    libvirt.qemuenv QEMU_AUDIO_TIMER_PERIOD: '150'
    libvirt.qemuenv QEMU_PA_SAMPLES: '1024', QEMU_PA_SERVER: '/run/user/1000/pulse/native'
  end
end

Box Formats

Version 1

This is the original format that most boxes currently use.

You can view an example box in the example_box/directory. That directory also contains instructions on how to build a box.

The box is a tarball containing:

• qcow2 image file named box.img
• metadata.json file describing box image (provider, virtual_size, format)
• Vagrantfile that does default settings for the provider-specific configuration for this provider

Version 2 (Experimental)

Due to the limitation of only being able to handle a single disk with the version 1 format, a new format was added to support boxes that need to specify multiple disks. This is still currently experimental and as such support for packaging has yet to be added. There is a script in the tools folder (tools/create_box_with_two_disks.sh) that should provide a guideline on how to create such a box for those that wish to experiment and provide early feedback.

At it's most basic, it expects an array of disks to allow a specific order to be presented. Disks will be attached in this order and as such assume device names base on this within the VM. The 'path' attribute is required, and is expected to be relative to the base of the box. This should allow placing the disk images within a nested directory within the box if it useful for those with a larger number of disks. The name allows overriding the target volume name that will be used in the libvirt storage pool. Note that vagrant-libvirt will still prefix the volume name with #{box_name}_vagrant_box_image_#{box_version}_ to avoid accidental clashes with other boxes.

Format and virtual size need no longer be specified as they are now retrieved directly from the provided image using qemu-img info ....

Example format:

{
  'disks': [
      {
          'path': 'disk1.img'
      },
      {
          'path': 'disk2.img',
          'name': 'secondary_disk'
      },
      {
          'path': 'disk3.img'
      }
  ],
  'provider': 'libvirt'
}

Create Box

If creating a box from a modified vagrant-libvirt machine, ensure that you have set the config.ssh.insert_key = false in the original Vagrantfile as otherwise Vagrant will replace the default connection key-pair that is required on first boot with one specific to the machine and prevent the default key from working on the exported result.

Vagrant.configure("2") do |config|
  # this setting is only recommended if planning to export the
  # resulting machine
  config.ssh.insert_key = false

  config.vm.define :test_vm do |test_vm|
    test_vm.vm.box = "fedora/32-cloud-base"
  end
end

To create a vagrant-libvirt box from a qcow2 image, run create_box.sh (located in the tools directory):

$ create_box.sh ubuntu14.qcow2

You can also create a box by using Packer. Packer templates for use with vagrant-libvirt are available at https://github.com/jakobadam/packer-qemu-templates. After cloning that project you can build a vagrant-libvirt box by running:

$ cd packer-qemu-templates
$ packer build ubuntu-14.04-server-amd64-vagrant.json

Package Box from VM

vagrant-libvirt has native support for vagrant package via libguestfs virt-sysprep. virt-sysprep operations can be customized via the VAGRANT_LIBVIRT_VIRT_SYSPREP_OPERATIONS environment variable; see the upstream documentation for further details especially on default sysprep operations enabled for your system.

Options to the virt-sysprep command call can be passed via VAGRANT_LIBVIRT_VIRT_SYSPREP_OPTIONS environment variable.

$ export VAGRANT_LIBVIRT_VIRT_SYSPREP_OPTIONS="--delete /etc/hostname"
$ vagrant package

For example, on Chef bento VMs that require SSH hostkeys already set (e.g. bento/debian-7) as well as leave existing LVM UUIDs untouched (e.g. bento/ubuntu-18.04), these can be packaged into vagrant-libvirt boxes like so:

$ export VAGRANT_LIBVIRT_VIRT_SYSPREP_OPERATIONS="defaults,-ssh-userdir,-ssh-hostkeys,-lvm-uuids"
$ vagrant package

Troubleshooting VMs

The first step for troubleshooting a VM image that appears to not boot correctly, or hangs waiting to get an IP, is to check it with a VNC viewer. A key thing to remember is that if the VM doesn't get an IP, then vagrant can't communicate with it to configure anything, so a problem at this stage is likely to come from the VM, but we'll outline the tools and common problems to help you troubleshoot that.

By default, when you create a new VM, a vnc server will listen on 127.0.0.1 on port TCP5900. If you connect with a vnc viewer you can see the boot process. If your VM isn't listening on 5900 by default, you can use virsh dumpxml to find out which port it's listening on, or can configure it with graphics_port and graphics_ip (see 'Domain Specific Options' above).

Note: Connecting with the console (virsh console) requires additional config, so some VMs may not show anything on the console at all, instead displaying it in the VNC console. The issue with the text console is that you also need to build the image used to tell the kernel to output to the console during boot, and typically most do not have this built in.

Problems we've seen in the past include:

• Forgetting to remove /etc/udev/rules.d/70-persistent-net.rules before packaging the VM
• VMs expecting a specific disk device to be connected

If you're still confused, check the Github Issues for this repo for anything that looks similar to your problem.

Github Issue #1032 contains some historical troubleshooting for VMs that appeared to hang.

Did you hit a problem that you'd like to note here to save time in the future? Please do!

Development

To work on the vagrant-libvirt plugin, clone this repository out, and use Bundler to get the dependencies:

$ git clone https://github.com/vagrant-libvirt/vagrant-libvirt.git
$ cd vagrant-libvirt
$ bundle install

Once you have the dependencies, verify the unit tests pass with rspec:

$ export VAGRANT_HOME=$(mktemp -d)
$ bundle exec rspec --fail-fast --color --format documentation

If those pass, you're ready to start developing the plugin.

Setting VAGRANT_HOME is to avoid issues with conflicting with other plugins/gems or data already present under ~/.vagrant.d.

Additionally if you wish to test against a specific version of vagrant you can control the version using the following before running the tests:

$ export VAGRANT_VERSION=v2.2.14

Note rvm is used by the maintainers to help provide an environment to test against multiple ruby versions that align with the ones used by vagrant for their embedded ruby depending on the release. You can see what version is used by looking at the current unit tests workflow.

You can test the plugin without installing it into your Vagrant environment by just creating a Vagrantfile in the top level of this directory (it is gitignored) that uses it. You can add the following line to your Vagrantfile while in development to ensure vagrant checks that the plugin is installed:

Vagrant.configure("2") do |config|
  config.vagrant.plugins = "vagrant-libvirt"
end

Or add the following to the top of the file to ensure that any required plugins are installed globally:

REQUIRED_PLUGINS = %w(vagrant-libvirt)
exit unless REQUIRED_PLUGINS.all? do |plugin|
  Vagrant.has_plugin?(plugin) || (
    puts "The #{plugin} plugin is required. Please install it with:"
    puts "$ vagrant plugin install #{plugin}"
    false
  )
end

Now you can use bundler to execute Vagrant:

$ bundle exec vagrant up --provider=libvirt

IMPORTANT NOTE: bundle is crucial. You need to use bundled Vagrant.

Contributing

1. Fork it
2. Create your feature branch (git checkout -b my-new-feature)
3. Commit your changes (git commit -am 'Add some feature')
4. Push to the branch (git push origin my-new-feature)
5. Create new Pull Request