Libvirt - Nova Virtualisation Driver

Overview

Libvirt is the most commonly used virtualisation driver in OpenStack. It uses libvirt, backed by QEMU and when available, KVM. Libvirt is executed in the nova_libvirt container, or as a daemon running on the host.

Hardware Virtualisation

Two values are supported for nova_compute_virt_type with libvirt - kvm and qemu, with kvm being the default.

For optimal performance, kvm is preferable, since many aspects of virtualisation can be offloaded to hardware. If it is not possible to enable hardware virtualisation (e.g. Virtualisation Technology (VT) BIOS configuration on Intel systems), qemu may be used to provide less performant software-emulated virtualisation.

SASL Authentication

The default configuration of Kolla Ansible is to run libvirt over TCP, authenticated with SASL. This should not be considered as providing a secure, encrypted channel, since the username/password SASL mechanisms available for TCP are no longer considered cryptographically secure. However, it does at least provide some authentication for the libvirt API. For a more secure encrypted channel, use libvirt TLS.

SASL is enabled according to the libvirt_enable_sasl flag, which defaults to true.

The username is configured via libvirt_sasl_authname, and defaults to nova. The password is configured via libvirt_sasl_password, and is generated with other passwords using kolla-mergepwd and kolla-genpwd and stored in passwords.yml.

The list of enabled authentication mechanisms is configured via libvirt_sasl_mech_list, and defaults to ["SCRAM-SHA-256"] if libvirt TLS is enabled, or ["DIGEST-MD5"] otherwise.

Host vs containerised libvirt

By default, Kolla Ansible deploys libvirt in a nova_libvirt container. In some cases it may be preferable to run libvirt as a daemon on the compute hosts instead.

Kolla Ansible does not currently support deploying and configuring libvirt as a host daemon. However, since the Yoga release, if a libvirt daemon has already been set up, then Kolla Ansible may be configured to use it. This may be achieved by setting enable_nova_libvirt_container to false.

When the firewall driver is set to openvswitch, libvirt will plug VMs directly into the integration bridge, br-int. To do this it uses the ovs-vsctl utility. The search path for this binary is controlled by the $PATH environment variable (as seen by the libvirt process). There are a few options to ensure that this binary can be found:

  • Set openvswitch_ovs_vsctl_wrapper_enabled to True. This will install a wrapper script to the path: /usr/bin/ovs-vsctl that will execute ovs-vsctl in the context of the openvswitch_vswitchd container. This option is useful if you do not have openvswitch installed on the host. It also has the advantage that the ovs-vsctl utility will match the version of the server.

  • Install openvswitch on the hypervisor. Kolla mounts /run/openvswitch from the host into the openvswitch_vswitchd container. This means that socket is in the location ovs-vsctl expects with its default options.

Migration from container to host

The kolla-ansible nova-libvirt-cleanup command may be used to clean up the nova_libvirt container and related items on hosts, once it has been disabled. This should be run after the compute service has been disabled, and all active VMs have been migrated away from the host.

By default, the command will fail if there are any VMs running on the host. If you are sure that it is safe to clean up the nova_libvirt container with running VMs, setting nova_libvirt_cleanup_running_vms_fatal to false will allow the command to proceed.

The nova_libvirt container has several associated Docker volumes: libvirtd, nova_libvirt_qemu and nova_libvirt_secrets. By default, these volumes are not cleaned up. If you are sure that the data in these volumes can be safely removed, setting nova_libvirt_cleanup_remove_volumes to true will cause the Docker volumes to be removed.

A future extension could support migration of existing VMs, but this is currently out of scope.

Libvirt TLS

The default configuration of Kolla Ansible is to run libvirt over TCP, with SASL authentication. As long as one takes steps to protect who can access the network this works well. However, in a less trusted environment one may want to use encryption when accessing the libvirt API. To do this we can enable TLS for libvirt and make nova use it. Mutual TLS is configured, providing authentication of clients via certificates. SASL authentication provides a further level of security.

Using libvirt TLS

Libvirt TLS can be enabled in Kolla Ansible by setting the following option in /etc/kolla/globals.yml:

libvirt_tls: "yes"

Creation of the TLS certificates is currently out-of-scope for Kolla Ansible. You will need to either use an existing Internal CA or you will need to generate your own offline CA. For the TLS communication to work correctly you will have to supply Kolla Ansible the following pieces of information:

  • cacert.pem

    • This is the CA’s public certificate that all of the client and server certificates are signed with. Libvirt and nova-compute will need this so they can verify that all the certificates being used were signed by the CA and should be trusted.

  • serverkey.pem (not used when using a host libvirt daemon)

    • This is the private key for the server, and is no different than the private key of a TLS certificate. It should be carefully protected, just like the private key of a TLS certificate.

  • servercert.pem (not used when using a host libvirt daemon)

    • This is the public certificate for the server. Libvirt will present this certificate to any connection made to the TLS port. This is no different than the public certificate part of a standard TLS certificate/key bundle.

  • clientkey.pem

    • This is the client private key, which nova-compute/libvirt will use when it is connecting to libvirt. Think of this as an SSH private key and protect it in a similar manner.

  • clientcert.pem

    • This is the client certificate that nova-compute/libvirt will present when it is connecting to libvirt. Think of this as the public side of an SSH key.

Kolla Ansible will search for these files for each compute node in the following locations and order on the host where Kolla Ansible is executed:

  • /etc/kolla/config/nova/nova-libvirt/<hostname>/

  • /etc/kolla/config/nova/nova-libvirt/

In most cases you will want to have a unique set of server and client certificates and keys per hypervisor and with a common CA certificate. In this case you would place each of the server/client certificate and key PEM files under /etc/kolla/config/nova/nova-libvirt/<hostname>/ and the CA certificate under /etc/kolla/config/nova/nova-libvirt/.

However, it is possible to make use of wildcard server certificate and a single client certificate that is shared by all servers. This will allow you to generate a single client certificate and a single server certificate that is shared across every hypervisor. In this case you would store everything under /etc/kolla/config/nova/nova-libvirt/.

Externally managed certificates

One more option for deployers who already have automation to get TLS certs onto servers is to disable certificate management under /etc/kolla/globals.yaml:

libvirt_tls_manage_certs: "no"

With this option disabled Kolla Ansible will simply assume that certificates and keys are already installed in their correct locations. Deployers will be responsible for making sure that the TLS certificates/keys get placed in to the correct container configuration directories on the servers so that they can get copied into the nova-compute and nova-libvirt containers. With this option disabled you will also be responsible for restarting the nova-compute and nova-libvirt containers when the certs are updated, as kolla-ansible will not be able to tell when the files have changed.