Secure live migration with QEMU-native TLS

Context

The encryption offered by nova’s libvirt.live_migration_tunnelled does
not secure all the different migration streams of a nova instance,
namely: guest RAM, device state, and disks (via NBD) when using
non-shared storage. Further, the “tunnelling via libvirtd” has inherent
limitations: (a) it cannot handle live migration of disks in a
non-shared storage setup (a.k.a. “block migration”); and (b) has a huge
performance overhead and latency, because it burns more CPU and memory
bandwidth due to increased number of data copies on both source and
destination hosts.

To solve this existing limitation, QEMU and libvirt have gained
(refer below <Prerequisites> for version details)
support for “native TLS”, i.e. TLS built into QEMU. This will secure all
data transports, including disks that are not on shared storage, without
incurring the limitations of the “tunnelled via libvirtd” transport.

To take advantage of the “native TLS” support in QEMU and libvirt,
nova has introduced new configuration attribute libvirt.live_migration_with_native_tls.

Prerequisites

1. Version requirement: This feature needs at least libvirt 4.4.0
   and QEMU 2.11.

2. A pre-configured TLS environment—i.e. CA, server, and client
   certificates, their file permissions, et al—must be “correctly”
   configured (typically by an installer tool) on all relevant compute
   nodes. To simplify your PKI (Public Key Infrastructure) setup, use
   deployment tools that take care of handling all the certificate
   lifecycle management. For example, refer to the “TLS
   everywhere” guide from the TripleO project.

3. Password-less SSH setup for all relevant compute nodes.

4. On all relevant compute nodes, ensure the TLS-related config
   attributes in /etc/libvirt/qemu.conf are in place:

   default_tls_x509_cert_dir = "/etc/pki/qemu"
   default_tls_x509_verify = 1

   If it is not already configured, modify
   /etc/sysconfig/libvirtd on both (ComputeNode1 &
   ComputeNode2) to listen for TCP/IP connections:

   LIBVIRTD_ARGS="--listen"

   Then, restart the libvirt daemon (also on both nodes):

   $ systemctl restart libvirtd

   Refer to the “Related information”
   section on a note about the other TLS-related configuration attributes
   in /etc/libvirt/qemu.conf.

Validating
your TLS environment on compute nodes

Assuming you have two compute hosts (ComputeNode1, and
ComputeNode2) run the virt-pki-validate tool (comes with the
libvirt-client package on your Linux distribution) on both
the nodes to ensure all the necessary PKI files are configured are
configured:

[ComputeNode1]$ virt-pki-validate
Found /usr/bin/certtool
Found CA certificate /etc/pki/CA/cacert.pem for TLS Migration Test
Found client certificate /etc/pki/libvirt/clientcert.pem for ComputeNode1
Found client private key /etc/pki/libvirt/private/clientkey.pem
Found server certificate /etc/pki/libvirt/servercert.pem for ComputeNode1
Found server private key /etc/pki/libvirt/private/serverkey.pem
Make sure /etc/sysconfig/libvirtd is setup to listen to
TCP/IP connections and restart the libvirtd service

[ComputeNode2]$ virt-pki-validate
Found /usr/bin/certtool
Found CA certificate /etc/pki/CA/cacert.pem for TLS Migration Test
Found client certificate /etc/pki/libvirt/clientcert.pem for ComputeNode2
Found client private key /etc/pki/libvirt/private/clientkey.pem
Found server certificate /etc/pki/libvirt/servercert.pem for ComputeNode2
Found server private key /etc/pki/libvirt/private/serverkey.pem
Make sure /etc/sysconfig/libvirtd is setup to listen to
TCP/IP connections and restart the libvirtd service

Other TLS
environment related checks on compute nodes

IMPORTANT: Ensure that the permissions of
certificate files and keys in /etc/pki/qemu/* directory on
both source and destination compute nodes to be the following
0640 with root:qemu as the group/user. For
example, on a Fedora-based system:

$ ls -lasrtZ /etc/pki/qemu
total 32
0 drwxr-xr-x. 10 root root system_u:object_r:cert_t:s0      110 Dec 10 10:39 ..
4 -rw-r-----.  1 root qemu unconfined_u:object_r:cert_t:s0 1464 Dec 10 11:08 ca-cert.pem
4 -rw-r-----.  1 root qemu unconfined_u:object_r:cert_t:s0 1558 Dec 10 11:08 server-cert.pem
4 -rw-r-----.  1 root qemu unconfined_u:object_r:cert_t:s0 1619 Dec 10 11:09 client-cert.pem
8 -rw-r-----.  1 root qemu unconfined_u:object_r:cert_t:s0 8180 Dec 10 11:09 client-key.pem
8 -rw-r-----.  1 root qemu unconfined_u:object_r:cert_t:s0 8177 Dec 11 05:35 server-key.pem
0 drwxr-xr-x.  2 root root unconfined_u:object_r:cert_t:s0  146 Dec 11 06:01 .

Performing the migration

1. On all relevant compute nodes, enable the libvirt.live_migration_with_native_tls
   configuration attribute and set the libvirt.live_migration_scheme
   configuration attribute to tls:

   [libvirt]
   live_migration_with_native_tls = true
   live_migration_scheme = tls

   Note

   Setting both libvirt.live_migration_with_native_tls
   and libvirt.live_migration_tunnelled at the
   same time is invalid (and disallowed).

   Note

   Not setting libvirt.live_migration_scheme to
   tls will result in libvirt using the unencrypted TCP
   connection without displaying any error or a warning in the logs.

   And restart the nova-compute service:

   $ systemctl restart openstack-nova-compute

2. Now that all TLS-related configuration is in place, migrate
   guests (with or without shared storage) from ComputeNode1
   to ComputeNode2. Refer to the live-migration-usage document
   on details about live migration.

Related information

• If you have the relevant libvirt and QEMU versions (mentioned in
  the “Prerequisites” section earlier), then
  using the libvirt.live_migration_with_native_tls
  is strongly recommended over the more limited libvirt.live_migration_tunnelled
  option, which is intended to be deprecated in future.

• There are in total nine TLS-related config options in
  /etc/libvirt/qemu.conf:

  default_tls_x509_cert_dir
  default_tls_x509_verify
  nbd_tls
  nbd_tls_x509_cert_dir
  migrate_tls_x509_cert_dir
  
  vnc_tls_x509_cert_dir
  spice_tls_x509_cert_dir
  vxhs_tls_x509_cert_dir
  chardev_tls_x509_cert_dir

  If you set both default_tls_x509_cert_dir and
  default_tls_x509_verify parameters for all certificates,
  there is no need to specify any of the other *_tls* config
  options.

  The intention (of libvirt) is that you can just use the
  default_tls_x509_* config attributes so that you don’t need
  to set any other *_tls* parameters, _unless
  you need different certificates for some services. The rationale for
  that is that some services (e.g. migration / NBD) are only exposed to
  internal infrastructure; while some sevices (VNC, Spice) might be
  exposed publically, so might need different certificates. For OpenStack
  this does not matter, though, we will stick with the defaults.

• If they are not already open, ensure you open up these TCP ports
  on your firewall: 16514 (where the authenticated and
  encrypted TCP/IP socket will be listening on) and
  49152-49215 (for regular migration) on all relevant compute
  nodes. (Otherwise you get
  error: internal error: unable to execute QEMU command 'drive-mirror': Failed to connect socket: No route to host).