Missing configuration option in Secure Live Migration guide leads to unencrypted traffic
The guide to enable secure live migration with QEMU-native tls on nova compute nodes missed an important config option. Without this option a hard-coded part in nova is triggerd which sets the default route to TCP instead of TLS. This leads to an unecrypted migration of the ram without throwing any kind of Error.
Affected Services / Software
Nova / Victoria, Ussuri, Train, Stein (might also be affected: Rocky, Queens, Pike, Ocata)
In the OpenStack guide to setup secure live migration with QEMU-native tls there are a few configuration options given, which have to be applied to nova compute nodes. After following the instructions and setting up everything it seems to work as expected. But after checking that libvirt is able to use tls using tcpdump to listen on the port for tls while manually executing libvirt commands, the same check for live migration of an instance through openstack fails. Listening on the port for unencrypted tcp-traffic shows that OpenStack still uses the unencrypted TCP path instead of the TLS one for the migration.
The reason for this is a patch from Ocata which adds the calculation of the live-migration-uri in code: https://review.opendev.org/c/openstack/nova/+/410817/ The config parameter ``live_migration_uri`` was deprecated in favor of ``live_migration_scheme`` and the default set to tcp. This leads to the problem that if none of these two config options are set, libvirt will always use the default tcp connection. To enable QEMU-native TLS to be used in nova one of them has to be set so that a TLS connection can be established. Currently the guide does not show that this is necessary and there was no other documentation indicating that these config options are important for the usage of QEMU-native TLS.
As there is no documentation which recognizes this and it is hard to find this problem as the migration happens even without those config option set - not stating that it is still unencrypted, it might have been unrecognized in various deployments, which followed the guide.
For deployments using secure live migration with QEMU-native TLS:
- Check the config of all nova compute nodes. The libvirt section needs to have either ``live_migration_uri`` (deprecated) or ``live_migration_scheme`` configured.
- If neither of those config options are present, add ``live_migration_scheme = tls`` to enable the use of the tls connection.
The guide for secure live migration was updated to reflect the necessary configuration options and now has a note, which warns users that not setting all config options may lead into a seemingly working deployment, which still uses unencrypted traffic for the ram-migration.
Contacts / References
Author: Josephine Seifert, secustack GmbH
This OSSN : https://wiki.openstack.org/wiki/OSSN/OSSN-0089
Original LaunchPad Bug : https://bugs.launchpad.net/nova/+bug/1919357
Mailing List : [Security] tag on firstname.lastname@example.org
OpenStack Security Project : https://launchpad.net/~openstack-ossg