Add LXD server UUID file

[roosterfish]

This takes the required bits from the Dell PowerFlex PR in order to introduce a new UUID file which is unique for every LXD server.

The UUID can be retrieved via the daemons State struct using the ServerUUID field. This is comparable with the contents of /etc/machine-id or /var/lib/dbus/machine-id which can be decoded from hexadecimal to its 128 bit UUID representation. However using those files has dependencies on dbus and systemd. See https://wiki.debian.org/MachineId.

Having a unique UUID file per daemon also ensures that multiple instances of LXD on the same machine don't use the same UUID.

[simondeziel]

@roosterfish I don't have much context on this but was wondering why another UUID would be needed? We already have one:

$ lxc launch --vm ubuntu-daily:22.04 j
Creating j
Starting j

$ lxc config show j | grep uuid
  volatile.uuid: b6b8ed20-dbec-4c72-947a-2abbfa97f586
  volatile.uuid.generation: b6b8ed20-dbec-4c72-947a-2abbfa97f586

$ lxc exec j -- cat /sys/devices/virtual/dmi/id/product_uuid
b6b8ed20-dbec-4c72-947a-2abbfa97f586

$ lxc exec j -- cat /etc/machine-id
b6b8ed20dbec4c72947a2abbfa97f586

I'm pretty sure the /sys/devices/virtual/dmi/id/product_uuid file is populated by the kernel itself, so no dbus nor systemd involved there. Could this file be used instead?

[tomponline]

@simondeziel this came up as part of #12304 as each cluster member (not per instance) requires a persistent unique UUID for NVME.

The latter two examples were discussed as we consulted with https://0pointer.de/blog/projects/ids.html during the discussion.

DMI ID was not suitable because if hardware changes (such as motherboard) it would change.

The nearest suitable one was /var/lib/dbus/machine-id (which /etc/machine-id is derived from), but we observed this wasn't present on all systems (such as Alpine which didn't have dbus installed by default).

We decided to just generated a UUID and store it persistently as the most reliable way to get a per-host ID without any external dependencies.

[simondeziel]

@simondeziel this came up as part of #12304 as each cluster member (not per instance) requires a persistent unique UUID for NVME.

Yep, that was the root of my confusion, I was thinking of a per-instance UUID, sorry for the noise.