[pve-devel] [PATCH container] fix #3443: setup: clear /etc/machine-id in post-create hook

Oguz Bektas o.bektas at proxmox.com
Wed May 26 11:40:23 CEST 2021


thanks for checking

On Tue, May 25, 2021 at 03:45:53PM +0200, Thomas Lamprecht wrote:
> On 25.05.21 15:17, Oguz Bektas wrote:
> > this way when new containers are created the will have a unique
> > /etc/machine-id
> > 
> why the dbus then? systemd talks explicitly only about /etc/machine-id
> in the docs and also in their CT interface [0], the thematic is only touched
> there though.
> The dbus one is most often a symlink to /etc/machine-id, removing that can
> break things...

in our ubuntu templates it exists as a regular file (not symlinked).

so if that's not removed before first boot, it's copied to /etc/machine-id
from there when the container boots.

`man machine-id` tells me:

       When a machine is booted with systemd(1) the ID of the machine will be established. If systemd.machine_id= or
       --machine-id= options (see first section) are specified, this value will be used. Otherwise, the value in
       /etc/machine-id will be used. If this file is empty or missing, systemd will attempt to use the D-Bus machine
       ID from /var/lib/dbus/machine-id, the value of the kernel command line option container_uuid, the KVM DMI
       product_uuid (on KVM systems), and finally a randomly generated UUID.

so without removing the dbus file we don't get a unique machine-id on container
creation, since the templates seem to have a hardcoded id in the dbus path by
default. we can also remove that but then we will have to make sure to do that
for all the relevant templates

> [0] https://systemd.io/CONTAINER_INTERFACE/
> > note that post_create_hook doesn't run for cloned containers so that
> > will need to be handled separately
> > 
> If you read my post you also read that we must not remove the file in the
> clone case.

yes this hook doesn't run at clone so that's fine.

> We currently always generate a new random MAC-address for all netX devies of
> a CT on clone, that suggests that we always want to truncate in the clone case,
> to ensure that IPv6 SLAAC, among other things, can work OK.
> We could add a "unique" param to the clone call, but until now this was never
> requested to be configurable.

looking at the clone_vm api call i wasn't sure where to modify the file during

would it make sense to add this as a config option? we could set this to
"uninitialized" in the container config by default. the "unique" param can then
decide if the machine-id would be copied or truncated at clone.

i'm open to ideas

> [1]: https://forum.proxmox.com/threads/bug-machine-id-etc-machine-id-not-unique-in-lxc-containers.89708/post-392258
> > @@ -491,6 +503,7 @@ sub pre_start_hook {
> >  sub post_create_hook {
> >      my ($self, $conf, $root_password, $ssh_keys) = @_;
> >  
> > +    $self->clear_machine_id($conf);
> only relevant for systemd envs. so it should be only called then, if we must call this
> in such a general place.
> Either called in in the respective distro setup or check if any of "/lib/systemd/systemd"
> or "/usr/lib/systemd/system" is executable for a heuristic to find out if the CT is
> systemd managed.

okay that makes sense i'll add that

> >      $self->template_fixup($conf);
> >  
> >      $self->randomize_crontab($conf);
> > 
> this now depends on the patch which changed the private randomize_crontab helper
> to a plugin method as the change is visible in the context, but that isn't mentioned
> anywhere...

guess i'll also remove this part based on your other email

More information about the pve-devel mailing list