CentOS - Sep 2021 - how to display/create DUID?

On Sunday, September 19, 2021 6:53:45 PM CEST Kenneth Porter
wrote:
> --On Sunday, September 19, 2021 3:02 PM +0200 hw <hw at gc-24.de>
wrote:
> > None of this is working because the server isn't running a DHCPv6
server,
> > and there seems to be no file in /var/lib/NetworkManager that would
seem
> > to be helpful.
> > 
> > Isn't there a tool that creates the DUID and prints it?  This
can't be
> > too difficult ...
> 
> I found this thread that suggests that NetworkManager computes it every
> time unless it's manually overridden:
> 
>
<https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/135>
That might mean that I would need to extract the functions from the soruces of 
dhclient to make a program that prints DUID(s) for the machine you run it on. 
... But that isn't so easy.  Maybe I can find out how to create a DUID and 
write something in perl; it doesn't seem to be too complicated in the
source.

The comment is interesting:

/*
 * The "best" default DUID, since we cannot predict any information
 * about the system (such as whether or not the hardware addresses are
 * integrated into the motherboard or similar), is the "LLT", link
local
 * plus time, DUID. For real stateless "LL" is better.
 *
 * Once generated, this duid is stored into the state database, and
 * retained across restarts.
 *
 * For the time being, there is probably a different state database for
 * every daemon, so this winds up being a per-interface identifier...which
 * is not how it is intended.  Upcoming rearchitecting the client should
 * address this "one daemon model."
 */

I don't understand what the point of a DUID is which is /not/ a
per-interface
identifier.  When I assign addresses via DHCP, I don't want them to end up 
being assigned anywhere else than to the interface they need to be assigned 
to.  What is intended with these DUIDs?
> Based on that, nmtui-connect might do what you want.
That seems to only allow to bring up or down a connection but doesn't
display
DUIDs?

On Wednesday, September 29, 2021 10:22:27 PM CEST hw
wrote:
> On Sunday, September 19, 2021 6:53:45 PM CEST Kenneth Porter wrote:
> > --On Sunday, September 19, 2021 3:02 PM +0200 hw <hw at
gc-24.de> wrote:
> > > None of this is working because the server isn't running a
DHCPv6
> > > server,
> > > and there seems to be no file in /var/lib/NetworkManager that
would seem
> > > to be helpful.
> > > 
> > > Isn't there a tool that creates the DUID and prints it?  This
can't be
> > > too difficult ...
> > 
> > I found this thread that suggests that NetworkManager computes it
every
> > time unless it's manually overridden:
> > 
> >
<https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/issues/135
> > >
> 
> That might mean that I would need to extract the functions from the soruces
> of dhclient to make a program that prints DUID(s) for the machine you run
> it on. ... But that isn't so easy.  Maybe I can find out how to create
a
> DUID and write something in perl; it doesn't seem to be too complicated
in
> the source.
> 
> The comment is interesting:
> 
> /*
>  * The "best" default DUID, since we cannot predict any
information
>  * about the system (such as whether or not the hardware addresses are
>  * integrated into the motherboard or similar), is the "LLT",
link local
>  * plus time, DUID. For real stateless "LL" is better.
>  *
>  * Once generated, this duid is stored into the state database, and
>  * retained across restarts.
>  *
>  * For the time being, there is probably a different state database for
>  * every daemon, so this winds up being a per-interface identifier...which
>  * is not how it is intended.  Upcoming rearchitecting the client should
>  * address this "one daemon model."
>  */
> 
> I don't understand what the point of a DUID is which is /not/ a
> per-interface identifier.  When I assign addresses via DHCP, I don't
want
> them to end up being assigned anywhere else than to the interface they need
> to be assigned to.  What is intended with these DUIDs?
> 
Hm, ok, here's an explanation: https://www.juniper.net/documentation/en_US/
junose15.1/topics/concept/dhcp-unique-id-servers-clients-overview.html

IIUC, this means that I could create a DUID (of type 1) with an arbitrary time 
value (Or where would I get the "correct" one?)  and enter it into the
configuration of a static lease in OPNsense (or any other DHCPv6 server), and 
it would be totally pointless unless I manage to force the client beforehand 
to use this DUID to ask for an IP address and unless I can make sure that the 
client always uses this DUID for a particular interface --- which isn't 
intended because the DUIDs aren't supposed to identify particular
interfaces.

Is that my lack of understanding or are these DUIDs really a rather stupid 
idea?

And how are we actually supposed to set up static leases with DHCPv6?