Bug#1092176: release-notes: systemd v257 has new rules for network interface names

Michael Biebl <[email protected]> Thu, 31 Jul 2025 17:18:34 +0200
Newsgroups gmane.linux.debian.devel.documentation
Message-ID <fdda0ed4-b59b-4b55-b378-53771d65ac17__11541.9303577923$1753976357$gmane$org@debian.org>
Am 30.07.25 um 06:03 schrieb Justin B Rye:
> Michael Biebl wrote:
>> aipark <[email protected]> wrote:
>>> I updated a staging bookworm VM with VirtIO NIC to trixie and
>>> temporarily lost network access because enp6s18 became ens18,
> [...]
>>
>> So, for a remote system where you don't want the naming scheme to change,
>> would it make sense to add a recommendation to pin it to the systemd version
>> the system was originally installed with?
>>
>> Say for a bookworm system, you add "net.naming_scheme=v252" to the kernel
>> command line.
>> Then you can dist-upgrade and boot into your new trixie system (including
>> the new kernel).
>>
>> And at a later point, you can decide to bump net.naming_scheme to v257 at a
>> more opportune time.
> 
> I could understand it if you were suggesting that any server accessed
> via the net should have a .link file.  But this pinning approach seems
> a strange strategy for the maintainers to be recommending.
> 
> Users who know that a specific upgrade is about to rename their
> network interfaces don't have a problem; they can just arrange to
> switch over to the new name (for instance by defining both in
> /etc/network/interfaces).

If you know (beforehand) that your network interface name is going to 
change (and your are affected by it), then obviously you can make the 
changes in advance. But how would you know?

> So this recommendation only really makes any sense if the idea is that
> *everybody* who cares about having reliable networking should keep
> this bodge in place from now on.

I indeed think that we should discuss if d-i should create such a 
configuration by default to pin the network naming scheme to the version 
the system was installed with. This would actually ensure that names 
remain stable (and predictable) and not change due to systemd/kernel 
updates.


> And if this is the official canonical way of getting predictable names
> out of the predictable names mechanism, why does systemd even *have*
> new versions of the naming scheme?


I assume the versioning scheme was explicitly added to avoid situations 
like the above.

Michael
OpenPGP_signature.asc (application/pgp-signature, 840 B)
-----BEGIN PGP SIGNATURE-----
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=BkiB
-----END PGP SIGNATURE-----