[pve-devel] applied: [PATCH ifupdown2] d/patches: add patch for transparent handling of interface altnames

[Christoph Heiss]

On Fri Jul 11, 2025 at 8:09 PM CEST, Thomas Lamprecht wrote:
> On Fri, 11 Jul 2025 18:25:12 +0200, Christoph Heiss wrote:
>> tl;dr: add transparent altname support by caching links under their
>> primary name as well as any potential altname, and translating all
>> interface names from the config into the primary interface name.
>>
>> The structure makes it also as easy as possible to translate altnames in
>> other corners of ifupdown2, if other cases should come up.
>>
>> [...]
>
> The amount of code changes are fewer than I  expected, and most of it reads
> pretty straight forward. While I'd expect a few bugs for edge cases to pop up,

Yep, I don't completely rule some edge cases either, but I done a sanity
check for other places where interface names are directly used - they
are normally retrieved by looking up the interface index, which always
returns the primary interface name.

So overall it's the manually specified interface names that are
problematic, thus the actual fallout/bugs should hopefully be pretty
constrained.

> the underlying design is IMO sound and a bit of testing of common stuff did not
> surface any issues.
>
> So, applied, thanks!
>
> Once the (release) dust settles it might be still worth to send a PR upstream,
> even though upstream went a bit silent, it still might help others and one
> never knows what wonders might happen ;-)

For future reference:
https://github.com/CumulusNetworks/ifupdown2/pull/330

>
> W.r.t the ports-condone-regex: I think keeping the current behavior is fine
> for now, these attributes are not that wide spread and if they really just
> affect reloads it won't trigger immediately after the first reboot after the
> major upgrade, which is when the interface renaming is normally happening, so
> admins get a chance to update these regexes then.
> If we want to check them too the only sensible solution (without thinking this
> through _that_ much though) would be to check if any of the interface names
> match the regex, and if, ignore the respective interface.

Alright, then I'd really just keep the current behavior for now, waiting
for a real usecase to pop up. At least I have never seen any
-condone-regex in any real-world /e/n/i, it really seems like a niche
and/or rather Cumulus-specific feature.