[IPv6]Re: [v6ops] New draft: "The IPv6 Loopback Address Prefix"

[Brian E Carpenter <[e-mail]>]

Michael,

On 27-Nov-25 10:47, Michael Sweet wrote:
> Brian,
> 
>> On Nov 26, 2025, at 3:25 PM, Brian E Carpenter <[e-mail]> wrote:
>> ...
>> (d) as far as my experiments went, mDNS on Linux has messed up support for link-local addresses, which all need a zoneid because of (b). Which is probably the real reason that ::/96 would work on Linux virtual interfaces but fe80::/10 wouldn't.
> 
> I haven't noticed any particular issues with my Linux laptop or VMs using Avahi, although for my home office the Wi-Fi and Ethernet are treated as a single LAN (Ubiquity DM Pro + a bunch of APs and switches) so it is possible I am just "lucky".  

The problems that I saw are described in the README at https://github.com/becarpenter/misc/tree/main/zelect

The worst issue was that if mDNS resolves test.local to a link-local IPv6 address (and no IPv4 address), which requires a non-zero Zone Index, then 'ping test.local' worked OK, but both 'ping6 test.local' and 'ping -6 test.local' failed with 'invalid argument'. Therefore, one code path was picking up the interface index correctly, and the other two weren't.

Similarly, Firefox worked as expected on http://test.local, but Chrome failed.

That indicated that indeed Avahi was doing the right thing. I didn't dig into the ping/ping6 source code, however.

There was also a gap in 'avahi-publish-address' which I noticed.

Since my test setup was ancient, I don't know if the problems still exist.

> Similarly, my Apple gear (which uses mDNSResponder) doesn't seem to have any issues.  Both return IPv6 results with the scope zone index set to the required interface.

I can't test Apple. Windows 10 seemed to be doing the right thing.

    Brian