Re: [v6ops] Our IPv6-only home network and future experiments

["Soni \"They/Them\" L." <[e-mail]>]

On 2024-04-13 08:15, Brian Candler wrote:
> On 13/04/2024 09:57, Tore Anderson wrote:
>> Having IPv6 nodes with a single global IPv6 address is not «bad 
>> practice», it is very common, even in networks using SLAAC.
>>
>> As for the CLAT as described in RFC6877, it was simply easier to 
>> implement in a bolt-on fashion by requiring a dedicated IPv6 address, 
>> since that meant the implementer did not have to care about things 
>> like sockets, TCP/UDP ports and such. Your proposed implementation 
>> goes directly into those areas though, so it does appear to me that 
>> you could do better than the original. I would therefore suggest you 
>> reconsider this decision. 
>
> I just want to reiterate that point.
>
> When composing my original reply, I starting thinking that a libc CLAT 
> implementation in an unprivileged process would have difficulty 
> getting an additional IPv6 address assigned - until I realised this 
> was completely unnecessary.  You can, and should, just let the kernel 
> do its default source address selection.
>
> Imagine that you had a dozen processes all with their own userland 
> CLAT: you wouldn't want, or need, a dozen additional IPv6 addresses 
> created. But neither is there any point in creating a single 
> additional IPv6 address for all the CLATs to share.
>

Oh. That makes sense. So the extra IPv6(s) would only be necessary for 
providing services to IPv4-only and CLATless dual-stack hosts, we're 
guessing?

As much as we are a SLAAC maximalist, making CLAT-in-libc use a separate 
IPv6 does seem more trouble than it's worth...