Re: IPv6 addressing APIs (Re: I-D Action: draft-fgont-6man-rfc4941bis-01.txt)

[Erik Kline <[e-mail]>]

On 29 March 2018 at 13:05, Brian E Carpenter
<[e-mail]> wrote:
> On 29/03/2018 13:20, Fernando Gont wrote:
>> On 03/27/2018 08:10 PM, Brian E Carpenter wrote:
>>>>     > FWIW, this has been the whole point of:
>>>>     > draft-gont-6man-address-usage-recommendations
>>>>
>>>> I think that we need to extend the IPv6 APIs so that applications
>>>> can either request stable addresses or demand a new privacy address.
>>>
>>> Yes. But unfortunately that's outside the scope of the IETF, since we gave
>>> away that API many years ago. I guess we could write a draft and find out
>>> how to get it into the POSIX process.
>>
>> I think there's a difference between an abstract API (that is certainly
>> within the scope of the IETF), and a concrete API (say, for different
>> OSes or languages), which are not.
>
> True. But in real life, we aren't going to revolutionize the basic design
> of the IPv6 socket API, so any new abstractions have to be mappable into
> the POSIX model.

True.

One thing that could have been done back in getaddrinfo days would
have been to recommend the existence of some async version.  That's
not necessarily easy in C, but nevertheless.

I started to collect some thoughts about a v6 addrs abstract API, but
only got as far as this half-baked document:

    https://github.com/ekline/draft-v6addrs-api/blob/master/draft.txt

I think it's probably possible to implement something as a library for
Linux, that would include a thread that listens on a netlink socket
for changes in addresses.

After I get out from under some more pressing work I'm hoping to back
to thinking about it more.