Re: appropriate length of fe80:: prefix and new IP-over-foo drafts

[Alexandre Petrescu <[e-mail]>]

Le 31/01/2019 à 13:31, Fernando Gont a écrit :
> On 31/1/19 09:25, Alexandre Petrescu wrote:
>>
>>
>> Le 31/01/2019 à 12:41, Fernando Gont a écrit :
>>> On 31/1/19 08:33, Alexandre Petrescu wrote:
>>>>
>>>>
>>>> Le 31/01/2019 à 12:17, Fernando Gont a écrit :
>>>>> On 31/1/19 07:24, Alexandre Petrescu wrote:
>>>>>>
>>>>>>
>>>>>> Le 30/01/2019 à 23:14, Brian E Carpenter a écrit :
>>>>>>> Section 5.3 "Creation of Link-Local Addresses" of RFC4862 refers to
>>>>>>> an interface identifier of N bits and an fe80::0 prefix "of
>>>>>>> appropriate length", which according to the addressing architecture
>>>>>>> is 128-N. But N is undefined by RFC4862 and cannot be derived from an
>>>>>>> RA because all this happens as soon as the interface is enabled.
>>>>>>> Therefore N must be predefined.
>>>>>>>
>>>>>>> As pioneered by RFC2464**, all ipv6-over-foo documents must therefore
>>>>>>> specify N for their link type. So far, everybody has specified 64, as
>>>>>>> far as I know. Is there a reason to do otherwise for the two drafts
>>>>>>> mentioned below?
>>>>>>
>>>>>> One of the two drafts involves a Network ID of length 24.
>>>>>>
>>>>>> The question is why not use 24 for InterfaceID?  Just because Ethernet
>>>>>> had it that way?
>>>>>
>>>>> You can certainly do that with OpenBSD. ;-)
>>>>
>>>> In linux too I can add (and we actually do) link-local addresses with
>>>> prefix length 32.
>>>
>>> They will not work if you try to employ them with BSDs. BSDs use bits in
>>> the range /16 /64 to store information such as the interface index.
>>> So... don't use prefixes shorter than /64 for link-locals, or you'll be
>>> asking for trouble.
>>
>> Ah?  according to which spec do the BSD implementers put bits in the
>> range 16-64?
> 
> They don't send those bits on the wire. But they do use those bits as
> described. -- According to no spec. If the link-local prefix is assumed
> to be /64, then you're free to use those bits *internally* as you wish.
> That's what BSDs do. Jinmei is quite likely in a better position ot
> comment on this.

I would like to kindly ask Jinmei to make sure to keep that internal 
such that to not forbid the end user to ifconfig add fe80:1::1/63

If possible.

Alex

> 
> Thanks,
>