Re: [ipwave] [Int-dir] side note RFC 4291 2nd par sec. 2.1 LL on loopback

[Alexandre Petrescu <[e-mail]>]

Le 16/04/2019 à 16:30, 神明達哉 a écrit :
> At Tue, 16 Apr 2019 12:52:37 +0200,
> Alexandre Petrescu <[e-mail] 
> <mailto:[e-mail]>> wrote:
> 
>  > >  > That RFC4291 section 2.1 says:
>  > >  > >    All interfaces are required to have at least one Link-Local 
> unicast
>  > >  > >    address (see Section 2.8 for additional required addresses).
>  > >  >
>  > >  > Sidenote: I think the loopback interface does not have a link-local
>  > >  > address.  Probably it is not all interfaces that must have at least
>  > > one ll(
>  > >
>  > > Indeed, there's some subtle point here.  RFC4007 somewhat tries to
>  > > clarifies it:
> 
>  > Well,  RFC4007 predates RFC4291, so it is little meaningful to say it
>  > clarifies it.
> 
> Could we be stop nitpicking?  I didn't say RFC4007 clarified RFC4291;
> I meant it clarified the requirement of having a link-local address on
> all interfaces, which I thought was clear from the context.  For that
> matter, the above exact same text has existed since RFC2373.  Even
> RFC1884 (the very first address architecture RFC) essentially says the
> same thing in Section 2.7.
> 
>  > Maybe we can say that RFC4291 must be updated given all these issues.
> 
>  > > BTW, whether a loopback interface has an fe80 address is actually
>  > > implementation dependent.  BSDs usually assign ::1 on a loopback
>  > > interface while also generating an fe80 address on it:
>  > >
>  > > % ifconfig lo0
>  > > lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> metric 0 mtu 16384
>  > >      options=600003<RXCSUM,TXCSUM,RXCSUM_IPV6,TXCSUM_IPV6>
>  > >      inet6 ::1 prefixlen 128
>  > >      inet6 fe80::1%lo0 prefixlen 64 scopeid 0x2
>  >
>  > Ah?  It self-forms that fe80 address without a sysadmin requesting it?
> 
> Yes.
> 
>  > In other OSs like linux, windows, there is no fe80::1 unless the
>  > sysadmin adds it.
>  >
>  > In that sense, this is an additional reason (together with the
>  > RFC4291-violating fe80:1::1 potentially breaking in BSD) to think that
>  > RFC4291 and co. are mostly for BSDs.
> 
> You're free to think so.  But that doesn't change the fact that
> fe80:1::1 violates the RFC, so you'll need to update it if you want it
> to be allowed.
> 
> Note also that MacOS is derived from BSD in case you don't
> remember/know it.  And, in fact it also generates both "fe80::1" and
> "::1" on the "lo0" interface.
> 
> Anyway, I don't think this discussion affects your draft?  If you want
> to continue it, 6man would be a better place, if only to reduce the
> noise in the original thread.

I would like to know from you whether the current state of the draft is 
ok with you at this time?

Alex

> 
> --
> JINMEI, Tatuya