Re: IPv6 Address Architecture update question

[Bob Hinden <[e-mail]>]

Jeroen,

At 05:24 AM 01/31/2005, Jeroen Massar wrote:
>On Mon, 2005-01-31 at 04:48 -0800, Bob Hinden wrote:
> > Hi,
> >
> > I am working on an update to the IPv6 address architecture.  In doing this
> > I am working through the comments on the previous draft.  One comment made
> > was to remove Section 2.5.5 "IPv6 Addresses with Embedded IPv4 Addresses"
> > from the document.  This would include removing the special case in the
> > textual representation (section 2.2, 3.).
>
>In http://www.ietf.org/rfc/rfc2373.txt it is section 2.5.4.
>Or am I looking at the wrong document? :)

Right it's section 2.5.4 in RFC2373.  The section changed a little in the 
new draft.

><SNIP>
>
> > Since I expect this is widely implemented, please be sure to report any
> > problem that might occur if this is to be removed from the
> > specification.
>
>I think that at least all the BSD's and most Linuxes are using this.
>They allow binding on :: (IPv6 any) and also accept IPv4 connections on
>the same socket, which are then represented in netstat etc
>as ::ffff:1.2.3.4. Both OS's nowadays have a bind-ipv6-only flag or
>similar, but many applications rely on this behaviour in that they
>either bind to the IPv6 any or the IPv4 any address (0.0.0.0).
>
>Removing it, thus would mean that all these applications are broken and
>need to be updated, which actually is true, having that programs should
>use multiple sockets and use getaddrinfo() to figure out the correct
>sockets to bind on. This has some programming overhead though and as
>many people are lazy they did not do it.

I am not sure they would be broken.  They would be doing something in 
addition to what is in the new RFC.

In any case, thanks for the comments.

Bob




>The mixing of the stacks, thus having IPv6 also accept IPv4 connections,
>did have some negative side effects for the applications that did/do
>support getaddrinfo() type binding and bind to multiple sockets, as they
>would first bind to IPv6 any, after which they bind to IPv4 any which
>would fail with socket already in use, giving an error message that is
>not entirely true.
>
>IMO the section can be removed and programmers need to be learned the
>correct thing for which I always very inclined to point people to Eva's
>excellent document at http://gsyc.escet.urjc.es/~eva/IPv6-web/ipv6.html
>and/or draft-ietf-v6ops-application-transition-02.txt
>
>The various Operating Systems that support this should be thought the
>same tricks too of course...
>
> >   This includes would it break other documents that refer to
> > the IPv6 address architecture specification.
>
>At least RFC3484 has a short section (3.3) on it, though it mentions
>handling them as 'global scope', which is basically the default case,
>for address selection.
>
>Greets,
>  Jeroen
>


--------------------------------------------------------------------
IETF IPv6 working group mailing list
[e-mail]
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------