Proposed F19 Feature: DualstackNetworking - proper dual stack IPv4 and IPv6 networking

[Pavel Simerda]

----- Original Message -----
> From: "Miloslav Trmač" <mitr at volny.cz>
> 
> On Wed, Jan 2, 2013 at 7:59 PM, Jaroslav Reznik <jreznik at redhat.com>
> wrote:
> > = Features/DualstackNetworking =
> > https://fedoraproject.org/wiki/Features/DualstackNetworking
> 
> (Sending to the list instead of using the wiki talk page to hopefully
> gather more information from networking experts.  The recent-ish
> bugzilla activity in this space does suggest that this is a good
> topic to revisit in detail, thanks for taking the effort on.)

Thanks!

> 1. So what is the _precise_ scope of the ''proposed feature''?  For
> example, does "glibc: name resolution must work properly" mean that
> you have some specific proposals to change glibc?

Yes.

I identified a bunch of problems in glibc and commented on previously identified problems:

http://sourceware.org/bugzilla/buglist.cgi?quicksearch=getaddrinfo&list_id=7854

Now I'm writing and submitting patches. But it will take time.

> (https://fedoraproject.org/wiki/Networking/NameResolution/ADDRCONFIG
> suggests changing glibc as one of the options.  Have you decided
> which
> option you propose?  In general, glibc maintainers' signoff or at
> least significant discussion would probably be required for any glibc
> change of this kind.  FWIW, historically it was recommended to have
> AI_ADDRCONFIG enabled, see
> http://www.akkadia.org/drepper/userapi-ipv6.html .)

It is now pretty clear that we need to do #3 and #4 (in this order) and then start recommending AI_ADDRCONFIG for connect(), sendto() and other stuff again. But I found more fundamental problems in getaddrinfo() while playing with the code and while reading IPv6-related bug reports.

Ulrich's documentation is unfortunately not useful as it was written without practical IPv6 skills and testing. I recommend to ignore it in favor of the resources on Fedora Wiki.

> 2. We already have a guideline requiring IPv6
> (https://fedoraproject.org/wiki/Packaging:Guidelines?rd=PackagingGuidelines#Networking_Support
> ); to the extent that this feature would propose specific ways to use
> the API, the packaging guideline should point to documentation that
> describes the correct way.

Good point. Specific recommedations that are evolving will be part of this feature.

> It would be great to get a wide review of the proposed approach before any extensive effort to
> modify many
> packages starts.

Actually, it's easier than that.

Most important packages either use getaddrinfo() or are moving towards it. Many of them (if not most) set AI_ADDRCONFIG. As we are now going to fix AI_ADDRCONFIG to make it actually useful and not harmful, that would mean only packages that are buggy already would need fixing.

Thank you for your feedback,

Pavel