[gentoo-dev] IPv6 support in network initscripts

[Peter Johanson <latexer@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 2173 bytes --]

hey all,

i've started doing a lot of work on IPv6 stuff in gentoo, to try to get
us up to snuff with regards to IPv6 support. I've been commiting a lot
more ebuilds for various things, and more are to come. BUT, an important
aspect needs fixing. 

init scripts.

The problem is we need better control of configuring both the IPv4 and
the IPv6 attributes of an interface. Specifically, we want IPv6 services
to start if IPv4 networking fails and vice versa.

Along with that, we also need some initscript control over establishing
tunnels, etc. That issue will be better tackled i think after we have a
solid general device configuration in place.

Solutions:
* Seperate net.ethX and net6.ethX scripts.
	Problems: When something fails in current net.ethX script,
	we do 'ifconfig ethX down'. If we do this with co-existing
	v4+v6, we bring down the other form of networking.

	Advantages: Lets us have a "provide net" and "provide net6".
	Other scripts can independantly depend on one or the other, or both,
	or have "need net \ use net6", etc.

	More easily allows for IPv6 only networks if in an IPv6 only
	environment. (example: LAN using 6to4 tools)

* Both IPv6 and IPv4 handled by an enhanced net.ethX script
	
	Problems: If v4 setup fails, does the script fail completely,
	or succeed? It can't succeed, because other things that might need
	v4 will subsequently fail. It can't die, because things that need
	the format that *didn't* fail will not be started, even though
	they could have been.

	Advantages: 
	* Less work to enable a new device. Only have to do 
	"rc-update add net.eth0 default" then configure /etc/conf.d/net

	* Marginally less time spent running scripts

* Other solutions?

I really prefer the first method, as it provides much finer grained
control. I've got a net6.ethX script mostly done, but i wanted to get
some feedback and opinions before fully implementing. The one major
sticking point is how to deal with the current "do something or die and
bring the interface down" situation. This behavior would obviously have
to change.

-pete


-- 
Peter Johanson
<latexer@gentoo.org>

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]