Re: Next steps (Was Re: Initial version of a SuSE backend)

[Tom Parker <palfrey tevp net>]

David Zeuthen wrote:
> Ideally, I'd like to see a plug-in based solution (both back-ends and
> front-ends), much like the VPN stuff we're doing now, leaving
> NetworkManager proper as a core that controls these plug-ins and deals
> with core stuff such as dhcp, named etc. Yes, this means wired and
> wireless networking would be plug-ins as well. Things would probably
> need to be layered too. I would start by considering the UI we need and
> the use that to design the API interfaces.

Related thought I had a while back (but hadn't gotten around to having a better 
look at due to current time constriants), was that currently we have a fairly 
hard-coded approach to what link is "better". I'm fairly sure it's still set to 
"if we've got a wired link that supports carrier detect and it's alive, pick 
that over *anything* else".

I had some thoughts regarding a better classification scheme for links, which 
becomes even more important if we start supporting lots of things. In order of 
worst->best:

1) Link does not support minimum features we need (e.g. wireless scan/link detect)
2) Link supports features, but we can't see any connections (e.g. scan returns 0 
APs/link detect is down)
3) Link supports features, has a possible connection but is restricted. e.g. you 
need to login somewhere (VPN for example) before you've got full access, or 
maybe it's just a walled garden somewhere.
4) Link has basic features, has a possible connection, and appears unrestricted. 
*Then* we start to sort on connection speed...

Overriding by user UI can jump from 1->3 (and possibly 4, but probably not). 
Also, a wireless connection with multiple APs should be considered as a group of 
connections, one per detected/known AP. They have the same basic features, but 
some may be in group 3 and others in 4.

How do we detect restrictions? Well, there's a couple of easy ways that might 
work e.g. attempt to resolve something.invalid. RFC 2606 guarantees that 
*.invalid will always be non-lookable, which might work for walled garden 
"everything points to one server" DNS. Or, check a couple of known sane lookups. 
If both gnome.org and microsoft.com resolve to the same address, then either the 
apocalypse has come or you're in a walled garden. I guess there are a few other 
tests that people could think of...

This looks like a better choice scheme to me than the current one, and more 
flexible for new protocols.

Tom
--
palfrey tevp net - http://tevp.net
Illegitimus non carborundum