Re: connecting to 6bone problem
On Tue, 1 Aug 2000, Bernd Eckenfels wrote:
> > We *MUST* now establish some routing into the tunnel - like with ethernet
> > the only available destination is fe80::192.87.106.15
>
> No you dont need to. Like with ethernet you can specify a device route and
> dont need to specify a gateway. This is especially tru for point-to-point
> devices/interfaces like ppp, ippp or tunnels. Since there is only one
> receiver you do not need to address and hardware address of a gateway. In
Well, to be clear, you do have to establish routing or nothing will go
into that tunnel! But the way I showed using fe80:: addresess is probably
more confusing than it is worth.. The better way is just:
ip -6 route add default dev surfnet
I later discovered I had the above wrong when I was trying it and
mistakenly assumed it did not work for some odd reason. I was glad to
later learn that is not the case :>
Careful though, not all tunnels are point to point, gre, sit and ipip all
have NBMA modes (wildcard remote address) for which the above is not
valid.
> in the tunneld packages. Nowhere, even on "linklevel" there is a reference
> to fe80::192.87.106.15.
It is only used internally to the kernel. When the device comes up it
establishes a 'network' route for fe80::/16 on the tunnel device, like all
ipv6 devices. The only two addresses the sit tunnel responds to are local
and remote mapped IPs.
> Using the "ipv4-like address ::192.87.106.15" is only used for auto tunnels
> which dont set up a remote ipv4 address. Usually the sit0 devce in linux
I covered this entire situation in the second half of my posting.. 'auto
tunnels' are properly called NBMA tunnels.
> kernel is working that way. You set up a route like
> ::/96
Done by the kernel when you bring sit0 up.. sit0 knows to treat 0::x.x.x.x
specially, otherwise you'd have to use constructs like:
ip -6 route add default via 0::ffff:1.1.1.1 dev sit0 onlink
AFAIK the ::/96 is just a shortcut!
ipip and gre NBMA devices have to use the above sort of thing.
> (which is by definition a zero prefix to mark ipv4 addresses)
Another interesting point is that 0::1.1.1.1 has the leading zeros to mark
an ipv4 address, but it is not a v4 mapped address (0::ffff:1.1.1.1).
AFAIK the 0::1.1.1.1 thing is a magic linux tunnel thing. When I first
read over this I mistakenly assumed it was a V4 mapped address and got
confused quickly :>
> and address them toi an imaginary ipv4 adress with the tunnel endpoint
> x.x.x.x. Even with this setup you dont need to specify any gateway address
> for the tunnel or interface.
The gateway bit was more for example than real usage - it has a certain
nice conceptual likeness to how ppp operates. After ppp comes up the
remote IP/network is reachable over the link - pppd establishes this route
AFAIK. After adding the gateway route you have the same situation with
the tunnel. The kernel optimizes the extra route away of course, even in
the case with ppp you don't need it.
Jason
Reply to: