[GRLUG] Testing a host's IPv6

Adam Tauno Williams awilliam at whitemice.org
Sat Jan 15 18:07:02 EST 2011


On Sat, 2011-01-15 at 15:33 -0500, Michael Mol wrote: 
> On Sat, Jan 15, 2011 at 3:16 PM, Adam Tauno Williams
> <awilliam at whitemice.org> wrote:
> > On Sat, 2011-01-15 at 15:03 -0500, Michael Mol wrote:
> >> eth0's IPv6 line says that its address is fe80::a800:ff:fedc:e7/64,
> >> and that the scope (where that address is valid/meaningful) is "Link".
> >> That means that it's only valid on the local Ethernet segment. (Well,
> >> technically, that means it's valid on the local 802 network segment,
> >> which Ethernet and wifi both fall under, along with things like token
> >> ring and FDDI.)
> > Actually, the link-local address is valid on any IPv6 link; even
> > point-to-point links like PPP or HDLC.
> I suppose it may be technically valid, but link-local IP addresses are
> only guaranteed to be unique in Link scope. If you have multiple
> non-bridged 802 networks (say, wired and wifi, and they're not
> directly connected), and a machine has a NIC on each network, the same
> link-local address may exist on both addresses, but be associated with
> different machines.
> So, really, you shouldn't route link-local scope packets. 

Correct, and there is no need to route such packets.  This is one reason
IPv6 specifically supports scopes;  The host knows that XYZ network
exists at the other end of a link-local-only addresses link - the host
routes the packets to that remote address out through the appropriate
interface's link-local address.  The receiver realizing it can
successfully forward that packet does so - nobody cares about the
addressing of the link-local-only link [which for WANs is a huge
improvement over IPv4].  The same link-local addresses can exist on
multiple circuits within the internetwork and that doesn't impact
routing.

This is much like if you could just re-use the same 192.168.x.x/252
network over and over again for your PtP links [which wouldn't work;
because IPv4 is IPv4 and not IPv6]

Setting up a central router in a hub-n-spoke network for IPv6 is *WAY*
less effort than doing the same for IPv4.  With IPv4 you either have to
make many /252 networks or use something odd like Cisco's unnumbered
interfaces [which brings some issues of its own].  For IPv6 you
configure the interfaces *you* care about [the local LAN interface] and
leave the rest for the routing protocol to figure out - which it will.
Because IPv6 scoping rocks.

> There's a
> site-local scope which seems similar in purpose to RFC 1918 addresses,
> but that's deprecated with no specified replacement I'm aware of.

Correct, site-local is deprecated.



-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.



More information about the grlug mailing list