ssh -6 error(s)

[Felipe Alfaro Solana]

On Tue, 2004-01-13 at 00:25, Zach Wilkinson wrote:
> Just the default after modprode ipv6
> 
> [root at server root]# ifconfig -a
> eth0      Link encap:Ethernet  HWaddr 00:40:F4:57:87:63  
>           inet addr:192.168.1.100  Bcast:192.168.1.255  Mask:255.255.255.0
>           inet6 addr: fe80::240:f4ff:fe57:8763/64 Scope:Link
                                                          ^^^^
The problem you're seeing is the same I had when starting to use IPv6.
IPv6 supports stateless autoconfiguration and has some peculiarities.

For what I've seen, it seems you are using link-local addresses. The
problem here is that all link-local addresses have the same network
prefix: fe80/10 (although Linux seems to use fe80::/64), which makes
impossible to know how to route packets when not specifying an outgoing
interface. If you have two network interfaces, both have a network
prefix of fe80::/10, so when trying to send out a packet to another
link-local address, how do you know which interface the packet should go
out, if you have two routing entries with a prefix fe80::/10?

When you using ping6 with a link-local addresses, you must supply the -I
<interface> argument to ping6 to specify which outgoing interface to use
for sending out the ICMPv6 ECHO packets and resolve the ambiguity.

# ping6 -I eth0 fe80::240:f4ff:fe57:8763

Also, link-local addresses are mainly used for autoconfiguration, like
prefix network assignment, using DHCPv6, using Router Discovery or
Neighbour Discovery. They are not normally used to high-level
communications. If you want to communicate clients using TCP/IPv6,
you'll need to assign both your server and workstation both IPv6
addresses with a greater scope, like a Site-local or Global address.
Site-local and Global addresses do have specific network prefixes and
both create specific routing entries, so packets can be sent and
received without having to specify and outgoing interface.

Site-local addresses are similary to private IPv4 addresses as used
today in private LAN's, like 10.0.0.0/8, 172.16.0.0/12 or
192.168.0.0/16, are unique to a routing domain, but not unique on the
global Internet, that is, won't escape their own routing domain. Global
addresses are unique on the whole Internet and are routable, that is,
they can pass through router domains.

Site-local addresses have the fec0::/64 prefix, while global addresses
usually have prefixes in the 2000::/64 range, as currently specified by
IANA during the transition from IPv4 to IPv6.

You can manually assign an IPv6 address to your interfaces by modifying
/etc/sysconfig/network-scripts/ifcfg-ethX, or by setting up an IPv6
Router Advertising daemon on your LAN segment, like radvd or zebra. In
my case, I opted out for zebra, as it also offers RIPng and BGP4 routing
protocols in addition to perform router advertising.

Hope this helps.