IPv6 Address Selection on Loopback Interfaces

[twgraham]

I have assigned three addresses on a loopback interface I want to use for management purposes (ie "logging source-interface Lo 0").  I have assigned a "primary" routable address from the 2604:: space, a backup address from 2001:: and a unique local address, of course from the FC00:: space.  I would like the local logs to be sourced from the FC00:: address while the UGA addresses would be used for remote monitoring and management.  Does anyone know what the rules for which address gets selected are?  The logging server I am using has a UGA in the 2604:: space and that is the address that is being used to source the syslog messages.  Is it that the 2604:: address is somehow deemed to be "closest".  I would think that a device would try to find a route using the unique-local address first.

[gfowler721]

Separate out the address on to different loopback interfaces.

lo0, lo1, lo2 etc.

[twgraham]

Well that would work but:

1) in the production world this would be a major change to the configuration standard baseline. Yes, I know that this is going to happen with IPv6 - a lot of things will change - but we would like to be as seamless as possible between the two protocols.

and

2) since I am in the lab I am more interested in finding out how IPv6 works and not just hacking a fix.

Since the log server is multihomed I did change the logging host address to use the site-unique FC00:: address and the router started using the FC00:: address to source the log messages.  Now my question is did that happen because of a subnet match or because the IPv6 path discovery determined the FC00:: interface identifier was the best for other reasons?

[gfowler721]

Check RFC 3484 Section 5.

Given multiple Global IPs on a single interface, I believe the source IP of the syslog packets would depend on IP of the syslog server.

If I'm reading this correctly, the Global IP with the most number of bits in common (reading from left to right) with the destination IP wins.