Automatic 6to4 IPv6 Tunnels

Question

Hey Guys,
I understand this implementation pretty well, I'm just having trouble understanding one important part. I understand that if you have R1<--->R2,
with IPv4 connectivity, and then R1 and R2 also each has an IPv6 network on the LAN side, you're address will be 2002::/48, but the bits after
2002:xxxx:xxxx will be the IPv4 source address of the tunnel in hexadecimal. I know you need a route saying to get to 2001:DEAD:BEEF::1/64
goto tunnel0 (etc etc). It will know how to get there automatically by going to 2002:303:303:x.y (Which I'm assuming is the IPv4 remote address in hex)
but how does it know to go to 2002:303:303: automatically........
Im assuming that R1 has physical IP going to R2 of 1.1.1.1 and R2 has a physical IP of 2.2.2.2 going to R1.

Answer 1

In the classic 6to4 scenario, you would be depending on the existence of two additional 3rd party relay routers.  The relay routers would be anycasting 192.88.99.0/24 on the v4 side and 2002::/16 on the v6 side.  Typically the sending client would only have v4 connectivity, not v6.  Some operating systems build in 6to4 tunneling, and some endpoints might be dual-stack, so the number of relays could be reduced.
Sender:
    1. client v6 encapsulated-->192.88.99.1 via next hop R1
    2. R1 -> dual stack relay A (advertising 192.88.99.0/24) via v4
    3. relay A -> v6 destination via R2
    4. R2 -> destination server (v6)
On the reply path,
    5a dual-stack server with embedded 6to4 encapsulates reply directly to client IPv4 address via R2
  or
    5b IPv6-only server sends native v6 reply to relay B at 2002::/16 via R2 using IPv6
    6a R2 forwards v4 packet toward final destination
    6b R2 forwards v6 packet toward dual stack relay B (advertising 2002::/16)
    7a relay B is not involved if the server did its own 6to4 encapsulation
    7b relay B encapsulates the v6 packet in a v4 envelope addressed to the decoded v4 address of the client
    8 R1 receives a v4 encapsulated packet via either R2 or relay B, depending on step 5 choice
    9 client decapsulates v6 reply from v4 envelope received from R1
Geof Huston and others have described why automatic tunnels like Teredo and 6to4 are a bad idea, e.g.
http://www.potaroo.net/ispcol/2010-12/6to4fail.html
-- Jim Leinweber, WI State Lab of Hygiene