>  |    however, it is not always possible to know peer's address beforehand:
>  |    - we don't always know peer's address (think about linklocal address
>  |      - fe80::/10), and
>  |    - we don't run NS/NA on L2 without hardware address, so
>  |    - we don't know the peer's address.
>Then what use is installing the route?   If you don't know the link
>local address of the remote end, simply don't install a route to it.
>
>On a p2p link the only people who are going to be able to send to a
>link local addr are the 2 end points, by definition you don't know his
>address to send to, thus you can't be sending any packets to him.
>If he sends packets to you (hoping that you will loop them back to him)
>then whether you fail to do that because of your proposed rule that
>makes a special case for p2p links, or whether you do it because there's
>no route to his link local, seems to make little practical difference.
>
>If you do, via a routing protocol, or any other way, manage to discover
>what the remote end's LL address is for the P2P link, then you can install
>the route, and then you would be able to send to it.

        suppose you have a tunnel between two machines, establish a tunnel
        between them, and would like to run ripng.  they have link-local
        address, La and Lb, respectively, on their interface.

        router A
          | La
          |
          | Lb
        router B

        normally router A does not know Lb, router B does not know La.
        they will learn about each other using routing protocols (ripng).

        to throw out ripng packets to the other end, we would need to have
        fe80::%interface/64 as well as ff02::%interface/64 to be installed
        onto the routing table.  they are normally installed by default by
        the operating system.  now, we have the problem i have described
        repeatedly (packet to non-existing destination will make a loop).

        am I confused in some way?  the above is reallife example for us,
        how do you setup routers in your setup?  do you tell the other end your
        linklocal address?  if you did that, you will have trouble replacing
        router in your end.

        or, are you proposing this scenario?
        - routing table has ff02::%interface/64
        - routing daemon discovers each other
        - routing daemon installs Lb%interface/128 once they discover each
          other
        none of the existing routing daemons (i know of) work this way.

itojun
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to