[OLSR-users] Rescue access

[Ignacio García Pérez]

> consider to setup and run the ssh daemon/client on any box in the
> path. Then you are able to ssh-hop from box to box without olsr
> routes.

It's not that easy in my setup. I have several network interfaces in each
node, and all are in the same subnet. If per-node routes are not properly
setup, the default route will route all reply packets to one interface. If
you try to connect from a node via one of the other interfaces, your packets
will arrive, but the replies will go out through the wrong interface, making
communication impossible.

> Your setup is
> much too compicated to really function in the field.

Um.... why?... I though it was just the opposite (it is SO simple it should
work in all scenarios). My only concern is that misuse can bring the whole
net to its knees.

Wait, now that I think about it, I was thinking about telnet-like access,
that is, interactive. If you try to transfer a file via this mechanism the
network will be probably overflowed.

To solve that, one possibility would be the following scheme:

1- Use the "simple flood" mechanism only initially in a ping-pong scheme to
learn a valid path to the node we want to reach.

2- Use a second type of packet, which is not flooded but contains the full
path embedded in the header. Nodes will pass the packets along the specified
route.

Problem here is that if the daemon wants to be transparent to programs
(which is what I intended by using the tun driver):

- A path needs to be discovered when we get a packet from tun0 for a node
for which we do not already have a calculated path.

- When do we "forget" that path?. We cannot make it permanent, because what
would happen if the topology changes?. Of course we could be analyzing TCP
packets and forget the path when no TCP connection is alive, but that is way
too much work, and anyway that would not work for UDP packets.

Another possibility is that each node periodically floods the net with a
"HELLO" packet. This packet would contain a path in the header, and before
rebroadcasting it, each node would add itself to the path. When those
packets arrive to each node, it can directly learn the path to the origin
from it.

I don't like this approach. It is fairly simple (though not as much as the
initial proposal), but would generate traffic in the network, and this is
definitely a no-no. The "rescue" scheme must be as unobstrusive as possible.

The more I think about it the more I believe the simplest, initial approach
is the best. Will definitely not work (or even temporarily kill the network)
if used simply to transfer a file, but is more than enough for manual
interactive telnet, which is indeed all I need to get into a node and
bootstrap a rescue process (maybe set up some routes manually, kill olsrd
and launch the previous version, whatever...).

Nacho.