On Fri, 4 Feb 2000, Terry Dawson wrote:
> to do is to send an ICMP redirect, but that for some reason it isn't
> able to do that.
In fact, the correct behaviour is to send an ICMP redirect, and resend
the packet anyway. But I'm not sure if the ICMP redirect is made if
the packet has a route to another gateway os is sent ever if the station
is directly reachable. I have to find my CD with all RFC.
> I'm going to draft an email to Alexey and see if he can shed some light
> on it. I think it might be an artefact of him cleaning the IP routing
> code at some point. In every network I know of, other than AX.25, the
> behaviour we want would be considered broken, unfortunately it's
> absolutely necessary for us.
>
Is also necessary in some wired networks. Say you have an internal LAN
and two subnets say 10.42.42.0/24 and 10.69.42.0/24, and have a router
to connect to another network via ppp, say 10.69.43.0/24.
Now, the use of two subnets is made because the network will grow, and
make an IP renumbering is difficult. But now to trim costs down you have
to use a simple hub to connect the system.
When a machine on 10.42.42.0/24 want to talk to a system on 10.69.42.0
network, will send a packet to the router. The router send a ICMP redirect
to the originating machine but it will send anyway the packet back on the
interface.
Now this behaviour is correct because if the TCP/IP styack on the sender
is correct the next packet will be sent directly to the redirected host,
because the route will be added pointing to the correct host, but the
original packet is not resent.
And anyway Windows 3.11 and Windows 95 normally discard the ICMP redirect
message.
The fact that the network is not the better one you could design is true,
but in some case make wired networks in this manner is a necessary kluge.
73
Mike
