>>>>> "Ignatios" == Ignatios Souvatzis <is@jocelyn.rhein.de> writes:

>>  Nope, doesn't work.  I tried changing the netmask to
>> 255.255.255.255 (which, in retrospect, seems like a very good
>> idea), but I get the same symptoms:  If that route exists when
>> cricket dials in, then pppd doesn't change it to point out the
>> correct interface.

Ignatios> why does such a route ever exists? do you put the machine
Ignatios> onto the Ethernet sometimes (in this case the route contains
Ignatios> the link level address, replacing what was the arp cache in
Ignatios> 4.3BSD) ?

The route gets cloned from the default route on fxp0, if I even ping
cricket while ppp0 is down.  Because cricket's IP is 153.90.240.162,
and the default route puts 153.90.240/24 on fxp0, the kernel believes
cricket is on the local net.  So when I do *anything* that tries to
get to cricket, and ppp0 is down, a route gets cloned that points to
fxp0.

So, if anybody tries to access cricket from finn, when cricket isn't
connected, then cricket will be unable to connect until finn is
rebooted.  I don't think I can guarantee that nobody will ever try
that.

Ignatios> The "proxyarp" parameter should handle this right. You'll
Ignatios> need it anyway to make cricket visible on the Ethernet.

Yeah, it should.  I've already got that in my config, though.  When
this route problem doesn't occur, proxyarp works just great.

In fact, proxyarp may make things worse.  Finn advertises (via ARP)
that it's the place to send traffic designated for cricket.  So, if
cricket connects to finn, and then sends out an HTTP request, and then
disconnects, then the router on finn's ethernet will now have an ARP
entry that points to finn.  So when the HTTP reply comes back, the
router will forward it to finn, and finn will (probably; I haven't
tested this) end up cloning that route entry on fxp0 again.

Chris

-- 
-----------------------------------------------------cjones@math.montana.edu
Chris Jones                                          cjones@honors.montana.edu
           Mad scientist at large                    cjones@nervana.montana.edu
"Is this going to be a stand-up programming session, sir, or another bug hunt?"