Manually set route keeps reverting to default

[Rich <openbsd64_(_at_)_rbentley_(_dot_)_com>]

Hello people,

I have seen a similar problem to this posted previously, but not quite the 
same. Also the description of my setup is quite long and I suspect it might 
actually be irrelevant to the actual problem, but here goes anyway...

SETUP
--------
I have three machines (well, three that are relevant to this), all running 
OBSD. they are arranged thus :

#1                             #2                                 #3
192.168.0.5/24          192.168.1.1/24                 192.168.0.64/24
alias 192.168.1.2/24   alias 192.168.0.160/24        |
|___________________|_____________________|


#1 is my laptop that I work from
#2 is a machine that I'm preparing to move somewhere else
#3 is an old laptop I use as a dial-on-demand / firewall connected to a modem

Because I'm preparing to move it, one of the machines (#2) is on a different 
subnet (192.168.1.0/24). To allow me to work on it in the meantime, I've got 
round the obvious addressing problem by adding an alias to #2 of 
192.168.0.160/24 (to put it on the same subnet as the other two machines) and 
fiddled the pf configuration in #2 so that it will redirect 192.168.0.160 -> 
192.168.1.1.

In turn, I've done something similar on #1 but in the opposite direction - ie 
- 
added an alias to allow it to listen to the 192.168.1.0/24 subnet and fiddled 
the pf config to redirect 192.168.1.2 -> 192.168.0.5.

The missing link is the routing table. I have added this route to #2 so that 
it can find #1...

   route add 192.168.0.5/32 192.168.1.2

...and I've added this route to #1 so that it can find #2...

  route add 192.168.1.1/32 192.168.0.160

I've not bothered doing the same with #3 but for what I'm doing, #2 shouldn't 
need to refer to a default route anyway, so it shouldn't be a problem.

PROBLEM
------------
Now the problem....

This all works fine except that now and again, the route I added on #1 
spontaneously disappears and gets replaced with....

   192.168.1.1   192.168.0.64

...ie - it decides to move the route that I have set up for #2 to the default 
machine #3. This is completely wrong of course because #1 is NOT going to 
find #2 via #3 !

I ran the route monitor and found that this happens immediately after the 
following error :

got message of size 108 on Fri Aug 13 18:19:52 2004
RTM_LOSING: Kernel Suspects Partitioning: len 108, pid: 0, seq 0, errno 0, 
flags:<UP,GATEWAY,HOST,DYNAMIC,DONE>
locks:  inits:
sockaddrs: <DST,GATEWAY>
 venus.netone.domain tosh

('venus.netone.domain' being machine #2 and 'tosh' being machine #3; the 
default route)

I've dug about and (correct me if I am wrong), this error seems to be viewed 
as 'normal' (?) / to be expected now and then - it seems to happen when I'm 
messing about with #2 and some service or other on machine #2 isn't 
responding for some reason (like I've just broken some config or other).

I have had it fail in the same way when I've not being messing about - the 
setup can run for days with no problems and then the route will suddenly 
switch. Or it can switch several times in the space of an hour - there seems 
to be no pattern.

My question of course, is how do I stop this happening ? Either stop the error 
happening and/or stop my route spontaneously changing like it does - 
preferably both. Machine #1 taking it on itself to change the route table is 
by far the most annoying problem !!!

I realise that the setup I've got is pretty crude and  abuses the addressing 
scheme somewhat but it does work - most of the time. I'm aware that a better 
better setup would be to stick a bridge (which I don't have) between #2 and 
the rest of the network or have a second NIC on #1 (which I don't have).

By the way, I am NOT running routed on any of the machines. Also, this occurs 
even when #3 is switched off so clearly that has nothing to do with it (not 
that I thought it did anyway).

regards,

Rich.