On Fri, Aug 30, 2002 at 09:24:25AM +1200, C Falconer wrote:
> Gidday all.  I'm going to describe a problem that I was having with my
> work machines, and I'm going to see who comes closest to the correct
> answer to the problem (which took three days to solve)
> 
> I have a Ppro 166 doing firewall duties.  It has eth0, 192.168.1.1 with
> netmask of 255.255.0.0 for internal, and eth1, 202.0.37.196 /
> 255.255.255.0 for the cablemodem side.  On the firewall, ping works fine
> to internet hosts, localhost, and internal hosts.  It is runningf 2.4.19
> with an iptables firewall script.
> 
> Internally there is a machine called belt, which is the main linux box. 
> It runs bind9, squid, exim, and samba.  belt has two nics
> eth0      Link encap:Ethernet  HWaddr 00:02:E3:16:7F:FE  
>           inet addr:192.168.1.2  Bcast:192.168.255.255  Mask:255.255.0.0
> eth1      Link encap:Ethernet  HWaddr 00:02:E3:15:D4:CB  
>           inet addr:192.168.1.12  Bcast:192.168.255.255 Mask:255.255.0.0
> 
> eth0 is bound to squid and other IP related stuff, eth1 is used for
> samba and windowsie stuff.  Both NICs are plugged into the same 100 Mbit
> switch.
> 
> Destination     Gateway       Genmask        Flags Metric Ref Use Iface
> 192.168.0.0     *             255.255.0.0     U     0      0    0 eth1
> default         192.168.1.1   0.0.0.0         UG    0      0    0 eth0
> 
> All internal pinging from belt works fine.  What seems to be the
> problem?

I'd haszard to guess it's because you're running two interfaces on the same
subnet, and you're binding to one ethernet interface, and the packets are being
sent with a source address of 192.168.1.12, and then you're sending them out
eth0, but receiving via eth1.  You should turn reverse path filtering off,
which seems to default to on for some brain-dead reason.  Or it's doing some
kind of load-balancing and using both interfaces.

If you wanted to use two interfaces, wouldn't it make more sense to make
them both into a bridge, and both be virtual interfaces, hooked up to two
different switches.  Assuming you can do STP on your switch.  Or maybe you
can just do trunking on your switch, and make one 200 megabit interface out
of both interfaces?

The other thing is that the ping problem could partially be rate-limiting
icmp-echo/icmp-reply packets.  (RedHat seems to default this way now days)

Also why are you using 192.168.0.0/16 when it's an area set aside for /24s?
It means you're sure to get a clash if you connect your network with
another using address space within 192.168.0.0/16.

Ben.

Reply via email to