🚀 go-pugleaf

James Binkley (jrb@rigel.cs.pdx.edu) wrote:
: Back to 2 subnets on 1 segment again:
: 
: ...
: 
: We have an experimental server on zymurgy that is sending 
: out a directed broadcast on net 2 as 140.140.2.255. 
: Zymurgy has the capability to have 2 ip addresses
: attached to a single enet device.
: 		140.140.1.240
: 		140.140.2.240
: It is sending the directed bcast out to 2.255 as an
: ethernet bcast.
: 
: For "no apparent reason" an aix box and an hp X terminal too 
: (both on net 1 with 1 interface and 1 ip address on that
: interface)
: are picking up the directed broadcast and forwarding it
: ...
: Any insights here?  Mine is: "Life with 2 subnets
: in the presence of buggy tcp/ip implementations can be
: interesting :->"

BSD 4.3 networking code bites again... 

An IP implementation shouldn't forward a frame that comes in with a
broadcast (or multicast, for that matter) LLA. Unfortunately, 4.3
networking code throws away the information that the frame came in
on a broadcast link address, so ... If for any reason it is decided
by IP that the frame should be forwarded, then it will be if there's
a default or appropriate route.

I don't know about the AIX box, but the HP Xterminal does have two
network interfaces, even if you aren't using it ... the SLIP interface
on the serial port is 'attached', even though it isn't active. Having
2 interfaces and finding that a frame has been sent to it which has an
IP address that isn't any kind of broadcast address that it understands
(in fact, the address isn't even for its subnet), it forwards to the
default gateway, the Cisco box. You're lucky that the gateway was the
default or an ICMP redirect would also have been sent out.

The AIX box is probably doing something similar - it's something that
plagues this vintage of TCP/IP software.  Berkeley's Networking Release
2 fixed this by keeping a LLA broadcast/multicast flag in the mbuf header
that follows the frame around. Forwarding can thus be prevented for LLA
broadcasts (and 2 subnets on the same wire will be safer :-)

Incidentally, the most recent software release for the HP X terminal has
the network code modified so that forwarding to the same interface is never
done - the frame is just dropped. Check around on the AIX box and see if
there is any indication of 2 interfaces being attached (lo0, loopback
doesn't count). The forwarding code should never be called if there is
only one interface (in BSD 4.3/Tahoe code anyway).

Regards,

Ted Rypma
HP Panacom Division

(Any opinions, etc. expressed are entirely my own.)
even for its