UDP packet strange behavior

(Fedora Core 4)

I've got a computer A with two Ethernets, connected to two distinct
networks (one corporate, one for development only). I make a socket
for UDP, and I "connect" this socket to a the IP of computer B that's
reached via one of those interfaces.

So I "send" one packet. Recipient B gets it, but it's source IP names
A's other interface! The recipient has no route to get a reply packet
back.

I "send" another packet. This time I get errno=113 - network
unreachable.

What is going on?

 

I suspect that when you called connect() on your datagram socket, it
did the implicit bind for you but happened to do it to the other IP on
the system. But you probably already figured-that out.

To be certain of the source IP, you probably need to bind() to it
first before you call connect(). Doing so will make your application
no longer dependent on stack-specific behaviour such as whether or not
it is a "strong" or a "weak" end-system model.

rick jones
--
The computing industry isn't as much a game of "Follow The Leader" as
it is one of "Ring Around the Rosy" or perhaps "Duck Duck Goose."
- Rick Jones
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

 

How is one computer supposed to know that the other can only reach it
using one particular interface? By what magic would you expect this to
automatically work?

 

I had expected it to supply the IP of the interface that transmitted
the packet.

 

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Please, on your Linux "computer A",
echo 1 >/proc/sys/net/ipv4/ip_forward
and try again

- --
Lew Pitcher

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (MingW32) - WinPT 0.11.12

iD8DBQFE3HV7agVFX4UWr64RAoT0AJ4+D3oF+7HTMAZY3IpMg4IsQO0OfgCeK85Z
vP814x9OKnzshJ1dv58+aI0=
=lyDQ
-----END PGP SIGNATURE-----

 

Actually, I probably shouldn't. I'm not allowed to route packets from
one interface to the other. The development network is supposed to be
somewhat isolated from the production network (and my machine's
existence on both is barely tolerated).

OTOH, the IP addresses don't overlap, and I'm sure I could achieve the
same effect with iptables. But ip_forward=0 was just so much easier...

 

Please, on your Linux "computer A",

Besides, by default, even without ip_forward set, the Linux system
with two NICs will accept traffic addressed to either IP on either
interface. The question is how to get the remote system to know to
reach that "remote" IP it should use the first system's other IP as a
gateway. So, that is where a static route added on the remote system
would come into play.

That, or having the application on the first system make that bind()
call to bind to the IP in the subnet common to both systems.

rick jones

 

You should think about packets as passing through interfaces.

Most operating systems will use the primary IP address of the "closest
interface" when they originate a packet. They also transmit the packet
on the closest interface. This assumes you didn't do something to
change this, such as 'bind'ing the socket to another address.

DS