One computer with two NICs in one segment = strange ARP behavior?

Hi,

i noticed some relly strange ARP behavior on my computer. The computer runs
Debian 3.0 with a 2.4.18 kernel.
I installed two networkcards and connected them to a hub. eth0 has
192.168.40.101 and eth1 has 192.168.40.102, both with a subnet mask of
255.255.255.0 and i didn't define a default gateway.

Now i am pinging these two cards from a second computer, that is also
connected to the hub and has a IP-Address of 192.168.40.1 and a netmask of
255.255.255.0.

Now i do the following things:

tron:~# ping 192.168.40.101
PING 192.168.40.101 (192.168.40.101): 56 data bytes
64 bytes from 192.168.40.101: icmp_seq=0 ttl=255 time=0.8 ms
64 bytes from 192.168.40.101: icmp_seq=1 ttl=255 time=0.4 ms

tron:~# arp -a
? (192.168.40.101) at 00:40:05:2D:ABC [ether] on eth2

then i type:

tron:~# ping 192.168.40.102
PING 192.168.40.102 (192.168.40.102): 56 data bytes
64 bytes from 192.168.40.102: icmp_seq=0 ttl=255 time=0.8 ms
64 bytes from 192.168.40.102: icmp_seq=1 ttl=255 time=0.4 ms

hennes.humungus.lokal (192.168.5.1) at 00:01:02:09:C7:61 [ether] on eth0
? (192.168.40.101) at 00:40:05:2D:ABC [ether] on eth2
? (192.168.40.102) at 00:40:05:2D:ABC [ether] on eth2


And that is strange isn't it? Both IP-Addresses in the ARP table have the
MAC-Address of the first (eth0) network card.

Then i did it the other way:

tron:~# ping 192.168.40.102
PING 192.168.40.102 (192.168.40.102): 56 data bytes
64 bytes from 192.168.40.102: icmp_seq=0 ttl=255 time=0.9 ms
64 bytes from 192.168.40.102: icmp_seq=1 ttl=255 time=0.4 ms

tron:~# arp -a
? (192.168.40.102) at 00:40:05:2D:ABC [ether] on eth2

I stop here because as you see, i get in the ARP-Table the MAC-Address of
the first card, though i pinged the second card.

When you use a default gateway entry, then you will always get the
MAC-Address of the card where the default gateway is set on.


Is this behavior normal? I mean, both cards are in the same segment and get
pinged from a computer that is also in the same segement. Each card should
answer the ping by it's own interface. Very strange. I know that it is not
normal to put two cards on a computer in the same segment + subnet and
normally you would do IP-Aliasing, but why does the Linux Kernel behave this
way? I didn't checked this behavior on Windows-Machines yet, but i will
check it later today.

Any comments welcomed.

Greetz,
Kamran Remin

> Not strange at all. If you check /sbin/route -n, you will likely find 2
> routes for 192.168.40.0/255.255.255.0 with the first one leading to eth0,
> so that is the one used. Also note that Linux knows what IPs are local to
> it and will answer for any of them from any interface.

Well, after reading this ... sounds logical.

Thanks,
Kamran Remin