Linux Ethernet Bridging - Dec 2008 - [Bridge] received packet with own address as source address

Trying to determine if I have a misconfiguration, misunderstanding or
have stumbled on a bug.

Have a CentOS 5.2 server with 2 e1000e nics in a balance-xor bond, check
this out: 


Test 1 - arping the bond device @ 172.16.0.117 from a second machine
---
root at foo [/root]# arping -c 1 -b -I eth0 172.16.0.117
ARPING 172.16.0.117 from 172.16.0.116 eth0
Unicast reply from 172.16.0.117 [00:15:17:70:A3:88]  0.607ms
Unicast reply from 172.16.0.117 [00:15:17:70:A3:88]  0.648ms
Sent 1 probes (1 broadcast(s))
Received 2 response(s)

Nothing logged to the kernel ring buffer on the server. All looks good
and happy.


Test 2 - now add a bridge @ 172.16.0.117 with the bond as a port:
---
root at foo [/root]# arping -c 1 -b -I eth0 172.16.0.117
ARPING 172.16.0.117 from 172.16.0.116 eth0
Unicast reply from 172.16.0.117 [00:15:17:70:A3:88]  0.603ms
Unicast reply from 172.16.0.117 [00:15:17:70:A3:88]  0.642ms
Sent 1 probes (1 broadcast(s))
Received 2 response(s)

Looks good, traffic moves, etc but now the following gets logged on the
server for every broadcast packet:

[26489.040112] bond0: received packet with  own address as source
address


Capturing the traffic for both tests and comparing, they look identical,
but the bridge wants to throw the warning level message in
net/bridge/br_fdb.c br_fdb_update(). The printk warning, being rate
limited, seems to add a noticable lag. Disabling the warning with a
patch made things happy again.

Couple questions from this for the more knowledgeable folks:

1) Are the two arp replies from a balance-xor bond normal?

2) Should the bridge handle this or should I just not be tying an ip to
the bridge in balance-xor mode?

Thanks for your time!
-- 
Matthew Kent \ SA \ bravenet.com

Matthew Kent <matt at bravenet.com> wrote:
>Trying to determine if I have a misconfiguration, misunderstanding or
>have stumbled on a bug.
>
>Have a CentOS 5.2 server with 2 e1000e nics in a balance-xor bond, check
>this out: 
	I suspect your switch is misconfigured.
	
	The balance-xor mode is nominally "Etherchannel compatible" and
the switch ports connected to bonding balance-xor should be in
Etherchannel mode ("Trunking", etc, but not "LACP" or
"802.3ad").

	If the switch doesn't know the ports are aggregated, it may very
well send broadcasts recieved on one port back out the other port, which
may be the cause of what you're seeing.  The switch might also whine
about flapping of the MAC address.

	If I set up bonding here with the switch unconfigured for
Etherchannel, I see the same behavior as this:

ARPING 172.16.0.117 from 172.16.0.116 eth0
Unicast reply from 172.16.0.117 [00:15:17:70:A3:88]  0.607ms
Unicast reply from 172.16.0.117 [00:15:17:70:A3:88]  0.648ms
Sent 1 probes (1 broadcast(s))
Received 2 response(s)

	Specifically the "Received 2 response(s)" part.  This happens
because the bond receives one copy of the packet on each port, and
responds to each.

	After I configure the switch ports correctly for Etherchannel,
there is only 1 response.  Without the switch configuration IPv6
addrconf also complains about duplicate address detected.

	-J

---
	-Jay Vosburgh, IBM Linux Technology Center, fubar at us.ibm.com