Bonding + Bridge : traffic passing through the wrong interface

Question

I'm using Ubuntu server 12.04, on a server with 6 NICs, grouped in 2 bundles :

• eth0 and eth2 are bundled using bonding mode 1, under interface name bond0 which has IP [network].8
• eth1, eth3, eth4 and eth5 are bundled using bonding mode 4 (802.3ad), under interface name bond1

bond1 will be used to connect VMs to our network : it is bridged through br0, which has IP [network].

Now, when I ping [network].5 from our network, everything seems to works, but our VMs have no network access.

After poking around for a while, I noticed that br0's IP ([network].5) is associated with bond1's MAC address, i.e.:

arping <[network].5>

returns

Unicast reply from <[network].5> [<bond0's MAC address>]  0.710ms

Also, while I'm pinging [network].5 :

tcpdump -i br0 icmp

Shows no ICMP traffic,

tcpdump -i bond1

show no traffic either, but

tcpdump -i bond0

shows the ICMP packets I'm sending using ping.

It's pretty obvious that packets are sent down the wrong tube. My question here is : why is it so and how can I fix that ?

Here is the content of my /etc/network/interfaces file :

# bond0 part :

auto eth0
    iface eth0 inet manual
    bond-master bond0

auto eth2
    iface eth2 inet manual
    bond-master bond0

auto bond0
    iface bond0 inet static
    address [network].8
    gateway [network].254
    netmask 255.255.254.0
    # bonding mode 1 :
    bond-mode balance-rr
    bond-slaves none

auto eth1
    iface eth1 inet manual
    bond-master bond1

auto eth3
    iface eth3 inet manual
    bond-master bond1

# bond1 and br0 part :

auto eth4
    iface eth4 inet manual
    bond-master bond1

auto eth5
    iface eth5 inet manual
    bond-master bond1

auto bond1
    iface bond1 inet manual
    # bonding mode 4 :
    bond-mode 802.3ad
    bond-slaves none
    bond-miimon 100
    bond-downdelay 200
    bond-updelay 200        
    bond_xmit_hash_policy layer2
    bond_lacp_rate fast

auto br0
    iface br0 inet static
    address [network].5
    netmask 255.255.254.0
    gateway [network].254
    bridge_ports bond1
    bridge_maxwait 5
    bridge_stp off
    bridge_fd 0

Please note that :

• 802.3ad link aggregation has been configured on the swich side
• We have verified muliple times that the right ports are connected
• The same issue occurs on 2 servers with the exact same hardware+software configuration

[EDIT] After several reboots, the opposite happens : bond0 is associated with bond1's mac address. This seems to happen randomly. When it does, VMs behind the bridge have access to our network and Internet.

Solution

After more searching, it appears the root of my problem is what is discussed here :

Serverfault : arp responses always go out a sinle NIC

As this question is much more precise and has several answers, that's where you should look if you have the same problem than me