Al Boldi skrev: > TLB mode doesn't need any switch support, yet the message persists when > part of the bridge. There is several bonding modes not needing explicit switch support, but all share the minor drawback that some packets (broadcasts and where the destination MAC location is unknown to the switch) will loop back to the host on the other interfaces in the bond. > Does this mean that the kernel bridge does not support bonding? It shouldn't care, but I cannot say I have tried the combination in detail yet. > Should an IP be bound to a dev? Well, it is not. All configred IP addresses is available on all IP enabled interfaces. The minor detail on which interface the IP address was added only tells where the implicit route to that network should be pointing.. >> The local IP stack will get somewhat upset if it receives IP packets >> with >> a source IP matching one of it's own interfaces, as this indicates >> something is wrong in the network either an IP conflict or packets being >> looped back to the sender host. > > Ping from the local host to a non-existent IP reproduces the message. Which fully matches what I have been suspecting wrt your bond all the time. This ping will initiate an ARP broadcast to learn the MAC of the pinged IP. This packet goes out on one interface in your bond and is broadcasted by the switch on all other ports, including your other interfaces in the bond. > Can the bridge handle broadcasts looping back? It handles it fine. The message is just a warning that something is odd in your network. It is quite interesting that you do not see the message when only using bonding without bridgeing. I have no explanation why you are not seeing this message when using bonding only. Regards Henrik - : send the line "unsubscribe linux-net" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
