netifd proto bonding: cannot create bond without IP/netmask

[bobafetthotmail]

Hi, I've tried to create a bond without an IP/netmask to add it as a slave of a bridge. For example I want to bond port1 and port2 of my router and make a bridge between the bond, port3 and port4. The bond is set as "active-backup" to provide redundancy. This allows me to use the OpenWrt router as an "adapter" to connect non-bonding-aware devices to an existing redundant/high availability network setup with two switches where all other nodes are bonding-aware.

But from what I have seen, this protocol (and the luci-proto) requires an Ip and netmask. I also tried deleting the ip and netmask from config file, restarted network service and no bond is created.

I know it is possible to make bonds without an IP/netmask and use them as slaves in a bridge in Linux, I've been doing this on OpenSUSE (also works on Debian) for a while. A IP-less bond can also be used as a target to create VLANs over the bond.

Is there something I can change in the protocol script to just make it ignore the need for IP/netmask and build the bond anwyay?

maintainer (of the bonding proto): @he-ma

[ghost]

@bobafetthotmail Hi, I'm not sure if this is a solution to your problem (as I have no equipment to test this), but could you try assigning IP 0.0.0.0 with netmask 0.0.0.0

[bobafetthotmail]

Tried and it does not work (it creates a bond with IP 0.0.0.0 and if I add that to a bridge it does not work.

I have commented out all parts about setting an IP in the script in /lib/netifd/proto/bonding.sh, and the bond1 I added comes up but it's not configured correctly if I look from cat /proc/net/bonding/bond1 (it shows no interfaces bound and it's not set as active-backup).

I've added ip-full package and then tried to do the commands manually like

ip link add bond1 type bond miimon 100 mode active-backup

and now it creates the bond correctly and it shows the right configuration, but then I can't add any slave to it with

ip link set dev ethernet3@eth0 master bond1

as it fails with

Cannot find device "ethernet3@eth0"

but that's how the "ethernet ports" are called in this device's onboard switch.

I'm suspecting the fact that this device is using DSA for switch configuration (not the old swconfig) is screwing up this netifd protocol.

So, I created a single-port bridge on top of port 3 and another on top of port 4, and created a single unmanaged interface for each so these bridges are created on boot:

Then I went in /etc/rc.local and added

ip link add dev bond1 type bond miimon 100 mode active-backup ip link set dev brport3 master bond1 ip link set dev brport4 master bond1

now it's all good in theory but neither the single-port bridges nor the bond1 never comes up even if I do ip link set dev brport3 up ip link set dev brport4 up ip link set dev bond1 up

So there is something weird going on with the DSA switch infrastructure that breaks bonding or I'm a noob with it.

And I'm back to the old commandeered Gateprotect GP-110 (which is x86-based network device with 4 ethernet ports connected to 4 different ethernet controllers) running OpenSUSE where the above is working fine

[ghost]

Thank you very much for your detailed explanations. So, I assume you solved your problem by your way. I'm also afraid I coudn't help you anymore as I have no experiences with DSA.

[Kritzefitz]

@bobafetthotmail ethernet3@eth0 is, as you guessed, a DSA port. It means there's a port ethernet3 that part of the physical switch eth0. It's a bit confusing, because when you refer to that port you just call it ethernet3, without the @eth0. So your ip link add should be:

ip link set dev ethernet3 master bond1

On my device I can use manual ip command to configure bonding over bridge ports this way and it works fine. However, I think it would be better if I could configure this declaratively in /etc/config/network and didn't have to rely on imperative setup in /etc/rc.local.

[justarandomgeek]

I realize i'm a bit late to this thread, but I wanted this also and found that commenting out

	# For static configuration we _MUST_ have an IP address
	#[ -z "$ipaddr" ] && {
	#	echo "$cfg" "No local IP address defined"
	#	proto_notify_error "$cfg" INVALID_LOCAL_ADDRESS
	#	proto_block_restart "$cfg"
	#	return
	#}

	#proto_add_ipv4_address "$ipaddr" "$netmask"

in /lib/netifd/proto/bonding.sh brings the bond up just fine, and then i can include it in a bridge and configure everything above that as normal - but in doing all this it strikes me, that perhaps a bond would be better treated as a "device" rather than an "interface", in the same way bridges are?

[webknjaz]

Subscribing for updates. I'm also puzzled as to why it isn't treated as a device.

[Dacesilian]

I'm also wondering, why bonding requires to set IP. I want to create bond (device), add it to a bridge and set IP to bridge, not bond.