Re: [PATCH net] net: do not bond/team netdevices which use ml_priv

On Wed, Aug 19, 2026 at 08:15:42PM +0200, Oliver Hartkopp wrote:

Hi Hangbin, all,

thanks for your feedback!

In fact checking for dev->ml_priv != NULL seems to be too hard and would
likely create some breaking setups.

I have sent a patch on the Linux-CAN ML that makes the CAN subsystem robust
against alien ARPHRD_CAN interfaces that might be created by TUN/TAP via
TUNSETLINK ioctl:

https://lore.kernel.org/linux-can/20260819160822.8256-1-socketcan@hartkopp.net/ (local)

The other question remains how to make bonding/team keep the fingers away
from incompatible net devices.

The current black listing approach from Jiale Yao (using a unified helper as
suggested by Jiri Pirko) seems to be the best way to go now:

https://lore.kernel.org/netdev/20260728151240.89434-1-yaojiale02@163.com/ (local)

+		if (port_dev->type == ARPHRD_CAN ||
+		    port_dev->type == ARPHRD_IEEE802154 ||
+		    port_dev->type == ARPHRD_IEEE802154_MONITOR) {
+			NL_SET_ERR_MSG(extack,
+				       "CAN and IEEE 802.15.4 devices can't be added as a team port");
+			netdev_err(dev, "Device %s is CAN or IEEE 802.15.4. These device types
can't be added as a team port\n",
+				   portname);
+			return -EINVAL;
+		}
+	}
+

This looks good to me to avoid making the checking to wide.

There might be other dev->types (e.g. ARPHRD_ARCNET) or a future white list
approach this unified helper could support. For either team and bonding!

@Jiale Yao: What would be a good name for such helper in
include/linux/if_arp.h ?

dev_is_not_ether_compatible(const struct net_device *dev)
dev_is_ether_compatible(const struct net_device *dev)
dev_has_special_l2_proto(const struct net_device *dev)
dev_has_l2_proto(const struct net_device *dev)
dev_is_pure_packet_bus(const struct net_device *dev)

or something like this?

I'm not good at naming. Leave it to you.

Thanks for your works!

Hangbin