Creating a bonding interface via the ip tool gives it the wrong MAC address

[Mikhail Morfikov <morfikov@gmail.com>]

A couple months ago I opened an issue on the Debian Bug Tracker[1] concerning
some weird network behavior, in which bonding interface was involved. Basically, 
what I wanted to achieve was to have two interfaces (eth0 and wlan0) of my 
laptop in the *active-backup* mode, and in order to make this work, the 
*fail_over_mac* has to be set to *none*. Then the two interfaces (and also the 
bond0 interface) should have the same MAC address, which is set based on the 
interface specified by the *primary* parameter (in this case eth0). 

This was working well in the past, but it stopped for some reason. When the 
bond0 interface is being set up via the /etc/network/interfaces file, it gets 
wrong MAC address, and it's always the same MAC (ca:16:91:ae:9a:ba).  

I didn't really know where the problem was (it looks like no one knows so far), 
but I recently moved from ifupdown to systemd-networkd, and I noticed that the 
issue went away, at least in the default config. But in my case, I had to 
create the bonding interface during the initramfs/initrd phase using the *ip* 
tool (the regular one, and not the one from busybox). And the problem came back, 
but in this case I couldn't really fix it by just restarting the network 
connection.

So I created manually the bond0 interface using the *ip* tool in the following 
way to check what will happen:

ip link add name bond0 type bond mode active-backup \
  miimon 200 \
  downdelay 400 \
  updelay 400 \
  primary eth0 \
  primary_reselect always \
  fail_over_mac none \
  min_links 1

and the interface got the MAC in question. That gave me the idea that something 
could be wrong with setting up/configuring the bonding interface via the *ip* 
tool because it works well with systemd-networkd, which I think doesn't use the 
tool to configure the network interfaces.

So why does this happen?


[1]: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=949062&archived=False&mbox=no&mboxmaint=no