Comment 5 for bug 1776013

Revision history for this message
Wladimir Mutel (mwg) wrote :

The problem I had was not related to netplan, networkd or dhcp client of any kind.
I had both my ISP links connected to a VLAN switch DLink DGS-1100-08 and multiplexed to a single Linux router interface with different VLAN numbers.
On Linux side, vlan.NNN interfaces were configured to separate each ISP traffic by its VLAN number.
These interfaces were managed by separate DHCP clients, requesting dynamic IPs from their corresponding ISP.
Both ISPs had a policy permitting only a single MAC address to request a DHCP lease on a single customer's port.
And this was combined with the fact that for some Ethernet protocols, DLink VLAN switches of this model do not obey VLAN isolation rules.
In part, among them were DLink's own discovery & loobpack detection protocols (Ether.proto number = 0x9000, using multicast MAC addresses prefixed with CF: byte).
My both ISPs had a plenty of DLink hardware in their Ethernet segments, and my DLink switch inadvertently served as a bridge for DLink discovery packets, allowing them to fly freely between two VLANs.
Which of course was detected by each other ISP's policy as 'stray MACs' on a single customer port, and caused my real MAC to be temporarily kicked out from their valid MAC set, which subsequently caused DHCP lease loss.
What resolved by problem once and for all, was installing dual-port PCIe Gigabit Ethernet adapter into my Linux router and physically plugging my ISP links into its separate ports.
Detailed discussion can be seen at http://forums.dlink.com/index.php?topic=74704.0