netplan-developers team mailing list archive

[Jason Heeris <jason.heeris@xxxxxxxxx>]

On Tue, Jun 12, 2018 at 3:07 AM, Steve Langasek 
<steve.langasek@xxxxxxxxxx> wrote

> But if that's the case, why do you say that using DHCP for this 
> interface is easier?  What DHCP-provided
> information *do* you want to use on this interface?

I'll disclaim up front that routing and interface configuration 
confuses me no end, so I might have made some assumptions when I set 
this up under ifconfig that don't really make sense. I had it in my 
head that I needed DHCP for (a) DNS addresses, so I could pass them on 
via DHCPD serving to my eth-connected device, and (b) to avoid extra 
config on the router. The router is a nice enough Asus device running 
Merlin, but still a bit clunky to work with.

Now that I dig into it, I think (a) is not true, because my DHCPD 
config (on the machine running Netplan, not the router) has nothing to 
do with whatever DHCP information is provided via wlx485d601f9b83 from 
the router, it's just statically set to the router's address (or I 
could set it to some external service). And (b) doesn't apply, I can 
just configure a static IP on this machine and router will happily 
allow it with no extra config.

But when I set a static address using:

     addresses: [192.168.1.110/24]
     gateway4: 192.168.1.1
     routes:
       - to: 0.0.0.0/0
         via: 192.168.1.1
     dhcp4: no

...even providing the gateway4 and routing settings that recreate the 
above routing table, my eth-connected device doesn't seem to get any 
DNS information from the DHCPS server (again, I mean the one on this 
machine, not the router). Also confusing is that while I didn't even 
touch the config for enp1s0f0, I now have this line in my routing table:

10.0.0.0/24 dev enp1s0f0 proto kernel scope link src 10.0.0.1

This remains even after totally reverting my Netplan config to what I 
had yesterday and rebooting. I am absolutely baffled by this. This is a 
relatively clean install, so there are no custom scripts messing with 
the routing tables that I've added.

Doing this made me realise also - I have to recreate those routing 
table entries I wanted to get rid of anyway. Otherwise the eth-wifi 
bridge can't work, because nothing will get routed out of the 
wlx485d601f9b83 interface. (The other part to the bridge is a "-A 
POSTROUTING -s 10.0.0.0/24 -o wlx485d601f9b83 -j MASQUERADE" entry in 
iptables' NAT table.) So I seem to be stuck with having everything 
think it can go out via wlx485d601f9b83, or nothing. This is getting 
beyond the scope of Netplan perhaps, but if you have advice I'll 
happily take pointers.

> I do see you have a DHCP-provided unicast route on this interface, 
> which is
> unusual:
>
>   192.168.1.1 dev wlx485d601f9b83 proto dhcp scope link src 
> 192.168.1.3 metric 600
>
> I'm not sure why you have this since it's redundant with the route 
> for the
> subnet as a whole which has a lower metric.  Perhaps you were 
> experimenting
> with the DHCP server config?

I had assumed this was Netplan's doing! I didn't put it there manually, 
and it's still there if I disable the DHCP server. No idea what it does 
or why it's there.

In between every config change I am running "sudo netplan apply" or 
even rebooting. Does Netplan's Networkd renderer clear all of its 
output files before re-rendering them? Or is there some state somewhere 
that's persisting, causing these unexpected routes?

Thanks for your help so far.

- Jason