networkd: force bringing up devices with no IP addresses

[cyphermox]

Signed-off-by: Mathieu Trudel-Lapierre mathieu.trudel-lapierre@canonical.com

[cyphermox]

This is failing tests on purpose for now, as it's up for review but might need more work to make sure things are working correctly.

The intent is to fix bringing up anonymous bridges, it seems to work reasonably well for that. Non-virtual interfaces will also be affected though, and we should make sure they are fine to be UP, NO-CARRIER in this case. Where people absolutely want the devices to be unconfigured, then we should expect the devices to not be listed at all in netplan YAML.

[raharper]

In some cases folks want something like lxdbr0 or virbr0 where the bridge itself has an address, but no members. This check looks like it won't set ConfigureWithoutCarrier=true if you assign any DHCP or IP values. I understand DHCP, as it needs an interface to emit the DHCP request, but static IP assignment should be OK, right?

[cyphermox]

That's a different issue; a bridge with no members will be handled separately in a rework of the validation code.

Here, it's exactly what we want to "configure" the interface is there's no carrier and no address; and afterwards we'll add "configuring with no carrier" for the interface-less case.

[georgmu]

A have different case where I need the ConfigureWithoutCarrier=true with an address configured, but maybe no carrier.
I have an interface which should serve as DHCP/DNS server via dnsmasq. The interface may have no carrier, but it should be started anyway because dnsmasq fails to bind the IP if the interface is not up on start of dnsmasq.

So I would propose to bring the interface up if there is an address configured, or to add a new option to force the start of the interface.

[codecov-io]

Codecov Report

Merging #34 into master will not change coverage.
The diff coverage is 100%.

@@          Coverage Diff          @@
##           master    #34   +/-   ##
=====================================
  Coverage     100%   100%           
=====================================
  Files          32     32           
  Lines        3974   3977    +3     
=====================================
+ Hits         3974   3977    +3

Impacted Files	Coverage Δ
tests/generator/test_common.py	100% <ø> (ø)	⬆️
tests/generator/test_ethernets.py	100% <100%> (ø)	⬆️
src/networkd.c	100% <100%> (ø)	⬆️
tests/generator/test_tunnels.py	100% <100%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fb71b87...d96a6a0. Read the comment docs.

[vorlonofportland]

I would expect there to be some documentation changes with this, as this is a behavior change.

[cyphermox]

Merged as of a5389e5

[KalenAnson]

@cyphermox Looking this closed PR over, I see that e681f1e did not make the cut (or was unintentionally omitted). I am referring specifically to configuring a physical device without a carrier. Master Ref: https://github.com/CanonicalLtd/netplan/blob/master/src/networkd.c#L576

The update to the documentation seems to indicate that the change in behavior in this PR would include the condition that the linked commit addresses, i.e. an interface specified in the netplan config will be configured by the renderer (regardless of carrier link state presumably).

If an interface is defined with an ID in a configuration file; it will be brought up by the applicable renderer. To not have netplan touch an interface at all, it should be completely omitted from the netplan configuration files.

Can you shed some light on why this additional behavior change was not introduced to how netplan functions?

It would be very convenient for an interface that has no carrier to still be able to receive a static address so that services that depend on it would not have to specify interface requirements in both their own config files as well as systemd service files in the form of 'after's, 'bindsto's and 'wants' (dhcp servers, dns servers, et.al.).

[sabado]

Switch to Netplan and mantain a bug from '17, put automatic updates by default in server edition. Did you guys have worked on Microsoft or what?

[vorlonofportland]

On November 12, 2020 7:23:05 AM PST, Ramiro Ricoy ***@***.***> wrote: Switch to Netplan and mantain a bug from '17, put automatic updates by default in server edition. Did you guys have worked on Microsoft or what? -- You are receiving this because you commented. Reply to this email directly or view it on GitHub: #34 (comment)

You're replying to a closed issue. Maybe you want to open a new issue for whatever bug you're referring to? -- Steve Langasek

[n-cc]

d4884cf adds support for networkd's ConfigureWithoutCarrier option via the ignore-carrier key.