Add ethernet settings on physical network interfaces

[AmandaCameron]

Fixes #723 -- I'm not entirely happy with the option name I chose, but I'm not great at naming things, so if there's a better name do let me know. Passes go test locally, but didn't try in a VM yet, need to get mkosi installed on my laptop first.

[AmandaCameron]

Testing in a local VM is turning out to be more complicated than I expected. Seems make test assumes incus is running locally, when I've only got it available remotely, or via a libvirt instance of incusos

[stgraber]

I don't think we want to get an extra entry in that struct for every type of offload, so having a sub-struct for it is probably best.

Something like:

type SystemNetworkOffloading struct {
    DisableGRO bool `json:"disable_gro,omitempty" yaml:"disable_gro,omitempty"`
    DisableGSO bool `json:"disable_gso,omitempty" yaml:"disable_gso,omitempty"`
    DisableIPv4TSO bool `json:"disable_ipv4_tso,omitempty" yaml:"disable_ipv4_tso,omitempty"`
    DisableIPv6TSO bool `json:"disable_ipv6_tso,omitempty" yaml:"disable_ipv6_tso,omitempty"`
}

For now they're all "disable" flags for offloads that should be there, but that could also start getting positive values for offloading that's typically disabled.

[stgraber]

Then both SystemNetworkInterface and SystemNetworkBond would need to get a Offloading SystemNetworkOffloading struct field to contain those flags.

[AmandaCameron]

Implemented separate switches for the ones you listed.

I went with making them just be unmentioned if they're not set to disabled, because I didn't want to risk the code causing them to be turned on when the kernel has them off for some reason, causing potential problems.

[AmandaCameron]

ah, forgot I switched my git email, will fix the DCO now

[stgraber]

Thanks, I'll take a look, tweak a bit and land it!

Other ethtool flags were brought up on the forum today (energy efficient ethernet) so I'm going to check if that makes sense to still fit under Offloading or if we need an even more generic thing like Ethernet or something instead for those flags.

[stgraber]

@gibmat given my additions on this one, can you give this a quick review?

[cfouche3005]

Does the WOL settings (#638) also related to this ? Should they also be add to this PR ?

[stgraber]

They'll most likely get added to that struct. I was going to take a look at that next once this one is in with the general infrastructure for this.

[stgraber]

@gibmat should be good to go. I updated the API and filed an issue so we can keep track switching to text/template in the future.

[AmandaCameron]

Will this require waiting for incus 1.21 or will this just work when the stable incusos branch is next updated? I ask because this being a blocker is exactly what motivated me to make the PR in the first place. (Buggy e1000e hw <-> driver interactions)

[stgraber]

@AmandaCameron it will be in the next testing and then stable IncusOS image.
We have a few more things landing right now but there's a good chance I'll tag a testing image later today or tomorrow and we'll be pushing something to stable this week.

[AmandaCameron]

Got it, hopefully I can switch my remaining two incus nodes to incusos sometime soon then. Having to physically go and detach/reatatch the ethernet cable on the single NUC i did as a canary is a bit tedius, and halted my migration entirely.