zebra should read IPv4 "class E" from kernel regardless of allow-reserved-ranges

[eqvinox]

It's not currently possible to set allow-reserved-ranges early enough to make the initial RIB read in zebra work correctly; config is only loaded after that has already completed.

But more generally speaking, there's an argument to be made here that zebra should always accept Class E from the kernel regardless of the setting; the kernel is the source of truth here and it's not particularly helpful to intentionally desync our RIB view. The switch should probably only control:

• what we permit to load as config (i.e. static routes and addresses in config)
• what we accept from routing protocol neighbors
• whether we feed Class E out into routing protocols

⇒ probably needs a bit of discussion

[donaldsharp]

I'm ok with accepting the kernels source of truth. It might make it fun in some of the protocols when we have a redistribute kernel command though.

[ton31337]

the kernel is the source of truth here and it's not particularly helpful to intentionally desync our RIB view.

Agree.

[donaldsharp]

Let's discuss this dduring today's call.

[eqvinox]

Let's discuss this dduring today's call.

Exactly why I filed this 😄

FWIW the switch should also control whether we redistribute routes into routing protocols. If it's not set, we can still pretend the routes don't exist when feeding out redistributes…

[eqvinox]

related: #16326

[davischw]

Should zebra remove class E addresses from a loopback interface configured by frr on daemon shutdown if the interface is still up (loopback should never flap)?