Right now, ipv6_prefix option in OVN IPAM implementation maps to EUI-based ("stateless") IPAM address generation. This is a good default and I don't propose to change it.

What I suggest here is to be able to use different (longer) prefixes. A specific use case here is an OpenStack cluster (18+) that runs on top of OCP with OVN-K. Existing (OSP-17 and earlier) clusters were deployed into a single /64 prefix. With OVN-K, now every OCP node requires its own /64 prefix (due to 1LS per worker).

This means that adoption of existing OSP 17 environments into OSP 18 is complicated because now the adoption also implies reconfiguration of underlying infrastructure for the new address mapping.

I assume the feature would be implemented by mimicking what "subnet" option is currently doing for IPv4 addresses, namely allocating addresses from the subnet range manually and tracking "free" addresses in a runtime calculated bitmap. (I am not sure the approach can be mimicked verbatim due to significantly larger sizes of ipv6 prefixes, so perhaps something more elaborate than a bitmap of free addresses would have to be used.)

Perhaps "subnet_v6" could be a good name for this option.

Note: to complete the loop for OpenStack purposes, we would have to also expose the new option in OVN-K. I haven't created a Feature Request for OVN-K, yet, to first allow FDP team to confirm, or deny, the request here. If this sounds like a good idea and OVN team is ready to work on it, I will follow up with OVN-K jira.

Relevant customer case: https://access.redhat.com/support/cases/#/case/03414509/discussion