This is why we have firewalls. My pfSense router e.g. blocks external access by virtue of it being a firewall with sensible defaults.

All routers w/ firewalls will/should have them.

Firewalls are fine but having the ability to make machines unroutable is even more powerful, no?

While I get that firewalls are probably safe like 99.99% of the time… I’ve gotta say, I just don’t trust software to not have vulns of some sort that someone important already knows how to break. So eg for a corporate network I would be hesitant to do this, for my personal network not so much (unlikely that a 3 letter agency gives a shit about what I do).

Additionally. Firewall changes can sometimes fail for whatever reasons so you might have accidentally exposed a node publicly for some time.

Making those nodes not reachable from the internet seems prudent. It’s like your office building has a public address but your office assignments don’t need to be public.

>Firewalls are fine but having the ability to make machines unroutable is even more powerful, no?

For the whole world there is absolutely no difference betwen routable address, non-routable address or even an absense of the machine... behind a firewall with drop all on a public interface.

Your statement doesn’t address the point you are responding to so it makes me think you completely missed the point you are replying to.

You are correct in theory. What you don’t seem to be taking into consideration are firewall vulnerabilities or other unknown things that could happen that make it different from a practical standpoint.

I started an ISP from the ground in 1996. I’ve seen a lot of weird stuff. You seem to just hand waive it away like firewalls are this perfect bastion of security.

Maybe you have more experience than I do?

where do you think the NAT and associated connection tracking is occurring exactly?

Anything smaller than a IPv6 /48 is effectively unroutable. Your ISP has to route smaller blocks, like /64s, privately (because only /48s hit public bgp). And then you have to route the /64 they assign you privately once again.

No. There is no difference between the two from a security point.

Even firewalls are just a last resort defense IMO. The primary thing is not having any crap listening on ports in the first place.

IPv4 with nat is effectively routing your 100,000 hosts through one of 4,294,967,296 possible IP prefixes (a /32).

IPv6 with a /64 routes one of 18,889,465,931,478,580,854,783 possible IP prefixes to your router. Everything that happens behind that is opaque. There are so many IPs in that prefix that people can't even guess which ones you're using. It's easier to guess internal IPv4 NAT addresses.

The internet can't see your IPs. "Publicly routable" doesn't mean much when you're talking about prefixes.

Unless you're planning to assign random IPv6 addresses for each connecton then it's not at all the same thing.

And anyways, when you have a NAT you're probably routing traffic through a level 4 balancer or jumphost or using a VPN somewhere in your LAN.

We assign IPv6 deterministically. Incrementing them is the most boring way to use IPv6 space.

You can also just start in the middle of a block and make a sequence of 10,000 IPs effectively unguessable.

I don't buy that keeping addresses private has much value, but IPv6 still isn't worse than NAT in that respect.

> Unless you're planning to assign random IPv6 addresses for each connecton then it's not at all the same thing.

SLAAC

SLAAC has an option to randomize, but it's a lot slower than per-connection.

As soon as one of the computers in the subnet initiates traffic outside the network, its IP address becomes public knowledge though, right? Doesn't matter how large the space is, that information leaks, whereas with NAT it doesn't.

Modern OS change their outgoing IPv6 address fairly often. They could grab 10,000 from the pool and rotate between them every second.

IPv6 addresses can be ephemeral? What, DHCP is assigning a whole range to each client? Didn't know that.

Pretty weird. For internal stuff I would be using IP addresses to identify peers. Ephemeral IP addresses in logs aren't very useful.