While there are ways to poke holes in NAT, it’s not really scalable. Also, you might be behind more than one level of NAT and not even realize it today. eg: When I ask a website for my IP, it shows something different than what my hotspot is assigned… and my hotspot is not reporting an RFC1918 address. This means I am already sharing ports with someone else on the public IPv4 address that the world sees. Also, no http proxy in the middle here.
As for obscurity of addresses, NAT is pretty easily guessed in most scenarios today. IPv6 has far more address space per network making it really hard to scan. That combined with privacy addresses that change constantly is a pretty compelling reason to use IPv6 over IPv4+NAT if what you care about is people not being able to guess your IP.
Right, but you can't do this without punching holes in your firewall, and I assert that's not a desirable tradeoff, at least for consumer use cases. As far as I can tell, you still need an arbiter with a public IP address.
> While there are ways to poke holes in NAT, it’s not really scalable.
Agreed, but this is a relatively infrequent problem. It seems like there is some belief that IPv6 is going to make p2p stuff painless, but for most use case it's still going to require poking holes in something; however, for a few use cases (e.g., 2 game consoles in the same network) it will be significantly better. I definitely agree that there's some benefit to foregoing NAT, but it doesn't seem like it will improve most use cases and it certainly doesn't seem like it will deliver the painless p2p experience that many people expect.
> As for obscurity of addresses, NAT is pretty easily guessed in most scenarios today. IPv6 has far more address space per network making it really hard to scan. That combined with privacy addresses that change constantly is a pretty compelling reason to use IPv6 over IPv4+NAT if what you care about is people not being able to guess your IP.
My concern about obscuring addresses was more about making fingerprinting more difficult (some website can't just see my IP address and associate it with my identity), albeit this isn't a well-founded concern, and it could be mitigated by rotating IP addresses.
That's true, but setting up these firewall rules dynamically is way easier than setting up NAT mappings (for example, UPnP through two different NATs never works properly).
And even for consumer use cases, gateways could provide a way to allow all traffic to a specific destination, as most operating systems should provide a proper firewall. Of course, there is still work to be done, the UI for these firewalls should be made better (for example, allowing an application to request to accept incoming packets, and letting the user choose if it should only for the LAN, or for the whole Internet, etc).
Indeed IPv6 by itself will not make p2p stuff painless, but it's still a better basis than IPv4.
This is ultimately an operating system issue. For most of the history of the web, we've used NAT routers and firewalls as a fig leaf over the operating system issue. What is it? Operating systems are extremely promiscuous about listening for traffic on a multitude of ports. Operating systems are promiscuous about including a vast number of daemons running in the background handling a variety of tasks. Operating systems are promiscuous about running a bunch of daemons that phone home all the time.
All of this stuff is completely opaque to the user. All of it occurs on a default opt-out basis. All of it requires an extraordinary amount of knowledge for the user to feasibly withdraw consent. This is the operating system problem.
In another world, I can envision computers running operating systems which are totally transparent and easily understood by their users. All running services would be opt-in and users would be fully aware of exactly what's happening on their machines. That would be the world where end-to-end internet connectivity is highly desirable.
Even then, you have the issue of bugs - not just in the programs themselves, but also in the kernel-mode stack and even in the hardware. As long as something is reachable from the Internet, it will get scanned and assaulted from the Internet - and the lower your attack surface is, the better.
I sort of glossed over this part so now I have a chance to elaborate. Alan Kay has put a ton of thought into this issue [1]. He firmly believes that we can build an operating system and application software with an extremely small footprint (LOC's) so that a single person can understand the whole thing.
Since he gave that talk, we've moved further and further away from Kay's vision. We've made things more and more complex, opaque, centralized, and difficult to change. We've given away our future to big tech companies. Heck, we've even given away the past. We've lost much of the freedom we had back in the 90's, let alone the 70's and 80's when Kay did so much of his work. We're going to have to work incredibly hard just to regain what we've lost.
Do we just go back to the software Middle Ages?
It's ultimately an issue at every layer, hence "defense in depth". Every layer does its part for security, we don't punt because some other layer ought to handle it.
And so many of these layers exist for legacy reasons, business expedience, and market failure. They don’t actually make things better.
You're talking about a different problem: How can I extend the concept of my "home network" to the devices that I use and trust regardless of where I am? I'd argue that this is something that suggests that VPN functionality should get built into gateway devices.
Regardless, I don't want scammers in Malaysia port-scanning my 10 year old printer that's never going to get a security update.
Bugs and therefore vulnerabilities are inevitable. The larger your attack surface, the more likely some rando is to find a vulnerability and exploit it. No walls is real convenient up until someone unexpected walks right in and trashes the place.
You can still have a DMZ, servers, and devices directly connected to the internet, but a gateway with a stateful firewall is a wonderful thing and your typical gateway with NAT helps makes things dead simple solving far more problems than it causes.
It's like saying, if a person walking around at night gets mugged, it's a "them" problem for not carrying a weapon to defend themselves. Uh, no, let's create an environment where even a completely unprotected child is safe. Oh wait, we already have.
https://www.f5.com/services/resources/white-papers/the-myth-...
https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
In the IPv4 case you have NAT and a firewall. If you have some software that you want others to connect to (communication, gaming, etc) you have to punch a whole through the firewall (via UPnP, PCP) and then the software has to use a bunch of protocols to figure out what the public IP address of your router is: see STUN, TURN, etc.
See "How NAT traversal works":
* https://tailscale.com/blog/how-nat-traversal-works/
* https://news.ycombinator.com/item?id=30707711 (2022)
* https://news.ycombinator.com/item?id=24241105 (2020)
With IPv6 you just have a firewall, which you punch a hole through when needed (UPnP, PCP) and you're done (because there's no futzing about with determining the network address). When the P2P session is done the whole is closed and you're protected again.
So if you have a 'home network', it cannot be reached from the Internet by default.
Note: you already have a device that's always on the Internet: your mobile phone. Lots of telcos are IPv6-only and you there's not NAT or firewall between it and the Internet.
The two biggest use cases for direct p2p connections are multiplayer games and video calling. Latency is unavoidable if your traffic has to bounce around a third-party
Not at all--end-to-end encryption is still a very desirable property. I certainly don't want a consumer router decrypting my browser traffic even if it is re-encrypting it to send to my device. I'll tolerate HTTP proxies on the server side when I'm administering the proxy and I need layer 7 routing, but I want to avoid it wherever possible.
> The two biggest use cases for direct p2p connections are multiplayer games and video calling.
You still have to punch a hole in your firewall either way. The only advantage ipv6 has is that you can have two hosts listening on the same port (whereas port-forwarding in a NAT context only works for 1-host-per-port).
Tangentially, I was never a big fan of player-hosted games anyway because they tended to be more vulnerable to cheating and the host always had an unfair advantage (or else a dramatic penalty in the case of lag compensation). Moreover, it's much easier to send a malicious packet directly to another player than it would be to send it to the server and convince the server to proxy it through bit-for-bit (although a poorly written game server might still do just that).
No you don't have to do hole punching.
With IPv4 it is much much harder.
What prevents you from putting the same validation logic into the client, thus rejecting malicious packets at the destination?
My point is that IPv6 restoring the end-to-end principle need not jeopardise the - real or perceived - security of multiplayer games.
If “you” refers to the user, then because the game isn’t architected to have a server running next to each client if the server binary is even distributed to users at all.
If “you” refers to the game publisher, then because they aren’t architecting it that way to begin with, because they aren’t thinking about running the server as a security feature.
Moreover, a game developer has incentives to protect its own servers; it has much less incentive to protect its end users. You might argue that it’s end users being hacked is bad for business, but most end users wouldn’t be able to attribute a hack to a particular piece of software or infrastructure if they even know they’re hacked in the first place (consider the rampant insecurity in the consumer router and iot spaces).
Different use cases beget different requirements.
p2p communications can be nice for latency sensitive communications. Sometimes it's faster to communicate from user A to user B directly instead of going from user A to server Z to user B (although, sometimes it's not faster... if latency is important, you really have to try all the accessible paths and use the best one, keeping in mind that paths may have asymmetric latency, so maybe you want A to send to B directly, but B should send to A through an intermediary; and path latency isn't static, so for a long session, if it's important, you need to probe throughout and change thigns around)
But, maybe you don't want your connection to be a full peer capable of receiving as well as initiating connections, you can run a stateful firewall on your end and drop incoming initiations. You'll still benefit from having end-to-end connectivity because it means your ISP can process your packets with basically no state, so there shouldn't be problems with connection state timing out and your connections being dropped without warning. If you run your own stateful firewall, you may still have that problem, but you might have less state required for a stateful firewall instead of a NAT, so maybe you can manage more connections.
This is a job for a firewall. NAT is not a firewall. You can easily filter incoming connections to untrusted devices when using IPv6, with the advantage that when you want to allow a certain kind of traffic in you can do so without messing around with port forwarding or dealing with multiple devices competing for access to standard port numbers on a single public IP address. That's assuming you actually get a public IP address; if you're behind CGNAT then port forwarding isn't even an option, since it would need to be configured on the ISP's side and not just in your router.
If you enable UPnP for automatic port forwarding, as most do, then NAT isn't blocking much of anything. The only difference between NAT with UPnP and IPv6 with no filter preventing incoming connections is in whether devices which open ports but don't set up forwarding can assume that incoming connections probably came from the same local network. However, it's considered poor practice to treat access to the local network as a means of authentication. (Note that with NAT alone if your router receives a packet addressed to your local network's private IP range, and not the routers public IP address, it will forward it unmodified; preventing that is a firewall function, not a NAT function.)
> and I think it's kind of nice that the Internet only knows the address of my router rather than that of my physical machine
If you use IPv6 with privacy extensions enabled then the Internet will only know your /64 network prefix, which is basically the same thing (unique per subscriber and subnet). The rest of the address will be randomly generated and short-lived, unless you choose to assign an additional long-lived address e.g. for a server.
> I'm not sure whether I care if that gateway is doing NAT as well or not? What can I do with a non-NAT-ing gateway?
Doing NAT isn't the problem, requiring NAT is. When the architecture requires NAT devices can't receive incoming connections without port forwarding even when you want them to. We've gotten rather good at working around NAT's limitations (not without cost), but with IPv6 those workarounds are unnecessary. For example, any peer-to-peer multiplayer game, video chat, or file transfer app where both sides are behind NAT depends on third-party servers for NAT traversal. (Note that the fact that this works at all without actually forwarding all data through the third-party servers shows that NAT is not a reliable system for preventing incoming UDP connections: it can be tricked into thinking a connection is already established.) With IPv6 you don't need the third-party servers as the peers can connect to each other directly.
This will never happen. NAT gets replaced with a stateful gateway still doing conntrack (look at OpenWRT...) and p2p works exactly the same. UPNP, port forwarding, STUN are still relevant and work the same... Except IPv6 hexadecimal addresses are a usability disaster and dual stack will forever be a security disaster. Worst technology ever.
Yeah, blocking incoming connections by default is a bad habit and needs to stop. It's fine for untrusted devices or private VLANs which shouldn't be accepting direct incoming connections in the first place (like cheap IoT gadgets), and should probably be additionally filtered to prevent inter-device connections and access to arbitrary Internet sites, but a laptop, phone, or tablet is perfectly capable of deciding on its own whether to accept or reject an incoming connection, and moreover as a mobile device must assume the network could be hostile anyway.
> Except IPv6 hexadecimal addresses are a usability disaster…
How are IPv6 addresses "a usability disaster" when you never see them? Just use DNS like a sane person.
> …and dual stack will forever be a security disaster.
That's a new one to me. How is dual-stack (IPv4+IPv6) any worse security-wise than any other situation where you have multiple "upstream" Internet connections, e.g. for failover or load balancing?
You don't trust "cheap IoT gadgets". I would like to be able to trust any/all my devices. But I don't.
I don't trust M$/Apple/Linux - AND any associated applications people might want to use at home(kodi, plex, screencast, NAS for example) - to be 100% perfect when it comes to "deciding on its own whether to accept or reject an incoming connection".
I see "block by default" as being a layer of security - one bit of defence in depth.
Happy to drop NAT (with it's IP<->port mapping complications) for a straight IPv6 firewall though.
EDIT: concision
If you've ever connected your phone or laptop to a public WiFi network (or for that matter, the cellular data network) then it's been exposed to an environment were there is no extra layer of protection from incoming connections beyond that implemented by the host itself. We generally expect that to work without major security issues. Non-mobile, "appliance"-type devices might need stronger filtering if they weren't designed to be connected directly to the Internet, but that assumption is becoming less common as more devices require authenticated connections rather than trusting the local network.
I would aim for a default block with allowList and agree with you that a non-IoT host using a UPnP-like mechanism (does UPnP cover IPv6 firewall like scenario?) is probably ok.
Ideally I'd like some kind of notification system where I can click "allow" for the firewall. (Maybe the firewall notifies my phone?) I think UPnP as it currently stands is a bit too hands off but can understand not every user wants to deal with this.
And we agree regards mobiles being in a default hostile environment and expecting it to work. But I see that as a matter of fit-for-purpose. I don't trust every computer I have to that level.
The miniupnpd UPnP daemon (used e.g. by OpenWRT) includes code[0] to handle IPv6 "pinhole" requests—not port forwarding, which isn't required for IPv6, but rather just opening a port in the firewall to permit incoming connections to a certain host.
[0] https://github.com/miniupnp/miniupnp/blob/b734f94bdf6ff555a2...
I wish I could upvote you multiple times. This interaction with you has been most enlightening. Thank you.