Always is your responsibility even with tailscale. Now it is his responsibility without tailscale.
replyAttack surface almost entirely goes away if you just require a good port knock first! I do that with my protection SSH servers and it works just fine. Also prevents anyone from knowing there even is an SSH server there. Or any server if it doesn't respond to pings either, and only uses reverse tunnels to accept public connections
replyWireguard doesn’t respond to malformed packets or handshakes from unknown devices by design.
replyOh, awesome!
replyIs there a way to automate port knocks in your .ssh/config? I think that’s the main reason why I haven’t added it to my home server.
replyTechnically you could use a ProxyCommand that does the knock and then runs nc to your target.
replyPort knocking is idiotic and amounts to Biff Tannen's "only I know how to start this car".
replyDunno, in combination with having an up to date system following best practices, port knocking is probably going to save you when bots start scanning the internet with the next SSH 0day. Any amount of friction will cause the bots to move to the next IP address.
replyJust moving SSH off a default port gets you 98% there.
replyAn attacker only has to get lucky once, so while I do put SSH on a non-default port too, it's not enough to avoid attacks on its own.
replyI mostly do it to quiet down the noise from the non-stop bots and scanners.
replyBeing off the IPv4 space completely can be even more effective at that, FWIW, but that can have other issues. Only feasible for some!
replyLLMs port knock for redteam scans. Have for a long time.
replyMy knock for SSH was 20 ports within 2 seconds. knockd resets the sequence if you hit any of its ports out of order. I generated the sequence with a CSPRNG.
replyThat's not a knock, it's a password :-)
replyI mean, it suffers from some of the same issues with knocks, like being replayable. Wonder if anyone's come up with a TOTP-like thing for port knocks, where the code changes every little bit...
replyThat’s cute. Have Claude do it. There is already an echo system of TOPT QR to communicate it, so you would just need to convert your six digit code to two or more hits on the 16bit port range.
replyI’m not even sure you need more than two. One seems bad because that’s just a lucky port scanner one time.
deleted
replyIt's more like "only I know how to find this keyhole". You still need the key then, you're just also adding some defense against picking.
replyIt would be idiotic to use the knock as the only authentication, which is why I don't.
To be quite honest, I just like the aesthetics of it.
You're sending your knock over the public network in the clear. It's observable by anything in the path. And then you're not the only one that can find the keyhole.
replyEntirely true, it's not true security, same as changing the ssh port. So, relying on it for any security is not useful but it's a good deterrent from automated scans that are not targeted towards you. Defense in depth means not necessarily discarding flawed defenses if they are in addition to proper defense.
replyWith AI, the complexity doesn't really matter anymore. Codex or Claude can easily configure complex site-to-site setups and peer configurations on mobile devices. I did this by hand before and it was hard and difficult to get right. Now I just tell AI and it does so perfectly, checks and verifies things. AI made life so easy.
replyNo amount of claude can handle the key thing that Tailscale fixes about wireguard, all of the connection to connection logic is handled be it direct, via all sorts of weird hole punching or via relays. Wireguard itself is horrible to do this with, half the time you're just looking at `wg status` wondering why the packet counter isn't incrementing today. It can never achieve the magic that is a direct connection between plane wifi and a host that's behind NAT on the ground.
replyI can’t think of a single time I’ve wanted to connect between two hosts when I haven’t specifically allowed one through a firewall
replyNAT are not firewalls.
replyFunctionally they are. Allow traffic aiming at this address and this port to arrive on this server.
replyNot that my WireGuard connection needs nat as it terminates on my router. And I use private ip ranges as I want to be sure that if WireGuard drops out for some reason traffic doesn’t route over the unencrypted path. Defence in depth.
[dead]
replyIt makes me happy that I can automate some menial tasks or replace a search engine sometimes, but this is just because I already have knowledge of the field and the skill to apply it. You're not wrong that it'll probably get even better, but I don't see a path right now to unskilled people being able to effectively use an agent. Only time will tell if that can happen.
replyI do disagree with GP's assertion that Claude can never make something comparable to Tailscale, but that's just because I believe I (and probably many others here) could probably prompt that, not because I think just anyone could. LLMs can one-shot a fair few things based on unskilled prompts, but I don't believe this is one of them yet.
Right now it's not unskilled. It's managers, PMs, TPM's, or people with some tech knowledge but who previously couldn't do these types tasks (not that they weren't capable, but it takes a lot of effort and time to get skilled in these areas and they didn't devote the time). They know what they want at a high level and now can do it with AI.
replyThis is mostly true, I think. I think this is a sweet spot for LLMs, too. Maybe with the right approach, a user could know or figure out how to articulate what they're missing, if the first attempt doesn't work well enough for them. I just worry that e.g. if they forget to review things like protocol details, they could silently end up with flaws and such, or jank. But yeah, there's potential where there wasn't before, and I'm all for that.
replyThis is true, but I still wouldn't discount the possibility of mistakes. It's good to know how your home network works. Even Opus 5.5 can make some pretty dumb mistakes, particularly if it doesn't have the full picture.
replyHumans also make mistakes. At this point, I trust AI more than humans when it comes to casually configuring VPNs. I've read many who exposed things by mistake on reddit. AI has nearly solved this, probably is a solved issue at this point.
replyI would not personally expose my devices over a VPN that I self-hosted and set up purely through AI unless I felt like I understood how it worked completely. But that's probably just personal paranoia.
replyJust go read reddit threads on people who do this by hand following half baked guides. There's a huge population doing this, now everyone is switching to AI. It's a big win and safer. Not everyone is a trained network engineer, but now they are with AI
replyIt makes no sense at all to configure a >3 nodes wireguard setup with straight wireguard.
replyWireguard is the building block for higher level tools like Tailscale to build on.