Hacker News Reader

Comment by devilbunny | Hacker News Reader

Tailscale has another interesting feature that I figured out entirely by accident: while the SSO planes (at least using Apple as SSO, rather than your own) may be blocked, the data planes and actual control planes usually are not. If your device is connected to your tailnet before joining a given WiFi, it will stay connected afterward.

The guest WiFi at work blocks OpenVPN connections, but established Tailscale slips by. I haven't tried straight Wireguard because I don't consider Tailscale having timing and volume data on me to be all that valuable to them, and they do mitigate the double-NAT situation. I do run a private peer relay for my tailnet but not a full DERP server, nor do I run Headscale.

Obviously, your personal security concerns play a role here, but I'm not doing anything I wouldn't do straight from my home network, so I see no reason to make my life harder. If you need that level of security, you need a different solution.

View on HN

While waiting for someone in the hospital I recently played the fun game of "how can I work around their firewall stopping me from connecting to tailscale" that they kindly provided.

It was just blocking new connections. Via SNI. Tailscale's control plane turn out not to care if SNI is sent. Tailscale's app let you set a custom control plane... like a local proxy that forwards connections to tailscale's servers without setting SNI.

This may very well be the system in use.

I've seen this effect in several places, not just my work.

Of note: I do not work in the tech sphere. I suspect that this particular loophole may be used by IT personnel to be able to tell the management "yes, we block VPN use" while letting them continue to use their own VPNs. I see no reason to complain.

I suspect there's less thought put into it than that.

There's probably a firewall vendor that has a product that does SNI inspection for blocking things like pornhub and the product comes with a list of sites that includes VPN control planes.

Well, yeah, they didn't roll their own. Offhand, I forget the product, but it's definitely off the shelf.

My point being that surely some of them have noticed the same thing I have, and it hasn't been stopped. I'm not going to raise the issue either way.

> I'm not going to raise the issue either way.

Except, you kinda just did

Not to them, in a way they can’t just ignore. I could be anyone here on HN.

That's why I said kinda.

It'd be funny if someone working there was a visitor here...and it doesn't matter who you are. I was thinking of them closing the loophole

I understand your point better now, but if that was really a risk I cared about, I wouldn’t have put it on the public Internet to begin with.

The worst they can do to me is make me tether, and my iPad will never hit that allotment. And, like I said, I think they use it themselves. So, no incentive to close their loophole.

Wait, tailscale survives connecting to a locked down wifi? That's insane. I remember not being able to use NordVPN at work. I'd just switch to 4G back then. But if you can't initiate a tailscale connection when connected to the office wifi, what does that mean?

Initiate while on mobile connection or tethered to one (or just leave it connected from home), use while on that WiFi.

EDIT: I figured this out because I brought my laptop from home to do a few things while at work that needed it. I noticed that my Tailscale connection (initially established at home) was working just fine. That's when I realized that it was the initial authentication that was blocked, not the service.

My phone is usually on my tailnet and my iPad is always on it (and using my home exit node), as a result. Using the exit node has a modest but noticeable effect on battery life, but just being connected is maybe 2% of battery a day. Negligible.

When I work at the local coffee shop I cannot SSH to my remote servers for work on their wifi, but if I connect to Tailscale and use my exit node at home I can. Lifesaver

I think this is mostly a Wireguard thing and not specifically a Tailscale thing. Wireguard does what they call "cryptokey routing" where if you prove you possess a key that the other peer knows, you get the traffic (subject to firewall, allowed IPs list, etc etc). Wireguard stores the most recent address:port that it heard from a particular cryptokey on, but it natively lets peers roam, as long as only one roams at a time.

My work guest WiFi network allows only IPv4 HTTPS on port 443 and their their own DNS. Everything else, including ICMP (ping) is blocked. Tailscale barely works as any persistant connection is dropped after 2-3 minutes.

Called this out and the security team said noone complains, that there is no use case and they do not want to deal with security risks.

And the ossification continues.

A TCP over websockets VPN would be fairly simple to write, or ask an AI to write for you

> IPv4 HTTPS on port 443

TCP or TCP and UDP?

SSTP can work if they don't look at the traffic too hard.

> Called this out and the security team said noone complains

Classic. And this probably works do every complaint. You need an irritated executive.

Reply on news.ycombinator.com