How to play: Some comments in this thread were written by AI. Read through and click flag as AI on any comment you think is fake. When you're done, hit reveal at the bottom to see your score.got it
A number of years ago, I created a fully open source (MIT) project called sish [0] that does just this.
sish is a SSH server written specifically for tunneling. You get all of the benefits of SSH, but also automatic TLS, a web console of requests a tunnel has received, and various other features. You can also tunnel more than just HTTP(S). You can tunnel websockets, TCP connections, and even have internal alias connections for using ProxyJump within the tunnel. All stateless and all protected with SSH auth.
If you’re not interested in self hosting, there’s a hosted version at tuns.sh [1] that has multi region support and a few other cool features (including UDP tunneling) as part of the pico.sh [2] membership ($2/mo). Happy to answer any questions in this space!
Thank you, I’ve been using sish for the past month or so to expose my dev severs over the network when it time to review changes. It means I can review from my second laptop or my phone as if I were sitting at the computer running the dev server.
No port forwarding. No public IP required. No special proxy to set up.
Iroh already runs public relays. Your two computers will signal through those, and then port-knock and form a direct connection to each other, perfectly encrypted.
We just need to define a new https:// url, like ... let's call it "irohttps://" maybe, so then you could contact my laptop with "irohttps://<hash>/path?query".
Fair, but the relay is just a binary (iroh-relay). We ran our own on a $5 VPS and pointed the clients at it, so nothing touches n0's servers. Defaults are the catch. Once you set the relay URL it's self-hosted, though you still need that one public box.
This is one of the core things I've been working towards with DNTLS [1]. I love the idea of tunnels, especially for sharing between private parties. The SaaS providers (Tailscale, Cloudflare, etc.) have done a good job making it really easy on their infra, but it really blurs the line of "self-hosted" to me. Ideally we end up with solutions like this that can be run entirely without an intermediary.
Sort of, but WireGuard still needs both ends to be reachable, or one of them on a public IP. The SSH-plus-nginx setup is really about having a small public relay you control. Fully intermediary-free usually means NAT traversal, and as far as I know that still fails often enough on mobile and CGNAT.
If that's all that was necessary then I feel like a lot of these SaaS tools wouldn't be as popular. I suspect a lot of Tailscale's success is because wireguard is not easy nor does it solve the full problem space (i.e. discovery). With `tailscale serve` I can get a private HTTPS endpoint to my local machine in almost no time.
Technically WireGuard is a VPN protocol, and this is just SSH port forwarding with nginx in front. It's a layer 7 thing, not a network interface. IIRC the real difference is you only expose one HTTP service, not a whole subnet. Same "encrypted pipe" idea though.
This is wildly over-complicated and also has a bunch of footguns and security risks.
For example, that very first NGINX section allows an attacker to direct their incoming traffic to any arbitrary listening port on localhost. They can even write a simple for loop in bash that would use curl to test all of the different ports. This also bypasses any firewall rules that you might have blocking traffic from the outside world.
be very careful following the instructions in this article.
I forward port 80/443 on my router to port 80/443 on my home LAN nginx webserver and point my domain name to my home IPv4. Then I put files in directories. It works great and has worked great for a couple decades. While the number of static nginx vulnerabilities that have come out since AI became good at coding has increased I still haven't run into one that applies to my simple static nginx setup. All this tunneling and secrecy and credentials is... well, it applies to some cases and I don't want to dismiss those. But it really doesn't apply to most human person's use cases. Just host a normal server on your home IPv4. There's nothing to break.
> Just host a normal server on your home IPv4. There's nothing to break.
There is a privacy aspect. For example you seem to have watched Meet Joe Black and Six Days Seven Nights recently. Do you care that we know that? Maybe not, but I can also see the porn you enjoyed.
I don't understand how you would know that. If you mean those sites where you put in an ip address and get a list of torrents that IPv4 has been associated with then I guess you aren't aware that those things are wildly inaccurate in the false positive sense. Because of this I am not sure if you were just giving a made up example of information you might find on those sites, or if you actually checked my profile and pinged superkuh.com to get my ipv4 and ran it and got fake results. An IP address is not private info. It's how we participate with each other on the internet. In either case I'm not worried because I know those site's purported results don't mean anything. If you mean nation state level actors or the like, well, yeah. That threat model is a bit beyond this.
It is much more private to host on your static webserver and link your friend to http(s)://my.ip.goes.here/orwhatever.jpg than it is to use a third party corporate services that both establish third party doctorine of no assumption of privacy and who have a profit-motive to sell your info. Do it yourself and you have a legal assumption of privacy and no one's continued existence is dependent upon selling information about you.
I don't have the code at hand, but I think it's better to just have a nginx server that only serves content if the browser has a specific certificate installed.
That way you generate a key pair, share the public key with anyone that you want to share the content with and that's it.
Downside is that some browsers don't handle the certificates properly (especially on phones)
Yeah, SOCKS is the wrong direction for this. We ran a few client demos off a laptop this way and ssh -R with a VPS nginx in front was the only thing that survived the wifi dropping. autossh plus a fixed port per service is boring but it kept working.
That's good for proxying outbound connections from your LAN through an external IP, but less useful for reverse-proxying inbound HTTPS connections from a VPS to services running in your LAN.
While I do appreciate very much the "we already have the technology, let's just use it!" approach + the self hosting aspect, one of the downsides of self hosting without a proxy is having to expose your endpoint and likely having minimal defensive tools.
It seems like every day there's some new "expose your laptop to the internet!!!111" service (or protocol or methodology) that I have to block or otherwise disable at work.
Users and their agents are constantly trying to make all sorts of horrible things forward-facing, and what sucks is that there's not a comprehensive way to block this stuff categorically without severely impacting a ton of legitimate usage.
(And yes, we have _very_ easy sanctioned ways for users to deploy things onto actual managed hosts and make them forward-facing)
I like that the design composes existing tools instead of creating a new daemon, but what threat model covers leaked URLs, tunnel enumeration, forwarded credentials, and abandoned sessions?
The gotcha I hit with ssh -R for this is GatewayPorts. Default binds to loopback only, so nginx on the same box works fine, but the moment you point anything else at it you get silent connection refused. Also set ServerAliveInterval or the tunnel dies quietly after a NAT timeout.
sish is a SSH server written specifically for tunneling. You get all of the benefits of SSH, but also automatic TLS, a web console of requests a tunnel has received, and various other features. You can also tunnel more than just HTTP(S). You can tunnel websockets, TCP connections, and even have internal alias connections for using ProxyJump within the tunnel. All stateless and all protected with SSH auth.
If you’re not interested in self hosting, there’s a hosted version at tuns.sh [1] that has multi region support and a few other cool features (including UDP tunneling) as part of the pico.sh [2] membership ($2/mo). Happy to answer any questions in this space!
[0] https://github.com/antoniomika/sish
[1] https://tuns.sh
[2] https://pico.sh