this post was submitted on 27 Mar 2024
34 points (88.6% liked)

Selfhosted

40708 readers
702 users here now

A place to share alternatives to popular online services that can be self-hosted without giving up privacy or locking you into a service you don't control.

Rules:

  1. Be civil: we're here to support and learn from one another. Insults won't be tolerated. Flame wars are frowned upon.

  2. No spam posting.

  3. Posts have to be centered around self-hosting. There are other communities for discussing hardware or home computing. If it's not obvious why your post topic revolves around selfhosting, please include details to make it clear.

  4. Don't duplicate the full text of your blog or github here. Just post the link for folks to click.

  5. Submission headline should match the article title (don’t cherry-pick information from the title to fit your agenda).

  6. No trolling.

Resources:

Any issues on the community? Report it using the report flag.

Questions? DM the mods!

founded 2 years ago
MODERATORS
[email protected]
[email protected]
[email protected]
[email protected]
[email protected]
[email protected]
34
How can I bypass CGNAT by using a VPS with a public IPv4 address? (lemmy.zip)
submitted 9 months ago by [email protected] to c/[email protected]
53 comments fedilink hide all child comments
 

I want to move away from Cloudflare tunnels, so I rented a cheap VPS from Hetzner and tried to follow this guide. Unfortunately, the WireGuard setup didn't work. I'm trying to forward all traffic from the VPS to my homeserver and vice versa. Are there any other ways to solve this issue?

VPS Info:

OS: Debian 12

Architecture: ARM64 / aarch64

RAM: 4 GB

Traffic: 20 TB

top 50 comments
sorted by: hot top controversial new old
[–] [email protected] 21 points 9 months ago* (last edited 9 months ago) (2 children)

You don't want to forward all traffic. You can do SNAT port forwards across the VPN, but that requires the clients in your LAN to use the VPS as their gateway (I do this for a few services that I can't run through a proxy; its clunky but works well).

Typically, you'll want to proxy requests to your services rather than forwarding traffic.

  1. Setup Wireguard or OpenVPN on the VPS as a server VPN. Allow whatever listener port in the firewall (I use ufw on Debian, but you can use iptables if you want)
  2. Install HAProxy or Nginx (or Nginx Proxy Manager) on the VPS to act as your frotnend. Those will listen on ports 80/443 and proxy requests to your backend servers. They'll also be responsible for SSL termination, and your public-facing certs will be set there.
  3. Point your DNS records for your services to the VPS's public IPv4
  4. On your LAN, configure your router to connect to the VPS as a VPN client and route into your LAN from the VPN subnet -or- install the VPN client (WG/OVPN) on each host
  5. In your VPS's reverse proxy (HAProxy, etc), set the backend server address and port to the VPN address of your host

I've done this since ~2013 (before CF tunnels were even a product) and has worked great.

My original use case was to setup direct connectivity between a Raspberry PI with a 3G dongle with a server a home on satellite internet. Both ends of that were behind CG-NAT, so this was the solution I came up with.

[–] [email protected] 3 points 9 months ago (1 children)

Out of curiosity, why not a simple reverse proxy on the VPS (that only adds client real IP to headers), tunneled to a full reverse proxy on the home server (that does host routing and everything else) through a SSH tunnel?

[–] [email protected] 1 points 9 months ago (1 children)

How would that kind of a setup look like?

[–] [email protected] 4 points 9 months ago* (last edited 9 months ago) (1 children)

Variant 1:

  • SSH tunnel established outgoing from home server to VPS_PUBLIC_IP:22, which makes an encrypted tunnel that "forwards" traffic from VPS_PUBLIC_IP:443 to HOME_LOCALHOST:443.
  • Full reverse proxy listening on HOME_LOCALHOST:443 and does everything (TLS termination, host routing, 3rd-party auth etc.)
  • Instead of running home proxy on the host you can ofc run it inside a container, just need to also run the ssh tunnel from inside that container.

Pro: very secure, VPS doesn't store any sensitive data (no TLS certificates, only a SSH public key) and the client connections pass through the VPS double-encrypted (TLS between client browser and home proxy, wrapped inside SSH).

Con: you don't get the client's IP. When the home apps receive the connections they appear to originate at the home end of the SSH tunnel, which is a private interface on the home server.

Variant 2 (in case you need client IPs):

  • SSH tunnel established same way as variant 1 but listens on VPS_LOCALHOST:PORT.
  • Simple reverse proxy on VPS_PUBLIC_IP:443. It terminates the TLS connections (decrypts them) using each domain's certificate. Adds the client IP to the HTTP headers. Forwards the connection into VPS_LOCALHOST:PORT which sends it to the home proxy.
  • Full reverse proxy at home set up same way as variant 1 except you can listen to 80 and not do any TLS termination because it's redundant at this point – the connection has already been decrypted and will arrive wrapped inside SSH.

Pro: by decrypting the TLS connection the simple proxy can add the client's IP to the HTTP headers, making it available to logs and apps at home.

Con: the VPS needs to store the TLS certificates for all the domains you're serving, you need to copy fresh certificates to the VPS whenever they expire, and the unencrypted connections are available on the VPS between the exit from TLS and the entry into the SSH tunnel.

Edit: Variant 3? proxy protocol

I've never tried this but apparently there's a so called proxy_protocol that can be used to attach information such as client IP to TLS connections without terminating them.

You would still need a VPS proxy and a home proxy like in variant 2, and they both need to support proxy protocol.

The frontend (VPS) proxy would forward connections in stream mode and use proxy protocol to add client info on the outside.

The backend (home) proxy would terminate TLS and do host routing etc. but also it can unpack client IP from the proxy protocol and place it in HTTP headers for apps and logs.

Pro: It's basically the best of both variant 1 and 2. TLS connections don't need to be terminated half-way, but you still get client IPs.

Please note that it's up to you to weigh the pros and cons of having the client IPs or not. In some circumstances it may actually be a feature to not log client IPs, for example If you expect you might be compelled to provide logs to someone.

[–] [email protected] 1 points 9 months ago (1 children)

Very interesting... How do I get started?

[–] [email protected] 1 points 9 months ago (1 children)

The SSH tunnel is just one command, but you may want to use autossh to restart it if it fails.

If you choose variant 2 you will need to configure a pass-through reverse proxy on the VPS that does TLS termination (uses correct certificates for each domain on 443). Look into nginx, caddy, traefik or haproxy.

For the full home proxy you will once again need a proxy but you'll additionally need to do host routing to direct each (sub)domain to the correct app. You'll probably want to use the same proxy as above to avoid learning two different proxies.

I would recommend either caddy (both) or nginx (vps) + nginx proxy manager (home) if you're a beginner.

[–] [email protected] 1 points 9 months ago (1 children)

How do I make the SSH tunnel forward traffic? It can't be as easy as just running ssh user@SERVER_IP in the terminal.

(I only need variant 1 btw)

[–] [email protected] 1 points 9 months ago* (last edited 9 months ago) (2 children)

You also add the -R parameter:

ssh -R SERVER_IP:443:HOME_PROXY_IP:HOME_PROXY_PORT user@SERVER_IP

https://linuxize.com/post/how-to-setup-ssh-tunneling/ (you want the "remote port forwarding"). ssh -R, -L and -D options are magical, more people should learn about them.

You may also need to open access to port 443 on the VPS. How you do that depends on the VPS service, check their documentation.

load more comments (2 replies)
[–] [email protected] 1 points 9 months ago (1 children)

The biggest obstacle for me is the connection between the VPS and my homeserver. I have tried this today and I tried pinging 10.0.0.2 (the homeserver IP via WireGuard) and get this as a result:

PING 10.0.0.2 (10.0.0.2) 56(84) bytes of data.
From 10.0.0.1 icmp_seq=1 Destination Host Unreachable
ping: sendmsg: Destination address required
From 10.0.0.1 icmp_seq=2 Destination Host Unreachable
ping: sendmsg: Destination address required
^C
***
10.0.0.2 ping statistics
***
2 packets transmitted, 0 received, +2 errors, 100% packet loss, time 1019ms

Not sure why though.

[–] [email protected] 2 points 9 months ago* (last edited 9 months ago) (1 children)

Can you post your WG config (masking the public IPs and private key if necessary)?

With wireguard, the allowed-ips setting is basically the routing table for it.

Also, you don't want to set the endpoint address (on the VPS) for your homeserver peer since it's behind NAT. You'll only want to set that on the 'client' side. Since you're behind NAT, you'll also want to set the persistent keepalive in the client peer so the tunnel remains open.

[–] [email protected] 2 points 9 months ago (1 children)

Hi, thank you so much for trying to help me, I really appreciate it!

VPS wg0.conf:

[Interface]
Address = 10.0.0.1/24
ListenPort = 51820
PrivateKey = REDACTED

PostUp = iptables -t nat -A PREROUTING -p tcp -i eth0 '!' --dport 22 -j DNAT --to-destination 10.0.0.2; iptables -t nat -A POSTROUTING -o eth0 -j SNAT --to-source SERVER_IP
PostUp = iptables -t nat -A PREROUTING -p udp -i eth0 '!' --dport 55107 -j DNAT --to-destination 10.0.0.2;

PostDown = iptables -t nat -D PREROUTING -p tcp -i eth0 '!' --dport 22 -j DNAT --to-destination 10.0.0.2; iptables -t nat -D POSTROUTING -o eth0 -j SNAT --to-source SERVER_IP
PostDown = iptables -t nat -D PREROUTING -p udp -i eth0 '!' --dport 55107 -j DNAT --to-destination 10.0.0.2;

[Peer]
PublicKey = REDACTED
AllowedIPs = 10.0.0.2/32

Homeserver wg0.conf:

[Interface]
Address = 10.0.0.2/24
PrivateKey = REDACTED
 
[Peer]
PublicKey = REDACTED
AllowedIPs = 0.0.0.0/0
PersistentKeepalive = 25
Endpoint = SERVER_IP:51820

(REDACTED would've been the public / private keys, SERVER_IP would've been the VPS IP.)

[–] [email protected] 3 points 9 months ago (1 children)

On the surface, that looks like it should work (assuming all the keys are correct and 51820/udp is open to the world on your VPS).

Can you ping the VPS's WG IP from your homeserver and get a response? If so, try pinging back from the VPS after that.

Until you get the bidirectional traffic going, you might try pulling out the iptables rules from your wireguard script and bringing everything back up clean.

[–] [email protected] 1 points 9 months ago (1 children)

I do not get a response when pinging the VPS's WG IP from my homeserver. It might have something to do with the firewall that my VPS provider (Hetzner) is using. I've now allowed the port 51820 on UDP and TCP and it's still the same as before... This is weird.

[–] [email protected] 2 points 9 months ago (1 children)

I'm not familiar with Hetzner, but I know people use them; haven't heard any kinds of blocks for WG traffic (though I've read they do block outbound SMTP).

Maybe double-check your public and private WG keys on both ends. If the keys aren't right, it doesn't give you any kind of error; the traffic is just silently dropped if it doesn't decrypt.

[–] [email protected] 2 points 9 months ago (1 children)

Hmm, the keys do match on the two different machines. I have no idea why this doesn't work...

[–] [email protected] 2 points 9 months ago* (last edited 9 months ago) (1 children)

Dumb question: you're starting wireguard right? lol

In most distros, it's systemctl start wg-quick@wg0 where wg0 is the name of the config file in /etc/wireguard

If so, then maybe double/triple check any firewalls / iptables rules. My VPS providers don't have any kind of firewall in front of the VM, but I'm not sure about Hetzner.

Maybe try stopping wireguard, starting a netcat listener on 51820 UDP and seeing if you can send to it from your homelab. This will validate that the UDP port is open and your lab can make the connection.

### VPS
user@vps:  nc -l -u VPS_PUBLIC_IP 51820

### Homelab
user@home:  echo "Testing" | nc -u VPS_PUBLIC_IP 51820

### If successful, VPS should show:
user@vps:  nc -l -u VPS_PUBLIC_IP 51820
Testing

I do know this is possible as I've made it work with CG-NAT on both ends (each end was a client and routed through the VPS).

[–] [email protected] 2 points 9 months ago (1 children)

The command you provided for the VPS returns UDP listen needs -p arg, so I just added -p right before the port number and then it worked. Running the homelab command returns no port[s] to connect to... Not good.

[–] [email protected] 3 points 9 months ago (1 children)

At least that points you to the problem: firewall somewhere.

Try a different port with your netcat test, perhaps? 51820 is the well-known WG port. Can't imagine they'd intentionally block it, but you never know.

Maybe Hetzner support can offer more guidance? Again, I'm not sure what or how they do network traffic before it gets to the VM. On all of mine, it's just a raw gateway and up to me to handle all port blocking.

If you figure that part out and are still stuck on the WG part, just shoot me a reply.

[–] [email protected] 2 points 9 months ago (1 children)

I tried to open the port 22 on UDP (yeah, I am getting pretty desperate over here...) and still get the message no port[s] to connect to... Someone else on this post commented that I should stop using iptables for opening ports and start using something else as a firewall. Should I try this approach?

[–] [email protected] 3 points 9 months ago (1 children)

Yeah, might be worth a shot. iptables is nice, but very verbose and somewhat obtuse.

I'd just clear out iptables completely and use ufw. Should be in Debian's package manager.

Here's a cheat sheet: https://www.digitalocean.com/community/tutorials/ufw-essentials-common-firewall-rules-and-commands

[–] [email protected] 2 points 9 months ago (1 children)

What do you mean with "clear out iptables completely"? Should I remove the iptables package with sudo apt remove iptables?

[–] [email protected] 3 points 9 months ago (1 children)

I believe iptables --flush should clear out any entries you've made. You can also reboot and clear them (unless you've got scripts bound to your interface up/down config that adds rules).

Basically just need to get any custom iptables rules you made out of there and then re-implement any FW rules with ufw

You can still use iptables alongside UFW, but I only use those for more complex things like port forwarding, masquerading, etc.

[–] [email protected] 2 points 9 months ago (1 children)

Alright, I switched to ufw and... it's still not working. sigh

Should we just try something completely different? WireGuard doesn't seem to be working on my VPS. Someone in the comments mentioned tunneling via SSH, sounds interesting.

[–] [email protected] 3 points 9 months ago* (last edited 9 months ago) (1 children)

That would work, but I've noticed performance isn't as good as a UDP VPN that uses the kernel's tun module. OpenVPN is also an option, but it's a LOT more involved to configure (I used to run it before Wireguard existed).

The oddest part is you can't get a netcat message through. That implies firewall somewhere.

What is the output of your ufw status ?

[–] [email protected] 2 points 9 months ago (1 children)

I've added some different ports for the future, but this is my ufw status:

Status: active

To                         Action      From
--                         ------      ----
OpenSSH                    ALLOW       Anywhere                  
51820                      ALLOW       Anywhere                  
2333                       ALLOW       Anywhere                  
80                         ALLOW       Anywhere                  
81                         ALLOW       Anywhere                  
443                        ALLOW       Anywhere                  
80/tcp                     ALLOW       Anywhere                  
OpenSSH (v6)               ALLOW       Anywhere (v6)             
51820 (v6)                 ALLOW       Anywhere (v6)             
2333 (v6)                  ALLOW       Anywhere (v6)             
80 (v6)                    ALLOW       Anywhere (v6)             
81 (v6)                    ALLOW       Anywhere (v6)             
443 (v6)                   ALLOW       Anywhere (v6)             
80/tcp (v6)                ALLOW       Anywhere (v6)
[–] [email protected] 1 points 9 months ago (1 children)

I can't recall if ufw opens both TCP and UDP or just TCP by default.

Try explicitly allowing 51820/udp with ufw allow 51820/udp

[–] [email protected] 2 points 9 months ago* (last edited 9 months ago)

I've added the firewall rule and it still says no port[s] to connect to whenever I run echo "Testing" | nc -u SERVER_IP -p 51820. I feel like you're trying to stay on a sinking ship, so I would suggest to try another method to see if we even can get the whole "bypass CGNAT with a VPS" thing to work at all.

Update: I've tried setting up SSH tunneling instead and it STILL doesn't work. I contacted Hetzner support about this issue and I'm hoping that they can resolve the firewall issues that I'm having.

[–] [email protected] 8 points 9 months ago (2 children)

I use rathole https://github.com/rapiz1/rathole
Its been solid.

[–] [email protected] 5 points 9 months ago* (last edited 9 months ago) (1 children)

I use this too, and it should be noted that this does not require wireguard or any VPN solution. Rathole can be served publicly, allowing a machine behind a NAT or firewall to connect.

[–] [email protected] 1 points 9 months ago (1 children)

I like that its really simple and obvious, with a good confif file structure.
Server forwards a port to a client.
Client forwards that to an ip:port.

If you need to know the real IP, its up to you to run reverse-proxies that support PROXY TCP headers or insert x-forward-for, or whatever.
Rathole does its thing, only its thing, and does it well.

load more comments (1 replies)
[–] [email protected] 3 points 9 months ago

This looks really interesting. I'll check it these days.

[–] [email protected] 6 points 9 months ago (1 children)

I managed this by using tailscale, with a kind of weird setup I think, but it just works.

I have tailscale on the VPS and my local server, let's say its tailscale name is potatoserver

Then with Caddy on the VPS i have something like:

mywebsite.com { reverse_proxy potatoserver:port }

And so mywebsite.com is accessible on the clearnet through the VPS

Though given you're getting rid of cloudflare tunnles I don't know if you'd want to get into Tailscale. There's Headscale too but I haven't worked with it so I can't comment

[–] [email protected] 3 points 9 months ago

I only use headscale. It just works and does not complain.

[–] [email protected] 3 points 9 months ago

Not sure exactly how good this would work for your use case of all traffic, but I use autossh and ssh reverse tunneling to forward a few local ports/services from my local machine to my VPS, where I can then proxy those ports in nginx or apache on the VPS. It might take a bit of extra configuration to go this route, but it's been reliable for years for me. Wireguard is probably the "newer, right way" to do what I'm doing, but personally I find using ssh tunnels a bit simpler to wrap my head around and manage.

Technically wireguard would have a touch less latency, but most of the latency will be due to the round trip distance between you and your VPS and the difference in protocols is comparatively negligible.

[–] [email protected] 3 points 8 months ago

This one works , I myself have done it cause my shitty isp needs a huge payment for a static public ip. A 5$ VPS was much cheaper . Server behind NAT I can help if you got any doubts

[–] [email protected] 2 points 9 months ago* (last edited 8 months ago)

Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I've seen in this thread:

Fewer Letters More Letters
CF CloudFlare
CGNAT Carrier-Grade NAT
DNS Domain Name Service/System
HTTP Hypertext Transfer Protocol, the Web
IP Internet Protocol
NAT Network Address Translation
Plex Brand of media server package
SSH Secure Shell for remote terminal access
SSL Secure Sockets Layer, for transparent encryption
TCP Transmission Control Protocol, most often over IP
TLS Transport Layer Security, supersedes SSL
UDP User Datagram Protocol, for real-time communications
VPN Virtual Private Network
VPS Virtual Private Server (opposed to shared hosting)
nginx Popular HTTP server

[Thread #635 for this sub, first seen 27th Mar 2024, 18:15] [FAQ] [Full list] [Contact] [Source code]

[–] [email protected] 2 points 9 months ago (1 children)

Do you have a working Wireguard connection? If so you can setup two reverse proxies.

[–] [email protected] 1 points 9 months ago (1 children)

Not really, pinging my homeserver via the VPS returns:

PING 10.0.0.2 (10.0.0.2) 56(84) bytes of data.
From 10.0.0.1 icmp_seq=1 Destination Host Unreachable
ping: sendmsg: Destination address required
From 10.0.0.1 icmp_seq=2 Destination Host Unreachable
ping: sendmsg: Destination address required
^C
***
10.0.0.2 ping statistics
***
2 packets transmitted, 0 received, +2 errors, 100% packet loss, time 1019ms
[–] [email protected] 2 points 9 months ago (1 children)

Forget iptables. You have a broken Wireguard setup. Did you verify that you have the proper keys and that Wireguard is allowed though the firewall?

[–] [email protected] 1 points 9 months ago (2 children)

I have no idea how to properly manage the firewall with Hetzner. I've opened the ports on the Hetzner management page and I ran several iptables commands to allow traffic from those ports. Still doesn't work. This is weird!

[–] [email protected] 1 points 9 months ago (1 children)

What firewall are you using in the VPS? It will likely be firewalld or ufw

[–] [email protected] 1 points 9 months ago (1 children)

Does iptables count as a firewall? You said that I should "forget" iptables. Is it that bad? It came preinstalled on the VPS. Should I switch? And if so, how?

[–] [email protected] 1 points 9 months ago (1 children)

Iptables is the low level mechanism that handles network routing. Firewall software just takes it up a layer so you can manage it without crazy long commands

[–] [email protected] 1 points 9 months ago* (last edited 9 months ago) (1 children)

Alright, sounds good. What firewall are you recommending me to use? I would like to use a firewall that's easy to manage.

Edit: I went with ufw.

[–] [email protected] 1 points 9 months ago

Sorry I didn't mean you should go out and create a firewall. (Although that's a good idea) I mean there is a issue with either your connectivity or your Wireguard config.

load more comments (1 replies)
[–] [email protected] 2 points 9 months ago* (last edited 9 months ago)

I had a similar problem like you. My provider only gives me a public IPv6 but no public IPv4. Im using a VPS with an IPv4 with jool to set up SIIT-DC https://nicmx.github.io/Jool/en/siit-dc.html

This converts all IPv4 traffic arriving at the VPS to IPv6 traffic which gets then directly routed to my homeserver.

Not sure if this setup would work for you. This is not a viable solution if you are completly behind a CGNAT without even a public IPv6.

Pro:

  • Works without any sensitive Data on the VPS (SSl certificates/passwords...)
  • Works for all IP based traffic (TCP,UDP,ICMP)
  • The original source IPv4 can be restored by the homeserver Contra:
  • AFAIK you cannot choose to only forward some TCP ports. Everything gets redirected.
  • You cannot access the VPS via IPv4 anymore since it gets redirected to your homeserver. (I only access my VPS via IPv6)
  • No (additional) encryption. (This is no problem for me since all my traffic is already e2e encrypted)
[–] [email protected] 1 points 9 months ago

Here https://wiki.gardiol.org/doku.php?id=router:ssh_tunnel

[–] [email protected] 1 points 9 months ago

Yeah you probably need a second IP address on the VPS though.

I open a wire guard tunnel from home to the VPS and then tunnel an Nginx ingress down the VPS.

[–] [email protected] 1 points 9 months ago* (last edited 9 months ago)

I did this. Works flawlessly for half year now. I have x86 thin client at home running all my stuff, it creates tunnel to my VPS (I use Free tier Oracle VPS - yes, it is a shit company, I know, no need to let me know again in the comments). Works like a charm. This GitHub repo has automated installer for Oracle, Amazon,... https://github.com/mochman/Bypass_CGNAT/wiki/Oracle-Cloud-(Automatic-Installer-Script) - it installs and configures Wireguard on both server (VPS) and client (your home machine).

load more comments
view more: next ›