My ISP uses CG NAT which is stopping me from reaching my internal network, so I’m thinking about using Tailscale to allow me to connect to my server and hence to my internal network.
But I’m not very comfortable giving 100% access to Tailscale to my internal network, so I was thinking if I could limit it only to what it requires to connect to the internet and to a wireguard service running in the same container. This would in turn connect to a wireguard server in the container’s host and provide me with full network access.
I know, as long as they have a service running in the server, even if inside a container, they can always be able to access the host. But even do I would feel safer if at least tried to contain it.
Does anyone know if this is possible? And can it be done through Docker Compose?
You must log in or register to comment.
Depending on what services you want to give access with, I have had great luck with an ultra cheap VPS
https://lowendbox.com/blog/1-vps-1-usd-vps-per-month/
Then I host my edge services on a container and use an ssh tunnel to the remote host which gives me an ipv4 and any port forward that I want.
For example I have my reverse proxy inside my network and my VPN server then I use a command like:
ssh -R 8080:localhost:80 public.example.com
Which would forward publicip:8080 to localhost:80
Read more here: https://www.ssh.com/academy/ssh/tunneling-example.
I use autossh to keep the tunnel alive at all times.
https://www.harding.motd.ca/autossh/
This is an ultra cheap way to get any ports you want and self host the whole thing. The remote VPS also doesn’t get any extra access to your local network and doesn’t initiate the connection so it doesn’t have credentials for your local network
There is no need, and you’re defeating the point of using tailscale. Use headscale if you cannot summit your anxiety around trusting tailscale.
Note that using headscale transfers the anxiety of contril from tailscale as a company to whatever vps you would be hosting the headscale on
Yes, you can run Tailscale in a container. You could create a second VLAN, attach it to your hosts interface, add a macvlan docker interface to the container and put it directly on your network.
If you have concerns about the software running on your host I would recommend getting a dedicated piece of hardware instead (rpi, zimaboard, etc).
How paranoid are you wanting to be? You can either go Headscale, or Tailnet Lock (my preference) to give your self some peace of mind. It completely depends on your threat model, which you didn’t mention.
There’s no point doing anything fancy like that - wireguard over Tailscale is pretty pointless, as Tailscale is literally wireguard with NAT traversal and authentication bolted on. Unless you enable subnetting, it can’t get more secure than that.
And even if you do enable subnetting (which you might wanna do if you need access to absolutely everything), you can use Tailscale ACLs to keep tighter control - say, from specific (tagged) devices.
Just setup wireguard on your server, add masquerading and ip forwarding. That single wireguard in, will give you full access to your lan
He can’t open ports because of the ISPs setup.
Edit NVM, read up on it. Seems like you have to run PCP protocal on IP4 to bypass thr CGNAT issues.
You can use any open port and port forward at the router, or is CG NAT only 80
