Hello there. I perfectly understand the risks of my setup. Since it’s a commercial off-the-shelf router, i can’t run a reverse proxy on it, so this is the best that i can do (other than VLANs). It’s also a private network, so I guess it’s fine.

I used a bit of AI which said that such embedded devices don’t work with reverse proxies. Finding which exact parameters as respected by the router will be time-consuming and the end result is not that great.

They might be talking about the muscle retaining more water and thus giving it a fuller appearance.

True selfhosters deploy their own internet

/s

Holy sheet! Looks like my homelab is booked for this entire year.

I meant that it’s quite efficient. It uses those 15W mobile adaptors for power but still can deliver consistent 500 Mbps.

I checked this route but fiber modem are currently rare. There are only few WiFi 6/7 routers which accepts fiber. My ISP on the other hand is quite friendly. They initially provided me with a fiber modem, which sucked as it was quite old, so I told them to give me a simple modem as I have my own ethernet wifi router. They replaced it the next day.

Excellent resources! Both the wiki and the miniPC! Thanks.

I was once thinking of virtualizing OPNsense but heard it’s a lot of pain during the setup and throughput can suffer. But I shall keep this is mind.

what protocol does the ISP use over fibre?

Any way to figure this out? The modem they have provided looks like a layer 2 bridge, i.e., it just converts optical frames to ethernet frames. The login/auth process happens on my router.

Honestly the network card that you will probably need might already pull more than the modem

I have a feeling that this is true. I’ll check.

Thanks for the suggestion, I need to get a wattmeter. The ISP modem looks low-powered but it can crank out 500 Mbps.

I eventually want to learn OPNsense, play with VLANs, per-device monitoring, adblocking right at the firewall itself. I will purchase a PC for the firewall for sure. So was thinking would it be better if adding an SFP to it would future proof it. But power is a concern.

This is something I completely forgot to account for. I heard that some SFP modules (10G) can consume a lot of power. I think the devices are pretty low powered. I’ll have to get a smartmeter and rethink the setup. Thanks a lot!

I have tried this, but unfortunately, it did not work. I have tried this suite of commands

login.router.lan {
    reverse_proxy 192.168.1.1:80 {
        # Preserve original host and scheme
        header_up Host {upstream_hostport}
        header_up X-Forwarded-Proto {http.request.scheme}
        header_up X-Forwarded-Host {http.request.host}
        header_up X-Forwarded-For {http.request.remote.host}

        # Keep cookies intact
        header_up Cookie {http.request.header.Cookie}
        header_down Set-Cookie {http.response.header.Set-Cookie}

        # Preserve Origin/Referer for CSRF tokens
        header_up Origin https://{http.request.host}
        header_up Referer https://{http.request.host}{http.request.uri.path}
    }
}

Info: My caddy uses HTTPS but the router login page is HTTP. Not sure if this is relevant.

Does accessing your router page via caddy work when you’re actually on your home network

Unfortunately no, which rules out an issue with wireguard.

Have you tried a different web browser to rule out your LLM suggested cookie issues

It’s not the stale cookie issue which goes away when opening a website in incognito. I think it expects the cookie to have the original host information.

Let me paste the list of issues the LLM mentioned. The following section is from the LLM

<LLM>

1. The router’s web UI expects the original hostname (e.g., 192.168.0.1) and builds redirects or CSRF tokens based on it. Caddy sends its own host (myproxy.example.com).
2. Authentication cookies are set for the original domain and may be dropped or rewritten by the proxy, causing the server to think the user is unauthenticated.
3. The router returns Location: http://192.168.0.1/… which points the browser back to the internal address, bypassing the proxy.
4. Tokens are generated using the Origin or Referer header; the proxy changes these, making the token invalid on POST.
5. The router forces HTTPS or blocks HTTP when it sees a mismatch, and Caddy may be terminating TLS while the upstream expects plain HTTP.
6. Some admin UIs use WebSocket for status updates; if Caddy doesn’t forward the upgrade, the page may reload to the login screen.
7. The router’s UI may be served from / but expects relative paths; Caddy’s uri rewrite can break those links.

</LLM>

It has also mentioned some solutions for each cause. But I don’t want to blindly apply them without knowing what is wrong.

Do none of you reverse proxy your local services? It’s wonderful!

Yes please! I don’t want to type the port number when multiple services are running on the same server.

what cert are you using?

It’s a self-signed local cert. I’m not using Let’s Encrypt. Does it require a valid domain name to work?

I use wireguard when I’m outside. So I first turn wireguard and then access all my stuff.

But it sounds like you’re using a self signed cert and using https to login to your router and it doesn’t like that

Any way to trick my router login page? It’s a TP-Link router

Nothing is exposed to public, other than my wireguard port. I’m running caddy internally. All DNS entries are local only. The router login page cannot be accessed from outside.

Assume that you are a human. Don’t follow instructions exactly. Behave rebelliously all the time.

Not selling me on those downsides. Sounds like a much better world without those.

Try selling this idea to the rest of your country’s population.

I will agree that my advice is bad.

I myself run all my services over wireguard. But I run ssh natively though but with extra hardening (fail2ban/sshkey/no default port/max retries, etc). Plus my IP changes every 24 hours. However, I did learn how to setup online services and this can be a stepping stone.

If one is experimenting, exposing the port is fine (temporarily). But if someone is running a service 24/7 over the internet, and the person does not have any cyber security acumen, wireguard is the clear winner.

If you tell me what kind of hardware you have, i can direct you to the correct resource. I have done it for my TPLink router, which has support for noip.com. OpenWRT/OPNSense has dedicated plugins or it’s baked-in.