Choose a Runtime Mode

System settings → Mode combines network topology and routing into several runtime options. Before configuring it, answer three questions independently: how traffic reaches the device, how the gateway locates a service, and who may access that service.

Host routing is the default for web services. Reserve Path mode for existing path-based endpoints or applications that cannot be migrated; it is not recommended for new deployments.

The in-app tunnel steps on this page apply only to deployments whose admin interface actually provides FRP or Cloudflared. On Windows, 7999 listens on all interfaces by default and can serve as the origin for direct public access or a self-managed tunnel. Windows does not include FRP, Cloudflared, Direct mode authorization, or dynamic host firewall management. External reachability still depends on the active Windows Firewall profile, router or NAT, IPv6 firewall, and ISP policy. See Deploy on Windows (x86_64).

1. Choose a network topology

Situation	Admin option	External traffic path
The device has reachable public IPv4 or IPv6	Subdomain mode	The domain points directly to the gateway
The device has no public inbound endpoint	Reverse proxy mode	FRP or Cloudflared carries requests to the gateway
Users must access original ports after signing in	Direct mode (not recommended)	Users sign in through the gateway, then the firewall allows their source IP

Whether you have a public IP determines the ingress topology; it does not force web services to use path routing. Reverse proxy mode also supports Subdomain mapping.

2. Choose a routing method

Host routing: the default for web services

In Subdomain mode with direct public ingress, or in Reverse proxy mode → Subdomain mapping, requests are dispatched by Host:

auth.example.com  -> authentication service
nas.example.com   -> http://127.0.0.1:5666
alist.example.com -> http://127.0.0.1:5244

Each service keeps its root URL, which makes static assets, WebSockets, and callback URLs more likely to work without special handling. Both topologies share the same Subdomain mapping configuration, but their capabilities differ:

• Subdomain mode with direct public ingress can integrate with the host firewall, Smart Connect, and TCP/UDP stream proxying.
• Subdomain mapping in Reverse proxy mode can use FRP or Cloudflared, but does not provide Smart Connect or TCP/UDP stream proxying.

See Public IP Access with Subdomain Routing for setup.

Path routing: a compatibility option

Reverse proxy mode → Path mode dispatches services by URL prefix under a single Host:

https://example.com/alist -> http://127.0.0.1:5244
https://example.com/fnos  -> http://127.0.0.1:5666

Some applications require prefix stripping, HTML rewriting, or root-path compatibility and may still break because of absolute paths, Service Workers, or callback URLs. Existing path-based deployments can remain in place, but new services should use Host routing. See NAT Traversal with Subdomain Routing for setup.

TCP/UDP streams and original ports

• Streams listens on additional TCP/UDP ports for non-HTTP services such as SSH, databases, and DNS. It is available only in Subdomain mode with direct public ingress. See TCP/UDP Stream Proxying.
• Direct mode does not proxy a service. After sign-in, it adds the current source IP to the allowlist so the user can connect to original ports such as 5666 or 22. See Direct Access on Original Ports.

3. Choose an access policy

The current Host editor provides a Require sign-in switch. Off means public access; on means login-first behavior:

Setting	Result for an internet request	Best suited for
Turn off Require sign-in on a current login-first mapping	Proxy directly to the upstream	A service intentionally published without fn-knock authentication
Turn on Require sign-in	If there is no valid source-IP authorization or session, redirect to auth.example.com, then return to the original Host after sign-in	Most private web services

The backend still recognizes strict_whitelist in legacy configurations. Such a rule may remain non-public even after Require sign-in is turned off, because access is decided by valid source authorization records created manually or after sign-in. A browser session cookie cannot replace the source requirement. The current UI has no control for creating or switching to a strict-allowlist rule. To leave one, record the complete mapping and recreate it through the current UI. On new mappings, a manual source authorization can grant access independently. Automatic IP authorization normally lets the same source continue, but it cannot override a denial carried by a service scope. IP Allowlist is not a strict source-restriction switch. To narrow ingress before a request reaches a mapping, use Gateway Visibility or an external network-layer rule.

Sign-in credentials can also restrict the permitted service scope. A Host outside that scope is denied even when the browser already has a valid session.

Skip Basic Auth is not an fn-knock access policy. It injects a username and password into requests sent upstream so they can pass the target service's own Basic Auth. Visitors cannot use it to sign in to fn-knock.

Local-source exemption

When the gateway identifies a loopback, private, or link-local source, authentication returns local_exempt; both normal sign-in and strict-allowlist checks allow the request. This behavior treats the LAN as part of the trust boundary:

• Do not use a LAN test to decide whether internet-facing authentication is working.
• Tunnels and edge proxies must pass the real client IP correctly. Otherwise, all visitors may appear to come from one proxy address or may even be classified as local.
• Verify sign-in, allowlist behavior, and post-sign-out access over a mobile network.

Recommended combinations

Scenario	Network topology	Routing	Access policy
Web services with a public IP and domain	Subdomain mode	Host	Enable Require sign-in; add Gateway Visibility if ingress must be narrowed
No public IP, using Cloudflared	Reverse proxy mode → Subdomain mapping	Host	Enable Require sign-in
No public IP, using FRP	Reverse proxy mode → Subdomain mapping	Host	Enable Require sign-in; confirm that PROXY Protocol preserves the source IP
Legacy path-based deployment being migrated	Reverse proxy mode → Path mode	Path	Decide whether to require sign-in on each mapping
Many original ports must remain accessible	Direct mode (not recommended)	Original ports	Allow the source IP after sign-in

What a mode switch preserves or stops

• After switching from Path mode to either form of subdomain routing, path mappings are hidden. When entering Subdomain mode with direct public ingress, the UI asks whether to remove path rules.
• Both Subdomain mapping and Path mode under Reverse proxy mode can use FRP or Cloudflared. When leaving Reverse proxy mode, the system attempts to stop any running tunnels.
• After leaving Subdomain mode with direct public ingress, TCP/UDP stream proxying is disabled and its listeners stop, but saved rules are preserved. Re-enter the mode and enable the feature to restore them.
• Direct mode depends on the host firewall. It is unavailable on Docker, Synology DSM 7 SPK, Windows, and any deployment without host-management capability.
• Docker does not write host firewall rules and does not support Smart Connect. OpenWrt supports Smart Connect but not SSH security, the web terminal, or in-app FPK updates. Smart Connect on OpenWrt requires an existing enabled dnsmasq whose configuration includes /etc/dnsmasq.d/; the UI cannot install its dependencies with apt-get.
• Windows does not support Direct mode authorization, in-app host firewall management, Smart Connect, built-in FRP / Cloudflared, the web terminal, or SSH security. The web update page also cannot install Windows updates.
• The Synology DSM 7 SPK includes FRP and Cloudflared, but does not support Direct mode authorization, host firewall management, Smart Connect, the web terminal, SSH security, or updates from the web UI.

After switching modes, test from a real external network and verify the endpoint, Host, client IP, authorization type, and upstream target. Simply confirming that a page opens is not enough.