Tailscale Exit Node vs Subnet Router: Which One You Need and How to Set Each Up
Priya Sen
November 19, 2026
You have Tailscale on your laptop and phone, and now you want more out of it. Maybe you want to reach the printer, NAS admin page or IP camera at home that cannot run Tailscale. Maybe you want your browsing on hotel Wi-Fi to come out of your home connection. Tailscale has two features with similar setup steps and similar-sounding names for these jobs, exit nodes and subnet routers, and it is easy to enable the wrong one and wonder why nothing changed.
This guide explains what each one routes, how to set each up on Linux, a Raspberry Pi and Docker, how to run both on one machine, and what to check when traffic does not flow. Commands follow Tailscale’s current knowledge base articles on exit nodes and subnet routers.
Exit nodes carry internet traffic, subnet routers carry LAN traffic
The cleanest way to tell them apart is to ask which destinations each one takes over.
A subnet router advertises specific private ranges, such as 192.168.1.0/24. Devices on your tailnet that accept the route send traffic for those addresses to the subnet router, which forwards it onto the local network. Everything else, including normal internet browsing, keeps using the device’s own connection. This is how you reach things that cannot run a Tailscale client: printers, switches, smart home hubs, a router’s admin page, or an old NAS.
An exit node offers to carry all internet traffic. When a device selects an exit node, its traffic to the public internet goes through the tailnet to that machine and leaves from its network, so websites see the exit node’s public IP. Nothing changes until a client explicitly selects that exit node; advertising one does not affect anyone by itself.
In short: if your question is “how do I reach 192.168.1.50 from outside,” you want a subnet router. If your question is “how do I make my traffic look like it is coming from home,” you want an exit node. What an exit node does and does not hide is covered in Tailscale exit nodes vs a commercial VPN, and why a subnet router beats opening ports in Tailscale subnet routers vs port forwarding.
Turning on IP forwarding for either role on Linux
Both roles make a Linux machine forward packets between interfaces, which the kernel does not do by default. Tailscale’s subnet router guide uses a drop-in sysctl file:
echo 'net.ipv4.ip_forward = 1' | sudo tee -a /etc/sysctl.d/99-tailscale.conf
echo 'net.ipv6.conf.all.forwarding = 1' | sudo tee -a /etc/sysctl.d/99-tailscale.conf
sudo sysctl -p /etc/sysctl.d/99-tailscale.conf
On systems without /etc/sysctl.d, the same lines go in /etc/sysctl.conf. If the machine uses firewalld, Tailscale’s docs note you may need to allow masquerading because of a known issue:
sudo firewall-cmd --permanent --add-masquerade
Tailscale’s exit node guide also recommends that your firewall deny forwarded traffic by default, so that only traffic Tailscale handles gets forwarded.
Setting up a subnet router
Install Tailscale on an always-on machine that sits on the LAN you want to reach. On a Raspberry Pi or most Linux distributions, the official install script works:
curl -fsSL https://tailscale.com/install.sh | sh
sudo tailscale up
Enable IP forwarding as above, then advertise your LAN range:
sudo tailscale set --advertise-routes=192.168.1.0/24
You can list several ranges separated by commas. Advertising is only half of it; the route must be approved. In the admin console, open Machines, select the device, find the Subnets section, select the routes and save.
Then check the clients. According to Tailscale’s docs, Android, iOS, macOS, tvOS and Windows pick up approved subnet routes automatically. Linux clients do not; they only see Tailscale IP addresses until you run:
sudo tailscale set --accept-routes
Do not run --accept-routes on the subnet router itself unless you have a reason. Tailscale’s docs advise using --advertise-routes alone for most setups, since accepting routes on routers can create inefficient paths.

Setting up an exit node
With IP forwarding enabled on the Linux machine, advertise it:
sudo tailscale set --advertise-exit-node
sudo tailscale up
Tailscale’s guide says that when you pass --advertise-exit-node to tailscale set, you must run tailscale up afterward. Then approve it: on the Machines page, open the device’s menu, choose Edit route settings and enable Use as exit node.
Exit nodes are not limited to Linux. Tailscale lists Linux, macOS, Windows, Android, iOS and tvOS as able to act as one, but notes that on Android, macOS and Windows exit nodes use userspace routing, which is slower, and that macOS and Windows machines must stay awake. For a home exit node, a small Linux box or Raspberry Pi wired to the router is the usual choice.
To use it from a client, pick it in the Tailscale app’s exit node menu. On Linux:
sudo tailscale set --exit-node=100.101.102.103 --exit-node-allow-lan-access=true
# stop using it
sudo tailscale set --exit-node=
The --exit-node-allow-lan-access flag keeps the client’s own local network reachable (the local printer, the hotel’s captive portal) while internet traffic goes through the exit node. Tailscale points out that this also exposes the device to the untrusted local network, so leave it off when you do not need it.
Approving routes automatically with autoApprovers
Clicking approve in the admin console is fine for one machine. If you rebuild routers often or manage several sites, put the approval in the tailnet policy file instead. Tailscale’s policy syntax supports an autoApprovers section:
"autoApprovers": {
"routes": {
"192.168.1.0/24": ["tag:router"]
},
"exitNode": ["tag:router"]
}
Any device tagged tag:router that advertises 192.168.1.0/24, or any smaller range inside it, is approved automatically, and the same tag can advertise an exit node. Tailscale recommends tags as approvers because routes approved through a user stop being advertised if the device is re-authenticated by someone else or that user is removed.
If you have customized your access rules, there is one more line to add for exit nodes. Tailscale’s docs explain that once you write your own policy, users no longer get exit node access unless a grant or ACL allows autogroup:internet as a destination. Granting access to the exit node device itself only lets you connect to that machine, not route internet traffic through it. Subnet routes have the same principle: users need a rule allowing the destination range, such as 192.168.1.0/24:*. Writing those rules for a family is covered in Tailscale ACLs for a household.
Running both roles on one machine, a Pi or a container
Nothing stops one machine from being both an exit node and a subnet router, and at home that is often the simplest arrangement:
sudo tailscale set --advertise-exit-node --advertise-routes=192.168.1.0/24
sudo tailscale up
Prefer tailscale set for changing these flags. tailscale up expects you to repeat every non-default flag you set earlier and refuses to apply a partial change, which is a common way to accidentally drop an advertised route. tailscale set changes only what you pass.
In Docker
The official tailscale/tailscale image is configured with environment variables. Tailscale’s Docker parameter reference lists TS_ROUTES for advertising subnet routes, TS_EXTRA_ARGS for flags without their own variable (including --advertise-exit-node), and TS_STATE_DIR, which must persist or the container registers as a new machine on every restart. Two defaults matter here: TS_USERSPACE is enabled by default, and TS_ACCEPT_DNS defaults to not accepting tailnet DNS settings. Kernel networking is faster but needs /dev/net/tun and extra capabilities.
services:
tailscale:
image: tailscale/tailscale:latest
hostname: home-gateway
network_mode: host
environment:
- TS_AUTHKEY=tskey-auth-REPLACE_ME
- TS_STATE_DIR=/var/lib/tailscale
- TS_USERSPACE=false
- TS_ROUTES=192.168.1.0/24
- TS_EXTRA_ARGS=--advertise-exit-node --advertise-tags=tag:router
volumes:
- ./tailscale-state:/var/lib/tailscale
devices:
- /dev/net/tun:/dev/net/tun
cap_add:
- NET_ADMIN
restart: unless-stopped
With host networking, the host kernel does the forwarding, so the sysctl settings from earlier still apply on the Docker host. Pin the image to a specific version rather than latest once it works, and keep the auth key out of version control. If you use a tag in --advertise-tags, the auth key or OAuth client must be allowed to apply it.

How DNS behaves with each feature
MagicDNS names such as nas.your-tailnet.ts.net resolve only on devices running Tailscale. Devices behind a subnet router do not run the client, so they have no MagicDNS names; you reach them by LAN IP, or by your LAN’s own DNS names if you set up split DNS.
Split DNS is the piece that makes a subnet router pleasant to use. In the admin console’s DNS page, add a nameserver pointing at your LAN’s DNS server (for example the router or a Pi-hole at 192.168.1.1) and restrict it to your local domain, such as home.arpa or lan. Tailnet devices then send only those lookups to the LAN resolver, through the subnet router. Without an approved subnet route covering that resolver’s IP, and an access rule that lets users reach it, the queries have nowhere to go.
Exit nodes handle DNS differently. Tailscale’s docs explain that an exit node runs a DNS server for the peers using it, so lookups follow your traffic instead of going to the local network’s resolver. On Windows exit nodes, that DNS server uses a system thread per query, which Tailscale flags as a concern for heavily used exit nodes.
When traffic does not flow: the usual causes
“Exit node selected, no internet”
- IP forwarding is off on the exit node, or did not survive a reboot. Check with
sysctl net.ipv4.ip_forward. - The host firewall drops forwarded traffic. UFW’s default forward policy is to drop, and
firewalldmay need the masquerade rule shown earlier. - The exit node was advertised but not approved in the admin console.
- Your policy file lacks
autogroup:internet, so the client may select the node but is not allowed to send internet traffic through it. - The exit node went to sleep. Laptops and desktops on macOS or Windows stop forwarding when they sleep.
“The route is approved but I cannot reach the LAN”
- The client is Linux and has not run
tailscale set --accept-routes. - The client is on a network that uses the same range. If the hotel and your home both use
192.168.1.0/24, the local network wins. Renumbering your home LAN to a less common range avoids this. - Two sites advertise the same range. Tailscale’s answer is 4via6:
tailscale debug via 7 10.1.1.0/24generates a unique IPv6 route for site 7, which you advertise instead, and with MagicDNS you reach hosts by names like10-1-1-16-via-7. Site IDs range from 0 to 65535 on current versions. - A more specific route’s router is offline. Tailscale uses longest-prefix matching and does not fall back to a broader route when the more specific router disappears.
“LAN devices see the router’s IP, not the client’s”
That is source NAT, and it is on by default for subnet routers so LAN devices can reply without knowing about Tailscale. If you need real client addresses in logs, Linux subnet routers support --snat-subnet-routes=false, but then your LAN router needs a return route for 100.64.0.0/10 pointing to the subnet router’s LAN IP, or replies will be lost.
Getting acceptable speed from an exit node or subnet router
An exit node at home is capped by your home upload speed, since every page you load comes down to the exit node and then back up to you. No tuning changes that. Beyond the link itself, Tailscale’s performance guidance for Linux routers and exit nodes is specific:
- Use Tailscale 1.54 or later on a Linux 6.2 or later kernel to benefit from UDP offload improvements.
- Enable
rx-udp-gro-forwardingand disablerx-gro-liston the network interface withethtool, and make it persistent through NetworkManager or networkd-dispatcher, since the setting does not survive a reboot by default. - Favor CPU clock speed over core count when picking hardware.
- Prefer kernel networking over userspace mode, which is why the Docker example above sets
TS_USERSPACE=false, and avoid Android, macOS or Windows as a busy exit node.
Also check whether connections are direct. If tailscale status shows traffic going through a relay rather than a direct connection, throughput will be limited regardless of the hardware, and fixing NAT traversal will help more than any setting on the router. If your exit node is a travel router rather than a home box, the DNS caveats in Tailscale exit node on a travel router are worth reading before you rely on it.