Rewrite README as a short project front page; add a Nix flake

The README is now a landing page: the idea (a mesh VPN with no server, joined with a network name and a secret), install and platform tables, exit nodes, and a short build section. Implementation detail stays in docs/. The flake builds tsng and the tray from source and ships a NixOS module (services.tsunagi). Dist install notes now show the real `join -n -s` syntax, and two links that pointed into the old README are redirected. Co-Authored-By: Claude <noreply@anthropic.com>
Windows installer: find the Wintun licence at the archive root
2026-10-03 17:47:44 +01:00 · 2026-10-03 14:38:16 +01:00 · 2026-10-03 14:16:05 +01:00 · 2026-10-03 13:43:01 +01:00
28 changed files with 1766 additions and 777 deletions
@@ -0,0 +1,5 @@
 # Windows: link the C runtime statically, so the agent and the tray start on a
 # machine that has no Visual C++ redistributable installed. The installer ships
 # nothing but our own files and Wintun.
 [target.x86_64-pc-windows-msvc]
 rustflags = ["-C", "target-feature=+crt-static"]
@@ -190,6 +190,29 @@ jobs:
            tar -C dist/pkg -czf "${archive}" "${{ matrix.asset }}"
          fi
      # The Windows installer: the agent as a service, Wintun, the tray and the
      # PATH entry in one setup program. Wintun is the signed library from
      # wintun.net, pinned by hash so a changed download fails the build.
      - name: Build the Windows installer
        if: runner.os == 'Windows'
        shell: pwsh
        run: |
          $ErrorActionPreference = 'Stop'
          $version = $env:GITHUB_REF_NAME.TrimStart('v')
          Invoke-WebRequest -Uri https://www.wintun.net/builds/wintun-0.14.1.zip -OutFile wintun.zip
          $hash = (Get-FileHash wintun.zip -Algorithm SHA256).Hash
          if ($hash -ne '07C256185D6EE3652E09FA55C0B673E2624B565E02C4B9091C79CA7D2F24EF51') {
            throw "wintun.zip has an unexpected hash: $hash"
          }
          Expand-Archive wintun.zip -DestinationPath wintun-dl
          choco install innosetup -y --no-progress
          $iscc = Join-Path ${env:ProgramFiles(x86)} 'Inno Setup 6\ISCC.exe'
          $stage = Join-Path $PWD 'dist/pkg/${{ matrix.asset }}'
          & $iscc "/DAppVersion=$version" "/DSourceDir=$stage" `
            "/DWintunDir=$(Join-Path $PWD 'wintun-dl/wintun/bin/amd64')" `
            "/O$PWD" dist/windows/tsunagi.iss
          if ($LASTEXITCODE -ne 0) { throw "ISCC failed with $LASTEXITCODE" }
      # A dependency-free .deb for the x86_64 and aarch64 Linux builds: it drops
      # the static binary, the systemd service and the polkit rule. Built with
      # dpkg-deb (present on the runner); see dist/deb/build-deb.sh.
@@ -251,6 +274,7 @@ jobs:
            ${{ matrix.asset }}-*.*
            *.deb
            *.pkg.tar.zst
            tsunagi-setup-*.exe
          if-no-files-found: error
  publish:
@@ -322,5 +346,5 @@ jobs:
              cat up.json >&2 || true
              rc=1
            fi
-          done < <(find artifacts -type f \( -name '*.tar.gz' -o -name '*.zip' -o -name '*.deb' -o -name '*.pkg.tar.zst' \) -print0)
+          done < <(find artifacts -type f \( -name '*.tar.gz' -o -name '*.zip' -o -name '*.deb' -o -name '*.pkg.tar.zst' -o -name 'tsunagi-setup-*.exe' \) -print0)
          exit "$rc"
@@ -192,6 +192,29 @@ jobs:
            tar -C dist/pkg -czf "${archive}" "${{ matrix.asset }}"
          fi
      # The Windows installer: the agent as a service, Wintun, the tray and the
      # PATH entry in one setup program. Wintun is the signed library from
      # wintun.net, pinned by hash so a changed download fails the build.
      - name: Build the Windows installer
        if: runner.os == 'Windows'
        shell: pwsh
        run: |
          $ErrorActionPreference = 'Stop'
          $version = $env:GITHUB_REF_NAME.TrimStart('v')
          Invoke-WebRequest -Uri https://www.wintun.net/builds/wintun-0.14.1.zip -OutFile wintun.zip
          $hash = (Get-FileHash wintun.zip -Algorithm SHA256).Hash
          if ($hash -ne '07C256185D6EE3652E09FA55C0B673E2624B565E02C4B9091C79CA7D2F24EF51') {
            throw "wintun.zip has an unexpected hash: $hash"
          }
          Expand-Archive wintun.zip -DestinationPath wintun-dl
          choco install innosetup -y --no-progress
          $iscc = Join-Path ${env:ProgramFiles(x86)} 'Inno Setup 6\ISCC.exe'
          $stage = Join-Path $PWD 'dist/pkg/${{ matrix.asset }}'
          & $iscc "/DAppVersion=$version" "/DSourceDir=$stage" `
            "/DWintunDir=$(Join-Path $PWD 'wintun-dl/wintun/bin/amd64')" `
            "/O$PWD" dist/windows/tsunagi.iss
          if ($LASTEXITCODE -ne 0) { throw "ISCC failed with $LASTEXITCODE" }
      # A dependency-free .deb for the x86_64 and aarch64 Linux builds: it drops
      # the static binary, the systemd service and the polkit rule. Built with
      # dpkg-deb (present on the runner); see dist/deb/build-deb.sh.
@@ -250,6 +273,7 @@ jobs:
            ${{ matrix.asset }}-*.*
            *.deb
            *.pkg.tar.zst
            tsunagi-setup-*.exe
          if-no-files-found: error
  publish:
@@ -278,7 +302,7 @@ jobs:
          # download-artifact nests each artifact under its own directory, so
          # the archives are artifacts/<asset>/<file>.
-          mapfile -t files < <(find artifacts -type f \( -name '*.tar.gz' -o -name '*.zip' -o -name '*.deb' -o -name '*.pkg.tar.zst' \))
+          mapfile -t files < <(find artifacts -type f \( -name '*.tar.gz' -o -name '*.zip' -o -name '*.deb' -o -name '*.pkg.tar.zst' -o -name 'tsunagi-setup-*.exe' \))
          printf 'asset: %s\n' "${files[@]}"
          prerelease=()
@@ -8,7 +8,7 @@ before touching the code. Do not restate the full design here; follow the links.
 A proof-of-concept agent library for small private mesh networks. A user
 supplies a network name and one shared secret; agents derive the same network
 space independently, find each other, prove membership and exchange control
-messages. Scope and non-scope are in [README.md](README.md).
+messages. What it does and does not do is in [docs/architecture.md](docs/architecture.md) and [docs/threat-model.md](docs/threat-model.md).
 There is no central server, no owner, no registration and no majority vote.
 Design accordingly: a majority is not a root of trust.
@@ -6250,6 +6250,7 @@ dependencies = [
 "tsunagi-tcp-tls",
 "tsunagi-wg-quic",
 "tsunagi-wireguard",
 "windows-service",
 ]
 [[package]]
@@ -7135,6 +7136,17 @@ dependencies = [
 "windows-link",
 ]
 [[package]]
 name = "windows-service"
 version = "0.8.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "857224b3b211c6f3616921f081ee54721ee3ad2ace2fac6a6337e032f7b4dcf2"
 dependencies = [
 "bitflags 2.13.2",
 "widestring",
 "windows-sys 0.61.2",
 ]
 [[package]]
 name = "windows-strings"
 version = "0.1.0"
@@ -1,766 +1,183 @@
-# tsunagi
+<p align="center">
-
+  <img src="dist/linux/tsunagi.svg" width="96" alt="">
-A proof-of-concept Rust library for **small private mesh networks** — a handful
+</p>
-of friends, home machines, a few servers. Units to dozens of participants, not
+
-thousands.
+<h1 align="center">tsunagi</h1>
-
+
-The end user configures exactly two things:
+<p align="center">
-
+  <b>A mesh VPN with no server.</b><br>
-```text
+  Pick a network name, share a secret, and your machines are on one private network.
-network_name
+</p>
-secret          # one shared secret; "password" and "secret" mean the same value
+
-```
+<p align="center">
-
+  <a href="https://github.com/house-of-vanity/tsunagi/releases/latest"><img alt="release" src="https://img.shields.io/github/v/release/house-of-vanity/tsunagi?include_prereleases"></a>
-From those, every agent independently derives the same network space. There is
+  <a href="https://github.com/house-of-vanity/tsunagi/actions/workflows/ci.yml"><img alt="CI" src="https://github.com/house-of-vanity/tsunagi/actions/workflows/ci.yml/badge.svg"></a>
-no central server, no network owner with special powers, no registration and no
+  <a href="LICENSE"><img alt="license" src="https://img.shields.io/badge/license-WTFPL-blue"></a>
-majority vote. Anyone who knows the parameters can join; nobody has to trust
+</p>
-anybody else.
+
-
+```console
-## What this proof of concept actually does
+laptop$ tsng join -n home
-
+joined `home`
 A working library with **real iroh connections** and integration tests:
 - persistent device identity stored in SQLite, stable across restarts;
 - several independent networks at once in one agent;
 - deterministic network identity derived from name + secret;
 - candidates supplied by a replaceable discovery component;
 - automatic first contact through the public Mainline DHT (BEP44);
 - real iroh connections plus an explicit mutual proof of network membership;
 - a small versioned control protocol: handshake, hostname/capability
  announcement, ping/pong;
 - automatic reconnect with bounded exponential backoff and jitter;
 - status snapshots, an event stream and honest diagnostics;
 - configuration restored after a restart;
 - correct behaviour when the disposable cache is missing or corrupt;
 - a **WireGuard data plane**, in userspace: its own key per network, an IPv6
  overlay with deterministically derived addresses and optional IPv4, real
  tunnels carried over iroh, and address ownership enforced rather than
  believed;
 - a **command line agent**, `tsng`, with a local control socket.
 ### What it deliberately does **not** do
 Not implemented, and not pretended to be: DNS, routing through
 intermediate participants, a full CRDT, dynamically loaded plugins, a system
 service, a complete CLI, or a local control socket. Snapshot synchronisation
 and signed revocations are designed for but not implemented — see
 [docs/sync-model.md](docs/sync-model.md). The WireGuard plugin's own limits,
 including that its system backend is Linux-only, are in
 [docs/wireguard.md](docs/wireguard.md#limits-and-future-work).
 **Control and data are separated logically, not physically.** Both ride on
 iroh, on different ALPNs and different connections, so the data plane inherits
 iroh's hole punching and relay fallback instead of reimplementing them — while
 the control protocol still knows nothing about packets and can keep a different
 transport underneath it later. Filtering user traffic remains the operating
 system's and the user's responsibility, not this library's.
 ## Requirements
 - Rust 1.91 or newer (iroh 1.2 requires it), edition 2024. Pinned dependencies
  in `Cargo.lock`.
 - No internet, no public DHT, no public relay, no administrator rights and no changes
  to OS network settings are needed to build or test.
 - WireGuard runs in userspace (boringtun): **no kernel module and no `wg`
  tool**. Only creating a real network interface needs `CAP_NET_ADMIN`, and
  `--no-tun` skips even that.
 ## Trying it on two machines
 On the first machine:
 ```bash
 cargo build --release
 ./target/release/tsng status          # this device, the agent, and this host
 ./target/release/tsng network         # the networks this device belongs to
 ```
 Two commands do the work, and they are separate on purpose: `up` runs the
 agent — one per device, for as long as you want it — and `join` decides
 what it belongs to, at any time, while it runs.
 ```bash
 # Terminal one: the agent. It prints its endpoint id and then serves.
 ./target/release/tsng up
 # Terminal two: make a network. With no --secret it invents one and prints
 # it, along with the line to send the other machine.
 ./target/release/tsng join --network lab
 ```
 On the second machine, start its agent and join the same network with the
 secret that was printed:
 ```bash
 ./target/release/tsng up
 ./target/release/tsng join --network lab --secret tsn1...
 ```
 **Name and secret are enough for the first meeting.** Mainline DHT discovery
 is enabled by default. Each active network publishes this device's endpoint
 and searches until the first authenticated connection. While connected it
 only republishes; after 60 seconds without any authenticated connection it
 searches again. Publication is approximately every five minutes, and records
 older than 15 minutes are ignored. DHT storage nodes may retain them longer.
 Use `--no-dht` to disable DHT, or `--peer <endpoint-id>` for optional manual
 bootstrap. `--reach local` disables public DHT regardless of the DHT flag.
 Endpoint records are public, signed, and contain no network secret. Discovery
 is not authentication: every connection must still prove network membership.
 Details and failure behavior are in [docs/mainline-dht.md](docs/mainline-dht.md).
 **One introduction is enough for the whole network.** Members tell each
 other about the members they know, and every author of a signed record is
 somebody to try, so a device pointed at one member ends up talking to all
 of them rather than to the one that happened to be on its command line.
 What travels is a candidate — an address somebody has seen — and it is
 authenticated by the handshake like any other; being introduced grants
 nothing. It is deliberately the same shape a lookup in a distributed hash
 table would return, so that is a source to add beside this one rather than
 a redesign.
 An agent with nobody to contact says so in `status` rather than sitting
 there looking patient.
 Within a few seconds both print something like:
 ```text
  + peer b47c958462 connected over direct rtt=Some(4.5ms)
  + data link to b47c958462 for wg-quic: direct via 192.0.2.7:41234, datagram 1382
 --- status ---
 control: 1 peer(s), 0 dial failure(s), 0 handshake failure(s)
 wg-quic: tsun0 on 10.13.37.69/24 mtu 1280, 1/1 tunnel(s) established
  4jO4kx9Z 10.13.37.237 handshake 3s ago tx=0 rx=0 dropped=0 path=direct via 192.0.2.7:41234
 ```
 `1/1 tunnel(s) established` means a real WireGuard handshake completed.
 ## Checking that it works
 From another shell on either machine:
 ```bash
 tsng status
 ```
 ```text
 endpoint  7d76ccbbc21bf30767e14422c0494740a2cecc02aa9f82d5b8d57bdae350e7fc
 hostname  tsunagi-7d76ccbbc2
 bound     0.0.0.0:41641
 network lab (z2o4qwrvnj3zb6st2aoqg4abf342j662q2ujttqrsmz22argk2ba)  active
  peer b345d5271b  tsunagi-b345d5271b  Direct  rtt 24ms
  overlay tsunz2o4qwrvnj3 fd09:…:c1c6/64 and 100.110.49.177 mtu 1280  1/1 tunnel(s) up
    SDsEb/WF  fd09:…:c4b / 100.65.243.53  handshake 4s ago  tx 0 rx 0  Direct via Ip(…)
 ```
 `1/1 tunnel(s) up` and a recent handshake mean the tunnel is live. Then send
 real traffic to the peer's overlay address:
 ```bash
 ping6 fd09:…:c4b        # or
 ping  100.65.243.53
 ```
 `tx` and `rx` in the status should start moving.
 **Addresses are allocated and then remembered.** The default range is
 `10.13.37.0/24`; the first member to join settles it and later members adopt
 what they find, so `--ipv4-range` only matters for whoever starts the network:
 ```bash
 tsng up --ipv4-range 10.44.0.0/16
 tsng up --ipv4-range none   # no data plane at all
 ```
 One agent has one interface, so two of its networks cannot both use that
 range. The second takes the range **derived from its own network id**: not
 picked locally — every member derives the same one from something they all
 already have — so it is an agreement rather than a guess, and a device in
 several networks gets an address in each.
 An address is claimed with a record signed by that member's persistent device
 key, stored, and merged between every replica. A member that disappears for a
 month comes back to the same address, because the claim outlived the session.
 No vote is involved — see
 [docs/wireguard.md](docs/wireguard.md#ipv4-allocated-signed-and-kept) and
 [docs/sync-model.md](docs/sync-model.md).
 Because the address is allocated at run time rather than derived, it is not
 known until the agent has started and agreed with its peers. The agent then
 assigns it to the interface itself.
 ## Levels
 The command line is split the way the design is. A bare `tsng up` needs
 only a network name and a secret; everything else sits under the level it
 belongs to, which `--help` shows as two sections:
 * **System** — what the agent itself does: how it reaches peers (`--reach`),
  the one overlay interface it owns (`--interface`, `--mtu`, `--no-tun`), the
  address range (`--ipv4-range`), and the local resolver (`--no-dns` to disable).
 * **Transport** — which protocols carry packets (`--protocol`, a list) and
  their own settings (`-o key=value`, or `-o protocol:key=value`).
 ```bash
 tsng protocols          # what this build can carry packets with
 ```
 ```text
 wg-quic  (wire version 4)
        what                  WireGuard's cryptography carried in iroh's QUIC datagrams
        -o keepalive=SECONDS  keeps a tunnel and its link warm through a NAT
        -o mtu=BYTES          largest packet a tunnel will carry, at least 576
 ```
 Each protocol declares its own settings, so the agent can list them without
 knowing anything about the protocol, and a setting no selected protocol takes
 is refused rather than ignored. `--protocol none` runs the control plane by
 itself.
 A pair of peers uses a protocol they both have **at the same wire version**.
 That is not the software version: two peers on different builds carry traffic
 for each other for as long as the bytes between them have not changed. A peer
 with nothing in common keeps its control plane — messages and signed state
 still flow — and simply has no data plane, which `tsng status` shows as a
 session with no agreed protocol.
 ## Names
 DNS is enabled by default and serves a local zone for **every network this device is in**,
 each named after the network, so members can be reached by name instead of
 by address:
 ```bash
 tsng up               # DNS is already enabled
 dig @127.0.0.1 -p 5354 music.lab
 ```
 An explicit choice is remembered across restarts. `--no-dns` or `tsng dns off`
 disables it; `--dns` or `tsng dns on` enables it again. Existing saved opt-outs
 remain respected. The `dns on` and `dns off` commands also take effect immediately
 on a running agent. `--dns` and `--no-dns` cannot be combined.
 The zone of a network is its name — `--dns-zone` is gone, because with
 several networks there is no single zone to name. A network name may contain
 dots, so `--network lab.internal` is how you get `music.lab.internal`.
 Names come from signed state, which is the point: **a member that is
 switched off still resolves**, because its claim outlived the session.
 The answers are IPv4 addresses, because that is what the overlay is. The
 *questions* are taken on `127.0.0.1` and `[::1]`, over UDP and TCP, so a
 resolver reaches it over whichever family it uses; both are published to the
 system resolver together. Loopback and nothing else: the zones are a view
 for the host running the agent, and binding an overlay address would put
 them in front of the whole mesh — an agent in two networks would then answer
 one network's questions about the other's names.
 The zone is the network name, and it is yours to choose. A name that
 shadows a real public domain is reported and then used: a network called
 `ru` warns that every public `.ru` name becomes unreachable from this host,
 and then does it. Anything that collides with nothing — which is most names
 — is said nothing about, because a warning that fires on every private name
 anybody picks is how people learn to ignore warnings.
 On Linux the agent tells systemd-resolved to send questions for that suffix
 here, over D-Bus, scoped to the overlay interface and as a *routing* domain
 so it never becomes the resolver for anything else. resolved drops the whole
 setting when the interface goes, and the interface goes with the agent.
 Only an interface the agent created, though. Under `--no-tun` there is no
 host interface at all, and the agent says so and serves the zone on
 loopback rather than configuring whatever else on the host happens to share
 the name.
 That last step needs permission that `CAP_NET_ADMIN` does not give:
 systemd-resolved asks polkit, and polkit decides by **user**, not by
 capability, so there is no way for the agent to arrange it from inside. On
 a desktop the refusal reads `Interactive authentication required`.
 Running as a system service is enough. Otherwise the agent prints the rule
 that grants it — the four actions it calls and nothing else — ready to
 paste:
 ```bash
 sudo tee /etc/polkit-1/rules.d/50-tsunagi-resolved.rules > /dev/null <<'RULE'
 polkit.addRule(function(action, subject) {
    var allowed = [
        "org.freedesktop.resolve1.set-dns-servers",
        "org.freedesktop.resolve1.set-domains",
        "org.freedesktop.resolve1.set-default-route",
        "org.freedesktop.resolve1.revert"
    ];
    if (allowed.indexOf(action.id) >= 0 && subject.user == "YOUR-USER") {
        return polkit.Result.YES;
    }
 });
 RULE
 ```
 **Without it the server still runs** — `tsng status` prints where it is
 listening and the exact `dig` line — so the automatic part is missing, not
 the feature. The refusal is said once rather than on every pass, and retried
 slowly, because nothing but a person will change it.
 The server is authoritative for its zone and nothing else. No recursion, no
 forwarding, no cache: pointing a resolver at it can never make it a route to
 the outside.
 ## Privileges
 On Linux the agent **manages its own overlay interface**. It creates the TUN
 interface, sets the MTU, brings it up and assigns both overlay addresses, all
 over netlink in process — no `ip` invocation, no shell, nothing that a
 remote peer could influence.
 That needs `CAP_NET_ADMIN`, granted once:
 ```bash
 sudo setcap cap_net_admin+p /usr/local/bin/tsng
 ```
 `+p` rather than `+ep`: the capability is then *permitted* but not
 *effective*, and the agent raises it only around the handful of netlink calls
 that need it — a few milliseconds at startup, and again if its address
 allocation changes. Everything else, including every byte from the network,
 is handled with it lowered. `+ep` works too; the agent lowers it on the way
 in.
 `tsng status` says which of these applies on the host it runs on, along
 with what the agent is doing. It grades each finding: **ok** for what works,
 **warn** for what the agent runs without and you can fix from the line it
 prints, **FAIL** for what it cannot work around. The words carry the grade as
 well as the colour, so the report reads the same piped to a file or on a
 terminal without colour, and it honours `NO_COLOR`.
 Members are listed online first, then the ones that are away. A member that
 is away is reported plainly rather than flagged: in a mesh of laptops it is
 the ordinary condition, not a fault. The signed state is what makes that
 sayable — it remembers who belongs while they are gone, so the report can say
 "offline, 10.13.37.99 still reserved for it" instead of leaving a
 dial-failure counter to imply it. Counters are history and are never graded:
 a peer that left and came back should not leave the report looking broken.
 The other two commands split along the line the system itself draws. `id` is
 this **device**: the key it signs with and the name it answers to. `network`
 is what it **belongs to**: which networks, their secrets, joining and
 leaving. A device outlives every network it is in, and a network outlives
 any device in it, so a command that mixed them had to be read twice.
 Every item takes the same shape, so there is nothing to remember: name it to
 see it, name it with a value to change it.
 ```
 tsng id                      everything about this device
 tsng id hostname             the name it answers to
 tsng id hostname mango       change it
 tsng id key                  the key it signs with
 tsng id key rotate           replace that key
 tsng network                 the networks this device belongs to
 tsng network join -n lab -s tsn1…   join one; adds it to a running agent
 tsng network join -n lab     resume one this device has, or make it on the spot
 tsng network stop lab        stop serving it, keeping everything
 tsng network start lab       serve it again, from where it left off
 tsng network leave lab       give up the address and name, then forget it
 tsng network secret          the secret of each joined network
 tsng network secret lab      just that one, for copying
 tsng network secret generate a fresh secret for a network that does not exist yet
 tsng dns                     whether the local resolver is serving, and what
 tsng dns on                  start it, now and after every restart
 tsng dns off                 stop it, now and after every restart
 ```
 **A network without a secret makes one.** `tsng join --network lab`
 resolves a bare name in the obvious way: if this device is already in
 exactly one network called `lab`, that one — so the name alone resumes what
 you have; if it is in none, a fresh random secret, printed in full with the
 line to send the other machine:
 ```
 joined `lab` (k2on43wadbi5x267vp6z3ogkm7nbjedfdoxyauxhtxhwqrprylba)
  secret  tsn1u7c…
-Run this on the other machine:
+server$ tsng join -n home -s tsn1u7c…
 server$ tsng id hostname server
-  tsng join --network lab --secret tsn1u7c…
+laptop$ ssh server.home
 Its agent has to be running. If it is not:
  tsng up --peer 91e83a6e2b7a…
 ```
-Two networks of one name is a thing that happens — a mistyped secret makes
+No account, no coordination server, no admin console, no keys to approve.
-one — so joining says whether the network was already here, and warns when
+Anyone who knows the **name** and the **secret** is in. Nobody else can find
-another configured network answers to the same name. A name is a label and
+the network, let alone enter it.
 the id is the identity.
-That is the ad-hoc case: one person makes a network and sends the command
+## Why
 round. The secret is printed *only* when the agent invented it — there is
 nowhere else to read it from — and never when it was supplied, because then
 it is already yours. Two networks of one name and no secret is the one case
 with no answer, and it says so instead of choosing.
-A secret is printed by `network secret` and nowhere else — not by `id`, not
+Tailscale, Headscale and ZeroTier are great, but each needs a control server:
-by `status`, not in a log, a `Debug` rendering or anything sent to a peer.
+theirs, or one you have to run and keep alive. tsunagi doesn't. Machines find
-Asking for it is deliberate, because these reports get pasted into chats.
+each other through the public BitTorrent DHT, connect directly (hole punching,
 with relay fallback), and keep the network state among themselves.
-**`up` takes no network at all** if you would rather decide later: it
+|                        | Tailscale          | Headscale        | ZeroTier            | **tsunagi**         |
-brings up the agent and whatever it is already configured for, and waits.
+| ---------------------- | ------------------ | ---------------- | ------------------- | ------------------- |
-That is the shape of a daemon in one terminal and `tsng network join`
+| Control server         | hosted by vendor   | you run it       | hosted or self-run  | **none**            |
-in another.
+| Account / sign-in      | yes                | yes              | yes                 | **no**              |
 | To join a network      | sign in, get approved | pre-auth key  | network ID, approval | **name + secret**  |
 | Direct connections     | yes                | yes              | yes                 | yes                 |
 | Exit nodes             | yes                | yes              | via routes          | yes                 |
-`network join` is also the answer to a question `up` cannot: a state
+## Features
 directory belongs to one live agent, so a second `tsng up` cannot add a
 network to the one already running. This adds it over the control socket and
 it starts at once. With no agent running it is written to the configuration
 and starts with the next `up`.
-The name is part of the signed state, so changing it revokes the previous
+- **Two things to know:** a network name and a secret. A secret is generated for you when you make a network.
-one: there is one record per author, a new version replaces the whole claim,
+- **Direct, encrypted links** between members (WireGuard cryptography), with relay fallback behind strict NATs.
-and no replica can keep the old name standing. Changing it while the agent
+- **Names for machines:** `server.home` just works, even while `server` is switched off.
-runs goes through the agent, which republishes and tells its peers straight
+- **Exit nodes:** send all your internet traffic through any member that offers one.
-away.
+- **Multi-hop:** members that can't reach each other directly are routed through a member that can.
 - **LAN games:** UDP broadcasts are relayed, so "LAN" lobbies show up across the mesh.
 - **Several networks at once** on one device.
 - **Tray app** for Windows, macOS and Linux; a command line for everything.
-Replacing the signing key makes this device a different member, and it loses
+## Install
 the address and name the old key held — nothing can sign on a retired key's
 behalf, and by design there is no authority that could overrule an author. So
 the outgoing key signs a release for every network on its way out, which
 frees them for whoever wants them next, and the whole thing commits at once.
-`status` and `id` both prefer a running agent, which is live and
+Grab a build from the [latest release](https://github.com/house-of-vanity/tsunagi/releases/latest).
 authoritative, and fall back to reading the state store when there is none.
 Reading takes no directory lock, so neither has to wait for the agent it is
 asking about — nor does either need one to be running.
-### It cleans up after itself
+| Platform              | How                                                                              |
 | --------------------- | -------------------------------------------------------------------------------- |
 | **Windows**           | Run `tsunagi-setup-<version>-x86_64.exe`. Installs the service, the tray and `tsng`. |
 | **macOS**             | `brew tap house-of-vanity/tap && brew install --cask tsunagi-gui`<br>CLI only: `brew install tsunagi` |
 | **Debian / Ubuntu**   | `sudo apt install ./tsunagi-gui_<version>_amd64.deb` (`tsunagi_…` for no tray)   |
 | **Arch**              | `sudo pacman -U tsunagi-gui-<version>-1-x86_64.pkg.tar.zst`                      |
 | **NixOS**             | [Flake and module](#nixos) below                                                 |
 | **Any other Linux**   | Unpack the tarball, then `sudo ./install.sh && sudo systemctl enable --now tsunagi` |
-The interface is tied to an open file descriptor and is deliberately **not**
+Linux packages run the agent as an unprivileged service user. Add yourself to
-made persistent, so the kernel removes it when the agent exits — on a clean
+the `tsunagi` group (`sudo usermod -aG tsunagi $USER`, then log in again) to
-shutdown, on a panic, on `SIGKILL`, on power loss alike. Keeping it is what
+use `tsng` and the tray without `sudo`.
 would take an action; removing it is the default.
-If something is left behind anyway — an interface made by an older version's
+## Quick start
 manual recipe, or one from a run killed in the instant between creating it and
 recording it — the next start **replaces it**, along with any stale addresses
 it carried. Two things are never touched:
-* an interface that is not a TUN, because the name colliding with somebody's
+```sh
-  bridge is not a reason to destroy the bridge;
+tsng join -n home                 # first machine: makes the network and prints a secret
-* a TUN that another process is holding open, because that is a working
+tsng join -n home -s tsn1…        # every other machine
-  overlay belonging to somebody else — most likely a second agent on this
+tsng status                       # who is online, over which path
  host, which should be given a different `--wg-prefix`.
 Both of those refuse with an explanation rather than guessing.
 Two settings the manual recipe used to need are gone with it.
 `keep_addr_on_down` existed only because an interface nobody held open lost
 carrier and had its IPv6 addresses flushed, and `nodad` only because duplicate
 address detection can never finish without carrier. An interface held open for
 its whole life has carrier for its whole life.
 ### The MTU is 1280
 The default interface MTU stays at 1280 on direct paths, relays and paths
 inside another VPN. Large encrypted packets are split into smaller QUIC
 datagrams and reassembled before reaching WireGuard. SSH and full-size TCP
 segments therefore do not require a manual MTU override or MSS adjustment.
 The original IP packet, including its DF bit, is preserved. See
 [docs/wireguard.md](docs/wireguard.md#mtu) for limits and compatibility.
 ### Summary
 | approach | agent runs as | notes |
 |---|---|---|
 | `setcap cap_net_admin+p` | ordinary user, one capability | recommended: nothing to prepare, nothing left behind. Lost on every rebuild or copy of the binary. |
 | systemd service | `User=`, `AmbientCapabilities=CAP_NET_ADMIN` | the same, for an installed service |
 | `sudo tsng up` | root | everything works, nothing is isolated |
 | macOS root LaunchDaemon | root | the native macOS deployment; see below |
 | `--no-tun` | ordinary user, no capabilities | tunnels run and handshake, traffic never reaches the OS |
 On Windows, put `wintun.dll` beside the executable and start `tsng up`
 from PowerShell or Command Prompt opened with **Run as administrator**.
 Run commands controlling that agent (`join`, `status`, `dns`, `network`)
 as the same Windows user with the same elevation. If Windows denies access
 to the agent's control pipe, the command reports a permission error instead
 of claiming the agent is absent or trying to edit its locked state.
 Creating a TUN without sufficient privileges also explains how to restart
 the agent with the required permissions.
 On macOS the agent creates a `utun` interface, configures it with `ifconfig`
 and `route`, and points the system resolver at the overlay by writing
 `/etc/resolver/<network>` files (which carry a port, so the DNS server needs no
 privileged socket). All of this needs **root**: macOS has no per-capability
 grant like Linux, and a `utun` plus `/etc/resolver` are root-only. The kernel
 assigns the interface its `utunN` name, so the name derived from the network id
 is only a request and the real one is adopted once the interface is up.
 Run it as root — `sudo tsng up` for a quick start, or install it as a root
 **LaunchDaemon** for the native always-on deployment. As a daemon the agent runs
 unprivileged commands (`join`, `status`, `dns`, `network`) over its control
 socket like any other platform; the socket lives at `/var/run/tsunagi/agent.sock`
 and the stores at `/var/db/tsunagi`, both fixed system paths so they do not
 depend on whose shell started `sudo`. A ready-to-edit plist and install notes
 are in [dist/macos/](dist/macos/). The broadcast route is installed; the
 inbound-UDP firewall allowance (a `pf` anchor) is not yet, so `status` reports
 that half as incomplete and names what to allow by hand. The decision logic that
 says *what* to change is shared and tested on every platform.
 ## Checks
 ```bash
 cargo fmt --all -- --check
 cargo clippy --locked --workspace --all-targets -- -D warnings
 cargo test --locked --workspace --all-targets
 ```
-The whole suite runs offline on loopback. Set `TSUNAGI_TEST_LOG=tsunagi=debug`
+Every member gets an address in `10.13.37.0/24` and a name like `<host>.home`.
-to see agent logs while a test runs.
+Machines are called `tsunagi-<id>` until you rename them: `tsng id hostname laptop`.
-There are also two runnable demos, which are demos and not substitutes for the
+### Exit nodes
 tests:
-```bash
+```sh
-cargo run --example two_agents       # control plane only
+tsng network exit-node home on    # on the machine that should be the exit
 tsng exit-node                    # on another one: who offers an exit
 tsng exit-node laptop             # send everything through it
 tsng exit-node off
 ```
-Both run with no privileges and change nothing on the host.
+The tray does the same with a checkbox. An exit node sees everything you send
 through it, so use only members you trust.
 [Details and per-OS notes](docs/exit-node.md).
-## Usage
+## Platforms
-```rust,no_run
+| Platform                  | Build        | CLI `tsng` | Tray | Exit node: offer / use | Packages                            |
-use std::sync::Arc;
+| ------------------------- | ------------ | :--------: | :--: | :--------------------: | ----------------------------------- |
-use tsunagi::config::{AgentConfig, StoragePaths, TransportPolicy};
+| Linux x86_64              | release      |     ✔      |  ✔   |        ✔ / ✔           | `.deb`, `.pkg.tar.zst`, tarball, Nix |
-use tsunagi::discovery::SharedMemoryDiscovery;
+| Linux aarch64             | release      |     ✔      |  –¹  |        ✔ / ✔           | `.deb`, `.pkg.tar.zst`, tarball, Nix |
-use tsunagi::identity::{NetworkName, NetworkSecret};
+| Linux armv7               | release      |     ✔      |  –¹  |        ✔ / ✔           | `.deb`, `.pkg.tar.zst`, tarball      |
-use tsunagi::proto::ControlMessage;
+| macOS, Apple silicon      | release      |     ✔      |  ✔   |        ✔ / ✔²          | Homebrew, tarball                    |
-use tsunagi::{Agent, Result};
+| macOS, Intel              | from source  |     ✔³     |  ✔³  |        ✔ / ✔²³         | –                                    |
 | Windows x86_64            | release      |     ✔      |  ✔   |        ✔ / ✔²          | installer, zip                       |
-// The library never starts a runtime, installs a logger, handles signals,
+<sup>¹ Not in the release; builds from source (and through Nix).
-// forks, or calls process::exit. The binary owns all of that.
+² Newer and less battle-tested than on Linux.
-#[tokio::main]
+³ Builds from source, not covered by CI.</sup>
 async fn main() -> Result<()> {
    let config = AgentConfig::new(StoragePaths::user_default()?)
        .with_transport(TransportPolicy::N0Defaults)
        .with_discovery(Arc::new(SharedMemoryDiscovery::new()));
-    let agent = Agent::spawn(config).await?;
+The Linux CLI is a single static binary: it runs on any distribution.
-    let name = NetworkName::new("kitchen-table")?;
+## NixOS
    let secret = NetworkSecret::generate();       // 32 random bytes
    println!("share this: {}", secret.encode().as_str());
-    let network = agent.join_network(&name, &secret).await?;
+The repository is a flake with a package and a module.
-    let mut events = agent.subscribe();
+```nix
-    tokio::spawn(async move {
+# flake.nix
-        while let Ok(event) = events.recv().await {
+{
-            println!("{event:?}");
+  inputs.tsunagi.url = "github:house-of-vanity/tsunagi";
  outputs = { nixpkgs, tsunagi, ... }: {
    nixosConfigurations.myhost = nixpkgs.lib.nixosSystem {
      modules = [
        tsunagi.nixosModules.default
        {
          services.tsunagi = {
            enable = true;
            tray.enable = true;      # optional, for desktops
            users = [ "alice" ];     # may use tsng and the tray without sudo
          };
        }
-    });
+      ];
-
+    };
-    for peer in agent.network_status(network).await?.connected_peers() {
+  };
        agent
            .send(network, peer, ControlMessage::Ping { seq: 1, payload: vec![] })
            .await?;
    }
    agent.shutdown().await;
    Ok(())
 }
 ```
-`TransportPolicy::LocalOnly` is the default, so a plain `AgentConfig::new` never
+Just trying it: `nix run github:house-of-vanity/tsunagi#tsunagi-cli -- --help`.
-reaches the internet by accident. Opt into `DirectOnly` or `N0Defaults`
+`packages.<system>.tsunagi` (with the tray) and `tsunagi-cli` are also exported,
-explicitly.
+along with `overlays.default`.
-### Through somebody in the middle
+## Build from source
-Tsunagi routes the overlay in userspace. A direct data link always wins;
+Rust 1.91 or newer.
 otherwise the shortest available path can pass through several members.
 Each authenticated member advertises only its own live, protocol-specific
 transport links. The control plane collects these into a graph and replaces
 the routing table when connectivity changes; announcements expire after
 90 seconds without refresh.
 Transit goes transport → router → transport, without entering WireGuard or
 the intermediate host's TUN. The payload stays encrypted between the original
 endpoints. TUN connects only the local OS to Tsunagi; the kernel needs only
 the overlay route through that interface, with no IP forwarding configuration.
 A 16-hop limit bounds loops while topology updates converge. Equal shortest
 paths use a stable flow hash derived before encryption, so packets of a TCP
 or UDP flow keep their path while topology is unchanged. Physical link changes
 do not replace end-to-end WireGuard tunnels or change the 1280-byte default MTU.
 `status` reports the hop count and first next hop for relayed paths.
 The hot path reads an immutable routing snapshot and sends directly to a cached
 transport handle. It takes no routing mutex, walks no graph and does not parse
 the encrypted payload. See [routing.md](docs/routing.md) for the architecture,
 limits and reproducible forwarding microbenchmark.
 This wire format requires all members to upgrade together (control ALPN 3,
 data ALPN 4); saved identities, network names, secrets and addresses survive.
 ### LAN game discovery
 IPv4 UDP broadcast relay is enabled by default for each network. The IP router
 sends one encrypted copy to each participating peer, including through multihop
 paths. It supports `255.255.255.255` and the overlay subnet's broadcast address;
 received broadcasts never trigger another fanout.
 ```sh
-tsng join -n games --no-broadcast
+cargo build --release --bin tsng               # the agent and CLI
-tsng join -n games --broadcast
+cargo build --release --bin tsunagi-tray       # the tray (Linux needs GTK 3 and libayatana-appindicator)
 tsng network broadcast games off
 tsng network broadcast games on
 ```
-The choice persists across restarts and a plain `join`. It can be changed while
+Windows needs [Wintun](https://www.wintun.net) next to `tsng.exe`.
-the agent runs; `status` shows it for each network. While broadcast is on, the
+Per-platform notes: [Linux](dist/linux/README.md), [Arch](dist/arch/README.md),
-agent routes `255.255.255.255` through the Tsunagi interface and allows inbound
+[macOS](dist/macos/README.md), [Windows](dist/windows/README.md).
 UDP from the overlay range on it (Linux and Windows, needs the same privilege as
 the interface), and removes both when it is turned off or the agent stops. Games
 bound to a physical address still bypass it. Physical LAN capture/subnet sharing
 is not implemented.
 See [broadcast.md](docs/broadcast.md) for domain isolation and future LAN gateways.
-## How peers find each other
+## How it works
-Two different lookups are involved, and only one of them is this project's:
+- Your name and secret derive a network identity. Members publish signed
  contact records to the public Mainline DHT under it, and find each other there.
 - Every connection proves knowledge of the secret before it's accepted.
  Traffic between members is end-to-end encrypted, relays included.
 - Addresses and names are claimed with each device's own key and replicated
  between members. There is no one in charge, so nothing to be taken down.
-**1. Resolving one endpoint's address — iroh's, and it works today.**
+The secret is the only key, so make it a good one (the generated ones are).
-With `--reach relay` or `--reach direct`, iroh publishes a signed
+The [threat model](docs/threat-model.md) says what this does and doesn't protect against.
 record of this endpoint's addresses, keyed by its endpoint id, to the public
 service run by Number 0 — "n0", the company behind iroh — at `dns.iroh.link`,
 over pkarr and DNS, and resolves other endpoints the same way. That is why `--peer <endpoint-id>` works with no address attached:
 iroh looks it up. None of that code is ours.
 **2. Finding who is in a network — Mainline DHT by default.**
 `NetworkDiscovery` supplies unverified candidates. The Mainline backend uses
 a signing key derived from the network name and secret to publish signed
 BEP44 records containing each writer's endpoint ID and addresses. Its 16
 slots provide starting candidates, not a complete membership list or a limit
 on network size. After the first authenticated connection, introductions and
 signed state reveal the other members. `StaticBootstrap` (`--peer`) remains
 available, and tests use either a local Mainline Testnet or in-memory discovery.
 What this means in practice:
 - With `relay` or `direct`, **your endpoint id and IP addresses are published
  to a public third-party service** (Number 0's, unless you change it). They are not secret, and the network secret is
  never published, but an observer of that service learns that your endpoint
  exists and where it is. `--reach local` publishes nothing.
 - With DHT enabled, endpoint records are also published to the public Mainline
  DHT. They are signed, not encrypted; membership still requires the secret.
 - A relay, when one is needed, sees the volume and timing of your traffic — not
  its contents. The default relays are Number 0's, in the US, EU and
  Asia-Pacific.
 ## Storage
 Two physically separate SQLite files, placed wherever the library's
 configuration says (`StoragePaths`). A future system service supplies its own
 paths; tests always use temporary directories.
 | file           | holds                                            | when damaged            |
 |----------------|--------------------------------------------------|-------------------------|
 | `state.sqlite` | device identity, network configuration, hostname | clear error, never reset |
 | `cache.sqlite` | address hints and other recoverable data          | discarded and recreated  |
 The `wg-quic` protocol keeps its own keys in its own store under `wg-quic/`,
 because a protocol's keys are neither the iroh identity nor the network
 secret.
 The command line agent puts everything under the platform's per-user
 directories by default; `--state-dir` and `--cache-dir` override them.
 One state directory belongs to one live agent, enforced with a real OS file
 lock rather than an existence check. That directory *is* the identity: one
 agent, one device key, one interface, and as many networks as you like on it
 — `tsng network join` adds them to the agent that is already running,
 which is why a second `tsng up` on the same directory is refused rather
 than made to work. A second agent on the same host is a second identity, and
 needs everything of its own: its own state and cache directories, its own
 interface name and its own overlay range. The library owns no globals, so
 several of them run side by side in one process as readily as in one host.
 ### Leaving, and starting over
 Membership outlives a session, so it also has to be possible to end it.
 ```bash
 tsng network                     # what this device belongs to
 tsng network leave lab           # give up the address and the name, then forget it
 tsng wipe --yes                  # remove everything and be a stranger again
 ```
 **Stopping is not leaving.** `stop` closes this network's sessions, takes
 its address off the interface and keeps it from starting again, and that is
 all: the configuration, the secret, the signed state and the protocol key
 stay exactly as they are, nothing is announced, and to the others this
 device is simply away — an ordinary condition they already handle, with its
 address and name still reserved for it. `start` picks it up where it left
 off. A network named on the `up` command line is started by that command
 whatever its stored state, and the start-up banner says so rather than
 letting a `stop` quietly come back.
 `leave` is the other one, and it removes, locally: the configuration and
 the secret, this network's signed records, its cached address hints and the
 protocol key it used there. What it keeps is this author's version counter
 for that network — a rejoin has to continue above the release, or every
 replica would treat the new claim as stale. Everything about *other*
 networks, and the device identity itself, is untouched.
 `leave` publishes a signed `Release` **first**, while the agent is running and
 its sessions are up, so the address and the name are freed for the others
 instead of staying reserved to a member that has gone. They pass the tombstone
 on, so a member that was away hears it from them. With no agent running,
 nothing can sign or send it: the command says so and refuses, and `--offline`
 drops the network locally while leaving the others holding the old claim. The
 protocol key for that network goes too — rejoining is joining, not resuming.
 `leave`, `stop`, `start`, `secret` and `broadcast` accept an exact stored network
 name, its full id, or a unique id prefix (including a copied report ellipsis).
 Names are case-sensitive. If several networks share a name, or a name matches
 another network's id prefix, the command lists matching names and full ids and
 changes nothing. Use a full id to choose explicitly; a full id always takes
 precedence over a name. No secret is needed to select a configured network.
 `wipe` removes both directories' contents: the device identity, every network,
 every signed record and everything a protocol kept beside them. It refuses
 while an agent is running, and refuses a directory with no `state.sqlite` in
 it, so a mistyped `--state-dir` cannot take somebody's documents with it.
 Without `--yes` it only says what it would remove. It is not a goodbye: nobody
 is told, because after it there is no key left to sign anything with. Leave the
 networks first if the addresses should be freed.
 ## Documentation
- [docs/architecture.md](docs/architecture.md) — module boundaries and runtime.
+[Architecture](docs/architecture.md) ·
- [docs/wireguard.md](docs/wireguard.md) — the WireGuard plugin: overlay
+[Protocol](docs/protocol.md) ·
-  addressing, announcements, backends, reconciliation.
+[WireGuard data plane](docs/wireguard.md) ·
- [docs/protocol.md](docs/protocol.md) — identity derivation, framing, handshake.
+[Routing](docs/routing.md) ·
- [docs/sync-model.md](docs/sync-model.md) — the planned signed-state model and
+[Exit nodes](docs/exit-node.md) ·
-  what is deliberately not built yet.
+[Threat model](docs/threat-model.md) ·
- [docs/exit-node.md](docs/exit-node.md) — exit nodes: sending all traffic through a member.
+[Contributing](AGENTS.md)
 - [docs/threat-model.md](docs/threat-model.md) — threat model and known limits.
 - [docs/testing.md](docs/testing.md) — what the suite covers and what it does not.
 - [AGENTS.md](AGENTS.md) — rules for anyone (human or agent) changing this repo.
 ## Security in one paragraph
 Membership is proved by an HMAC over a transcript keyed by a value derived from
 the shared secret, bound to the specific iroh connection through the TLS
 exporter, to the network id, to both endpoint identities and to distinct role
 labels. This targets high-entropy secrets: there is no PAKE here, so a short
 human passphrase is guessable offline by anyone who can reach the handshake.
 Anyone who knows the secret is a full participant and can create many
 identities. Read [docs/threat-model.md](docs/threat-model.md) before relying on
 any of this.
 ## Licence
@@ -800,6 +800,11 @@ impl Agent {
        // itself stays, because the agent owns it and other networks may
        // still be using it.
        self.inner.routes.remove_network(network_id);
        if let Some(interface) = self.inner.interface.get() {
            interface
                .set_underlay(network_id, Vec::new(), Vec::new())
                .await;
        }
        if let Some(interface) = self.inner.interface.get()
            && let Err(err) = interface.sync_addresses().await
        {
@@ -712,6 +712,34 @@ impl Runtime {
        self.start_dials();
        // Last, so it carries what this round learned.
        self.introduce_peers();
        self.refresh_underlay().await;
    }
    /// Hands the interface the underlay destinations this network depends
    /// on: the verified direct IPv4 paths of every control connection, and
    /// the relays in use. Only hosts that cannot exempt the agent from an
    /// exit node's default route read it.
    async fn refresh_underlay(&self) {
        let Some(interface) = self.params.interface.clone() else {
            return;
        };
        let mut addresses = Vec::new();
        let mut hosts = Vec::new();
        for session in self.sessions.values() {
            for path in snapshot_connection(&session.conn).paths {
                match path.remote {
                    PathAddr::Ip(std::net::SocketAddr::V4(socket)) => addresses.push(*socket.ip()),
                    PathAddr::Relay(url) => hosts.extend(relay_host(&url)),
                    _ => {}
                }
            }
        }
        for url in self.params.adapter.snapshot().relay_urls {
            hosts.extend(relay_host(&url));
        }
        interface
            .set_underlay(self.network_id, addresses, hosts)
            .await;
    }
    fn add_candidate(&mut self, candidate: Candidate) {
@@ -2448,3 +2476,48 @@ mod tests {
        assert_eq!(chosen_restored.copied(), Some("wg"));
    }
 }
 /// The host name of a relay URL such as `https://relay.example:443/`.
 fn relay_host(url: &str) -> Option<String> {
    let rest = url.split_once("://").map_or(url, |(_, rest)| rest);
    let authority = rest.split(['/', '?', '#']).next()?;
    let host = authority.rsplit_once('@').map_or(authority, |(_, h)| h);
    let host = if host.starts_with('[') {
        host.split_once(']')?.0.trim_start_matches('[')
    } else {
        host.split(':').next()?
    };
    (!host.is_empty()
        && host.len() <= 253
        && host
            .chars()
            .all(|c| c.is_ascii_alphanumeric() || c == '.' || c == '-' || c == ':'))
    .then(|| host.to_ascii_lowercase())
 }
 #[cfg(test)]
 mod relay_host_tests {
    use super::relay_host;
    #[test]
    fn the_host_is_taken_out_of_a_relay_url() {
        assert_eq!(
            relay_host("https://relay.example.com./").as_deref(),
            Some("relay.example.com.")
        );
        assert_eq!(
            relay_host("https://Euw1-1.relay.iroh.network:443/x").as_deref(),
            Some("euw1-1.relay.iroh.network")
        );
        assert_eq!(
            relay_host("http://192.0.2.7:3340").as_deref(),
            Some("192.0.2.7")
        );
        assert_eq!(
            relay_host("https://[2001:db8::1]:443/").as_deref(),
            Some("2001:db8::1")
        );
        assert_eq!(relay_host("https:///").as_deref(), None);
        assert_eq!(relay_host("https://bad host/").as_deref(), None);
    }
 }
@@ -39,6 +39,13 @@ pub struct ExitHostPlan {
    /// reads it, to keep this agent's own traffic *to* the overlay inside it
    /// where the host has no per-user routing (macOS).
    pub overlay: Vec<Ipv4Range>,
    /// Underlay addresses this agent talks to directly and must keep reaching
    /// outside the overlay's default route. Only filled while `client` is set
    /// and only for hosts that report [`ExitHostRules::needs_bypass`].
    pub bypass: Vec<std::net::Ipv4Addr>,
    /// Host names (relay servers) that must stay reachable the same way; the
    /// host rules resolve them.
    pub bypass_hosts: Vec<String>,
 }
 impl ExitHostPlan {
@@ -76,6 +83,13 @@ pub trait ExitHostRules: Send + Sync + std::fmt::Debug + 'static {
    /// when nothing is there: it runs on the shutdown path, and once at
    /// startup to take away what a crashed run left.
    fn clear<'a>(&'a self, interface: &'a str) -> BoxFuture<'a, ()>;
    /// Whether using an exit node on this host needs the underlay addresses
    /// the agent talks to, because the host cannot exempt the agent's own
    /// traffic from the default route by user id (Windows). The default is no.
    fn needs_bypass(&self) -> bool {
        false
    }
 }
 /// The tag every firewall object of one offered range carries.
@@ -103,6 +117,7 @@ pub struct MockExitRules {
    calls: Arc<Mutex<Vec<String>>>,
    forwarding: Arc<Mutex<Option<bool>>>,
    failure: Arc<Mutex<Option<String>>>,
    needs_bypass: Arc<Mutex<bool>>,
 }
 impl MockExitRules {
@@ -118,6 +133,11 @@ impl MockExitRules {
        *lock(&self.forwarding) = forwarding;
    }
    /// Whether this host asks for the underlay addresses, like Windows.
    pub fn set_needs_bypass(&self, needs: bool) {
        *lock(&self.needs_bypass) = needs;
    }
    /// Makes every step fail with this reason from now on.
    pub fn fail_with(&self, reason: Option<String>) {
        *lock(&self.failure) = reason;
@@ -142,6 +162,10 @@ fn lock<T>(mutex: &Mutex<T>) -> std::sync::MutexGuard<'_, T> {
 }
 impl ExitHostRules for MockExitRules {
    fn needs_bypass(&self) -> bool {
        *lock(&self.needs_bypass)
    }
    fn apply<'a>(&'a self, plan: &'a ExitHostPlan) -> BoxFuture<'a, ExitHostReport> {
        Box::pin(async move {
            lock(&self.calls).push(format!("apply:{}", plan.interface));
@@ -209,6 +233,7 @@ mod tests {
            offer: vec![range()],
            client: true,
            overlay: vec![range()],
            ..Default::default()
        };
        let report = rules.apply(&plan).await;
        assert_eq!(report.offer[0].1, RuleOutcome::Applied);
@@ -224,6 +249,7 @@ mod tests {
                offer: vec![range()],
                client: false,
                overlay: Vec::new(),
                ..Default::default()
            })
            .await;
        assert_eq!(report.forwarding, Some(false));
@@ -623,6 +623,7 @@ destination: default\n\
            offer: if offer { vec![range()] } else { Vec::new() },
            client,
            overlay: vec![range()],
            ..Default::default()
        }
    }
@@ -162,12 +162,36 @@ impl WindowsHostRules {
    }
 }
 /// The IPv4 addresses a host name resolves to; empty when it does not.
 fn resolve_host(host: &str) -> Vec<std::net::Ipv4Addr> {
    use std::net::{IpAddr, ToSocketAddrs};
    (host, 443)
        .to_socket_addrs()
        .map(|addrs| {
            addrs
                .filter_map(|addr| match addr.ip() {
                    IpAddr::V4(ip) => Some(ip),
                    IpAddr::V6(_) => None,
                })
                .collect()
        })
        .unwrap_or_default()
 }
 impl ExitHostRules for WindowsHostRules {
    fn needs_bypass(&self) -> bool {
        true
    }
    fn apply<'a>(&'a self, plan: &'a ExitHostPlan) -> BoxFuture<'a, ExitHostReport> {
        Box::pin(async move {
            let job = plan.clone();
            tokio::task::spawn_blocking(move || {
-                windows_exit::apply_plan(&|script| run_powershell_output(script), &job)
+                windows_exit::apply_plan(
                    &|script| run_powershell_output(script),
                    &resolve_host,
                    &job,
                )
            })
            .await
            .unwrap_or_else(|err| {
@@ -23,24 +23,46 @@
 //!
 //! # Using
 //!
-//! Not available. Windows has no per-user or per-process routing, so the
+//! Windows has no per-user or per-process routing, so the Linux scheme (an
-//! agent's own QUIC, relay and DHT traffic cannot be kept out of a default
+//! exemption for the agent's user) does not carry over. Instead the agent's
-//! route through the overlay: it would loop into the tunnel it carries. Rather
+//! own underlay traffic is exempted by *destination*:
-//! than install something that stops the agent working, the using side
+//!
-//! reports why it is not in place and changes nothing.
+//! ```text
 //! New-NetRoute 0.0.0.0/1   on the overlay interface, metric 1
 //! New-NetRoute 128.0.0.0/1 on the overlay interface, metric 1
 //! New-NetRoute <ip>/32 via the physical default gateway, metric 7373
 //! ```
 //!
 //! The halves out-rank the default route without replacing it, so every more
 //! specific route (LAN, another VPN) keeps working. A `/32` per address the
 //! agent talks to directly (the verified direct paths of its connections, and
 //! the relays it uses, resolved when the rules are applied) keeps that traffic
 //! on the physical network instead of looping into the tunnel it carries. The
 //! metric 7373 marks those routes as ours, which is how stale ones are found
 //! and removed. Everything is written to the active store, so a reboot clears
 //! it. IPv6 is blackholed with `::/1` and `8000::/1` on the overlay interface,
 //! which has no IPv6: applications fall back to IPv4 after a delay, where
 //! Linux and macOS answer at once. The bypass follows the agent's paths as
 //! they change; a path that appears after the last apply is reached through
 //! the tunnel until the next one.
 //!
 //! The bypass routes are written before the halves, and removed after them.
 use super::RuleOutcome;
 use std::net::Ipv4Addr;
 use super::exit::{ExitHostPlan, ExitHostReport, exit_tag, exit_tag_prefix};
 use crate::state::Ipv4Range;
-/// Why the using side is not installed.
+/// The route metric that marks a bypass route as this agent's.
-pub(super) const CLIENT_UNAVAILABLE: &str = "using an exit node is not available on Windows yet: \
+const BYPASS_METRIC: u32 = 7373;
     Windows cannot keep the agent's own traffic out of a default route through the overlay, \
     so it would loop";
 /// Runs a PowerShell script and returns what it printed.
 pub(super) type RunScript<'a> = &'a dyn Fn(&str) -> Result<String, String>;
 /// Resolves a host name to its IPv4 addresses.
 pub(super) type Resolve<'a> = &'a dyn Fn(&str) -> Vec<Ipv4Addr>;
 /// A PowerShell single-quoted literal, with any embedded quote doubled.
 fn ps_literal(value: &str) -> String {
    format!("'{}'", value.replace('\'', "''"))
@@ -86,6 +108,125 @@ fn forwarding_script(interface: &str) -> String {
    )
 }
 /// Prints `<interface index> <next hop>` of the best default route that is
 /// not through the overlay interface.
 fn gateway_script(interface: &str) -> String {
    let alias = ps_literal(interface);
    format!(
        "$ErrorActionPreference = 'Stop'\n\
         $alias = {alias}\n\
         $best = Get-NetRoute -DestinationPrefix '0.0.0.0/0' -AddressFamily IPv4 \
         -PolicyStore ActiveStore | Where-Object {{ $_.InterfaceAlias -ne $alias -and \
         $_.NextHop -ne '0.0.0.0' }} | Sort-Object {{ $_.RouteMetric + \
         (Get-NetIPInterface -InterfaceIndex $_.InterfaceIndex -AddressFamily IPv4).InterfaceMetric }} | \
         Select-Object -First 1\n\
         if (-not $best) {{ throw 'the host has no default gateway outside the overlay' }}\n\
         \"$($best.InterfaceIndex) $($best.NextHop)\"\n"
    )
 }
 /// The default gateway as the script above printed it.
 fn parse_gateway(text: &str) -> Option<(u32, Ipv4Addr)> {
    let mut words = text.split_whitespace();
    let index = words.next()?.parse().ok()?;
    let hop = words.next()?.parse().ok()?;
    words.next().is_none().then_some((index, hop))
 }
 /// Replaces the bypass routes with one per address, through the gateway.
 fn bypass_script(index: u32, gateway: Ipv4Addr, addresses: &[Ipv4Addr]) -> String {
    let list: Vec<String> = addresses
        .iter()
        .map(|address| ps_literal(&address.to_string()))
        .collect();
    format!(
        "$ErrorActionPreference = 'Stop'\n\
         Get-NetRoute -AddressFamily IPv4 -PolicyStore ActiveStore -ErrorAction SilentlyContinue | \
         Where-Object {{ $_.RouteMetric -eq {BYPASS_METRIC} -and $_.DestinationPrefix.EndsWith('/32') }} | \
         Remove-NetRoute -Confirm:$false -ErrorAction SilentlyContinue\n\
         foreach ($ip in @({})) {{\n\
         New-NetRoute -DestinationPrefix \"$ip/32\" -InterfaceIndex {index} -NextHop '{gateway}' \
         -RouteMetric {BYPASS_METRIC} -PolicyStore ActiveStore -ErrorAction SilentlyContinue | Out-Null\n\
         }}\n",
        list.join(", ")
    )
 }
 /// Sends everything else through the overlay interface.
 fn halves_script(interface: &str) -> String {
    let alias = ps_literal(interface);
    format!(
        "$ErrorActionPreference = 'Stop'\n\
         $alias = {alias}\n\
         foreach ($p in '0.0.0.0/1', '128.0.0.0/1') {{\n\
         Get-NetRoute -DestinationPrefix $p -InterfaceAlias $alias -PolicyStore ActiveStore \
         -ErrorAction SilentlyContinue | Remove-NetRoute -Confirm:$false -ErrorAction SilentlyContinue\n\
         New-NetRoute -DestinationPrefix $p -InterfaceAlias $alias -NextHop '0.0.0.0' \
         -RouteMetric 1 -PolicyStore ActiveStore | Out-Null\n\
         }}\n\
         foreach ($p in '::/1', '8000::/1') {{\n\
         Get-NetRoute -DestinationPrefix $p -InterfaceAlias $alias -PolicyStore ActiveStore \
         -ErrorAction SilentlyContinue | Remove-NetRoute -Confirm:$false -ErrorAction SilentlyContinue\n\
         New-NetRoute -DestinationPrefix $p -InterfaceAlias $alias -NextHop '::' \
         -RouteMetric 1 -PolicyStore ActiveStore -ErrorAction SilentlyContinue | Out-Null\n\
         }}\n"
    )
 }
 /// Removes the halves and every bypass route. What is not there is the state
 /// wanted.
 fn remove_client_script(interface: &str) -> String {
    let alias = ps_literal(interface);
    format!(
        "$ErrorActionPreference = 'SilentlyContinue'\n\
         $alias = {alias}\n\
         foreach ($p in '0.0.0.0/1', '128.0.0.0/1', '::/1', '8000::/1') {{\n\
         Get-NetRoute -DestinationPrefix $p -InterfaceAlias $alias -PolicyStore ActiveStore | \
         Remove-NetRoute -Confirm:$false\n\
         }}\n\
         Get-NetRoute -AddressFamily IPv4 -PolicyStore ActiveStore | \
         Where-Object {{ $_.RouteMetric -eq {BYPASS_METRIC} -and $_.DestinationPrefix.EndsWith('/32') }} | \
         Remove-NetRoute -Confirm:$false\n\
         exit 0\n"
    )
 }
 /// Every address to keep direct: the ones given and what the names resolve
 /// to, sorted and without duplicates.
 fn bypass_addresses(
    resolve: Resolve<'_>,
    addresses: &[Ipv4Addr],
    hosts: &[String],
 ) -> Vec<Ipv4Addr> {
    let mut all: std::collections::BTreeSet<Ipv4Addr> = addresses.iter().copied().collect();
    for host in hosts {
        all.extend(resolve(host));
    }
    all.into_iter()
        .filter(|a| !a.is_loopback() && !a.is_unspecified() && !a.is_multicast())
        .collect()
 }
 fn apply_client(
    run: RunScript<'_>,
    resolve: Resolve<'_>,
    plan: &ExitHostPlan,
 ) -> Result<(), String> {
    let interface = plan.interface.as_str();
    let gateway = run(&gateway_script(interface))?;
    let (index, hop) = parse_gateway(&gateway)
        .ok_or_else(|| format!("cannot read the default gateway from `{}`", gateway.trim()))?;
    let addresses = bypass_addresses(resolve, &plan.bypass, &plan.bypass_hosts);
    run(&bypass_script(index, hop, &addresses))?;
    if let Err(why) = run(&halves_script(interface)) {
        // Do not leave a half-installed set: bypass routes alone are harmless
        // but the next apply starts clean.
        let _ = run(&remove_client_script(interface));
        return Err(why);
    }
    Ok(())
 }
 fn parse_forwarding(text: &str) -> Option<bool> {
    match text.trim().to_ascii_lowercase().as_str() {
        "enabled" | "1" => Some(true),
@@ -95,7 +236,11 @@ fn parse_forwarding(text: &str) -> Option<bool> {
 }
 /// Makes the host match the plan.
-pub(super) fn apply_plan(run: RunScript<'_>, plan: &ExitHostPlan) -> ExitHostReport {
+pub(super) fn apply_plan(
    run: RunScript<'_>,
    resolve: Resolve<'_>,
    plan: &ExitHostPlan,
 ) -> ExitHostReport {
    let interface = plan.interface.as_str();
    // What an earlier plan left and this one no longer wants.
    let pruned = run(&prune_script(interface, &plan.offer));
@@ -113,6 +258,16 @@ pub(super) fn apply_plan(run: RunScript<'_>, plan: &ExitHostPlan) -> ExitHostRep
            (*range, outcome)
        })
        .collect();
    let client = if plan.client {
        Some(match apply_client(run, resolve, plan) {
            Ok(()) => RuleOutcome::Applied,
            Err(why) => RuleOutcome::Failed(why),
        })
    } else {
        // Whatever an earlier plan left must not stay in force.
        let _ = run(&remove_client_script(interface));
        None
    };
    let forwarding = if plan.offer.is_empty() {
        None
    } else {
@@ -123,9 +278,7 @@ pub(super) fn apply_plan(run: RunScript<'_>, plan: &ExitHostPlan) -> ExitHostRep
    ExitHostReport {
        offer,
        forwarding,
-        client: plan
+        client,
            .client
            .then(|| RuleOutcome::Failed(CLIENT_UNAVAILABLE.to_string())),
    }
 }
@@ -134,6 +287,9 @@ pub(super) fn clear_all(run: RunScript<'_>, interface: &str) {
    if let Err(why) = run(&prune_script(interface, &[])) {
        tracing::warn!(interface, "cannot remove the exit node NAT: {why}");
    }
    if let Err(why) = run(&remove_client_script(interface)) {
        tracing::warn!(interface, "cannot remove the exit node routes: {why}");
    }
 }
 #[cfg(test)]
@@ -154,6 +310,7 @@ mod tests {
            offer: vec![range()],
            client,
            overlay: vec![range()],
            ..Default::default()
        }
    }
@@ -210,14 +367,16 @@ mod tests {
                String::new()
            })
        };
-        let report = apply_plan(&run, &plan(false));
+        let report = apply_plan(&run, &no_names, &plan(false));
        assert_eq!(report.offer, vec![(range(), RuleOutcome::Applied)]);
        assert_eq!(report.forwarding, Some(false));
        assert_eq!(report.client, None);
        let calls = calls.borrow();
        assert!(calls[0].contains("StartsWith"), "{calls:?}");
        assert!(calls[1].contains("New-NetNat"), "{calls:?}");
-        assert!(calls[2].contains("Get-NetIPInterface"), "{calls:?}");
+        // Without a client, the routes of an earlier plan are taken away.
        assert!(calls[2].contains("RouteMetric -eq 7373"), "{calls:?}");
        assert!(calls[3].contains("Get-NetIPInterface"), "{calls:?}");
    }
    #[test]
@@ -229,39 +388,216 @@ mod tests {
                Ok(String::new())
            }
        };
-        let report = apply_plan(&run, &plan(false));
+        let report = apply_plan(&run, &no_names, &plan(false));
        let RuleOutcome::Failed(why) = &report.offer[0].1 else {
            panic!("{report:?}");
        };
        assert!(why.contains("already exists"), "{why}");
    }
    fn no_names(_: &str) -> Vec<Ipv4Addr> {
        Vec::new()
    }
    fn client_plan() -> ExitHostPlan {
        ExitHostPlan {
            interface: "tsun0".into(),
            offer: Vec::new(),
            client: true,
            overlay: vec![range()],
            bypass: vec![Ipv4Addr::new(138, 201, 61, 182)],
            bypass_hosts: vec!["relay.example".into()],
        }
    }
    #[test]
-    fn using_an_exit_node_is_refused_and_changes_nothing() {
+    fn the_gateway_is_read_from_the_script_output() {
-        let calls = RefCell::new(0usize);
+        assert_eq!(
-        let run = |script: &str| -> Result<String, String> {
+            parse_gateway("12 192.168.1.1\r\n"),
-            *calls.borrow_mut() += 1;
+            Some((12, Ipv4Addr::new(192, 168, 1, 1)))
-            // No script may touch routes.
+        );
-            assert!(!script.contains("Route"), "{script}");
+        assert_eq!(parse_gateway(""), None);
-            Ok(String::new())
+        assert_eq!(parse_gateway("12"), None);
        assert_eq!(parse_gateway("12 gw"), None);
        assert_eq!(parse_gateway("12 1.1.1.1 extra"), None);
        let script = gateway_script("tsun0");
        assert!(script.contains("-ne $alias"));
        assert!(script.contains("-ne '0.0.0.0'"));
    }
    #[test]
    fn bypass_routes_are_marked_host_routes_through_the_physical_gateway() {
        let script = bypass_script(
            12,
            Ipv4Addr::new(192, 168, 1, 1),
            &[Ipv4Addr::new(1, 2, 3, 4), Ipv4Addr::new(5, 6, 7, 8)],
        );
        assert!(script.contains("@('1.2.3.4', '5.6.7.8')"), "{script}");
        assert!(script.contains("-InterfaceIndex 12 -NextHop '192.168.1.1'"));
        assert!(script.contains("-RouteMetric 7373"));
        assert!(script.contains("-PolicyStore ActiveStore"));
        // Stale marked routes go first, and only /32 ones with our metric.
        assert!(script.find("Remove-NetRoute").unwrap() < script.find("New-NetRoute").unwrap());
        assert!(script.contains("RouteMetric -eq 7373"));
        assert!(script.contains("EndsWith('/32')"));
    }
    #[test]
    fn the_halves_go_through_the_overlay_and_ipv6_is_blackholed() {
        let script = halves_script("tsun0");
        assert!(script.contains("'0.0.0.0/1', '128.0.0.0/1'"));
        assert!(script.contains("-InterfaceAlias $alias -NextHop '0.0.0.0'"));
        assert!(script.contains("'::/1', '8000::/1'"));
        assert!(script.contains("-PolicyStore ActiveStore"));
        assert!(
            !script.contains("0.0.0.0/0"),
            "the default route is left alone"
        );
        assert!(halves_script("x'y").contains("'x''y'"));
    }
    #[test]
    fn addresses_are_resolved_merged_and_cleaned() {
        let resolve = |host: &str| -> Vec<Ipv4Addr> {
            match host {
                "relay.example" => vec![Ipv4Addr::new(9, 9, 9, 9), Ipv4Addr::new(1, 2, 3, 4)],
                _ => Vec::new(),
            }
        };
-        let report = apply_plan(&run, &plan(true));
+        let all = bypass_addresses(
            &resolve,
            &[
                Ipv4Addr::new(1, 2, 3, 4),
                Ipv4Addr::LOCALHOST,
                Ipv4Addr::UNSPECIFIED,
            ],
            &["relay.example".into(), "gone.example".into()],
        );
        assert_eq!(
            all,
            vec![Ipv4Addr::new(1, 2, 3, 4), Ipv4Addr::new(9, 9, 9, 9)]
        );
    }
    #[test]
    fn the_bypass_is_written_before_the_halves_that_need_it() {
        let calls = RefCell::new(Vec::new());
        let run = |script: &str| -> Result<String, String> {
            calls.borrow_mut().push(script.to_string());
            Ok(if script.contains("Select-Object -First 1") {
                "12 192.168.1.1\n".to_string()
            } else {
                String::new()
            })
        };
        let resolve = |_: &str| vec![Ipv4Addr::new(9, 9, 9, 9)];
        let report = apply_plan(&run, &resolve, &client_plan());
        assert_eq!(report.client, Some(RuleOutcome::Applied));
        let calls = calls.borrow();
        let at = |needle: &str| calls.iter().position(|c| c.contains(needle)).unwrap();
        assert!(at("Select-Object -First 1") < at("-RouteMetric 7373"));
        assert!(at("-RouteMetric 7373") < at("'0.0.0.0/1', '128.0.0.0/1'"));
        let bypass = &calls[at("-RouteMetric 7373")];
        assert!(
            bypass.contains("'9.9.9.9'") && bypass.contains("'138.201.61.182'"),
            "{bypass}"
        );
    }
    #[test]
    fn nothing_goes_through_the_overlay_when_the_bypass_cannot_be_written() {
        let run = |script: &str| -> Result<String, String> {
            if script.contains("Select-Object -First 1") {
                Ok("12 192.168.1.1".to_string())
            } else if script.contains("-RouteMetric 7373") && script.contains("foreach ($ip") {
                Err("Access is denied".to_string())
            } else {
                assert!(!script.contains("-NextHop '0.0.0.0'"), "{script}");
                Ok(String::new())
            }
        };
        let report = apply_plan(&run, &no_names, &client_plan());
        assert_eq!(
            report.client,
            Some(RuleOutcome::Failed("Access is denied".into()))
        );
    }
    #[test]
    fn a_missing_gateway_is_a_failure_that_installs_no_route() {
        let run = |script: &str| -> Result<String, String> {
            assert!(!script.contains("New-NetRoute"), "{script}");
            if script.contains("Select-Object -First 1") {
                Err("the host has no default gateway outside the overlay".into())
            } else {
                Ok(String::new())
            }
        };
        let report = apply_plan(&run, &no_names, &client_plan());
        let Some(RuleOutcome::Failed(why)) = report.client else {
            panic!("{report:?}");
        };
-        assert!(why.contains("not available on Windows"), "{why}");
+        assert!(why.contains("no default gateway"), "{why}");
-        assert!(why.contains("loop"), "{why}");
+    }
    #[test]
    fn a_failed_halves_step_removes_what_was_written() {
        let calls = RefCell::new(Vec::new());
        let run = |script: &str| -> Result<String, String> {
            calls.borrow_mut().push(script.to_string());
            if script.contains("Select-Object -First 1") {
                Ok("12 192.168.1.1".to_string())
            } else if script.contains("'128.0.0.0/1'") && script.contains("New-NetRoute") {
                Err("boom".to_string())
            } else {
                Ok(String::new())
            }
        };
        let report = apply_plan(&run, &no_names, &client_plan());
        assert!(matches!(report.client, Some(RuleOutcome::Failed(_))));
        assert!(calls.borrow().last().unwrap().contains("Remove-NetRoute"));
    }
    #[test]
    fn dropping_the_client_removes_its_routes_and_clearing_does_too() {
        let calls = RefCell::new(Vec::new());
        let run = |script: &str| -> Result<String, String> {
            calls.borrow_mut().push(script.to_string());
            Ok(String::new())
        };
        let report = apply_plan(&run, &no_names, &plan(false));
        assert_eq!(report.client, None);
        assert!(
            calls
                .borrow()
                .iter()
                .any(|c| c.contains("RouteMetric -eq 7373"))
        );
        let script = remove_client_script("tsun0");
        assert!(script.contains("'0.0.0.0/1', '128.0.0.0/1', '::/1', '8000::/1'"));
        assert!(script.contains("exit 0"));
        // Only routes on our interface, or marked ours, are touched.
        assert!(script.contains("-InterfaceAlias $alias"));
        calls.borrow_mut().clear();
        clear_all(&run, "tsun0");
        assert!(
            calls
                .borrow()
                .iter()
                .any(|c| c.contains("RouteMetric -eq 7373"))
        );
    }
    #[test]
    fn clearing_prunes_everything_for_the_interface() {
-        let seen = RefCell::new(String::new());
+        let seen = RefCell::new(Vec::<String>::new());
        let run = |script: &str| -> Result<String, String> {
-            *seen.borrow_mut() = script.to_string();
+            seen.borrow_mut().push(script.to_string());
            Ok(String::new())
        };
        clear_all(&run, "tsun0");
-        assert!(seen.borrow().contains("$keep = @()"));
+        let seen = seen.borrow();
-        assert!(seen.borrow().contains("'tsunagi-exit:tsun0:'"));
+        assert!(seen[0].contains("$keep = @()"));
        assert!(seen[0].contains("'tsunagi-exit:tsun0:'"));
    }
 }
@@ -164,6 +164,17 @@ pub struct Interface {
    rules_report: std::sync::Mutex<Option<BroadcastRulesReport>>,
    /// What the exit-node rules were last asked for, and what came of it.
    exit_state: std::sync::Mutex<ExitState>,
    /// Where each network's runtime reaches its peers and relays outside the
    /// overlay; read only by hosts that cannot exempt the agent from the
    /// exit node's default route.
    underlay: std::sync::Mutex<std::collections::HashMap<NetworkId, Underlay>>,
 }
 /// Underlay destinations one network's runtime currently depends on.
 #[derive(Debug, Default, Clone, PartialEq, Eq)]
 struct Underlay {
    addresses: std::collections::BTreeSet<std::net::Ipv4Addr>,
    hosts: std::collections::BTreeSet<String>,
 }
 /// The exit-node host rules as this interface last left them.
@@ -275,6 +286,7 @@ impl Interface {
            host_rules,
            rules_report: std::sync::Mutex::new(None),
            exit_state: std::sync::Mutex::new(ExitState::default()),
            underlay: std::sync::Mutex::new(std::collections::HashMap::new()),
        })
    }
@@ -350,11 +362,20 @@ impl Interface {
            .collect();
        ranges.sort_by_key(|range| (range.base, range.prefix_len));
        ranges.dedup();
        let client = self.routes.exit_via().is_some();
        let overlay = self.routes.overlay_ranges();
        let (bypass, bypass_hosts) = if client && exit.needs_bypass() {
            self.underlay_targets(&overlay)
        } else {
            (Vec::new(), Vec::new())
        };
        let wanted = ExitHostPlan {
            interface: self.name.clone(),
            offer: ranges,
-            client: self.routes.exit_via().is_some(),
+            client,
-            overlay: self.routes.overlay_ranges(),
+            overlay,
            bypass,
            bypass_hosts,
        };
        let first = !std::mem::replace(&mut self.exit_state().cleaned, true);
@@ -426,6 +447,68 @@ impl Interface {
        state.report = Some(report);
    }
    /// Tells the interface where a network's runtime reaches peers and
    /// relays outside the overlay. When the host needs those kept out of the
    /// exit node's default route and the set changed, the rules follow.
    pub async fn set_underlay(
        &self,
        network: NetworkId,
        addresses: impl IntoIterator<Item = std::net::Ipv4Addr>,
        hosts: impl IntoIterator<Item = String>,
    ) {
        let new = Underlay {
            addresses: addresses.into_iter().collect(),
            hosts: hosts.into_iter().collect(),
        };
        let changed = {
            let mut map = match self.underlay.lock() {
                Ok(guard) => guard,
                Err(poisoned) => poisoned.into_inner(),
            };
            if new == Underlay::default() {
                map.remove(&network).is_some()
            } else {
                map.insert(network, new.clone()) != Some(new)
            }
        };
        let needed = changed
            && self.routes.exit_via().is_some()
            && self
                .host_rules
                .as_ref()
                .and_then(|rules| rules.exit_rules())
                .is_some_and(|exit| exit.needs_bypass());
        if needed {
            self.sync_exit_rules().await;
        }
    }
    /// Every underlay destination worth keeping direct, minus anything that
    /// is the overlay itself, loopback or link-local.
    fn underlay_targets(
        &self,
        overlay: &[crate::state::Ipv4Range],
    ) -> (Vec<std::net::Ipv4Addr>, Vec<String>) {
        let map = match self.underlay.lock() {
            Ok(guard) => guard,
            Err(poisoned) => poisoned.into_inner(),
        };
        let mut addresses = std::collections::BTreeSet::new();
        let mut hosts = std::collections::BTreeSet::new();
        for underlay in map.values() {
            addresses.extend(underlay.addresses.iter().copied().filter(|address| {
                !address.is_loopback()
                    && !address.is_link_local()
                    && !address.is_multicast()
                    && !address.is_unspecified()
                    && !address.is_broadcast()
                    && !overlay.iter().any(|range| range.contains(*address))
            }));
            hosts.extend(underlay.hosts.iter().cloned());
        }
        (addresses.into_iter().collect(), hosts.into_iter().collect())
    }
    fn exit_state(&self) -> std::sync::MutexGuard<'_, ExitState> {
        match self.exit_state.lock() {
            Ok(guard) => guard,
@@ -1343,6 +1426,65 @@ mod tests {
        assert_eq!(interface.counters().received, 1);
    }
    #[tokio::test]
    async fn the_underlay_reaches_the_rules_only_for_hosts_that_need_it_and_only_while_using_one() {
        let rules = MockHostRules::new();
        let (interface, routes, id) = managed_interface(&rules, false).await;
        let relay = || vec!["relay.example".to_string()];
        let direct = || {
            vec![
                Ipv4Addr::new(138, 201, 61, 182),
                Ipv4Addr::new(10, 13, 37, 9), // the overlay itself
                Ipv4Addr::LOCALHOST,
                Ipv4Addr::new(169, 254, 1, 1),
            ]
        };
        // A host that does not ask never gets them, so Linux and macOS do not
        // re-apply their rules whenever a path moves.
        routes.set_exit(id, using(peer(2)));
        interface.set_underlay(id, direct(), relay()).await;
        interface.sync_exit_rules().await;
        let plan = rules.exit().installed().unwrap();
        assert!(plan.client);
        assert!(plan.bypass.is_empty() && plan.bypass_hosts.is_empty());
        // A host that does gets the usable ones, and follows changes.
        rules.exit().set_needs_bypass(true);
        interface.set_underlay(id, direct(), relay()).await;
        interface.sync_exit_rules().await;
        let plan = rules.exit().installed().unwrap();
        assert_eq!(plan.bypass, vec![Ipv4Addr::new(138, 201, 61, 182)]);
        assert_eq!(plan.bypass_hosts, relay());
        let applies = || {
            rules
                .exit()
                .calls()
                .iter()
                .filter(|c| c.starts_with("apply"))
                .count()
        };
        let before = applies();
        interface.set_underlay(id, direct(), relay()).await;
        assert_eq!(applies(), before, "an unchanged set is not applied again");
        interface
            .set_underlay(id, vec![Ipv4Addr::new(5, 6, 7, 8)], relay())
            .await;
        assert_eq!(applies(), before + 1);
        assert_eq!(
            rules.exit().installed().unwrap().bypass,
            vec![Ipv4Addr::new(5, 6, 7, 8)]
        );
        // Not using an exit node: nothing to keep direct.
        routes.set_exit(id, Default::default());
        interface.sync_exit_rules().await;
        interface
            .set_underlay(id, vec![Ipv4Addr::new(1, 1, 1, 1)], relay())
            .await;
        assert!(rules.exit().installed().is_none());
    }
    #[tokio::test]
    async fn the_exit_rules_follow_the_routing_table_and_leave_with_the_interface() {
        let rules = MockHostRules::new();
@@ -34,6 +34,9 @@ netwatch = "0.19.3"
 [target.'cfg(unix)'.dependencies]
 nix = { version = "0.31", features = ["user"] }
 [target."cfg(windows)".dependencies]
 windows-service = "0.8"
 [dev-dependencies]
 tokio = { version = "1.53", features = ["rt", "rt-multi-thread", "sync", "time", "macros", "process"] }
 tempfile.workspace = true
@@ -11,6 +11,9 @@ use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::Duration;
 #[cfg(windows)]
 mod winservice;
 use clap::{Args, Parser, Subcommand, ValueEnum};
 use tsunagi::agent::Event;
 use tsunagi::config::{AgentConfig, StoragePaths, TransportPolicy};
@@ -84,6 +87,10 @@ enum Command {
    ExitNode(ExitNodeArgs),
    /// Removes everything this device has stored and starts over.
    Wipe(WipeArgs),
    /// Runs the agent under the Windows service manager. Registered by the
    /// installer, not typed.
    #[command(hide = true)]
    Service(Box<UpArgs>),
 }
 #[derive(Debug, Args)]
@@ -757,12 +764,26 @@ fn main() -> std::process::ExitCode {
    let is_debug = log_filter.contains("debug") || log_filter.contains("trace");
-    tracing_subscriber::fmt()
+    let builder = tracing_subscriber::fmt()
        .with_env_filter(tracing_subscriber::EnvFilter::new(&log_filter))
        .with_target(is_debug)
-        .compact()
+        .compact();
-        .with_writer(std::io::stderr)
+    // A service has no console: it writes to a file.
-        .init();
+    match matches!(cli.command, Command::Service(_))
        .then(open_service_log)
        .flatten()
    {
        Some(file) => builder
            .with_ansi(false)
            .with_writer(std::sync::Mutex::new(file))
            .init(),
        None => builder.with_writer(std::io::stderr).init(),
    }
    #[cfg(windows)]
    if let Command::Service(args) = cli.command {
        return run_service(*args);
    }
    // The library never starts a runtime; this binary owns it.
    let runtime = match tokio::runtime::Builder::new_multi_thread()
@@ -785,6 +806,89 @@ fn main() -> std::process::ExitCode {
    }
 }
 /// The log file of the Windows service, `%ProgramData%\Tsunagi\logs\tsng.log`.
 ///
 /// The previous one is kept as `tsng.log.old` once it passes a few megabytes,
 /// so a service that logs for months does not fill the disk. `None` anywhere
 /// else, and when the file cannot be opened.
 fn open_service_log() -> Option<std::fs::File> {
    if !cfg!(windows) {
        return None;
    }
    let root = std::env::var_os("ProgramData")?;
    open_log_in(&PathBuf::from(root).join("Tsunagi").join("logs"))
 }
 /// Opens `tsng.log` in a directory for appending, moving a large one aside.
 fn open_log_in(dir: &std::path::Path) -> Option<std::fs::File> {
    std::fs::create_dir_all(dir).ok()?;
    let log = dir.join("tsng.log");
    if std::fs::metadata(&log).is_ok_and(|meta| meta.len() > SERVICE_LOG_LIMIT) {
        let _ = std::fs::rename(&log, dir.join("tsng.log.old"));
    }
    std::fs::OpenOptions::new()
        .create(true)
        .append(true)
        .open(log)
        .ok()
 }
 /// Past this size the service log is moved aside, keeping one older file.
 const SERVICE_LOG_LIMIT: u64 = 8 * 1024 * 1024;
 /// The directories and control socket of the Windows service, unless the
 /// service command line names its own.
 ///
 /// A service runs as the system account, whose profile is the wrong home for
 /// a device's identity, so the state is kept under `%ProgramData%\Tsunagi`,
 /// where the installer locks it to administrators and the system. The control
 /// socket stays at the system-wide default every client already looks at,
 /// which is what lets a tray and an unelevated `tsng status` find the service.
 #[cfg_attr(not(windows), allow(dead_code))]
 fn service_defaults(mut args: UpArgs, program_data: &std::path::Path) -> UpArgs {
    let root = program_data.join("Tsunagi");
    args.paths
        .state_dir
        .get_or_insert_with(|| root.join("data"));
    args.paths
        .cache_dir
        .get_or_insert_with(|| root.join("cache"));
    args.control_socket
        .get_or_insert_with(tsunagi::ipc::default_control_socket_path);
    args
 }
 /// Hands the process to the Windows service manager, with `up` as the body.
 #[cfg(windows)]
 fn run_service(args: UpArgs) -> std::process::ExitCode {
    let program_data = std::env::var_os("ProgramData")
        .map(PathBuf::from)
        .filter(|path| path.is_absolute())
        .unwrap_or_else(|| PathBuf::from(r"C:\ProgramData"));
    let args = service_defaults(args, &program_data);
    let args = std::sync::Mutex::new(Some(args));
    let body = move || -> Result<(), String> {
        let Some(args) = args.lock().ok().and_then(|mut slot| slot.take()) else {
            return Err("the service was started twice".to_string());
        };
        let runtime = tokio::runtime::Builder::new_multi_thread()
            .enable_all()
            .build()
            .map_err(|err| format!("cannot start the async runtime: {err}"))?;
        runtime.block_on(up(args)).map_err(|err| err.to_string())
    };
    match winservice::run(Box::new(body)) {
        Ok(()) => std::process::ExitCode::SUCCESS,
        Err(err) => {
            eprintln!(
                "error: cannot run as a service ({err}); `tsng service` is started by the \
                 Windows service manager, use `tsng up` to run the agent yourself"
            );
            std::process::ExitCode::FAILURE
        }
    }
 }
 async fn run(command: Command) -> Result<(), Box<dyn std::error::Error>> {
    match command {
        Command::Id(args) => id(args).await,
@@ -796,6 +900,11 @@ async fn run(command: Command) -> Result<(), Box<dyn std::error::Error>> {
        Command::Dns(args) => dns_command(args).await,
        Command::ExitNode(args) => exit_node_command(args).await,
        Command::Wipe(args) => wipe(args).await,
        // Taken in `main`, before there is a runtime: the service manager
        // wants the process's thread, not an async task.
        Command::Service(_) => Err("`tsng service` is started by the Windows service \
                                    manager; use `tsng up` to run the agent yourself"
            .into()),
    }
 }
@@ -4900,7 +5009,14 @@ async fn stop_signal() -> &'static str {
            _ = terminate.recv() => "terminated",
        }
    }
-    #[cfg(not(unix))]
+    #[cfg(windows)]
    {
        tokio::select! {
            _ = tokio::signal::ctrl_c() => "interrupted",
            _ = winservice::stop_requested() => "stopped by the service manager",
        }
    }
    #[cfg(not(any(unix, windows)))]
    {
        let _ = tokio::signal::ctrl_c().await;
        "interrupted"
@@ -5943,3 +6059,93 @@ mod network_context_tests {
        assert_eq!(standing, NetworkStanding::Stopped);
    }
 }
 #[cfg(test)]
 mod service_log_tests {
    #![allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)]
    use std::io::Write;
    use super::*;
    #[test]
    fn the_log_appends_and_a_large_one_is_moved_aside() {
        let dir = tempfile::tempdir().unwrap();
        let logs = dir.path().join("logs");
        let mut file = open_log_in(&logs).unwrap();
        writeln!(file, "one").unwrap();
        drop(file);
        let mut file = open_log_in(&logs).unwrap();
        writeln!(file, "two").unwrap();
        drop(file);
        assert_eq!(
            std::fs::read_to_string(logs.join("tsng.log")).unwrap(),
            "one\ntwo\n"
        );
        let big = std::fs::File::options()
            .append(true)
            .open(logs.join("tsng.log"))
            .unwrap();
        big.set_len(SERVICE_LOG_LIMIT + 1).unwrap();
        drop(big);
        let file = open_log_in(&logs).unwrap();
        assert_eq!(file.metadata().unwrap().len(), 0);
        assert!(logs.join("tsng.log.old").exists());
    }
    #[test]
    fn a_log_that_cannot_be_opened_is_none() {
        let dir = tempfile::tempdir().unwrap();
        let file = dir.path().join("file");
        std::fs::write(&file, b"x").unwrap();
        // A directory cannot be made under a regular file.
        assert!(open_log_in(&file.join("logs")).is_none());
    }
    #[test]
    fn the_service_keeps_its_state_under_program_data_unless_told_otherwise() {
        use clap::Parser;
        let parse = |extra: &[&str]| {
            let mut words = vec!["tsng", "service"];
            words.extend_from_slice(extra);
            match Cli::try_parse_from(words).unwrap().command {
                Command::Service(args) => *args,
                other => panic!("{other:?}"),
            }
        };
        let root = std::path::Path::new("/pd");
        let args = service_defaults(parse(&[]), root);
        assert_eq!(
            args.paths.state_dir,
            Some(root.join("Tsunagi").join("data"))
        );
        assert_eq!(
            args.paths.cache_dir,
            Some(root.join("Tsunagi").join("cache"))
        );
        assert_eq!(
            args.control_socket,
            Some(tsunagi::ipc::default_control_socket_path())
        );
        let args = service_defaults(
            parse(&["--state-dir", "/mine", "--control-socket", "/mine/sock"]),
            root,
        );
        assert_eq!(args.paths.state_dir, Some(PathBuf::from("/mine")));
        assert_eq!(args.control_socket, Some(PathBuf::from("/mine/sock")));
    }
    #[test]
    fn the_service_command_is_refused_off_the_service_manager() {
        use clap::Parser;
        let cli = Cli::try_parse_from(["tsng", "service"]).unwrap();
        assert!(matches!(cli.command, Command::Service(_)));
        let runtime = tokio::runtime::Builder::new_current_thread()
            .build()
            .unwrap();
        let err = runtime.block_on(run(cli.command)).unwrap_err();
        assert!(err.to_string().contains("service manager"), "{err}");
    }
 }
@@ -0,0 +1,121 @@
 //! The agent as a Windows service.
 //!
 //! `tsng service` is what the installer registers with the service control
 //! manager (`sc create`); nobody types it. The manager starts the process,
 //! which must call into the dispatcher quickly and then report its state, so
 //! this is the only place the Windows service API is touched. Everything the
 //! agent does is the same `up` that runs in a console; the only differences
 //! are that it stops when the manager says so rather than on Ctrl-C, and that
 //! it logs to a file because a service has no console.
 //!
 //! The one `unsafe` here is not ours: `define_windows_service!` expands to the
 //! `extern "system"` entry point the manager calls, which has to be unsafe.
 #![allow(unsafe_code)]
 use std::ffi::OsString;
 use std::sync::OnceLock;
 use std::time::Duration;
 use windows_service::service::{
    ServiceControl, ServiceControlAccept, ServiceExitCode, ServiceState, ServiceStatus, ServiceType,
 };
 use windows_service::service_control_handler::{
    self, ServiceControlHandlerResult, ServiceStatusHandle,
 };
 use windows_service::{define_windows_service, service_dispatcher};
 /// The name the installer registers the service under.
 pub const SERVICE_NAME: &str = "Tsunagi";
 /// How long the manager waits for a stop before it gives up on us. The agent's
 /// own shutdown is bounded and shorter.
 const STOP_WAIT: Duration = Duration::from_secs(25);
 /// Set when the manager asks the service to stop.
 static STOP: tokio::sync::Notify = tokio::sync::Notify::const_new();
 /// What `up` does, handed over by `main` because the manager's entry point
 /// cannot take arguments.
 type Body = Box<dyn Fn() -> Result<(), String> + Send + Sync>;
 static BODY: OnceLock<Body> = OnceLock::new();
 /// Where the manager's status handle is kept, so the stop handler can report.
 static STATUS: OnceLock<ServiceStatusHandle> = OnceLock::new();
 /// Resolves when the manager has asked the service to stop.
 pub async fn stop_requested() {
    STOP.notified().await;
 }
 define_windows_service!(ffi_service_main, service_main);
 /// Hands the process to the service control manager and returns when the
 /// service has stopped. Fails at once when not started by the manager, which
 /// is how running `tsng service` in a console is reported.
 pub fn run(body: Body) -> Result<(), windows_service::Error> {
    let _ = BODY.set(body);
    service_dispatcher::start(SERVICE_NAME, ffi_service_main)
 }
 fn service_main(_arguments: Vec<OsString>) {
    let result = serve();
    let code = match &result {
        Ok(()) => ServiceExitCode::Win32(0),
        Err(err) => {
            tracing::error!("the service failed: {err}");
            // A non-zero exit lets the recovery actions restart it.
            ServiceExitCode::ServiceSpecific(1)
        }
    };
    if let Some(handle) = STATUS.get() {
        let _ = handle.set_service_status(status(ServiceState::Stopped, code, Duration::ZERO));
    }
 }
 fn serve() -> Result<(), String> {
    let handle = service_control_handler::register(SERVICE_NAME, |control| match control {
        ServiceControl::Stop | ServiceControl::Shutdown => {
            if let Some(handle) = STATUS.get() {
                let _ = handle.set_service_status(status(
                    ServiceState::StopPending,
                    ServiceExitCode::Win32(0),
                    STOP_WAIT,
                ));
            }
            STOP.notify_one();
            ServiceControlHandlerResult::NoError
        }
        ServiceControl::Interrogate => ServiceControlHandlerResult::NoError,
        _ => ServiceControlHandlerResult::NotImplemented,
    })
    .map_err(|err| format!("cannot register with the service control manager: {err}"))?;
    let _ = STATUS.set(handle);
    handle
        .set_service_status(status(
            ServiceState::Running,
            ServiceExitCode::Win32(0),
            Duration::ZERO,
        ))
        .map_err(|err| format!("cannot report the service as running: {err}"))?;
    match BODY.get() {
        Some(body) => body(),
        None => Err("the service has nothing to run".to_string()),
    }
 }
 fn status(state: ServiceState, exit_code: ServiceExitCode, wait_hint: Duration) -> ServiceStatus {
    ServiceStatus {
        service_type: ServiceType::OWN_PROCESS,
        current_state: state,
        controls_accepted: if state == ServiceState::Running {
            ServiceControlAccept::STOP | ServiceControlAccept::SHUTDOWN
        } else {
            ServiceControlAccept::empty()
        },
        exit_code,
        checkpoint: 0,
        wait_hint,
        process_id: None,
    }
 }
@@ -22,7 +22,7 @@ service's socket at `/run/tsunagi/agent.sock`):
 ```sh
 sudo tsng status
-sudo tsng join <network-name> <tsn1…secret>
+sudo tsng join -n <network-name> -s <tsn1…secret>
 ```
 ## How it runs
@@ -28,7 +28,7 @@ group, so after `sudo usermod -aG tsunagi "$USER"` (and logging in again) no
 ```sh
 sudo tsng status
-sudo tsng join <network-name> <tsn1…secret>
+sudo tsng join -n <network-name> -s <tsn1…secret>
 ```
 ## As your own user — one capability, no service
@@ -48,7 +48,7 @@ Done. Start the agent with:
 Control it with sudo (the CLI finds the service's socket automatically):
    sudo tsng status
-    sudo tsng join <network-name> <tsn1…secret>
+    sudo tsng join -n <network-name> -s <tsn1…secret>
 EOF
 else
    cat <<EOF
@@ -15,7 +15,7 @@ socket and do **not** need `sudo`:
 ```sh
 tsng status
-tsng join <network-name> <tsn1…secret>
+tsng join -n <network-name> -s <tsn1…secret>
 ```
 The agent uses fixed system paths, so the commands find it regardless of whose
@@ -1,5 +1,34 @@
 # Running tsunagi on Windows
 ## The installer (recommended)
 `tsunagi-setup-<version>-x86_64.exe` does everything in one go. Run it as an
 administrator (it asks):
 - installs `tsng.exe`, the tray app and **Wintun** to `C:\Program Files\Tsunagi`;
 - registers and starts the **Tsunagi** Windows service (it starts at boot and
  restarts after a crash), running as the system account;
 - adds `tsng` to the PATH and an inbound Windows Firewall rule for the agent
  (both are tasks you can untick);
 - starts the tray icon at sign-in.
 Where things live: the device's identity and networks in
 `%ProgramData%\Tsunagi\data` (readable only by administrators and the system),
 the log in `%ProgramData%\Tsunagi\logs\tsng.log`. The service keeps its own
 state: networks joined by an agent you started by hand (`tsng up`, which keeps
 its data in your user profile) are not carried over; join them again with
 `tsng join`. The installer stops such an agent, because it cannot run beside the
 service.
 `tsng status`, `tsng join` and the tray work from an ordinary, non-elevated
 terminal. Manage the service with the Services app or `sc stop Tsunagi` /
 `sc start Tsunagi`. Silent install: `/VERYSILENT /SUPPRESSMSGBOXES`; choose
 tasks with `/TASKS="!autostart,addtopath,firewall"`. Uninstalling removes the
 service, the firewall rule and the PATH entry, and asks before deleting the
 data.
 ## By hand (the zip)
 This archive holds `tsng.exe`. The overlay interface is a Wintun adapter, so
 the agent needs two things to carry real traffic:
@@ -0,0 +1,242 @@
 ; Inno Setup script for the Tsunagi Windows installer.
 ;
 ; One installer that leaves a working machine:
 ;   * the agent (tsng.exe), as a Windows service that starts at boot
 ;   * Wintun, the driver library behind the overlay interface
 ;   * the tray application, started at sign-in
 ;   * `tsng` on the PATH, and an inbound Defender Firewall rule for the agent
 ;
 ; Built in CI (see .github/workflows/release.yml); locally:
 ;   ISCC.exe /DAppVersion=0.1.0 /DSourceDir=<dir with tsng.exe and tsunagi-tray.exe>
 ;            /DWintunDir=<wintun\bin\amd64> /O<output dir> dist\windows\tsunagi.iss
 ; Silent install: tsunagi-setup-<version>-x86_64.exe /VERYSILENT /SUPPRESSMSGBOXES
 ; Pick tasks with /TASKS="!autostart,addtopath,firewall".
 #ifndef AppVersion
  #define AppVersion "0.0.0"
 #endif
 #ifndef SourceDir
  #error SourceDir must point at the directory holding tsng.exe and tsunagi-tray.exe
 #endif
 #ifndef WintunDir
  #error WintunDir must point at the directory holding wintun.dll
 #endif
 #define ServiceName "Tsunagi"
 [Setup]
 AppId={{F635E5DD-250D-499F-B0B3-AFC7CC64A15D}
 AppName=Tsunagi
 AppVersion={#AppVersion}
 AppPublisher=Tsunagi
 DefaultDirName={autopf}\Tsunagi
 DefaultGroupName=Tsunagi
 DisableProgramGroupPage=yes
 PrivilegesRequired=admin
 ArchitecturesAllowed=x64compatible
 ArchitecturesInstallIn64BitMode=x64compatible
 OutputBaseFilename=tsunagi-setup-{#AppVersion}-x86_64
 Compression=lzma2
 SolidCompression=yes
 WizardStyle=modern
 ChangesEnvironment=yes
 UninstallDisplayName=Tsunagi
 ; The service and the tray are stopped by hand below, so the installer does
 ; not need to ask Windows to close applications.
 CloseApplications=no
 [Tasks]
 Name: "autostart"; Description: "Start the tray icon when I sign in"; GroupDescription: "Tsunagi:"
 Name: "addtopath"; Description: "Add tsng to the PATH"; GroupDescription: "Tsunagi:"
 Name: "firewall"; Description: "Allow the agent through Windows Firewall (direct connections between devices)"; GroupDescription: "Tsunagi:"
 [Files]
 Source: "{#SourceDir}\tsng.exe"; DestDir: "{app}"; Flags: ignoreversion
 Source: "{#SourceDir}\tsunagi-tray.exe"; DestDir: "{app}"; Flags: ignoreversion
 ; Wintun is the unmodified, signed library from wintun.net; its licence allows
 ; shipping it with an application and asks for the licence text beside it.
 ; In the archive the text sits at its root, two levels above bin\amd64.
 Source: "{#WintunDir}\wintun.dll"; DestDir: "{app}"; Flags: ignoreversion
 Source: "{#WintunDir}\..\..\LICENSE.txt"; DestDir: "{app}\licenses"; DestName: "wintun-LICENSE.txt"; Flags: ignoreversion
 [Dirs]
 ; Identity and network secrets live here; locked down below.
 Name: "{commonappdata}\Tsunagi\data"
 Name: "{commonappdata}\Tsunagi\cache"
 Name: "{commonappdata}\Tsunagi\logs"
 [Icons]
 Name: "{group}\Tsunagi"; Filename: "{app}\tsunagi-tray.exe"
 Name: "{commonstartup}\Tsunagi"; Filename: "{app}\tsunagi-tray.exe"; Tasks: autostart
 [Registry]
 Root: HKLM; Subkey: "SYSTEM\CurrentControlSet\Control\Session Manager\Environment"; \
    ValueType: expandsz; ValueName: "Path"; ValueData: "{olddata};{app}"; \
    Check: NeedsAddPath('{app}'); Tasks: addtopath
 [Run]
 ; Only the system account and administrators may read the state: it holds the
 ; device's key and every network's secret. SIDs, so it works in any language.
 Filename: "{sys}\icacls.exe"; \
    Parameters: """{commonappdata}\Tsunagi\data"" /inheritance:r /grant:r *S-1-5-18:(OI)(CI)F *S-1-5-32-544:(OI)(CI)F"; \
    Flags: runhidden; StatusMsg: "Protecting the agent's data..."
 Filename: "{sys}\icacls.exe"; \
    Parameters: """{commonappdata}\Tsunagi\cache"" /inheritance:r /grant:r *S-1-5-18:(OI)(CI)F *S-1-5-32-544:(OI)(CI)F"; \
    Flags: runhidden
 Filename: "{sys}\icacls.exe"; \
    Parameters: """{commonappdata}\Tsunagi\logs"" /inheritance:r /grant:r *S-1-5-18:(OI)(CI)F *S-1-5-32-544:(OI)(CI)F"; \
    Flags: runhidden
 ; The tray runs as the person who installed, not elevated, like it does at sign-in.
 Filename: "{app}\tsunagi-tray.exe"; Description: "Start the Tsunagi tray"; \
    Flags: nowait postinstall skipifsilent runasoriginaluser
 [UninstallRun]
 Filename: "{sys}\netsh.exe"; Parameters: "advfirewall firewall delete rule name=""Tsunagi"""; \
    Flags: runhidden; RunOnceId: "RemoveFirewallRule"
 [Code]
 const
  ServiceName = '{#ServiceName}';
  EnvKey = 'SYSTEM\CurrentControlSet\Control\Session Manager\Environment';
 function Sc(const Params: String): Integer;
 var
  Code: Integer;
 begin
  if Exec(ExpandConstant('{sys}\sc.exe'), Params, '', SW_HIDE, ewWaitUntilTerminated, Code) then
    Result := Code
  else
    Result := -1;
 end;
 function ServiceExists: Boolean;
 begin
  // 1060 is ERROR_SERVICE_DOES_NOT_EXIST.
  Result := Sc('query ' + ServiceName) <> 1060;
 end;
 procedure StopService;
 var
  Code: Integer;
 begin
  // `net stop` waits until the service has stopped, or says it was not running.
  Exec(ExpandConstant('{sys}\net.exe'), 'stop ' + ServiceName, '', SW_HIDE, ewWaitUntilTerminated, Code);
 end;
 procedure StopTray;
 var
  Code: Integer;
 begin
  Exec(ExpandConstant('{sys}\taskkill.exe'), '/F /IM tsunagi-tray.exe', '', SW_HIDE, ewWaitUntilTerminated, Code);
 end;
 // An agent started by hand (`tsng up`) holds the control pipe and the Wintun
 // adapter, and the file about to be replaced: it cannot run beside the service.
 procedure StopStrayAgents;
 var
  Code: Integer;
 begin
  Exec(ExpandConstant('{sys}\taskkill.exe'), '/F /IM tsng.exe', '', SW_HIDE, ewWaitUntilTerminated, Code);
 end;
 function NeedsAddPath(Param: String): Boolean;
 var
  Current: String;
 begin
  if not RegQueryStringValue(HKLM, EnvKey, 'Path', Current) then
  begin
    Result := True;
    exit;
  end;
  Result := Pos(';' + Uppercase(ExpandConstant(Param)) + ';', ';' + Uppercase(Current) + ';') = 0;
 end;
 procedure RemoveFromPath;
 var
  Current, Entry: String;
  At: Integer;
 begin
  if not RegQueryStringValue(HKLM, EnvKey, 'Path', Current) then
    exit;
  Entry := ';' + ExpandConstant('{app}');
  At := Pos(Uppercase(Entry), Uppercase(Current));
  if At > 0 then
  begin
    Delete(Current, At, Length(Entry));
    RegWriteExpandStringValue(HKLM, EnvKey, 'Path', Current);
  end;
 end;
 // Stops what is about to be replaced: the files are in use while it runs.
 function PrepareToInstall(var NeedsRestart: Boolean): String;
 begin
  StopTray;
  if ServiceExists then
    StopService;
  StopStrayAgents;
  Result := '';
 end;
 procedure InstallService;
 var
  App: String;
 begin
  App := ExpandConstant('{app}\tsng.exe');
  // The command line is `tsng service`: it keeps its state under
  // %ProgramData%\Tsunagi and serves the system-wide control socket.
  if ServiceExists then
    Sc('config ' + ServiceName + ' binPath= "\"' + App + '\" service" start= auto obj= LocalSystem')
  else
    Sc('create ' + ServiceName + ' binPath= "\"' + App + '\" service" start= auto obj= LocalSystem DisplayName= "Tsunagi agent"');
  Sc('description ' + ServiceName + ' "Tsunagi: private mesh network agent"');
  // Restart after a crash, backing off, and forget the failures after a day.
  Sc('failure ' + ServiceName + ' reset= 86400 actions= restart/5000/restart/5000/restart/30000');
  Sc('start ' + ServiceName);
 end;
 procedure InstallFirewallRule;
 var
  Code: Integer;
  Netsh: String;
 begin
  Netsh := ExpandConstant('{sys}\netsh.exe');
  Exec(Netsh, 'advfirewall firewall delete rule name="Tsunagi"', '', SW_HIDE, ewWaitUntilTerminated, Code);
  if WizardIsTaskSelected('firewall') then
    Exec(Netsh,
      'advfirewall firewall add rule name="Tsunagi" dir=in action=allow enable=yes profile=any program="' +
      ExpandConstant('{app}\tsng.exe') + '"',
      '', SW_HIDE, ewWaitUntilTerminated, Code);
 end;
 procedure CurStepChanged(CurStep: TSetupStep);
 begin
  if CurStep = ssPostInstall then
  begin
    InstallFirewallRule;
    InstallService;
  end;
 end;
 procedure CurUninstallStepChanged(CurUninstallStep: TUninstallStep);
 begin
  if CurUninstallStep = usUninstall then
  begin
    StopTray;
    if ServiceExists then
    begin
      StopService;
      Sc('delete ' + ServiceName);
    end;
    StopStrayAgents;
    RemoveFromPath;
  end;
  if CurUninstallStep = usPostUninstall then
  begin
    // The data is the device's identity and its networks: kept unless asked,
    // and never removed by a silent uninstall.
    if (not UninstallSilent) and
       (MsgBox('Also delete this device''s identity, networks and logs?' + #13#10 +
               'Choose No to keep them for a later reinstall.', mbConfirmation, MB_YESNO or MB_DEFBUTTON2) = IDYES) then
      DelTree(ExpandConstant('{commonappdata}\Tsunagi'), True, True, True);
  end;
 end;
@@ -82,18 +82,25 @@ is built from `pf` and `route`. The agent runs as root there.
 ## Windows
-Offering only. Each offered range gets a WinNAT object
+- **Offering:** each offered range gets a WinNAT object
-`tsunagi-exit:<interface>:<range>` (`New-NetNat`), removed with the offer, the
+  `tsunagi-exit:<interface>:<range>` (`New-NetNat`), removed with the offer, the
-network and the agent. Forwarding is read for the overlay interface and never
+  network and the agent. Forwarding is read for the overlay interface and never
-changed; Windows forwards per interface, so enable it on the ones involved
+  changed; Windows forwards per interface, so enable it on the ones involved
-(`Get-NetIPInterface | Set-NetIPInterface -Forwarding Enabled`, administrator
+  (`Get-NetIPInterface | Set-NetIPInterface -Forwarding Enabled`, administrator
-PowerShell). Some Windows versions allow one NAT per host; a second range then
+  PowerShell). Some Windows versions allow one NAT per host; a second range then
-reports its own failure.
+  reports its own failure.
-
+- **Using:** Windows has no per-user routing, so the agent's own traffic is kept
-Using one is **not available**: Windows has no per-user routing, so the agent's
+  off the tunnel *by destination*. `0.0.0.0/1` and `128.0.0.0/1` go through the
-own traffic cannot be kept out of a default route through the overlay and would
+  overlay interface; a `/32` route with metric 7373 (the marker for "ours")
-loop. `tsng exit-node <name>` reports that and installs nothing. This was
+  through the physical default gateway is added for every address the agent
-written without a Windows host to run it on.
+  talks to directly: verified direct paths of its connections, plus the relays
  it uses, resolved when the rules are applied. The set follows path changes. A
  path that appears after the last apply goes through the tunnel until the next
  one. IPv6 is blackholed with `::/1` and `8000::/1` on the overlay interface,
  so applications fall back to IPv4 after a delay rather than at once.
  Everything is in the active store: a reboot clears it.
 - Not verified on a live host when this was written. To inspect:
  `Get-NetRoute -PolicyStore ActiveStore | ? { $_.RouteMetric -eq 7373 -or $_.InterfaceAlias -like 'tsun*' }`.
 ## What it does not do
@@ -26,8 +26,8 @@ system can hand us IP packets, and even that is behind a trait
 | `ManagedTunFactory` | `CAP_NET_ADMIN` | traffic actually reaches the OS |
 `ManagedTunFactory` creates the interface and configures it; see
-*Provisioning the interface* below and *Privileges* in
+*Provisioning the interface* below and Privileges in
-[../README.md](../README.md#privileges). With no capability the agent runs
+[../dist/linux/README.md](../dist/linux/README.md). With no capability the agent runs
 with `--no-tun`: everything but the last hop into the kernel still works.
 [boringtun]: https://docs.rs/boringtun
@@ -0,0 +1,27 @@
 {
  "nodes": {
    "nixpkgs": {
      "locked": {
        "lastModified": 1790769526,
        "narHash": "sha256-5AIVFLRcx8YAwbFTVPTs7YhwJH7JyHfAQsvXmpovcXI=",
        "owner": "NixOS",
        "repo": "nixpkgs",
        "rev": "c9fe7d12cd78d1adcd12dd15e24432dde5b155a0",
        "type": "github"
      },
      "original": {
        "owner": "NixOS",
        "ref": "nixpkgs-unstable",
        "repo": "nixpkgs",
        "type": "github"
      }
    },
    "root": {
      "inputs": {
        "nixpkgs": "nixpkgs"
      }
    }
  },
  "root": "root",
  "version": 7
 }
@@ -0,0 +1,39 @@
 {
  description = "tsunagi: serverless mesh VPN. A network name and a secret are all it takes.";
  inputs.nixpkgs.url = "github:NixOS/nixpkgs/nixpkgs-unstable";
  outputs =
    { self, nixpkgs }:
    let
      systems = [
        "x86_64-linux"
        "aarch64-linux"
      ];
      forAllSystems = f: nixpkgs.lib.genAttrs systems (system: f nixpkgs.legacyPackages.${system});
    in
    {
      packages = forAllSystems (pkgs: rec {
        tsunagi = pkgs.callPackage ./nix/package.nix { };
        tsunagi-cli = pkgs.callPackage ./nix/package.nix { withTray = false; };
        default = tsunagi;
      });
      overlays.default = final: _prev: {
        tsunagi = final.callPackage ./nix/package.nix { };
        tsunagi-cli = final.callPackage ./nix/package.nix { withTray = false; };
      };
      nixosModules.default = import ./nix/module.nix self;
      devShells = forAllSystems (pkgs: {
        default = pkgs.mkShell {
          inputsFrom = [ self.packages.${pkgs.stdenv.hostPlatform.system}.tsunagi ];
          packages = [
            pkgs.rustfmt
            pkgs.clippy
          ];
        };
      });
    };
 }
@@ -0,0 +1,132 @@
 flake:
 {
  config,
  lib,
  pkgs,
  ...
 }:
 let
  cfg = config.services.tsunagi;
  socket = "/run/tsunagi/agent.sock";
 in
 {
  options.services.tsunagi = {
    enable = lib.mkEnableOption "the tsunagi mesh agent";
    package = lib.mkOption {
      type = lib.types.package;
      default =
        if cfg.tray.enable then
          flake.packages.${pkgs.stdenv.hostPlatform.system}.tsunagi
        else
          flake.packages.${pkgs.stdenv.hostPlatform.system}.tsunagi-cli;
      defaultText = lib.literalExpression "tsunagi.packages.\${system}.tsunagi (tsunagi-cli without the tray)";
      description = "The tsunagi package to run.";
    };
    tray.enable = lib.mkEnableOption "the tray client, started with every graphical session";
    users = lib.mkOption {
      type = lib.types.listOf lib.types.str;
      default = [ ];
      example = [ "alice" ];
      description = ''
        Users allowed to talk to the agent without sudo (`tsng status`, `tsng join`,
        the tray). They are added to the `tsunagi` group.
      '';
    };
    protocols = lib.mkOption {
      type = lib.types.nullOr (lib.types.listOf lib.types.str);
      default = null;
      example = [
        "wg-quic"
        "wg"
      ];
      description = "Packet protocols for `tsng up --protocol`, best first. Null keeps the agent's default.";
    };
  };
  config = lib.mkIf cfg.enable {
    users.groups.tsunagi = { };
    users.users = {
      tsunagi = {
        isSystemUser = true;
        group = "tsunagi";
        home = "/var/lib/tsunagi";
        description = "tsunagi agent";
      };
    }
    // lib.genAttrs cfg.users (_: {
      extraGroups = [ "tsunagi" ];
    });
    systemd.services.tsunagi = {
      description = "tsunagi mesh agent";
      documentation = [ "https://github.com/house-of-vanity/tsunagi" ];
      wantedBy = [ "multi-user.target" ];
      wants = [ "network-online.target" ];
      after = [ "network-online.target" ];
      environment = {
        TSUNAGI_STATE_DIR = "/var/lib/tsunagi";
        TSUNAGI_CACHE_DIR = "/var/cache/tsunagi";
        TSUNAGI_CONTROL_SOCKET = socket;
        TSUNAGI_CONTROL_GROUP = "tsunagi";
      };
      serviceConfig = {
        ExecStart = lib.concatStringsSep " " (
          [ "${cfg.package}/bin/tsng up" ]
          ++ lib.optional (cfg.protocols != null) "--protocol ${lib.concatStringsSep "," cfg.protocols}"
        );
        User = "tsunagi";
        Group = "tsunagi";
        AmbientCapabilities = [
          "CAP_NET_ADMIN"
          "CAP_NET_BIND_SERVICE"
        ];
        CapabilityBoundingSet = [
          "CAP_NET_ADMIN"
          "CAP_NET_BIND_SERVICE"
        ];
        StateDirectory = "tsunagi";
        StateDirectoryMode = "0700";
        CacheDirectory = "tsunagi";
        RuntimeDirectory = "tsunagi";
        RuntimeDirectoryMode = "0755";
        Restart = "on-failure";
        RestartSec = 5;
      };
    };
    # Exit nodes: the agent looks for iptables only in fixed FHS directories.
    systemd.tmpfiles.rules = [
      "L+ /usr/bin/iptables - - - - ${pkgs.iptables}/bin/iptables"
    ];
    # Names of mesh members are served locally and handed to the system resolver.
    services.resolved.enable = lib.mkDefault true;
    # polkit decides by group, not capability, who may configure systemd-resolved.
    security.polkit.extraConfig = ''
      polkit.addRule(function(action, subject) {
          var allowed = [
              "org.freedesktop.resolve1.set-dns-servers",
              "org.freedesktop.resolve1.set-domains",
              "org.freedesktop.resolve1.set-default-route",
              "org.freedesktop.resolve1.revert"
          ];
          if (allowed.indexOf(action.id) >= 0 && subject.isInGroup("tsunagi")) {
              return polkit.Result.YES;
          }
      });
    '';
    environment.systemPackages = [ cfg.package ];
    environment.sessionVariables.TSUNAGI_CONTROL_SOCKET = socket;
    environment.etc."xdg/autostart/tsunagi-tray.desktop" = lib.mkIf cfg.tray.enable {
      source = "${cfg.package}/share/applications/tsunagi-tray.desktop";
    };
  };
 }
@@ -0,0 +1,94 @@
 {
  withTray ? true,
  lib,
  rustPlatform,
  pkg-config,
  copyDesktopItems,
  wrapGAppsHook3,
  gtk3,
  libayatana-appindicator,
  libglvnd,
  libxkbcommon,
  vulkan-loader,
  wayland,
  xdotool,
  libx11,
  libxcursor,
  libxi,
  libxrandr,
 }:
 let
  # The tray is an egui app that loads graphics, windowing and appindicator
  # libraries at run time, so they have to be put on its library path by hand.
  trayRuntimeLibs = [
    libayatana-appindicator
    libglvnd
    libxkbcommon
    vulkan-loader
    wayland
    libx11
    libxcursor
    libxi
    libxrandr
  ];
 in
 rustPlatform.buildRustPackage {
  pname = if withTray then "tsunagi" else "tsunagi-cli";
  version = "0.1.0";
  src = lib.cleanSource ../.;
  cargoLock.lockFile = ../Cargo.lock;
  cargoBuildFlags = [
    "--bin"
    "tsng"
  ]
  ++ lib.optionals withTray [
    "--bin"
    "tsunagi-tray"
  ];
  # The suite opens real sockets and interfaces; it runs in CI instead.
  doCheck = false;
  nativeBuildInputs = [ pkg-config ] ++ lib.optionals withTray [
    copyDesktopItems
    wrapGAppsHook3
  ];
  buildInputs = lib.optionals withTray [
    gtk3
    libayatana-appindicator
    xdotool
  ];
  dontWrapGApps = true;
  postInstall = ''
    install -Dm644 dist/linux/50-tsunagi-resolved.rules \
      "$out/share/doc/tsunagi/50-tsunagi-resolved.rules"
  ''
  + lib.optionalString withTray ''
    install -Dm644 dist/linux/tsunagi.svg \
      "$out/share/icons/hicolor/scalable/apps/tsunagi.svg"
    install -Dm644 dist/linux/tsunagi-tray.desktop \
      "$out/share/applications/tsunagi-tray.desktop"
  '';
  postFixup = lib.optionalString withTray ''
    wrapProgram "$out/bin/tsunagi-tray" \
      "''${gappsWrapperArgs[@]}" \
      --prefix LD_LIBRARY_PATH : ${lib.makeLibraryPath trayRuntimeLibs}:/run/opengl-driver/lib
  '';
  meta = {
    description = "Serverless mesh VPN: join a network with a name and a secret";
    homepage = "https://github.com/house-of-vanity/tsunagi";
    license = lib.licenses.wtfpl;
    mainProgram = "tsng";
    platforms = [
      "x86_64-linux"
      "aarch64-linux"
    ];
  };
 }
Author	SHA1	Message	Date
Aleksandr BogomiakovandClaude	c184ef0914	Rewrite README as a short project front page; add a Nix flake The README is now a landing page: the idea (a mesh VPN with no server, joined with a network name and a secret), install and platform tables, exit nodes, and a short build section. Implementation detail stays in docs/. The flake builds tsng and the tray from source and ships a NixOS module (services.tsunagi). Dist install notes now show the real `join -n -s` syntax, and two links that pointed into the old README are redirected. Co-Authored-By: Claude <noreply@anthropic.com>	2026-10-03 17:47:44 +01:00
Aleksandr BogomiakovandClaude	e6ec62bcba	Windows installer: find the Wintun licence at the archive root Co-Authored-By: Claude <noreply@anthropic.com>	2026-10-03 14:38:16 +01:00
UltradesuandClaude Sonnet 5.5	21328bf818	Windows installer: agent service, Wintun, tray and PATH in one setup program Running on Windows needed wintun.dll by hand, an elevated console and a manually started agent; without the DLL the agent silently ran with no interface (and so no exit node or routing at all). `tsng service` (hidden) runs the same `up` under the Windows service manager: src/winservice.rs registers with the SCM through the windows-service crate, reports Running/StopPending/Stopped, and maps a stop request onto the existing shutdown path (stop_signal). The only unsafe is the extern "system" entry point that define_windows_service! expands to, allowed in that module alone. A service has no console, so it logs to %ProgramData%\Tsunagi\logs\tsng.log (one older file kept past 8 MiB). Its state defaults to %ProgramData%\Tsunagi\{data,cache} and its control socket to the system-wide default every client already uses, so an unelevated `tsng status` and the tray find it. dist/windows/tsunagi.iss (Inno Setup) installs tsng.exe, the tray and Wintun 0.14.1 (pinned by SHA-256 in CI, licence text shipped), registers the "Tsunagi" service as LocalSystem with auto start and restart-on-failure, locks the data/cache/log directories to SYSTEM and Administrators by SID, and by tasks adds tsng to the PATH, an inbound firewall rule for the agent and a tray autostart. It stops the service, the tray and any hand-started agent before replacing files; uninstall removes the service, rule and PATH entry and asks before deleting the data (never on a silent uninstall). The release workflows (GitHub and Gitea) build it on the Windows job and publish tsunagi-setup-<version>-x86_64.exe. .cargo/config.toml links the MSVC C runtime statically so nothing needs a Visual C++ redistributable on a clean machine. Compatibility: an agent run by hand keeps its data in the user profile; the service has its own, so networks must be joined again once. Checks: cargo fmt --check, clippy -D warnings, cargo test --workspace passed on macOS (one unrelated wireguard restart test failed once under load and passed on three reruns). winservice.rs was cargo-checked for x86_64-pc-windows-msvc in a scratch crate; the whole tsunagi crate cannot be built for Windows from macOS (aws-lc needs the Windows SDK). The .iss script and the CI step were NOT compiled or run: no Inno Setup or Windows here. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-03 14:16:05 +01:00
UltradesuandClaude Sonnet 5.5	7c176e0b74	Exit nodes on Windows: use a member as exit node via destination bypass routes Choosing an exit node on Windows did nothing: the using side was refused, because Windows has no per-user routing and the agent's own traffic would loop into the tunnel it carries. Now the overlay interface gets 0.0.0.0/1 and 128.0.0.0/1 routes (and ::/1, 8000::/1 as an IPv6 blackhole), and every underlay address the agent talks to directly gets a /32 route through the physical default gateway, marked with route metric 7373 so stale ones are found and removed. Bypass routes are written before the halves and removed after them; a failed step removes what was written. All routes are in the active store. The addresses come from the runtime: each network runtime hands the interface the verified direct IPv4 paths of its control connections and the relay hosts in use (Interface::set_underlay, refreshed every dial round); relay names are resolved when the rules are applied. ExitHostRules::needs_bypass (true only on Windows) gates it, so Linux and macOS plans never carry or re-apply on path changes. ExitHostPlan gains `bypass` and `bypass_hosts`; networks leaving forget their entry. Limits: a path that appears after the last apply goes through the tunnel until the next apply; data-plane-only addresses are not collected separately (the data links ride the same iroh paths). Nothing was run on Windows. Checks: cargo fmt --check, cargo clippy --locked --workspace --all-targets -D warnings and cargo test --locked --workspace --all-targets passed on macOS; scripts and ordering are unit-tested with an injected runner and resolver, and Interface::set_underlay with the mock host. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>	2026-10-03 13:43:01 +01:00