Files
tsunagi/README.md
T
tsunagiandClaude Opus 5 b23e832a73 Manage the overlay interface instead of asking for it
The agent printed a list of `ip` commands and asked a human to run them.
That is fragile in the way hand-held setup always is: a persistent TUN
does not survive a reboot, a changed address allocation needs another
manual round, and a run that died leaves a half-configured interface the
next run trips over.

On Linux the agent now creates the interface, sets the MTU, brings it up
and assigns both overlay addresses itself, over netlink in process. No
`ip` is invoked, so nothing this path does can be influenced by PATH, a
shell, or anything a remote peer said.

Cleanup stops being an action. The interface is tied to an open file
descriptor and is deliberately not persistent, so the kernel removes it
when the agent goes — cleanly, by panic, by SIGKILL or by power loss
alike. That also retires `keep_addr_on_down` and `nodad`, which existed
only because an interface nobody held open lost carrier.

Anything still left behind is repaired rather than tripped over: an
abandoned TUN is replaced along with its stale addresses. Two cases
refuse instead of guessing — a link that is not a TUN, because a name
collision is no reason to destroy somebody's bridge, and a TUN another
process holds open, because that is a working overlay belonging to
someone else.

CAP_NET_ADMIN is kept out of the effective set except around the calls
that use it. Two facts shape how: capabilities are per thread, and
netlink checks the credentials of whichever thread calls sendmsg, which
with an async client is the connection task rather than the caller. So
netlink runs on one dedicated thread with a current-thread runtime where
nothing is polled outside a block_on, and opening the TUN descriptor is
synchronous with no await between the guard and its release.

The decision of what to change is a pure function, tested on every
platform; only the execution is behind the provisioner trait. macOS and
Windows get an implementation that refuses with an explanation and falls
back to attaching to a prepared interface, plus a mock host the tests
drive the whole plugin lifecycle against.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-21 14:33:19 +01:00

17 KiB

tsunagi

A proof-of-concept Rust library for small private mesh networks — a handful of friends, home machines, a few servers. Units to dozens of participants, not thousands.

The end user configures exactly two things:

network_name
secret          # one shared secret; "password" and "secret" mean the same value

From those, every agent independently derives the same network space. There is no central server, no network owner with special powers, no registration and no majority vote. Anyone who knows the parameters can join; nobody has to trust anybody else.

What this proof of concept actually does

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;
  • 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, tsunagi, with a local control socket.

What it deliberately does not do

Not implemented, and not pretended to be: Mainline DHT, 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. The WireGuard plugin's own limits, including that its system backend is Linux-only, are in docs/wireguard.md.

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 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:

cargo build --release
./target/release/tsunagi secret          # prints tsn1...; share it privately
./target/release/tsunagi doctor          # what this host can and cannot do

./target/release/tsunagi up --network lab --secret "$SECRET" --wireguard

It prints its endpoint id and then waits. On the second machine, pass that id:

./target/release/tsunagi up --network lab --secret "$SECRET" --wireguard \
  --peer <endpoint-id-from-the-first-machine>

Within a few seconds both print something like:

  + peer b47c958462 connected over Direct rtt=Some(4.5ms)
  + data link to b47c958462 for wireguard: Direct via Ip(…), datagram 1382

--- status ---
control: 1 peer(s), 0 dial failure(s), 0 handshake failure(s)
wireguard: tsunkkcp43lmdje on fd15:1d9e:fa21:f201:…/64 mtu 1280, 1/1 tunnel(s) established
  4jO4kx9Z fd15:1d9e:fa21:f201:… handshake 3s ago tx=0 rx=0 dropped=0 path=Direct via Ip(…)

1/1 tunnel(s) established means a real WireGuard handshake completed.

Checking that it works

From another shell on either machine:

tsunagi status
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:

ping6 fd09:…:c4b        # or
ping  100.65.243.53

tx and rx in the status should start moving.

IPv6 works out of the box: each member's address is derived from the network id and collides with essentially nothing.

IPv4 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:

tsunagi up --network lab --secret "$SECRET" --wireguard --ipv4-range 10.44.0.0/16
tsunagi up --network lab --secret "$SECRET" --wireguard --ipv4-range none   # IPv6 only

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 and 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. On Linux the agent assigns it itself as soon as it has one; elsewhere tsunagi tun-setup prints the ip address add line once the claim is in state.sqlite.

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:

sudo setcap cap_net_admin+p /usr/local/bin/tsunagi

+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.

tsunagi doctor says which of these applies on the host it runs on.

It cleans up after itself

The interface is tied to an open file descriptor and is deliberately not made persistent, so the kernel removes it when the agent exits — on a clean shutdown, on a panic, on SIGKILL, on power loss alike. Keeping it is what would take an action; removing it is the default.

If something is left behind anyway — an interface made by an older version's 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 bridge is not a reason to destroy the bridge;
  • a TUN that another process is holding open, because that is a working overlay belonging to somebody else — most likely a second agent on this host, which should be given a different --wg-prefix.

Both of those refuse with an explanation rather than guessing.

Running without the capability

--interface attach (or auto, which falls back on its own) opens an interface prepared beforehand and needs no privileges at all. Ask the agent what to run:

tsunagi tun-setup --network lab --secret "$SECRET"
# Interface tsunjwc6dcrtmo5, address fd80:1210:f724:f620:d1bb:f982:3b6e:19bd/64, mtu 1280
# Run once as root; then run `tsunagi up` as ab.

sudo ip tuntap add dev tsunjwc6dcrtmo5 mode tun user ab
sudo ip link set dev tsunjwc6dcrtmo5 mtu 1280 up
sudo sysctl -qw net.ipv6.conf.tsunjwc6dcrtmo5.keep_addr_on_down=1
sudo ip -6 address add fd80:1210:f724:f620:d1bb:f982:3b6e:19bd/64 dev tsunjwc6dcrtmo5 nodad

user ab is the point: the interface is persistent and owned by that user, so tsunagi up afterwards opens it with no privileges and no capabilities.

The last two settings are what the managed path does not need. A persistent TUN has no carrier until a process attaches to it; Linux flushes IPv6 addresses from an interface that loses carrier unless keep_addr_on_down is set, and duplicate address detection can never finish without carrier, so the address would sit there tentative and unusable without nodad. An interface the agent creates and holds open has carrier for its whole life, so neither applies. IPv4 needs neither in either case: Linux keeps IPv4 addresses across carrier loss and IPv4 has no duplicate address detection.

With IPv4 enabled, tun-setup adds an ip address add line once there is an address to print — it reads the signed claim back out of state.sqlite, which does not disturb a running agent.

The MTU is 1280 because that is the minimum IPv6 requires (RFC 8200). Linux disables IPv6 entirely on an interface below it — the per-device /proc/sys/net/ipv6 entries vanish and adding an address fails with Invalid argument — so a smaller MTU cannot work at all. The agent refuses one rather than letting it fail later.

Summary

approach agent runs as notes
setcap cap_net_admin+p ordinary user, one capability recommended on Linux: 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
tsunagi tun-setup then --interface attach ordinary user, no capabilities one privileged setup per host; needed on macOS and Windows, where no provisioner is implemented yet
sudo tsunagi up root everything works, nothing is isolated
--no-tun ordinary user, no capabilities tunnels run and handshake, traffic never reaches the OS

Not implemented yet. macOS and Windows have no provisioner: both need real platform work — utun and SystemConfiguration on one, the IP Helper API and a Wintun adapter on the other. On those the agent says so and falls back to attaching to a prepared interface. The decision logic that says what to change is shared and tested on every platform; only the execution is per-platform.

Checks

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 to see agent logs while a test runs.

There are also two runnable demos, which are demos and not substitutes for the tests:

cargo run --example two_agents       # control plane only
cargo run --example wireguard_mesh   # a WireGuard overlay carrying a real packet

Both run with no privileges and change nothing on the host.

Usage

use std::sync::Arc;
use tsunagi::config::{AgentConfig, StoragePaths, TransportPolicy};
use tsunagi::discovery::SharedMemoryDiscovery;
use tsunagi::identity::{NetworkName, NetworkSecret};
use tsunagi::proto::ControlMessage;
use tsunagi::{Agent, Result};

// The library never starts a runtime, installs a logger, handles signals,
// forks, or calls process::exit. The binary owns all of that.
#[tokio::main]
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?;

    let name = NetworkName::new("kitchen-table")?;
    let secret = NetworkSecret::generate();       // 32 random bytes
    println!("share this: {}", secret.encode().as_str());

    let network = agent.join_network(&name, &secret).await?;

    let mut events = agent.subscribe();
    tokio::spawn(async move {
        while let Ok(event) = events.recv().await {
            println!("{event:?}");
        }
    });

    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 reaches the internet by accident. Opt into DirectOnly or N0Defaults explicitly.

How peers find each other

Two different lookups are involved, and only one of them is this project's:

1. Resolving one endpoint's address — iroh's, and it works today. With --transport relay or --transport direct, iroh publishes a signed 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 — ours, and it is still manual. NetworkDiscovery maps a secret-derived DiscoveryKey to a set of candidate members. Two backends exist: StaticBootstrap (what --peer feeds) and an in-memory one for tests. The planned Mainline DHT backend, which would let members find each other from the network secret alone, is not implemented. So today you bootstrap by passing one peer's id; after that the mesh is whatever those agents reach.

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. --transport local publishes nothing.
  • 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 WireGuard plugin keeps its own keys in its own wireguard.sqlite, wherever its configuration points, because plugin 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.

Documentation

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 before relying on any of this.

Licence

MIT OR Apache-2.0.