feat(discovery): native-protocol LAN auto-discovery over mDNS

Both the unified host (serve --native) and standalone m3-host now advertise the
native punktfunk/1 service over mDNS (_punktfunk._udp) — the analogue of the
GameStream _nvstream._tcp advert. TXT records carry proto, the host cert
fingerprint (fp, the value clients pin), the pairing requirement
(pair=required|optional), and the host id. New crate::discovery module, wired
into m3::serve so both host entry points get it; best-effort, never blocks
streaming (--connect always works).

Client gains `punktfunk-client-rs --discover [SECS]`: browses the LAN and prints
each host (name, addr:port, pairing, fingerprint), then exits. Apple clients
browse the same service natively via NWBrowser (service type + TXT keys are the
contract).

Validated cross-LAN: the dev box discovered the GNOME-box appliance
(pair=required) and a standalone synthetic host (pair=optional); fingerprint and
pairing state correct in both.

Also refresh the now-stale sendmmsg caveat in the bitrate doc (batched/paced send
landed + validated to 1 Gbps) and mark the encode|send thread split done in §12.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CLAUDE.md

@@ -43,6 +43,10 @@ Low-latency desktop/game streaming stack, Linux-first, with a shared Rust protoc
   no offline dictionary attack) establishes mutual trust:
   clients present persistent identities via QUIC client auth, the host stores paired
   fingerprints (`punktfunk1-paired.json`) and can gate sessions with `--require-pairing`.
+  **LAN auto-discovery**: both `serve --native` and `m3-host` advertise the native service over
+  mDNS (`_punktfunk._udp`, `crate::discovery`) with TXT `proto`/`fp`(cert fingerprint to
+  pin)/`pair`(required|optional)/`id`; `punktfunk-client-rs --discover` lists hosts, Apple clients
+  browse the same service via NWBrowser (validated cross-LAN 2026-06-12).
   **Mid-stream mode renegotiation**: `Reconfigure` on the still-open control stream — the
   host rebuilds output+encoder at the new mode in ~90 ms while the data plane runs on
   (validated live: one .h265 with 720p and 1080p segments). Measured on-box at 720p120: 1680/1680 frames, **p50 0.83 ms**

Cargo.lock

@@ -1953,6 +1953,7 @@ name = "punktfunk-client-rs"
 version = "0.0.1"
 dependencies = [
  "anyhow",
+ "mdns-sd",
  "opus",
  "punktfunk-core",
  "quinn",

crates/punktfunk-client-rs/Cargo.toml

@@ -15,6 +15,9 @@ tokio = { version = "1", features = ["rt-multi-thread", "net", "time", "macros"]
 anyhow = "1"
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
+# LAN host discovery (`--discover`): browse the native `_punktfunk._udp` mDNS service the host
+# advertises (same crate/version the host advertises with).
+mdns-sd = "0.20"
 
 # Linux-only: --mic-test's Opus encoder (libopus). The mic UPLINK itself is portable —
 # only this synthetic-tone test rig needs the encoder.

crates/punktfunk-client-rs/src/main.rs

@@ -31,8 +31,12 @@
 //! host honors it where available (DualSense needs Linux UHID), else falls back to X-Box 360,
 //! and reports the resolved choice in its Welcome (logged as `session offer … gamepad=…`).
 //!
+//! `--discover [SECS]` browses the LAN for native (`_punktfunk._udp`) hosts the host advertises
+//! over mDNS, prints each (name, addr:port, pairing requirement, cert fingerprint to pin), and
+//! exits without connecting.
+//!
 //! Usage: `punktfunk-client-rs [--connect HOST:PORT] [--mode WxHxFPS] [--out FILE] [--input-test]
-//!         [--pin HEX] [--compositor NAME] [--gamepad NAME]`
+//!         [--pin HEX] [--compositor NAME] [--gamepad NAME] | --discover [SECS]`
 //! (M4 adds VAAPI decode + wgpu present on this skeleton.)
 
 use anyhow::{anyhow, Context, Result};
@@ -75,6 +79,9 @@ struct Args {
     /// `--speed-test KBPS:MS` — after the stream starts, ask the host for a `MS`-millisecond
     /// bandwidth probe burst at `KBPS`, then report measured throughput + loss.
     speed_test: Option<(u32, u32)>,
+    /// `--discover [SECS]` — browse the LAN for native (`_punktfunk._udp`) hosts for `SECS`
+    /// seconds (default 4), print what's found, and exit. No connection is made.
+    discover: Option<u64>,
 }
 
 fn parse_mode(m: &str) -> Option<Mode> {
@@ -191,6 +198,12 @@ fn parse_args() -> Args {
             let (kbps, ms) = s.split_once(':')?;
             Some((kbps.parse().ok()?, ms.parse().ok()?))
         }),
+        // `--discover` may be a bare flag or carry a seconds value (`--discover 8`); only treat
+        // the following token as a count when it parses as a number (else it's the next flag).
+        discover: argv
+            .iter()
+            .any(|a| a == "--discover")
+            .then(|| get("--discover").and_then(|s| s.parse().ok()).unwrap_or(4)),
     }
 }
 
@@ -215,6 +228,10 @@ fn main() {
 }
 
 fn run(args: Args) -> Result<()> {
+    // Discovery mode: browse the LAN for native hosts, print them, and exit (no connection).
+    if let Some(secs) = args.discover {
+        return discover(secs);
+    }
     // Pairing mode: run the PIN ceremony and print the fingerprint to pin, then exit.
     if let Some(pin) = &args.pair {
         let (host, port) = args
@@ -245,6 +262,75 @@ fn run(args: Args) -> Result<()> {
     rt.block_on(session(args))
 }
 
+/// Browse the LAN for native (`_punktfunk._udp`) hosts for `secs` seconds and print them, then
+/// exit — the discovery side of the host's mDNS advert (host crate `discovery.rs`). TXT keys:
+/// `fp` (host cert fingerprint to pin), `pair` (required|optional), `id` (stable host id).
+fn discover(secs: u64) -> Result<()> {
+    use mdns_sd::{ServiceDaemon, ServiceEvent};
+    use std::collections::BTreeMap;
+    use std::time::{Duration, Instant};
+
+    let daemon = ServiceDaemon::new().context("create mDNS daemon")?;
+    let receiver = daemon
+        .browse("_punktfunk._udp.local.")
+        .context("browse _punktfunk._udp")?;
+    tracing::info!(secs, "browsing for native punktfunk/1 hosts (_punktfunk._udp)…");
+    // One row per host, keyed by the stable uniqueid (falls back to the fullname) so the same
+    // host re-advertising or answering on several interfaces collapses to a single entry.
+    let mut hosts: BTreeMap<String, String> = BTreeMap::new();
+    let deadline = Instant::now() + Duration::from_secs(secs);
+    loop {
+        let remaining = deadline.saturating_duration_since(Instant::now());
+        if remaining.is_zero() {
+            break;
+        }
+        // Timeout == time left to the deadline: an event returns immediately, otherwise the recv
+        // returns Err exactly at the deadline (or if the daemon channel closes) and we stop.
+        match receiver.recv_timeout(remaining) {
+            Ok(ServiceEvent::ServiceResolved(info)) => {
+                let props = info.get_properties();
+                let val = |k: &str| props.get_property_val_str(k).unwrap_or("").to_string();
+                let addr = info
+                    .get_addresses()
+                    .iter()
+                    .next()
+                    .map(|a| a.to_string())
+                    .unwrap_or_else(|| "?".into());
+                let fp = val("fp");
+                let fp_short = fp.get(..16).unwrap_or(fp.as_str());
+                let name = info.get_fullname().split('.').next().unwrap_or("?").to_string();
+                let id = val("id");
+                let key = if id.is_empty() {
+                    info.get_fullname().to_string()
+                } else {
+                    id
+                };
+                let row = format!(
+                    "  {name:<24} {addr}:{:<6} pair={:<9} fp={fp_short}…",
+                    info.get_port(),
+                    val("pair"),
+                );
+                hosts.insert(key, row);
+            }
+            Ok(_) => {} // SearchStarted / ServiceFound / removals — ignore
+            Err(_) => break,
+        }
+    }
+    let _ = daemon.shutdown();
+    if hosts.is_empty() {
+        println!("no native punktfunk/1 hosts found on the LAN ({secs}s)");
+    } else {
+        println!("native punktfunk/1 hosts ({}):", hosts.len());
+        for row in hosts.values() {
+            println!("{row}");
+        }
+        println!(
+            "\nconnect with: punktfunk-client-rs --connect <addr:port> [--pin <fp> | --pair <PIN>]"
+        );
+    }
+    Ok(())
+}
+
 async fn session(args: Args) -> Result<()> {
     let remote: std::net::SocketAddr = args.connect.parse().context("--connect host:port")?;
     let identity = load_or_create_identity()?;

crates/punktfunk-host/src/discovery.rs

@@ -0,0 +1,65 @@
+//! mDNS advertisement of the native punktfunk/1 service so native clients auto-discover the
+//! host — the native-protocol analogue of the GameStream `_nvstream._tcp` advert
+//! ([`crate::gamestream::mdns`]).
+//!
+//! The service type is **`_punktfunk._udp.local.`** (UDP because punktfunk/1 is QUIC, and the
+//! advertised port is the QUIC control/data port a client `--connect`s). TXT records carry:
+//! - `proto` — the wire protocol id ([`NATIVE_PROTO`]), so a future incompatible revision is
+//!   distinguishable by discovery alone;
+//! - `fp` — the host certificate SHA-256 (lowercase hex), the exact value a client pins. mDNS is
+//!   unauthenticated, so this is advisory — TOFU/pinning still verifies it on connect — but it
+//!   lets a picker show the fingerprint and pre-pin a chosen host;
+//! - `pair` — `required` or `optional`, so a client can tell up front whether it must run the PIN
+//!   pairing ceremony before it can stream;
+//! - `id` — the stable host uniqueid (dedup across IPs / re-advertises).
+
+use anyhow::{Context, Result};
+use mdns_sd::{ServiceDaemon, ServiceInfo};
+use std::collections::HashMap;
+use std::net::IpAddr;
+
+/// The native-protocol mDNS service type. Clients browse this to find punktfunk/1 hosts.
+pub const NATIVE_SERVICE: &str = "_punktfunk._udp.local.";
+
+/// Wire protocol id advertised in the `proto` TXT record.
+pub const NATIVE_PROTO: &str = "punktfunk/1";
+
+/// Holds the mDNS daemon; dropping it unregisters the service.
+pub struct Advert {
+    _daemon: ServiceDaemon,
+}
+
+/// Advertise the native host on the LAN. `fingerprint` is the host cert SHA-256 (lowercase hex);
+/// `require_pairing` tells a discovering client whether it must pair before it can stream.
+pub fn advertise_native(
+    hostname: &str,
+    ip: IpAddr,
+    port: u16,
+    fingerprint: &str,
+    require_pairing: bool,
+    uniqueid: &str,
+) -> Result<Advert> {
+    let daemon = ServiceDaemon::new().context("create mDNS daemon")?;
+    let host_name = format!("{hostname}.local.");
+    let mut props: HashMap<String, String> = HashMap::new();
+    props.insert("proto".into(), NATIVE_PROTO.into());
+    props.insert("fp".into(), fingerprint.to_string());
+    props.insert(
+        "pair".into(),
+        if require_pairing { "required" } else { "optional" }.into(),
+    );
+    props.insert("id".into(), uniqueid.to_string());
+    let service = ServiceInfo::new(NATIVE_SERVICE, hostname, &host_name, ip, port, props)
+        .context("build native mDNS ServiceInfo")?;
+    daemon
+        .register(service)
+        .context("register native mDNS service")?;
+    tracing::info!(
+        service = "_punktfunk._udp",
+        port,
+        host = %host_name,
+        pair = if require_pairing { "required" } else { "optional" },
+        "native punktfunk/1 mDNS advertising"
+    );
+    Ok(Advert { _daemon: daemon })
+}

crates/punktfunk-host/src/m3.rs

@@ -159,6 +159,27 @@ pub(crate) async fn serve(opts: M3Options, np: Arc<NativePairing>) -> Result<()>
         "punktfunk/1 host listening (QUIC) — clients pin this fingerprint"
     );
 
+    // mDNS: advertise the native service so clients auto-discover this host (the analogue of the
+    // GameStream _nvstream advert; both run in the unified host). Held for the host's lifetime —
+    // dropping `_advert` unregisters. Best-effort: a discovery failure must not stop streaming
+    // (manual `--connect HOST:PORT` always works), so we log and continue.
+    let _advert = match crate::gamestream::Host::detect() {
+        Ok(h) => crate::discovery::advertise_native(
+            &h.hostname,
+            h.local_ip,
+            opts.port,
+            &fingerprint_hex(&fingerprint),
+            opts.require_pairing,
+            &h.uniqueid,
+        )
+        .map_err(|e| tracing::warn!(error = %format!("{e:#}"), "native mDNS advertise failed (continuing)"))
+        .ok(),
+        Err(e) => {
+            tracing::warn!(error = %format!("{e:#}"), "host detect for mDNS failed (continuing)");
+            None
+        }
+    };
+
     // One audio capturer for the whole host lifetime, handed from session to session
     // (avoids a PipeWire stream setup per session — see AudioCapSlot).
     let audio_cap: AudioCapSlot = Arc::new(std::sync::Mutex::new(None));
@@ -236,10 +257,10 @@ const DEFAULT_BITRATE_KBPS: u32 = 20_000;
 /// above unusable, and a **2 Gbps** ceiling is generous headroom over the 1 Gbps+ target that
 /// GF(2¹⁶) Leopard FEC was built to reach — it lifts the GF(2⁸)/~1 Gbps wall, and at 1 Gbps a frame
 /// is only a few-hundred shards in one block (far under the 65535 limit). Enough for 5K@240 with
-/// margin. Resolved value is echoed in `Welcome::bitrate_kbps`. CAVEAT: the native data plane still
+/// margin. Resolved value is echoed in `Welcome::bitrate_kbps`. The native data plane batches sends
-/// does one `send()` syscall per packet (no `sendmmsg` batching / paced send thread yet), so
+/// (`sendmmsg`) and paces each frame on a dedicated send thread (microburst cap), validated to a
-/// sustained multi-hundred-Mbps can show send-buffer drops (counted as `packets_send_dropped`)
+/// clean 1 Gbps with zero send-buffer drops; sustained overruns are still counted as
-/// until that lands — see the 1 Gbps work in the roadmap.
+/// `packets_send_dropped`.
 const MIN_BITRATE_KBPS: u32 = 500;
 const MAX_BITRATE_KBPS: u32 = 2_000_000;
 

crates/punktfunk-host/src/main.rs

@@ -15,6 +15,7 @@
 
 mod audio;
 mod capture;
+mod discovery;
 mod encode;
 mod gamestream;
 mod inject;
@@ -437,6 +438,8 @@ NOTES:
     'portal' needs headless Sway + xdg-desktop-portal-wlr running in this session
     (see docs/linux-setup.md). 'synthetic' needs no capture session and always runs.
     Encoded AUs are written to a playable file AND (unless --no-loopback) fed through a
-    punktfunk_core host→client loopback that reassembles and byte-verifies each one."
+    punktfunk_core host→client loopback that reassembles and byte-verifies each one.
+    Both 'serve --native' and 'm3-host' advertise the native service over mDNS
+    (_punktfunk._udp) for client auto-discovery — 'punktfunk-client-rs --discover' lists them."
     );
 }

docs/roadmap.md

@@ -289,17 +289,42 @@ buffer; `sendmmsg`/`recvmmsg` batching; the capture-timestamp anchor placement.
   the ~150 Mbps@60 frame size where drops began). Plus **per-frame instrumentation** (PUNKTFUNK_PERF):
   `encode_us` + `pace_us` p50/p99/max + immediate-vs-paced counts, so the cap is tunable against real
   numbers. **Validate with the LAN soak before raising the cap** (`send_dropped` must stay 0).
+- **Done & live (`b295a5b`; validated on the GNOME box 2026-06-12):** **encode|send thread split**
+  on the native path — a dedicated `send_loop` thread owns the `Session` and does seal+pace+send+
+  probes; the encode thread captures+encodes+handles reconfig and hands `FrameMsg` over a bounded
+  `sync_channel(3)` with backpressure. Removes the serialization (~2–8 ms @60–120 fps) and is the
+  substrate the slice wrapper needs. Real-NIC soak (host on the Ubuntu/GNOME box, client over the
+  LAN): `send_dropped=0` at 720p60 / 1080p120, and a 1 Gbps probe pushed 625 MB in 5 s clean.
 - **Bigger bets (ordered, deferred — need real-NIC/GPU/Mac validation):**
-  1. **Encode|send thread split** on the native path (port GameStream's `spawn_sender` + depth-2
+  1. **Wall-clock skew handshake + glass-to-glass probe** (`tools/latency-probe`) — measures the two
-     channel; `seal_frame` stays on the encode thread, `send_sealed` on a send thread) — removes the
-     serialization (~2–8 ms @60–120 fps), and is the substrate the slice wrapper needs.
-  2. **Wall-clock skew handshake + glass-to-glass probe** (`tools/latency-probe`) — measures the two
      biggest unmeasured terms (render→capture, decode→present); client present-stamp vs the AU's
      `pts_ns` (already attached).
-  3. **CUDA stream+event** to drop one of two redundant `cuCtxSynchronize` in `submit_cuda` (keep the
+  2. **CUDA stream+event** to drop one of two redundant `cuCtxSynchronize` in `submit_cuda` (keep the
      copy) — ~0.1–0.4 ms@720p, ~1 ms@5K; only if per-stage timing proves the sync is on the path.
-  4. **Stage-2 Apple presenter** (`VTDecompressionSession` → `CAMetalLayer`, hand-paced) — ~0.5 refresh
+  3. **Stage-2 Apple presenter** (`VTDecompressionSession` → `CAMetalLayer`, hand-paced) — ~0.5 refresh
      off the present tail (biggest client win at 60 Hz); gate on the probe proving present is real.
-  5. **NVENC slice-mode wrapper** (roadmap §2 sub-frame pipelining) — per-slice transmit overlaps
+  4. **NVENC slice-mode wrapper** (roadmap §2 sub-frame pipelining) — per-slice transmit overlaps
      encode+send within a frame (~3–6 ms at 4K/5K/IDR); large + driver-ABI-fragile, on top of the
      thread split, only after measurement justifies it.
+
+## 13. Native-protocol LAN auto-discovery ✅ *(done — 2026-06-12, validated cross-LAN)*
+
+The native protocol had no discovery — clients connected by `--connect HOST:PORT` only, while
+GameStream already auto-discovered via mDNS (`_nvstream._tcp`). Now both the unified host
+(`serve --native`) and standalone `m3-host` advertise the native service over mDNS:
+
+- **Service**: `_punktfunk._udp.local.` (UDP — punktfunk/1 is QUIC; the advertised port is the QUIC
+  control/data port). Host side: `crate::discovery::advertise_native`, wired into `m3::serve` so
+  both host entry points get it; best-effort (a discovery failure never blocks streaming —
+  `--connect` always works). The advert is held for the host's lifetime (RAII unregister).
+- **TXT records**: `proto=punktfunk/1`, `fp=<host cert SHA-256>` (the value a client pins — advisory
+  over unauthenticated mDNS, TOFU/pinning still verifies on connect), `pair=required|optional`
+  (so a picker knows up front whether the PIN ceremony is needed), `id=<host uniqueid>` (dedup).
+- **Client**: `punktfunk-client-rs --discover [SECS]` browses and prints each host (name, addr:port,
+  pairing, fingerprint), then exits. Apple clients browse the same service natively via NWBrowser
+  (Bonjour) — no Rust-connector dependency; this section's service type + TXT keys are the contract.
+- **Validated**: cross-LAN — dev box discovered the GNOME-box appliance
+  (`home-worker-3 192.168.1.248:9777 pair=required fp=1dcf3a…`) and a standalone synthetic host
+  (`pair=optional`); fingerprint + pairing state correct in both.
+- **Next** (not done): wire NWBrowser discovery into the Apple client UI (host picker); the
+  host-side contract above is all it needs.