Tailnet: interactive browser login + LAN↔tailnet fallback

Browser login — the auth key is now optional on both platforms. When the
embedded node starts with no key, TailnetNode asks the backend (LocalAPI
backendStatus — IPN state, deliberately not filesystem heuristics: tsnet
writes logs and a machine key on every start, registered or not) whether a
login is needed, triggers login-interactive, surfaces the auth URL through a
new statusStream()/.needsLogin, and waits for Running (4-minute deadline,
generation-fenced against stop/restart). The Mac auto-opens the login page
from Settings ▸ Remote and shows a re-open button; the iPhone auto-opens only
during user-initiated pairing (a background reconnect that suddenly needs a
login must not eject the user to Safari — Settings ▸ Tailscale carries the
link). The remote-access toggle now reflects the in-flight start instead of
snapping off for the whole login window, and toggling off mid-start is
honored at both commit points (before and after host.start).

LAN↔tailnet fallback — picking Tailnet now keeps LAN on too: the host runs
both listeners under a new CompositeSyncListener (merged accept stream; one
child ending doesn't end the rest) and the pairing QR carries both hints.
The phone builds an ordered candidate chain — LAN first (QR hint or Bonjour),
tailnet second — and walks it on pair and reconnect, so a phone that leaves
the Mac's Wi‑Fi rolls over to the tailnet and rolls back when it returns.
A per-channel 4s connect guard (readiness-checking, bound to exactly its
channel) keeps a stale LAN hint from hanging the chain; a stale-client guard
in consume() keeps a replaced client's tail events from advancing it; chain
exhaustion during pairing lands in a terminal failure instead of spinning on
"Connecting…"; routine pre-fallback handshake errors no longer flash the red
error bubble. "Connected · LAN / Tailnet" shows whichever transport won.

Multi-agent review: 11 confirmed findings (incl. the login gate being dead
code via tsnet's eager state-dir writes, and two connect-timeout races), all
fixed and re-verified. Suite green (24 sync-related tests incl. 3 new
CompositeSyncListener tests); macOS + iOS builds clean.

Co-Authored-By: Claude Fable 5 <[email protected]>

NucleicRemote/NucleicRemote/Models/RemoteStore.swift

@@ -155,6 +155,9 @@ final class RemoteStore: ObservableObject {
     private var activeTransport: SyncTransportHint = .lan
     /// The phone's embedded Tailscale node state, for Settings (nil = not running).
     @Published private(set) var tailnetStatus: String?
+    /// The Tailscale interactive-login URL while the node waits for a browser login (a
+    /// first tailnet start with no auth key). Auto-opened; Settings shows a re-open button.
+    @Published private(set) var tailnetLoginURL: URL?
     private var seenSeq: Set<UInt64> = []
     private var reconnectAttempts = 0
 
@@ -290,99 +293,168 @@ final class RemoteStore: ObservableObject {
             """)
     }
 
-    /// Pair from a scanned QR (SYNC §4.2): connect over the transport the QR names — LAN
+    /// One dialable way to reach the Mac. A connect builds an ordered candidate list — LAN
-    /// (explicit hint first, else Bonjour) or the Mac's tailnet IP via the phone's embedded
+    /// first (cheapest when reachable), then the tailnet (SYNC §3.2's LAN-then-fallback
-    /// Tailscale node — run XXpsk0, and on success pin the host key for future IK reconnects.
+    /// ordering) — and `attempt` walks it until one carries a session.
+    private enum TransportAttempt {
+        case lan(NWEndpoint)
+        case tailnet(host: String, port: UInt16)
+    }
+
+    /// The in-progress connect: remaining candidates plus everything needed to start a
+    /// client on whichever one succeeds. Cleared on `.ready` (chain done) and by teardown.
+    private struct ConnectPlan {
+        var remaining: [TransportAttempt]
+        let hostStaticKey: Data
+        let mode: SyncClient.Mode
+        let deviceID: String
+        let pairingPayload: PairingPayload?
+    }
+    private var connectPlan: ConnectPlan?
+    /// Kills a LAN attempt whose TCP connect just hangs (stale IP hint) so the chain can
+    /// move on — NWConnection's own timeout is far too slow for a fallback decision.
+    private var lanConnectTimeout: Task<Void, Never>?
+
+    /// Pair from a scanned QR (SYNC §4.2): try the QR's transports in order (LAN hint or
+    /// Bonjour first, then the Mac's tailnet IP via the phone's embedded node), run XXpsk0,
+    /// and on success pin the host key for future IK reconnects.
     func pair(with payload: PairingPayload) {
         teardown()
         connectivity = .connecting
         hostName = payload.hostName
-        let deviceID = IdentityStore.deviceID()
+        guard let hint = payload.transportHint else {
-        switch payload.transportHint {
+            connectivity = .failed("This pairing code needs a newer version of Nucleic Remote.")
-        case .lan:
-            activeTransport = .lan
-            guard let endpoint = resolveEndpoint(
-                fingerprint: payload.hostStaticKey.fingerprintHex,
-                lanHost: payload.lanHost, lanPort: payload.lanPort)
-            else { connectivity = .failed("No Mac found on this network"); return }
-            startClient(
-                channel: makeChannel(endpoint), hostStaticKey: payload.hostStaticKey,
-                mode: .pair(secret: payload.pairingSecret), deviceID: deviceID,
-                pairingPayload: payload)
-        case .tailnet:
-            activeTransport = .tailnet
-            guard let tailnetHost = payload.tailnetHost, let tailnetPort = payload.tailnetPort else {
-                connectivity = .failed("The pairing code is missing the Mac's tailnet address.")
             return
         }
-            connectTask = Task { [weak self] in
+        guard hint != .relay else {
-                guard let self else { return }
-                do {
-                    let channel = try await self.tailnetChannel(host: tailnetHost, port: tailnetPort)
-                    guard !Task.isCancelled else { channel.close(); return }
-                    self.startClient(
-                        channel: channel, hostStaticKey: payload.hostStaticKey,
-                        mode: .pair(secret: payload.pairingSecret), deviceID: deviceID,
-                        pairingPayload: payload)
-                } catch {
-                    guard !Task.isCancelled else { return }
-                    self.connectivity = .failed(error.localizedDescription)
-                }
-            }
-        case .relay:
             connectivity = .failed("Relay connections aren't supported yet.")
-        case nil:
+            return
-            connectivity = .failed("This pairing code needs a newer version of Nucleic Remote.")
         }
+        let candidates = buildCandidates(
+            fingerprint: payload.hostStaticKey.fingerprintHex,
+            lanHost: payload.lanHost, lanPort: payload.lanPort,
+            tailnet: hint == .tailnet ? (payload.tailnetHost, payload.tailnetPort) : nil)
+        guard !candidates.isEmpty else {
+            connectivity = .failed(hint == .tailnet && !TailnetSupport.isBuiltIn
+                ? TailnetError.notBuiltIn.errorDescription ?? "Tailscale support isn't built in"
+                : "No Mac found on this network")
+            return
+        }
+        connectPlan = ConnectPlan(
+            remaining: candidates, hostStaticKey: payload.hostStaticKey,
+            mode: .pair(secret: payload.pairingSecret), deviceID: IdentityStore.deviceID(),
+            pairingPayload: payload)
+        _ = tryNextCandidate()
     }
 
-    /// Reconnect to the already-paired host using IK against the pinned static key, over
+    /// Reconnect to the already-paired host using IK against the pinned static key: LAN
-    /// whichever transport the pairing recorded.
+    /// when reachable, else the pairing's tailnet hint — so a phone that leaves the Mac's
+    /// Wi‑Fi rolls over to the tailnet and rolls back when it returns.
     func reconnect() {
         guard let host = IdentityStore.loadPairedHost() else { connectivity = .unpaired; return }
         teardown()
         connectivity = reconnectAttempts == 0 ? .connecting : .reconnecting
         hostName = host.hostName
-        switch host.transportHint {
+        guard host.transportHint != .relay else {
-        case .lan:
+            connectivity = .failed("Relay connections aren't supported yet.")
-            activeTransport = .lan
-            guard let endpoint = resolveEndpoint(
-                fingerprint: host.fingerprint, lanHost: host.lanHost, lanPort: host.lanPort)
-            else { connectivity = .hostOffline; scheduleRetry(); return }
-            startClient(
-                channel: makeChannel(endpoint), hostStaticKey: host.hostStaticKey,
-                mode: .reconnect, deviceID: host.deviceID, pairingPayload: nil)
-        case .tailnet:
-            activeTransport = .tailnet
-            guard let tailnetHost = host.tailnetHost, let tailnetPort = host.tailnetPort else {
-                connectivity = .failed("Missing tailnet address — pair with your Mac again.")
             return
         }
+        let candidates = buildCandidates(
+            fingerprint: host.fingerprint,
+            lanHost: host.lanHost, lanPort: host.lanPort,
+            tailnet: host.transportHint == .tailnet ? (host.tailnetHost, host.tailnetPort) : nil)
+        guard !candidates.isEmpty else {
+            connectivity = .hostOffline
+            scheduleRetry()
+            return
+        }
+        connectPlan = ConnectPlan(
+            remaining: candidates, hostStaticKey: host.hostStaticKey,
+            mode: .reconnect, deviceID: host.deviceID, pairingPayload: nil)
+        _ = tryNextCandidate()
+    }
+
+    /// LAN first (explicit hint, else a Bonjour match), tailnet second when the pairing
+    /// carries one and this build can dial it.
+    private func buildCandidates(
+        fingerprint: String?, lanHost: String?, lanPort: UInt16?,
+        tailnet: (host: String?, port: UInt16?)?
+    ) -> [TransportAttempt] {
+        var candidates: [TransportAttempt] = []
+        if let endpoint = resolveEndpoint(fingerprint: fingerprint, lanHost: lanHost, lanPort: lanPort) {
+            candidates.append(.lan(endpoint))
+        }
+        if let tailnet, let host = tailnet.host, let port = tailnet.port, TailnetSupport.isBuiltIn {
+            candidates.append(.tailnet(host: host, port: port))
+        }
+        return candidates
+    }
+
+    /// Pop and dial the next candidate. False when the plan is exhausted (or gone) — the
+    /// caller then applies its terminal failure handling.
+    private func tryNextCandidate() -> Bool {
+        guard var plan = connectPlan, !plan.remaining.isEmpty else { return false }
+        let next = plan.remaining.removeFirst()
+        connectPlan = plan
+        attempt(next, plan: plan)
+        return true
+    }
+
+    private func attempt(_ candidate: TransportAttempt, plan: ConnectPlan) {
+        teardownClient()
+        switch candidate {
+        case .lan(let endpoint):
+            activeTransport = .lan
+            let channel = makeChannel(endpoint)
+            // Close the channel if TCP isn't up within the window (a stale IP hint would
+            // otherwise hang the chain on NWConnection's slow timeout); the finished stream
+            // then advances to the next candidate. The timer checks readiness itself — it's
+            // bound to exactly this channel, so a stale timer can never hit a later attempt.
+            lanConnectTimeout?.cancel()
+            lanConnectTimeout = Task { [weak channel] in
+                try? await Task.sleep(for: .seconds(4))
+                guard !Task.isCancelled, let channel, !channel.isReady else { return }
+                channel.close()
+            }
+            startClient(
+                channel: channel, hostStaticKey: plan.hostStaticKey,
+                mode: plan.mode, deviceID: plan.deviceID, pairingPayload: plan.pairingPayload)
+        case .tailnet(let host, let port):
+            activeTransport = .tailnet
             connectTask = Task { [weak self] in
                 guard let self else { return }
                 do {
-                    let channel = try await self.tailnetChannel(host: tailnetHost, port: tailnetPort)
+                    let channel = try await self.tailnetChannel(host: host, port: port)
                     guard !Task.isCancelled else { channel.close(); return }
                     self.startClient(
-                        channel: channel, hostStaticKey: host.hostStaticKey,
+                        channel: channel, hostStaticKey: plan.hostStaticKey,
-                        mode: .reconnect, deviceID: host.deviceID, pairingPayload: nil)
+                        mode: plan.mode, deviceID: plan.deviceID, pairingPayload: plan.pairingPayload)
                 } catch {
                     guard !Task.isCancelled else { return }
+                    self.tailnetAttemptFailed(error, isPairing: plan.pairingPayload != nil)
+                }
+            }
+        }
+    }
+
+    /// The tailnet is always the last candidate, so its failure ends the chain: terminal
+    /// for pairing and for anything retrying can't fix; otherwise offline + backoff.
+    private func tailnetAttemptFailed(_ error: Error, isPairing: Bool) {
+        if tryNextCandidate() { return }
+        if isPairing {
+            connectivity = .failed(error.localizedDescription)
+            return
+        }
         switch error {
         case TailnetError.notBuiltIn, TailnetError.notConfigured:
-                        // Retrying can't fix a missing auth key or a build without Tailscale.
+            connectivity = .failed(error.localizedDescription)
-                        self.connectivity = .failed(error.localizedDescription)
         default:
-                        self.connectivity = .hostOffline
+            connectivity = .hostOffline
-                        self.scheduleRetry()
+            scheduleRetry()
-                    }
-                }
-            }
-        case .relay:
-            connectivity = .failed("Relay connections aren't supported yet.")
         }
     }
 
+
     /// Create the `SyncClient` on an established channel and start consuming its events —
     /// the tail of every connect path, LAN or tailnet, pair or reconnect.
     private func startClient(
@@ -404,14 +476,38 @@ final class RemoteStore: ObservableObject {
     private func tailnetChannel(host: String, port: UInt16) async throws -> FDFrameChannel {
         guard TailnetSupport.isBuiltIn else { throw TailnetError.notBuiltIn }
         let config = Self.phoneTailnetConfig()
-        if config.authKey == nil, !config.hasExistingState {
-            throw TailnetError.notConfigured("Add your Tailscale auth key in Settings ▸ Tailscale first.")
-        }
         tailnetStatus = "Starting…"
+        // Mirror node status while the start is in flight. A first start with no auth key
+        // goes through the interactive browser login; pairing usually runs from the scanner
+        // sheet — not Settings — so take the user straight to the approval page.
+        let watcher = Task { [weak self] in
+            for await status in await TailnetNode.shared.statusStream() {
+                guard let self, !Task.isCancelled else { break }
+                self.tailnetStatus = status.label
+                if case .needsLogin(let url) = status, let loginURL = URL(string: url) {
+                    if self.tailnetLoginURL != loginURL {
+                        self.tailnetLoginURL = loginURL
+                        // Eject to Safari only for user-initiated pairing — the user is
+                        // actively watching. A routine reconnect that suddenly needs a
+                        // login (node revoked, state wiped) must not yank them out of the
+                        // app; Settings ▸ Tailscale carries the login link instead.
+                        if self.connectPlan?.pairingPayload != nil {
+                            UIApplication.shared.open(loginURL, options: [:], completionHandler: nil)
+                        }
+                    }
+                } else {
+                    self.tailnetLoginURL = nil
+                }
+            }
+        }
+        defer {
+            watcher.cancel()
+            tailnetLoginURL = nil
+        }
         do {
             try await TailnetNode.shared.ensureRunning(config: config)
         } catch TailnetError.timedOut(let message) {
-            // A login timeout won't fix itself — retrying would just block 45s per lap.
+            // A login/auth timeout won't fix itself — retrying would just block per lap.
             // Rethrow as .notConfigured so reconnect() treats it as terminal, not offline.
             tailnetStatus = await TailnetNode.shared.status.label
             throw TailnetError.notConfigured(message)
@@ -825,7 +921,11 @@ final class RemoteStore: ObservableObject {
         eventTask = Task { [weak self] in
             let stream = await client.start()
             for await event in stream {
-                await self?.handle(event, pairingPayload: pairingPayload)
+                // A replaced client's tail events (`.failed` is always chased by `.closed`)
+                // must not leak into the new attempt — they'd advance the candidate chain
+                // or schedule retries against a connection that no longer exists.
+                guard let self, self.client === client else { break }
+                await self.handle(event, pairingPayload: pairingPayload)
             }
         }
     }
@@ -836,6 +936,7 @@ final class RemoteStore: ObservableObject {
             break
         case .ready(let welcome):
             reconnectAttempts = 0
+            connectPlan = nil  // the chain found its transport
             connectivity = .connected(activeTransport)
             hostName = welcome.host.hostName
             capabilities = welcome.capabilities
@@ -917,12 +1018,28 @@ final class RemoteStore: ObservableObject {
             // Losing an approval race isn't an error worth interrupting for; the card
             // collapses on the matching `approvalResolved`.
             guard error.code != .alreadyResolved else { break }
+            // While the connect chain is still resolving a transport, a pre-ready error
+            // (e.g. "handshake closed" from a LAN probe about to fall back to tailnet) is
+            // routine, not news — the follow-on `.closed` advances the chain silently.
+            guard connectPlan == nil else { break }
             showError(error.message, sessionID: error.sessionID)
         case .failed(let message):
+            // Another candidate may still carry the session (e.g. LAN died → tailnet).
+            if tryNextCandidate() { break }
+            connectPlan = nil
             connectivity = .failed(message)
             scheduleRetry()
         case .closed:
-            if connectivity.isLive { connectivity = .reconnecting }
+            if !connectivity.isLive, tryNextCandidate() { break }
+            if connectivity.isLive {
+                connectivity = .reconnecting
+            } else if connectPlan?.pairingPayload != nil {
+                // A pairing chain died silently (every candidate closed pre-welcome).
+                // There's no retry loop before a pairing succeeds, so without a terminal
+                // state this would sit on "Connecting…" forever.
+                connectivity = .failed("Couldn't connect to your Mac — check that it's reachable, then scan again.")
+            }
+            connectPlan = nil
             scheduleRetry()
         }
     }
@@ -963,6 +1080,15 @@ final class RemoteStore: ObservableObject {
         retryTask = nil
         connectTask?.cancel()
         connectTask = nil
+        connectPlan = nil
+        teardownClient()
+    }
+
+    /// Drop just the current client/channel — what moving to the next transport candidate
+    /// needs, without discarding the rest of the plan.
+    private func teardownClient() {
+        lanConnectTimeout?.cancel()
+        lanConnectTimeout = nil
         eventTask?.cancel()
         eventTask = nil
         if let client { Task { await client.disconnect() } }

NucleicRemote/NucleicRemote/Net/NWFrameChannel.swift

@@ -17,6 +17,16 @@ final class NWFrameChannel: FrameChannel, @unchecked Sendable {
     var onReady: (@Sendable () -> Void)?
     var onFailed: (@Sendable (String) -> Void)?
 
+    /// Whether TCP reached `.ready` — latched, thread-safe. The connect-timeout guard
+    /// polls this instead of relying on a callback that could race connection start.
+    private let stateLock = NSLock()
+    private var ready = false
+    var isReady: Bool {
+        stateLock.lock()
+        defer { stateLock.unlock() }
+        return ready
+    }
+
     init(endpoint: NWEndpoint) {
         let params = NWParameters.tcp
         params.includePeerToPeer = true
@@ -28,7 +38,7 @@ final class NWFrameChannel: FrameChannel, @unchecked Sendable {
 
         connection.stateUpdateHandler = { [weak self] state in
             switch state {
-            case .ready: self?.onReady?()
+            case .ready: self?.markReady(); self?.onReady?()
             case .failed(let error): self?.onFailed?("\(error)"); self?.continuation.finish()
             case .cancelled: self?.continuation.finish()
             default: break
@@ -38,6 +48,12 @@ final class NWFrameChannel: FrameChannel, @unchecked Sendable {
         receiveLoop()
     }
 
+    private func markReady() {
+        stateLock.lock()
+        ready = true
+        stateLock.unlock()
+    }
+
     func frames() -> AsyncStream<Data> { stream }
 
     func send(_ frame: Data) {

NucleicRemote/NucleicRemote/Views/SettingsView.swift

@@ -56,8 +56,9 @@ struct SettingsView: View {
                     if !store.hostName.isEmpty {
                         LabeledContent("Mac", value: store.hostName)
                     }
-                    if let transport = IdentityStore.loadPairedHost()?.transportHint {
+                    // What this pairing can dial — the live transport is in Status above.
-                        LabeledContent("Transport", value: transport.label)
+                    if let host = IdentityStore.loadPairedHost() {
+                        LabeledContent("Transports", value: host.tailnetHost != nil ? "LAN + Tailnet" : "LAN")
                     }
                     Button("Reconnect") { store.reconnect() }
                         .disabled(!store.isPaired)
@@ -78,6 +79,11 @@ struct SettingsView: View {
                         if let status = store.tailnetStatus {
                             LabeledContent("Node", value: status)
                         }
+                        if let loginURL = store.tailnetLoginURL {
+                            Link(destination: loginURL) {
+                                Label("Open Tailscale login", systemImage: "arrow.up.forward.app")
+                            }
+                        }
                     } else {
                         Text("This build doesn't include Tailscale support.")
                             .foregroundStyle(.secondary)
@@ -85,7 +91,7 @@ struct SettingsView: View {
                 } header: {
                     Text("Tailscale")
                 } footer: {
-                    Text("Needed only when your Mac shares over Tailscale (Mac ▸ Settings ▸ Remote ▸ Connect via). Create an auth key in the Tailscale admin console (Settings ▸ Keys); it's kept in the Keychain and used once to join your tailnet — after that the phone stays registered.")
+                    Text("Needed only when your Mac shares over Tailscale (Mac ▸ Settings ▸ Remote ▸ Connect via). Leave the key empty to approve this iPhone in your browser on first connect, or create an auth key in the Tailscale admin console (kept in the Keychain, used once to join your tailnet).")
                 }
 
                 Section("This device") {