Fix restore-path main↔bleQueue deadlock and courier drop amplification (#1425)

* Fix restore-path main↔bleQueue deadlock and courier drop amplification A device froze permanently in a two-phone test. Debugger stacks showed an ABBA deadlock: the main actor was in bleQueue.sync (delivery-ack send → broadcastPacket → readLinkState) while bleQueue was in main.sync (captureBluetoothStatus reading backgroundTimeRemaining). The load that lined the two edges up came from a courier-drop amplification storm: drop dedup was in-memory only while the outbox driving 120s re-deposits is persisted, so every relaunch republished the same undelivered DM as a fresh 24h relay drop and every gateway relaunch re-fetched the whole backlog — ~20 copies of one DM delivered in 40ms, each triggering decrypt + delivery + ack + handshake work. Fixes, in rank order: - Edge B (P0): captureBluetoothStatus no longer main.syncs from bleQueue; backgroundTimeRemaining is sampled on main and cached behind a lock. Invariant documented: bleQueue must NEVER sync-dispatch to main. - Edge A (P0, defense in depth): sendDeliveryAck / sendReadReceipt / sendPrivateMessage / sendNoisePayload / triggerHandshake hop to messageQueue like sendMessage, so no main-actor call path reaches readLinkState's bleQueue.sync. - Drop dedup (P1): publishedDropKeys and seenDropEventIDs persist across relaunches (new BridgeDropDedupStore, entries expire with the 24h NIP-40 drop window; wiped on panic) — one drop per message ID per 24h regardless of relaunch count. - Receiver dedup (P1): openCourierEnvelope dedups on the inner private message ID before delivery, so a duplicate copy costs one decrypt and never re-delivers, re-acks, or re-triggers a handshake. - Handshake gating (P2): queued acks initiate a Noise handshake only for reachable peers; mail from absent/rotated identities no longer turns each copy into a mesh-wide handshake flood (the ack stays queued and flushes when a session eventually establishes). - Outbox (P3): re-enqueueing a queued message ID carries over its depositedCourierKeys so resends stop re-burning the same courier slots. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Review fixes: offline-drop durability, gateway handoff retry, restore-log freshness, coalesced persist Adversarial review of the storm/deadlock PR surfaced four issues: - Offline blackhole (must-fix): a deposit made while relays were down persisted its dedup key even though the drop only sat in the in-memory pending queue — app killed before reconnect meant the relaunch lost the drop but the persisted key blocked every re-deposit for 24h. The persisted snapshot now excludes keys still pending; they become durable only when flushPendingDrops actually publishes them. - Gateway handoff: seen-event IDs were consumed before the deliverToPeer handoff; a failed handoff (peer walked away) permanently dropped the event for a single-gateway island. deliverToPeer now reports whether the handoff was attempted, and a failure releases the seen slot so a relaunch or backlog redelivery retries. - Restore-path logs: central/peripheral-restore captures logged the init sentinel bgRemaining=∞. The cache is now seeded in init's main-thread branch and restore captures route through the sampler, which refreshes the cached budget before logging. - Persist cost: the dedup record was a full JSON encode + atomic write on the main actor per mutation (once per event during a backlog re-fetch). Writes now coalesce behind a 1s window, flushed immediately on background/terminate; panic wipe stays immediate. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Remove BoundedIDSet.remove, orphaned by the ExpiringIDSet migration The drop-dedup sets that needed slot release moved to ExpiringIDSet; remaining BoundedIDSet users only insert and check. Periphery caught it. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: jack <jackjackbits@users.noreply.github.com> Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-26 11:05:20 +00:00 · 2026-07-09 18:23:46 +02:00
parent 304460ee83
commit f8ab0a7dc0
13 changed files with 754 additions and 55 deletions
@@ -129,6 +129,15 @@ final class BLEService: NSObject {
    // Application state tracking (thread-safe)
    #if os(iOS)
    private var isAppActive: Bool = true  // Assume active initially
+    /// Last `UIApplication.shared.backgroundTimeRemaining` sampled on the
+    /// main thread, cached so bleQueue status logs can read it without ever
+    /// dispatching to main (see `captureBluetoothStatus` for the invariant).
+    private let backgroundTimeLock = NSLock()
+    private var _cachedBackgroundTimeRemaining: TimeInterval = .greatestFiniteMagnitude
+    private var cachedBackgroundTimeRemaining: TimeInterval {
+        backgroundTimeLock.lock(); defer { backgroundTimeLock.unlock() }
+        return _cachedBackgroundTimeRemaining
+    }
    #endif
    
    // MARK: - Core BLE Objects
@@ -176,6 +185,13 @@ final class BLEService: NSObject {

    // Ingress link tracking for duplicate and last-hop suppression
    private var ingressLinks = BLEIngressLinkRegistry()
+    // Inner message IDs of recently opened courier envelopes. Redundant
+    // copies of one message ride different envelopes (each seal uses a fresh
+    // ephemeral key, and bridge drops multiply across relays/couriers), so
+    // envelope-level dedup can't catch them; dedup on the inner ID before
+    // delivery so a duplicate costs one decrypt instead of a delivery + ack
+    // + handshake each. Owned by collectionsQueue barriers.
+    private var openedCourierMessageIDs = BoundedIDSet(capacity: TransportConfig.courierOpenedMessageIDCap)
    private let logRateLimiter = BLELogRateLimiter(defaultMinimumInterval: 5)

    private var pendingPeripheralWrites = BLEOutboundWriteBuffer()
@@ -265,12 +281,18 @@ final class BLEService: NSObject {
        
        // Set up application state tracking (iOS only)
        #if os(iOS)
-        // Check initial state on main thread
+        // Check initial state on main thread. The background-budget cache is
+        // seeded here too: a background-restore launch captures Bluetooth
+        // status before any lifecycle notification fires, and the init-time
+        // sentinel would log a meaningless bgRemaining=∞ for exactly the
+        // wake window that matters.
        if Thread.isMainThread {
            isAppActive = UIApplication.shared.applicationState == .active
+            refreshCachedBackgroundTimeRemaining()
        } else {
            DispatchQueue.main.sync {
                isAppActive = UIApplication.shared.applicationState == .active
+                refreshCachedBackgroundTimeRemaining()
            }
        }
        
@@ -777,7 +799,11 @@ final class BLEService: NSObject {
    }
    
    func triggerHandshake(with peerID: PeerID) {
-        initiateNoiseHandshake(with: peerID)
+        // Callers are on the main actor; the handshake broadcast sync-waits
+        // on bleQueue for link state, so hop off main first.
+        messageQueue.async { [weak self] in
+            self?.initiateNoiseHandshake(with: peerID)
+        }
    }
    
    // MARK: Noise identity/session access (narrow Transport wrappers)
@@ -932,6 +958,15 @@ final class BLEService: NSObject {

    
    func sendReadReceipt(_ receipt: ReadReceipt, to peerID: PeerID) {
+        // Hop like sendMessage: callers are often on the main actor, and the
+        // send path sync-waits on bleQueue for link state — the main thread
+        // must never block on bleQueue (see captureBluetoothStatus).
+        if DispatchQueue.getSpecific(key: messageQueueKey) == nil {
+            messageQueue.async { [weak self] in
+                self?.sendReadReceipt(receipt, to: peerID)
+            }
+            return
+        }
        let payload = BLENoisePayloadFactory.readReceipt(originalMessageID: receipt.originalMessageID)

        if noiseService.hasEstablishedSession(with: peerID) {
@@ -942,11 +977,15 @@ final class BLEService: NSObject {
                SecureLogger.error("Failed to send read receipt: \(error)")
            }
        } else {
-            // Queue for after handshake and initiate if needed
+            // Queue for after handshake; initiate only while the peer is
+            // around to answer (see sendDeliveryAck — absent senders must
+            // not turn queued acks into handshake floods).
            collectionsQueue.sync(flags: .barrier) {
                pendingNoiseSessionQueues.appendTypedPayload(payload, for: peerID)
            }
-            if !noiseService.hasSession(with: peerID) { initiateNoiseHandshake(with: peerID) }
+            if !noiseService.hasSession(with: peerID), isPeerReachable(peerID) {
+                initiateNoiseHandshake(with: peerID)
+            }
            SecureLogger.debug("🕒 Queued READ receipt for \(peerID.id.prefix(8))… until handshake completes", category: .session)
        }
    }
@@ -1395,6 +1434,15 @@ final class BLEService: NSObject {
    }
    
    func sendDeliveryAck(for messageID: String, to peerID: PeerID) {
+        // Hop like sendMessage: callers are often on the main actor, and the
+        // send path sync-waits on bleQueue for link state — the main thread
+        // must never block on bleQueue (see captureBluetoothStatus).
+        if DispatchQueue.getSpecific(key: messageQueueKey) == nil {
+            messageQueue.async { [weak self] in
+                self?.sendDeliveryAck(for: messageID, to: peerID)
+            }
+            return
+        }
        let payload = BLENoisePayloadFactory.delivered(messageID: messageID)

        if noiseService.hasEstablishedSession(with: peerID) {
@@ -1404,11 +1452,18 @@ final class BLEService: NSObject {
                SecureLogger.error("Failed to send delivery ACK: \(error)")
            }
        } else {
-            // Queue for after handshake and initiate if needed
+            // Queue for after handshake; initiate only while the peer is
+            // around to answer — couriered/bridged mail routinely arrives
+            // from absent (or rotated) identities, and every duplicate copy
+            // initiating a handshake broadcast turns one undeliverable ack
+            // into a mesh-wide flood. The queued ack flushes whenever a
+            // session eventually establishes.
            collectionsQueue.sync(flags: .barrier) {
                pendingNoiseSessionQueues.appendTypedPayload(payload, for: peerID)
            }
-            if !noiseService.hasSession(with: peerID) { initiateNoiseHandshake(with: peerID) }
+            if !noiseService.hasSession(with: peerID), isPeerReachable(peerID) {
+                initiateNoiseHandshake(with: peerID)
+            }
            SecureLogger.debug("🕒 Queued DELIVERED ack for \(peerID.id.prefix(8))… until handshake completes", category: .session)
        }
    }
@@ -1680,7 +1735,10 @@ extension BLEService: CBCentralManagerDelegate {
            recentPeripheralCache.record(peripheral, peripheralID: identifier, at: Date())
        }

-        captureBluetoothStatus(context: "central-restore")
+        // Via the sampler (not a direct capture): it refreshes the cached
+        // background budget on main first, so the restore log shows the real
+        // wake window instead of the init sentinel.
+        logBluetoothStatus("central-restore")

        if central.state == .poweredOn {
            startScanning()
@@ -2439,7 +2497,8 @@ extension BLEService: CBPeripheralManagerDelegate {
            }
        }

-        captureBluetoothStatus(context: "peripheral-restore")
+        // Via the sampler for a fresh background budget (see central-restore).
+        logBluetoothStatus("peripheral-restore")

        if peripheral.state == .poweredOn && !peripheral.isAdvertising {
            peripheral.startAdvertising(buildAdvertisementData())
@@ -2720,19 +2779,37 @@ extension BLEService {
    }

    private func logBluetoothStatus(_ context: String) {
-        bleQueue.async { [weak self] in
-            guard let self = self else { return }
-            self.captureBluetoothStatus(context: context)
-        }
+        scheduleBluetoothStatusSample(after: 0, context: context)
    }

    private func scheduleBluetoothStatusSample(after delay: TimeInterval, context: String) {
-        bleQueue.asyncAfter(deadline: .now() + delay) { [weak self] in
-            guard let self = self else { return }
-            self.captureBluetoothStatus(context: context)
+        #if os(iOS)
+        // Sample the main-actor background budget first (async hop, never a
+        // sync wait), then log from bleQueue off the cache — bleQueue must
+        // never block on main (see captureBluetoothStatus).
+        DispatchQueue.main.asyncAfter(deadline: .now() + delay) { [weak self] in
+            guard let self else { return }
+            self.refreshCachedBackgroundTimeRemaining()
+            self.bleQueue.async { self.captureBluetoothStatus(context: context) }
        }
+        #else
+        bleQueue.asyncAfter(deadline: .now() + delay) { [weak self] in
+            self?.captureBluetoothStatus(context: context)
+        }
+        #endif
    }

+    #if os(iOS)
+    /// Main thread only (reads main-actor UIApplication state).
+    private func refreshCachedBackgroundTimeRemaining() {
+        dispatchPrecondition(condition: .onQueue(.main))
+        let seconds = UIApplication.shared.backgroundTimeRemaining
+        backgroundTimeLock.lock()
+        _cachedBackgroundTimeRemaining = seconds
+        backgroundTimeLock.unlock()
+    }
+    #endif
+
    private func captureBluetoothStatus(context: String) {
        assert(DispatchQueue.getSpecific(key: bleQueueKey) != nil, "captureBluetoothStatus must run on bleQueue")

@@ -2750,11 +2827,15 @@ extension BLEService {
        }

        #if os(iOS)
-        var backgroundDescriptor = ""
-        var backgroundSeconds: TimeInterval = 0
-        DispatchQueue.main.sync {
-            backgroundSeconds = UIApplication.shared.backgroundTimeRemaining
-        }
+        // INVARIANT: bleQueue must NEVER sync-dispatch to the main thread.
+        // The main actor sync-waits on bleQueue along the send paths
+        // (readLinkState), so a main.sync here completes an ABBA deadlock —
+        // field-verified as a permanent freeze when a courier-drop storm put
+        // an ack send (main → bleQueue.sync) up against a status capture
+        // (bleQueue → main.sync). backgroundTimeRemaining is main-actor
+        // state, so it is sampled on main and cached.
+        let backgroundSeconds = cachedBackgroundTimeRemaining
+        let backgroundDescriptor: String
        if backgroundSeconds == .greatestFiniteMagnitude {
            backgroundDescriptor = " bgRemaining=∞"
        } else {
@@ -3044,6 +3125,15 @@ extension BLEService {

    
    private func sendNoisePayload(_ typedPayload: Data, to peerID: PeerID) {
+        // Hop like sendMessage: the Transport-facing wrappers (verify/vouch/
+        // group payloads) call this from the main actor, and the send path
+        // sync-waits on bleQueue for link state.
+        if DispatchQueue.getSpecific(key: messageQueueKey) == nil {
+            messageQueue.async { [weak self] in
+                self?.sendNoisePayload(typedPayload, to: peerID)
+            }
+            return
+        }
        guard noiseService.hasSession(with: peerID) else {
            // No session yet - queue the payload SYNCHRONOUSLY before initiating handshake
            // to prevent race where fast handshake completion drains empty queue
@@ -3291,6 +3381,23 @@ extension BLEService {
                SecureLogger.warning("⚠️ Courier envelope carried unsupported payload type", category: .session)
                return
            }
+            let payload = Data(typedPayload.dropFirst())
+            guard let innerMessageID = PrivateMessagePacket.decode(from: payload)?.messageID else {
+                SecureLogger.warning("⚠️ Courier envelope carried undecodable private message", category: .session)
+                return
+            }
+            // Redundant copies of one message arrive as distinct envelopes
+            // (fresh seal each: mesh couriers, bridge drops across relays),
+            // so dedup here on the inner message ID — before delivery, ack,
+            // and handshake work. A duplicate costs only the decrypt above
+            // and at most one ack ever goes out per message ID.
+            let firstOpen = collectionsQueue.sync(flags: .barrier) {
+                openedCourierMessageIDs.insert(innerMessageID)
+            }
+            guard firstOpen else {
+                SecureLogger.debug("📦 Dropping duplicate courier envelope for message \(innerMessageID.prefix(8))…", category: .session)
+                return
+            }
            // Couriered mail arrives while the sender is absent, so the UI's
            // block check can't resolve their fingerprint from a live session.
            // Gate here, where the full static key is in hand.
@@ -3306,7 +3413,6 @@ extension BLEService {
            let shortID = PeerID(publicKey: senderStaticKey)
            let isKnownOnMesh = collectionsQueue.sync { peerRegistry.info(for: shortID) != nil }
            let senderPeerID = isKnownOnMesh ? shortID : PeerID(hexData: senderStaticKey)
-            let payload = Data(typedPayload.dropFirst())
            SecureLogger.debug("📦 Opened courier envelope from \(senderPeerID.id.prefix(8))…", category: .session)
            sfMetrics?.record(.courierOpened)
            notifyUI { [weak self] in
@@ -3745,6 +3851,7 @@ extension BLEService {
    #if os(iOS)
    @objc private func appDidBecomeActive() {
        isAppActive = true
+        refreshCachedBackgroundTimeRemaining()
        // Restart scanning with allow duplicates when app becomes active
        if centralManager?.state == .poweredOn {
            centralManager?.stopScan()
@@ -3758,6 +3865,7 @@ extension BLEService {

    @objc private func appDidEnterBackground() {
        isAppActive = false
+        refreshCachedBackgroundTimeRemaining()
        // Restart scanning without allow duplicates in background
        if centralManager?.state == .poweredOn {
            centralManager?.stopScan()
@@ -3851,6 +3959,15 @@ extension BLEService {
    // MARK: Private Message Handling
    
    private func sendPrivateMessage(_ content: String, to recipientID: PeerID, messageID: String) {
+        // Hop like sendMessage: the Transport-facing wrappers call this from
+        // the main actor (router sends, favorite notifications), and the send
+        // path sync-waits on bleQueue for link state.
+        if DispatchQueue.getSpecific(key: messageQueueKey) == nil {
+            messageQueue.async { [weak self] in
+                self?.sendPrivateMessage(content, to: recipientID, messageID: messageID)
+            }
+            return
+        }
        // Sessions and wire recipient IDs are keyed by the short 16-hex form;
        // callers may pass the full 64-hex noise key (mirrors sendFilePrivate).
        let recipientID = recipientID.toShort()
@@ -32,14 +32,4 @@ struct BoundedIDSet {
        }
        return true
    }
-
-    /// Releases a previously inserted ID so it can be re-added later. Used
-    /// when a tentatively-recorded item is later abandoned (e.g. a queued
-    /// drop evicted before it was ever published) and must become retryable.
-    mutating func remove(_ id: String) {
-        guard members.remove(id) != nil else { return }
-        if let index = order.firstIndex(of: id) {
-            order.remove(at: index)
-        }
-    }
 }
@@ -9,6 +9,11 @@
 import BitFoundation
 import BitLogger
 import Foundation
+#if os(iOS)
+import UIKit
+#elseif os(macOS)
+import AppKit
+#endif

 /// Courier delivery over the internet bridge: sealed courier envelopes are
 /// parked on relays as kind-1401 "drops" tagged with their rotating
@@ -48,6 +53,10 @@ final class BridgeCourierService: ObservableObject {
        /// Encoded envelope cap for a drop (16 KiB ciphertext + TLV slack).
        static let maxDropEnvelopeBytes = 20 * 1024
        static let maxTrackedIDs = 512
+        /// Coalescing window for dedup-record writes: a backlog re-fetch
+        /// mutates the seen set once per event, and each snapshot save is a
+        /// full JSON encode + atomic write on the main actor.
+        static let dedupPersistCoalesceSeconds: TimeInterval = 1.0
    }

    static let shared = BridgeCourierService()
@@ -70,7 +79,9 @@ final class BridgeCourierService: ObservableObject {
    /// Opens a drop addressed to us (tag verified inside).
    var openEnvelope: (@MainActor (CourierEnvelope) -> Void)?
    /// Hands a drop to a matching local peer as a directed courier packet.
-    var deliverToPeer: (@MainActor (CourierEnvelope, PeerID) -> Void)?
+    /// Returns false when the handoff could not even be attempted (peer no
+    /// longer reachable), so the drop event stays retryable.
+    var deliverToPeer: (@MainActor (CourierEnvelope, PeerID) -> Bool)?
    /// Held envelopes eligible for (re)publish, honoring the cooldown.
    var heldEnvelopes: (@MainActor (TimeInterval) -> [CourierEnvelope])?
    /// Timer injection for tests; nil arms a real `Task`.
@@ -81,21 +92,90 @@ final class BridgeCourierService: ObservableObject {
    private(set) var myTagsHex: Set<String> = []
    private(set) var watchedPeerTags: [(peerID: PeerID, tagsHex: Set<String>)] = []
    private(set) var pendingDrops: [(envelope: CourierEnvelope, dedupKey: String?)] = []
-    /// Message IDs already published as drops (sender-side dedup).
-    private var publishedDropKeys: BoundedIDSet
-    /// Drop event IDs already handled (multi-relay dedup).
-    private var seenDropEventIDs: BoundedIDSet
+    /// Message IDs already published as drops (sender-side dedup) and drop
+    /// event IDs already handled (multi-relay dedup). Both persist across
+    /// relaunches: relays hold drops for the full 24h NIP-40 window and the
+    /// persisted outbox keeps re-depositing, so in-memory-only dedup meant
+    /// every relaunch republished the same message as a fresh drop and every
+    /// gateway relaunch re-delivered the whole backlog (field-verified
+    /// amplification storm). Entries age out with the 24h drop window.
+    private var publishedDropKeys: ExpiringIDSet
+    private var seenDropEventIDs: ExpiringIDSet
    private var subscriptionOpen = false
    private var lastSubscribedTags: Set<String> = []
    private var refreshTimerArmed = false
    private var lastAnnounceRefresh = Date.distantPast

    private let now: () -> Date
+    private let dedupStore: BridgeDropDedupStore

-    init(now: @escaping () -> Date = Date.init) {
+    private var dedupPersistScheduled = false
+
+    init(now: @escaping () -> Date = Date.init, dedupStore: BridgeDropDedupStore? = nil) {
        self.now = now
-        self.publishedDropKeys = BoundedIDSet(capacity: Limits.maxTrackedIDs)
-        self.seenDropEventIDs = BoundedIDSet(capacity: Limits.maxTrackedIDs)
+        self.dedupStore = dedupStore ?? BridgeDropDedupStore(persistsToDisk: !TestEnvironment.isRunningTests)
+        let snapshot = self.dedupStore.load()
+        let date = now()
+        self.publishedDropKeys = ExpiringIDSet(
+            capacity: Limits.maxTrackedIDs,
+            lifetime: CourierEnvelope.maxLifetimeSeconds,
+            entries: snapshot.publishedDropKeys,
+            now: date
+        )
+        self.seenDropEventIDs = ExpiringIDSet(
+            capacity: Limits.maxTrackedIDs,
+            lifetime: CourierEnvelope.maxLifetimeSeconds,
+            entries: snapshot.seenDropEventIDs,
+            now: date
+        )
+        // A coalesced dedup write scheduled just before a background kill
+        // would be lost; flush when the app backgrounds or terminates.
+        #if os(iOS)
+        let flushNotifications = [UIApplication.didEnterBackgroundNotification, UIApplication.willTerminateNotification]
+        #else
+        let flushNotifications = [NSApplication.willTerminateNotification]
+        #endif
+        for name in flushNotifications {
+            NotificationCenter.default.addObserver(forName: name, object: nil, queue: .main) { [weak self] _ in
+                MainActor.assumeIsolated { self?.flushDedupSnapshot() }
+            }
+        }
+    }
+
+    /// Schedules a coalesced write of the dedup record (see
+    /// `Limits.dedupPersistCoalesceSeconds`); lifecycle notifications flush
+    /// any scheduled write before a background kill could drop it.
+    private func persistDedup() {
+        guard !dedupPersistScheduled else { return }
+        dedupPersistScheduled = true
+        Task { @MainActor [weak self] in
+            try? await Task.sleep(nanoseconds: UInt64(Limits.dedupPersistCoalesceSeconds * 1_000_000_000))
+            guard let self else { return }
+            self.dedupPersistScheduled = false
+            self.flushDedupSnapshot()
+        }
+    }
+
+    /// Writes the dedup record now. Sender keys still sitting in the
+    /// in-memory `pendingDrops` queue are excluded: their drop is not durable
+    /// until it actually reaches a relay, and persisting the key early would
+    /// turn "app killed before relays connected" into a silent 24h blackhole
+    /// (the relaunch loses the queued drop but the persisted key blocks every
+    /// re-deposit). `flushPendingDrops` re-persists once they publish.
+    func flushDedupSnapshot() {
+        let pendingKeys = Set(pendingDrops.compactMap(\.dedupKey))
+        dedupStore.save(BridgeDropDedupStore.Snapshot(
+            publishedDropKeys: publishedDropKeys.entries.filter { !pendingKeys.contains($0.key) },
+            seenDropEventIDs: seenDropEventIDs.entries
+        ))
+    }
+
+    /// Panic wipe: forget queued drops and the persisted dedup record.
+    func wipe() {
+        pendingDrops.removeAll()
+        publishedDropKeys = ExpiringIDSet(capacity: Limits.maxTrackedIDs, lifetime: CourierEnvelope.maxLifetimeSeconds)
+        seenDropEventIDs = ExpiringIDSet(capacity: Limits.maxTrackedIDs, lifetime: CourierEnvelope.maxLifetimeSeconds)
+        dedupStore.wipe()
    }

    // MARK: - Sender role
@@ -108,14 +188,15 @@ final class BridgeCourierService: ObservableObject {
    @discardableResult
    func depositDrop(content: String, messageID: String, recipientNoiseKey: Data) -> Bool {
        guard bridgeEnabled?() ?? false else { return false }
-        guard !publishedDropKeys.contains(messageID) else { return false }
+        guard !publishedDropKeys.contains(messageID, now: now()) else { return false }
        guard let envelope = sealEnvelope?(content, messageID, recipientNoiseKey) else { return false }
        // An envelope that can't encode within the drop size caps fails the
        // same way on every attempt (size is a function of the content, not
        // of the sealing); consume the dedup slot so the retry sweep stops
        // re-running Noise sealing on a drop that can never ship.
        guard let encoded = envelope.encode(), encoded.count <= Limits.maxDropEnvelopeBytes else {
-            publishedDropKeys.insert(messageID)
+            publishedDropKeys.insert(messageID, now: now())
+            persistDedup()
            return false
        }
        // Only consume the sender-side dedup slot once the drop is durably
@@ -124,7 +205,8 @@ final class BridgeCourierService: ObservableObject {
        // router's retry sweep can attempt a fresh deposit rather than
        // marking the message "carried" and blocking retries forever.
        guard publishDrop(envelope, messageID: messageID) else { return false }
-        publishedDropKeys.insert(messageID)
+        publishedDropKeys.insert(messageID, now: now())
+        persistDedup()
        return true
    }

@@ -153,7 +235,10 @@ final class BridgeCourierService: ObservableObject {
                let evicted = pendingDrops.removeFirst()
                // The oldest queued drop is being dropped before it ever
                // published; release its dedup slot so it stays retryable.
-                if let key = evicted.dedupKey { publishedDropKeys.remove(key) }
+                if let key = evicted.dedupKey {
+                    publishedDropKeys.remove(key)
+                    persistDedup()
+                }
            }
            return true
        }
@@ -180,8 +265,13 @@ final class BridgeCourierService: ObservableObject {
        for item in queued where !publishDrop(item.envelope, messageID: item.dedupKey) {
            // Compose failed with relays up: release the slot so the router's
            // retry sweep can attempt a fresh deposit.
-            if let key = item.dedupKey { publishedDropKeys.remove(key) }
+            if let key = item.dedupKey {
+                publishedDropKeys.remove(key)
+            }
        }
+        // Flushed keys just became durable (published, so no longer excluded
+        // as pending) or were released above; either way the record changed.
+        persistDedup()
    }

    // MARK: - Subscription (recipient + gateway watch)
@@ -265,7 +355,8 @@ final class BridgeCourierService: ObservableObject {
    func handleDropEvent(_ event: NostrEvent) {
        guard bridgeEnabled?() ?? false else { return }
        guard event.kind == NostrProtocol.EventKind.courierDrop.rawValue else { return }
-        guard seenDropEventIDs.insert(event.id) else { return }
+        guard seenDropEventIDs.insert(event.id, now: now()) else { return }
+        persistDedup()
        guard let data = Data(base64Encoded: event.content),
              data.count <= Limits.maxDropEnvelopeBytes,
              let envelope = CourierEnvelope.decode(data),
@@ -285,7 +376,14 @@ final class BridgeCourierService: ObservableObject {
        }
        if let match = watchedPeerTags.first(where: { $0.tagsHex.contains(tagHex) }) {
            SecureLogger.info("📦🌉 Courier drop fetched for local peer \(match.peerID.id.prefix(8))…", category: .session)
-            deliverToPeer?(envelope, match.peerID)
+            if deliverToPeer?(envelope, match.peerID) != true {
+                // The best-effort handoff never left this device (the peer
+                // walked away between the relay fetch and the mesh send).
+                // Release the seen slot so a relaunch or backlog redelivery
+                // retries — a single-gateway island has no other carrier.
+                seenDropEventIDs.remove(event.id)
+                persistDedup()
+            }
        }
    }

@@ -0,0 +1,137 @@
+//
+// BridgeDropDedupStore.swift
+// bitchat
+//
+// This is free and unencumbered software released into the public domain.
+// For more information, see <https://unlicense.org>
+//
+
+import BitLogger
+import Foundation
+
+/// ID set with per-entry timestamps: entries expire after `lifetime` and the
+/// oldest are evicted past `capacity`. The bridge's dedup caches use this
+/// instead of `BoundedIDSet` because their contents persist across relaunches
+/// and must age out with the 24h drop window they guard.
+struct ExpiringIDSet {
+    private(set) var entries: [String: Date]
+    let capacity: Int
+    let lifetime: TimeInterval
+
+    init(capacity: Int, lifetime: TimeInterval, entries: [String: Date] = [:], now: Date = Date()) {
+        self.capacity = capacity
+        self.lifetime = lifetime
+        self.entries = entries
+        prune(now: now)
+    }
+
+    func contains(_ id: String, now: Date) -> Bool {
+        guard let recorded = entries[id] else { return false }
+        return now.timeIntervalSince(recorded) <= lifetime
+    }
+
+    /// Returns false when the ID was already present (and unexpired).
+    @discardableResult
+    mutating func insert(_ id: String, now: Date) -> Bool {
+        guard !contains(id, now: now) else { return false }
+        entries[id] = now
+        prune(now: now)
+        return true
+    }
+
+    /// Releases a previously inserted ID so it can be re-added later (e.g. a
+    /// queued drop evicted before it ever published must become retryable).
+    mutating func remove(_ id: String) {
+        entries.removeValue(forKey: id)
+    }
+
+    private mutating func prune(now: Date) {
+        if entries.contains(where: { now.timeIntervalSince($0.value) > lifetime }) {
+            entries = entries.filter { now.timeIntervalSince($0.value) <= lifetime }
+        }
+        let overflow = entries.count - capacity
+        guard overflow > 0 else { return }
+        for (id, _) in entries.sorted(by: { $0.value < $1.value }).prefix(overflow) {
+            entries.removeValue(forKey: id)
+        }
+    }
+}
+
+/// Disk persistence for the bridge courier's drop-dedup record. Relays hold
+/// drops for the full 24h NIP-40 window and redeliver them on every launch,
+/// and the 120s outbox sweep re-deposits anything undelivered — so with
+/// in-memory-only dedup every relaunch republished the same message as a
+/// fresh drop (fresh throwaway seal, undeduplicatable downstream) and every
+/// gateway relaunch re-delivered the whole backlog. Field-verified: ~20
+/// copies of one DM delivered in 40ms fed the storm behind a permanent
+/// device freeze. Persisting both sides caps this at one drop per message
+/// ID per 24h regardless of relaunch count.
+///
+/// Contents are opaque IDs (message UUIDs, relay event IDs) — no plaintext,
+/// no peer identities — so until-first-unlock protection matches
+/// `NostrProcessedEventStore`, and the file must load during a
+/// locked-background restoration relaunch. Wiped on panic with the rest of
+/// the courier state.
+final class BridgeDropDedupStore {
+    struct Snapshot: Codable {
+        var publishedDropKeys: [String: Date]
+        var seenDropEventIDs: [String: Date]
+    }
+
+    private let fileURL: URL?
+
+    /// - Parameter fileURL: Overrides the on-disk location (tests). Ignored
+    ///   when `persistsToDisk` is false.
+    init(persistsToDisk: Bool = true, fileURL: URL? = nil) {
+        self.fileURL = persistsToDisk ? (fileURL ?? Self.defaultFileURL()) : nil
+    }
+
+    func load() -> Snapshot {
+        guard let fileURL,
+              let data = try? Data(contentsOf: fileURL),
+              let snapshot = try? JSONDecoder().decode(Snapshot.self, from: data) else {
+            return Snapshot(publishedDropKeys: [:], seenDropEventIDs: [:])
+        }
+        return snapshot
+    }
+
+    func save(_ snapshot: Snapshot) {
+        guard let fileURL else { return }
+        guard !(snapshot.publishedDropKeys.isEmpty && snapshot.seenDropEventIDs.isEmpty) else {
+            try? FileManager.default.removeItem(at: fileURL)
+            return
+        }
+        do {
+            try FileManager.default.createDirectory(
+                at: fileURL.deletingLastPathComponent(),
+                withIntermediateDirectories: true
+            )
+            let data = try JSONEncoder().encode(snapshot)
+            var options: Data.WritingOptions = [.atomic]
+            #if os(iOS)
+            options.insert(.completeFileProtectionUntilFirstUserAuthentication)
+            #endif
+            try data.write(to: fileURL, options: options)
+        } catch {
+            SecureLogger.error("Failed to persist bridge drop dedup record: \(error)", category: .session)
+        }
+    }
+
+    /// Panic wipe: forget which drops we published or handled.
+    func wipe() {
+        guard let fileURL else { return }
+        try? FileManager.default.removeItem(at: fileURL)
+    }
+
+    private static func defaultFileURL() -> URL? {
+        guard let base = try? FileManager.default.url(
+            for: .applicationSupportDirectory,
+            in: .userDomainMask,
+            appropriateFor: nil,
+            create: true
+        ) else { return nil }
+        return base
+            .appendingPathComponent("courier", isDirectory: true)
+            .appendingPathComponent("bridge-drop-dedup.json")
+    }
+}
@@ -351,9 +351,15 @@ final class MessageRouter {
    }

    private func enqueue(_ message: QueuedMessage, for peerID: PeerID) {
+        var message = message
        var queue = outbox[peerID] ?? []
-        // Re-sending an already-queued ID replaces the entry (keeps attempt count fresh)
-        queue.removeAll { $0.messageID == message.messageID }
+        // Re-sending an already-queued ID replaces the entry (keeps attempt
+        // count fresh) but must not forget which couriers already carry it,
+        // or the replacement re-burns the same courier slots.
+        if let existing = queue.firstIndex(where: { $0.messageID == message.messageID }) {
+            message.depositedCourierKeys.formUnion(queue[existing].depositedCourierKeys)
+            queue.remove(at: existing)
+        }
        queue.append(message)

        // Enforce per-peer size limit with FIFO eviction
@@ -345,6 +345,9 @@ enum TransportConfig {
    // Cooldown between speculative multi-hop handovers of the same envelope
    // toward a recipient heard only via relayed announces.
    static let courierRemoteHandoverCooldownSeconds: TimeInterval = 10 * 60
+    // Recently opened courier inner message IDs kept for receiver-side dedup
+    // (redundant copies ride distinct seals, so only the inner ID matches).
+    static let courierOpenedMessageIDCap: Int = 512

    // One-time prekey bundles (forward-secret courier sealing)
    // Own gossip-sync round for bundles: modest cadence, bounded peer count,
@@ -1173,6 +1173,7 @@ final class ChatViewModel: ObservableObject, BitchatDelegate, TransportEventDele
        // our own queued outbox, the carried public history, and the
        // counters describing all of it
        CourierStore.shared.wipe()
+        BridgeCourierService.shared.wipe()
        messageRouter.wipeOutbox()
        GossipMessageArchive.wipeDefault()
        StoreAndForwardMetrics.shared.reset()
@@ -567,7 +567,7 @@ private extension ChatViewModelBootstrapper {
            bleService?.openBridgedCourierEnvelope(envelope)
        }
        courier.deliverToPeer = { [weak bleService] envelope, peerID in
-            bleService?.deliverBridgedEnvelope(envelope, to: peerID)
+            bleService?.deliverBridgedEnvelope(envelope, to: peerID) ?? false
        }
        courier.heldEnvelopes = { cooldown in
            CourierStore.shared.envelopesForBridgePublish(cooldown: cooldown)