v1.16.5

changelog.md

@@ -1,5 +1,20 @@
 # Changelog
 
+## 2026-03-31 - 1.16.5 - fix(rust-userspace-nat)
+improve TCP session backpressure, buffering, and idle cleanup in userspace NAT
+
+- apply proper bridge-channel backpressure by reserving channel capacity before consuming smoltcp TCP data
+- defer bridge sender initialization until the bridge task starts and track TCP session activity timestamps
+- cap per-session pending TCP send buffers at 512KB and abort stalled sessions when clients cannot keep up
+- add idle TCP session cleanup and switch NAT polling to a dynamic smoltcp-driven delay
+
+## 2026-03-31 - 1.16.4 - fix(server)
+register preloaded WireGuard clients as peers on server startup
+
+- Adds configured clients from the runtime registry to the WireGuard listener when the server starts.
+- Ensures clients loaded from config can complete WireGuard handshakes without requiring separate peer registration.
+- Logs a warning if automatic peer registration fails for an individual client.
+
 ## 2026-03-31 - 1.16.3 - fix(rust-nat)
 defer TCP bridge startup until handshake completion and buffer partial NAT socket writes
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@push.rocks/smartvpn",
-  "version": "1.16.3",
+  "version": "1.16.5",
   "private": false,
   "description": "A VPN solution with TypeScript control plane and Rust data plane daemon",
   "type": "module",

rust/src/server.rs

@@ -372,6 +372,28 @@ impl VpnServer {
         }
 
         info!("VPN server started (transport: {})", transport_mode);
+
+        // Register pre-loaded clients (from config.clients) as WG peers.
+        // The WG listener only starts with config.wg_peers; clients loaded into the
+        // registry need to be dynamically added so WG handshakes work.
+        if self.wg_command_tx.is_some() {
+            let registry = state.client_registry.read().await;
+            for entry in registry.list() {
+                if let (Some(ref wg_key), Some(ref ip_str)) = (&entry.wg_public_key, &entry.assigned_ip) {
+                    let peer_config = crate::wireguard::WgPeerConfig {
+                        public_key: wg_key.clone(),
+                        preshared_key: None,
+                        allowed_ips: vec![format!("{}/32", ip_str)],
+                        endpoint: None,
+                        persistent_keepalive: Some(25),
+                    };
+                    if let Err(e) = self.add_wg_peer(peer_config).await {
+                        warn!("Failed to register pre-loaded WG peer for {}: {}", entry.client_id, e);
+                    }
+                }
+            }
+        }
+
         Ok(())
     }
 

rust/src/userspace_nat.rs

@@ -17,6 +17,10 @@ use crate::acl;
 use crate::server::{DestinationPolicyConfig, ServerState};
 use crate::tunnel;
 
+/// Maximum size of per-session pending send buffer (512KB = 8x socket buffer).
+/// Sessions exceeding this are aborted — the client cannot keep up.
+const TCP_PENDING_SEND_MAX: usize = 512 * 1024;
+
 // ============================================================================
 // Virtual IP device for smoltcp
 // ============================================================================
@@ -121,7 +125,8 @@ struct SessionKey {
 
 struct TcpSession {
     smoltcp_handle: SocketHandle,
-    bridge_data_tx: mpsc::Sender<Vec<u8>>,
+    /// Channel to send data to the bridge task. None until bridge starts.
+    bridge_data_tx: Option<mpsc::Sender<Vec<u8>>>,
     #[allow(dead_code)]
     client_ip: Ipv4Addr,
     /// Bridge task has been spawned (deferred until handshake completes)
@@ -132,6 +137,8 @@ struct TcpSession {
     pending_send: Vec<u8>,
     /// Session is closing (FIN in progress), don't accept new SYNs
     closing: bool,
+    /// Last time data flowed through this session (for idle timeout)
+    last_activity: tokio::time::Instant,
 }
 
 struct UdpSession {
@@ -385,17 +392,15 @@ impl NatEngine {
 
         let handle = self.sockets.add(socket);
 
-        // Channel for sending data from NAT engine to bridge task
-        let (data_tx, _data_rx) = mpsc::channel::<Vec<u8>>(256);
-
         let session = TcpSession {
             smoltcp_handle: handle,
-            bridge_data_tx: data_tx,
+            bridge_data_tx: None,
             client_ip: key.src_ip,
             bridge_started: false,
             connect_addr,
             pending_send: Vec::new(),
             closing: false,
+            last_activity: tokio::time::Instant::now(),
         };
         self.tcp_sessions.insert(key.clone(), session);
 
@@ -470,9 +475,8 @@ impl NatEngine {
                 let socket = self.sockets.get_mut::<tcp::Socket>(session.smoltcp_handle);
                 if socket.is_active() {
                     session.bridge_started = true;
-                    // Recreate the data channel — the old receiver was dropped
                     let (data_tx, data_rx) = mpsc::channel::<Vec<u8>>(256);
-                    session.bridge_data_tx = data_tx;
+                    session.bridge_data_tx = Some(data_tx);
                     let btx = bridge_tx_clone.clone();
                     let k = key.clone();
                     let addr = session.connect_addr;
@@ -503,15 +507,29 @@ impl NatEngine {
 
         // Bridge: read data from smoltcp TCP sockets → send to bridge tasks
         let mut closed_tcp: Vec<SessionKey> = Vec::new();
-        let mut tcp_outbound: Vec<(mpsc::Sender<Vec<u8>>, Vec<u8>)> = Vec::new();
+        let mut active_tcp: Vec<SessionKey> = Vec::new();
         for (key, session) in &self.tcp_sessions {
             let socket = self.sockets.get_mut::<tcp::Socket>(session.smoltcp_handle);
             if session.bridge_started && socket.can_recv() {
-                let sender = session.bridge_data_tx.clone();
-                let _ = socket.recv(|data| {
-                    tcp_outbound.push((sender.clone(), data.to_vec()));
-                    (data.len(), ())
-                });
+                if let Some(ref sender) = session.bridge_data_tx {
+                    // Reserve channel slot BEFORE consuming from smoltcp.
+                    // If the channel is full, we don't consume — smoltcp's RX buffer
+                    // fills up, it stops advertising TCP window space, and the VPN
+                    // client's TCP stack backs off. Proper end-to-end backpressure.
+                    match sender.try_reserve() {
+                        Ok(permit) => {
+                            let _ = socket.recv(|data| {
+                                permit.send(data.to_vec());
+                                (data.len(), ())
+                            });
+                            active_tcp.push(key.clone());
+                        }
+                        Err(_) => {
+                            debug!("NAT: bridge channel full for {}:{} -> {}:{}, applying backpressure",
+                                key.src_ip, key.src_port, key.dst_ip, key.dst_port);
+                        }
+                    }
+                }
             }
             // Detect closed connections
             if !socket.is_open() && !socket.is_listening() {
@@ -519,10 +537,11 @@ impl NatEngine {
             }
         }
 
-        // Send TCP data to bridge tasks (outside borrow of self.tcp_sessions)
-        for (sender, data) in tcp_outbound {
-            if sender.try_send(data).is_err() {
-                debug!("NAT: bridge channel full, TCP data dropped");
+        // Update last_activity for sessions that had data flow
+        let now = tokio::time::Instant::now();
+        for key in active_tcp {
+            if let Some(session) = self.tcp_sessions.get_mut(&key) {
+                session.last_activity = now;
             }
         }
 
@@ -561,39 +580,31 @@ impl NatEngine {
         match msg {
             BridgeMessage::TcpData { key, data } => {
                 if let Some(session) = self.tcp_sessions.get_mut(&key) {
+                    session.last_activity = tokio::time::Instant::now();
+                    // Append to pending buffer, then flush as much as possible
+                    session.pending_send.extend_from_slice(&data);
                     let socket =
                         self.sockets.get_mut::<tcp::Socket>(session.smoltcp_handle);
-                    if socket.can_send() {
-                        // Try to write directly first
-                        let all_data = if session.pending_send.is_empty() {
-                            &data
-                        } else {
-                            session.pending_send.extend_from_slice(&data);
-                            &session.pending_send.clone()
-                        };
-                        match socket.send_slice(all_data) {
-                            Ok(written) if written < all_data.len() => {
-                                // Partial write — buffer the rest
-                                if session.pending_send.is_empty() {
-                                    session.pending_send = data[written..].to_vec();
-                                } else {
-                                    session.pending_send.drain(..written);
-                                }
-                            }
-                            Ok(_) => {
-                                // Full write — clear any pending data
-                                session.pending_send.clear();
-                            }
-                            Err(_) => {
-                                // Write failed — buffer everything
-                                if session.pending_send.is_empty() {
-                                    session.pending_send = data;
-                                }
+                    if socket.can_send() && !session.pending_send.is_empty() {
+                        match socket.send_slice(&session.pending_send) {
+                            Ok(written) if written > 0 => {
+                                session.pending_send.drain(..written);
                             }
+                            _ => {}
                         }
-                    } else {
-                        // Can't send yet — buffer for later
-                        session.pending_send.extend_from_slice(&data);
+                    }
+                    // Cap check — abort session if client can't keep up
+                    if session.pending_send.len() > TCP_PENDING_SEND_MAX {
+                        warn!(
+                            "NAT: TCP session {}:{} -> {}:{} pending buffer exceeded {}KB, aborting",
+                            key.src_ip, key.src_port, key.dst_ip, key.dst_port,
+                            TCP_PENDING_SEND_MAX / 1024
+                        );
+                        let socket =
+                            self.sockets.get_mut::<tcp::Socket>(session.smoltcp_handle);
+                        socket.abort();
+                        session.pending_send.clear();
+                        session.closing = true;
                     }
                 }
             }
@@ -644,6 +655,29 @@ impl NatEngine {
         }
     }
 
+    fn cleanup_idle_tcp_sessions(&mut self) {
+        let timeout = Duration::from_secs(300); // 5 minutes
+        let now = tokio::time::Instant::now();
+        let expired: Vec<SessionKey> = self
+            .tcp_sessions
+            .iter()
+            .filter(|(_, s)| now.duration_since(s.last_activity) > timeout)
+            .map(|(k, _)| k.clone())
+            .collect();
+
+        for key in expired {
+            if let Some(session) = self.tcp_sessions.remove(&key) {
+                let socket = self.sockets.get_mut::<tcp::Socket>(session.smoltcp_handle);
+                socket.abort();
+                self.sockets.remove(session.smoltcp_handle);
+                warn!(
+                    "NAT: TCP session timed out {}:{} -> {}:{}",
+                    key.src_ip, key.src_port, key.dst_ip, key.dst_port
+                );
+            }
+        }
+    }
+
     /// Main async event loop for the NAT engine.
     pub async fn run(
         mut self,
@@ -651,9 +685,13 @@ impl NatEngine {
         mut shutdown_rx: mpsc::Receiver<()>,
     ) -> Result<()> {
         info!("Userspace NAT engine started");
-        let mut timer = tokio::time::interval(Duration::from_millis(50));
+        let default_poll_delay = Duration::from_millis(50);
         let mut cleanup_timer = tokio::time::interval(Duration::from_secs(10));
 
+        // Dynamic poll timer — reset after each event using smoltcp's poll_delay()
+        let poll_sleep = tokio::time::sleep(default_poll_delay);
+        tokio::pin!(poll_sleep);
+
         loop {
             tokio::select! {
                 Some(packet) = packet_rx.recv() => {
@@ -664,18 +702,26 @@ impl NatEngine {
                     self.handle_bridge_message(msg);
                     self.process().await;
                 }
-                _ = timer.tick() => {
+                () = &mut poll_sleep => {
                     // Periodic poll for smoltcp maintenance (TCP retransmit, etc.)
                     self.process().await;
                 }
                 _ = cleanup_timer.tick() => {
                     self.cleanup_idle_udp_sessions();
+                    self.cleanup_idle_tcp_sessions();
                 }
                 _ = shutdown_rx.recv() => {
                     info!("Userspace NAT engine shutting down");
                     break;
                 }
             }
+
+            // Reset poll delay based on smoltcp's actual timer needs
+            let now = self.smoltcp_now();
+            let delay = self.iface.poll_delay(now, &self.sockets)
+                .map(|d| Duration::from_millis(d.total_millis()))
+                .unwrap_or(default_poll_delay);
+            poll_sleep.as_mut().reset(tokio::time::Instant::now() + delay);
         }
 
         Ok(())

ts/00_commitinfo_data.ts

@@ -3,6 +3,6 @@
  */
 export const commitinfo = {
   name: '@push.rocks/smartvpn',
-  version: '1.16.3',
+  version: '1.16.5',
   description: 'A VPN solution with TypeScript control plane and Rust data plane daemon'
 }