chore: release v0.4.4 on master

A Scheduled Task stores an absolute path. Checking that a task named
"OSTP TUN Helper" exists said nothing about whether its <Command> still points
at the helper we are about to run, and the paths do drift: a dev build
registers target\debug\ostp-tun-helper.exe, an installer registers Program
Files, and moving or reinstalling the app leaves the old path behind.

That failed silently in the worst way. schtasks /Run reports success for
merely ACCEPTING the launch request — a task whose exe no longer exists fails
afterwards, out of band, with nothing returned to us. So launch_as_admin
returned Ok, and the caller then sat in its 60-second connect loop before
reporting "Timeout connecting to helper." On every connect, permanently, with
no way out except deleting the task by hand.

The check now reads the registered <Command> back and compares it to the exe,
re-registering through the existing /F overwrite when they differ: one consent
prompt, once, instead of a permanent silent breakage.

The path is read via /Query /XML rather than /FO LIST /V because the list
format's field labels are localized — "Task To Run" is "Задача для запуска" on
a Russian Windows — while XML tag names are not. schtasks emits UTF-16LE with
a BOM there, which is decoded explicitly, with UTF-8 tolerated as a fallback.
Both paths are canonicalized before comparison so casing, `..` and 8.3 short
names do not read as a mismatch; a path that cannot be canonicalized no longer
exists, which is itself grounds to re-register.

.github/workflows/release.yml

@@ -477,12 +477,19 @@ jobs:
         working-directory: ostp-gui
         run: |
           npm install
+          # TUN mode shells out to this helper, elevated via pkexec. Only the
+          # Windows job used to build it, so the Linux package shipped without
+          # it and TUN could never start.
+          cargo build -p ostp-tun-helper --release --target ${{ matrix.target }} --manifest-path ../Cargo.toml
           npx tauri build --no-bundle --target ${{ matrix.target }}
 
       - name: Package Portable Tarball
         run: |
+          set -euo pipefail
           mkdir ostp-linux-gui-${{ matrix.arch }}
           cp ostp-gui/src-tauri/target/${{ matrix.target }}/release/ostp-gui ostp-linux-gui-${{ matrix.arch }}/
+          # The GUI looks for the helper next to its own executable first.
+          cp target/${{ matrix.target }}/release/ostp-tun-helper ostp-linux-gui-${{ matrix.arch }}/
           tar -czf ostp-linux-gui-${{ matrix.arch }}.tar.gz ostp-linux-gui-${{ matrix.arch }}
 
       - name: Upload to GitHub Release

.release-state.json

@@ -1,6 +1,6 @@
 {
-  "target_version": "0.4.2",
+  "target_version": "0.4.4",
   "branch": "master",
   "alpha_iteration": 0,
-  "beta_iteration": 5
+  "beta_iteration": 0
 }

Cargo.lock

@@ -1386,7 +1386,7 @@ checksum = "c08d65885ee38876c4f86fa503fb49d7b507c2b62552df7c70b2fce627e06381"
 
 [[package]]
 name = "ostp"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "base64",
@@ -1409,7 +1409,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-client"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "base64",
@@ -1440,7 +1440,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-core"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "bytes",
@@ -1474,7 +1474,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-server"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "axum",
@@ -1507,7 +1507,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-tun"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "libc",
@@ -1519,7 +1519,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-tun-helper"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "chrono",

Cargo.toml

@@ -12,7 +12,7 @@ resolver = "2"
 [workspace.package]
 edition = "2021"
 license = "AGPL-3.0"
-version = "0.4.2"
+version = "0.4.4"
 
 [workspace.dependencies]
 anyhow = "1.0"

docs/relay-config-example.json

@@ -20,9 +20,15 @@
   // Адрес следующего узла в цепочке — UDP
   "upstream_udp": "TARGET_SERVER_IP:50000",
 
-  // URL API конечного (целевого) сервера для синхронизации access_keys
-  // Должен быть доступен с этого relay-сервера (можно через SSH-туннель)
-  "upstream_api_url": "http://TARGET_SERVER_IP:9090",
+  // URL API конечного (целевого) сервера для синхронизации access_keys.
+  // Должен быть доступен с этого relay-сервера (можно через SSH-туннель).
+  //
+  // ВАЖНО: URL обязан включать секретный путь панели (api.webpath целевого
+  // сервера). Management API смонтирован ВНУТРИ этого пути — именно он скрывает
+  // панель от сканеров, — поэтому голый host:port попадает в несуществующий
+  // маршрут, и синхронизация падает с 404 ещё до проверки токена.
+  // Это тот же адрес, по которому вы открываете веб-панель.
+  "upstream_api_url": "http://TARGET_SERVER_IP:9090/TARGET_SERVER_WEBPATH",
 
   // Bearer-токен для доступа к API целевого сервера
   // Должен совпадать с api.token в конфиге target-сервера

ostp-client/src/bridge.rs

@@ -23,6 +23,12 @@ use crate::tunnel::{ProxyEvent, ProxyToClientMsg};
 /// candidate address is tried.
 const UOT_CONNECT_TIMEOUT: Duration = Duration::from_secs(4);
 
+/// How long to keep retrying a resume-triggered reconnect before handing the
+/// problem back to the ordinary stall path. That path is what releases the
+/// system proxy, so this is really a bound on how long the machine may be left
+/// with no working internet at all after waking.
+const RESUME_RECONNECT_GIVE_UP: Duration = Duration::from_secs(45);
+
 static SOCKET_PROTECTOR: std::sync::OnceLock<Box<dyn Fn(i32) -> bool + Send + Sync>> = std::sync::OnceLock::new();
 
 pub fn set_socket_protector<F>(f: F)
@@ -137,6 +143,21 @@ pub struct Bridge {
     last_rtt_ms: f64,
     last_sample_at: Instant,
     last_valid_recv: Instant,
+    /// Set when a suspend/resume is detected, cleared once a reconnect actually
+    /// succeeds. Waking is precisely when the network is least likely to be
+    /// ready — Wi-Fi has not reassociated yet — so a single attempt fired
+    /// milliseconds after resume usually fails, and a one-shot forced reconnect
+    /// then fell back to the ordinary 25s stall heuristic. That heuristic keys
+    /// off a monotonic clock which does not advance while the machine is
+    /// asleep, so it could take a further 25s of real uptime to fire, or not
+    /// fire at all. Retrying until success removes the dependency on either.
+    forced_reconnect_pending: bool,
+    last_forced_reconnect_try: Instant,
+    /// Wall-clock start of the current resume-reconnect campaign, used to bound
+    /// it. Wall clock rather than Instant because the monotonic clock does not
+    /// advance across suspend on Windows, so it cannot measure anything that
+    /// begins at wake.
+    forced_reconnect_started: Option<SystemTime>,
 }
 
 impl Bridge {
@@ -173,6 +194,9 @@ impl Bridge {
             last_rtt_ms: 0.0,
             last_sample_at: Instant::now(),
             last_valid_recv: Instant::now(),
+            forced_reconnect_pending: false,
+            last_forced_reconnect_try: Instant::now(),
+            forced_reconnect_started: None,
         })
     }
 
@@ -255,7 +279,64 @@ impl Bridge {
                         let _ = tx.send(UiEvent::Log(format!(
                             "Resumed after ~{}s suspend — forcing clean reconnect", wall_gap.as_secs()
                         ))).await;
+                        self.forced_reconnect_pending = true;
+                        self.forced_reconnect_started = Some(SystemTime::now());
+                        self.last_forced_reconnect_try = Instant::now() - Duration::from_secs(60);
+                    }
+
+                    // Give up if resume reconnects keep failing. Retrying forever
+                    // looks harmless but is not: the system proxy stays pointed at
+                    // our local listener the whole time, so the machine has NO
+                    // working internet — not merely no tunnel — while the UI sits
+                    // on "connecting". Handing the retry to the ordinary keepalive
+                    // path restores the proxy through its hard-timeout branch,
+                    // which force=true deliberately skips.
+                    //
+                    // Measured on the wall clock: Instant does not advance across
+                    // suspend on Windows (QPC stops), so a monotonic deadline can
+                    // not bound anything that starts at wake.
+                    if self.forced_reconnect_pending {
+                        let pending_for = self
+                            .forced_reconnect_started
+                            .and_then(|t| t.elapsed().ok())
+                            .unwrap_or_default();
+                        if pending_for > RESUME_RECONNECT_GIVE_UP {
+                            self.forced_reconnect_pending = false;
+                            self.forced_reconnect_started = None;
+                            let _ = tx.send(UiEvent::Log(format!(
+                                "Reconnect after suspend failed for {}s — releasing the system \
+                                 proxy so normal traffic works; will keep retrying in the \
+                                 background",
+                                pending_for.as_secs()
+                            ))).await;
+                            // Make the ordinary stall path fire on the next
+                            // keepalive tick: it is the one that tears the proxy
+                            // back down (or, with kill switch on, deliberately
+                            // keeps blocking).
+                            self.last_valid_recv = Instant::now()
+                                .checked_sub(Duration::from_secs(3600))
+                                .unwrap_or_else(Instant::now);
+                        }
+                    }
+
+                    // Keep retrying a resume-triggered reconnect until one lands.
+                    // The first attempt fires within half a second of waking, when
+                    // the NIC is typically still reassociating, so treating it as
+                    // one-shot left the tunnel dead until some other timer noticed.
+                    if self.running
+                        && self.forced_reconnect_pending
+                        && self.last_forced_reconnect_try.elapsed() >= Duration::from_secs(3)
+                    {
+                        self.last_forced_reconnect_try = Instant::now();
                         self.handle_keepalive(true, &mut sessions_opt, &mut udp_rx_opt, &mut proxy_guard, &mut stream_map, &tx, &proxy_tx, &mut proxy_rx).await;
+                        // handle_keepalive refreshes last_valid_recv only when a
+                        // session was actually established, so this is a real
+                        // success check rather than "we tried".
+                        if self.last_valid_recv.elapsed() < Duration::from_secs(3) {
+                            self.forced_reconnect_pending = false;
+                            self.forced_reconnect_started = None;
+                            let _ = tx.send(UiEvent::Log("Reconnected after suspend".into())).await;
+                        }
                     }
                     if self.running {
                         self.emit_metrics(&tx).await;
@@ -272,7 +353,20 @@ impl Bridge {
                     }
                 }
                 proxy_ev = proxy_rx.recv(), if self.running && sessions_opt.as_ref().map(|s| {
-                    s.iter().any(|ses| ses.machine.in_flight_count() < ses.machine.cwnd_packets().clamp(16, 16384))
+                    // Upper bound matches MAX_CWND_PACKETS in ostp-core's congestion
+                    // controller. The old 16384 ceiling let ~20 MB sit in flight,
+                    // which on a mobile uplink is minutes of buffered queue rather
+                    // than throughput — the app kept handing over data long after
+                    // the path had stopped draining it.
+                    // Two independent gates. cwnd bounds how much may be in
+                    // flight; pacing bounds how FAST it is released. Without the
+                    // second, a full window goes out back-to-back and lands in
+                    // the bottleneck's buffer as standing queue rather than
+                    // throughput — the thing that produced multi-second RTT.
+                    s.iter().any(|ses| {
+                        ses.machine.in_flight_count() < ses.machine.cwnd_packets().clamp(16, 1024)
+                            && ses.machine.can_pace_packet()
+                    })
                 }).unwrap_or(true) => {
                     self.handle_proxy_event(proxy_ev, &mut sessions_opt, &mut stream_map, &tx, &proxy_tx).await;
                 }
@@ -926,7 +1020,21 @@ impl Bridge {
             Ok(addrs) => addrs.collect(),
             Err(e) => return Err(anyhow::anyhow!("failed to resolve server address {}: {}", self.server_addr, e)),
         };
-        resolved_addrs.sort_by_key(|addr| if addr.is_ipv6() { 0 } else { 1 });
+        // IPv4 first. Addresses are tried strictly in order, each burning its
+        // full retry budget before the next is touched, so this ordering decides
+        // how long a bad family stalls the whole connect. Mobile carriers
+        // routinely hand out IPv6 with no working route and BLACKHOLE it rather
+        // than rejecting, so every IPv6 candidate costs the full timeout budget
+        // — with several AAAA records the working IPv4 address was not reached
+        // for tens of seconds. (The same ordering bug was already fixed on the
+        // server's outbound path and in the UoT connect.)
+        resolved_addrs.sort_by_key(|addr| if addr.is_ipv6() { 1 } else { 0 });
+
+        // NAT64 is a fallback for IPv6-only networks. Retrying it per failing
+        // address multiplied an already-long connect: each attempt re-runs a DNS
+        // lookup and another full round of handshake retries, for a path that
+        // either works for the whole network or for none of it.
+        let mut nat64_attempted = false;
 
         let mut last_err = anyhow::anyhow!("no IP addresses resolved for {}", self.server_addr);
 
@@ -939,7 +1047,8 @@ impl Bridge {
             let socket = match self.try_connect_transport(target_ip, port).await {
                 Ok(sock) => sock,
                 Err(e) => {
-                    if let std::net::IpAddr::V4(ipv4) = target_ip {
+                    if let (std::net::IpAddr::V4(ipv4), false) = (target_ip, nat64_attempted) {
+                        nat64_attempted = true;
                         tx.send(UiEvent::Log(format!("Direct IPv4 connection failed: {}. Trying NAT64 fallback...", e))).await.ok();
                         let nat64_ipv6 = synthesize_nat64(ipv4).await;
                         match self.try_connect_transport(std::net::IpAddr::V6(nat64_ipv6), port).await {
@@ -1020,7 +1129,8 @@ impl Bridge {
             let (final_socket, size) = if success {
                 (socket, size)
             } else {
-                if let std::net::IpAddr::V4(ipv4) = target_ip {
+                if let (std::net::IpAddr::V4(ipv4), false) = (target_ip, nat64_attempted) {
+                    nat64_attempted = true;
                     tx.send(UiEvent::Log("Direct IPv4 handshake timed out. Trying NAT64 fallback...".to_string())).await.ok();
                     let nat64_ipv6 = synthesize_nat64(ipv4).await;
                     match self.try_connect_transport(std::net::IpAddr::V6(nat64_ipv6), port).await {
@@ -1264,8 +1374,19 @@ fn next_profile(current: TrafficProfile) -> TrafficProfile {
 }
 
 async fn synthesize_nat64(ip: std::net::Ipv4Addr) -> std::net::Ipv6Addr {
+    // Well-known prefix (RFC 6052), used if discovery doesn't answer in time.
     let mut prefix = [0x00, 0x64, 0xff, 0x9b, 0, 0, 0, 0, 0, 0, 0, 0];
-    if let Ok(addrs) = tokio::net::lookup_host("ipv4only.arpa:80").await {
+    // Bound the discovery lookup. This runs on exactly the networks that are
+    // already misbehaving, where the resolver can hang for tens of seconds
+    // before giving up — unbounded, it was a large part of why connecting over
+    // a broken mobile network took minutes. Falling back to the well-known
+    // prefix is strictly better than waiting.
+    let discovery = tokio::time::timeout(
+        Duration::from_secs(2),
+        tokio::net::lookup_host("ipv4only.arpa:80"),
+    )
+    .await;
+    if let Ok(Ok(addrs)) = discovery {
         for addr in addrs {
             if let std::net::SocketAddr::V6(v6) = addr {
                 let octets = v6.ip().octets();

ostp-client/src/config.rs

@@ -418,19 +418,22 @@ pub struct RelayServerConfig {
     pub upstream_tcp: String,
     /// Upstream address for UDP traffic
     pub upstream_udp: String,
-    /// Target server's API URL, for key sync
+    // ── Deprecated ──────────────────────────────────────────────────────────
+    // The relay used to authenticate clients itself and pulled the access-key
+    // list from the target server's management API to do it. It no longer does:
+    // sessions are authenticated end-to-end by the target server, and a relay
+    // that re-checks credentials only adds a weaker second gate plus a copy of
+    // the key list on a machine that does not need one. These are kept solely
+    // so existing relay configs still parse; they are ignored.
+    #[serde(default)]
     pub upstream_api_url: String,
-    /// Bearer token for the target server's API
     #[serde(default)]
     pub upstream_api_token: String,
-    /// Key sync interval in seconds (default 30)
-    #[serde(default = "default_sync_interval")]
+    #[serde(default)]
     pub sync_interval_secs: u64,
     pub debug: Option<bool>,
 }
 
-fn default_sync_interval() -> u64 { 30 }
-
 /// Supports both a single string "0.0.0.0:50000" and an array
 /// ["0.0.0.0:50000", "[::]:50000"].
 #[derive(Debug, Deserialize, Serialize, Clone)]

ostp-core/src/congestion.rs

@@ -39,6 +39,9 @@ pub struct CongestionController {
     loss_count: u32,
     /// Pacing rate: bytes per second
     pacing_rate: u64,
+    /// Token-bucket allowance for pacing, in bytes.
+    pacing_tokens: f64,
+    pacing_last_refill: Instant,
     /// MTU estimate (used for cwnd → packet count conversion)
     mtu: u64,
     /// Min RTT expiry: re-probe after 10 seconds
@@ -65,6 +68,20 @@ const MIN_CWND_PACKETS: u64 = 2;
 /// Min RTT expiry window (after which we re-probe)
 const MIN_RTT_EXPIRY: Duration = Duration::from_secs(10);
 /// Minimum RTO (RFC 6298: 1s in TCP; we use 50ms since we own the protocol)
+/// Absolute ceiling on the congestion window, in packets. At a ~1200-byte MTU
+/// this is roughly 1.2 MB in flight — already far above the bandwidth-delay
+/// product of any link this protocol realistically runs over, so anything
+/// beyond it is standing queue, not throughput. The client previously allowed
+/// up to 16384 packets (~20 MB), which on a mobile uplink is minutes of buffer.
+const MAX_CWND_PACKETS: u64 = 1024;
+/// SRTT/min_rtt ratio at which slow start stops. Doubling is what fills a deep
+/// buffer fastest, so growth must end when the queue starts building rather
+/// than waiting for a loss that a deep buffer may never produce.
+const RTT_INFLATION_EXIT_SLOW_START: f64 = 2.0;
+/// SRTT/min_rtt ratio treated as a standing queue that must be actively drained.
+const RTT_INFLATION_BACKOFF: f64 = 4.0;
+/// How much pacing allowance may accumulate, expressed as time-at-rate.
+const PACING_BURST: Duration = Duration::from_millis(10);
 const RTO_MIN: Duration = Duration::from_millis(50);
 /// Maximum RTO
 const RTO_MAX: Duration = Duration::from_secs(16);
@@ -113,9 +130,50 @@ impl CongestionController {
             min_rtt_stamp: now,
             slow_start_losses: 0,
             slow_start_loss_window_start: now,
+            pacing_tokens: (INITIAL_CWND_PACKETS * mtu) as f64,
+            pacing_last_refill: now,
         }
     }
 
+    /// Bytes of pacing allowance available right now, without consuming any.
+    ///
+    /// Read-only so the send path can use it as an admission check before it
+    /// commits to building a datagram.
+    pub fn pacing_available(&self) -> f64 {
+        let elapsed = self.pacing_last_refill.elapsed().as_secs_f64();
+        (self.pacing_tokens + elapsed * self.pacing_rate as f64).min(self.pacing_burst())
+    }
+
+    /// Whether at least one full-size packet may be released right now.
+    pub fn can_pace_packet(&self) -> bool {
+        self.pacing_available() >= self.mtu as f64
+    }
+
+    /// Ceiling on accumulated allowance.
+    ///
+    /// Pacing intervals here are fractions of a millisecond, so releasing
+    /// strictly one packet at a time would need a sub-millisecond timer per
+    /// packet. Instead we allow a short burst — the same trade every real
+    /// pacing implementation makes — sized so the loop's existing ~10ms wakeups
+    /// can still saturate the configured rate, with a small floor so a
+    /// cold/low estimate can never wedge sending entirely.
+    fn pacing_burst(&self) -> f64 {
+        let by_rate = self.pacing_rate as f64 * PACING_BURST.as_secs_f64();
+        by_rate.max((self.mtu * 4) as f64)
+    }
+
+    /// Refill from elapsed time and deduct `bytes`. Called on the real send
+    /// path; allowance is permitted to go negative so an oversized packet still
+    /// pays for itself rather than being released for free.
+    fn consume_pacing(&mut self, bytes: u64) {
+        let now = Instant::now();
+        let elapsed = now.duration_since(self.pacing_last_refill).as_secs_f64();
+        self.pacing_last_refill = now;
+        self.pacing_tokens =
+            (self.pacing_tokens + elapsed * self.pacing_rate as f64).min(self.pacing_burst())
+                - bytes as f64;
+    }
+
     /// Returns the current congestion window in bytes.
     pub fn cwnd(&self) -> u64 {
         self.cwnd
@@ -167,6 +225,11 @@ impl CongestionController {
     /// Record that we sent `bytes` of data.
     pub fn on_send(&mut self, bytes: u64) {
         self.bytes_in_flight = self.bytes_in_flight.saturating_add(bytes);
+        // Charge the pacing bucket here rather than at the admission check, so
+        // every byte that actually reaches the wire is paid for exactly once —
+        // including retransmits, which are precisely what must not be allowed
+        // to bypass the rate limit and pile into an already-full queue.
+        self.consume_pacing(bytes);
     }
 
     /// Record that `bytes` were acknowledged but WITHOUT a usable RTT sample
@@ -198,9 +261,46 @@ impl CongestionController {
 
     /// Congestion-window growth shared by both ACK paths (slow start / probe).
     fn grow_window(&mut self, bytes: u64) {
-        // State machine
+        // ── Delay-based congestion signal ────────────────────────────────────
+        // A loss-only controller is blind on a deeply-buffered path, and mobile
+        // carrier buffers are very deep: they absorb a burst instead of dropping
+        // it, so no loss is ever signalled and cwnd keeps growing. The queue —
+        // not the link — is what grows, and the standing delay it adds shows up
+        // as RTT inflating far above the path's floor. Left unchecked this is a
+        // positive feedback loop: bigger queue -> larger RTT samples -> larger
+        // SRTT -> larger RTO -> retransmits pile on -> bigger queue, which is
+        // how a session ends up reporting multi-second (even multi-minute) RTT
+        // and stalls video until the buffer finally drains or the user
+        // reconnects. Treat sustained RTT inflation as congestion in its own
+        // right, exactly as it is.
+        let inflation = if self.rtt_initialized && !self.min_rtt.is_zero() {
+            self.srtt.as_secs_f64() / self.min_rtt.as_secs_f64()
+        } else {
+            1.0
+        };
+
+        if inflation >= RTT_INFLATION_BACKOFF {
+            // Standing queue is severe — actively drain it.
+            self.cwnd = (self.cwnd / 2).max(MIN_CWND_PACKETS * self.mtu);
+            self.ssthresh = self.cwnd;
+            self.phase = Phase::ProbeBandwidth;
+            tracing::debug!(cwnd = self.cwnd, inflation, "congestion: draining standing queue");
+            self.clamp_cwnd();
+            return;
+        }
+
         match self.phase {
             Phase::SlowStart => {
+                // Exponential doubling is what fills a deep buffer fastest, so
+                // leave slow start as soon as the queue starts to build rather
+                // than waiting for the loss that may never come.
+                if inflation >= RTT_INFLATION_EXIT_SLOW_START {
+                    self.ssthresh = self.cwnd;
+                    self.phase = Phase::ProbeBandwidth;
+                    tracing::debug!(cwnd = self.cwnd, inflation, "congestion: RTT inflation ended slow start");
+                    self.clamp_cwnd();
+                    return;
+                }
                 // Exponential growth: increase cwnd by acked bytes (doubles per RTT)
                 self.cwnd = self.cwnd.saturating_add(bytes);
                 if self.cwnd >= self.ssthresh {
@@ -213,6 +313,21 @@ impl CongestionController {
                 self.cwnd = self.cwnd.saturating_add(bytes * self.mtu / self.cwnd.max(1));
             }
         }
+
+        self.clamp_cwnd();
+    }
+
+    /// Hard ceiling on the congestion window.
+    ///
+    /// Independent of any estimate: no real path this protocol runs over has a
+    /// bandwidth-delay product anywhere near this, so a window above it is
+    /// buffered queue rather than data in transit. Without it, slow start on a
+    /// buffer that never drops could grow the window into the tens of megabytes.
+    fn clamp_cwnd(&mut self) {
+        let ceiling = MAX_CWND_PACKETS.saturating_mul(self.mtu);
+        if self.cwnd > ceiling {
+            self.cwnd = ceiling;
+        }
     }
 
     /// Record a loss event.
@@ -332,6 +447,94 @@ mod tests {
         assert!(cc.cwnd() < initial);
     }
 
+    /// The bufferbloat case: a deep buffer absorbs everything, so NOTHING is
+    /// ever lost, but the standing queue inflates RTT. A loss-only controller
+    /// grows cwnd forever here — which is how a session ends up reporting
+    /// multi-second RTT and stalling video.
+    #[test]
+    fn test_rtt_inflation_halts_growth_without_any_loss() {
+        let mut cc = CongestionController::new(1200);
+
+        // Establish a low path floor; this becomes min_rtt.
+        for _ in 0..4 {
+            cc.on_send(1200);
+            cc.on_ack(1200, Duration::from_millis(20));
+        }
+        let cwnd_before = cc.cwnd();
+
+        // Queue builds: RTT climbs far above the floor, still zero loss.
+        for _ in 0..20 {
+            cc.on_send(1200);
+            cc.on_ack(1200, Duration::from_millis(400));
+        }
+
+        assert!(
+            cc.cwnd() <= cwnd_before,
+            "cwnd kept growing while the queue was inflating RTT ({} -> {})",
+            cwnd_before,
+            cc.cwnd()
+        );
+    }
+
+    /// Pacing must actually bound the release rate: draining the bucket has to
+    /// deny the next packet. Without this the congestion window alone decides,
+    /// and a whole window leaves back-to-back.
+    #[test]
+    fn test_pacing_bucket_denies_once_drained() {
+        let mut cc = CongestionController::new(1200);
+        assert!(cc.can_pace_packet(), "a fresh controller must allow sending");
+
+        // Spend well beyond one burst allowance.
+        let burst_bytes = cc.pacing_available();
+        let mut spent = 0.0;
+        while spent <= burst_bytes + 1200.0 {
+            cc.on_send(1200);
+            spent += 1200.0;
+        }
+
+        assert!(
+            !cc.can_pace_packet(),
+            "pacing allowed unbounded sending: {} bytes still available after spending {}",
+            cc.pacing_available(),
+            spent
+        );
+    }
+
+    /// The allowance must refill over time, or sending would stall permanently
+    /// once the first burst is spent.
+    #[test]
+    fn test_pacing_bucket_refills_over_time() {
+        let mut cc = CongestionController::new(1200);
+        while cc.can_pace_packet() {
+            cc.on_send(1200);
+        }
+        assert!(!cc.can_pace_packet());
+
+        std::thread::sleep(Duration::from_millis(25));
+        assert!(
+            cc.can_pace_packet(),
+            "pacing bucket never refilled; sending would be stuck forever"
+        );
+    }
+
+    /// cwnd must never exceed the absolute ceiling, however long slow start
+    /// runs unopposed — above it the window is buffered queue, not throughput.
+    #[test]
+    fn test_cwnd_never_exceeds_absolute_ceiling() {
+        let mut cc = CongestionController::new(1200);
+        // Constant RTT: no inflation signal, so only the hard cap can stop this.
+        for _ in 0..5000 {
+            cc.on_send(1200);
+            cc.on_ack(1200, Duration::from_millis(30));
+        }
+        assert!(
+            cc.cwnd() <= MAX_CWND_PACKETS * 1200,
+            "cwnd {} exceeded the {}-packet ceiling",
+            cc.cwnd(),
+            MAX_CWND_PACKETS
+        );
+    }
+
     #[test]
     fn test_isolated_slow_start_loss_does_not_exit_slow_start() {
         // A single dropped packet (wireless noise, a brief handover blip) is

ostp-core/src/protocol.rs

@@ -4,6 +4,11 @@ use thiserror::Error;
 use std::collections::{BTreeMap, VecDeque};
 use std::time::{Duration, Instant};
 
+/// Upper bound on a single frame's retransmit timer, after exponential backoff
+/// is applied to the adaptive RTO. Past this the session is dead from the
+/// user's point of view, and waiting longer only delays recovery.
+const MAX_EFFECTIVE_RTO: Duration = Duration::from_secs(8);
+
 use crate::congestion::CongestionController;
 use crate::crypto::{NoiseRole, NoiseSession, SessionCipher};
 use crate::framing::{AdaptivePadder, FrameHeader, FrameKind, FramedPacket, PaddingStrategy};
@@ -183,6 +188,16 @@ impl ProtocolMachine {
         self.cc.cwnd_packets() as usize
     }
 
+    /// Whether the pacing bucket currently allows releasing another packet.
+    ///
+    /// The congestion window bounds how much may be UNACKNOWLEDGED; it says
+    /// nothing about how fast that window is emptied onto the wire. Sending a
+    /// whole window back-to-back is what drives a deep buffer into standing
+    /// queue, so admission is gated on both.
+    pub fn can_pace_packet(&self) -> bool {
+        self.cc.can_pace_packet()
+    }
+
     pub fn on_send(&mut self, bytes: u64) {
         self.cc.on_send(bytes);
     }
@@ -675,8 +690,15 @@ impl ProtocolMachine {
                 break;
             }
 
+            // Exponential backoff, but bounded in absolute terms. base_rto is
+            // itself adaptive and can reach RTO_MAX (16s) on a congested path;
+            // multiplying that by the 64x backoff cap yields a frame that sits
+            // unretransmitted for ~17 MINUTES, long past the point where the
+            // session is simply dead to the user. Cap the product so backoff
+            // stays a backoff rather than an outage.
             let backoff_factor = 1u64 << (frame.retries as u64).min(6);
-            let effective_rto = Duration::from_millis(base_rto_ms.saturating_mul(backoff_factor));
+            let effective_rto = Duration::from_millis(base_rto_ms.saturating_mul(backoff_factor))
+                .min(MAX_EFFECTIVE_RTO);
 
             if now.duration_since(frame.last_sent) >= effective_rto {
                 // Only burn the retry counter and reset the RTO timer when the

ostp-flutter/pubspec.yaml

@@ -16,7 +16,7 @@ publish_to: 'none' # Remove this line if you wish to publish to pub.dev
 # https://developer.apple.com/library/archive/documentation/General/Reference/InfoPlistKeyReference/Articles/CoreFoundationKeys.html
 # In Windows, build-name is used as the major, minor, and patch parts
 # of the product and file versions while build-number is used as the build suffix.
-version: 0.4.2+25
+version: 0.4.4+31
 
 environment:
   sdk: ^3.11.4

ostp-gui/package.json

@@ -1,7 +1,7 @@
 {
   "name": "ostp-gui",
   "private": true,
-  "version": "0.4.2",
+  "version": "0.4.4",
   "type": "module",
   "scripts": {
     "tauri": "tauri",

ostp-gui/src-tauri/Cargo.lock

@@ -2665,7 +2665,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-client"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "base64 0.22.1",
@@ -2696,7 +2696,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-core"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "bytes",
@@ -2713,7 +2713,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-gui"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "json_comments",
@@ -2733,7 +2733,7 @@ dependencies = [
 
 [[package]]
 name = "ostp-tun"
-version = "0.4.2"
+version = "0.4.4"
 dependencies = [
  "anyhow",
  "libc",

ostp-gui/src-tauri/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "ostp-gui"
-version = "0.4.2"
+version = "0.4.4"
 description = "OSTP desktop GUI"
 authors = ["ospab"]
 edition = "2021"

ostp-gui/src-tauri/src/lib.rs

@@ -204,19 +204,34 @@ fn get_wintun_install_path() -> String {
     String::new()
 }
 
+/// A `Command` for a console program, with the console window suppressed.
+///
+/// The GUI is a windowed-subsystem binary, so every console child it spawns
+/// pops up a console window for as long as that child runs. With `reg`,
+/// `tasklist` and `schtasks` all being invoked from here, that surfaced as
+/// windows flashing on screen — worst while polling for the scheduled task,
+/// which could spawn twenty of them in a row.
+#[cfg(target_os = "windows")]
+fn quiet_command(program: &str) -> std::process::Command {
+    use std::os::windows::process::CommandExt;
+    const CREATE_NO_WINDOW: u32 = 0x0800_0000;
+    let mut cmd = std::process::Command::new(program);
+    cmd.creation_flags(CREATE_NO_WINDOW);
+    cmd
+}
+
 /// Sets or removes the app from Windows startup (HKCU\...\Run).
 #[tauri::command]
 fn set_autostart(enable: bool) -> Result<(), String> {
     #[cfg(target_os = "windows")]
     {
-        use std::process::Command;
         let key = r"HKCU\Software\Microsoft\Windows\CurrentVersion\Run";
         let app_name = "OSTP";
         if enable {
             let exe = std::env::current_exe()
                 .map_err(|e| format!("Cannot get exe path: {}", e))?;
             let exe_str = format!("\"{}\"", exe.to_string_lossy());
-            let out = Command::new("reg")
+            let out = quiet_command("reg")
                 .args(["add", key, "/v", app_name, "/t", "REG_SZ", "/d", &exe_str, "/f"])
                 .output()
                 .map_err(|e| format!("reg add failed: {}", e))?;
@@ -224,28 +239,71 @@ fn set_autostart(enable: bool) -> Result<(), String> {
                 return Err(String::from_utf8_lossy(&out.stderr).to_string());
             }
         } else {
-            let _ = Command::new("reg")
+            let _ = quiet_command("reg")
                 .args(["delete", key, "/v", app_name, "/f"])
                 .output();
         }
     }
+    #[cfg(target_os = "linux")]
+    {
+        // XDG autostart: desktop environments launch every .desktop file in
+        // ~/.config/autostart on login. This is the portable equivalent of the
+        // HKCU Run key above and needs no elevation.
+        let path = linux_autostart_path().ok_or("Cannot determine the autostart directory")?;
+        if enable {
+            let exe = std::env::current_exe().map_err(|e| format!("Cannot get exe path: {}", e))?;
+            if let Some(dir) = path.parent() {
+                std::fs::create_dir_all(dir)
+                    .map_err(|e| format!("Cannot create {}: {}", dir.display(), e))?;
+            }
+            let entry = format!(
+                "[Desktop Entry]\n\
+                 Type=Application\n\
+                 Name=OSTP\n\
+                 Exec=\"{}\"\n\
+                 Terminal=false\n\
+                 X-GNOME-Autostart-enabled=true\n",
+                exe.display()
+            );
+            std::fs::write(&path, entry)
+                .map_err(|e| format!("Cannot write {}: {}", path.display(), e))?;
+        } else if path.exists() {
+            std::fs::remove_file(&path)
+                .map_err(|e| format!("Cannot remove {}: {}", path.display(), e))?;
+        }
+    }
     Ok(())
 }
 
+/// Path of the XDG autostart entry, honouring XDG_CONFIG_HOME.
+#[cfg(target_os = "linux")]
+fn linux_autostart_path() -> Option<PathBuf> {
+    let base = std::env::var_os("XDG_CONFIG_HOME")
+        .map(PathBuf::from)
+        .filter(|p| !p.as_os_str().is_empty())
+        .or_else(|| std::env::var_os("HOME").map(|h| PathBuf::from(h).join(".config")))?;
+    Some(base.join("autostart").join("ostp.desktop"))
+}
+
 /// Checks if the app is currently in Windows startup.
 #[tauri::command]
 fn get_autostart() -> bool {
     #[cfg(target_os = "windows")]
     {
-        use std::process::Command;
         let key = r"HKCU\Software\Microsoft\Windows\CurrentVersion\Run";
-        let out = Command::new("reg")
+        let out = quiet_command("reg")
             .args(["query", key, "/v", "OSTP"])
             .output();
         if let Ok(o) = out {
             return o.status.success();
         }
     }
+    #[cfg(target_os = "linux")]
+    {
+        if let Some(path) = linux_autostart_path() {
+            return path.exists();
+        }
+    }
     false
 }
 
@@ -254,8 +312,7 @@ fn get_autostart() -> bool {
 fn list_running_processes() -> Vec<String> {
     #[cfg(target_os = "windows")]
     {
-        use std::process::Command;
-        if let Ok(out) = Command::new("tasklist")
+        if let Ok(out) = quiet_command("tasklist")
             .args(["/FO", "CSV", "/NH"])
             .output()
         {
@@ -625,13 +682,18 @@ async fn start_tun_via_helper(
     raw: &ClientConfigRaw,
     app: tauri::AppHandle,
 ) -> Result<bool, String> {
+    // TUN goes through a privileged helper. Elevation is implemented for
+    // Windows (UAC) and Linux (polkit/pkexec); anywhere else launch_as_admin
+    // reports that plainly rather than letting this fail later as a confusing
+    // missing-file error.
     let port = {
         let listener = std::net::TcpListener::bind("127.0.0.1:0").map_err(|e| format!("Bind error: {}", e))?;
         listener.local_addr().unwrap().port()
     };
 
     let auth_token = rand::random::<u64>().to_string();
-    let helper_exe = find_helper_exe().ok_or_else(|| "ostp-tun-helper.exe not found.".to_string())?;
+    let helper_exe = find_helper_exe()
+        .ok_or_else(|| format!("{HELPER_EXE_NAME} not found next to the app or in target/."))?;
     launch_as_admin(&helper_exe, &auth_token, port).map_err(|e| format!("Failed to launch helper: {}", e))?;
     tokio::time::sleep(std::time::Duration::from_millis(1500)).await;
 
@@ -705,21 +767,32 @@ struct HelperPipeState {
     error_msg: Option<String>,
 }
 
+/// Executable name of the TUN helper for the current platform.
+///
+/// The ".exe" suffix was hardcoded, so on Linux every lookup below searched for
+/// a file that cannot exist and the GUI reported the helper as missing on a
+/// platform where it ships without an extension.
+const HELPER_EXE_NAME: &str = if cfg!(windows) {
+    "ostp-tun-helper.exe"
+} else {
+    "ostp-tun-helper"
+};
+
 fn find_helper_exe() -> Option<PathBuf> {
     if let Ok(exe) = std::env::current_exe() {
         if let Some(dir) = exe.parent() {
             // 1. Release/Production adjacent
-            let candidate = dir.join("ostp-tun-helper.exe");
+            let candidate = dir.join(HELPER_EXE_NAME);
             if candidate.exists() { return Some(candidate); }
-            
+
             // 2. Tauri target directory fallback
             // e.g. from ostp-gui/src-tauri/target/debug/deps/
             let mut parent = dir;
             while let Some(p) = parent.parent() {
                 if p.file_name().map(|n| n == "target").unwrap_or(false) {
-                    let deb = p.join("debug").join("ostp-tun-helper.exe");
+                    let deb = p.join("debug").join(HELPER_EXE_NAME);
                     if deb.exists() { return Some(deb); }
-                    let rel = p.join("release").join("ostp-tun-helper.exe");
+                    let rel = p.join("release").join(HELPER_EXE_NAME);
                     if rel.exists() { return Some(rel); }
                 }
                 parent = p;
@@ -729,13 +802,13 @@ fn find_helper_exe() -> Option<PathBuf> {
     // 3. Current working directory target fallback
     let cwd = std::env::current_dir().unwrap_or_default();
     let candidates = [
-        cwd.join("ostp-tun-helper.exe"),
-        cwd.join("target").join("debug").join("ostp-tun-helper.exe"),
-        cwd.join("target").join("release").join("ostp-tun-helper.exe"),
-        cwd.join("..").join("target").join("debug").join("ostp-tun-helper.exe"),
-        cwd.join("..").join("target").join("release").join("ostp-tun-helper.exe"),
-        cwd.join("..").join("..").join("target").join("debug").join("ostp-tun-helper.exe"),
-        cwd.join("..").join("..").join("target").join("release").join("ostp-tun-helper.exe"),
+        cwd.join(HELPER_EXE_NAME),
+        cwd.join("target").join("debug").join(HELPER_EXE_NAME),
+        cwd.join("target").join("release").join(HELPER_EXE_NAME),
+        cwd.join("..").join("target").join("debug").join(HELPER_EXE_NAME),
+        cwd.join("..").join("target").join("release").join(HELPER_EXE_NAME),
+        cwd.join("..").join("..").join("target").join("debug").join(HELPER_EXE_NAME),
+        cwd.join("..").join("..").join("target").join("release").join(HELPER_EXE_NAME),
     ];
     for path in &candidates {
         if path.exists() { return Some(path.clone()); }
@@ -743,8 +816,272 @@ fn find_helper_exe() -> Option<PathBuf> {
     None
 }
 
+/// Name of the Scheduled Task that runs the helper elevated without a prompt.
+#[cfg(target_os = "windows")]
+const HELPER_TASK_NAME: &str = "OSTP TUN Helper";
+
+/// Fixed path the GUI writes launch parameters to, and the task's command line
+/// reads them from.
+///
+/// A Scheduled Task stores a FIXED command line, so the per-launch port and
+/// token cannot travel as arguments. The file lives under the user's own
+/// LOCALAPPDATA: the helper runs elevated but as the SAME user, so this keeps
+/// the token inside the trust boundary it already had — no other user can read
+/// it, which would not be true of a shared location.
+#[cfg(target_os = "windows")]
+fn helper_args_file() -> PathBuf {
+    let base = std::env::var_os("LOCALAPPDATA")
+        .map(PathBuf::from)
+        .unwrap_or_else(std::env::temp_dir);
+    base.join("OSTP").join("helper-args.json")
+}
+
+/// Minimal XML text escaping for the values interpolated into the task
+/// definition. Paths and usernames are attacker-irrelevant here but can easily
+/// contain `&`, which would otherwise produce invalid XML and a confusing
+/// schtasks parse failure.
+#[cfg(target_os = "windows")]
+fn xml_escape(s: &str) -> String {
+    s.replace('&', "&amp;")
+        .replace('<', "&lt;")
+        .replace('>', "&gt;")
+        .replace('"', "&quot;")
+        .replace('\'', "&apos;")
+}
+
+/// Reverse of [`xml_escape`]. `&amp;` must be undone last or `&amp;lt;` would
+/// come back as `<`.
+#[cfg(target_os = "windows")]
+fn xml_unescape(s: &str) -> String {
+    s.replace("&quot;", "\"")
+        .replace("&apos;", "'")
+        .replace("&lt;", "<")
+        .replace("&gt;", ">")
+        .replace("&amp;", "&")
+}
+
+/// The exe path currently baked into the registered task, if any.
+///
+/// Queried as XML rather than `/FO LIST /V`: the list format's field labels are
+/// localized (on a Russian Windows "Task To Run" is "Задача для запуска"),
+/// whereas XML tag names are fixed. schtasks writes UTF-16LE with a BOM here,
+/// but tolerate UTF-8 in case that ever changes.
+#[cfg(target_os = "windows")]
+fn helper_task_command() -> Option<String> {
+    let out = quiet_command("schtasks")
+        .args(["/Query", "/TN", HELPER_TASK_NAME, "/XML"])
+        .output()
+        .ok()?;
+    if !out.status.success() {
+        return None;
+    }
+
+    let text = if out.stdout.starts_with(&[0xFF, 0xFE]) {
+        let units: Vec<u16> = out.stdout[2..]
+            .chunks_exact(2)
+            .map(|c| u16::from_le_bytes([c[0], c[1]]))
+            .collect();
+        String::from_utf16_lossy(&units)
+    } else {
+        String::from_utf8_lossy(&out.stdout).into_owned()
+    };
+
+    let start = text.find("<Command>")? + "<Command>".len();
+    let end = text[start..].find("</Command>")? + start;
+    Some(xml_unescape(text[start..end].trim()))
+}
+
+/// Whether a task is registered AND still points at the exe we are about to run.
+///
+/// The path matters as much as the name. A task registered by a dev build (or
+/// by an install that has since moved) keeps its original `<Command>`, and
+/// `schtasks /Run` reports success merely for *accepting* the request — a task
+/// whose exe no longer exists fails asynchronously and silently. Trusting the
+/// name alone therefore bought a 60-second "Timeout connecting to helper" on
+/// every single connect, permanently, until the task was deleted by hand.
+/// Re-registering costs one consent prompt and fixes it for good.
+#[cfg(target_os = "windows")]
+fn helper_task_matches(exe: &std::path::Path) -> bool {
+    let Some(registered) = helper_task_command() else {
+        return false;
+    };
+    let registered = registered.trim().trim_matches('"');
+
+    // Canonicalize both sides when possible so `..`, short 8.3 names and
+    // casing differences do not read as a mismatch. A missing file cannot be
+    // canonicalized — which is itself a mismatch worth re-registering over.
+    match (
+        std::fs::canonicalize(registered),
+        std::fs::canonicalize(exe),
+    ) {
+        (Ok(a), Ok(b)) => a == b,
+        _ => registered.eq_ignore_ascii_case(&exe.display().to_string()),
+    }
+}
+
+/// Register the Scheduled Task. This is the ONLY step that needs elevation, and
+/// it happens once per machine; every later tunnel start reuses the task.
+///
+/// RunLevel=HIGHEST makes the task run elevated, and because a task launch is
+/// not an elevation request, Windows shows no consent dialog for it.
+#[cfg(target_os = "windows")]
+fn install_helper_task(exe: &std::path::Path) -> anyhow::Result<()> {
+    let args_file = helper_args_file();
+    if let Some(dir) = args_file.parent() {
+        std::fs::create_dir_all(dir)?;
+    }
+
+    // Register from an XML definition rather than /TR. The command line would
+    // otherwise need the exe path and the args path quoted INSIDE an already
+    // quoted /TR value, escaped again through ShellExecuteW — a notoriously
+    // brittle chain when either path contains a space, which both of these do
+    // by default (Program Files, and usernames with spaces). XML also lets the
+    // battery and time-limit settings below be stated explicitly.
+    let user = format!(
+        "{}\\{}",
+        std::env::var("USERDOMAIN").unwrap_or_else(|_| "%COMPUTERNAME%".into()),
+        std::env::var("USERNAME").unwrap_or_default()
+    );
+    let xml = format!(
+        r#"<?xml version="1.0" encoding="UTF-16"?>
+<Task version="1.2" xmlns="http://schemas.microsoft.com/windows/2004/02/mit/task">
+  <RegistrationInfo>
+    <Description>Runs the OSTP TUN helper elevated so enabling the tunnel does not prompt for consent every time.</Description>
+  </RegistrationInfo>
+  <Principals>
+    <Principal id="Author">
+      <UserId>{user}</UserId>
+      <LogonType>InteractiveToken</LogonType>
+      <RunLevel>HighestAvailable</RunLevel>
+    </Principal>
+  </Principals>
+  <Settings>
+    <MultipleInstancesPolicy>Parallel</MultipleInstancesPolicy>
+    <DisallowStartIfOnBatteries>false</DisallowStartIfOnBatteries>
+    <StopIfGoingOnBatteries>false</StopIfGoingOnBatteries>
+    <StartWhenAvailable>false</StartWhenAvailable>
+    <RunOnlyIfNetworkAvailable>false</RunOnlyIfNetworkAvailable>
+    <ExecutionTimeLimit>PT0S</ExecutionTimeLimit>
+    <Enabled>true</Enabled>
+    <Hidden>false</Hidden>
+    <AllowHardTerminate>true</AllowHardTerminate>
+  </Settings>
+  <Actions Context="Author">
+    <Exec>
+      <Command>{exe}</Command>
+      <Arguments>--args-file "{args}"</Arguments>
+    </Exec>
+  </Actions>
+</Task>
+"#,
+        user = xml_escape(&user),
+        exe = xml_escape(&exe.display().to_string()),
+        args = xml_escape(&args_file.display().to_string()),
+    );
+
+    // schtasks /Create /XML expects UTF-16LE with a BOM.
+    let xml_path = std::env::temp_dir().join(format!("ostp_task_{}.xml", rand::random::<u32>()));
+    let mut utf16: Vec<u8> = vec![0xFF, 0xFE];
+    for unit in xml.encode_utf16() {
+        utf16.extend_from_slice(&unit.to_le_bytes());
+    }
+    std::fs::write(&xml_path, &utf16)?;
+
+    // Registering a HighestAvailable task is itself privileged: this is the one
+    // prompt, and it happens once per machine.
+    //
+    // Elevate through PowerShell's Start-Process -Wait rather than
+    // ShellExecuteW. ShellExecuteW returns as soon as the elevated process is
+    // LAUNCHED, so the XML below was being deleted while schtasks was still
+    // starting up — registration then failed, leaving the user with a consent
+    // prompt that accomplished nothing, followed by a second prompt from the
+    // fallback path. -Wait makes the deletion safe and lets the exit code be
+    // checked instead of guessed at by polling.
+    //
+    // ArgumentList takes an array, so the task name and XML path never need
+    // quoting or escaping through a command line, only PowerShell's own
+    // single-quote doubling.
+    let ps = format!(
+        "$p = Start-Process -FilePath 'schtasks.exe' -Verb RunAs -Wait -PassThru \
+         -WindowStyle Hidden -ArgumentList @('/Create','/TN','{}','/XML','{}','/F'); \
+         exit $p.ExitCode",
+        ps_quote(HELPER_TASK_NAME),
+        ps_quote(&xml_path.display().to_string()),
+    );
+
+    let status = quiet_command("powershell")
+        .args(["-NoProfile", "-NonInteractive", "-WindowStyle", "Hidden", "-Command", &ps])
+        .status();
+
+    // schtasks has exited by now, so this is safe.
+    let _ = std::fs::remove_file(&xml_path);
+
+    match status {
+        Ok(s) if s.success() => {}
+        Ok(s) => anyhow::bail!(
+            "registering the scheduled task failed (exit code {:?}). A declined consent prompt \
+             reports 1223.",
+            s.code()
+        ),
+        Err(e) => anyhow::bail!("could not run powershell to register the task: {e}"),
+    }
+
+    if helper_task_matches(exe) {
+        Ok(())
+    } else {
+        anyhow::bail!("schtasks reported success but the task does not point at {}", exe.display())
+    }
+}
+
+/// Escape a value for embedding in a PowerShell single-quoted string.
+#[cfg(target_os = "windows")]
+fn ps_quote(s: &str) -> String {
+    s.replace('\'', "''")
+}
+
 #[cfg(target_os = "windows")]
 fn launch_as_admin(exe: &std::path::PathBuf, token: &str, port: u16) -> anyhow::Result<()> {
+    // Preferred path: hand the parameters over in a file and trigger the
+    // pre-registered task, which runs elevated with no prompt. Falls back to a
+    // direct elevated launch when the task is absent (first ever run, or the
+    // user removed it) — and that first run is also where the task gets created,
+    // so the prompt appears once rather than on every connect.
+    let args_file = helper_args_file();
+    if let Some(dir) = args_file.parent() {
+        let _ = std::fs::create_dir_all(dir);
+    }
+    let payload = serde_json::json!({ "port": port, "token": token });
+    let wrote_args = std::fs::write(&args_file, payload.to_string()).is_ok();
+
+    if wrote_args {
+        if !helper_task_matches(exe) {
+            if let Err(e) = install_helper_task(exe) {
+                eprintln!("[OSTP] could not register the helper task ({e}); falling back to a direct elevated launch");
+            }
+        }
+        if helper_task_matches(exe) {
+            let run = quiet_command("schtasks")
+                .args(["/Run", "/TN", HELPER_TASK_NAME])
+                .output();
+            match run {
+                Ok(o) if o.status.success() => return Ok(()),
+                Ok(o) => eprintln!(
+                    "[OSTP] schtasks /Run failed: {}",
+                    String::from_utf8_lossy(&o.stderr).trim()
+                ),
+                Err(e) => eprintln!("[OSTP] schtasks /Run could not start: {e}"),
+            }
+        }
+        // Falling through: remove the file so a stale token is not left behind.
+        let _ = std::fs::remove_file(&args_file);
+    }
+
+    launch_as_admin_direct(exe, token, port)
+}
+
+/// The original one-prompt-per-launch path, kept as the fallback.
+#[cfg(target_os = "windows")]
+fn launch_as_admin_direct(exe: &std::path::PathBuf, token: &str, port: u16) -> anyhow::Result<()> {
     use std::ffi::OsStr;
     use std::os::windows::ffi::OsStrExt;
     use std::ptr::null_mut;
@@ -797,8 +1134,50 @@ fn launch_as_admin(exe: &std::path::PathBuf, token: &str, port: u16) -> anyhow::
     Ok(())
 }
 
-#[cfg(not(target_os = "windows"))]
-fn launch_as_admin(_exe: &PathBuf, _token: &str, _port: u16) -> Result<()> { anyhow::bail!("Windows only."); }
+#[cfg(target_os = "linux")]
+fn launch_as_admin(exe: &PathBuf, token: &str, port: u16) -> Result<()> {
+    use std::os::unix::fs::PermissionsExt;
+    use std::process::Command;
+
+    // Same shape as the Windows path: the token goes through a file rather than
+    // argv, so it never shows up in the process list.
+    let token_file = std::env::temp_dir().join(format!("ostp_auth_{}.tmp", rand::random::<u32>()));
+    std::fs::write(&token_file, token)?;
+    // Unlike Windows, /tmp is world-readable here, and this token authenticates
+    // control of the privileged tunnel helper — restrict it to the owner.
+    let _ = std::fs::set_permissions(&token_file, std::fs::Permissions::from_mode(0o600));
+
+    // pkexec is polkit's front-end: in a desktop session it raises a graphical
+    // authentication dialog. sudo is not an option from a GUI process, which has
+    // no terminal to prompt on.
+    match Command::new("pkexec")
+        .arg(exe)
+        .arg("--port")
+        .arg(port.to_string())
+        .arg("--token-file")
+        .arg(&token_file)
+        .spawn()
+    {
+        Ok(_) => Ok(()),
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
+            let _ = std::fs::remove_file(&token_file);
+            anyhow::bail!(
+                "pkexec was not found, so the TUN helper cannot be granted the privileges it \
+                 needs. Install polkit (package \"policykit-1\" on Debian/Ubuntu, \"polkit\" on \
+                 Fedora/Arch), or use proxy mode, which needs no elevation."
+            )
+        }
+        Err(e) => {
+            let _ = std::fs::remove_file(&token_file);
+            Err(e.into())
+        }
+    }
+}
+
+#[cfg(not(any(target_os = "windows", target_os = "linux")))]
+fn launch_as_admin(_exe: &PathBuf, _token: &str, _port: u16) -> Result<()> {
+    anyhow::bail!("TUN mode needs a privileged helper, which is implemented on Windows and Linux only. Use proxy mode on this platform.");
+}
 
 #[cfg(target_os = "windows")]
 fn show_error_dialog(msg: &str) {

ostp-gui/src-tauri/tauri.conf.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://schema.tauri.app/config/2",
   "productName": "ostp-gui",
-  "version": "0.4.2",
+  "version": "0.4.4",
   "identifier": "com.ospab.ostp",
   "build": {
     "frontendDist": "../src"
@@ -11,9 +11,11 @@
     "windows": [
       {
         "title": "OSTP",
-        "width": 360,
-        "height": 680,
-        "resizable": false
+        "width": 400,
+        "height": 720,
+        "minWidth": 360,
+        "minHeight": 560,
+        "resizable": true
       }
     ],
     "security": {

ostp-gui/src/main.js

@@ -660,6 +660,13 @@ function loadSettingsIntoForm() {
   updateClientVisibility();
 }
 
+// Last values actually pushed to the OS / backend, so repeated saves that did
+// not change them stay free. Undefined until the first save, which is correct:
+// the first one should apply.
+let lastAppliedAutostart;
+let lastAppliedTunnelConfig;
+let hotReloadTimer;
+
 function collectAndSaveSettings() {
   const s = {
     tun:          inTun.checked,
@@ -686,19 +693,41 @@ function collectAndSaveSettings() {
     fragChunk:    parseInt(inFragChunk.value) || 2,
     fragSleep:    !isNaN(parseInt(inFragSleep.value)) ? parseInt(inFragSleep.value) : 2,
   };
+  // Cheap and local: safe to run on every debounced keystroke.
   saveClientSettings(s);
   updateClientVisibility();
 
-  // Set autostart
-  invoke('set_autostart', { enable: s.launchStartup }).catch(() => {});
+  // Everything below talks to the OS or restarts the tunnel. Running it per
+  // keystroke is what made typing in the exclusion fields lag by seconds: the
+  // 400ms debounce fires during natural pauses in typing, and each firing hit
+  // the Windows registry and then tore down and rebuilt the tunnel.
 
-  // Hot-reload exclusions if connected
+  // Only touch autostart when it actually changed — this is a registry write.
+  if (s.launchStartup !== lastAppliedAutostart) {
+    lastAppliedAutostart = s.launchStartup;
+    invoke('set_autostart', { enable: s.launchStartup }).catch(() => {});
+  }
+
+  // Hot-reload the tunnel only when something it actually reads has changed,
+  // and on a much longer debounce: a reload is disruptive, so it should land
+  // once the user has stopped editing rather than between keystrokes.
   if (appState === 'connected') {
-    const cfg = buildConfig();
-    if (cfg) {
-      invoke('save_config', { jsonContent: JSON.stringify(cfg, null, 2) })
-        .then(() => invoke('reload_tunnel'))
-        .catch(() => {});
+    const tunnelRelevant = JSON.stringify([
+      s.tun, s.killSwitch, s.mux, s.muxSessions, s.mtu, s.dns, s.socks,
+      s.exDomains, s.exIps, s.exProcs, s.junkEnabled, s.junkPcMin, s.junkPcMax,
+      s.junkPsMin, s.junkPsMax, s.tcpFrag, s.fragChunk, s.fragSleep,
+    ]);
+    if (tunnelRelevant !== lastAppliedTunnelConfig) {
+      clearTimeout(hotReloadTimer);
+      hotReloadTimer = setTimeout(() => {
+        lastAppliedTunnelConfig = tunnelRelevant;
+        const cfg = buildConfig();
+        if (cfg) {
+          invoke('save_config', { jsonContent: JSON.stringify(cfg, null, 2) })
+            .then(() => invoke('reload_tunnel'))
+            .catch(() => {});
+        }
+      }, 1500);
     }
   }
 }

ostp-gui/src/styles.css

@@ -99,6 +99,13 @@ a { text-decoration: none; }
 .app-root {
   position: relative;
   width: 100%;
+  /* The window is resizable so users on desktops where the toolkit does not
+     apply our DPI scaling (WebKitGTK on HiDPI Linux renders the configured
+     size as raw pixels, giving a postage-stamp window) can size it themselves.
+     Capping and centring the column keeps the intended narrow layout instead of
+     stretching controls across a wide window. */
+  max-width: 460px;
+  margin: 0 auto;
   height: 100%;
   display: flex;
   flex-direction: column;

ostp-server/src/dispatcher.rs

@@ -263,8 +263,19 @@ impl Dispatcher {
         self.peer_machines
             .iter()
             .map(|(&sid, ps)| {
-                let cwnd = (ps.machine.cwnd_packets() as i64).clamp(16, 16384);
+                // Ceiling matches MAX_CWND_PACKETS in ostp-core. The old 16384
+                // allowed ~20 MB outstanding toward one client — on a mobile
+                // downlink that is standing queue, not throughput, and it is the
+                // download direction that carries video.
+                let cwnd = (ps.machine.cwnd_packets() as i64).clamp(16, 1024);
                 let in_flight = ps.machine.in_flight_count() as i64;
+                // Pacing gates the RATE, cwnd only the outstanding amount. With
+                // the pacing bucket empty, report no headroom so the relay
+                // reader pauses instead of handing over another chunk that would
+                // leave back-to-back.
+                if !ps.machine.can_pace_packet() {
+                    return (sid, 0);
+                }
                 (sid, cwnd - in_flight)
             })
             .collect()

ostp-server/src/outbound.rs

@@ -48,10 +48,23 @@ pub async fn connect_target(
             }
             if action == OutboundAction::Proxy {
                 let proxy_addr = format!("{}:{}", outbound.address, outbound.port);
-                return match outbound.protocol.as_str() {
+                // Case-insensitive: a config saying "SOCKS5" means the same thing
+                // as "socks5", and silently treating it as unknown is a trap.
+                return match outbound.protocol.to_ascii_lowercase().as_str() {
                     "socks5" => connect_via_socks5(&proxy_addr, target).await,
                     "http" => connect_via_http(&proxy_addr, target).await,
-                    _ => connect_direct(target, connect_timeout).await,
+                    // FAIL CLOSED. This used to fall through to a direct
+                    // connection, so any unrecognised protocol string — a typo,
+                    // a case difference, an empty value — silently sent ALL TCP
+                    // straight out of the server while the operator believed it
+                    // was proxied. Combined with the same bug on the UDP path,
+                    // that is how one session ends up presenting two different
+                    // exit addresses to the remote site.
+                    other => Err(anyhow::anyhow!(
+                        "outbound.protocol is \"{other}\", which is not a supported proxy type \
+                         (expected \"socks5\" or \"http\"); refusing to connect to {target} \
+                         directly, because the rules asked for the proxy"
+                    )),
                 };
             }
         }
@@ -370,10 +383,22 @@ pub async fn connect_udp_target(
             }
             if action == OutboundAction::Proxy {
                 let proxy_addr = format!("{}:{}", outbound.address, outbound.port);
-                if outbound.protocol == "socks5" {
+                if outbound.protocol.eq_ignore_ascii_case("socks5") {
                     return connect_udp_via_socks5(&proxy_addr, server_udp).await;
                 }
-                // HTTP CONNECT does not support UDP. Fallback to direct.
+                // FAIL CLOSED. HTTP CONNECT genuinely cannot carry UDP — but the
+                // answer to that is not to send the datagrams in the clear. The
+                // previous "fallback to direct" honoured a Proxy rule by
+                // egressing from the server's own address, so with an HTTP
+                // upstream every UDP flow (QUIC, DNS) leaked while TCP stayed
+                // proxied, presenting two exit IPs to the same remote site.
+                return Err(anyhow::anyhow!(
+                    "outbound rules route UDP to {target} through the proxy, but the upstream \
+                     protocol is \"{}\", which cannot carry UDP. Refusing to send directly. \
+                     Use a socks5 upstream, or add an explicit udp rule with action \"direct\" \
+                     or \"block\" so the intent is recorded in the config.",
+                    outbound.protocol
+                ));
             }
         }
     }

ostp-server/src/relay_node.rs

@@ -1,403 +1,460 @@
-//! Authenticated Relay Node
+//! Transparent relay node.
 //!
-//! Принимает входящие UDP/TCP (UoT) соединения от клиентов,
-//! валидирует HMAC-подпись клиента, используя ключи синхронизированные с upstream-сервера,
-//! и слепо пробрасывает авторизованный трафик к целевому upstream-серверу.
+//! Forwards traffic to a fixed upstream OSTP server:
 //!
-//! Архитектура цепочек:
-//!   Клиент -> [Relay 1] -> [Relay 2] -> ... -> [Target Server]
-//! Каждый Relay скачивает access_keys напрямую с Target Server API.
+//!   Client -> [Relay] -> [Target server]
+//!
+//! ## Why this performs no authentication of its own
+//!
+//! The previous design had the relay authenticate clients itself, with an
+//! HMAC handshake and a background job that pulled the access-key list from the
+//! target server's management API. That was wrong on two counts.
+//!
+//! It did not work: no OSTP client has ever produced those credentials. The TCP
+//! path expected an HTTP request (`GET /stream` with an `Authorization: Bearer`
+//! header) and the UDP path expected a `timestamp || HMAC` preamble, while the
+//! client sends junk frames followed by length-prefixed OSTP frames, and an
+//! obfuscated Noise handshake, respectively. Every connection was rejected.
+//!
+//! It was also weak where it did apply: the HMAC covered only an 8-byte
+//! timestamp, so a captured signature was a bearer token that anyone could
+//! replay from any address for the length of the clock-skew window. And the
+//! HTTP handshake was a plaintext `GET /stream` on the wire, a greppable
+//! signature in a protocol whose entire premise is that no byte is
+//! recognisable.
+//!
+//! Authentication belongs where it is cryptographically meaningful: the target
+//! server already authenticates every session end-to-end via Noise with a PSK
+//! derived from the access key, and silently drops anything that fails. A relay
+//! that re-checks credentials adds a second, weaker gate and a copy of the key
+//! list on a machine that has no need for it. So this relay makes no security
+//! decisions at all — it is a pipe, and says so.
+//!
+//! What it does need is protection against being used as a resource sink, which
+//! is what the session cap and admission rate limit below are for. It forwards
+//! only to one fixed upstream and returns replies only to the sender, so it is
+//! not a reflector: the amplification factor is one.
 
-use anyhow::Result;
-use bytes::Bytes;
-use hmac::{Hmac, Mac};
-use sha2::Sha256;
+use anyhow::{Context, Result};
 use std::collections::HashMap;
 use std::net::SocketAddr;
-use std::sync::{Arc, RwLock};
-use std::time::{Duration, Instant, SystemTime, UNIX_EPOCH};
-use tokio::io::{AsyncReadExt, AsyncWriteExt};
+use std::sync::Arc;
+use std::time::{Duration, Instant};
 use tokio::net::{TcpListener, TcpStream, UdpSocket};
 use tokio::sync::Mutex;
 
-/// Конфигурация Relay-узла.
+/// Configuration for a relay node.
 #[derive(Debug, Clone)]
 pub struct RelayConfig {
-    /// Адрес(а) для прослушивания входящих соединений (UDP + TCP).
+    /// Address(es) to accept client traffic on (UDP and TCP both bind here).
     pub listen_addrs: Vec<String>,
-    /// Адрес upstream TCP для пересылки (обычно тот же порт, что и у target-сервера).
+    /// Upstream target for TCP (UoT) traffic.
     pub upstream_tcp: String,
-    /// Адрес upstream UDP.
+    /// Upstream target for UDP traffic.
     pub upstream_udp: String,
-    /// URL API target-сервера для получения access_keys.
-    /// Пример: "http://127.0.0.1:9090"
-    pub upstream_api_url: String,
-    /// Bearer-токен для аутентификации на API target-сервера.
-    pub upstream_api_token: String,
-    /// Интервал синхронизации ключей (секунды).
-    pub sync_interval_secs: u64,
 }
 
-type SharedKeys = Arc<RwLock<Vec<String>>>;
+/// Maximum concurrent UDP client sessions. Each holds one upstream socket and
+/// one reader task, so this bounds both file descriptors and tasks.
+const MAX_UDP_SESSIONS: usize = 4096;
+/// A UDP session with no traffic for this long is reclaimed. Mobile NAT
+/// bindings are typically shorter-lived than this, so it is generous enough not
+/// to break roaming clients.
+const UDP_SESSION_IDLE: Duration = Duration::from_secs(120);
+/// Maximum concurrent relayed TCP connections.
+const MAX_TCP_CONNECTIONS: usize = 4096;
+/// Sustained rate (and burst ceiling) for admitting NEW sessions, per second.
+/// Established sessions are never rate limited; this only bounds how fast an
+/// unknown source can cause state to be allocated.
+const NEW_SESSION_RATE: f64 = 200.0;
+/// How long to wait for the upstream TCP connection before giving up.
+const UPSTREAM_CONNECT_TIMEOUT: Duration = Duration::from_secs(8);
 
-/// Точка входа Relay-узла.
-pub async fn run_relay_node(cfg: RelayConfig) -> Result<()> {
-    let shared_keys: SharedKeys = Arc::new(RwLock::new(Vec::new()));
+/// Token bucket bounding how fast new sessions may be created.
+struct AdmissionLimiter {
+    tokens: f64,
+    last_refill: Instant,
+}
 
-    // Первоначальная синхронизация ключей
-    if let Err(e) = sync_keys(&cfg, &shared_keys).await {
-        tracing::warn!("Relay: initial key sync failed: {}. Will retry.", e);
-    } else {
-        let count = shared_keys.read().unwrap_or_else(|e| e.into_inner()).len();
-        tracing::info!("Relay: synced {} access key(s) from upstream API", count);
+impl AdmissionLimiter {
+    fn new() -> Self {
+        Self { tokens: NEW_SESSION_RATE, last_refill: Instant::now() }
     }
 
-    // Фоновый синхронизатор ключей
-    let cfg_clone = cfg.clone();
-    let keys_clone = shared_keys.clone();
+    /// Consume one admission slot, or report that the caller should drop.
+    fn try_admit(&mut self) -> bool {
+        let now = Instant::now();
+        let elapsed = now.duration_since(self.last_refill).as_secs_f64();
+        self.last_refill = now;
+        self.tokens = (self.tokens + elapsed * NEW_SESSION_RATE).min(NEW_SESSION_RATE);
+        if self.tokens >= 1.0 {
+            self.tokens -= 1.0;
+            true
+        } else {
+            false
+        }
+    }
+}
+
+/// Entry point.
+pub async fn run_relay_node(cfg: RelayConfig) -> Result<()> {
+    let udp_cfg = cfg.clone();
     tokio::spawn(async move {
-        loop {
-            tokio::time::sleep(Duration::from_secs(cfg_clone.sync_interval_secs)).await;
-            match sync_keys(&cfg_clone, &keys_clone).await {
-                Ok(count) => tracing::debug!("Relay: refreshed {} access key(s)", count),
-                Err(e) => tracing::warn!("Relay: key sync error: {}", e),
-            }
+        if let Err(e) = run_udp_relay(udp_cfg).await {
+            tracing::error!("Relay UDP loop error: {e}");
         }
     });
 
-    // Запуск UDP relay
-    {
-        let cfg_udp = cfg.clone();
-        let keys_udp = shared_keys.clone();
-        tokio::spawn(async move {
-            if let Err(e) = run_udp_relay(cfg_udp, keys_udp).await {
-                tracing::error!("Relay UDP loop error: {}", e);
-            }
-        });
-    }
-
-    // Запуск TCP (UoT) relay
-    run_tcp_relay(cfg, shared_keys).await
+    run_tcp_relay(cfg).await
 }
 
-/// Синхронизация access_keys с upstream API.
-async fn sync_keys(cfg: &RelayConfig, shared_keys: &SharedKeys) -> Result<usize> {
-    let url = format!("{}/api/users", cfg.upstream_api_url.trim_end_matches('/'));
+// ── UDP ──────────────────────────────────────────────────────────────────────
 
-    let client = reqwest::Client::builder()
-        .timeout(Duration::from_secs(10))
-        .build()?;
-
-    let mut req = client.get(&url);
-    if !cfg.upstream_api_token.is_empty() {
-        req = req.header("Authorization", format!("Bearer {}", cfg.upstream_api_token));
-    }
-
-    let resp = req.send().await?;
-    if !resp.status().is_success() {
-        anyhow::bail!("API returned HTTP {}", resp.status());
-    }
-
-    #[derive(serde::Deserialize)]
-    struct UserStatsSnapshot {
-        access_key: String,
-    }
-
-    #[derive(serde::Deserialize)]
-    struct ApiResponse {
-        ok: bool,
-        data: Option<Vec<UserStatsSnapshot>>,
-    }
-
-    let body: ApiResponse = resp.json().await?;
-    if !body.ok {
-        anyhow::bail!("API returned error ok=false");
-    }
-
-    let keys: Vec<String> = body.data.unwrap_or_default().into_iter().map(|u| u.access_key).collect();
-    let count = keys.len();
-    {
-        let mut lock = shared_keys.write().unwrap();
-        *lock = keys;
-    }
-    Ok(count)
+struct UdpSession {
+    upstream: Arc<UdpSocket>,
+    last_seen: Instant,
 }
 
-/// Проверяет HMAC-подпись клиента по набору ключей.
-/// Возвращает true если хотя бы один ключ подходит.
-fn verify_hmac(ts_bytes: &[u8; 8], provided_mac: &[u8], keys: &[String]) -> bool {
-    let client_ts = u64::from_be_bytes(*ts_bytes);
-    let now = SystemTime::now()
-        .duration_since(UNIX_EPOCH)
-        .unwrap()
-        .as_secs();
-
-    // Защита от replay: ±60 секунд
-    if client_ts > now + 30 || client_ts < now.saturating_sub(60) {
-        return false;
-    }
-
-    for key in keys {
-        if let Ok(mut mac) = Hmac::<Sha256>::new_from_slice(key.as_bytes()) {
-            mac.update(ts_bytes);
-            if mac.verify_slice(provided_mac).is_ok() {
-                return true;
-            }
-        }
-    }
-    false
-}
-
-// ── UDP Relay ────────────────────────────────────────────────────────────────
-
-async fn run_udp_relay(cfg: RelayConfig, shared_keys: SharedKeys) -> Result<()> {
-    // NAT-таблица: client_addr -> (upstream_socket, last_seen)
-    let nat_table: Arc<Mutex<HashMap<SocketAddr, (Arc<UdpSocket>, Instant)>>> =
+async fn run_udp_relay(cfg: RelayConfig) -> Result<()> {
+    // client address -> the upstream socket carrying that client's flow
+    let sessions: Arc<Mutex<HashMap<SocketAddr, UdpSession>>> =
         Arc::new(Mutex::new(HashMap::new()));
+    let limiter = Arc::new(Mutex::new(AdmissionLimiter::new()));
 
     for bind_addr in &cfg.listen_addrs {
-        let sock = UdpSocket::bind(bind_addr).await?;
-        tracing::info!("Relay UDP listening on {}", bind_addr);
-        let sock = Arc::new(sock);
-        let upstream_udp = cfg.upstream_udp.clone();
-        let keys = shared_keys.clone();
-        let nat = nat_table.clone();
+        let sock = Arc::new(
+            UdpSocket::bind(bind_addr)
+                .await
+                .with_context(|| format!("relay: failed to bind UDP on {bind_addr}"))?,
+        );
+        tracing::info!("Relay UDP listening on {bind_addr} -> {}", cfg.upstream_udp);
+
+        let upstream_addr = cfg.upstream_udp.clone();
+        let sessions = sessions.clone();
+        let limiter = limiter.clone();
 
         tokio::spawn(async move {
             let mut buf = vec![0u8; 65535];
             loop {
-                let (n, peer) = match sock.recv_from(&mut buf).await {
-                    Ok(v) => v,
-                    Err(_) => continue,
-                };
-
-                let packet = Bytes::copy_from_slice(&buf[..n]);
-
-                // Быстрая проверка: первый UDP-пакет от нового клиента содержит Noise handshake.
-                // Мы берём из него первые 8 байт как timestamp + 32 байта MAC.
-                // Если пакет достаточно длинный, проверяем подпись.
-                // Для уже авторизованных клиентов (есть в NAT) — пропускаем проверку.
-                {
-                    let nat_lock = nat.lock().await;
-                    if !nat_lock.contains_key(&peer) {
-                        drop(nat_lock);
-
-                        // Пакет должен быть >= 40 байт (8 ts + 32 hmac) для первичной проверки
-                        if packet.len() < 40 {
-                            tracing::debug!("Relay UDP: dropping short packet from {}", peer);
-                            continue;
-                        }
-
-                        let ts_bytes: [u8; 8] = packet[0..8].try_into().unwrap();
-                        let provided_mac = &packet[8..40];
-                        let keys_guard = keys.read().unwrap_or_else(|e| e.into_inner());
-
-                        if !verify_hmac(&ts_bytes, provided_mac, &keys_guard) {
-                            tracing::debug!("Relay UDP: unauthorized probe from {}, dropped", peer);
-                            continue;
-                        }
-                        tracing::debug!("Relay UDP: authorized new client {}", peer);
-                    }
-                }
-
-                // Находим или создаём upstream socket для этого клиента
-                let upstream_sock = {
-                    let mut nat_lock = nat.lock().await;
-                    if let Some(entry) = nat_lock.get_mut(&peer) {
-                        entry.1 = Instant::now();
-                        entry.0.clone()
-                    } else {
-                        // Новый upstream socket для этого клиента
-                        let usock = match UdpSocket::bind("0.0.0.0:0").await {
-                            Ok(s) => Arc::new(s),
-                            Err(e) => {
-                                tracing::warn!("Relay UDP: failed to bind upstream socket: {}", e);
-                                continue;
-                            }
-                        };
-                        if usock.connect(&upstream_udp).await.is_err() {
-                            tracing::warn!("Relay UDP: failed to connect to upstream {}", upstream_udp);
-                            continue;
-                        }
-
-                        nat_lock.insert(peer, (usock.clone(), Instant::now()));
-
-                        // Задача: читаем ответы от upstream и отправляем клиенту
-                        let usock_rx = usock.clone();
-                        let client_sock = sock.clone();
-                        let peer_addr = peer;
-                        tokio::spawn(async move {
-                            let mut rbuf = vec![0u8; 65535];
-                            loop {
-                                match usock_rx.recv(&mut rbuf).await {
-                                    Ok(n) => {
-                                        let _ = client_sock.send_to(&rbuf[..n], peer_addr).await;
-                                    }
-                                    Err(_) => break,
-                                }
-                            }
-                        });
-
-                        usock
-                    }
-                };
-
-                // Пересылаем пакет в upstream
-                let _ = upstream_sock.send(&packet).await;
-            }
-        });
-    }
-
-    // Периодически чистим устаревшие NAT записи (timeout 120 сек)
-    loop {
-        tokio::time::sleep(Duration::from_secs(30)).await;
-        let mut nat_lock = nat_table.lock().await;
-        let now = Instant::now();
-        nat_lock.retain(|_, (_, last)| now.duration_since(*last) < Duration::from_secs(120));
-    }
-}
-
-// ── TCP (UoT) Relay ──────────────────────────────────────────────────────────
-
-async fn run_tcp_relay(cfg: RelayConfig, shared_keys: SharedKeys) -> Result<()> {
-    for bind_addr in &cfg.listen_addrs {
-        let listener = TcpListener::bind(bind_addr).await?;
-        tracing::info!("Relay TCP (UoT) listening on {}", bind_addr);
-
-        let upstream_tcp = cfg.upstream_tcp.clone();
-        let keys = shared_keys.clone();
-
-        tokio::spawn(async move {
-            loop {
-                let (stream, peer_addr) = match listener.accept().await {
+                let (len, peer) = match sock.recv_from(&mut buf).await {
                     Ok(v) => v,
                     Err(e) => {
-                        tracing::warn!("Relay TCP accept error: {}", e);
+                        tracing::warn!("Relay UDP recv error: {e}");
                         continue;
                     }
                 };
 
-                let upstream = upstream_tcp.clone();
-                let keys_clone = keys.clone();
-
-                tokio::spawn(async move {
-                    if let Err(e) = handle_tcp_client(stream, peer_addr, upstream, keys_clone).await {
-                        tracing::debug!("Relay TCP client {} closed: {}", peer_addr, e);
+                // Fast path: an established session just forwards.
+                {
+                    let mut map = sessions.lock().await;
+                    if let Some(session) = map.get_mut(&peer) {
+                        session.last_seen = Instant::now();
+                        let upstream = session.upstream.clone();
+                        drop(map);
+                        let _ = upstream.send(&buf[..len]).await;
+                        continue;
                     }
+                }
+
+                // New client: bounded by both a hard cap and an admission rate,
+                // so a flood of spoofed sources cannot exhaust sockets or tasks.
+                {
+                    let map = sessions.lock().await;
+                    if map.len() >= MAX_UDP_SESSIONS {
+                        continue;
+                    }
+                }
+                if !limiter.lock().await.try_admit() {
+                    continue;
+                }
+
+                let upstream = match new_upstream_socket(&upstream_addr).await {
+                    Ok(s) => s,
+                    Err(e) => {
+                        tracing::warn!("Relay UDP: cannot reach upstream {upstream_addr}: {e}");
+                        continue;
+                    }
+                };
+
+                sessions.lock().await.insert(
+                    peer,
+                    UdpSession { upstream: upstream.clone(), last_seen: Instant::now() },
+                );
+
+                // Reverse direction for this client.
+                let back_sock = sock.clone();
+                let sessions_rx = sessions.clone();
+                tokio::spawn(async move {
+                    let mut rbuf = vec![0u8; 65535];
+                    loop {
+                        match upstream.recv(&mut rbuf).await {
+                            Ok(n) => {
+                                if back_sock.send_to(&rbuf[..n], peer).await.is_err() {
+                                    break;
+                                }
+                                if let Some(s) = sessions_rx.lock().await.get_mut(&peer) {
+                                    s.last_seen = Instant::now();
+                                }
+                            }
+                            Err(_) => break,
+                        }
+                    }
+                    sessions_rx.lock().await.remove(&peer);
+                });
+
+                let _ = sessions
+                    .lock()
+                    .await
+                    .get(&peer)
+                    .map(|s| s.upstream.clone())
+                    .unwrap()
+                    .send(&buf[..len])
+                    .await;
+            }
+        });
+    }
+
+    // Reclaim idle sessions. Dropping the entry closes the upstream socket,
+    // which ends that session's reader task.
+    loop {
+        tokio::time::sleep(Duration::from_secs(30)).await;
+        let now = Instant::now();
+        let mut map = sessions.lock().await;
+        let before = map.len();
+        map.retain(|_, s| now.duration_since(s.last_seen) < UDP_SESSION_IDLE);
+        let reclaimed = before - map.len();
+        if reclaimed > 0 {
+            tracing::debug!("Relay UDP: reclaimed {reclaimed} idle session(s), {} active", map.len());
+        }
+    }
+}
+
+/// One upstream socket per client, `connect`ed so replies can be read with
+/// `recv` and cannot come from anywhere else.
+async fn new_upstream_socket(upstream: &str) -> Result<Arc<UdpSocket>> {
+    // Resolve first, then bind the SAME address family. Binding "[::]:0" and
+    // connecting to an IPv4 upstream fails anywhere IPV6_V6ONLY defaults on
+    // (Windows, and many Linux configurations) — which is every deployment with
+    // an IPv4 target server, i.e. the common case.
+    let addr: SocketAddr = tokio::net::lookup_host(upstream)
+        .await
+        .with_context(|| format!("resolve upstream {upstream}"))?
+        .next()
+        .ok_or_else(|| anyhow::anyhow!("upstream {upstream} resolved to no addresses"))?;
+
+    let bind: SocketAddr = if addr.is_ipv6() {
+        "[::]:0".parse().expect("valid literal")
+    } else {
+        "0.0.0.0:0".parse().expect("valid literal")
+    };
+
+    let sock = UdpSocket::bind(bind).await?;
+    sock.connect(addr)
+        .await
+        .with_context(|| format!("connect to upstream {addr}"))?;
+    Ok(Arc::new(sock))
+}
+
+// ── TCP (UoT) ────────────────────────────────────────────────────────────────
+
+async fn run_tcp_relay(cfg: RelayConfig) -> Result<()> {
+    let live = Arc::new(std::sync::atomic::AtomicUsize::new(0));
+
+    for bind_addr in &cfg.listen_addrs {
+        let listener = TcpListener::bind(bind_addr)
+            .await
+            .with_context(|| format!("relay: failed to bind TCP on {bind_addr}"))?;
+        tracing::info!("Relay TCP (UoT) listening on {bind_addr} -> {}", cfg.upstream_tcp);
+
+        let upstream = cfg.upstream_tcp.clone();
+        let live = live.clone();
+
+        tokio::spawn(async move {
+            loop {
+                let (client, peer) = match listener.accept().await {
+                    Ok(v) => v,
+                    Err(e) => {
+                        tracing::warn!("Relay TCP accept error: {e}");
+                        continue;
+                    }
+                };
+
+                use std::sync::atomic::Ordering;
+                if live.load(Ordering::Relaxed) >= MAX_TCP_CONNECTIONS {
+                    // Close immediately rather than queueing unbounded work.
+                    drop(client);
+                    continue;
+                }
+                live.fetch_add(1, Ordering::Relaxed);
+
+                let upstream = upstream.clone();
+                let live = live.clone();
+                tokio::spawn(async move {
+                    if let Err(e) = splice_tcp(client, &upstream).await {
+                        tracing::debug!("Relay TCP {peer} closed: {e}");
+                    }
+                    live.fetch_sub(1, Ordering::Relaxed);
                 });
             }
         });
     }
 
-    // Держим поток живым
     futures_util::future::pending::<()>().await;
     Ok(())
 }
 
-/// Обработка одного TCP (UoT) соединения.
+/// Splice a client connection to the upstream, byte for byte.
 ///
-/// Алгоритм:
-/// 1. Читаем HTTP-заголовки (фейковый WebSocket upgrade).
-/// 2. Извлекаем HMAC-подпись из Authorization: Bearer.
-/// 3. Проверяем подпись по синхронизированным ключам.
-/// 4. Если авторизован — открываем соединение к upstream и пайпим потоки.
-async fn handle_tcp_client(
-    mut client: TcpStream,
-    peer_addr: SocketAddr,
-    upstream_addr: String,
-    shared_keys: SharedKeys,
-) -> Result<()> {
-    // Читаем HTTP-заголовки (до \r\n\r\n)
-    let mut header_buf = vec![0u8; 4096];
-    let mut header_len = 0usize;
-
-    loop {
-        let n = client.read(&mut header_buf[header_len..]).await?;
-        if n == 0 {
-            anyhow::bail!("connection closed before handshake");
-        }
-        header_len += n;
-        if header_buf[..header_len].windows(4).any(|w| w == b"\r\n\r\n") {
-            break;
-        }
-        if header_len >= header_buf.len() {
-            anyhow::bail!("headers too large");
-        }
-    }
-
-    let headers_str = String::from_utf8_lossy(&header_buf[..header_len]);
-
-    // Быстрая проверка: должен быть GET /stream
-    if !headers_str.starts_with("GET /stream HTTP/1.1\r\n") {
-        // Возвращаем 404 как обычный сервер (anti-scan)
-        let _ = client.write_all(b"HTTP/1.1 404 Not Found\r\nContent-Length: 9\r\nConnection: close\r\n\r\nNot Found").await;
-        anyhow::bail!("invalid request from {}", peer_addr);
-    }
-
-    // Извлекаем HMAC-подпись
-    let mut sig_b64 = None;
-    for line in headers_str.lines() {
-        let lower = line.to_ascii_lowercase();
-        if lower.starts_with("authorization: bearer ") {
-            sig_b64 = Some(line[22..].trim().to_string());
-        } else if lower.starts_with("cookie: ostp_token=") {
-            sig_b64 = Some(line[19..].trim().to_string());
-        }
-    }
-
-    let sig_b64 = match sig_b64 {
-        Some(s) => s,
-        None => {
-            let _ = client.write_all(b"HTTP/1.1 404 Not Found\r\nContent-Length: 9\r\nConnection: close\r\n\r\nNot Found").await;
-            anyhow::bail!("missing authorization from {}", peer_addr);
-        }
-    };
-
-    let sig_bytes = base64::Engine::decode(
-        &base64::engine::general_purpose::STANDARD_NO_PAD,
-        &sig_b64,
+/// Nothing is parsed or rewritten: the relay must stay agnostic to the payload,
+/// both because the payload is an opaque encrypted stream and because any
+/// parsing would be a place for the relay to disagree with the endpoints.
+async fn splice_tcp(mut client: TcpStream, upstream_addr: &str) -> Result<()> {
+    let mut upstream = tokio::time::timeout(
+        UPSTREAM_CONNECT_TIMEOUT,
+        TcpStream::connect(upstream_addr),
     )
-    .map_err(|_| anyhow::anyhow!("invalid base64 from {}", peer_addr))?;
+    .await
+    .map_err(|_| anyhow::anyhow!("upstream {upstream_addr} connect timed out"))?
+    .with_context(|| format!("connect to upstream {upstream_addr}"))?;
 
-    if sig_bytes.len() < 40 {
-        let _ = client.write_all(b"HTTP/1.1 401 Unauthorized\r\nContent-Length: 12\r\nConnection: close\r\n\r\nUnauthorized").await;
-        anyhow::bail!("signature too short from {}", peer_addr);
-    }
+    // Both sides carry latency-sensitive framed traffic; Nagle would add delay
+    // for no benefit on an already-batched stream.
+    let _ = client.set_nodelay(true);
+    let _ = upstream.set_nodelay(true);
 
-    let ts_bytes: [u8; 8] = sig_bytes[0..8].try_into().unwrap();
-    let provided_mac = &sig_bytes[8..];
-
-    // Проверяем по синхронизированным ключам
-    let authorized = {
-        let keys = shared_keys.read().unwrap_or_else(|e| e.into_inner());
-        verify_hmac(&ts_bytes, provided_mac, &keys)
-    };
-
-    if !authorized {
-        let _ = client.write_all(b"HTTP/1.1 404 Not Found\r\nContent-Length: 9\r\nConnection: close\r\n\r\nNot Found").await;
-        anyhow::bail!("unauthorized client {}", peer_addr);
-    }
-
-    tracing::info!("Relay TCP: authorized client {}, forwarding to {}", peer_addr, upstream_addr);
-
-    // Подключаемся к upstream
-    let mut upstream = TcpStream::connect(&upstream_addr).await
-        .map_err(|e| anyhow::anyhow!("failed to connect to upstream {}: {}", upstream_addr, e))?;
-
-    // Пересылаем upstream заголовки AS-IS (он сам проверит подпись)
-    upstream.write_all(&header_buf[..header_len]).await?;
-
-    // Пайпим оба потока: client <-> upstream
-    let (mut cr, mut cw) = client.into_split();
-    let (mut ur, mut uw) = upstream.into_split();
-
-    let c2u = tokio::spawn(async move {
-        let _ = tokio::io::copy(&mut cr, &mut uw).await;
-    });
-    let u2c = tokio::spawn(async move {
-        let _ = tokio::io::copy(&mut ur, &mut cw).await;
-    });
-
-    let _ = tokio::join!(c2u, u2c);
+    tokio::io::copy_bidirectional(&mut client, &mut upstream).await?;
     Ok(())
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    /// The admission limiter is what replaced per-client authentication as the
+    /// defence against resource abuse, so it has to actually stop admitting.
+    #[test]
+    fn admission_limiter_stops_at_the_burst_ceiling() {
+        let mut limiter = AdmissionLimiter::new();
+        let mut admitted = 0usize;
+        // Ask for far more than one burst without letting time pass.
+        for _ in 0..(NEW_SESSION_RATE as usize * 3) {
+            if limiter.try_admit() {
+                admitted += 1;
+            }
+        }
+        assert!(
+            admitted <= NEW_SESSION_RATE as usize + 1,
+            "admitted {admitted} sessions in one instant, ceiling is {NEW_SESSION_RATE}"
+        );
+        assert!(admitted > 0, "limiter admitted nothing at all");
+    }
+
+    /// It must also refill, or the relay would accept a burst once and then
+    /// refuse every client forever.
+    #[test]
+    fn admission_limiter_refills_over_time() {
+        let mut limiter = AdmissionLimiter::new();
+        while limiter.try_admit() {}
+        assert!(!limiter.try_admit(), "bucket should be empty");
+
+        std::thread::sleep(Duration::from_millis(50));
+        assert!(
+            limiter.try_admit(),
+            "limiter never refilled; the relay would stop accepting new clients"
+        );
+    }
+
+    /// End-to-end through the real UDP path: a client datagram reaches the
+    /// upstream and the reply comes back to that same client. This is the whole
+    /// job of the relay, and it is what the previous implementation could not do
+    /// with a real client, because it demanded credentials no client sends.
+    #[tokio::test]
+    async fn udp_relay_forwards_both_directions() {
+        // Stand-in upstream that echoes with a marker.
+        let upstream = UdpSocket::bind("127.0.0.1:0").await.unwrap();
+        let upstream_addr = upstream.local_addr().unwrap();
+        tokio::spawn(async move {
+            let mut buf = [0u8; 1500];
+            while let Ok((n, from)) = upstream.recv_from(&mut buf).await {
+                let mut reply = b"echo:".to_vec();
+                reply.extend_from_slice(&buf[..n]);
+                let _ = upstream.send_to(&reply, from).await;
+            }
+        });
+
+        let relay_listen = {
+            let probe = UdpSocket::bind("127.0.0.1:0").await.unwrap();
+            let a = probe.local_addr().unwrap();
+            drop(probe);
+            a
+        };
+
+        tokio::spawn(run_udp_relay(RelayConfig {
+            listen_addrs: vec![relay_listen.to_string()],
+            upstream_tcp: upstream_addr.to_string(),
+            upstream_udp: upstream_addr.to_string(),
+        }));
+        tokio::time::sleep(Duration::from_millis(150)).await;
+
+        // A plain OSTP-looking datagram: no credentials, no preamble.
+        let client = UdpSocket::bind("127.0.0.1:0").await.unwrap();
+        client.send_to(b"opaque-payload", relay_listen).await.unwrap();
+
+        let mut buf = [0u8; 1500];
+        let (n, _) = tokio::time::timeout(Duration::from_secs(3), client.recv_from(&mut buf))
+            .await
+            .expect("relay did not deliver a reply within 3s")
+            .unwrap();
+
+        assert_eq!(
+            &buf[..n],
+            b"echo:opaque-payload",
+            "relay did not forward the payload verbatim in both directions"
+        );
+    }
+
+    /// Same for TCP: bytes must cross unmodified in both directions, with no
+    /// handshake demanded of the client.
+    #[tokio::test]
+    async fn tcp_relay_splices_both_directions() {
+        use tokio::io::{AsyncReadExt, AsyncWriteExt};
+
+        let upstream = TcpListener::bind("127.0.0.1:0").await.unwrap();
+        let upstream_addr = upstream.local_addr().unwrap();
+        tokio::spawn(async move {
+            if let Ok((mut sock, _)) = upstream.accept().await {
+                let mut buf = [0u8; 128];
+                if let Ok(n) = sock.read(&mut buf).await {
+                    let mut reply = b"echo:".to_vec();
+                    reply.extend_from_slice(&buf[..n]);
+                    let _ = sock.write_all(&reply).await;
+                }
+            }
+        });
+
+        let relay_listen = {
+            let probe = TcpListener::bind("127.0.0.1:0").await.unwrap();
+            let a = probe.local_addr().unwrap();
+            drop(probe);
+            a
+        };
+
+        tokio::spawn(run_tcp_relay(RelayConfig {
+            listen_addrs: vec![relay_listen.to_string()],
+            upstream_tcp: upstream_addr.to_string(),
+            upstream_udp: upstream_addr.to_string(),
+        }));
+        tokio::time::sleep(Duration::from_millis(150)).await;
+
+        let mut client = TcpStream::connect(relay_listen).await.unwrap();
+        client.write_all(b"opaque-stream").await.unwrap();
+
+        let mut buf = [0u8; 128];
+        let n = tokio::time::timeout(Duration::from_secs(3), client.read(&mut buf))
+            .await
+            .expect("relay did not deliver a reply within 3s")
+            .unwrap();
+
+        assert_eq!(&buf[..n], b"echo:opaque-stream");
+    }
+}

ostp-server/src/router.rs

@@ -47,12 +47,29 @@ impl Router {
         
         let mut proxy = None;
         if let Some(ref c) = cfg {
-            if c.enabled && c.protocol == "socks5" {
-                let proxy_addr = format!("{}:{}", c.address, c.port);
-                if let Ok(p) = crate::outbound::connect_udp_via_socks5(&proxy_addr, server_udp.clone()).await {
-                    proxy = Some(Arc::new(p));
-                } else if self.debug {
-                    tracing::warn!("Failed to establish SOCKS5 UDP Associate");
+            if c.enabled {
+                if c.protocol == "socks5" {
+                    let proxy_addr = format!("{}:{}", c.address, c.port);
+                    match crate::outbound::connect_udp_via_socks5(&proxy_addr, server_udp.clone()).await {
+                        Ok(p) => proxy = Some(Arc::new(p)),
+                        // Warn unconditionally, not only under `debug`. Every UDP
+                        // flow the rules want proxied is now dropped instead of
+                        // sent, so an operator who cannot see this has a session
+                        // where TCP works and UDP silently does not.
+                        Err(e) => tracing::warn!(
+                            "SOCKS5 UDP ASSOCIATE to {proxy_addr} failed: {e}. UDP that the \
+                             outbound rules route through the proxy will be DROPPED (it is not \
+                             sent directly, which would expose this server's address)."
+                        ),
+                    }
+                } else {
+                    tracing::warn!(
+                        "Upstream proxy protocol is '{}', which cannot carry UDP. UDP matching \
+                         a Proxy rule will be DROPPED. Use a socks5 upstream for UDP, or add an \
+                         explicit udp rule with action \"direct\" or \"block\" to make the \
+                         intent explicit.",
+                        c.protocol
+                    );
                 }
             }
         }
@@ -87,9 +104,28 @@ impl UdpSessionRouter {
                     return Err(anyhow::anyhow!("blocked by outbound udp rule: {}", target));
                 }
                 if action == crate::outbound::OutboundAction::Proxy {
-                    if let Some(p) = &self.proxy {
-                        return p.send_to(data, target).await;
-                    }
+                    return match &self.proxy {
+                        Some(p) => p.send_to(data, target).await,
+                        // FAIL CLOSED. This used to fall through to the direct
+                        // socket, so whenever the UDP proxy was unavailable —
+                        // the SOCKS5 UDP ASSOCIATE failed, or the upstream is an
+                        // HTTP proxy, which cannot carry UDP at all — every UDP
+                        // datagram silently egressed from the server's own
+                        // address while TCP still went through the proxy. The
+                        // session then had two different exit IPs, which is what
+                        // Google flags and why YouTube (QUIC, i.e. UDP/443)
+                        // geolocated to the server instead of the proxy exit.
+                        //
+                        // A rule that says "proxy" must never be satisfied by
+                        // sending in the clear: a dropped datagram is visible and
+                        // debuggable, a deanonymising leak is neither.
+                        None => Err(anyhow::anyhow!(
+                            "outbound rule requires the proxy for UDP to {target}, but no UDP \
+                             proxy is available (SOCKS5 UDP ASSOCIATE failed, or the upstream \
+                             is an HTTP proxy, which cannot carry UDP) - dropping rather than \
+                             leaking the server's own address"
+                        )),
+                    };
                 }
             }
         }

ostp-tun-helper/build.rs

@@ -4,6 +4,21 @@
 // or launched via ShellExecuteW("runas").
 
 fn main() {
+    // Key off the TARGET, not the host. In a build script `cfg(windows)`
+    // describes the machine doing the building, so cross-compiling the helper
+    // from Windows to Linux took this branch and failed with "Can only compile
+    // resource file when target_env is gnu or msvc". CARGO_CFG_TARGET_OS is the
+    // target being built for, which is what actually decides whether a Windows
+    // manifest belongs in the binary.
+    let target_os = std::env::var("CARGO_CFG_TARGET_OS").unwrap_or_default();
+    if target_os != "windows" {
+        return;
+    }
+    // Second gate, on the HOST: winres is declared under
+    // [target.'cfg(windows)'.build-dependencies], and build-dependencies are
+    // resolved against the host triple, so the crate simply does not exist when
+    // building on Linux. Referencing it unconditionally would fail to compile
+    // there even though the target check above already passed.
     #[cfg(windows)]
     {
         let mut res = winres::WindowsResource::new();

ostp-tun-helper/src/main.rs

@@ -24,6 +24,14 @@ fn log_to_file(msg: &str) {
 
 
 
+/// Launch parameters handed over in a file rather than on the command line.
+/// See the `--args-file` handling in `main` for why.
+#[derive(Deserialize)]
+struct HelperArgs {
+    port: u16,
+    token: String,
+}
+
 #[derive(Deserialize)]
 #[serde(tag = "cmd", rename_all = "lowercase")]
 enum GuiCmd {
@@ -76,6 +84,28 @@ async fn main() -> Result<()> {
                 let _ = std::fs::remove_file(path); // securely delete after reading
             }
         }
+        // Both port and token from one file. A Scheduled Task stores a FIXED
+        // command line, so anything that varies per launch cannot be passed as
+        // an argument — the GUI writes this file immediately before triggering
+        // the task instead. That indirection is what lets the task be created
+        // once (a single UAC prompt) and reused for every later connect without
+        // prompting again.
+        if args[i] == "--args-file" && i + 1 < args.len() {
+            let path = &args[i + 1];
+            match std::fs::read_to_string(path) {
+                Ok(content) => {
+                    let _ = std::fs::remove_file(path); // single use
+                    match serde_json::from_str::<HelperArgs>(&content) {
+                        Ok(parsed) => {
+                            port = parsed.port;
+                            expected_token = parsed.token;
+                        }
+                        Err(e) => log_to_file(&format!("Failed to parse --args-file: {e}")),
+                    }
+                }
+                Err(e) => log_to_file(&format!("Failed to read --args-file {path}: {e}")),
+            }
+        }
     }
 
     log_to_file("Helper started (TCP mode)");

ostp/src/main.rs

@@ -28,6 +28,12 @@ enum Commands {
     Init {
         mode: String,
     },
+    /// Hash a password for the web panel's `api.password_hash` config field
+    #[command(name = "hash-password", alias = "hp")]
+    HashPassword {
+        /// The password to hash. Omit to be prompted (keeps it out of shell history).
+        password: Option<String>,
+    },
     /// Generate a new secure access key
     #[command(name = "gk", alias = "generate-key")]
     GenerateKey {
@@ -793,18 +799,16 @@ fn run_setup_wizard(config_path: &std::path::Path) -> Result<()> {
             let listen   = wizard_prompt("Listen address (host:port)", "0.0.0.0:50000");
             let upstream = wizard_prompt("Upstream server address (host:port)", "");
             if upstream.is_empty() { anyhow::bail!("Upstream address cannot be empty."); }
-            let api_url  = wizard_prompt("Upstream server API URL (e.g. http://1.2.3.4:9090)", "");
-            let api_token = wizard_prompt("Upstream API token (leave blank if none)", "");
 
             wizard_step(2, TOTAL, "Saving configuration");
+            // No credentials are collected: the relay forwards transparently and
+            // authenticates nothing, so it needs neither the target's API nor a
+            // copy of the access keys.
             let relay_json = serde_json::json!({
                 "mode": "relay",
                 "listen": listen,
                 "upstream_tcp": upstream,
                 "upstream_udp": upstream,
-                "upstream_api_url": api_url,
-                "upstream_api_token": api_token,
-                "sync_interval_secs": 30,
                 "debug": false
             });
 
@@ -920,6 +924,38 @@ async fn run_app() -> Result<()> {
         match cmd {
             Commands::Setup { init } => { args.setup = true; args.init = init; }
             Commands::Init { mode } => { args.init = Some(mode); }
+            Commands::HashPassword { password } => {
+                // The panel stores only a hash, and until now nothing in the CLI
+                // could produce one: `ostp init server` writes password_hash: ""
+                // and the only generator lived inside the Unix-only Server+Panel
+                // wizard branch, leaving no supported way to set up API auth on a
+                // plain server.
+                let password = match password {
+                    Some(p) => p,
+                    None => {
+                        print!("Password: ");
+                        use std::io::Write as _;
+                        std::io::stdout().flush().ok();
+                        let mut buf = String::new();
+                        std::io::stdin().read_line(&mut buf)?;
+                        buf.trim_end_matches(['\r', '\n']).to_string()
+                    }
+                };
+                if password.is_empty() {
+                    anyhow::bail!("password must not be empty");
+                }
+                // Must match api.rs's handle_login byte for byte.
+                let hash = format!(
+                    "{:x}",
+                    <sha2::Sha256 as sha2::Digest>::digest(password.as_bytes())
+                );
+                println!();
+                println!("Add this to the \"api\" section of your config:");
+                println!();
+                println!("  \"password_hash\": \"{hash}\"");
+                println!();
+                return Ok(());
+            }
             Commands::GenerateKey { format, count } => { args.generate_key = true; args.format = format; args.count = count; }
             Commands::Links => { args.links = true; }
             Commands::Check => { args.check = true; }
@@ -1099,7 +1135,9 @@ async fn run_app() -> Result<()> {
                         println!("  Listen: {:?}", r.listen.primary().cyan());
                         println!("  Upstream TCP: {}", r.upstream_tcp.cyan());
                         println!("  Upstream UDP: {}", r.upstream_udp.cyan());
-                        println!("  API sync: {}", r.upstream_api_url.yellow());
+                        if !r.upstream_api_url.is_empty() {
+                            println!("  {}", "upstream_api_url is set but no longer used - safe to remove".yellow());
+                        }
                     }
                 }
             }
@@ -1175,9 +1213,9 @@ async fn run_app() -> Result<()> {
   "listen": "0.0.0.0:50000",
   "upstream_tcp": "TARGET_SERVER_IP:50000",
   "upstream_udp": "TARGET_SERVER_IP:50000",
-  "upstream_api_url": "http://TARGET_SERVER_IP:9090",
-  "upstream_api_token": "YOUR_API_TOKEN_HERE",
-  "sync_interval_secs": 30,
+  // The relay forwards transparently and holds no keys: sessions are
+  // authenticated end-to-end by the target server, which drops anything that
+  // fails. Nothing else needs configuring here.
   "debug": false
 }"#.to_string()
         } else {
@@ -1376,14 +1414,18 @@ async fn run_app() -> Result<()> {
             println!("{} Starting relay node on {:?}", "[ostp]".cyan().bold(), listen_addrs);
             println!("{} Upstream TCP: {}", "[ostp]".cyan().bold(), relay_cfg.upstream_tcp);
             println!("{} Upstream UDP: {}", "[ostp]".cyan().bold(), relay_cfg.upstream_udp);
-            println!("{} Key sync API: {}", "[ostp]".cyan().bold(), relay_cfg.upstream_api_url);
+            if !relay_cfg.upstream_api_url.is_empty() {
+                println!(
+                    "{} Note: upstream_api_url is no longer used and can be removed. The relay \
+                     forwards transparently; sessions are authenticated end-to-end by the target \
+                     server.",
+                    "[ostp]".yellow().bold()
+                );
+            }
             let relay_config = ostp_server::RelayConfig {
                 listen_addrs,
                 upstream_tcp: relay_cfg.upstream_tcp,
                 upstream_udp: relay_cfg.upstream_udp,
-                upstream_api_url: relay_cfg.upstream_api_url,
-                upstream_api_token: relay_cfg.upstream_api_token,
-                sync_interval_secs: relay_cfg.sync_interval_secs,
             };
             ostp_server::relay_node::run_relay_node(relay_config).await?;
         }