v25.12.0

changelog.md

@@ -1,5 +1,148 @@
 # Changelog
 
+## 2026-03-19 - 25.12.0 - feat(proxy-protocol)
+add PROXY protocol v2 support to the Rust passthrough listener and streamline TypeScript proxy protocol exports
+
+- detect and parse PROXY protocol v2 headers in the Rust TCP listener, including TCP and UDP address families
+- add Rust v2 header generation, incomplete-header handling, and broader parser test coverage
+- remove deprecated TypeScript proxy protocol parser exports and tests, leaving shared type definitions only
+
+## 2026-03-17 - 25.11.24 - fix(rustproxy-http)
+improve async static file serving, websocket handshake buffering, and shared metric metadata handling
+
+- convert static file serving to async filesystem operations and await directory/file checks
+- preserve and forward bytes read past the WebSocket handshake header terminator to avoid dropping buffered upstream data
+- reuse Arc<str> values for route and source identifiers across counting bodies and metric reporting
+- standardize backend key propagation across H1/H2 forwarding, retry, and fallback paths for consistent logging and metrics
+
+## 2026-03-17 - 25.11.23 - fix(rustproxy-http,rustproxy-metrics)
+reduce per-frame metrics overhead by batching body byte accounting
+
+- Buffer HTTP body byte counts and flush them every 64 KB, at end of stream, and on drop to keep totals accurate while preserving throughput sampling.
+- Skip zero-value counter updates in metrics collection to avoid unnecessary atomic and DashMap operations for the unused direction.
+
+## 2026-03-17 - 25.11.22 - fix(rustproxy-http)
+reuse healthy HTTP/2 upstream connections after requests with bodies
+
+- Registers successful HTTP/2 connections in the pool regardless of whether the proxied request included a body
+- Continues to avoid pooling upstream connections that returned 502 Bad Gateway responses
+
+## 2026-03-17 - 25.11.21 - fix(rustproxy-http)
+reuse pooled HTTP/2 connections for requests with and without bodies
+
+- remove the bodyless-request restriction from HTTP/2 pool checkout
+- always return successful HTTP/2 senders to the connection pool after requests
+
+## 2026-03-17 - 25.11.20 - fix(rustproxy-http)
+avoid downgrading cached backend protocol on H2 stream errors
+
+- Treat HTTP/2 stream-level failures as retryable request errors instead of evidence that the backend only supports HTTP/1.1
+- Keep protocol cache entries unchanged after successful H2 handshakes so future requests continue using HTTP/2
+- Lower log severity for this fallback path from warning to debug while still recording backend H2 failure metrics
+
+## 2026-03-16 - 25.11.19 - fix(rustproxy-http)
+avoid reusing pooled HTTP/2 connections for requests with bodies to prevent upload flow-control stalls
+
+- Limit HTTP/2 pool checkout to bodyless requests such as GET, HEAD, and DELETE
+- Skip re-registering HTTP/2 connections in the pool after requests that send a body
+- Prevent stalled uploads caused by depleted connection-level flow control windows on reused HTTP/2 connections
+
+## 2026-03-16 - 25.11.18 - fix(repo)
+no changes to commit
+
+
+## 2026-03-16 - 25.11.17 - fix(rustproxy-http)
+prevent stale HTTP/2 connection drivers from evicting newer pooled connections
+
+- add generation IDs to pooled HTTP/2 senders so pool removal only affects the matching connection
+- update HTTP/2 proxy and retry paths to register generation-tagged connections and skip eviction before registration completes
+
+## 2026-03-16 - 25.11.16 - fix(repo)
+no changes to commit
+
+
+## 2026-03-16 - 25.11.15 - fix(rustproxy-http)
+implement vectored write support for backend streams
+
+- Add poll_write_vectored forwarding for both plain and TLS backend stream variants
+- Expose is_write_vectored so the proxy can correctly report vectored write capability
+
+## 2026-03-16 - 25.11.14 - fix(rustproxy-http)
+forward vectored write support in ShutdownOnDrop AsyncWrite wrapper
+
+- Implements poll_write_vectored by delegating to the wrapped writer
+- Exposes is_write_vectored so the wrapper preserves underlying AsyncWrite capabilities
+
+## 2026-03-16 - 25.11.13 - fix(rustproxy-http)
+remove hot-path debug logging from HTTP/1 connection pool hits
+
+- Stops emitting debug logs when reusing HTTP/1 idle connections in the connection pool.
+- Keeps pool hit behavior unchanged while reducing overhead on a frequently executed path.
+
+## 2026-03-16 - 25.11.12 - fix(rustproxy-http)
+remove connection pool hit logging and keep logging limited to actual failures
+
+- Removes debug and warning logs for HTTP/2 connection pool hits and age checks.
+- Keeps pool behavior unchanged while reducing noisy per-request logging in the Rust HTTP proxy layer.
+
+## 2026-03-16 - 25.11.11 - fix(rustproxy-http)
+improve HTTP/2 proxy error logging with warning-level connection failures and debug error details
+
+- Adds debug-formatted error fields to HTTP/2 handshake, retry, fallback, and request failure logs
+- Promotes upstream HTTP/2 connection error logs from debug to warn to improve operational visibility
+
+## 2026-03-16 - 25.11.10 - fix(rustproxy-http)
+validate pooled HTTP/2 connections asynchronously before reuse and evict stale senders
+
+- Add an async ready() check with a 500ms timeout before reusing pooled HTTP/2 senders to catch GOAWAY/RST states before forwarding requests
+- Return connection age from the HTTP/2 pool checkout path and log warnings for older pooled connections
+- Evict pooled HTTP/2 senders when they are closed, exceed max age, fail readiness validation, or time out during readiness checks
+
+## 2026-03-16 - 25.11.9 - fix(rustproxy-routing)
+reduce hot-path allocations in routing, metrics, and proxy protocol handling
+
+- skip HTTP header map construction unless a route on the current port uses header matching
+- reuse computed client IP strings during HTTP route matching to avoid redundant allocations
+- optimize per-route and per-IP metric updates with get-first lookups to avoid unnecessary String creation on existing entries
+- replace heap-allocated PROXY protocol peek and discard buffers with stack-allocated buffers in the TCP listener
+- improve domain matcher case-insensitive wildcard checks while preserving glob fallback behavior
+
+## 2026-03-16 - 25.11.8 - fix(rustproxy-http)
+prevent premature idle timeouts during streamed HTTP responses and ensure TLS close_notify is sent on dropped connections
+
+- track active streaming response bodies so the HTTP idle watchdog does not close connections mid-transfer
+- add a ShutdownOnDrop wrapper for TLS-terminated HTTP connections to send shutdown on drop and avoid improperly terminated TLS sessions
+- apply the shutdown wrapper in passthrough TLS terminate and terminate+reencrypt HTTP handling
+
+## 2026-03-16 - 25.11.7 - fix(rustproxy)
+prevent TLS route reload certificate mismatches and tighten passthrough connection handling
+
+- Load updated TLS configs before swapping the route manager so newly visible routes always have their certificates available.
+- Add timeouts when peeking initial decrypted data after TLS handshake to avoid leaked idle connections.
+- Raise dropped, blocked, unmatched, and errored passthrough connection events from debug to warn for better operational visibility.
+
+## 2026-03-16 - 25.11.6 - fix(rustproxy-http,rustproxy-passthrough)
+improve upstream connection cleanup and graceful tunnel shutdown
+
+- Evict pooled HTTP/2 connections when their driver exits and shorten the maximum pooled H2 age to reduce reuse of stale upstream connections.
+- Strip hop-by-hop headers from backend responses before forwarding to HTTP/2 clients to avoid invalid H2 response handling.
+- Replace immediate task aborts in WebSocket and TCP tunnel watchdogs with cancellation-driven graceful shutdown plus timed fallback aborts.
+- Use non-blocking semaphore acquisition in the TCP listener so connection limits do not stall the accept loop for the entire port.
+
+## 2026-03-16 - 25.11.5 - fix(repo)
+no changes to commit
+
+
+## 2026-03-15 - 25.11.4 - fix(rustproxy-http)
+report streamed HTTP and WebSocket bytes per chunk for real-time throughput metrics
+
+- Update CountingBody to record bytes immediately on each data frame instead of aggregating until completion or drop
+- Record WebSocket tunnel traffic inside both copy loops and remove the final aggregate byte report to keep throughput metrics current
+
+## 2026-03-15 - 25.11.3 - fix(repo)
+no changes to commit
+
+
 ## 2026-03-15 - 25.11.2 - fix(rustproxy-http)
 avoid reusing HTTP/1 senders during streaming responses and relax HTTP/2 keep-alive timeouts
 

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@push.rocks/smartproxy",
-  "version": "25.11.2",
+  "version": "25.12.0",
   "private": false,
   "description": "A powerful proxy package with unified route-based configuration for high traffic management. Features include SSL/TLS support, flexible routing patterns, WebSocket handling, advanced security options, and automatic ACME certificate management.",
   "main": "dist_ts/index.js",

rust/crates/rustproxy-http/src/connection_pool.rs

@@ -4,13 +4,13 @@
 //! HTTP/2 connections are multiplexed (clone the sender for each request).
 
 use std::sync::Arc;
+use std::sync::atomic::{AtomicU64, Ordering};
 use std::time::{Duration, Instant};
 
 use bytes::Bytes;
 use dashmap::DashMap;
 use http_body_util::combinators::BoxBody;
 use hyper::client::conn::{http1, http2};
-use tracing::debug;
 
 /// Maximum idle connections per backend key.
 const MAX_IDLE_PER_KEY: usize = 16;
@@ -20,7 +20,8 @@ const IDLE_TIMEOUT: Duration = Duration::from_secs(90);
 const EVICTION_INTERVAL: Duration = Duration::from_secs(30);
 /// Maximum age for pooled HTTP/2 connections before proactive eviction.
 /// Prevents staleness from backends that close idle connections (e.g. nginx GOAWAY).
-const MAX_H2_AGE: Duration = Duration::from_secs(300);
+/// 120s is well within typical server GOAWAY windows (nginx: ~60s idle, envoy: ~60s).
+const MAX_H2_AGE: Duration = Duration::from_secs(120);
 
 /// Identifies a unique backend endpoint.
 #[derive(Clone, Debug, Hash, Eq, PartialEq)]
@@ -37,10 +38,13 @@ struct IdleH1 {
     idle_since: Instant,
 }
 
-/// A pooled HTTP/2 sender (multiplexed, Clone-able).
+/// A pooled HTTP/2 sender (multiplexed, Clone-able) with a generation tag.
 struct PooledH2 {
     sender: http2::SendRequest<BoxBody<Bytes, hyper::Error>>,
     created_at: Instant,
+    /// Unique generation ID. Connection drivers use this to only remove their OWN
+    /// entry, preventing phantom eviction when multiple connections share the same key.
+    generation: u64,
 }
 
 /// Backend connection pool.
@@ -49,6 +53,8 @@ pub struct ConnectionPool {
     h1_pool: Arc<DashMap<PoolKey, Vec<IdleH1>>>,
     /// HTTP/2 multiplexed connections indexed by backend key.
     h2_pool: Arc<DashMap<PoolKey, PooledH2>>,
+    /// Monotonic generation counter for H2 pool entries.
+    h2_generation: AtomicU64,
     /// Handle for the background eviction task.
     eviction_handle: Option<tokio::task::JoinHandle<()>>,
 }
@@ -68,6 +74,7 @@ impl ConnectionPool {
         Self {
             h1_pool,
             h2_pool,
+            h2_generation: AtomicU64::new(0),
             eviction_handle: Some(eviction_handle),
         }
     }
@@ -81,7 +88,7 @@ impl ConnectionPool {
         while let Some(idle) = idles.pop() {
             // Check if the connection is still alive and ready
             if idle.idle_since.elapsed() < IDLE_TIMEOUT && idle.sender.is_ready() && !idle.sender.is_closed() {
-                debug!("Pool hit (h1): {}:{}", key.host, key.port);
+                // H1 pool hit — no logging on hot path
                 return Some(idle.sender);
             }
             // Stale or closed — drop it
@@ -114,40 +121,56 @@ impl ConnectionPool {
 
     /// Try to get a cloned HTTP/2 sender for the given key.
     /// HTTP/2 senders are Clone-able (multiplexed), so we clone rather than remove.
-    pub fn checkout_h2(&self, key: &PoolKey) -> Option<http2::SendRequest<BoxBody<Bytes, hyper::Error>>> {
+    pub fn checkout_h2(&self, key: &PoolKey) -> Option<(http2::SendRequest<BoxBody<Bytes, hyper::Error>>, Duration)> {
         let entry = self.h2_pool.get(key)?;
         let pooled = entry.value();
+        let age = pooled.created_at.elapsed();
 
-        // Check if the h2 connection is still alive and not too old
-        if pooled.sender.is_closed() || pooled.created_at.elapsed() >= MAX_H2_AGE {
+        if pooled.sender.is_closed() || age >= MAX_H2_AGE {
             drop(entry);
             self.h2_pool.remove(key);
             return None;
         }
 
         if pooled.sender.is_ready() {
-            debug!("Pool hit (h2): {}:{}", key.host, key.port);
-            return Some(pooled.sender.clone());
+            return Some((pooled.sender.clone(), age));
         }
         None
     }
 
-    /// Remove a dead HTTP/2 sender from the pool.
+    /// Remove a dead HTTP/2 sender from the pool (unconditional).
     /// Called when `send_request` fails to prevent subsequent requests from reusing the stale sender.
     pub fn remove_h2(&self, key: &PoolKey) {
         self.h2_pool.remove(key);
     }
 
-    /// Register an HTTP/2 sender in the pool. Since h2 is multiplexed,
-    /// only one sender per key is stored (it's Clone-able).
-    pub fn register_h2(&self, key: PoolKey, sender: http2::SendRequest<BoxBody<Bytes, hyper::Error>>) {
+    /// Remove an HTTP/2 sender ONLY if the current entry has the expected generation.
+    /// This prevents phantom eviction: when multiple connections share the same key,
+    /// an old connection's driver won't accidentally remove a newer connection's entry.
+    pub fn remove_h2_if_generation(&self, key: &PoolKey, expected_gen: u64) {
+        if let Some(entry) = self.h2_pool.get(key) {
+            if entry.value().generation == expected_gen {
+                drop(entry); // release DashMap ref before remove
+                self.h2_pool.remove(key);
+            }
+            // else: a newer connection replaced ours — don't touch it
+        }
+    }
+
+    /// Register an HTTP/2 sender in the pool. Returns the generation ID for this entry.
+    /// The caller should pass this generation to the connection driver so it can use
+    /// `remove_h2_if_generation` instead of `remove_h2` to avoid phantom eviction.
+    pub fn register_h2(&self, key: PoolKey, sender: http2::SendRequest<BoxBody<Bytes, hyper::Error>>) -> u64 {
+        let gen = self.h2_generation.fetch_add(1, Ordering::Relaxed);
         if sender.is_closed() {
-            return;
+            return gen;
         }
         self.h2_pool.insert(key, PooledH2 {
             sender,
             created_at: Instant::now(),
+            generation: gen,
         });
+        gen
     }
 
     /// Background eviction loop — runs every EVICTION_INTERVAL to remove stale connections.

rust/crates/rustproxy-http/src/counting_body.rs

@@ -9,27 +9,37 @@ use bytes::Bytes;
 use http_body::Frame;
 use rustproxy_metrics::MetricsCollector;
 
+/// Flush accumulated bytes to the metrics collector every 64 KB.
+/// This reduces per-frame DashMap shard-locked reads from ~15 to ~1 per 4 frames
+/// (assuming typical 16 KB upload frames).  The 1 Hz throughput sampler still sees
+/// data within one sampling period even at low transfer rates.
+const BYTE_FLUSH_THRESHOLD: u64 = 65_536;
+
 /// Wraps any `http_body::Body` and counts data bytes passing through.
 ///
-/// When the body is fully consumed or dropped, accumulated byte counts
-/// are reported to the `MetricsCollector`.
+/// Bytes are accumulated and flushed to the `MetricsCollector` every
+/// [`BYTE_FLUSH_THRESHOLD`] bytes (and on Drop) so the throughput tracker
+/// (sampled at 1 Hz) reflects real-time data flow without per-frame overhead.
 ///
 /// The inner body is pinned on the heap to support `!Unpin` types like `hyper::body::Incoming`.
 pub struct CountingBody<B> {
     inner: Pin<Box<B>>,
-    counted_bytes: AtomicU64,
     metrics: Arc<MetricsCollector>,
-    route_id: Option<String>,
-    source_ip: Option<String>,
+    route_id: Option<Arc<str>>,
+    source_ip: Option<Arc<str>>,
     /// Whether we count bytes as "in" (request body) or "out" (response body).
     direction: Direction,
-    /// Whether we've already reported the bytes (to avoid double-reporting on drop).
-    reported: bool,
+    /// Accumulated bytes not yet flushed to the metrics collector.
+    pending_bytes: u64,
     /// Optional connection-level activity tracker. When set, poll_frame updates this
     /// to keep the idle watchdog alive during active body streaming (uploads/downloads).
     connection_activity: Option<Arc<AtomicU64>>,
     /// Start instant for computing elapsed ms for connection_activity.
     activity_start: Option<std::time::Instant>,
+    /// Optional active-request counter. When set, CountingBody increments on creation
+    /// and decrements on Drop, keeping the HTTP idle watchdog aware that a response
+    /// body is still streaming (even after the request handler has returned).
+    active_requests: Option<Arc<AtomicU64>>,
 }
 
 /// Which direction the bytes flow.
@@ -46,20 +56,20 @@ impl<B> CountingBody<B> {
     pub fn new(
         inner: B,
         metrics: Arc<MetricsCollector>,
-        route_id: Option<String>,
-        source_ip: Option<String>,
+        route_id: Option<Arc<str>>,
+        source_ip: Option<Arc<str>>,
         direction: Direction,
     ) -> Self {
         Self {
             inner: Box::pin(inner),
-            counted_bytes: AtomicU64::new(0),
             metrics,
             route_id,
             source_ip,
             direction,
-            reported: false,
+            pending_bytes: 0,
             connection_activity: None,
             activity_start: None,
+            active_requests: None,
         }
     }
 
@@ -72,18 +82,23 @@ impl<B> CountingBody<B> {
         self
     }
 
-    /// Report accumulated bytes to the metrics collector.
-    fn report(&mut self) {
-        if self.reported {
+    /// Set the active-request counter for the HTTP idle watchdog.
+    /// CountingBody increments on creation and decrements on Drop, ensuring the
+    /// idle watchdog sees an "active request" while the response body streams.
+    pub fn with_active_requests(mut self, counter: Arc<AtomicU64>) -> Self {
+        counter.fetch_add(1, Ordering::Relaxed);
+        self.active_requests = Some(counter);
+        self
+    }
+
+    /// Flush accumulated bytes to the metrics collector.
+    #[inline]
+    fn flush_pending(&mut self) {
+        if self.pending_bytes == 0 {
             return;
         }
-        self.reported = true;
-
-        let bytes = self.counted_bytes.load(Ordering::Relaxed);
-        if bytes == 0 {
-            return;
-        }
-
+        let bytes = self.pending_bytes;
+        self.pending_bytes = 0;
         let route_id = self.route_id.as_deref();
         let source_ip = self.source_ip.as_deref();
         match self.direction {
@@ -93,12 +108,6 @@ impl<B> CountingBody<B> {
     }
 }
 
-impl<B> Drop for CountingBody<B> {
-    fn drop(&mut self) {
-        self.report();
-    }
-}
-
 // CountingBody is Unpin because inner is Pin<Box<B>> (always Unpin).
 impl<B> Unpin for CountingBody<B> {}
 
@@ -118,8 +127,13 @@ where
         match this.inner.as_mut().poll_frame(cx) {
             Poll::Ready(Some(Ok(frame))) => {
                 if let Some(data) = frame.data_ref() {
-                    this.counted_bytes.fetch_add(data.len() as u64, Ordering::Relaxed);
-                    // Keep the connection-level idle watchdog alive during body streaming
+                    let len = data.len() as u64;
+                    this.pending_bytes += len;
+                    if this.pending_bytes >= BYTE_FLUSH_THRESHOLD {
+                        this.flush_pending();
+                    }
+                    // Keep the connection-level idle watchdog alive on every frame
+                    // (this is just one atomic store — cheap enough per-frame)
                     if let (Some(activity), Some(start)) = (&this.connection_activity, &this.activity_start) {
                         activity.store(start.elapsed().as_millis() as u64, Ordering::Relaxed);
                     }
@@ -128,8 +142,8 @@ where
             }
             Poll::Ready(Some(Err(e))) => Poll::Ready(Some(Err(e))),
             Poll::Ready(None) => {
-                // Body is fully consumed — report now
-                this.report();
+                // End of stream — flush any remaining bytes
+                this.flush_pending();
                 Poll::Ready(None)
             }
             Poll::Pending => Poll::Pending,
@@ -144,3 +158,15 @@ where
         self.inner.size_hint()
     }
 }
+
+impl<B> Drop for CountingBody<B> {
+    fn drop(&mut self) {
+        // Flush any remaining accumulated bytes so totals stay accurate
+        self.flush_pending();
+        // Decrement the active-request counter so the HTTP idle watchdog
+        // knows this response body is no longer streaming.
+        if let Some(ref counter) = self.active_requests {
+            counter.fetch_sub(1, Ordering::Relaxed);
+        }
+    }
+}

rust/crates/rustproxy-http/src/lib.rs

@@ -9,6 +9,7 @@ pub mod protocol_cache;
 pub mod proxy_service;
 pub mod request_filter;
 pub mod response_filter;
+pub mod shutdown_on_drop;
 pub mod template;
 pub mod upstream_selector;
 

rust/crates/rustproxy-http/src/shutdown_on_drop.rs

@@ -0,0 +1,102 @@
+//! Wrapper that ensures TLS close_notify is sent when the stream is dropped.
+//!
+//! When hyper drops an HTTP connection (backend error, timeout, normal H2 close),
+//! the underlying TLS stream is dropped WITHOUT `shutdown()`. tokio-rustls cannot
+//! send `close_notify` in Drop (requires async). This wrapper tracks whether
+//! `poll_shutdown` was called and, if not, spawns a background task to send it.
+
+use std::io;
+use std::pin::Pin;
+use std::task::{Context, Poll};
+
+use tokio::io::{AsyncRead, AsyncWrite, ReadBuf};
+
+/// Wraps an AsyncRead+AsyncWrite stream and ensures `shutdown()` is called when
+/// dropped, even if the caller (e.g. hyper) doesn't explicitly shut down.
+///
+/// This guarantees TLS `close_notify` is sent for TLS-wrapped streams, preventing
+/// "GnuTLS recv error (-110): The TLS connection was non-properly terminated" errors.
+pub struct ShutdownOnDrop<S: AsyncRead + AsyncWrite + Unpin + Send + 'static> {
+    inner: Option<S>,
+    shutdown_called: bool,
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin + Send + 'static> ShutdownOnDrop<S> {
+    /// Create a new wrapper around the given stream.
+    pub fn new(stream: S) -> Self {
+        Self {
+            inner: Some(stream),
+            shutdown_called: false,
+        }
+    }
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin + Send + 'static> AsyncRead for ShutdownOnDrop<S> {
+    fn poll_read(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &mut ReadBuf<'_>,
+    ) -> Poll<io::Result<()>> {
+        Pin::new(self.get_mut().inner.as_mut().unwrap()).poll_read(cx, buf)
+    }
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin + Send + 'static> AsyncWrite for ShutdownOnDrop<S> {
+    fn poll_write(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &[u8],
+    ) -> Poll<io::Result<usize>> {
+        Pin::new(self.get_mut().inner.as_mut().unwrap()).poll_write(cx, buf)
+    }
+
+    fn poll_write_vectored(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        bufs: &[io::IoSlice<'_>],
+    ) -> Poll<io::Result<usize>> {
+        Pin::new(self.get_mut().inner.as_mut().unwrap()).poll_write_vectored(cx, bufs)
+    }
+
+    fn is_write_vectored(&self) -> bool {
+        self.inner.as_ref().unwrap().is_write_vectored()
+    }
+
+    fn poll_flush(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+    ) -> Poll<io::Result<()>> {
+        Pin::new(self.get_mut().inner.as_mut().unwrap()).poll_flush(cx)
+    }
+
+    fn poll_shutdown(
+        self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+    ) -> Poll<io::Result<()>> {
+        let this = self.get_mut();
+        let result = Pin::new(this.inner.as_mut().unwrap()).poll_shutdown(cx);
+        if result.is_ready() {
+            this.shutdown_called = true;
+        }
+        result
+    }
+}
+
+impl<S: AsyncRead + AsyncWrite + Unpin + Send + 'static> Drop for ShutdownOnDrop<S> {
+    fn drop(&mut self) {
+        // If shutdown was already called (hyper closed properly), nothing to do.
+        // If not (hyper dropped without shutdown — e.g. H2 close, error, timeout),
+        // spawn a background task to send close_notify / TCP FIN.
+        if !self.shutdown_called {
+            if let Some(mut stream) = self.inner.take() {
+                tokio::spawn(async move {
+                    let _ = tokio::time::timeout(
+                        std::time::Duration::from_secs(2),
+                        tokio::io::AsyncWriteExt::shutdown(&mut stream),
+                    ).await;
+                    // stream is dropped here — all resources freed
+                });
+            }
+        }
+    }
+}

rust/crates/rustproxy-metrics/src/collector.rs

@@ -259,51 +259,87 @@ impl MetricsCollector {
     /// Called per-chunk in the TCP copy loop. Only touches AtomicU64 counters —
     /// no Mutex is taken. The throughput trackers are fed during `sample_all()`.
     pub fn record_bytes(&self, bytes_in: u64, bytes_out: u64, route_id: Option<&str>, source_ip: Option<&str>) {
-        self.total_bytes_in.fetch_add(bytes_in, Ordering::Relaxed);
-        self.total_bytes_out.fetch_add(bytes_out, Ordering::Relaxed);
-
-        // Accumulate into lock-free pending throughput counters
-        self.global_pending_tp_in.fetch_add(bytes_in, Ordering::Relaxed);
-        self.global_pending_tp_out.fetch_add(bytes_out, Ordering::Relaxed);
-
-        if let Some(route_id) = route_id {
-            self.route_bytes_in
-                .entry(route_id.to_string())
-                .or_insert_with(|| AtomicU64::new(0))
-                .fetch_add(bytes_in, Ordering::Relaxed);
-            self.route_bytes_out
-                .entry(route_id.to_string())
-                .or_insert_with(|| AtomicU64::new(0))
-                .fetch_add(bytes_out, Ordering::Relaxed);
-
-            // Accumulate into per-route pending throughput counters (lock-free)
-            let entry = self.route_pending_tp
-                .entry(route_id.to_string())
-                .or_insert_with(|| (AtomicU64::new(0), AtomicU64::new(0)));
-            entry.0.fetch_add(bytes_in, Ordering::Relaxed);
-            entry.1.fetch_add(bytes_out, Ordering::Relaxed);
+        // Short-circuit: only touch counters for the direction that has data.
+        // CountingBody always calls with one direction zero — skipping the zero
+        // direction avoids ~50% of DashMap shard-locked reads per call.
+        if bytes_in > 0 {
+            self.total_bytes_in.fetch_add(bytes_in, Ordering::Relaxed);
+            self.global_pending_tp_in.fetch_add(bytes_in, Ordering::Relaxed);
+        }
+        if bytes_out > 0 {
+            self.total_bytes_out.fetch_add(bytes_out, Ordering::Relaxed);
+            self.global_pending_tp_out.fetch_add(bytes_out, Ordering::Relaxed);
         }
 
+        // Per-route tracking: use get() first (zero-alloc fast path for existing entries),
+        // fall back to entry() with to_string() only on the rare first-chunk miss.
+        if let Some(route_id) = route_id {
+            if bytes_in > 0 {
+                if let Some(counter) = self.route_bytes_in.get(route_id) {
+                    counter.fetch_add(bytes_in, Ordering::Relaxed);
+                } else {
+                    self.route_bytes_in.entry(route_id.to_string())
+                        .or_insert_with(|| AtomicU64::new(0))
+                        .fetch_add(bytes_in, Ordering::Relaxed);
+                }
+            }
+            if bytes_out > 0 {
+                if let Some(counter) = self.route_bytes_out.get(route_id) {
+                    counter.fetch_add(bytes_out, Ordering::Relaxed);
+                } else {
+                    self.route_bytes_out.entry(route_id.to_string())
+                        .or_insert_with(|| AtomicU64::new(0))
+                        .fetch_add(bytes_out, Ordering::Relaxed);
+                }
+            }
+
+            // Accumulate into per-route pending throughput counters (lock-free)
+            if let Some(entry) = self.route_pending_tp.get(route_id) {
+                if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
+            } else {
+                let entry = self.route_pending_tp.entry(route_id.to_string())
+                    .or_insert_with(|| (AtomicU64::new(0), AtomicU64::new(0)));
+                if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
+            }
+        }
+
+        // Per-IP tracking: same get()-first pattern to avoid String allocation on hot path.
         if let Some(ip) = source_ip {
             // Only record per-IP stats if the IP still has active connections.
             // This prevents orphaned entries when record_bytes races with
             // connection_closed (which evicts all per-IP data on last close).
             if self.ip_connections.contains_key(ip) {
-                self.ip_bytes_in
-                    .entry(ip.to_string())
-                    .or_insert_with(|| AtomicU64::new(0))
-                    .fetch_add(bytes_in, Ordering::Relaxed);
-                self.ip_bytes_out
-                    .entry(ip.to_string())
-                    .or_insert_with(|| AtomicU64::new(0))
-                    .fetch_add(bytes_out, Ordering::Relaxed);
+                if bytes_in > 0 {
+                    if let Some(counter) = self.ip_bytes_in.get(ip) {
+                        counter.fetch_add(bytes_in, Ordering::Relaxed);
+                    } else {
+                        self.ip_bytes_in.entry(ip.to_string())
+                            .or_insert_with(|| AtomicU64::new(0))
+                            .fetch_add(bytes_in, Ordering::Relaxed);
+                    }
+                }
+                if bytes_out > 0 {
+                    if let Some(counter) = self.ip_bytes_out.get(ip) {
+                        counter.fetch_add(bytes_out, Ordering::Relaxed);
+                    } else {
+                        self.ip_bytes_out.entry(ip.to_string())
+                            .or_insert_with(|| AtomicU64::new(0))
+                            .fetch_add(bytes_out, Ordering::Relaxed);
+                    }
+                }
 
                 // Accumulate into per-IP pending throughput counters (lock-free)
-                let entry = self.ip_pending_tp
-                    .entry(ip.to_string())
-                    .or_insert_with(|| (AtomicU64::new(0), AtomicU64::new(0)));
-                entry.0.fetch_add(bytes_in, Ordering::Relaxed);
-                entry.1.fetch_add(bytes_out, Ordering::Relaxed);
+                if let Some(entry) = self.ip_pending_tp.get(ip) {
+                    if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                    if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
+                } else {
+                    let entry = self.ip_pending_tp.entry(ip.to_string())
+                        .or_insert_with(|| (AtomicU64::new(0), AtomicU64::new(0)));
+                    if bytes_in > 0 { entry.0.fetch_add(bytes_in, Ordering::Relaxed); }
+                    if bytes_out > 0 { entry.1.fetch_add(bytes_out, Ordering::Relaxed); }
+                }
             }
         }
     }

rust/crates/rustproxy-passthrough/src/forwarder.rs

@@ -97,16 +97,25 @@ pub async fn forward_bidirectional_with_timeouts(
     let last_activity = Arc::new(AtomicU64::new(0));
     let start = std::time::Instant::now();
 
+    // Per-connection cancellation token: the watchdog cancels this instead of
+    // aborting tasks, so the copy loops can shut down gracefully (TCP FIN instead
+    // of RST, TLS close_notify if the stream is TLS-wrapped).
+    let conn_cancel = CancellationToken::new();
+
     let la1 = Arc::clone(&last_activity);
     let initial_len = initial_data.map_or(0u64, |d| d.len() as u64);
     let metrics_c2b = metrics.clone();
+    let cc1 = conn_cancel.clone();
     let c2b = tokio::spawn(async move {
         let mut buf = vec![0u8; 65536];
         let mut total = initial_len;
         loop {
-            let n = match client_read.read(&mut buf).await {
-                Ok(0) | Err(_) => break,
-                Ok(n) => n,
+            let n = tokio::select! {
+                result = client_read.read(&mut buf) => match result {
+                    Ok(0) | Err(_) => break,
+                    Ok(n) => n,
+                },
+                _ = cc1.cancelled() => break,
             };
             if backend_write.write_all(&buf[..n]).await.is_err() {
                 break;
@@ -117,19 +126,27 @@ pub async fn forward_bidirectional_with_timeouts(
                 ctx.collector.record_bytes(n as u64, 0, ctx.route_id.as_deref(), ctx.source_ip.as_deref());
             }
         }
-        let _ = backend_write.shutdown().await;
+        // Graceful shutdown with timeout (sends TCP FIN / TLS close_notify)
+        let _ = tokio::time::timeout(
+            std::time::Duration::from_secs(2),
+            backend_write.shutdown(),
+        ).await;
         total
     });
 
     let la2 = Arc::clone(&last_activity);
     let metrics_b2c = metrics;
+    let cc2 = conn_cancel.clone();
     let b2c = tokio::spawn(async move {
         let mut buf = vec![0u8; 65536];
         let mut total = 0u64;
         loop {
-            let n = match backend_read.read(&mut buf).await {
-                Ok(0) | Err(_) => break,
-                Ok(n) => n,
+            let n = tokio::select! {
+                result = backend_read.read(&mut buf) => match result {
+                    Ok(0) | Err(_) => break,
+                    Ok(n) => n,
+                },
+                _ = cc2.cancelled() => break,
             };
             if client_write.write_all(&buf[..n]).await.is_err() {
                 break;
@@ -140,14 +157,20 @@ pub async fn forward_bidirectional_with_timeouts(
                 ctx.collector.record_bytes(0, n as u64, ctx.route_id.as_deref(), ctx.source_ip.as_deref());
             }
         }
-        let _ = client_write.shutdown().await;
+        // Graceful shutdown with timeout (sends TCP FIN / TLS close_notify)
+        let _ = tokio::time::timeout(
+            std::time::Duration::from_secs(2),
+            client_write.shutdown(),
+        ).await;
         total
     });
 
-    // Watchdog: inactivity, max lifetime, and cancellation
+    // Watchdog: inactivity, max lifetime, and cancellation.
+    // First cancels the per-connection token for graceful shutdown (FIN/close_notify),
+    // then falls back to abort if the tasks are stuck (e.g. on a blocked write_all).
     let la_watch = Arc::clone(&last_activity);
-    let c2b_handle = c2b.abort_handle();
-    let b2c_handle = b2c.abort_handle();
+    let c2b_abort = c2b.abort_handle();
+    let b2c_abort = b2c.abort_handle();
     let watchdog = tokio::spawn(async move {
         let check_interval = std::time::Duration::from_secs(5);
         let mut last_seen = 0u64;
@@ -155,16 +178,12 @@ pub async fn forward_bidirectional_with_timeouts(
             tokio::select! {
                 _ = cancel.cancelled() => {
                     debug!("Connection cancelled by shutdown");
-                    c2b_handle.abort();
-                    b2c_handle.abort();
                     break;
                 }
                 _ = tokio::time::sleep(check_interval) => {
                     // Check max lifetime
                     if start.elapsed() >= max_lifetime {
                         debug!("Connection exceeded max lifetime, closing");
-                        c2b_handle.abort();
-                        b2c_handle.abort();
                         break;
                     }
 
@@ -174,8 +193,6 @@ pub async fn forward_bidirectional_with_timeouts(
                         let elapsed_since_activity = start.elapsed().as_millis() as u64 - current;
                         if elapsed_since_activity >= inactivity_timeout.as_millis() as u64 {
                             debug!("Connection inactive for {}ms, closing", elapsed_since_activity);
-                            c2b_handle.abort();
-                            b2c_handle.abort();
                             break;
                         }
                     }
@@ -183,6 +200,13 @@ pub async fn forward_bidirectional_with_timeouts(
                 }
             }
         }
+        // Phase 1: Signal copy loops to exit gracefully (allows FIN/close_notify)
+        conn_cancel.cancel();
+        // Phase 2: Wait for graceful shutdown (2s shutdown timeout + 2s margin)
+        tokio::time::sleep(std::time::Duration::from_secs(4)).await;
+        // Phase 3: Force-abort if still stuck (e.g. blocked on write_all)
+        c2b_abort.abort();
+        b2c_abort.abort();
     });
 
     let bytes_in = c2b.await.unwrap_or(0);

rust/crates/rustproxy-passthrough/src/proxy_protocol.rs

@@ -1,4 +1,4 @@
-use std::net::SocketAddr;
+use std::net::{IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr};
 use thiserror::Error;
 
 #[derive(Debug, Error)]
@@ -9,9 +9,11 @@ pub enum ProxyProtocolError {
     UnsupportedVersion,
     #[error("Parse error: {0}")]
     Parse(String),
+    #[error("Incomplete header: need {0} bytes, got {1}")]
+    Incomplete(usize, usize),
 }
 
-/// Parsed PROXY protocol v1 header.
+/// Parsed PROXY protocol header (v1 or v2).
 #[derive(Debug, Clone)]
 pub struct ProxyProtocolHeader {
     pub source_addr: SocketAddr,
@@ -24,14 +26,29 @@ pub struct ProxyProtocolHeader {
 pub enum ProxyProtocol {
     Tcp4,
     Tcp6,
+    Udp4,
+    Udp6,
     Unknown,
 }
 
+/// Transport type for PROXY v2 header generation.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum ProxyV2Transport {
+    Stream,   // TCP
+    Datagram, // UDP
+}
+
+/// PROXY protocol v2 signature (12 bytes).
+const PROXY_V2_SIGNATURE: [u8; 12] = [
+    0x0D, 0x0A, 0x0D, 0x0A, 0x00, 0x0D, 0x0A, 0x51, 0x55, 0x49, 0x54, 0x0A,
+];
+
+// ===== v1 (text format) =====
+
 /// Parse a PROXY protocol v1 header from data.
 ///
 /// Format: `PROXY TCP4 <src_ip> <dst_ip> <src_port> <dst_port>\r\n`
 pub fn parse_v1(data: &[u8]) -> Result<(ProxyProtocolHeader, usize), ProxyProtocolError> {
-    // Find the end of the header line
     let line_end = data
         .windows(2)
         .position(|w| w == b"\r\n")
@@ -56,10 +73,10 @@ pub fn parse_v1(data: &[u8]) -> Result<(ProxyProtocolHeader, usize), ProxyProtoc
         _ => return Err(ProxyProtocolError::UnsupportedVersion),
     };
 
-    let src_ip: std::net::IpAddr = parts[2]
+    let src_ip: IpAddr = parts[2]
         .parse()
         .map_err(|_| ProxyProtocolError::Parse("Invalid source IP".to_string()))?;
-    let dst_ip: std::net::IpAddr = parts[3]
+    let dst_ip: IpAddr = parts[3]
         .parse()
         .map_err(|_| ProxyProtocolError::Parse("Invalid destination IP".to_string()))?;
     let src_port: u16 = parts[4]
@@ -75,7 +92,6 @@ pub fn parse_v1(data: &[u8]) -> Result<(ProxyProtocolHeader, usize), ProxyProtoc
         protocol,
     };
 
-    // Consumed bytes = line + \r\n
     Ok((header, line_end + 2))
 }
 
@@ -97,10 +113,219 @@ pub fn is_proxy_protocol_v1(data: &[u8]) -> bool {
     data.starts_with(b"PROXY ")
 }
 
+// ===== v2 (binary format) =====
+
+/// Check if data starts with a PROXY protocol v2 header.
+pub fn is_proxy_protocol_v2(data: &[u8]) -> bool {
+    data.len() >= 12 && data[..12] == PROXY_V2_SIGNATURE
+}
+
+/// Parse a PROXY protocol v2 binary header.
+///
+/// Binary format:
+/// - [0..12]  signature (12 bytes)
+/// - [12]     version (high nibble) + command (low nibble)
+/// - [13]     address family (high nibble) + transport (low nibble)
+/// - [14..16] address block length (big-endian u16)
+/// - [16..]   address block (variable, depends on family)
+pub fn parse_v2(data: &[u8]) -> Result<(ProxyProtocolHeader, usize), ProxyProtocolError> {
+    if data.len() < 16 {
+        return Err(ProxyProtocolError::Incomplete(16, data.len()));
+    }
+
+    // Validate signature
+    if data[..12] != PROXY_V2_SIGNATURE {
+        return Err(ProxyProtocolError::InvalidHeader);
+    }
+
+    // Version (high nibble of byte 12) must be 0x2
+    let version = (data[12] >> 4) & 0x0F;
+    if version != 2 {
+        return Err(ProxyProtocolError::UnsupportedVersion);
+    }
+
+    // Command (low nibble of byte 12)
+    let command = data[12] & 0x0F;
+    // 0x0 = LOCAL, 0x1 = PROXY
+    if command > 1 {
+        return Err(ProxyProtocolError::Parse(format!("Unknown command: {}", command)));
+    }
+
+    // Address family (high nibble) + transport (low nibble) of byte 13
+    let family = (data[13] >> 4) & 0x0F;
+    let transport = data[13] & 0x0F;
+
+    // Address block length
+    let addr_len = u16::from_be_bytes([data[14], data[15]]) as usize;
+    let total_len = 16 + addr_len;
+
+    if data.len() < total_len {
+        return Err(ProxyProtocolError::Incomplete(total_len, data.len()));
+    }
+
+    // LOCAL command: no real addresses, return unspecified
+    if command == 0 {
+        return Ok((
+            ProxyProtocolHeader {
+                source_addr: SocketAddr::new(IpAddr::V4(Ipv4Addr::UNSPECIFIED), 0),
+                dest_addr: SocketAddr::new(IpAddr::V4(Ipv4Addr::UNSPECIFIED), 0),
+                protocol: ProxyProtocol::Unknown,
+            },
+            total_len,
+        ));
+    }
+
+    // PROXY command: parse addresses based on family + transport
+    let addr_block = &data[16..16 + addr_len];
+
+    match (family, transport) {
+        // AF_INET (0x1) + STREAM (0x1) = TCP4
+        (0x1, 0x1) => {
+            if addr_len < 12 {
+                return Err(ProxyProtocolError::Parse("IPv4 address block too short".to_string()));
+            }
+            let src_ip = Ipv4Addr::new(addr_block[0], addr_block[1], addr_block[2], addr_block[3]);
+            let dst_ip = Ipv4Addr::new(addr_block[4], addr_block[5], addr_block[6], addr_block[7]);
+            let src_port = u16::from_be_bytes([addr_block[8], addr_block[9]]);
+            let dst_port = u16::from_be_bytes([addr_block[10], addr_block[11]]);
+            Ok((
+                ProxyProtocolHeader {
+                    source_addr: SocketAddr::new(IpAddr::V4(src_ip), src_port),
+                    dest_addr: SocketAddr::new(IpAddr::V4(dst_ip), dst_port),
+                    protocol: ProxyProtocol::Tcp4,
+                },
+                total_len,
+            ))
+        }
+        // AF_INET (0x1) + DGRAM (0x2) = UDP4
+        (0x1, 0x2) => {
+            if addr_len < 12 {
+                return Err(ProxyProtocolError::Parse("IPv4 address block too short".to_string()));
+            }
+            let src_ip = Ipv4Addr::new(addr_block[0], addr_block[1], addr_block[2], addr_block[3]);
+            let dst_ip = Ipv4Addr::new(addr_block[4], addr_block[5], addr_block[6], addr_block[7]);
+            let src_port = u16::from_be_bytes([addr_block[8], addr_block[9]]);
+            let dst_port = u16::from_be_bytes([addr_block[10], addr_block[11]]);
+            Ok((
+                ProxyProtocolHeader {
+                    source_addr: SocketAddr::new(IpAddr::V4(src_ip), src_port),
+                    dest_addr: SocketAddr::new(IpAddr::V4(dst_ip), dst_port),
+                    protocol: ProxyProtocol::Udp4,
+                },
+                total_len,
+            ))
+        }
+        // AF_INET6 (0x2) + STREAM (0x1) = TCP6
+        (0x2, 0x1) => {
+            if addr_len < 36 {
+                return Err(ProxyProtocolError::Parse("IPv6 address block too short".to_string()));
+            }
+            let src_ip = Ipv6Addr::from(<[u8; 16]>::try_from(&addr_block[0..16]).unwrap());
+            let dst_ip = Ipv6Addr::from(<[u8; 16]>::try_from(&addr_block[16..32]).unwrap());
+            let src_port = u16::from_be_bytes([addr_block[32], addr_block[33]]);
+            let dst_port = u16::from_be_bytes([addr_block[34], addr_block[35]]);
+            Ok((
+                ProxyProtocolHeader {
+                    source_addr: SocketAddr::new(IpAddr::V6(src_ip), src_port),
+                    dest_addr: SocketAddr::new(IpAddr::V6(dst_ip), dst_port),
+                    protocol: ProxyProtocol::Tcp6,
+                },
+                total_len,
+            ))
+        }
+        // AF_INET6 (0x2) + DGRAM (0x2) = UDP6
+        (0x2, 0x2) => {
+            if addr_len < 36 {
+                return Err(ProxyProtocolError::Parse("IPv6 address block too short".to_string()));
+            }
+            let src_ip = Ipv6Addr::from(<[u8; 16]>::try_from(&addr_block[0..16]).unwrap());
+            let dst_ip = Ipv6Addr::from(<[u8; 16]>::try_from(&addr_block[16..32]).unwrap());
+            let src_port = u16::from_be_bytes([addr_block[32], addr_block[33]]);
+            let dst_port = u16::from_be_bytes([addr_block[34], addr_block[35]]);
+            Ok((
+                ProxyProtocolHeader {
+                    source_addr: SocketAddr::new(IpAddr::V6(src_ip), src_port),
+                    dest_addr: SocketAddr::new(IpAddr::V6(dst_ip), dst_port),
+                    protocol: ProxyProtocol::Udp6,
+                },
+                total_len,
+            ))
+        }
+        // AF_UNSPEC or unknown
+        (0x0, _) => Ok((
+            ProxyProtocolHeader {
+                source_addr: SocketAddr::new(IpAddr::V4(Ipv4Addr::UNSPECIFIED), 0),
+                dest_addr: SocketAddr::new(IpAddr::V4(Ipv4Addr::UNSPECIFIED), 0),
+                protocol: ProxyProtocol::Unknown,
+            },
+            total_len,
+        )),
+        _ => Err(ProxyProtocolError::Parse(format!(
+            "Unsupported family/transport: 0x{:X}{:X}",
+            family, transport
+        ))),
+    }
+}
+
+/// Generate a PROXY protocol v2 binary header.
+pub fn generate_v2(
+    source: &SocketAddr,
+    dest: &SocketAddr,
+    transport: ProxyV2Transport,
+) -> Vec<u8> {
+    let transport_nibble: u8 = match transport {
+        ProxyV2Transport::Stream => 0x1,
+        ProxyV2Transport::Datagram => 0x2,
+    };
+
+    match (source.ip(), dest.ip()) {
+        (IpAddr::V4(src_ip), IpAddr::V4(dst_ip)) => {
+            let mut buf = Vec::with_capacity(28);
+            buf.extend_from_slice(&PROXY_V2_SIGNATURE);
+            buf.push(0x21); // version 2, PROXY command
+            buf.push(0x10 | transport_nibble); // AF_INET + transport
+            buf.extend_from_slice(&12u16.to_be_bytes()); // addr block length
+            buf.extend_from_slice(&src_ip.octets());
+            buf.extend_from_slice(&dst_ip.octets());
+            buf.extend_from_slice(&source.port().to_be_bytes());
+            buf.extend_from_slice(&dest.port().to_be_bytes());
+            buf
+        }
+        (IpAddr::V6(src_ip), IpAddr::V6(dst_ip)) => {
+            let mut buf = Vec::with_capacity(52);
+            buf.extend_from_slice(&PROXY_V2_SIGNATURE);
+            buf.push(0x21); // version 2, PROXY command
+            buf.push(0x20 | transport_nibble); // AF_INET6 + transport
+            buf.extend_from_slice(&36u16.to_be_bytes()); // addr block length
+            buf.extend_from_slice(&src_ip.octets());
+            buf.extend_from_slice(&dst_ip.octets());
+            buf.extend_from_slice(&source.port().to_be_bytes());
+            buf.extend_from_slice(&dest.port().to_be_bytes());
+            buf
+        }
+        // Mixed IPv4/IPv6: map IPv4 to IPv6-mapped address
+        _ => {
+            let src_v6 = match source.ip() {
+                IpAddr::V4(v4) => v4.to_ipv6_mapped(),
+                IpAddr::V6(v6) => v6,
+            };
+            let dst_v6 = match dest.ip() {
+                IpAddr::V4(v4) => v4.to_ipv6_mapped(),
+                IpAddr::V6(v6) => v6,
+            };
+            let src6 = SocketAddr::new(IpAddr::V6(src_v6), source.port());
+            let dst6 = SocketAddr::new(IpAddr::V6(dst_v6), dest.port());
+            generate_v2(&src6, &dst6, transport)
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
 
+    // ===== v1 tests =====
+
     #[test]
     fn test_parse_v1_tcp4() {
         let header = b"PROXY TCP4 192.168.1.100 10.0.0.1 12345 443\r\n";
@@ -126,4 +351,130 @@ mod tests {
         assert!(is_proxy_protocol_v1(b"PROXY TCP4 ..."));
         assert!(!is_proxy_protocol_v1(b"GET / HTTP/1.1"));
     }
+
+    // ===== v2 tests =====
+
+    #[test]
+    fn test_is_proxy_protocol_v2() {
+        assert!(is_proxy_protocol_v2(&PROXY_V2_SIGNATURE));
+        assert!(!is_proxy_protocol_v2(b"PROXY TCP4 ..."));
+        assert!(!is_proxy_protocol_v2(b"short"));
+    }
+
+    #[test]
+    fn test_parse_v2_tcp4() {
+        let source: SocketAddr = "198.51.100.10:54321".parse().unwrap();
+        let dest: SocketAddr = "203.0.113.25:8443".parse().unwrap();
+        let header = generate_v2(&source, &dest, ProxyV2Transport::Stream);
+
+        assert_eq!(header.len(), 28);
+        let (parsed, consumed) = parse_v2(&header).unwrap();
+        assert_eq!(consumed, 28);
+        assert_eq!(parsed.protocol, ProxyProtocol::Tcp4);
+        assert_eq!(parsed.source_addr, source);
+        assert_eq!(parsed.dest_addr, dest);
+    }
+
+    #[test]
+    fn test_parse_v2_udp4() {
+        let source: SocketAddr = "10.0.0.1:12345".parse().unwrap();
+        let dest: SocketAddr = "10.0.0.2:53".parse().unwrap();
+        let header = generate_v2(&source, &dest, ProxyV2Transport::Datagram);
+
+        assert_eq!(header.len(), 28);
+        assert_eq!(header[13], 0x12); // AF_INET + DGRAM
+
+        let (parsed, consumed) = parse_v2(&header).unwrap();
+        assert_eq!(consumed, 28);
+        assert_eq!(parsed.protocol, ProxyProtocol::Udp4);
+        assert_eq!(parsed.source_addr, source);
+        assert_eq!(parsed.dest_addr, dest);
+    }
+
+    #[test]
+    fn test_parse_v2_tcp6() {
+        let source: SocketAddr = "[2001:db8::1]:54321".parse().unwrap();
+        let dest: SocketAddr = "[2001:db8::2]:443".parse().unwrap();
+        let header = generate_v2(&source, &dest, ProxyV2Transport::Stream);
+
+        assert_eq!(header.len(), 52);
+        assert_eq!(header[13], 0x21); // AF_INET6 + STREAM
+
+        let (parsed, consumed) = parse_v2(&header).unwrap();
+        assert_eq!(consumed, 52);
+        assert_eq!(parsed.protocol, ProxyProtocol::Tcp6);
+        assert_eq!(parsed.source_addr, source);
+        assert_eq!(parsed.dest_addr, dest);
+    }
+
+    #[test]
+    fn test_generate_v2_tcp4_byte_layout() {
+        let source: SocketAddr = "1.2.3.4:1000".parse().unwrap();
+        let dest: SocketAddr = "5.6.7.8:443".parse().unwrap();
+        let header = generate_v2(&source, &dest, ProxyV2Transport::Stream);
+
+        assert_eq!(&header[0..12], &PROXY_V2_SIGNATURE);
+        assert_eq!(header[12], 0x21); // v2, PROXY
+        assert_eq!(header[13], 0x11); // AF_INET, STREAM
+        assert_eq!(u16::from_be_bytes([header[14], header[15]]), 12); // addr len
+        assert_eq!(&header[16..20], &[1, 2, 3, 4]); // src ip
+        assert_eq!(&header[20..24], &[5, 6, 7, 8]); // dst ip
+        assert_eq!(u16::from_be_bytes([header[24], header[25]]), 1000); // src port
+        assert_eq!(u16::from_be_bytes([header[26], header[27]]), 443); // dst port
+    }
+
+    #[test]
+    fn test_generate_v2_udp4_byte_layout() {
+        let source: SocketAddr = "10.0.0.1:5000".parse().unwrap();
+        let dest: SocketAddr = "10.0.0.2:53".parse().unwrap();
+        let header = generate_v2(&source, &dest, ProxyV2Transport::Datagram);
+
+        assert_eq!(header[12], 0x21); // v2, PROXY
+        assert_eq!(header[13], 0x12); // AF_INET, DGRAM (UDP)
+    }
+
+    #[test]
+    fn test_parse_v2_local_command() {
+        // Build a LOCAL command header (no addresses)
+        let mut header = Vec::new();
+        header.extend_from_slice(&PROXY_V2_SIGNATURE);
+        header.push(0x20); // v2, LOCAL
+        header.push(0x00); // AF_UNSPEC
+        header.extend_from_slice(&0u16.to_be_bytes()); // 0-length address block
+
+        let (parsed, consumed) = parse_v2(&header).unwrap();
+        assert_eq!(consumed, 16);
+        assert_eq!(parsed.protocol, ProxyProtocol::Unknown);
+        assert_eq!(parsed.source_addr.port(), 0);
+    }
+
+    #[test]
+    fn test_parse_v2_incomplete() {
+        let data = &PROXY_V2_SIGNATURE[..8]; // only 8 bytes
+        assert!(parse_v2(data).is_err());
+    }
+
+    #[test]
+    fn test_parse_v2_wrong_version() {
+        let mut header = Vec::new();
+        header.extend_from_slice(&PROXY_V2_SIGNATURE);
+        header.push(0x11); // version 1, not 2
+        header.push(0x11);
+        header.extend_from_slice(&12u16.to_be_bytes());
+        header.extend_from_slice(&[0u8; 12]);
+        assert!(matches!(parse_v2(&header), Err(ProxyProtocolError::UnsupportedVersion)));
+    }
+
+    #[test]
+    fn test_v2_roundtrip_with_trailing_data() {
+        let source: SocketAddr = "192.168.1.1:8080".parse().unwrap();
+        let dest: SocketAddr = "10.0.0.1:443".parse().unwrap();
+        let mut data = generate_v2(&source, &dest, ProxyV2Transport::Stream);
+        data.extend_from_slice(b"GET / HTTP/1.1\r\n"); // trailing app data
+
+        let (parsed, consumed) = parse_v2(&data).unwrap();
+        assert_eq!(consumed, 28);
+        assert_eq!(parsed.source_addr, source);
+        assert_eq!(&data[consumed..], b"GET / HTTP/1.1\r\n");
+    }
 }

rust/crates/rustproxy-passthrough/src/tcp_listener.rs

@@ -465,21 +465,19 @@ impl TcpListenerManager {
                         Ok((stream, peer_addr)) => {
                             let ip = peer_addr.ip();
 
-                            // Global connection limit — acquire semaphore permit with timeout
-                            let permit = match tokio::time::timeout(
-                                std::time::Duration::from_secs(5),
-                                conn_semaphore.clone().acquire_owned(),
-                            ).await {
-                                Ok(Ok(permit)) => permit,
-                                Ok(Err(_)) => {
-                                    // Semaphore closed — shouldn't happen, but be safe
-                                    debug!("Connection semaphore closed, dropping connection from {}", peer_addr);
+                            // Global connection limit — non-blocking check.
+                            // MUST NOT block the accept loop: a blocking acquire would stall
+                            // ALL connections to this port (not just the one over limit), because
+                            // listener.accept() is not polled while we await the semaphore.
+                            let permit = match conn_semaphore.clone().try_acquire_owned() {
+                                Ok(permit) => permit,
+                                Err(tokio::sync::TryAcquireError::NoPermits) => {
+                                    warn!("Global connection limit reached, dropping connection from {}", peer_addr);
                                     drop(stream);
                                     continue;
                                 }
-                                Err(_) => {
-                                    // Timeout — global limit reached
-                                    debug!("Global connection limit reached, dropping connection from {}", peer_addr);
+                                Err(tokio::sync::TryAcquireError::Closed) => {
+                                    warn!("Connection semaphore closed, dropping connection from {}", peer_addr);
                                     drop(stream);
                                     continue;
                                 }
@@ -487,7 +485,7 @@ impl TcpListenerManager {
 
                             // Check per-IP limits and rate limiting
                             if !conn_tracker.try_accept(&ip) {
-                                debug!("Rejected connection from {} (per-IP limit or rate limit)", peer_addr);
+                                warn!("Rejected connection from {} (per-IP limit or rate limit)", peer_addr);
                                 drop(stream);
                                 drop(permit);
                                 continue;
@@ -519,7 +517,7 @@ impl TcpListenerManager {
                                     stream, port, peer_addr, rm, m, tc, sa, hp, cc, cn, sr, rc,
                                 ).await;
                                 if let Err(e) = result {
-                                    debug!("Connection error from {}: {}", peer_addr, e);
+                                    warn!("Connection error from {}: {}", peer_addr, e);
                                 }
                             });
                         }
@@ -563,8 +561,9 @@ impl TcpListenerManager {
         // Non-proxy connections skip the peek entirely (no latency cost).
         let mut effective_peer_addr = peer_addr;
         if !conn_config.proxy_ips.is_empty() && conn_config.proxy_ips.contains(&peer_addr.ip()) {
-            // Trusted proxy IP — peek for PROXY protocol header
-            let mut proxy_peek = vec![0u8; 256];
+            // Trusted proxy IP — peek for PROXY protocol header.
+            // Use stack-allocated buffers (PROXY v1 headers are max ~108 bytes).
+            let mut proxy_peek = [0u8; 256];
             let pn = match tokio::time::timeout(
                 std::time::Duration::from_millis(conn_config.initial_data_timeout_ms),
                 stream.peek(&mut proxy_peek),
@@ -574,18 +573,30 @@ impl TcpListenerManager {
                 Err(_) => return Err("Initial data timeout (proxy protocol peek)".into()),
             };
 
-            if pn > 0 && crate::proxy_protocol::is_proxy_protocol_v1(&proxy_peek[..pn]) {
-                match crate::proxy_protocol::parse_v1(&proxy_peek[..pn]) {
-                    Ok((header, consumed)) => {
-                        debug!("PROXY protocol: real client {} -> {}", header.source_addr, header.dest_addr);
-                        effective_peer_addr = header.source_addr;
-                        // Consume the proxy protocol header bytes
-                        let mut discard = vec![0u8; consumed];
-                        stream.read_exact(&mut discard).await?;
+            if pn > 0 {
+                if crate::proxy_protocol::is_proxy_protocol_v1(&proxy_peek[..pn]) {
+                    match crate::proxy_protocol::parse_v1(&proxy_peek[..pn]) {
+                        Ok((header, consumed)) => {
+                            debug!("PROXY v1: real client {} -> {}", header.source_addr, header.dest_addr);
+                            effective_peer_addr = header.source_addr;
+                            let mut discard = [0u8; 128];
+                            stream.read_exact(&mut discard[..consumed]).await?;
+                        }
+                        Err(e) => {
+                            debug!("Failed to parse PROXY v1 header: {}", e);
+                        }
                     }
-                    Err(e) => {
-                        debug!("Failed to parse PROXY protocol header: {}", e);
-                        // Not a PROXY protocol header, continue normally
+                } else if crate::proxy_protocol::is_proxy_protocol_v2(&proxy_peek[..pn]) {
+                    match crate::proxy_protocol::parse_v2(&proxy_peek[..pn]) {
+                        Ok((header, consumed)) => {
+                            debug!("PROXY v2: real client {} -> {} ({:?})", header.source_addr, header.dest_addr, header.protocol);
+                            effective_peer_addr = header.source_addr;
+                            let mut discard = [0u8; 256];
+                            stream.read_exact(&mut discard[..consumed]).await?;
+                        }
+                        Err(e) => {
+                            debug!("Failed to parse PROXY v2 header: {}", e);
+                        }
                     }
                 }
             }
@@ -664,7 +675,7 @@ impl TcpListenerManager {
                             if !rustproxy_http::request_filter::RequestFilter::check_ip_security(
                                 security, &peer_addr.ip(),
                             ) {
-                                debug!("Connection from {} blocked by route security", peer_addr);
+                                warn!("Connection from {} blocked by route security", peer_addr);
                                 return Ok(());
                             }
                         }
@@ -810,7 +821,7 @@ impl TcpListenerManager {
         let route_match = match route_match {
             Some(rm) => rm,
             None => {
-                debug!("No route matched for port {} domain {:?}", port, domain);
+                warn!("No route matched for port {} domain {:?} from {}", port, domain, peer_addr);
                 if is_http {
                     // Send a proper HTTP error instead of dropping the connection
                     use tokio::io::AsyncWriteExt;
@@ -844,7 +855,7 @@ impl TcpListenerManager {
                 security,
                 &peer_addr.ip(),
             ) {
-                debug!("Connection from {} blocked by route security", peer_addr);
+                warn!("Connection from {} blocked by route security", peer_addr);
                 return Ok(());
             }
         }
@@ -987,13 +998,18 @@ impl TcpListenerManager {
                     Err(_) => return Err("TLS handshake timeout".into()),
                 };
 
-                // Peek at decrypted data to determine if HTTP
+                // Peek at decrypted data to determine if HTTP.
+                // Timeout prevents connection leak if client completes TLS
+                // but never sends application data (scanners, health probes, slow-loris).
                 let mut buf_stream = tokio::io::BufReader::new(tls_stream);
                 let peeked = {
                     use tokio::io::AsyncBufReadExt;
-                    match buf_stream.fill_buf().await {
-                        Ok(data) => sni_parser::is_http(data),
-                        Err(_) => false,
+                    match tokio::time::timeout(
+                        std::time::Duration::from_millis(conn_config.initial_data_timeout_ms),
+                        buf_stream.fill_buf(),
+                    ).await {
+                        Ok(Ok(data)) => sni_parser::is_http(data),
+                        Ok(Err(_)) | Err(_) => false,
                     }
                 };
 
@@ -1011,7 +1027,11 @@ impl TcpListenerManager {
                         "TLS Terminate + HTTP: {} -> {}:{} (domain: {:?})",
                         peer_addr, target_host, target_port, domain
                     );
-                    http_proxy.handle_io(buf_stream, peer_addr, port, cancel.clone()).await;
+                    // Wrap in ShutdownOnDrop to ensure TLS close_notify is sent
+                    // even if hyper drops the connection without calling shutdown
+                    // (e.g. H2 close, backend error, idle timeout drain).
+                    let wrapped = rustproxy_http::shutdown_on_drop::ShutdownOnDrop::new(buf_stream);
+                    http_proxy.handle_io(wrapped, peer_addr, port, cancel.clone()).await;
                 } else {
                     debug!(
                         "TLS Terminate + TCP: {} -> {}:{} (domain: {:?})",
@@ -1062,13 +1082,18 @@ impl TcpListenerManager {
                     Err(_) => return Err("TLS handshake timeout".into()),
                 };
 
-                // Peek at decrypted data to detect protocol
+                // Peek at decrypted data to detect protocol.
+                // Timeout prevents connection leak if client completes TLS
+                // but never sends application data (scanners, health probes, slow-loris).
                 let mut buf_stream = tokio::io::BufReader::new(tls_stream);
                 let is_http_data = {
                     use tokio::io::AsyncBufReadExt;
-                    match buf_stream.fill_buf().await {
-                        Ok(data) => sni_parser::is_http(data),
-                        Err(_) => false,
+                    match tokio::time::timeout(
+                        std::time::Duration::from_millis(conn_config.initial_data_timeout_ms),
+                        buf_stream.fill_buf(),
+                    ).await {
+                        Ok(Ok(data)) => sni_parser::is_http(data),
+                        Ok(Err(_)) | Err(_) => false,
                     }
                 };
 
@@ -1088,7 +1113,10 @@ impl TcpListenerManager {
                         "TLS Terminate+Reencrypt + HTTP: {} (domain: {:?})",
                         peer_addr, domain
                     );
-                    http_proxy.handle_io(buf_stream, peer_addr, port, cancel.clone()).await;
+                    // Wrap in ShutdownOnDrop to ensure TLS close_notify is sent
+                    // even if hyper drops the connection without calling shutdown.
+                    let wrapped = rustproxy_http::shutdown_on_drop::ShutdownOnDrop::new(buf_stream);
+                    http_proxy.handle_io(wrapped, peer_addr, port, cancel.clone()).await;
                 } else {
                     // Non-HTTP: TLS-to-TLS tunnel (existing behavior for raw TCP protocols)
                     debug!(
@@ -1396,15 +1424,24 @@ impl TcpListenerManager {
         let last_activity = Arc::new(AtomicU64::new(0));
         let start = std::time::Instant::now();
 
+        // Per-connection cancellation token: the watchdog cancels this instead of
+        // aborting tasks, so the copy loops can shut down gracefully (TLS close_notify
+        // for terminate/reencrypt mode, TCP FIN for passthrough mode).
+        let conn_cancel = CancellationToken::new();
+
         let la1 = Arc::clone(&last_activity);
         let metrics_c2b = metrics.clone();
+        let cc1 = conn_cancel.clone();
         let c2b = tokio::spawn(async move {
             let mut buf = vec![0u8; 65536];
             let mut total = 0u64;
             loop {
-                let n = match client_read.read(&mut buf).await {
-                    Ok(0) | Err(_) => break,
-                    Ok(n) => n,
+                let n = tokio::select! {
+                    result = client_read.read(&mut buf) => match result {
+                        Ok(0) | Err(_) => break,
+                        Ok(n) => n,
+                    },
+                    _ = cc1.cancelled() => break,
                 };
                 if backend_write.write_all(&buf[..n]).await.is_err() {
                     break;
@@ -1418,19 +1455,27 @@ impl TcpListenerManager {
                     ctx.collector.record_bytes(n as u64, 0, ctx.route_id.as_deref(), ctx.source_ip.as_deref());
                 }
             }
-            let _ = backend_write.shutdown().await;
+            // Graceful shutdown with timeout (sends TLS close_notify / TCP FIN)
+            let _ = tokio::time::timeout(
+                std::time::Duration::from_secs(2),
+                backend_write.shutdown(),
+            ).await;
             total
         });
 
         let la2 = Arc::clone(&last_activity);
         let metrics_b2c = metrics;
+        let cc2 = conn_cancel.clone();
         let b2c = tokio::spawn(async move {
             let mut buf = vec![0u8; 65536];
             let mut total = 0u64;
             loop {
-                let n = match backend_read.read(&mut buf).await {
-                    Ok(0) | Err(_) => break,
-                    Ok(n) => n,
+                let n = tokio::select! {
+                    result = backend_read.read(&mut buf) => match result {
+                        Ok(0) | Err(_) => break,
+                        Ok(n) => n,
+                    },
+                    _ = cc2.cancelled() => break,
                 };
                 if client_write.write_all(&buf[..n]).await.is_err() {
                     break;
@@ -1444,14 +1489,20 @@ impl TcpListenerManager {
                     ctx.collector.record_bytes(0, n as u64, ctx.route_id.as_deref(), ctx.source_ip.as_deref());
                 }
             }
-            let _ = client_write.shutdown().await;
+            // Graceful shutdown with timeout (sends TLS close_notify / TCP FIN)
+            let _ = tokio::time::timeout(
+                std::time::Duration::from_secs(2),
+                client_write.shutdown(),
+            ).await;
             total
         });
 
-        // Watchdog task: check for inactivity, max lifetime, and cancellation
+        // Watchdog task: check for inactivity, max lifetime, and cancellation.
+        // First cancels the per-connection token for graceful shutdown (close_notify/FIN),
+        // then falls back to abort if the tasks are stuck (e.g. on a blocked write_all).
         let la_watch = Arc::clone(&last_activity);
-        let c2b_handle = c2b.abort_handle();
-        let b2c_handle = b2c.abort_handle();
+        let c2b_abort = c2b.abort_handle();
+        let b2c_abort = b2c.abort_handle();
         let watchdog = tokio::spawn(async move {
             let check_interval = std::time::Duration::from_secs(5);
             let mut last_seen = 0u64;
@@ -1459,16 +1510,12 @@ impl TcpListenerManager {
                 tokio::select! {
                     _ = cancel.cancelled() => {
                         debug!("Split-stream connection cancelled by shutdown");
-                        c2b_handle.abort();
-                        b2c_handle.abort();
                         break;
                     }
                     _ = tokio::time::sleep(check_interval) => {
                         // Check max lifetime
                         if start.elapsed() >= max_lifetime {
                             debug!("Connection exceeded max lifetime, closing");
-                            c2b_handle.abort();
-                            b2c_handle.abort();
                             break;
                         }
 
@@ -1479,8 +1526,6 @@ impl TcpListenerManager {
                             let elapsed_since_activity = start.elapsed().as_millis() as u64 - current;
                             if elapsed_since_activity >= inactivity_timeout.as_millis() as u64 {
                                 debug!("Connection inactive for {}ms, closing", elapsed_since_activity);
-                                c2b_handle.abort();
-                                b2c_handle.abort();
                                 break;
                             }
                         }
@@ -1488,6 +1533,13 @@ impl TcpListenerManager {
                     }
                 }
             }
+            // Phase 1: Signal copy loops to exit gracefully (allows close_notify/FIN)
+            conn_cancel.cancel();
+            // Phase 2: Wait for graceful shutdown (2s shutdown timeout + 2s margin)
+            tokio::time::sleep(std::time::Duration::from_secs(4)).await;
+            // Phase 3: Force-abort if still stuck (e.g. blocked on write_all)
+            c2b_abort.abort();
+            b2c_abort.abort();
         });
 
         let bytes_in = c2b.await.unwrap_or(0);

rust/crates/rustproxy-routing/src/matchers/domain.rs

@@ -6,25 +6,28 @@
 /// - `example.com` exact match
 /// - `**.example.com` matches any depth of subdomain
 pub fn domain_matches(pattern: &str, domain: &str) -> bool {
-    let pattern = pattern.trim().to_lowercase();
-    let domain = domain.trim().to_lowercase();
+    let pattern = pattern.trim();
+    let domain = domain.trim();
 
     if pattern == "*" {
         return true;
     }
 
-    if pattern == domain {
+    if pattern.eq_ignore_ascii_case(domain) {
         return true;
     }
 
     // Wildcard patterns
-    if pattern.starts_with("*.") {
+    if pattern.starts_with("*.") || pattern.starts_with("*.") {
         let suffix = &pattern[2..]; // e.g., "example.com"
         // Match exact parent or any single-level subdomain
-        if domain == suffix {
+        if domain.eq_ignore_ascii_case(suffix) {
             return true;
         }
-        if domain.ends_with(&format!(".{}", suffix)) {
+        if domain.len() > suffix.len() + 1
+            && domain.as_bytes()[domain.len() - suffix.len() - 1] == b'.'
+            && domain[domain.len() - suffix.len()..].eq_ignore_ascii_case(suffix)
+        {
             // Check it's a single level subdomain for `*.`
             let prefix = &domain[..domain.len() - suffix.len() - 1];
             return !prefix.contains('.');
@@ -35,11 +38,22 @@ pub fn domain_matches(pattern: &str, domain: &str) -> bool {
     if pattern.starts_with("**.") {
         let suffix = &pattern[3..];
         // Match exact parent or any depth of subdomain
-        return domain == suffix || domain.ends_with(&format!(".{}", suffix));
+        if domain.eq_ignore_ascii_case(suffix) {
+            return true;
+        }
+        if domain.len() > suffix.len() + 1
+            && domain.as_bytes()[domain.len() - suffix.len() - 1] == b'.'
+            && domain[domain.len() - suffix.len()..].eq_ignore_ascii_case(suffix)
+        {
+            return true;
+        }
+        return false;
     }
 
-    // Use glob-match for more complex patterns
-    glob_match::glob_match(&pattern, &domain)
+    // Use glob-match for more complex patterns (case-insensitive via lowercasing)
+    let pattern_lower = pattern.to_lowercase();
+    let domain_lower = domain.to_lowercase();
+    glob_match::glob_match(&pattern_lower, &domain_lower)
 }
 
 /// Check if a domain matches any of the given patterns.

rust/crates/rustproxy-routing/src/route_manager.rs

@@ -60,6 +60,16 @@ impl RouteManager {
         manager
     }
 
+    /// Check if any route on the given port uses header matching.
+    /// Used to skip expensive header HashMap construction when no route needs it.
+    pub fn any_route_has_headers(&self, port: u16) -> bool {
+        if let Some(indices) = self.port_index.get(&port) {
+            indices.iter().any(|&idx| self.routes[idx].route_match.headers.is_some())
+        } else {
+            false
+        }
+    }
+
     /// Find the best matching route for the given context.
     pub fn find_route<'a>(&'a self, ctx: &MatchContext<'_>) -> Option<RouteMatchResult<'a>> {
         // Get routes for this port

rust/crates/rustproxy/src/lib.rs

@@ -632,15 +632,13 @@ impl RustProxy {
         let new_manager = Arc::new(new_manager);
         self.route_table.store(Arc::clone(&new_manager));
 
-        // Update listener manager
+        // Update listener manager.
+        // IMPORTANT: TLS configs must be swapped BEFORE the route manager so that
+        // new routes only become visible after their certs are loaded. The reverse
+        // order (routes first) creates a window where connections match new routes
+        // but get the old TLS acceptor, causing cert mismatches.
         if let Some(ref mut listener) = self.listener_manager {
-            listener.update_route_manager(Arc::clone(&new_manager));
-            // Cancel connections on routes that were removed or disabled
-            listener.invalidate_removed_routes(&active_route_ids);
-            // Prune HTTP proxy caches (rate limiters, regex cache, round-robin counters)
-            listener.prune_http_proxy_caches(&active_route_ids);
-
-            // Update TLS configs
+            // 1. Update TLS configs first (so new certs are available before new routes)
             let mut tls_configs = Self::extract_tls_configs(&routes);
             if let Some(ref cm_arc) = self.cert_manager {
                 let cm = cm_arc.lock().await;
@@ -661,6 +659,13 @@ impl RustProxy {
             }
             listener.set_tls_configs(tls_configs);
 
+            // 2. Now swap the route manager (new routes become visible with certs already loaded)
+            listener.update_route_manager(Arc::clone(&new_manager));
+            // Cancel connections on routes that were removed or disabled
+            listener.invalidate_removed_routes(&active_route_ids);
+            // Prune HTTP proxy caches (rate limiters, regex cache, round-robin counters)
+            listener.prune_http_proxy_caches(&active_route_ids);
+
             // Add new ports
             for port in &new_ports {
                 if !old_ports.contains(port) {

test/test.proxy-protocol.ts

@@ -1,133 +0,0 @@
-import { expect, tap } from '@git.zone/tstest/tapbundle';
-import * as smartproxy from '../ts/index.js';
-import { ProxyProtocolParser } from '../ts/core/utils/proxy-protocol.js';
-
-tap.test('PROXY protocol v1 parser - valid headers', async () => {
-  // Test TCP4 format
-  const tcp4Header = Buffer.from('PROXY TCP4 192.168.1.1 10.0.0.1 56324 443\r\n', 'ascii');
-  const tcp4Result = ProxyProtocolParser.parse(tcp4Header);
-  
-  expect(tcp4Result.proxyInfo).property('protocol').toEqual('TCP4');
-  expect(tcp4Result.proxyInfo).property('sourceIP').toEqual('192.168.1.1');
-  expect(tcp4Result.proxyInfo).property('sourcePort').toEqual(56324);
-  expect(tcp4Result.proxyInfo).property('destinationIP').toEqual('10.0.0.1');
-  expect(tcp4Result.proxyInfo).property('destinationPort').toEqual(443);
-  expect(tcp4Result.remainingData.length).toEqual(0);
-  
-  // Test TCP6 format
-  const tcp6Header = Buffer.from('PROXY TCP6 2001:db8::1 2001:db8::2 56324 443\r\n', 'ascii');
-  const tcp6Result = ProxyProtocolParser.parse(tcp6Header);
-  
-  expect(tcp6Result.proxyInfo).property('protocol').toEqual('TCP6');
-  expect(tcp6Result.proxyInfo).property('sourceIP').toEqual('2001:db8::1');
-  expect(tcp6Result.proxyInfo).property('sourcePort').toEqual(56324);
-  expect(tcp6Result.proxyInfo).property('destinationIP').toEqual('2001:db8::2');
-  expect(tcp6Result.proxyInfo).property('destinationPort').toEqual(443);
-  
-  // Test UNKNOWN protocol
-  const unknownHeader = Buffer.from('PROXY UNKNOWN\r\n', 'ascii');
-  const unknownResult = ProxyProtocolParser.parse(unknownHeader);
-  
-  expect(unknownResult.proxyInfo).property('protocol').toEqual('UNKNOWN');
-  expect(unknownResult.proxyInfo).property('sourceIP').toEqual('');
-  expect(unknownResult.proxyInfo).property('sourcePort').toEqual(0);
-});
-
-tap.test('PROXY protocol v1 parser - with remaining data', async () => {
-  const headerWithData = Buffer.concat([
-    Buffer.from('PROXY TCP4 192.168.1.1 10.0.0.1 56324 443\r\n', 'ascii'),
-    Buffer.from('GET / HTTP/1.1\r\n', 'ascii')
-  ]);
-  
-  const result = ProxyProtocolParser.parse(headerWithData);
-  
-  expect(result.proxyInfo).property('protocol').toEqual('TCP4');
-  expect(result.proxyInfo).property('sourceIP').toEqual('192.168.1.1');
-  expect(result.remainingData.toString()).toEqual('GET / HTTP/1.1\r\n');
-});
-
-tap.test('PROXY protocol v1 parser - invalid headers', async () => {
-  // Not a PROXY protocol header
-  const notProxy = Buffer.from('GET / HTTP/1.1\r\n', 'ascii');
-  const notProxyResult = ProxyProtocolParser.parse(notProxy);
-  expect(notProxyResult.proxyInfo).toBeNull();
-  expect(notProxyResult.remainingData).toEqual(notProxy);
-  
-  // Invalid protocol
-  expect(() => {
-    ProxyProtocolParser.parse(Buffer.from('PROXY INVALID 1.1.1.1 2.2.2.2 80 443\r\n', 'ascii'));
-  }).toThrow();
-  
-  // Wrong number of fields
-  expect(() => {
-    ProxyProtocolParser.parse(Buffer.from('PROXY TCP4 192.168.1.1 10.0.0.1 56324\r\n', 'ascii'));
-  }).toThrow();
-  
-  // Invalid port
-  expect(() => {
-    ProxyProtocolParser.parse(Buffer.from('PROXY TCP4 192.168.1.1 10.0.0.1 99999 443\r\n', 'ascii'));
-  }).toThrow();
-  
-  // Invalid IP for protocol
-  expect(() => {
-    ProxyProtocolParser.parse(Buffer.from('PROXY TCP4 2001:db8::1 10.0.0.1 56324 443\r\n', 'ascii'));
-  }).toThrow();
-});
-
-tap.test('PROXY protocol v1 parser - incomplete headers', async () => {
-  // Header without terminator
-  const incomplete = Buffer.from('PROXY TCP4 192.168.1.1 10.0.0.1 56324 443', 'ascii');
-  const result = ProxyProtocolParser.parse(incomplete);
-  
-  expect(result.proxyInfo).toBeNull();
-  expect(result.remainingData).toEqual(incomplete);
-  
-  // Header exceeding max length - create a buffer that actually starts with PROXY
-  const longHeader = Buffer.from('PROXY TCP4 ' + '1'.repeat(100), 'ascii');
-  expect(() => {
-    ProxyProtocolParser.parse(longHeader);
-  }).toThrow();
-});
-
-tap.test('PROXY protocol v1 generator', async () => {
-  // Generate TCP4 header
-  const tcp4Info = {
-    protocol: 'TCP4' as const,
-    sourceIP: '192.168.1.1',
-    sourcePort: 56324,
-    destinationIP: '10.0.0.1',
-    destinationPort: 443
-  };
-  
-  const tcp4Header = ProxyProtocolParser.generate(tcp4Info);
-  expect(tcp4Header.toString('ascii')).toEqual('PROXY TCP4 192.168.1.1 10.0.0.1 56324 443\r\n');
-  
-  // Generate TCP6 header
-  const tcp6Info = {
-    protocol: 'TCP6' as const,
-    sourceIP: '2001:db8::1',
-    sourcePort: 56324,
-    destinationIP: '2001:db8::2',
-    destinationPort: 443
-  };
-  
-  const tcp6Header = ProxyProtocolParser.generate(tcp6Info);
-  expect(tcp6Header.toString('ascii')).toEqual('PROXY TCP6 2001:db8::1 2001:db8::2 56324 443\r\n');
-  
-  // Generate UNKNOWN header
-  const unknownInfo = {
-    protocol: 'UNKNOWN' as const,
-    sourceIP: '',
-    sourcePort: 0,
-    destinationIP: '',
-    destinationPort: 0
-  };
-  
-  const unknownHeader = ProxyProtocolParser.generate(unknownInfo);
-  expect(unknownHeader.toString('ascii')).toEqual('PROXY UNKNOWN\r\n');
-});
-
-// Skipping integration tests for now - focus on unit tests
-// Integration tests would require more complex setup and teardown
-
-export default tap.start();

ts/00_commitinfo_data.ts

@@ -3,6 +3,6 @@
  */
 export const commitinfo = {
   name: '@push.rocks/smartproxy',
-  version: '25.11.2',
+  version: '25.12.0',
   description: 'A powerful proxy package with unified route-based configuration for high traffic management. Features include SSL/TLS support, flexible routing patterns, WebSocket handling, advanced security options, and automatic ACME certificate management.'
 }

ts/core/utils/index.ts

@@ -15,4 +15,3 @@ export * from './lifecycle-component.js';
 export * from './binary-heap.js';
 export * from './enhanced-connection-pool.js';
 export * from './socket-utils.js';
-export * from './proxy-protocol.js';

ts/core/utils/proxy-protocol.ts

@@ -1,129 +0,0 @@
-import * as plugins from '../../plugins.js';
-import { logger } from './logger.js';
-import { ProxyProtocolParser as ProtocolParser, type IProxyInfo, type IProxyParseResult } from '../../protocols/proxy/index.js';
-
-// Re-export types from protocols for backward compatibility
-export type { IProxyInfo, IProxyParseResult } from '../../protocols/proxy/index.js';
-
-/**
- * Parser for PROXY protocol v1 (text format)
- * Spec: https://www.haproxy.org/download/1.8/doc/proxy-protocol.txt
- * 
- * This class now delegates to the protocol parser but adds 
- * smartproxy-specific features like socket reading and logging
- */
-export class ProxyProtocolParser {
-  static readonly PROXY_V1_SIGNATURE = ProtocolParser.PROXY_V1_SIGNATURE;
-  static readonly MAX_HEADER_LENGTH = ProtocolParser.MAX_HEADER_LENGTH;
-  static readonly HEADER_TERMINATOR = ProtocolParser.HEADER_TERMINATOR;
-  
-  /**
-   * Parse PROXY protocol v1 header from buffer
-   * Returns proxy info and remaining data after header
-   */
-  static parse(data: Buffer): IProxyParseResult {
-    // Delegate to protocol parser
-    return ProtocolParser.parse(data);
-  }
-  
-  /**
-   * Generate PROXY protocol v1 header
-   */
-  static generate(info: IProxyInfo): Buffer {
-    // Delegate to protocol parser
-    return ProtocolParser.generate(info);
-  }
-  
-  /**
-   * Validate IP address format
-   */
-  private static isValidIP(ip: string, protocol: 'TCP4' | 'TCP6' | 'UNKNOWN'): boolean {
-    return ProtocolParser.isValidIP(ip, protocol);
-  }
-  
-  /**
-   * Attempt to read a complete PROXY protocol header from a socket
-   * Returns null if no PROXY protocol detected or incomplete
-   */
-  static async readFromSocket(socket: plugins.net.Socket, timeout: number = 5000): Promise<IProxyParseResult | null> {
-    return new Promise((resolve) => {
-      let buffer = Buffer.alloc(0);
-      let resolved = false;
-      
-      const cleanup = () => {
-        socket.removeListener('data', onData);
-        socket.removeListener('error', onError);
-        clearTimeout(timer);
-      };
-      
-      const timer = setTimeout(() => {
-        if (!resolved) {
-          resolved = true;
-          cleanup();
-          resolve({
-            proxyInfo: null,
-            remainingData: buffer
-          });
-        }
-      }, timeout);
-      
-      const onData = (chunk: Buffer) => {
-        buffer = Buffer.concat([buffer, chunk]);
-        
-        // Check if we have enough data
-        if (!buffer.toString('ascii', 0, Math.min(6, buffer.length)).startsWith(this.PROXY_V1_SIGNATURE)) {
-          // Not PROXY protocol
-          resolved = true;
-          cleanup();
-          resolve({
-            proxyInfo: null,
-            remainingData: buffer
-          });
-          return;
-        }
-        
-        // Try to parse
-        try {
-          const result = this.parse(buffer);
-          if (result.proxyInfo) {
-            // Successfully parsed
-            resolved = true;
-            cleanup();
-            resolve(result);
-          } else if (buffer.length > this.MAX_HEADER_LENGTH) {
-            // Header too long
-            resolved = true;
-            cleanup();
-            resolve({
-              proxyInfo: null,
-              remainingData: buffer
-            });
-          }
-          // Otherwise continue reading
-        } catch (error) {
-          // Parse error
-          logger.log('error', `PROXY protocol parse error: ${error.message}`);
-          resolved = true;
-          cleanup();
-          resolve({
-            proxyInfo: null,
-            remainingData: buffer
-          });
-        }
-      };
-      
-      const onError = (error: Error) => {
-        logger.log('error', `Socket error while reading PROXY protocol: ${error.message}`);
-        resolved = true;
-        cleanup();
-        resolve({
-          proxyInfo: null,
-          remainingData: buffer
-        });
-      };
-      
-      socket.on('data', onData);
-      socket.on('error', onError);
-    });
-  }
-}

ts/protocols/proxy/index.ts

@@ -1,7 +1,6 @@
 /**
  * PROXY Protocol Module
- * HAProxy PROXY protocol implementation
+ * Type definitions for HAProxy PROXY protocol v1/v2
  */
 
-export * from './types.js';
-export * from './parser.js';
+export * from './types.js';

ts/protocols/proxy/parser.ts

@@ -1,183 +0,0 @@
-/**
- * PROXY Protocol Parser
- * Implementation of HAProxy PROXY protocol v1 (text format)
- * Spec: https://www.haproxy.org/download/1.8/doc/proxy-protocol.txt
- */
-
-import type { IProxyInfo, IProxyParseResult, TProxyProtocol } from './types.js';
-
-/**
- * PROXY protocol parser
- */
-export class ProxyProtocolParser {
-  static readonly PROXY_V1_SIGNATURE = 'PROXY ';
-  static readonly MAX_HEADER_LENGTH = 107; // Max length for v1 header
-  static readonly HEADER_TERMINATOR = '\r\n';
-  
-  /**
-   * Parse PROXY protocol v1 header from buffer
-   * Returns proxy info and remaining data after header
-   */
-  static parse(data: Buffer): IProxyParseResult {
-    // Check if buffer starts with PROXY signature
-    if (!data.toString('ascii', 0, 6).startsWith(this.PROXY_V1_SIGNATURE)) {
-      return {
-        proxyInfo: null,
-        remainingData: data
-      };
-    }
-    
-    // Find header terminator
-    const headerEndIndex = data.indexOf(this.HEADER_TERMINATOR);
-    if (headerEndIndex === -1) {
-      // Header incomplete, need more data
-      if (data.length > this.MAX_HEADER_LENGTH) {
-        // Header too long, invalid
-        throw new Error('PROXY protocol header exceeds maximum length');
-      }
-      return {
-        proxyInfo: null,
-        remainingData: data
-      };
-    }
-    
-    // Extract header line
-    const headerLine = data.toString('ascii', 0, headerEndIndex);
-    const remainingData = data.slice(headerEndIndex + 2); // Skip \r\n
-    
-    // Parse header
-    const parts = headerLine.split(' ');
-    
-    if (parts.length < 2) {
-      throw new Error(`Invalid PROXY protocol header format: ${headerLine}`);
-    }
-    
-    const [signature, protocol] = parts;
-    
-    // Validate protocol
-    if (!['TCP4', 'TCP6', 'UNKNOWN'].includes(protocol)) {
-      throw new Error(`Invalid PROXY protocol: ${protocol}`);
-    }
-    
-    // For UNKNOWN protocol, ignore addresses
-    if (protocol === 'UNKNOWN') {
-      return {
-        proxyInfo: {
-          protocol: 'UNKNOWN',
-          sourceIP: '',
-          sourcePort: 0,
-          destinationIP: '',
-          destinationPort: 0
-        },
-        remainingData
-      };
-    }
-    
-    // For TCP4/TCP6, we need all 6 parts
-    if (parts.length !== 6) {
-      throw new Error(`Invalid PROXY protocol header format: ${headerLine}`);
-    }
-    
-    const [, , srcIP, dstIP, srcPort, dstPort] = parts;
-    
-    // Validate and parse ports
-    const sourcePort = parseInt(srcPort, 10);
-    const destinationPort = parseInt(dstPort, 10);
-    
-    if (isNaN(sourcePort) || sourcePort < 0 || sourcePort > 65535) {
-      throw new Error(`Invalid source port: ${srcPort}`);
-    }
-    
-    if (isNaN(destinationPort) || destinationPort < 0 || destinationPort > 65535) {
-      throw new Error(`Invalid destination port: ${dstPort}`);
-    }
-    
-    // Validate IP addresses
-    const protocolType = protocol as TProxyProtocol;
-    if (!this.isValidIP(srcIP, protocolType)) {
-      throw new Error(`Invalid source IP for ${protocol}: ${srcIP}`);
-    }
-    
-    if (!this.isValidIP(dstIP, protocolType)) {
-      throw new Error(`Invalid destination IP for ${protocol}: ${dstIP}`);
-    }
-    
-    return {
-      proxyInfo: {
-        protocol: protocolType,
-        sourceIP: srcIP,
-        sourcePort,
-        destinationIP: dstIP,
-        destinationPort
-      },
-      remainingData
-    };
-  }
-  
-  /**
-   * Generate PROXY protocol v1 header
-   */
-  static generate(info: IProxyInfo): Buffer {
-    if (info.protocol === 'UNKNOWN') {
-      return Buffer.from(`PROXY UNKNOWN\r\n`, 'ascii');
-    }
-    
-    const header = `PROXY ${info.protocol} ${info.sourceIP} ${info.destinationIP} ${info.sourcePort} ${info.destinationPort}\r\n`;
-    
-    if (header.length > this.MAX_HEADER_LENGTH) {
-      throw new Error('Generated PROXY protocol header exceeds maximum length');
-    }
-    
-    return Buffer.from(header, 'ascii');
-  }
-  
-  /**
-   * Validate IP address format
-   */
-  static isValidIP(ip: string, protocol: TProxyProtocol): boolean {
-    if (protocol === 'TCP4') {
-      return this.isIPv4(ip);
-    } else if (protocol === 'TCP6') {
-      return this.isIPv6(ip);
-    }
-    return false;
-  }
-  
-  /**
-   * Check if string is valid IPv4
-   */
-  static isIPv4(ip: string): boolean {
-    const parts = ip.split('.');
-    if (parts.length !== 4) return false;
-    
-    for (const part of parts) {
-      const num = parseInt(part, 10);
-      if (isNaN(num) || num < 0 || num > 255 || part !== num.toString()) {
-        return false;
-      }
-    }
-    return true;
-  }
-  
-  /**
-   * Check if string is valid IPv6
-   */
-  static isIPv6(ip: string): boolean {
-    // Basic IPv6 validation
-    const ipv6Regex = /^(([0-9a-fA-F]{1,4}:){7}[0-9a-fA-F]{1,4}|([0-9a-fA-F]{1,4}:){1,7}:|([0-9a-fA-F]{1,4}:){1,6}:[0-9a-fA-F]{1,4}|([0-9a-fA-F]{1,4}:){1,5}(:[0-9a-fA-F]{1,4}){1,2}|([0-9a-fA-F]{1,4}:){1,4}(:[0-9a-fA-F]{1,4}){1,3}|([0-9a-fA-F]{1,4}:){1,3}(:[0-9a-fA-F]{1,4}){1,4}|([0-9a-fA-F]{1,4}:){1,2}(:[0-9a-fA-F]{1,4}){1,5}|[0-9a-fA-F]{1,4}:((:[0-9a-fA-F]{1,4}){1,6})|:((:[0-9a-fA-F]{1,4}){1,7}|:)|fe80:(:[0-9a-fA-F]{0,4}){0,4}%[0-9a-zA-Z]{1,}|::(ffff(:0{1,4}){0,1}:){0,1}((25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9])\.){3}(25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9])|([0-9a-fA-F]{1,4}:){1,4}:((25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9])\.){3}(25[0-5]|(2[0-4]|1{0,1}[0-9]){0,1}[0-9]))$/;
-    return ipv6Regex.test(ip);
-  }
-  
-  /**
-   * Create a connection ID string for tracking
-   */
-  static createConnectionId(connectionInfo: {
-    sourceIp?: string;
-    sourcePort?: number;
-    destIp?: string;
-    destPort?: number;
-  }): string {
-    const { sourceIp, sourcePort, destIp, destPort } = connectionInfo;
-    return `${sourceIp}:${sourcePort}-${destIp}:${destPort}`;
-  }
-}

ts/protocols/proxy/types.ts

@@ -11,7 +11,7 @@ export type TProxyProtocolVersion = 'v1' | 'v2';
 /**
  * Connection protocol type
  */
-export type TProxyProtocol = 'TCP4' | 'TCP6' | 'UNKNOWN';
+export type TProxyProtocol = 'TCP4' | 'TCP6' | 'UDP4' | 'UDP6' | 'UNKNOWN';
 
 /**
  * Interface representing parsed PROXY protocol information