perf(data-plane): restore native throughput after host portability (#2452)

* perf(core): make data-plane idle check constant time

Avoid scanning every DashMap shard for each peer packet when no data-plane flows are active.

Publish the flow count before insertion and release it after removal so an Acquire load is a safe O(1) idle signal. Reject count overflow and underflow instead of silently saturating.

* test(perf): add repeatable two-node netns benchmark

Create isolated underlay namespaces, pin both EasyTier cores and iperf3 endpoints, and measure a single TCP flow in both directions over either UDP or TCP peer transport.

Keep every iperf3 JSON result and emit directional medians while cleaning up processes and namespaces on every exit path.

* perf(tcp): preserve native owned stream halves

Let each VirtualTcpSocket adapter consume itself into independent read and write halves. Portable adapters retain the generic shared split as a default.

Use lock-free Tokio owned halves for native TCP and Unix streams so tunnel I/O no longer takes the generic split mutex on every poll. Cover full-duplex traffic and write-half shutdown.

* perf(packet): preserve ownership across the Host seam

Introduce an opaque, move-only HostPacket that retains core packet storage while exposing only the raw IP payload. Clear private headers before handing storage back to a native TUN adapter.

Use an ownership-preserving bounded channel for native ingress and egress. Keep explicit copy adapters for Vec and WASI boundaries, and verify allocation identity, backpressure, shutdown, and end-to-end delivery.

* perf(udp): preserve packet ownership through sessions

Carry EasyTier tunnel packets through UDP session queues as owned values. Reuse the existing tunnel header for session framing instead of copying payloads into a second packet and rebuilding them on receive.

Keep completion delivery for the public datagram socket API while removing the unused completion channel from streaming tunnel sends. Avoid the unconditional receive-side clone before QUIC routing is known.

* perf(peer): publish packet filters as immutable snapshots

Replace per-packet async and synchronous registry locks with ArcSwap snapshots. Permanent filters now need no activity checks, while managed registrations retain explicit acquire/release visibility.

Closing a managed registration marks it inactive before atomically removing it. Existing snapshots keep in-flight filters alive, and registration mutations prune inactive entries while preserving newest-first order.

* perf(instance): give native hosts direct packet egress

Let the core create one bounded HostPacket channel and transfer its receiver directly to a PacketEgressHost during startup. Native TUN runtimes now consume that receiver without the intermediate PacketSink channel and forwarding task.

Keep PacketSinkEgress as the compatibility adapter for callback and test hosts, and make receiver installation one-shot across desktop, mobile, and disabled runtimes.

* perf(crypto): restore accelerated native AEAD backends

Move Ring and OpenSSL implementations behind the core Encryptor seam.
Portable builds continue selecting only supported backends.

Restore historical precedence: OpenSSL, Ring, then RustCrypto. Keep
backend availability consistent across secure transports and cover
fixed-nonce wire compatibility between implementations.

* perf(udp): receive native datagrams into owned buffers

Extend the portable UDP socket seam with an owned-datagram receive path.
Keep a compatible default for portable hosts. Native Unix sockets write
recvmsg output directly into the final BytesMut allocation.

This removes the per-packet stack-to-heap copy introduced by the portable
socket boundary without exposing native socket resources to core.

* perf(data-plane): remove portable hot-path overhead

Restore native throughput lost while generalizing the host and UDP
session layers.

Read packet policy once per send, update traffic counters through
registry guards, and preserve packet ownership while UDP dispatch
borrows stable session state.

Move UDP shutdown monitoring into a control task so forwarding avoids
a select future per packet. Bound native datagram storage to 8 KiB,
reject oversized sends, and drop truncated Unix receives.

Keep accelerated AEAD selection warning-free when portable crypto
features are also built. Cover session bounds, truncation, and idle
shutdown with regression tests.

* fix(udp): preserve portable datagram receive semantics

Keep the public portable receive capacity at the theoretical UDP
maximum instead of silently shrinking it to the native fast-path limit.

Apply the 8 KiB session boundary after a complete portable receive,
so Windows cannot turn an oversized datagram into a fatal listener
error and other adapters cannot dispatch a truncated prefix.

Cover dropping an oversized packet while the same portable socket
continues to deliver the following valid datagram.

* fix(ci): align feature gating with backend selection

Compile the Ring implementation in production only when OpenSSL is not
selected, while retaining it for cross-backend unit tests.

Remove stale test imports and assert UDP dispatch results so the strict
workspace Clippy job passes without suppressing diagnostics.

easytier-core/Cargo.toml

@@ -50,6 +50,7 @@ prost = "0.14.3"
 prost-types = "0.14.3"
 rand = "0.8.5"
 quanta = "0.12"
+ring = { version = "0.17", optional = true }
 rustls = { version = "0.23", default-features = false, features = ["ring", "std", "tls12"], optional = true }
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1"
@@ -79,11 +80,14 @@ zerocopy = { version = "0.7.32", features = ["derive", "simd"] }
 zstd = { version = "0.13", optional = true }
 aes-gcm = { version = "0.10.3", optional = true }
 chacha20poly1305 = { version = "0.10.1", optional = true }
+openssl = { version = "0.10", optional = true, features = ["vendored"] }
 
 [features]
 default = ["aes-gcm", "endpoint-discovery", "extended-services", "management", "tcp-hole-punch"]
 aes-gcm = ["dep:aes-gcm"]
 chacha20 = ["dep:chacha20poly1305"]
+openssl-crypto = ["dep:openssl"]
+ring-crypto = ["dep:ring"]
 config-write = []
 endpoint-discovery = [
     "dep:http-body-util",

easytier-core/src/gateway/dataplane/flow.rs

@@ -117,10 +117,21 @@ impl<V> FlowTable<V> {
         self.count.load(Ordering::Relaxed)
     }
 
+    /// Returns whether no flow is visible or being published.
+    ///
+    /// New entries reserve their count before they become visible, while
+    /// removals release their count after the entry is gone. Consequently a
+    /// zero observed here is a safe fast-path signal without inspecting every
+    /// DashMap shard.
+    pub fn is_idle(&self) -> bool {
+        self.count.load(Ordering::Acquire) == 0
+    }
+
     pub fn len(&self) -> usize {
         self.entries.len()
     }
 
+    #[cfg(test)]
     pub fn is_empty(&self) -> bool {
         self.entries.is_empty()
     }
@@ -241,13 +252,13 @@ impl<V> FlowTable<V> {
     fn increment_count(&self) -> FlowCountChange {
         let previous = self
             .count
-            .fetch_update(Ordering::Relaxed, Ordering::Relaxed, |count| {
+            .fetch_update(Ordering::AcqRel, Ordering::Acquire, |count| {
                 count.checked_add(1)
             })
-            .unwrap_or_else(|count| count);
+            .expect("flow count overflow");
         FlowCountChange {
             previous,
-            current: previous.saturating_add(1),
+            current: previous + 1,
         }
     }
 
@@ -283,13 +294,13 @@ impl<V> FlowTable<V> {
 
         let previous = self
             .count
-            .fetch_update(Ordering::Relaxed, Ordering::Relaxed, |count| {
-                Some(count.saturating_sub(delta))
+            .fetch_update(Ordering::AcqRel, Ordering::Acquire, |count| {
+                count.checked_sub(delta)
             })
-            .unwrap_or_else(|count| count);
+            .expect("flow count underflow");
         FlowCountChange {
             previous,
-            current: previous.saturating_sub(delta),
+            current: previous - delta,
         }
     }
 }
@@ -334,6 +345,7 @@ mod tests {
         assert!(!inserted.replaced);
         assert_eq!(inserted.count.previous, 0);
         assert_eq!(inserted.count.current, 1);
+        assert!(!table.is_idle());
         assert_eq!(table.with_entry(&entry, |value| *value), Some("first"));
 
         let replaced = table.insert(entry.clone(), "second");
@@ -346,6 +358,7 @@ mod tests {
         assert!(removed.removed);
         assert_eq!(removed.count.previous, 1);
         assert_eq!(removed.count.current, 0);
+        assert!(table.is_idle());
 
         let missing = table.remove(&entry);
         assert!(!missing.removed);

easytier-core/src/gateway/dataplane/mod.rs

@@ -208,8 +208,7 @@ where
     H: VirtualTcpSocketFactory + VirtualTcpListenerFactory + VirtualUdpSocketFactory,
 {
     async fn try_process_packet_from_peer(&self, packet: ZCPacket) -> Option<ZCPacket> {
-        let entry_count = self.entries.count();
-        if entry_count == 0 && self.entries.is_empty() {
+        if self.entries.is_idle() {
             if tracing::enabled!(tracing::Level::TRACE)
                 && let Some(hdr) = packet.peer_manager_header()
                 && matches!(
@@ -244,7 +243,7 @@ where
                         ?tcp_src_port,
                         ?tcp_dst_port,
                         ?tcp_flags,
-                        entry_count,
+                        entry_count = 0,
                         "data plane fast gate passed packet from peer"
                     );
                 } else {
@@ -252,7 +251,7 @@ where
                         packet_type = hdr.packet_type,
                         from_peer_id = hdr.from_peer_id.get(),
                         to_peer_id = hdr.to_peer_id.get(),
-                        entry_count,
+                        entry_count = 0,
                         "data plane fast gate passed non-ipv4 packet from peer"
                     );
                 }

easytier-core/src/gateway/dataplane/tests.rs

@@ -12,10 +12,11 @@ use super::*;
 use crate::{
     config::peers::PeerRuntimeSnapshot,
     config::{IpPrefix, NetworkIdentity},
-    host::testkit::TestHost,
-    peers::{
-        PacketRecvChanReceiver, create_packet_recv_chan, peer_manager::PortablePeerManagerConfig,
+    host::{
+        packet::{HostPacketReceiver, host_packet_channel},
+        testkit::TestHost,
     },
+    peers::peer_manager::PortablePeerManagerConfig,
     tunnel::ring::RingTunnelRegistry,
 };
 
@@ -49,7 +50,7 @@ fn test_gateway() -> Arc<DataPlaneRuntime<TestHost>> {
 struct DataPlaneEndpoint {
     gateway: Arc<DataPlaneRuntime<TestHost>>,
     peer_manager: Arc<PeerManagerCore>,
-    _packet_receiver: PacketRecvChanReceiver,
+    _packet_receiver: HostPacketReceiver,
     ip: cidr::Ipv4Inet,
 }
 
@@ -73,7 +74,7 @@ fn data_plane_endpoint(host: Arc<TestHost>, ip: cidr::Ipv4Inet) -> DataPlaneEndp
         crate::config::runtime::CoreRuntimeConfig::default(),
         Arc::new(peer_config.snapshot.clone()),
     );
-    let (packet_sender, packet_receiver) = create_packet_recv_chan();
+    let (packet_sender, packet_receiver) = host_packet_channel();
     let peer_manager = Arc::new(
         PeerManagerCore::new_portable_for_test(peer_config, packet_sender)
             .expect("build portable peer manager"),

easytier-core/src/gateway/proxy/tcp_proxy_service.rs

@@ -471,7 +471,12 @@ impl<R: TcpProxyRuntime + 'static, F: VirtualTcpListenerFactory, C: TcpProxyDest
 
             #[cfg(not(feature = "proxy-smoltcp-stack"))]
             tracing::error!("smoltcp packet received but proxy-smoltcp-stack is disabled");
-        } else if let Err(err) = self.peer_manager.get_nic_channel().send(packet).await {
+        } else if let Err(err) = self
+            .peer_manager
+            .get_nic_channel()
+            .send(crate::host::packet::HostPacket::from_core_packet(packet))
+            .await
+        {
             tracing::error!(?err, "send to nic failed");
         }
 

easytier-core/src/gateway/proxy/wrapped_transport.rs

@@ -462,7 +462,7 @@ mod tests {
         config::peers::{HostRoutingPolicy, PeerRuntimeConfig, PeerRuntimeSnapshot},
         config::runtime::CoreRuntimeConfig,
         config::{CoreConfig, NetworkIdentity, NodeConfig},
-        peers::{create_packet_recv_chan, peer_manager::PortablePeerManagerConfig},
+        peers::peer_manager::PortablePeerManagerConfig,
     };
 
     use super::*;
@@ -745,7 +745,7 @@ mod tests {
     }
 
     fn wrapped_transport_peer_manager() -> Arc<PeerManagerCore> {
-        let (packet_tx, _packet_rx) = create_packet_recv_chan();
+        let (packet_tx, _packet_rx) = crate::host::packet::host_packet_channel();
         Arc::new(
             PeerManagerCore::new_portable_for_test(
                 PortablePeerManagerConfig::new(PeerRuntimeConfig {

easytier-core/src/host/packet.rs

@@ -1,24 +1,143 @@
-use std::{io, sync::Arc, task::Poll};
+use std::{fmt, io, sync::Arc, task::Poll};
 
 use async_trait::async_trait;
 use tokio::sync::mpsc;
 
+use crate::packet::{ZCPacket, ZCPacketType};
+
 use super::socket::{HostOperationId, HostSocketRuntime};
 
-/// Receives raw IP packet bytes leaving the EasyTier peer graph.
+/// An owned raw IP packet crossing the Host packet seam.
+///
+/// The backing allocation and any core-private headroom stay opaque. Native
+/// adapters can move the packet without copying its payload, while ABI adapters
+/// can borrow [`Self::payload`] at the point where a copy is unavoidable.
+pub struct HostPacket {
+    inner: ZCPacket,
+}
+
+pub(crate) type HostPacketSender = mpsc::Sender<HostPacket>;
+
+/// The single bounded receive side for packets leaving a core instance.
+///
+/// Hosts own this receiver for the instance lifetime. Keeping the channel
+/// opaque prevents platform adapters from depending on core packet headers or
+/// replacing the bounded backpressure contract.
+pub struct HostPacketReceiver {
+    receiver: mpsc::Receiver<HostPacket>,
+}
+
+impl HostPacketReceiver {
+    /// Wraps a Host-owned bounded packet channel.
+    ///
+    /// This is intended for native adapters that already receive owned
+    /// [`HostPacket`] values from an in-process packet sink.
+    pub fn new(receiver: mpsc::Receiver<HostPacket>) -> Self {
+        Self { receiver }
+    }
+
+    pub async fn recv(&mut self) -> Option<HostPacket> {
+        self.receiver.recv().await
+    }
+}
+
+pub(crate) fn host_packet_channel() -> (HostPacketSender, HostPacketReceiver) {
+    let (sender, receiver) = mpsc::channel(128);
+    (sender, HostPacketReceiver::new(receiver))
+}
+
+impl HostPacket {
+    /// Copies raw IP bytes into core-owned packet storage.
+    ///
+    /// Native packet devices should use [`Self::from_tun_packet`] instead.
+    pub fn copy_from_payload(payload: &[u8]) -> Self {
+        Self {
+            inner: ZCPacket::new_with_payload(payload),
+        }
+    }
+
+    /// Moves a packet read from a native TUN adapter into the Host seam.
+    pub fn from_tun_packet(packet: ZCPacket) -> Self {
+        debug_assert_eq!(packet.packet_type(), ZCPacketType::NIC);
+        Self { inner: packet }
+    }
+
+    pub fn payload(&self) -> &[u8] {
+        self.inner.payload()
+    }
+
+    pub fn len(&self) -> usize {
+        self.inner.payload_len()
+    }
+
+    pub fn is_empty(&self) -> bool {
+        self.inner.payload().is_empty()
+    }
+
+    /// Moves a packet into a native TUN adapter without copying its payload.
+    ///
+    /// Core-private headers are cleared before the backing allocation crosses
+    /// the seam. The returned packet retains its original payload offset so the
+    /// native writer can reuse the existing headroom for packet-info framing.
+    pub fn into_tun_packet(mut self) -> ZCPacket {
+        let payload_offset = self.inner.payload_offset();
+        self.inner.mut_inner()[..payload_offset].fill(0);
+        self.inner
+    }
+
+    pub(crate) fn from_core_packet(packet: ZCPacket) -> Self {
+        Self { inner: packet }
+    }
+
+    pub(crate) fn into_core_packet(self) -> ZCPacket {
+        self.inner
+    }
+}
+
+impl fmt::Debug for HostPacket {
+    fn fmt(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result {
+        formatter
+            .debug_struct("HostPacket")
+            .field("len", &self.len())
+            .finish()
+    }
+}
+
+/// Receives owned IP packets leaving the EasyTier peer graph.
 ///
 /// The host decides whether packets go to a TUN device, a Go callback, or a
-/// different packet backend. Core's internal packet headers never cross this
-/// boundary, and core never performs platform I/O directly.
+/// different packet backend. Core's internal packet headers remain hidden, and
+/// core never performs platform I/O directly.
 #[async_trait]
 pub trait PacketSink: Send + Sync + 'static {
-    async fn write_packet(&self, packet: Vec<u8>) -> anyhow::Result<()>;
+    async fn write_packet(&self, packet: HostPacket) -> anyhow::Result<()>;
 }
 
 #[async_trait]
 impl PacketSink for mpsc::Sender<Vec<u8>> {
-    async fn write_packet(&self, packet: Vec<u8>) -> anyhow::Result<()> {
-        self.send(packet)
+    async fn write_packet(&self, packet: HostPacket) -> anyhow::Result<()> {
+        self.send(packet.payload().to_vec())
+            .await
+            .map_err(|_| anyhow::anyhow!("packet sink channel is closed"))
+    }
+}
+
+/// Ownership-preserving in-process adapter for native packet runtimes.
+pub struct HostPacketChannelSink {
+    sender: mpsc::Sender<HostPacket>,
+}
+
+impl HostPacketChannelSink {
+    pub fn new(sender: mpsc::Sender<HostPacket>) -> Self {
+        Self { sender }
+    }
+}
+
+#[async_trait]
+impl PacketSink for HostPacketChannelSink {
+    async fn write_packet(&self, packet: HostPacket) -> anyhow::Result<()> {
+        self.sender
+            .send(packet)
             .await
             .map_err(|_| anyhow::anyhow!("packet sink channel is closed"))
     }
@@ -97,9 +216,9 @@ impl<I> PacketSink for HostPacketSink<I>
 where
     I: HostPacketIo,
 {
-    async fn write_packet(&self, packet: Vec<u8>) -> anyhow::Result<()> {
+    async fn write_packet(&self, packet: HostPacket) -> anyhow::Result<()> {
         loop {
-            match self.io.try_write_packet(self.handle, &packet) {
+            match self.io.try_write_packet(self.handle, packet.payload()) {
                 Ok(()) => return Ok(()),
                 Err(error) if error.kind() == io::ErrorKind::WouldBlock => {
                     self.wait_writable().await?;
@@ -202,7 +321,9 @@ mod tests {
     #[tokio::test]
     async fn admits_complete_packet_without_readiness_wait() {
         let (_runtime, io, sink) = test_sink(true);
-        sink.write_packet(vec![1, 2, 3, 4]).await.unwrap();
+        sink.write_packet(HostPacket::copy_from_payload(&[1, 2, 3, 4]))
+            .await
+            .unwrap();
 
         let state = io.state.lock().unwrap();
         assert_eq!(
@@ -215,7 +336,10 @@ mod tests {
     #[tokio::test]
     async fn waits_for_capacity_then_admits_packet_once() {
         let (runtime, io, sink) = test_sink(false);
-        let task = tokio::spawn(async move { sink.write_packet(vec![5, 6, 7]).await });
+        let task = tokio::spawn(async move {
+            sink.write_packet(HostPacket::copy_from_payload(&[5, 6, 7]))
+                .await
+        });
         tokio::task::yield_now().await;
         assert!(io.state.lock().unwrap().packets.is_empty());
         assert_eq!(runtime.inner.wakers.len(), 1);
@@ -238,7 +362,7 @@ mod tests {
     async fn dropping_pending_waiter_removes_waker_and_host_state() {
         let (runtime, io, sink) = test_sink(false);
         let operation = {
-            let mut write = Box::pin(sink.write_packet(vec![7, 8]));
+            let mut write = Box::pin(sink.write_packet(HostPacket::copy_from_payload(&[7, 8])));
             assert!(futures::poll!(&mut write).is_pending());
             assert_eq!(runtime.inner.wakers.len(), 1);
             let operation = io.waiter();
@@ -257,7 +381,7 @@ mod tests {
     async fn dropping_ready_waiter_does_not_admit_packet() {
         let (runtime, io, sink) = test_sink(false);
         let operation = {
-            let mut write = Box::pin(sink.write_packet(vec![8, 9]));
+            let mut write = Box::pin(sink.write_packet(HostPacket::copy_from_payload(&[8, 9])));
             assert!(futures::poll!(&mut write).is_pending());
             let operation = io.waiter();
             io.set_writable();
@@ -272,10 +396,46 @@ mod tests {
             assert!(state.waiters.is_empty());
             assert_eq!(state.cancelled, vec![operation]);
         }
-        sink.write_packet(vec![8, 9]).await.unwrap();
+        sink.write_packet(HostPacket::copy_from_payload(&[8, 9]))
+            .await
+            .unwrap();
         assert_eq!(
             io.state.lock().unwrap().packets,
             vec![(HostPacketSinkHandle(41), vec![8, 9])]
         );
     }
+
+    #[test]
+    fn native_packet_round_trip_preserves_allocation_and_hides_headers() {
+        let mut core_packet = ZCPacket::new_with_payload(b"owned payload");
+        let payload_offset = core_packet.payload_offset();
+        core_packet.mut_inner()[..payload_offset].fill(0xa5);
+        let payload_pointer = core_packet.payload().as_ptr();
+
+        let host_packet = HostPacket::from_core_packet(core_packet);
+        assert_eq!(host_packet.payload().as_ptr(), payload_pointer);
+        assert_eq!(host_packet.payload(), b"owned payload");
+
+        let tun_packet = host_packet.into_tun_packet();
+        assert_eq!(tun_packet.payload().as_ptr(), payload_pointer);
+        assert_eq!(tun_packet.payload(), b"owned payload");
+        assert!(
+            tun_packet.inner()[..payload_offset]
+                .iter()
+                .all(|byte| *byte == 0)
+        );
+    }
+
+    #[test]
+    fn tun_packet_moves_into_core_without_copying_payload() {
+        let tun_packet = ZCPacket::new_with_payload(b"from tun");
+        let payload_pointer = tun_packet.payload().as_ptr();
+
+        let host_packet = HostPacket::from_tun_packet(tun_packet);
+        assert_eq!(host_packet.payload().as_ptr(), payload_pointer);
+        let core_packet = host_packet.into_core_packet();
+
+        assert_eq!(core_packet.payload().as_ptr(), payload_pointer);
+        assert_eq!(core_packet.payload(), b"from tun");
+    }
 }

easytier-core/src/instance/lifecycle.rs

@@ -103,9 +103,13 @@ where
             .await;
 
         self.start_listener().await?;
-        if let Some(packet_egress) = &self.packet_egress {
-            packet_egress.start()?;
-        }
+        let packet_receiver = self
+            .packet_receiver
+            .lock()
+            .await
+            .take()
+            .ok_or_else(|| anyhow::anyhow!("packet egress is one-shot and already started"))?;
+        self.packet_egress.start(packet_receiver).await?;
         self.peer_manager.run().await.map_err(anyhow::Error::from)?;
         self.direct.run();
         #[cfg(feature = "tcp-hole-punch")]
@@ -184,9 +188,7 @@ where
         // before clearing PeerManager resources.
         self.instance_runtime.shutdown().await;
         self.peer_manager.clear_resources().await;
-        if let Some(packet_egress) = &self.packet_egress {
-            packet_egress.stop().await;
-        }
+        self.packet_egress.stop().await;
     }
 
     /// Starts the complete instance through one serial composition path.

easytier-core/src/instance/mod.rs

@@ -60,7 +60,10 @@ use crate::{
     },
     events::CoreEventSink,
     gateway::dhcp::DhcpIpv4Host,
-    host::dns::{DnsRecordResolver, DnsResolver},
+    host::{
+        dns::{DnsRecordResolver, DnsResolver},
+        packet::{HostPacketReceiver, PacketSink, host_packet_channel},
+    },
     listener::{
         AcceptedSocketHandler, ExternalListenerFactory, ExternalListenerRequest, ListenerFactory,
         RunningListenerRegistry,
@@ -74,7 +77,6 @@ use crate::{
     peers::{
         admission::{PeerAcceptedTunnelHandler, RawAcceptedTransportHandler},
         context::PeerStunInfoSource,
-        create_packet_recv_chan,
         credential_manager::CredentialStorage,
         peer_manager::{PeerManagerCore, PortablePeerManagerConfig},
         public_ipv6::{CorePublicIpv6Runtime, PublicIpv6Host},
@@ -108,12 +110,12 @@ use crate::gateway::vpn_portal::VpnPortalModule;
 use crate::gateway::{
     DataPlaneRuntime, DataPlaneSession, PortForwardAdapter, Socks5GatewayAdapter,
 };
-use crate::host::packet::PacketSink;
 #[cfg(feature = "public-ipv6-provider")]
 use crate::peers::public_ipv6::provider::PublicIpv6ProviderRuntime;
 pub use config::CoreInstanceHostConfig;
 use management_state::ManagementState;
-use packet_io::PacketEgress;
+pub use packet_io::PacketEgressHost;
+use packet_io::PacketSinkEgress;
 pub use packet_plane::CorePacketPlane;
 
 /// Complete Host capability set required by one portable core instance.
@@ -277,7 +279,7 @@ where
     stun_override: Option<Arc<dyn StunSocketMapper<<H as VirtualUdpSocketFactory>::Socket>>>,
     dns: Arc<dyn StunDnsRuntime>,
     process_runtime: Arc<CoreProcessRuntime>,
-    packet_sink: Arc<dyn PacketSink>,
+    pub packet_egress: Arc<dyn PacketEgressHost>,
     pub instance_runtime: Arc<dyn InstanceRuntimeHost>,
     pub events: Arc<dyn CoreEventSink>,
     pub credential_storage: Option<Arc<dyn CredentialStorage>>,
@@ -313,6 +315,22 @@ where
         dns: Arc<dyn StunDnsRuntime>,
         packet_sink: Arc<dyn PacketSink>,
         process_runtime: Arc<CoreProcessRuntime>,
+    ) -> Self {
+        Self::new_with_packet_egress(
+            host,
+            dns,
+            Arc::new(PacketSinkEgress::new(packet_sink)),
+            process_runtime,
+        )
+    }
+
+    /// Creates a host bundle whose packet runtime owns the core's single
+    /// bounded egress receiver directly.
+    pub fn new_with_packet_egress(
+        host: Arc<H>,
+        dns: Arc<dyn StunDnsRuntime>,
+        packet_egress: Arc<dyn PacketEgressHost>,
+        process_runtime: Arc<CoreProcessRuntime>,
     ) -> Self {
         Self {
             host,
@@ -321,7 +339,7 @@ where
             stun_override: None,
             dns,
             process_runtime,
-            packet_sink,
+            packet_egress,
             instance_runtime: Arc::new(()),
             events: Arc::new(()),
             credential_storage: None,
@@ -395,7 +413,8 @@ where
     proxy_cidr_monitor: ProxyCidrMonitorRuntime,
     #[cfg(feature = "dhcp-ipv4")]
     dhcp_ipv4: DhcpIpv4Runtime,
-    pub(super) packet_egress: Option<PacketEgress>,
+    pub(super) packet_egress: Arc<dyn PacketEgressHost>,
+    pub(super) packet_receiver: Mutex<Option<HostPacketReceiver>>,
     pub(super) peer_center: Arc<PeerCenterInstance>,
     #[cfg(feature = "public-ipv6-provider")]
     public_ipv6_provider: PublicIpv6ProviderRuntime,
@@ -463,7 +482,7 @@ where
     ) -> anyhow::Result<Arc<Self>> {
         let initial_acl = validate_core_instance_config(&config)?;
         let instance_name = config.instance_name;
-        let (packet_tx, packet_rx) = create_packet_recv_chan();
+        let (packet_tx, packet_rx) = host_packet_channel();
         let runtime_config = CoreRuntimeConfigStore::new(
             config.connectivity.runtime.clone(),
             Arc::new(config.peer.snapshot.clone()),
@@ -516,7 +535,7 @@ where
                 stun_override: _,
             dns,
             process_runtime,
-            packet_sink,
+            packet_egress,
             instance_runtime,
             events,
             credential_storage: _,
@@ -779,7 +798,8 @@ where
             proxy_cidr_monitor,
             #[cfg(feature = "dhcp-ipv4")]
             dhcp_ipv4: DhcpIpv4Runtime::new(),
-            packet_egress: Some(PacketEgress::new(packet_rx, packet_sink)),
+            packet_egress,
+            packet_receiver: Mutex::new(Some(packet_rx)),
             peer_center,
             #[cfg(feature = "public-ipv6-provider")]
             public_ipv6_provider,
@@ -887,6 +907,7 @@ where
     fn drop(&mut self) {
         self.cancel.cancel();
         self.instance_runtime.request_shutdown();
+        self.packet_egress.request_stop();
     }
 }
 

easytier-core/src/instance/packet_io.rs

@@ -3,10 +3,9 @@ use std::{
     sync::{Arc, Mutex},
 };
 
-use tokio::{sync::mpsc, task::JoinHandle};
+use tokio::task::JoinHandle;
 
-use crate::host::packet::PacketSink;
-use crate::packet::ZCPacket;
+use crate::host::packet::{HostPacketReceiver, PacketSink};
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub(crate) struct IpPacketMeta {
@@ -64,32 +63,41 @@ fn parse_ipv6_packet(packet: &[u8]) -> anyhow::Result<IpPacketMeta> {
     })
 }
 
-pub(crate) struct PacketEgress {
-    receiver: Mutex<Option<mpsc::Receiver<ZCPacket>>>,
-    sink: Arc<dyn PacketSink>,
-    task: Mutex<Option<JoinHandle<()>>>,
+#[async_trait::async_trait]
+pub trait PacketEgressHost: Send + Sync + 'static {
+    async fn start(&self, receiver: HostPacketReceiver) -> anyhow::Result<()>;
+
+    async fn stop(&self);
+
+    fn request_stop(&self) {}
 }
 
-impl PacketEgress {
-    pub(crate) fn new(receiver: mpsc::Receiver<ZCPacket>, sink: Arc<dyn PacketSink>) -> Self {
+pub(crate) struct PacketSinkEgress {
+    sink: Arc<dyn PacketSink>,
+    task: Mutex<Option<JoinHandle<()>>>,
+    started: std::sync::atomic::AtomicBool,
+}
+
+impl PacketSinkEgress {
+    pub(crate) fn new(sink: Arc<dyn PacketSink>) -> Self {
         Self {
-            receiver: Mutex::new(Some(receiver)),
             sink,
             task: Mutex::new(None),
+            started: std::sync::atomic::AtomicBool::new(false),
         }
     }
+}
 
-    pub(crate) fn start(&self) -> anyhow::Result<()> {
-        let mut receiver = self
-            .receiver
-            .lock()
-            .unwrap()
-            .take()
-            .ok_or_else(|| anyhow::anyhow!("packet egress is one-shot and already started"))?;
+#[async_trait::async_trait]
+impl PacketEgressHost for PacketSinkEgress {
+    async fn start(&self, mut receiver: HostPacketReceiver) -> anyhow::Result<()> {
+        if self.started.swap(true, std::sync::atomic::Ordering::AcqRel) {
+            anyhow::bail!("packet egress is one-shot and already started");
+        }
         let sink = self.sink.clone();
         let task = tokio::spawn(async move {
             while let Some(packet) = receiver.recv().await {
-                if let Err(error) = sink.write_packet(packet.payload().to_vec()).await {
+                if let Err(error) = sink.write_packet(packet).await {
                     tracing::warn!(?error, "host packet sink rejected an egress packet");
                 }
             }
@@ -98,17 +106,22 @@ impl PacketEgress {
         Ok(())
     }
 
-    pub(crate) async fn stop(&self) {
+    async fn stop(&self) {
         let task = self.task.lock().unwrap().take();
         if let Some(task) = task {
             task.abort();
             let _ = task.await;
         }
-        self.receiver.lock().unwrap().take();
+    }
+
+    fn request_stop(&self) {
+        if let Some(task) = self.task.lock().unwrap().take() {
+            task.abort();
+        }
     }
 }
 
-impl Drop for PacketEgress {
+impl Drop for PacketSinkEgress {
     fn drop(&mut self) {
         if let Some(task) = self.task.lock().unwrap().take() {
             task.abort();
@@ -119,6 +132,7 @@ impl Drop for PacketEgress {
 #[cfg(test)]
 mod tests {
     use crate::foundation::time::{Duration, timeout};
+    use crate::host::packet::{HostPacket, host_packet_channel};
 
     use super::*;
 
@@ -169,13 +183,13 @@ mod tests {
 
     #[tokio::test]
     async fn packet_egress_forwards_to_host_sink_and_joins_on_stop() {
-        let (core_tx, core_rx) = mpsc::channel(1);
-        let (host_tx, mut host_rx) = mpsc::channel(1);
-        let egress = PacketEgress::new(core_rx, Arc::new(host_tx));
-        egress.start().unwrap();
+        let (core_tx, core_rx) = host_packet_channel();
+        let (host_tx, mut host_rx) = tokio::sync::mpsc::channel(1);
+        let egress = PacketSinkEgress::new(Arc::new(host_tx));
+        egress.start(core_rx).await.unwrap();
 
         core_tx
-            .send(ZCPacket::new_with_payload(b"packet"))
+            .send(HostPacket::copy_from_payload(b"packet"))
             .await
             .unwrap();
         let packet = timeout(Duration::from_secs(1), host_rx.recv())
@@ -185,6 +199,7 @@ mod tests {
         assert_eq!(packet, b"packet");
 
         egress.stop().await;
-        assert!(egress.start().is_err());
+        let (_core_tx, core_rx) = host_packet_channel();
+        assert!(egress.start(core_rx).await.is_err());
     }
 }

easytier-core/src/instance/packet_plane.rs

@@ -6,6 +6,7 @@ use crate::{
     config::runtime::CoreRuntimeConfigStore,
     gateway::magic_dns::{MagicDnsRouteSnapshot, MagicDnsRouteSource},
     gateway::proxy::cidr_monitor::{ProxyCidrDiff, collect_proxy_cidr_diff},
+    host::packet::HostPacket,
     peers::peer_manager::PeerManagerCore,
 };
 
@@ -40,30 +41,22 @@ impl CorePacketPlane {
         }
     }
 
-    pub async fn send_ip_packet(&self, packet: Vec<u8>) -> anyhow::Result<()> {
-        let meta = parse_ip_packet(&packet)?;
+    pub async fn send_ip_packet(&self, packet: HostPacket) -> anyhow::Result<()> {
+        let meta = parse_ip_packet(packet.payload())?;
         let source_is_local = self.peer_manager.is_local_virtual_ip(&meta.source);
         if matches!(meta.source, IpAddr::V6(ip) if ip.is_unicast_link_local()) && !source_is_local {
             return Ok(());
         }
         self.peer_manager
-            .send_msg_by_ip(
-                crate::packet::ZCPacket::new_with_payload(&packet),
-                meta.destination,
-                source_is_local,
-            )
+            .send_msg_by_ip(packet.into_core_packet(), meta.destination, source_is_local)
             .await
             .map_err(Into::into)
     }
 
-    pub async fn send_local_ip_packet(&self, packet: Vec<u8>) -> anyhow::Result<()> {
-        let destination = parse_ip_packet(&packet)?.destination;
+    pub async fn send_local_ip_packet(&self, packet: HostPacket) -> anyhow::Result<()> {
+        let destination = parse_ip_packet(packet.payload())?.destination;
         self.peer_manager
-            .send_msg_by_ip(
-                crate::packet::ZCPacket::new_with_payload(&packet),
-                destination,
-                true,
-            )
+            .send_msg_by_ip(packet.into_core_packet(), destination, true)
             .await
             .map_err(Into::into)
     }

easytier-core/src/packet/mod.rs

@@ -663,6 +663,16 @@ impl ZCPacket {
             + UDP_TUNNEL_HEADER_SIZE..]
     }
 
+    pub fn udp_payload_bytes(mut self) -> BytesMut {
+        self.inner.advance(
+            self.packet_type
+                .get_packet_offsets()
+                .udp_tunnel_header_offset
+                + UDP_TUNNEL_HEADER_SIZE,
+        );
+        self.inner
+    }
+
     pub fn payload_len(&self) -> usize {
         self.inner.len() - self.payload_offset()
     }

easytier-core/src/peers/conn/peer_session.rs

@@ -421,7 +421,18 @@ mod tests {
     }
 
     #[test]
-    #[cfg(all(feature = "aes-gcm", feature = "chacha20"))]
+    #[cfg(all(
+        any(
+            feature = "aes-gcm",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        ),
+        any(
+            feature = "chacha20",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        )
+    ))]
     fn peer_session_supports_asymmetric_algorithms() {
         let a: PeerId = 10;
         let b: PeerId = 20;

easytier-core/src/peers/context.rs

@@ -528,6 +528,27 @@ impl Default for TrustedKeyMapManager {
 /// `PeerContext` is intentionally scoped to `easytier-core::peers`; other core
 /// modules should depend on their own narrow DTOs or traits instead of treating
 /// this as a core-wide global context.
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
+pub(crate) struct PeerPacketPolicy {
+    pub(crate) disable_relay_data: bool,
+    pub(crate) p2p_only: bool,
+    pub(crate) latency_first: bool,
+    pub(crate) disable_p2p: bool,
+    pub(crate) lazy_p2p: bool,
+}
+
+impl PeerPacketPolicy {
+    fn from_flags(flags: &FlagsInConfig) -> Self {
+        Self {
+            disable_relay_data: flags.disable_relay_data,
+            p2p_only: flags.p2p_only,
+            latency_first: flags.latency_first && !flags.p2p_only,
+            disable_p2p: flags.disable_p2p,
+            lazy_p2p: flags.lazy_p2p,
+        }
+    }
+}
+
 pub(crate) trait PeerContext: Send + Sync {
     fn host_routing_policy(&self) -> HostRoutingPolicy {
         HostRoutingPolicy::default()
@@ -543,8 +564,12 @@ pub(crate) trait PeerContext: Send + Sync {
         FlagsInConfig::default()
     }
 
+    fn packet_policy(&self) -> PeerPacketPolicy {
+        PeerPacketPolicy::from_flags(&self.flags())
+    }
+
     fn disable_relay_data(&self) -> bool {
-        self.flags().disable_relay_data
+        self.packet_policy().disable_relay_data
     }
 
     fn secure_mode(&self) -> Option<SecureModeConfig> {
@@ -583,15 +608,6 @@ pub(crate) trait PeerContext: Send + Sync {
         }
     }
 
-    fn p2p_only(&self) -> bool {
-        self.flags().p2p_only
-    }
-
-    fn latency_first(&self) -> bool {
-        let flags = self.flags();
-        flags.latency_first && !flags.p2p_only
-    }
-
     fn proxy_cidrs(&self) -> Vec<Ipv4Cidr> {
         Vec::new()
     }
@@ -746,6 +762,10 @@ impl PeerContext for CorePeerContext {
         self.snapshot().flags.clone()
     }
 
+    fn packet_policy(&self) -> PeerPacketPolicy {
+        PeerPacketPolicy::from_flags(&self.snapshot().flags)
+    }
+
     fn host_routing_policy(&self) -> HostRoutingPolicy {
         self.snapshot().runtime.host_routing
     }

easytier-core/src/peers/peer_manager.rs

@@ -9,8 +9,8 @@ use std::{
 };
 
 use anyhow::Context;
+use arc_swap::ArcSwap;
 use dashmap::DashMap;
-use parking_lot::RwLock as SyncRwLock;
 use quanta::Instant;
 use serde::{Deserialize, Serialize};
 use tokio::sync::{
@@ -26,6 +26,7 @@ use crate::{
     config::{P2pPolicyFlags, PeerId, ProxyNetworkConfig},
     events::CoreEventSink,
     foundation::task::ExternalTaskSignal,
+    host::packet::{HostPacket, HostPacketSender},
     packet::{
         CompressorAlgo, PacketType, ZCPacket,
         compressor::{Compressor as _, DefaultCompressor},
@@ -42,8 +43,7 @@ use crate::{
 };
 
 use super::{
-    BoxNicPacketFilter, BoxPeerPacketFilter, PacketRecvChan, PacketRecvChanReceiver,
-    PeerPacketFilter,
+    BoxNicPacketFilter, BoxPeerPacketFilter, PacketRecvChanReceiver, PeerPacketFilter,
     acl::AclFilter,
     conn::{
         peer_conn::{PeerConn, PeerConnId},
@@ -52,7 +52,7 @@ use super::{
     },
     context::{
         ArcPeerContext, CorePeerContext, CorePeerContextAdapters, NetworkIdentity, PeerContext,
-        PeerStunInfoSource,
+        PeerPacketPolicy, PeerStunInfoSource,
     },
     credential_manager::{CredentialManager, CredentialStorage},
     error::Error,
@@ -401,7 +401,7 @@ pub(crate) async fn close_untrusted_credential_peers<F>(
 }
 
 struct NicPacketProcessor {
-    nic_channel: PacketRecvChan,
+    nic_channel: HostPacketSender,
 }
 
 #[async_trait::async_trait]
@@ -420,7 +420,10 @@ impl PeerPacketFilter for NicPacketProcessor {
                 return None;
             }
             tracing::trace!(?packet, "send packet to nic channel");
-            let _ = self.nic_channel.send(packet).await;
+            let _ = self
+                .nic_channel
+                .send(HostPacket::from_core_packet(packet))
+                .await;
             None
         } else {
             Some(packet)
@@ -448,26 +451,58 @@ impl PeerPacketFilter for PeerRpcPacketProcessor {
     }
 }
 
+#[derive(Clone)]
 pub(crate) struct PeerPipelineEntry {
-    active: Arc<AtomicBool>,
-    filter: Arc<SyncRwLock<Option<Arc<dyn PeerPacketFilter + Send + Sync>>>>,
+    active: Option<Arc<AtomicBool>>,
+    filter: Arc<dyn PeerPacketFilter + Send + Sync>,
 }
 
+#[derive(Clone)]
 pub(crate) struct NicPipelineEntry {
-    active: Arc<AtomicBool>,
-    filter: Arc<SyncRwLock<Option<Arc<dyn super::NicPacketFilter + Send + Sync>>>>,
+    active: Option<Arc<AtomicBool>>,
+    filter: Arc<dyn super::NicPacketFilter + Send + Sync>,
+}
+
+type PeerPacketPipeline = Arc<ArcSwap<Vec<PeerPipelineEntry>>>;
+type NicPacketPipeline = Arc<ArcSwap<Vec<NicPipelineEntry>>>;
+
+impl PeerPipelineEntry {
+    fn filter_if_active(&self) -> Option<&Arc<dyn PeerPacketFilter + Send + Sync>> {
+        if self
+            .active
+            .as_ref()
+            .is_some_and(|active| !active.load(Ordering::Acquire))
+        {
+            return None;
+        }
+        Some(&self.filter)
+    }
+}
+
+impl NicPipelineEntry {
+    fn filter_if_active(&self) -> Option<&Arc<dyn super::NicPacketFilter + Send + Sync>> {
+        if self
+            .active
+            .as_ref()
+            .is_some_and(|active| !active.load(Ordering::Acquire))
+        {
+            return None;
+        }
+        Some(&self.filter)
+    }
 }
 
 #[derive(Clone)]
 pub(crate) struct PipelineRegistrationGuard {
     active: Arc<AtomicBool>,
-    release_filter: Arc<dyn Fn() + Send + Sync>,
+    unregister: Arc<dyn Fn() + Send + Sync>,
 }
 
 impl PipelineRegistrationGuard {
     pub fn close(&self) {
-        self.active.store(false, Ordering::Release);
-        (self.release_filter)();
+        if self.active.swap(false, Ordering::AcqRel) {
+            (self.unregister)();
+        }
     }
 }
 
@@ -477,36 +512,52 @@ impl Drop for PipelineRegistrationGuard {
     }
 }
 
-fn permanent_peer_pipeline_entry(filter: BoxPeerPacketFilter) -> Arc<PeerPipelineEntry> {
-    Arc::new(PeerPipelineEntry {
-        active: Arc::new(AtomicBool::new(true)),
-        filter: Arc::new(SyncRwLock::new(Some(Arc::from(filter)))),
-    })
+fn permanent_peer_pipeline_entry(filter: BoxPeerPacketFilter) -> PeerPipelineEntry {
+    PeerPipelineEntry {
+        active: None,
+        filter: Arc::from(filter),
+    }
 }
 
-fn permanent_nic_pipeline_entry(filter: BoxNicPacketFilter) -> Arc<NicPipelineEntry> {
-    Arc::new(NicPipelineEntry {
-        active: Arc::new(AtomicBool::new(true)),
-        filter: Arc::new(SyncRwLock::new(Some(Arc::from(filter)))),
-    })
+fn permanent_nic_pipeline_entry(filter: BoxNicPacketFilter) -> NicPipelineEntry {
+    NicPipelineEntry {
+        active: None,
+        filter: Arc::from(filter),
+    }
 }
 
 fn managed_peer_pipeline_entry(
     filter: BoxPeerPacketFilter,
-) -> (Arc<PeerPipelineEntry>, PipelineRegistrationGuard) {
+    pipeline: &PeerPacketPipeline,
+) -> (PeerPipelineEntry, PipelineRegistrationGuard) {
     let active = Arc::new(AtomicBool::new(true));
-    let filter = Arc::new(SyncRwLock::new(Some(Arc::from(filter))));
-    let release_filter = filter.clone();
+    let weak_pipeline = Arc::downgrade(pipeline);
+    let registration = active.clone();
     (
-        Arc::new(PeerPipelineEntry {
-            active: active.clone(),
-            filter,
-        }),
+        PeerPipelineEntry {
+            active: Some(active.clone()),
+            filter: Arc::from(filter),
+        },
         PipelineRegistrationGuard {
             active,
-            release_filter: Arc::new(move || {
-                let filter = release_filter.write().take();
-                drop(filter);
+            unregister: Arc::new(move || {
+                let Some(pipeline) = weak_pipeline.upgrade() else {
+                    return;
+                };
+                pipeline.rcu(|current| {
+                    Arc::new(
+                        current
+                            .iter()
+                            .filter(|entry| {
+                                entry
+                                    .active
+                                    .as_ref()
+                                    .is_none_or(|active| !Arc::ptr_eq(active, &registration))
+                            })
+                            .cloned()
+                            .collect(),
+                    )
+                });
             }),
         },
     )
@@ -515,63 +566,101 @@ fn managed_peer_pipeline_entry(
 #[cfg(any(feature = "proxy-packet", test))]
 fn managed_nic_pipeline_entry(
     filter: BoxNicPacketFilter,
-) -> (Arc<NicPipelineEntry>, PipelineRegistrationGuard) {
+    pipeline: &NicPacketPipeline,
+) -> (NicPipelineEntry, PipelineRegistrationGuard) {
     let active = Arc::new(AtomicBool::new(true));
-    let filter = Arc::new(SyncRwLock::new(Some(Arc::from(filter))));
-    let release_filter = filter.clone();
+    let weak_pipeline = Arc::downgrade(pipeline);
+    let registration = active.clone();
     (
-        Arc::new(NicPipelineEntry {
-            active: active.clone(),
-            filter,
-        }),
+        NicPipelineEntry {
+            active: Some(active.clone()),
+            filter: Arc::from(filter),
+        },
         PipelineRegistrationGuard {
             active,
-            release_filter: Arc::new(move || {
-                let filter = release_filter.write().take();
-                drop(filter);
+            unregister: Arc::new(move || {
+                let Some(pipeline) = weak_pipeline.upgrade() else {
+                    return;
+                };
+                pipeline.rcu(|current| {
+                    Arc::new(
+                        current
+                            .iter()
+                            .filter(|entry| {
+                                entry
+                                    .active
+                                    .as_ref()
+                                    .is_none_or(|active| !Arc::ptr_eq(active, &registration))
+                            })
+                            .cloned()
+                            .collect(),
+                    )
+                });
             }),
         },
     )
 }
 
-#[cfg(any(feature = "proxy-packet", test))]
-async fn remove_managed_nic_pipeline_entry(
-    pipeline: &RwLock<Vec<Arc<NicPipelineEntry>>>,
-    registration: &PipelineRegistrationGuard,
-) {
-    registration.close();
-    pipeline
-        .write()
-        .await
-        .retain(|entry| !Arc::ptr_eq(&entry.active, &registration.active));
+fn append_peer_pipeline(pipeline: &PeerPacketPipeline, entry: PeerPipelineEntry) {
+    pipeline.rcu(|current| {
+        let mut next = Vec::with_capacity(current.len() + 1);
+        next.extend(
+            current
+                .iter()
+                .filter(|entry| {
+                    entry
+                        .active
+                        .as_ref()
+                        .is_none_or(|active| active.load(Ordering::Acquire))
+                })
+                .cloned(),
+        );
+        next.push(entry.clone());
+        Arc::new(next)
+    });
+}
+
+fn append_nic_pipeline(pipeline: &NicPacketPipeline, entry: NicPipelineEntry) {
+    pipeline.rcu(|current| {
+        let mut next = Vec::with_capacity(current.len() + 1);
+        next.extend(
+            current
+                .iter()
+                .filter(|entry| {
+                    entry
+                        .active
+                        .as_ref()
+                        .is_none_or(|active| active.load(Ordering::Acquire))
+                })
+                .cloned(),
+        );
+        next.push(entry.clone());
+        Arc::new(next)
+    });
 }
 
 async fn init_packet_process_pipeline(
-    peer_packet_process_pipeline: &RwLock<Vec<Arc<PeerPipelineEntry>>>,
-    nic_channel: PacketRecvChan,
+    peer_packet_process_pipeline: &PeerPacketPipeline,
+    nic_channel: HostPacketSender,
     peer_rpc_tspt_sender: UnboundedSender<ZCPacket>,
 ) {
     // for tun/tap ip/eth packet.
-    peer_packet_process_pipeline
-        .write()
-        .await
-        .push(permanent_peer_pipeline_entry(Box::new(
-            NicPacketProcessor { nic_channel },
-        )));
+    append_peer_pipeline(
+        peer_packet_process_pipeline,
+        permanent_peer_pipeline_entry(Box::new(NicPacketProcessor { nic_channel })),
+    );
 
     // for peer rpc packet
-    peer_packet_process_pipeline
-        .write()
-        .await
-        .push(permanent_peer_pipeline_entry(Box::new(
-            PeerRpcPacketProcessor {
-                peer_rpc_tspt_sender,
-            },
-        )));
+    append_peer_pipeline(
+        peer_packet_process_pipeline,
+        permanent_peer_pipeline_entry(Box::new(PeerRpcPacketProcessor {
+            peer_rpc_tspt_sender,
+        })),
+    );
 }
 
 async fn add_route<T>(
-    peer_packet_process_pipeline: &RwLock<Vec<Arc<PeerPipelineEntry>>>,
+    peer_packet_process_pipeline: &PeerPacketPipeline,
     peers: Arc<PeerMap>,
     foreign_network_client: Arc<ForeignNetworkClient>,
     foreign_network_manager: Arc<ForeignNetworkManager>,
@@ -581,10 +670,10 @@ async fn add_route<T>(
     T: Route + PeerPacketFilter + Send + Sync + 'static,
 {
     // for route
-    peer_packet_process_pipeline
-        .write()
-        .await
-        .push(permanent_peer_pipeline_entry(Box::new(route.clone())));
+    append_peer_pipeline(
+        peer_packet_process_pipeline,
+        permanent_peer_pipeline_entry(Box::new(route.clone())),
+    );
 
     let _route_id = route
         .open(Box::new(PeerManagerRouteInterface {
@@ -614,9 +703,9 @@ pub struct PeerManagerCore {
     peers: Arc<PeerMap>,
     peer_rpc_mgr: Arc<super::peer_rpc::PeerRpcManager>,
     peer_rpc_tspt: Arc<RpcTransport>,
-    peer_packet_process_pipeline: Arc<RwLock<Vec<Arc<PeerPipelineEntry>>>>,
-    nic_packet_process_pipeline: Arc<RwLock<Vec<Arc<NicPipelineEntry>>>>,
-    nic_channel: PacketRecvChan,
+    peer_packet_process_pipeline: PeerPacketPipeline,
+    nic_packet_process_pipeline: NicPacketPipeline,
+    nic_channel: HostPacketSender,
     route_algo_inst: RouteAlgoInst,
     foreign_network_client: Arc<ForeignNetworkClient>,
     foreign_network_manager: Arc<ForeignNetworkManager>,
@@ -671,7 +760,7 @@ impl PeerManagerCore {
         mut config: PortablePeerManagerConfig,
         runtime_config: CoreRuntimeConfigStore,
         stun_info_source: Arc<dyn PeerStunInfoSource>,
-        nic_channel: PacketRecvChan,
+        nic_channel: HostPacketSender,
         public_ipv6_runtime: Arc<CorePublicIpv6Runtime>,
         events: Arc<dyn CoreEventSink>,
         credential_storage: Option<Arc<dyn CredentialStorage>>,
@@ -800,7 +889,7 @@ impl PeerManagerCore {
         my_peer_id: PeerId,
         core_context: Arc<CorePeerContext>,
         public_ipv6_runtime: Arc<dyn PublicIpv6Runtime>,
-        nic_channel: PacketRecvChan,
+        nic_channel: HostPacketSender,
         encryptor: Arc<dyn Encryptor + 'static>,
         is_secure_mode_enabled: bool,
         data_compress_algo: CompressorAlgo,
@@ -928,8 +1017,8 @@ impl PeerManagerCore {
                 }
             },
         ));
-        let peer_packet_process_pipeline = Arc::new(RwLock::new(Vec::new()));
-        let nic_packet_process_pipeline = Arc::new(RwLock::new(Vec::new()));
+        let peer_packet_process_pipeline = Arc::new(ArcSwap::from_pointee(Vec::new()));
+        let nic_packet_process_pipeline = Arc::new(ArcSwap::from_pointee(Vec::new()));
         let exit_nodes = Arc::new(RwLock::new(exit_nodes));
         let relay_peer_map = super::relay_peer_map::new_relay_peer_map(
             peers.clone(),
@@ -1171,7 +1260,7 @@ impl PeerManagerCore {
         self.peer_session_store.clone()
     }
 
-    pub fn get_nic_channel(&self) -> PacketRecvChan {
+    pub(crate) fn get_nic_channel(&self) -> HostPacketSender {
         self.nic_channel.clone()
     }
 
@@ -1271,28 +1360,27 @@ impl PeerManagerCore {
 
     pub async fn add_packet_process_pipeline(&self, pipeline: BoxPeerPacketFilter) {
         // newest pipeline will be executed first
-        self.peer_packet_process_pipeline
-            .write()
-            .await
-            .push(permanent_peer_pipeline_entry(pipeline));
+        append_peer_pipeline(
+            &self.peer_packet_process_pipeline,
+            permanent_peer_pipeline_entry(pipeline),
+        );
     }
 
     pub async fn add_nic_packet_process_pipeline(&self, pipeline: BoxNicPacketFilter) {
         // newest pipeline will be executed first
-        self.nic_packet_process_pipeline
-            .write()
-            .await
-            .push(permanent_nic_pipeline_entry(pipeline));
+        append_nic_pipeline(
+            &self.nic_packet_process_pipeline,
+            permanent_nic_pipeline_entry(pipeline),
+        );
     }
 
     pub(crate) async fn add_managed_packet_process_pipeline(
         &self,
         pipeline: BoxPeerPacketFilter,
     ) -> PipelineRegistrationGuard {
-        let (entry, guard) = managed_peer_pipeline_entry(pipeline);
-        let mut pipelines = self.peer_packet_process_pipeline.write().await;
-        pipelines.retain(|pipeline| pipeline.active.load(Ordering::Acquire));
-        pipelines.push(entry);
+        let (entry, guard) =
+            managed_peer_pipeline_entry(pipeline, &self.peer_packet_process_pipeline);
+        append_peer_pipeline(&self.peer_packet_process_pipeline, entry);
         guard
     }
 
@@ -1301,10 +1389,9 @@ impl PeerManagerCore {
         &self,
         pipeline: BoxNicPacketFilter,
     ) -> PipelineRegistrationGuard {
-        let (entry, guard) = managed_nic_pipeline_entry(pipeline);
-        let mut pipelines = self.nic_packet_process_pipeline.write().await;
-        pipelines.retain(|pipeline| pipeline.active.load(Ordering::Acquire));
-        pipelines.push(entry);
+        let (entry, guard) =
+            managed_nic_pipeline_entry(pipeline, &self.nic_packet_process_pipeline);
+        append_nic_pipeline(&self.nic_packet_process_pipeline, entry);
         guard
     }
 
@@ -1313,7 +1400,7 @@ impl PeerManagerCore {
         &self,
         registration: &PipelineRegistrationGuard,
     ) {
-        remove_managed_nic_pipeline_entry(&self.nic_packet_process_pipeline, registration).await;
+        registration.close();
     }
 
     pub async fn add_route<T>(&self, route: Arc<T>)
@@ -1321,7 +1408,7 @@ impl PeerManagerCore {
         T: Route + PeerPacketFilter + Send + Sync + 'static,
     {
         add_route(
-            self.peer_packet_process_pipeline.as_ref(),
+            &self.peer_packet_process_pipeline,
             self.peers.clone(),
             self.foreign_network_client.clone(),
             self.foreign_network_manager.clone(),
@@ -1332,16 +1419,28 @@ impl PeerManagerCore {
     }
 
     pub async fn remove_nic_packet_process_pipeline(&self, id: String) -> Result<(), Error> {
-        let mut pipelines = self.nic_packet_process_pipeline.write().await;
-        if let Some(pos) = pipelines.iter().position(|pipeline| {
-            let filter = pipeline.filter.read().clone();
-            filter.is_some_and(|filter| filter.id() == id)
-        }) {
-            pipelines.remove(pos);
-            Ok(())
-        } else {
-            Err(Error::NotFound)
-        }
+        let snapshot = self.nic_packet_process_pipeline.load_full();
+        let Some(target) = snapshot
+            .iter()
+            .find(|entry| {
+                entry
+                    .filter_if_active()
+                    .is_some_and(|filter| filter.id() == id)
+            })
+            .map(|entry| entry.filter.clone())
+        else {
+            return Err(Error::NotFound);
+        };
+        self.nic_packet_process_pipeline.rcu(|current| {
+            Arc::new(
+                current
+                    .iter()
+                    .filter(|entry| !Arc::ptr_eq(&entry.filter, &target))
+                    .cloned()
+                    .collect(),
+            )
+        });
+        Ok(())
     }
 
     pub async fn send_msg_for_proxy(
@@ -1428,10 +1527,9 @@ impl PeerManagerCore {
 
     pub(crate) async fn clear_resources(&self) {
         self.stop().await;
-        let mut peer_pipeline = self.peer_packet_process_pipeline.write().await;
-        peer_pipeline.clear();
-        let mut nic_pipeline = self.nic_packet_process_pipeline.write().await;
-        nic_pipeline.clear();
+        self.peer_packet_process_pipeline
+            .store(Arc::new(Vec::new()));
+        self.nic_packet_process_pipeline.store(Arc::new(Vec::new()));
 
         self.peer_rpc_mgr.rpc_server().registry().unregister_all();
     }
@@ -1490,7 +1588,7 @@ impl PeerManagerCore {
         }
 
         init_packet_process_pipeline(
-            self.peer_packet_process_pipeline.as_ref(),
+            &self.peer_packet_process_pipeline,
             self.nic_channel.clone(),
             self.peer_rpc_tspt.packet_sender(),
         )
@@ -2061,7 +2159,7 @@ pub(crate) struct PeerOutboundPacketRouter {
     route: ArcRoute,
     foreign_network_client: Arc<ForeignNetworkClient>,
     relay_peer_map: Arc<RelayPeerMap>,
-    nic_packet_process_pipeline: Arc<RwLock<Vec<Arc<NicPipelineEntry>>>>,
+    nic_packet_process_pipeline: NicPacketPipeline,
     encryptor: Arc<dyn Encryptor>,
     data_compress_algo: CompressorAlgo,
     exit_nodes: Arc<RwLock<Vec<IpAddr>>>,
@@ -2081,7 +2179,7 @@ impl PeerOutboundPacketRouter {
         route: ArcRoute,
         foreign_network_client: Arc<ForeignNetworkClient>,
         relay_peer_map: Arc<RelayPeerMap>,
-        nic_packet_process_pipeline: Arc<RwLock<Vec<Arc<NicPipelineEntry>>>>,
+        nic_packet_process_pipeline: NicPacketPipeline,
         encryptor: Arc<dyn Encryptor>,
         data_compress_algo: CompressorAlgo,
         exit_nodes: Arc<RwLock<Vec<IpAddr>>>,
@@ -2128,12 +2226,17 @@ impl PeerOutboundPacketRouter {
         }
     }
 
-    fn mark_recent_traffic(&self, dst_peer_id: PeerId) {
-        let flags = self.context.flags();
-        self.recent_traffic
-            .mark(dst_peer_id, flags.disable_p2p, flags.lazy_p2p, |peer_id| {
-                self.has_directly_connected_conn(peer_id)
-            });
+    fn mark_recent_traffic_with_policy(
+        &self,
+        dst_peer_id: PeerId,
+        packet_policy: PeerPacketPolicy,
+    ) {
+        self.recent_traffic.mark(
+            dst_peer_id,
+            packet_policy.disable_p2p,
+            packet_policy.lazy_p2p,
+            |peer_id| self.has_directly_connected_conn(peer_id),
+        );
     }
 
     async fn run_nic_packet_process_pipeline(&self, data: &mut ZCPacket) -> bool {
@@ -2148,12 +2251,9 @@ impl PeerOutboundPacketRouter {
             return false;
         }
 
-        for pipeline in self.nic_packet_process_pipeline.read().await.iter().rev() {
-            if !pipeline.active.load(Ordering::Acquire) {
-                continue;
-            }
-            let filter = pipeline.filter.read().clone();
-            if let Some(filter) = filter {
+        let pipelines = self.nic_packet_process_pipeline.load_full();
+        for pipeline in pipelines.iter().rev() {
+            if let Some(filter) = pipeline.filter_if_active() {
                 let _ = filter.try_process_packet_from_nic(data).await;
             }
         }
@@ -2161,8 +2261,12 @@ impl PeerOutboundPacketRouter {
         true
     }
 
-    fn check_p2p_only_before_send(&self, dst_peer_id: PeerId) -> Result<(), Error> {
-        if self.context.p2p_only() && !self.peers.has_peer(dst_peer_id) {
+    fn check_p2p_only_before_send(
+        &self,
+        dst_peer_id: PeerId,
+        packet_policy: PeerPacketPolicy,
+    ) -> Result<(), Error> {
+        if packet_policy.p2p_only && !self.peers.has_peer(dst_peer_id) {
             return Err(Error::RouteError(None));
         }
         Ok(())
@@ -2235,8 +2339,9 @@ impl PeerOutboundPacketRouter {
         mut msg: ZCPacket,
         dst_peer_id: PeerId,
     ) -> Result<(), Error> {
-        self.mark_recent_traffic(dst_peer_id);
-        self.check_p2p_only_before_send(dst_peer_id)?;
+        let packet_policy = self.context.packet_policy();
+        self.mark_recent_traffic_with_policy(dst_peer_id, packet_policy);
+        self.check_p2p_only_before_send(dst_peer_id, packet_policy)?;
 
         self.counters
             .compress_tx_bytes_before
@@ -2397,9 +2502,10 @@ impl PeerOutboundPacketRouter {
         if !self.run_nic_packet_process_pipeline(&mut msg).await {
             return Ok(());
         }
+        let packet_policy = self.context.packet_policy();
         let cur_to_peer_id = msg.peer_manager_header().unwrap().to_peer_id.into();
         if cur_to_peer_id != 0 {
-            self.mark_recent_traffic(cur_to_peer_id);
+            self.mark_recent_traffic_with_policy(cur_to_peer_id, packet_policy);
             return send_msg_internal(
                 self.peers.as_ref(),
                 &self.foreign_network_client,
@@ -2437,10 +2543,9 @@ impl PeerOutboundPacketRouter {
             .compress_tx_bytes_after
             .add(msg.buf_len() as u64);
 
-        let is_latency_first = self.context.latency_first();
         msg.mut_peer_manager_header()
             .unwrap()
-            .set_latency_first(is_latency_first)
+            .set_latency_first(packet_policy.latency_first)
             .set_exit_node(is_exit_node);
 
         let mut errs: Vec<Error> = vec![];
@@ -2449,9 +2554,9 @@ impl PeerOutboundPacketRouter {
         let should_mark_recent_traffic = should_mark_recent_traffic_for_fanout(total_dst_peers);
         for (i, peer_id) in dst_peers.iter().enumerate() {
             if should_mark_recent_traffic {
-                self.mark_recent_traffic(*peer_id);
+                self.mark_recent_traffic_with_policy(*peer_id, packet_policy);
             }
-            if let Err(e) = self.check_p2p_only_before_send(*peer_id) {
+            if let Err(e) = self.check_p2p_only_before_send(*peer_id, packet_policy) {
                 errs.push(e);
                 continue;
             }
@@ -2523,7 +2628,7 @@ pub(crate) struct PeerPacketRouter {
     packet_recv: PacketRecvChanReceiver,
     my_peer_id: PeerId,
     peers: Arc<PeerMap>,
-    peer_packet_process_pipeline: Arc<RwLock<Vec<Arc<PeerPipelineEntry>>>>,
+    peer_packet_process_pipeline: PeerPacketPipeline,
     foreign_client: Arc<ForeignNetworkClient>,
     relay_peer_map: Arc<RelayPeerMap>,
     foreign_network_manager: Arc<ForeignNetworkManager>,
@@ -2545,7 +2650,7 @@ impl PeerPacketRouter {
         packet_recv: PacketRecvChanReceiver,
         my_peer_id: PeerId,
         peers: Arc<PeerMap>,
-        peer_packet_process_pipeline: Arc<RwLock<Vec<Arc<PeerPipelineEntry>>>>,
+        peer_packet_process_pipeline: PeerPacketPipeline,
         foreign_client: Arc<ForeignNetworkClient>,
         relay_peer_map: Arc<RelayPeerMap>,
         foreign_network_manager: Arc<ForeignNetworkManager>,
@@ -2786,12 +2891,9 @@ impl PeerPacketRouter {
             let mut processed = false;
             let mut zc_packet = Some(ret);
             tracing::trace!(?zc_packet, "try_process_packet_from_peer");
-            for pipeline in self.peer_packet_process_pipeline.read().await.iter().rev() {
-                if !pipeline.active.load(Ordering::Acquire) {
-                    continue;
-                }
-                let filter = pipeline.filter.read().clone();
-                if let Some(filter) = filter {
+            let pipelines = self.peer_packet_process_pipeline.load_full();
+            for pipeline in pipelines.iter().rev() {
+                if let Some(filter) = pipeline.filter_if_active() {
                     zc_packet = filter
                         .try_process_packet_from_peer(zc_packet.unwrap())
                         .await;
@@ -3099,17 +3201,15 @@ mod tests {
     use crate::{
         config::runtime::CoreRuntimeConfig,
         config::{CoreConfig, IpPrefix, NetworkIdentity, NodeConfig, ProxyNetworkConfig},
-        peers::{
-            context::{PeerContext, PeerEvent},
-            create_packet_recv_chan,
-        },
+        host::packet::{HostPacketSender, host_packet_channel},
+        peers::context::{PeerContext, PeerEvent},
         proto::common::{PeerFeatureFlag, StunInfo},
     };
 
     impl PeerManagerCore {
         pub(crate) fn new_portable_for_test(
             config: PortablePeerManagerConfig,
-            nic_channel: PacketRecvChan,
+            nic_channel: HostPacketSender,
         ) -> anyhow::Result<Self> {
             let runtime_config = CoreRuntimeConfigStore::new(
                 CoreRuntimeConfig::default(),
@@ -3182,40 +3282,35 @@ mod tests {
         }
     }
 
-    #[tokio::test]
-    async fn managed_nic_pipeline_removal_waits_for_readers_and_drops_filter() {
+    #[test]
+    fn managed_nic_pipeline_removal_preserves_in_flight_snapshot() {
         let drops = Arc::new(AtomicUsize::new(0));
+        let pipeline = Arc::new(ArcSwap::from_pointee(Vec::new()));
         let (entry, registration) =
-            managed_nic_pipeline_entry(Box::new(DropCountingNicFilter(drops.clone())));
-        let pipeline = Arc::new(RwLock::new(vec![entry]));
-        let reader = pipeline.read().await;
-        let active_filter = reader[0].filter.read().clone().unwrap();
-        let remove_pipeline = pipeline.clone();
-        let remove_registration = registration.clone();
-        let removal = tokio::spawn(async move {
-            remove_managed_nic_pipeline_entry(&remove_pipeline, &remove_registration).await;
-        });
+            managed_nic_pipeline_entry(Box::new(DropCountingNicFilter(drops.clone())), &pipeline);
+        append_nic_pipeline(&pipeline, entry);
+        let reader = pipeline.load_full();
 
-        tokio::task::yield_now().await;
-        assert!(!removal.is_finished());
+        registration.close();
+
+        assert!(pipeline.load().is_empty());
         assert_eq!(drops.load(Ordering::Relaxed), 0);
 
         drop(reader);
-        drop(active_filter);
-        removal.await.unwrap();
-        assert!(pipeline.read().await.is_empty());
         assert_eq!(drops.load(Ordering::Relaxed), 1);
     }
 
     #[test]
     fn managed_pipeline_guard_releases_filter_without_a_runtime() {
         let drops = Arc::new(AtomicUsize::new(0));
+        let pipeline = Arc::new(ArcSwap::from_pointee(Vec::new()));
         let (entry, registration) =
-            managed_nic_pipeline_entry(Box::new(DropCountingNicFilter(drops.clone())));
+            managed_nic_pipeline_entry(Box::new(DropCountingNicFilter(drops.clone())), &pipeline);
+        append_nic_pipeline(&pipeline, entry);
 
         drop(registration);
 
-        assert!(entry.filter.read().is_none());
+        assert!(pipeline.load().is_empty());
         assert_eq!(drops.load(Ordering::Relaxed), 1);
     }
 
@@ -3258,7 +3353,7 @@ mod tests {
     fn build_portable_config_for_test(
         config: PortablePeerManagerConfig,
     ) -> anyhow::Result<PeerManagerCore> {
-        let (packet_tx, _packet_rx) = create_packet_recv_chan();
+        let (packet_tx, _packet_rx) = host_packet_channel();
         PeerManagerCore::new_portable_for_test(config, packet_tx)
     }
 
@@ -3305,7 +3400,11 @@ mod tests {
         );
     }
 
-    #[cfg(not(feature = "aes-gcm"))]
+    #[cfg(not(any(
+        feature = "aes-gcm",
+        feature = "openssl-crypto",
+        feature = "ring-crypto"
+    )))]
     #[tokio::test]
     async fn portable_peer_manager_rejects_requested_unavailable_aes() {
         let mut config = PortablePeerManagerConfig::new(portable_runtime_config("portable-net"));
@@ -3382,7 +3481,7 @@ mod tests {
         let public_ipv6_runtime =
             CorePublicIpv6Runtime::new(runtime_config.clone(), Arc::new(()), Arc::new(()));
         let events = Arc::new(CountingPeerEventSink::default());
-        let (packet_tx, _packet_rx) = create_packet_recv_chan();
+        let (packet_tx, _packet_rx) = host_packet_channel();
 
         let core = PeerManagerCore::new(
             config,
@@ -3417,7 +3516,7 @@ mod tests {
             }),
         };
         config.snapshot.set_acl_groups(Some(&acl));
-        let (packet_tx, _packet_rx) = create_packet_recv_chan();
+        let (packet_tx, _packet_rx) = host_packet_channel();
 
         let core = PeerManagerCore::new_portable_for_test(config, packet_tx).unwrap();
 
@@ -3563,7 +3662,7 @@ mod tests {
     async fn portable_peer_manager_rejects_inconsistent_network_names() {
         let mut runtime = portable_runtime_config("identity-net");
         runtime.core.node.network_name = "node-net".to_owned();
-        let (packet_tx, _packet_rx) = create_packet_recv_chan();
+        let (packet_tx, _packet_rx) = host_packet_channel();
 
         let result = PeerManagerCore::new_portable_for_test(
             PortablePeerManagerConfig::new(runtime),

easytier-core/src/peers/traffic_metrics.rs

@@ -37,6 +37,13 @@ enum CachedPeerTrafficCounters {
 }
 
 impl CachedPeerTrafficCounters {
+    fn add_sample(&self, bytes: u64) {
+        match self {
+            CachedPeerTrafficCounters::Unknown(counters)
+            | CachedPeerTrafficCounters::Resolved(counters) => counters.add_sample(bytes),
+        }
+    }
+
     fn counters(&self) -> TrafficCounters {
         match self {
             CachedPeerTrafficCounters::Unknown(counters)
@@ -95,7 +102,7 @@ impl LogicalTrafficMetrics {
         if let Some(entry) = self.per_peer.get(&peer_id)
             && entry.value().is_resolved()
         {
-            entry.value().counters().add_sample(bytes);
+            entry.value().add_sample(bytes);
             return;
         }
 

easytier-core/src/socket/tcp.rs

@@ -6,12 +6,28 @@ use tokio::io::{AsyncRead, AsyncWrite};
 
 use crate::socket::{IpVersion, SocketContext, SocketListener};
 
+pub type VirtualTcpReadHalf = Box<dyn AsyncRead + Unpin + Send + 'static>;
+pub type VirtualTcpWriteHalf = Box<dyn AsyncWrite + Unpin + Send + 'static>;
+pub type VirtualTcpSplit = (VirtualTcpReadHalf, VirtualTcpWriteHalf);
+
 /// A core-visible TCP stream endpoint.
 ///
 /// Implementations are runtime adapters over concrete TCP stream types. This
 /// trait deliberately stays below tunnel framing: it only exposes stream I/O and
 /// socket addresses.
 pub trait VirtualTcpSocket: AsyncRead + AsyncWrite + Unpin + Send + 'static {
+    /// Consumes the stream into independently owned read and write halves.
+    ///
+    /// Portable adapters may use the generic shared split. Native adapters
+    /// should override this when their runtime provides lock-free owned halves.
+    fn into_split(self) -> VirtualTcpSplit
+    where
+        Self: Sized,
+    {
+        let (reader, writer) = tokio::io::split(self);
+        (Box::new(reader), Box::new(writer))
+    }
+
     fn local_addr(&self) -> io::Result<SocketAddr>;
 
     fn peer_addr(&self) -> io::Result<SocketAddr>;

easytier-core/src/socket/udp/layer.rs

@@ -18,8 +18,9 @@ use super::{
     UDP_SESSION_CONNECT_TIMEOUT, UDP_SESSION_QUEUE_CAPACITY, UDP_SESSION_RESEND_INTERVAL,
     packet::{
         EasyTierUdpPacketKind, UdpDatagramClassification, UdpSessionPacketKind,
-        classify_udp_datagram, extract_dst_addr_from_v4_hole_punch_packet,
-        extract_v6_hole_punch_packet, new_sack_packet, new_syn_packet,
+        classify_session_udp_datagram, classify_udp_datagram,
+        extract_dst_addr_from_v4_hole_punch_packet, extract_v6_hole_punch_packet, new_sack_packet,
+        new_syn_packet,
     },
     session::{
         ClassifiedUdpSessionAccept, ClassifiedUdpSessionAccepts, ClassifiedUdpSessionKey,
@@ -32,8 +33,9 @@ use super::{
         dispatch_payload_to_session, udp_session_registry_entry,
     },
     virtual_socket::{
-        NoopUdpSessionStunResponder, PreferredIpv6Source, UdpSessionStunResponder,
-        UdpSocketRecvMeta, UdpSocketSendMeta, VirtualUdpSocket, VirtualUdpSocketFactory,
+        MAX_UDP_SESSION_DATAGRAM_SIZE, NoopUdpSessionStunResponder, PreferredIpv6Source,
+        UdpSessionStunResponder, UdpSocketRecvMeta, UdpSocketSendMeta, VirtualUdpSocket,
+        VirtualUdpSocketFactory,
     },
 };
 
@@ -443,11 +445,10 @@ pub(super) async fn udp_session_layer_recv_task<S, R>(
     S: VirtualUdpSocket,
     R: UdpSessionStunResponder<S>,
 {
-    let mut buf = [0u8; 65535];
     let control_permits = Arc::new(Semaphore::new(UDP_SESSION_QUEUE_CAPACITY));
     loop {
-        let (len, remote_addr, recv_meta) = match socket.recv_from_with_meta(&mut buf).await {
-            Ok(ret) => ret,
+        let datagram = match socket.recv_datagram().await {
+            Ok(datagram) => datagram,
             Err(err) => {
                 tracing::debug!(?err, "udp session recv loop stopped");
                 let _ = session_shutdown_tx.send(true);
@@ -457,12 +458,25 @@ pub(super) async fn udp_session_layer_recv_task<S, R>(
                 break;
             }
         };
-
-        let payload = BytesMut::from(&buf[..len]);
-        let datagram = UdpSessionDatagram::new(payload.clone(), recv_meta);
+        let payload = datagram.payload;
+        let remote_addr = datagram.remote_addr;
+        let recv_meta = datagram.meta;
+        if payload.len() > MAX_UDP_SESSION_DATAGRAM_SIZE {
+            tracing::debug!(
+                datagram_len = payload.len(),
+                max_datagram_len = MAX_UDP_SESSION_DATAGRAM_SIZE,
+                ?remote_addr,
+                "dropping oversized udp session datagram"
+            );
+            continue;
+        }
         let quic_key = ClassifiedUdpSessionKey::new(UdpSessionProtocol::Quic, remote_addr);
         if classified_sessions.contains_key(&quic_key) {
-            dispatch_existing_classified_udp_datagram(&classified_sessions, quic_key, datagram);
+            dispatch_existing_classified_udp_datagram(
+                &classified_sessions,
+                quic_key,
+                UdpSessionDatagram::new(payload, recv_meta),
+            );
             continue;
         }
         match classify_udp_datagram(payload) {
@@ -500,23 +514,24 @@ pub(super) async fn udp_session_layer_recv_task<S, R>(
                 kind,
                 conn_id,
                 packet,
-                fallback,
             } => {
-                let consumed = dispatch_easy_tier_udp_datagram(
-                    socket.clone(),
+                let unconsumed = dispatch_easy_tier_udp_datagram(
+                    &socket,
                     &sessions,
                     &pending_connects,
                     &mux_accepted,
                     &control,
-                    control_permits.clone(),
+                    &control_permits,
                     remote_addr,
                     kind,
                     conn_id,
-                    &packet,
+                    packet,
                     recv_meta,
-                    session_shutdown_tx.subscribe(),
+                    &session_shutdown_tx,
                 );
-                if !consumed {
+                if let Some(packet) = unconsumed {
+                    let datagram = BytesMut::from(packet.into_bytes());
+                    let fallback = classify_session_udp_datagram(&datagram);
                     dispatch_session_udp_datagram(
                         socket.clone(),
                         &classified_sessions,
@@ -524,7 +539,7 @@ pub(super) async fn udp_session_layer_recv_task<S, R>(
                         session_shutdown_tx.subscribe(),
                         remote_addr,
                         fallback,
-                        UdpSessionDatagram::new(packet.into_bytes().into(), recv_meta),
+                        UdpSessionDatagram::new(datagram, recv_meta),
                     );
                 }
             }
@@ -537,14 +552,14 @@ fn dispatch_existing_classified_udp_datagram(
     key: ClassifiedUdpSessionKey,
     datagram: UdpSessionDatagram,
 ) {
-    let Some(entry) = classified_sessions
-        .get(&key)
-        .map(|entry| entry.value().clone())
-    else {
+    let Some(entry) = classified_sessions.get(&key) else {
         return;
     };
 
-    if !dispatch_payload_to_session(&entry.incoming, datagram, UdpSessionEnqueuePolicy::Reliable) {
+    let dispatched =
+        dispatch_payload_to_session(&entry.incoming, datagram, UdpSessionEnqueuePolicy::Reliable);
+    drop(entry);
+    if !dispatched {
         close_classified_udp_session(classified_sessions, key);
         tracing::debug!(?key, "classified udp session data queue closed");
     }
@@ -552,73 +567,84 @@ fn dispatch_existing_classified_udp_datagram(
 
 #[allow(clippy::too_many_arguments)]
 fn dispatch_easy_tier_udp_datagram<S>(
-    socket: Arc<S>,
+    socket: &Arc<S>,
     sessions: &Arc<UdpSessionRegistry>,
     pending_connects: &Arc<PendingUdpSessionConnects>,
     mux_accepted: &mpsc::Sender<UdpSession>,
     control: &mpsc::Sender<UdpSessionLayerControl>,
-    control_permits: Arc<Semaphore>,
+    control_permits: &Arc<Semaphore>,
     remote_addr: SocketAddr,
     kind: EasyTierUdpPacketKind,
     conn_id: u32,
-    packet: &ZCPacket,
+    packet: ZCPacket,
     recv_meta: UdpSocketRecvMeta,
-    session_shutdown: watch::Receiver<bool>,
-) -> bool
+    session_shutdown: &watch::Sender<bool>,
+) -> Option<ZCPacket>
 where
     S: VirtualUdpSocket,
 {
-    match kind {
+    let consumed = match kind {
         EasyTierUdpPacketKind::Data => {
-            dispatch_data_packet(sessions, remote_addr, conn_id, packet, recv_meta)
+            return dispatch_data_packet(sessions, remote_addr, conn_id, packet, recv_meta).err();
         }
         EasyTierUdpPacketKind::Syn => handle_new_easy_tier_mux_connect(
-            socket,
+            socket.clone(),
             sessions.clone(),
             mux_accepted.clone(),
             remote_addr,
             conn_id,
-            packet,
-            session_shutdown,
+            &packet,
+            session_shutdown.subscribe(),
         ),
         EasyTierUdpPacketKind::Sack => {
-            dispatch_sack_packet(sessions, pending_connects, remote_addr, conn_id, packet)
+            dispatch_sack_packet(sessions, pending_connects, remote_addr, conn_id, &packet)
         }
         EasyTierUdpPacketKind::HolePunch => {
             dispatch_hole_punch_packet(pending_connects, remote_addr)
         }
-        EasyTierUdpPacketKind::V4HolePunch => {
-            dispatch_v4_hole_punch_control(socket, control_permits, control, remote_addr, packet)
-        }
-        EasyTierUdpPacketKind::V6HolePunch => {
-            dispatch_v6_hole_punch_control(socket, control_permits, control, remote_addr, packet)
-        }
-    }
+        EasyTierUdpPacketKind::V4HolePunch => dispatch_v4_hole_punch_control(
+            socket.clone(),
+            control_permits.clone(),
+            control,
+            remote_addr,
+            &packet,
+        ),
+        EasyTierUdpPacketKind::V6HolePunch => dispatch_v6_hole_punch_control(
+            socket.clone(),
+            control_permits.clone(),
+            control,
+            remote_addr,
+            &packet,
+        ),
+    };
+    if consumed { None } else { Some(packet) }
 }
 
 pub(super) fn dispatch_data_packet(
     sessions: &UdpSessionRegistry,
     peer_addr: SocketAddr,
     conn_id: u32,
-    packet: &ZCPacket,
+    packet: ZCPacket,
     recv_meta: UdpSocketRecvMeta,
-) -> bool {
+) -> Result<(), ZCPacket> {
     let key = UdpSessionKey::new(peer_addr, conn_id);
-    let Some(entry) = sessions.get(&key).map(|entry| entry.value().clone()) else {
-        return false;
+    let Some(entry) = sessions.get(&key) else {
+        return Err(packet);
     };
 
-    let payload = UdpSessionDatagram::new(BytesMut::from(packet.udp_payload()), recv_meta);
     let policy = if packet.is_lossy() {
         UdpSessionEnqueuePolicy::Lossy
     } else {
         UdpSessionEnqueuePolicy::Reliable
     };
-    if !dispatch_payload_to_session(&entry.incoming, payload, policy) {
+    let payload = UdpSessionDatagram::from_easytier_packet(packet, recv_meta);
+    let dispatched = dispatch_payload_to_session(&entry.incoming, payload, policy);
+    drop(entry);
+    if !dispatched {
         close_udp_session(sessions, key);
         tracing::debug!(?key, "udp session data queue closed");
     }
-    true
+    Ok(())
 }
 
 fn dispatch_session_udp_datagram<S>(
@@ -660,15 +686,14 @@ fn dispatch_classified_udp_datagram<S>(
     S: VirtualUdpSocket,
 {
     let key = ClassifiedUdpSessionKey::new(protocol, remote_addr);
-    if let Some(entry) = classified_sessions
-        .get(&key)
-        .map(|entry| entry.value().clone())
-    {
-        if !dispatch_payload_to_session(
+    if let Some(entry) = classified_sessions.get(&key) {
+        let dispatched = dispatch_payload_to_session(
             &entry.incoming,
             datagram,
             UdpSessionEnqueuePolicy::Reliable,
-        ) {
+        );
+        drop(entry);
+        if !dispatched {
             close_classified_udp_session(classified_sessions, key);
             tracing::debug!(?key, "classified udp session data queue closed");
         }
@@ -712,11 +737,12 @@ fn dispatch_classified_udp_datagram<S>(
         }
         dashmap::mapref::entry::Entry::Occupied(entry) => {
             let entry = entry.get().clone();
-            if !dispatch_payload_to_session(
+            let dispatched = dispatch_payload_to_session(
                 &entry.incoming,
                 datagram,
                 UdpSessionEnqueuePolicy::Reliable,
-            ) {
+            );
+            if !dispatched {
                 close_classified_udp_session(classified_sessions, key);
                 tracing::debug!(?key, "classified udp session data queue closed");
             }

easytier-core/src/socket/udp/mod.rs

@@ -29,7 +29,8 @@ pub(crate) use session::{
     UdpSessionTunnelParts,
 };
 pub use virtual_socket::{
-    NoopUdpSessionStunResponder, PreferredIpv6Source, UdpBindOptions, UdpSessionStunResponder,
+    MAX_UDP_DATAGRAM_SIZE, MAX_UDP_SESSION_DATAGRAM_SIZE, NoopUdpSessionStunResponder,
+    PreferredIpv6Source, UdpBindOptions, UdpSessionStunResponder, UdpSocketDatagram,
     UdpSocketPurpose, UdpSocketRecvMeta, UdpSocketSendMeta, VirtualUdpSocket,
     VirtualUdpSocketFactory, send_v4_hole_punch_control_packet, send_v6_hole_punch_control_packet,
 };

easytier-core/src/socket/udp/packet.rs

@@ -161,7 +161,6 @@ pub(super) enum UdpDatagramClassification {
         kind: EasyTierUdpPacketKind,
         conn_id: u32,
         packet: ZCPacket,
-        fallback: UdpSessionPacketKind,
     },
     SessionPacket {
         kind: UdpSessionPacketKind,
@@ -216,7 +215,7 @@ pub(super) enum EasyTierUdpDatagramInspectError {
     },
 }
 
-fn classify_session_udp_datagram(data: &[u8]) -> UdpSessionPacketKind {
+pub(super) fn classify_session_udp_datagram(data: &[u8]) -> UdpSessionPacketKind {
     if is_wireguard_packet(data) {
         UdpSessionPacketKind::Classified(UdpSessionProtocol::WireGuard)
     } else if is_quic_packet(data) {
@@ -335,12 +334,11 @@ pub(super) fn classify_udp_datagram(datagram: BytesMut) -> UdpDatagramClassifica
         return UdpDatagramClassification::Stun(datagram);
     }
 
-    let fallback = classify_session_udp_datagram(&datagram);
     let easytier = match inspect_easytier_udp_datagram(&datagram) {
         Ok(Some(easytier)) => easytier,
         Ok(None) => {
             return UdpDatagramClassification::SessionPacket {
-                kind: fallback,
+                kind: classify_session_udp_datagram(&datagram),
                 datagram,
             };
         }
@@ -361,7 +359,7 @@ pub(super) fn classify_udp_datagram(datagram: BytesMut) -> UdpDatagramClassifica
                 }
             }
             return UdpDatagramClassification::SessionPacket {
-                kind: fallback,
+                kind: classify_session_udp_datagram(&datagram),
                 datagram,
             };
         }
@@ -372,7 +370,6 @@ pub(super) fn classify_udp_datagram(datagram: BytesMut) -> UdpDatagramClassifica
         kind: easytier.kind,
         conn_id: easytier.conn_id,
         packet,
-        fallback,
     }
 }
 

easytier-core/src/socket/udp/session.rs

@@ -14,33 +14,63 @@ use tokio::{
     task::JoinHandle,
 };
 
-use crate::socket::ring::{RingSocket, RingSocketReceiver, RingSocketSendError, RingSocketSender};
+use crate::{
+    packet::{UdpPacketType, ZCPacket, ZCPacketType},
+    socket::ring::{RingSocket, RingSocketReceiver, RingSocketSendError, RingSocketSender},
+};
 
 use super::{
-    UDP_SESSION_QUEUE_CAPACITY,
+    MAX_UDP_SESSION_DATAGRAM_SIZE, UDP_SESSION_QUEUE_CAPACITY,
     packet::{new_data_packet, udp_session_payload_len},
     virtual_socket::{PreferredIpv6Source, UdpBindOptions, UdpSocketRecvMeta, VirtualUdpSocket},
 };
 
-#[derive(Debug, Clone, PartialEq, Eq)]
+#[derive(Debug)]
 pub(crate) struct UdpSessionDatagram {
-    pub(crate) payload: BytesMut,
+    payload: UdpSessionDatagramPayload,
     pub(crate) dst_ip: Option<IpAddr>,
 }
 
+#[derive(Debug)]
+enum UdpSessionDatagramPayload {
+    Bytes(BytesMut),
+    EasyTierPacket(ZCPacket),
+}
+
 impl UdpSessionDatagram {
     pub(crate) fn new(payload: BytesMut, meta: UdpSocketRecvMeta) -> Self {
         Self {
-            payload,
+            payload: UdpSessionDatagramPayload::Bytes(payload),
             dst_ip: meta.dst_ip,
         }
     }
+
+    pub(crate) fn from_easytier_packet(packet: ZCPacket, meta: UdpSocketRecvMeta) -> Self {
+        Self {
+            payload: UdpSessionDatagramPayload::EasyTierPacket(packet),
+            dst_ip: meta.dst_ip,
+        }
+    }
+
+    pub(crate) fn payload(&self) -> &[u8] {
+        match &self.payload {
+            UdpSessionDatagramPayload::Bytes(payload) => payload,
+            UdpSessionDatagramPayload::EasyTierPacket(packet) => packet.udp_payload(),
+        }
+    }
+
+    pub(crate) fn into_tunnel_packet(self) -> Result<ZCPacket, BytesMut> {
+        match self.payload {
+            UdpSessionDatagramPayload::Bytes(payload) => Err(payload),
+            UdpSessionDatagramPayload::EasyTierPacket(packet) => Ok(packet),
+        }
+    }
 }
 
 impl From<BytesMut> for UdpSessionDatagram {
     fn from(payload: BytesMut) -> Self {
         Self {
-            payload,
+            payload: UdpSessionDatagramPayload::Bytes(payload),
             dst_ip: None,
         }
     }
@@ -245,9 +275,12 @@ pub struct UdpSession {
     pub(super) _cleanup: UdpSessionCleanup,
 }
 
-pub(crate) struct UdpSessionOutbound {
-    pub(crate) payload: BytesMut,
-    pub(crate) completion: oneshot::Sender<io::Result<usize>>,
+pub(crate) enum UdpSessionOutbound {
+    Datagram {
+        payload: BytesMut,
+        completion: oneshot::Sender<io::Result<usize>>,
+    },
+    TunnelPacket(ZCPacket),
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
@@ -258,13 +291,38 @@ pub(crate) enum UdpSessionCodec {
 
 impl UdpSessionCodec {
     pub(crate) fn validate_payload(&self, payload: &[u8]) -> io::Result<()> {
+        self.validate_datagram_size(payload.len())?;
         if matches!(self, Self::EasyTierData { .. }) {
             udp_session_payload_len(payload)?;
         }
         Ok(())
     }
 
+    fn validate_datagram_size(&self, payload_len: usize) -> io::Result<()> {
+        let header_len = match self {
+            Self::EasyTierData { .. } => crate::packet::UDP_TUNNEL_HEADER_SIZE,
+            Self::Identity => 0,
+        };
+        let datagram_len = payload_len.checked_add(header_len).ok_or_else(|| {
+            io::Error::new(
+                io::ErrorKind::InvalidInput,
+                "udp session datagram size overflow",
+            )
+        })?;
+        if datagram_len > MAX_UDP_SESSION_DATAGRAM_SIZE {
+            return Err(io::Error::new(
+                io::ErrorKind::InvalidInput,
+                format!(
+                    "udp session datagram too large: {datagram_len}, max: \
+                     {MAX_UDP_SESSION_DATAGRAM_SIZE}"
+                ),
+            ));
+        }
+        Ok(())
+    }
+
     fn encode(&self, payload: &[u8]) -> io::Result<BytesMut> {
+        self.validate_payload(payload)?;
         match self {
             Self::EasyTierData { conn_id } => {
                 Ok(new_data_packet(*conn_id, payload)?.into_bytes().into())
@@ -272,6 +330,23 @@ impl UdpSessionCodec {
             Self::Identity => Ok(BytesMut::from(payload)),
         }
     }
+
+    fn encode_tunnel_packet(&self, packet: ZCPacket) -> io::Result<bytes::Bytes> {
+        let mut packet = packet.convert_type(ZCPacketType::UDP);
+        self.validate_datagram_size(packet.udp_payload().len())?;
+        match self {
+            Self::EasyTierData { conn_id } => {
+                let payload_len = udp_session_payload_len(packet.udp_payload())?;
+                let header = packet.mut_udp_tunnel_header().unwrap();
+                header.conn_id.set(*conn_id);
+                header.msg_type = UdpPacketType::Data as u8;
+                header.padding = 0;
+                header.len.set(payload_len);
+                Ok(packet.into_bytes())
+            }
+            Self::Identity => Ok(packet.udp_payload_bytes().freeze()),
+        }
+    }
 }
 
 #[derive(Clone)]
@@ -387,8 +462,12 @@ impl UdpSession {
             peer_addr,
             codec,
             rings.session_send_rx,
+            close.clone(),
+        ));
+        let shutdown_task = tokio::spawn(close_udp_session_on_shutdown(
             shutdown,
             close.clone(),
+            send_task.abort_handle(),
         ));
 
         Self {
@@ -402,7 +481,7 @@ impl UdpSession {
             _cleanup: UdpSessionCleanup {
                 session_close: Some(close),
                 shutdown: None,
-                tasks: vec![send_task],
+                tasks: vec![send_task, shutdown_task],
                 layer_guard: None,
             },
         }
@@ -471,7 +550,7 @@ impl UdpSessionSocket for UdpSession {
             return Err(udp_session_closed_error());
         }
         let (completion, sent) = oneshot::channel();
-        let outbound = UdpSessionOutbound {
+        let outbound = UdpSessionOutbound::Datagram {
             payload: BytesMut::from(data),
             completion,
         };
@@ -508,8 +587,8 @@ impl UdpSessionSocket for UdpSession {
                 .ok_or_else(udp_session_closed_error)?
                 .map_err(ring_socket_error_to_io)?,
         };
-        let len = payload.payload.len().min(buf.len());
-        buf[..len].copy_from_slice(&payload.payload[..len]);
+        let len = payload.payload().len().min(buf.len());
+        buf[..len].copy_from_slice(&payload.payload()[..len]);
         Ok((
             len,
             UdpSessionRecvMeta {
@@ -618,56 +697,86 @@ async fn forward_udp_session_to_socket<S>(
     peer_addr: SocketAddr,
     codec: UdpSessionCodec,
     mut outgoing: RingSocketReceiver<UdpSessionOutbound>,
-    mut shutdown: watch::Receiver<bool>,
     close: UdpSessionClose,
 ) where
     S: VirtualUdpSocket,
 {
     loop {
-        tokio::select! {
-            biased;
-            _ = shutdown.changed() => {
+        let Some(outbound) = outgoing.next().await else {
+            break;
+        };
+        let outbound = match outbound {
+            Ok(outbound) => outbound,
+            Err(err) => {
+                tracing::debug!(?err, ?peer_addr, "udp session outgoing ring closed");
                 close.close();
                 break;
             }
-            outbound = outgoing.next() => {
-                let Some(outbound) = outbound else {
-                    break;
-                };
-                let outbound = match outbound {
-                    Ok(outbound) => outbound,
+        };
+        let (datagram, completion) = match outbound {
+            UdpSessionOutbound::Datagram {
+                payload,
+                completion,
+            } => {
+                let payload_len = payload.len();
+                let datagram = match codec.encode(&payload) {
+                    Ok(datagram) => datagram.freeze(),
                     Err(err) => {
-                        tracing::debug!(?err, ?peer_addr, "udp session outgoing ring closed");
+                        tracing::debug!(
+                            ?err,
+                            ?peer_addr,
+                            ?codec,
+                            "udp session datagram encode error"
+                        );
+                        let _ = completion.send(Err(err));
                         close.close();
                         break;
                     }
                 };
-                let payload_len = outbound.payload.len();
-                let datagram = match codec.encode(&outbound.payload) {
+                (datagram, Some((completion, payload_len)))
+            }
+            UdpSessionOutbound::TunnelPacket(packet) => {
+                let datagram = match codec.encode_tunnel_packet(packet) {
                     Ok(datagram) => datagram,
                     Err(err) => {
-                        tracing::debug!(?err, ?peer_addr, ?codec, "udp session datagram encode error");
-                        let _ = outbound.completion.send(Err(err));
+                        tracing::debug!(?err, ?peer_addr, ?codec, "udp tunnel packet encode error");
                         close.close();
                         break;
                     }
                 };
-                match socket.send_to(&datagram, peer_addr).await {
-                    Ok(_) => {
-                        let _ = outbound.completion.send(Ok(payload_len));
-                    }
-                    Err(err) => {
-                        tracing::debug!(?err, ?peer_addr, "udp session send error");
-                        let _ = outbound.completion.send(Err(err));
-                        close.close();
-                        break;
-                    }
+                (datagram, None)
+            }
+        };
+        match socket.send_to(&datagram, peer_addr).await {
+            Ok(_) => {
+                if let Some((completion, payload_len)) = completion {
+                    let _ = completion.send(Ok(payload_len));
                 }
             }
+            Err(err) => {
+                tracing::debug!(?err, ?peer_addr, "udp session send error");
+                if let Some((completion, _)) = completion {
+                    let _ = completion.send(Err(err));
+                }
+                close.close();
+                break;
+            }
         }
     }
 }
 
+async fn close_udp_session_on_shutdown(
+    mut shutdown: watch::Receiver<bool>,
+    close: UdpSessionClose,
+    send_task: tokio::task::AbortHandle,
+) {
+    if !*shutdown.borrow() {
+        let _ = shutdown.changed().await;
+    }
+    close.close();
+    send_task.abort();
+}
+
 pub(super) fn dispatch_payload_to_session(
     incoming: &Arc<StdMutex<RingSocketSender<UdpSessionDatagram>>>,
     payload: impl Into<UdpSessionDatagram>,

easytier-core/src/socket/udp/tests.rs

@@ -945,6 +945,30 @@ async fn udp_layer_routes_quic_like_easytier_packet_to_existing_quic_session() {
     assert_eq!(layer.active_classified_session_count(), 1);
 }
 
+#[tokio::test]
+async fn udp_layer_drops_oversized_datagram_without_stopping_portable_socket() {
+    let local_addr = SocketAddr::from(([127, 0, 0, 1], 12000));
+    let peer_addr = SocketAddr::from(([127, 0, 0, 1], 12001));
+    let socket = Arc::new(AutoSackVirtualUdpSocket::new(local_addr));
+    let layer = UdpSessionLayer::new(socket.clone());
+    let session = layer
+        .open_classified_session(UdpSessionProtocol::Quic, peer_addr)
+        .unwrap();
+    socket.incoming.lock().unwrap().extend([
+        (vec![0xAA; MAX_UDP_SESSION_DATAGRAM_SIZE + 1], peer_addr),
+        (b"after-oversized".to_vec(), peer_addr),
+    ]);
+    socket.incoming_notify.notify_one();
+
+    let mut buf = [0; 32];
+    let len = tokio::time::timeout(Duration::from_secs(1), session.recv(&mut buf))
+        .await
+        .unwrap()
+        .unwrap();
+
+    assert_eq!(&buf[..len], b"after-oversized");
+}
+
 #[tokio::test]
 async fn udp_layer_keeps_easy_tier_syn_out_of_wireguard_session() {
     let local_addr = SocketAddr::from(([127, 0, 0, 1], 12000));
@@ -1021,13 +1045,37 @@ async fn easy_tier_mux_udp_session_rejects_oversized_payload_before_enqueue() {
         sessions,
     );
 
-    let payload = vec![0; u16::MAX as usize + 1];
+    let payload = vec![0; MAX_UDP_SESSION_DATAGRAM_SIZE - UDP_TUNNEL_HEADER_SIZE + 1];
     let err = session.send(&payload).await.unwrap_err();
 
     assert_eq!(err.kind(), io::ErrorKind::InvalidInput);
     assert!(socket.sent().is_empty());
 }
 
+#[test]
+fn udp_session_codecs_enforce_datagram_boundary() {
+    let identity = UdpSessionCodec::Identity;
+    assert!(
+        identity
+            .validate_payload(&vec![0; MAX_UDP_SESSION_DATAGRAM_SIZE])
+            .is_ok()
+    );
+    let err = identity
+        .validate_payload(&vec![0; MAX_UDP_SESSION_DATAGRAM_SIZE + 1])
+        .unwrap_err();
+    assert_eq!(err.kind(), io::ErrorKind::InvalidInput);
+
+    let easy_tier = UdpSessionCodec::EasyTierData {
+        conn_id: 0x1122_3344,
+    };
+    let max_payload = MAX_UDP_SESSION_DATAGRAM_SIZE - UDP_TUNNEL_HEADER_SIZE;
+    assert!(easy_tier.validate_payload(&vec![0; max_payload]).is_ok());
+    let err = easy_tier
+        .validate_payload(&vec![0; max_payload + 1])
+        .unwrap_err();
+    assert_eq!(err.kind(), io::ErrorKind::InvalidInput);
+}
+
 #[tokio::test]
 async fn easy_tier_mux_udp_session_send_failure_closes_session() {
     let local_addr = SocketAddr::from(([127, 0, 0, 1], 12000));
@@ -1074,26 +1122,35 @@ async fn easy_tier_mux_udp_session_receives_only_peer_data_payloads() {
         sessions.clone(),
     );
 
-    dispatch_data_packet(
-        &sessions,
-        unexpected_addr,
-        conn_id,
-        &new_data_packet(conn_id, b"wrong-peer").unwrap(),
-        Default::default(),
+    assert!(
+        dispatch_data_packet(
+            &sessions,
+            unexpected_addr,
+            conn_id,
+            new_data_packet(conn_id, b"wrong-peer").unwrap(),
+            Default::default(),
+        )
+        .is_err()
     );
-    dispatch_data_packet(
-        &sessions,
-        peer_addr,
-        conn_id + 1,
-        &new_data_packet(conn_id + 1, b"wrong-conn").unwrap(),
-        Default::default(),
+    assert!(
+        dispatch_data_packet(
+            &sessions,
+            peer_addr,
+            conn_id + 1,
+            new_data_packet(conn_id + 1, b"wrong-conn").unwrap(),
+            Default::default(),
+        )
+        .is_err()
     );
-    dispatch_data_packet(
-        &sessions,
-        peer_addr,
-        conn_id,
-        &new_data_packet(conn_id, b"payload").unwrap(),
-        Default::default(),
+    assert!(
+        dispatch_data_packet(
+            &sessions,
+            peer_addr,
+            conn_id,
+            new_data_packet(conn_id, b"payload").unwrap(),
+            Default::default(),
+        )
+        .is_ok()
     );
 
     let mut buf = [0; 16];
@@ -1163,6 +1220,26 @@ async fn dropping_udp_session_layer_closes_session_recv() {
     assert_eq!(err.kind(), io::ErrorKind::UnexpectedEof);
 }
 
+#[tokio::test]
+async fn idle_udp_session_closes_when_shutdown_is_signaled() {
+    let local_addr = SocketAddr::from(([127, 0, 0, 1], 12000));
+    let peer_addr = SocketAddr::from(([127, 0, 0, 1], 12001));
+    let key = UdpSessionKey::new(peer_addr, 0x1122_3344);
+    let socket = Arc::new(MockVirtualUdpSocket::new(local_addr, Vec::new()));
+    let sessions = Arc::new(DashMap::new());
+    let (session, shutdown_tx) = create_test_easy_tier_mux_session(socket, key, sessions.clone());
+
+    shutdown_tx.send(true).unwrap();
+
+    let mut buf = [0; 16];
+    let err = tokio::time::timeout(Duration::from_secs(1), session.recv(&mut buf))
+        .await
+        .unwrap()
+        .unwrap_err();
+    assert_eq!(err.kind(), io::ErrorKind::UnexpectedEof);
+    assert!(!sessions.contains_key(&key));
+}
+
 #[tokio::test]
 async fn udp_session_recv_loop_error_closes_registered_sessions() {
     let local_addr = SocketAddr::from(([127, 0, 0, 1], 12000));
@@ -1591,12 +1668,15 @@ async fn sack_from_actual_remote_rekeys_pending_session_before_data_dispatch() {
         },
     );
 
-    dispatch_data_packet(
-        &sessions,
-        expected_addr,
-        conn_id,
-        &new_data_packet(conn_id, b"pre-sack").unwrap(),
-        Default::default(),
+    assert!(
+        dispatch_data_packet(
+            &sessions,
+            expected_addr,
+            conn_id,
+            new_data_packet(conn_id, b"pre-sack").unwrap(),
+            Default::default(),
+        )
+        .is_err()
     );
     dispatch_sack_packet(
         &sessions,
@@ -1605,12 +1685,15 @@ async fn sack_from_actual_remote_rekeys_pending_session_before_data_dispatch() {
         conn_id,
         &new_sack_packet(conn_id, magic),
     );
-    dispatch_data_packet(
-        &sessions,
-        actual_addr,
-        conn_id,
-        &new_data_packet(conn_id, b"payload").unwrap(),
-        Default::default(),
+    assert!(
+        dispatch_data_packet(
+            &sessions,
+            actual_addr,
+            conn_id,
+            new_data_packet(conn_id, b"payload").unwrap(),
+            Default::default(),
+        )
+        .is_ok()
     );
 
     assert!(sessions.contains_key(&actual_key));
@@ -1624,7 +1707,7 @@ async fn sack_from_actual_remote_rekeys_pending_session_before_data_dispatch() {
         .await
         .unwrap()
         .unwrap();
-    assert_eq!(payload.payload, BytesMut::from(&b"payload"[..]));
+    assert_eq!(payload.payload(), b"payload");
 }
 
 #[tokio::test]

easytier-core/src/socket/udp/virtual_socket.rs

@@ -5,6 +5,7 @@ use std::{
 };
 
 use async_trait::async_trait;
+use bytes::BytesMut;
 use serde::{Deserialize, Serialize};
 
 use crate::socket::{IpVersion, SocketContext};
@@ -22,6 +23,23 @@ pub struct UdpSocketSendMeta {
     pub src_ifindex: Option<u32>,
 }
 
+/// Largest UDP datagram that portable socket implementations must receive.
+pub const MAX_UDP_DATAGRAM_SIZE: usize = u16::MAX as usize;
+
+/// Largest datagram accepted by the UDP session/multiplexer data plane.
+///
+/// EasyTier, WireGuard, and QUIC datagrams are bounded by their transport MTU.
+/// Keeping this capacity explicit avoids allocating the theoretical UDP maximum
+/// for every packet on native hosts that can detect truncation.
+pub const MAX_UDP_SESSION_DATAGRAM_SIZE: usize = 8 * 1024;
+
+#[derive(Debug)]
+pub struct UdpSocketDatagram {
+    pub payload: BytesMut,
+    pub remote_addr: SocketAddr,
+    pub meta: UdpSocketRecvMeta,
+}
+
 #[async_trait]
 pub trait VirtualUdpSocket: Send + Sync + 'static {
     fn local_addr(&self) -> std::io::Result<SocketAddr>;
@@ -51,6 +69,23 @@ pub trait VirtualUdpSocket: Send + Sync + 'static {
         let (len, addr) = self.recv_from(buf).await?;
         Ok((len, addr, UdpSocketRecvMeta::default()))
     }
+
+    /// Receives one datagram into an owned buffer.
+    ///
+    /// Portable hosts can use this default implementation. Native hosts should
+    /// override it when their socket API can write directly into owned storage,
+    /// avoiding a second allocation and copy at the Host boundary.
+    async fn recv_datagram(&self) -> std::io::Result<UdpSocketDatagram> {
+        let mut payload = BytesMut::new();
+        payload.resize(MAX_UDP_DATAGRAM_SIZE, 0);
+        let (len, remote_addr, meta) = self.recv_from_with_meta(&mut payload).await?;
+        payload.truncate(len);
+        Ok(UdpSocketDatagram {
+            payload,
+            remote_addr,
+            meta,
+        })
+    }
 }
 
 #[async_trait]

easytier-core/src/tunnel/encrypt/mod.rs

@@ -2,9 +2,21 @@ use crate::{config::EncryptionAlgorithm, packet::ZCPacket};
 use std::{collections::hash_map::DefaultHasher, hash::Hasher, sync::Arc};
 
 #[cfg(feature = "aes-gcm")]
+#[cfg_attr(
+    any(feature = "openssl-crypto", feature = "ring-crypto"),
+    allow(dead_code)
+)]
 pub mod aes_gcm;
 #[cfg(feature = "chacha20")]
+#[cfg_attr(
+    any(feature = "openssl-crypto", feature = "ring-crypto"),
+    allow(dead_code)
+)]
 pub mod chacha20;
+#[cfg(feature = "openssl-crypto")]
+mod openssl;
+#[cfg(all(feature = "ring-crypto", any(not(feature = "openssl-crypto"), test)))]
+mod ring;
 
 pub mod xor;
 
@@ -119,47 +131,123 @@ fn unavailable_encryptor(algorithm: &str) -> Arc<dyn Encryptor> {
     })
 }
 
-fn algorithm_is_available(algorithm: EncryptionAlgorithm) -> bool {
+pub(crate) fn algorithm_is_available(algorithm: EncryptionAlgorithm) -> bool {
     match algorithm {
         EncryptionAlgorithm::Xor => true,
-        EncryptionAlgorithm::AesGcm | EncryptionAlgorithm::Aes256Gcm => cfg!(feature = "aes-gcm"),
-        EncryptionAlgorithm::ChaCha20 => cfg!(feature = "chacha20"),
+        EncryptionAlgorithm::AesGcm | EncryptionAlgorithm::Aes256Gcm => cfg!(any(
+            feature = "aes-gcm",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        )),
+        EncryptionAlgorithm::ChaCha20 => cfg!(any(
+            feature = "chacha20",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        )),
     }
 }
 
+fn is_aead_algorithm(algorithm: EncryptionAlgorithm) -> bool {
+    matches!(
+        algorithm,
+        EncryptionAlgorithm::AesGcm
+            | EncryptionAlgorithm::Aes256Gcm
+            | EncryptionAlgorithm::ChaCha20
+    )
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum AeadBackend {
+    #[cfg(feature = "openssl-crypto")]
+    OpenSsl,
+    #[cfg(all(not(feature = "openssl-crypto"), feature = "ring-crypto"))]
+    Ring,
+    #[cfg(all(
+        not(feature = "openssl-crypto"),
+        not(feature = "ring-crypto"),
+        any(feature = "aes-gcm", feature = "chacha20")
+    ))]
+    RustCrypto,
+}
+
+#[cfg(feature = "openssl-crypto")]
+fn preferred_aead_backend(algorithm: EncryptionAlgorithm) -> Option<AeadBackend> {
+    is_aead_algorithm(algorithm).then_some(AeadBackend::OpenSsl)
+}
+
+#[cfg(all(not(feature = "openssl-crypto"), feature = "ring-crypto"))]
+fn preferred_aead_backend(algorithm: EncryptionAlgorithm) -> Option<AeadBackend> {
+    is_aead_algorithm(algorithm).then_some(AeadBackend::Ring)
+}
+
+#[cfg(all(
+    not(feature = "openssl-crypto"),
+    not(feature = "ring-crypto"),
+    any(feature = "aes-gcm", feature = "chacha20")
+))]
+fn preferred_aead_backend(algorithm: EncryptionAlgorithm) -> Option<AeadBackend> {
+    (is_aead_algorithm(algorithm) && algorithm_is_available(algorithm))
+        .then_some(AeadBackend::RustCrypto)
+}
+
+#[cfg(not(any(
+    feature = "openssl-crypto",
+    feature = "ring-crypto",
+    feature = "aes-gcm",
+    feature = "chacha20"
+)))]
+fn preferred_aead_backend(_algorithm: EncryptionAlgorithm) -> Option<AeadBackend> {
+    None
+}
+
+#[allow(unreachable_patterns)]
 fn create_aes_128(key: [u8; 16]) -> Arc<dyn Encryptor> {
-    #[cfg(feature = "aes-gcm")]
-    {
-        Arc::new(aes_gcm::AesGcmCipher::new_128(key))
-    }
-    #[cfg(not(feature = "aes-gcm"))]
-    {
-        let _ = key;
-        unavailable_encryptor("aes-gcm")
+    match preferred_aead_backend(EncryptionAlgorithm::AesGcm) {
+        #[cfg(feature = "openssl-crypto")]
+        Some(AeadBackend::OpenSsl) => Arc::new(openssl::OpenSslCipher::new_aes128_gcm(key)),
+        #[cfg(all(not(feature = "openssl-crypto"), feature = "ring-crypto"))]
+        Some(AeadBackend::Ring) => Arc::new(ring::RingCipher::new_aes128_gcm(key)),
+        #[cfg(all(
+            not(feature = "openssl-crypto"),
+            not(feature = "ring-crypto"),
+            feature = "aes-gcm"
+        ))]
+        Some(AeadBackend::RustCrypto) => Arc::new(aes_gcm::AesGcmCipher::new_128(key)),
+        _ => unavailable_encryptor("aes-gcm"),
     }
 }
 
+#[allow(unreachable_patterns)]
 fn create_aes_256(key: [u8; 32]) -> Arc<dyn Encryptor> {
-    #[cfg(feature = "aes-gcm")]
-    {
-        Arc::new(aes_gcm::AesGcmCipher::new_256(key))
-    }
-    #[cfg(not(feature = "aes-gcm"))]
-    {
-        let _ = key;
-        unavailable_encryptor("aes-256-gcm")
+    match preferred_aead_backend(EncryptionAlgorithm::Aes256Gcm) {
+        #[cfg(feature = "openssl-crypto")]
+        Some(AeadBackend::OpenSsl) => Arc::new(openssl::OpenSslCipher::new_aes256_gcm(key)),
+        #[cfg(all(not(feature = "openssl-crypto"), feature = "ring-crypto"))]
+        Some(AeadBackend::Ring) => Arc::new(ring::RingCipher::new_aes256_gcm(key)),
+        #[cfg(all(
+            not(feature = "openssl-crypto"),
+            not(feature = "ring-crypto"),
+            feature = "aes-gcm"
+        ))]
+        Some(AeadBackend::RustCrypto) => Arc::new(aes_gcm::AesGcmCipher::new_256(key)),
+        _ => unavailable_encryptor("aes-256-gcm"),
     }
 }
 
+#[allow(unreachable_patterns)]
 fn create_chacha20(key: [u8; 32]) -> Arc<dyn Encryptor> {
-    #[cfg(feature = "chacha20")]
-    {
-        Arc::new(chacha20::ChaCha20Cipher::new(key))
-    }
-    #[cfg(not(feature = "chacha20"))]
-    {
-        let _ = key;
-        unavailable_encryptor("chacha20")
+    match preferred_aead_backend(EncryptionAlgorithm::ChaCha20) {
+        #[cfg(feature = "openssl-crypto")]
+        Some(AeadBackend::OpenSsl) => Arc::new(openssl::OpenSslCipher::new_chacha20(key)),
+        #[cfg(all(not(feature = "openssl-crypto"), feature = "ring-crypto"))]
+        Some(AeadBackend::Ring) => Arc::new(ring::RingCipher::new_chacha20(key)),
+        #[cfg(all(
+            not(feature = "openssl-crypto"),
+            not(feature = "ring-crypto"),
+            feature = "chacha20"
+        ))]
+        Some(AeadBackend::RustCrypto) => Arc::new(chacha20::ChaCha20Cipher::new(key)),
+        _ => unavailable_encryptor("chacha20"),
     }
 }
 
@@ -203,6 +291,28 @@ pub fn create_encryptor(
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::packet::{StandardAeadTail, ZCPacket};
+
+    fn assert_interoperable(left: &dyn Encryptor, right: &dyn Encryptor) {
+        let plaintext = b"cross-backend compatibility";
+        let nonce = [9; StandardAeadTail::NONCE_SIZE];
+        let mut left_packet = ZCPacket::new_with_payload(plaintext);
+        left_packet.fill_peer_manager_hdr(1, 2, 1);
+        let mut right_packet = ZCPacket::new_with_payload(plaintext);
+        right_packet.fill_peer_manager_hdr(1, 2, 1);
+
+        left.encrypt_with_nonce(&mut left_packet, Some(&nonce))
+            .unwrap();
+        right
+            .encrypt_with_nonce(&mut right_packet, Some(&nonce))
+            .unwrap();
+        assert_eq!(left_packet.payload(), right_packet.payload());
+
+        left.decrypt(&mut right_packet).unwrap();
+        right.decrypt(&mut left_packet).unwrap();
+        assert_eq!(left_packet.payload(), plaintext);
+        assert_eq!(right_packet.payload(), plaintext);
+    }
 
     #[test]
     fn network_secret_key_derivation_is_stable() {
@@ -229,7 +339,11 @@ mod tests {
         assert!(!effective_algorithm_uses_xor("aes-gcm"));
     }
 
-    #[cfg(not(feature = "aes-gcm"))]
+    #[cfg(not(any(
+        feature = "aes-gcm",
+        feature = "openssl-crypto",
+        feature = "ring-crypto"
+    )))]
     #[test]
     fn unavailable_aes_is_known_but_rejected() {
         assert_eq!(
@@ -238,14 +352,22 @@ mod tests {
         );
     }
 
-    #[cfg(feature = "aes-gcm")]
+    #[cfg(any(
+        feature = "aes-gcm",
+        feature = "openssl-crypto",
+        feature = "ring-crypto"
+    ))]
     #[test]
     fn compiled_aes_is_available() {
         validate_algorithm("aes-gcm").unwrap();
         validate_algorithm("aes-256-gcm").unwrap();
     }
 
-    #[cfg(not(feature = "chacha20"))]
+    #[cfg(not(any(
+        feature = "chacha20",
+        feature = "openssl-crypto",
+        feature = "ring-crypto"
+    )))]
     #[test]
     fn unavailable_chacha20_is_known_but_rejected() {
         assert_eq!(
@@ -256,12 +378,102 @@ mod tests {
         );
     }
 
-    #[cfg(feature = "chacha20")]
+    #[cfg(any(
+        feature = "chacha20",
+        feature = "openssl-crypto",
+        feature = "ring-crypto"
+    ))]
     #[test]
     fn compiled_chacha20_is_available() {
         validate_algorithm("chacha20").unwrap();
     }
 
+    #[test]
+    fn accelerated_backends_take_precedence_over_rustcrypto() {
+        #[cfg(feature = "openssl-crypto")]
+        assert_eq!(
+            preferred_aead_backend(EncryptionAlgorithm::AesGcm),
+            Some(AeadBackend::OpenSsl)
+        );
+
+        #[cfg(all(not(feature = "openssl-crypto"), feature = "ring-crypto"))]
+        assert_eq!(
+            preferred_aead_backend(EncryptionAlgorithm::AesGcm),
+            Some(AeadBackend::Ring)
+        );
+
+        #[cfg(all(
+            not(feature = "openssl-crypto"),
+            not(feature = "ring-crypto"),
+            feature = "aes-gcm"
+        ))]
+        assert_eq!(
+            preferred_aead_backend(EncryptionAlgorithm::AesGcm),
+            Some(AeadBackend::RustCrypto)
+        );
+    }
+
+    #[cfg(all(feature = "ring-crypto", feature = "aes-gcm"))]
+    #[test]
+    fn ring_and_rustcrypto_aes_are_interoperable() {
+        assert_interoperable(
+            &ring::RingCipher::new_aes128_gcm([1; 16]),
+            &aes_gcm::AesGcmCipher::new_128([1; 16]),
+        );
+        assert_interoperable(
+            &ring::RingCipher::new_aes256_gcm([2; 32]),
+            &aes_gcm::AesGcmCipher::new_256([2; 32]),
+        );
+    }
+
+    #[cfg(all(feature = "ring-crypto", feature = "chacha20"))]
+    #[test]
+    fn ring_and_rustcrypto_chacha20_are_interoperable() {
+        assert_interoperable(
+            &ring::RingCipher::new_chacha20([3; 32]),
+            &chacha20::ChaCha20Cipher::new([3; 32]),
+        );
+    }
+
+    #[cfg(all(feature = "openssl-crypto", feature = "aes-gcm"))]
+    #[test]
+    fn openssl_and_rustcrypto_aes_are_interoperable() {
+        assert_interoperable(
+            &openssl::OpenSslCipher::new_aes128_gcm([1; 16]),
+            &aes_gcm::AesGcmCipher::new_128([1; 16]),
+        );
+        assert_interoperable(
+            &openssl::OpenSslCipher::new_aes256_gcm([2; 32]),
+            &aes_gcm::AesGcmCipher::new_256([2; 32]),
+        );
+    }
+
+    #[cfg(all(feature = "openssl-crypto", feature = "chacha20"))]
+    #[test]
+    fn openssl_and_rustcrypto_chacha20_are_interoperable() {
+        assert_interoperable(
+            &openssl::OpenSslCipher::new_chacha20([3; 32]),
+            &chacha20::ChaCha20Cipher::new([3; 32]),
+        );
+    }
+
+    #[cfg(all(feature = "openssl-crypto", feature = "ring-crypto"))]
+    #[test]
+    fn openssl_and_ring_algorithms_are_interoperable() {
+        assert_interoperable(
+            &openssl::OpenSslCipher::new_aes128_gcm([1; 16]),
+            &ring::RingCipher::new_aes128_gcm([1; 16]),
+        );
+        assert_interoperable(
+            &openssl::OpenSslCipher::new_aes256_gcm([2; 32]),
+            &ring::RingCipher::new_aes256_gcm([2; 32]),
+        );
+        assert_interoperable(
+            &openssl::OpenSslCipher::new_chacha20([3; 32]),
+            &ring::RingCipher::new_chacha20([3; 32]),
+        );
+    }
+
     #[test]
     fn invalid_algorithm_is_rejected() {
         assert_eq!(

easytier-core/src/tunnel/encrypt/openssl.rs

@@ -0,0 +1,159 @@
+use openssl::symm::{Cipher, Crypter, Mode};
+use rand::RngCore as _;
+use zerocopy::{AsBytes as _, FromBytes as _, FromZeroes as _};
+
+use crate::packet::{StandardAeadTail, ZCPacket};
+
+use super::{Encryptor, Error};
+
+#[derive(Clone)]
+pub struct OpenSslCipher {
+    cipher: OpenSslCipherKind,
+}
+
+#[derive(Clone, Copy)]
+enum OpenSslCipherKind {
+    Aes128Gcm([u8; 16]),
+    Aes256Gcm([u8; 32]),
+    ChaCha20Poly1305([u8; 32]),
+}
+
+impl OpenSslCipher {
+    pub fn new_aes128_gcm(key: [u8; 16]) -> Self {
+        Self {
+            cipher: OpenSslCipherKind::Aes128Gcm(key),
+        }
+    }
+
+    pub fn new_aes256_gcm(key: [u8; 32]) -> Self {
+        Self {
+            cipher: OpenSslCipherKind::Aes256Gcm(key),
+        }
+    }
+
+    pub fn new_chacha20(key: [u8; 32]) -> Self {
+        Self {
+            cipher: OpenSslCipherKind::ChaCha20Poly1305(key),
+        }
+    }
+
+    fn cipher_and_key(&self) -> (Cipher, &[u8]) {
+        match &self.cipher {
+            OpenSslCipherKind::Aes128Gcm(key) => (Cipher::aes_128_gcm(), key),
+            OpenSslCipherKind::Aes256Gcm(key) => (Cipher::aes_256_gcm(), key),
+            OpenSslCipherKind::ChaCha20Poly1305(key) => (Cipher::chacha20_poly1305(), key),
+        }
+    }
+}
+
+impl Encryptor for OpenSslCipher {
+    fn decrypt(&self, packet: &mut ZCPacket) -> Result<(), Error> {
+        let header = packet.peer_manager_header().unwrap();
+        if !header.is_encrypted() {
+            return Ok(());
+        }
+
+        let payload_len = packet.payload().len();
+        if payload_len < StandardAeadTail::SIZE {
+            return Err(Error::PacketTooShort(payload_len));
+        }
+
+        let (cipher, key) = self.cipher_and_key();
+        let tail = StandardAeadTail::ref_from_suffix(packet.payload()).unwrap();
+        let mut decrypter = Crypter::new(cipher, Mode::Decrypt, key, Some(&tail.nonce))
+            .map_err(|_| Error::DecryptionFailed)?;
+        decrypter
+            .set_tag(&tail.tag)
+            .map_err(|_| Error::DecryptionFailed)?;
+
+        let text_len = payload_len - StandardAeadTail::SIZE;
+        let mut output = vec![0; text_len + cipher.block_size()];
+        let mut written = decrypter
+            .update(&packet.payload()[..text_len], &mut output)
+            .map_err(|_| Error::DecryptionFailed)?;
+        written += decrypter
+            .finalize(&mut output[written..])
+            .map_err(|_| Error::DecryptionFailed)?;
+
+        packet.mut_payload()[..written].copy_from_slice(&output[..written]);
+        packet
+            .mut_peer_manager_header()
+            .unwrap()
+            .set_encrypted(false);
+        let old_len = packet.buf_len();
+        packet
+            .mut_inner()
+            .truncate(old_len - (payload_len - written));
+        Ok(())
+    }
+
+    fn encrypt(&self, packet: &mut ZCPacket) -> Result<(), Error> {
+        self.encrypt_with_nonce(packet, None)
+    }
+
+    fn encrypt_with_nonce(&self, packet: &mut ZCPacket, nonce: Option<&[u8]>) -> Result<(), Error> {
+        let header = packet.peer_manager_header().unwrap();
+        if header.is_encrypted() {
+            tracing::warn!(?packet, "packet is already encrypted");
+            return Ok(());
+        }
+
+        let (cipher, key) = self.cipher_and_key();
+        let mut tail = StandardAeadTail::new_zeroed();
+        match nonce {
+            Some(nonce) => {
+                tail.nonce = nonce.try_into().map_err(|_| Error::EncryptionFailed)?;
+            }
+            None => rand::thread_rng().fill_bytes(&mut tail.nonce),
+        }
+
+        let mut encrypter = Crypter::new(cipher, Mode::Encrypt, key, Some(&tail.nonce))
+            .map_err(|_| Error::EncryptionFailed)?;
+        let payload_len = packet.payload().len();
+        let mut output = vec![0; payload_len + cipher.block_size()];
+        let mut written = encrypter
+            .update(packet.payload(), &mut output)
+            .map_err(|_| Error::EncryptionFailed)?;
+        written += encrypter
+            .finalize(&mut output[written..])
+            .map_err(|_| Error::EncryptionFailed)?;
+        packet.mut_payload()[..written].copy_from_slice(&output[..written]);
+        encrypter
+            .get_tag(&mut tail.tag)
+            .map_err(|_| Error::EncryptionFailed)?;
+
+        packet
+            .mut_peer_manager_header()
+            .unwrap()
+            .set_encrypted(true);
+        packet.mut_inner().extend_from_slice(tail.as_bytes());
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn round_trip(cipher: OpenSslCipher) {
+        let plaintext = b"openssl accelerated packet";
+        let mut packet = ZCPacket::new_with_payload(plaintext);
+        packet.fill_peer_manager_hdr(1, 2, 1);
+
+        cipher
+            .encrypt_with_nonce(&mut packet, Some(&[3; StandardAeadTail::NONCE_SIZE]))
+            .unwrap();
+        assert!(packet.peer_manager_header().unwrap().is_encrypted());
+
+        cipher.decrypt(&mut packet).unwrap();
+        assert_eq!(packet.payload(), plaintext);
+        assert!(!packet.peer_manager_header().unwrap().is_encrypted());
+    }
+
+    #[test]
+    fn openssl_algorithms_round_trip() {
+        round_trip(OpenSslCipher::new_aes128_gcm([1; 16]));
+        round_trip(OpenSslCipher::new_aes256_gcm([2; 32]));
+        round_trip(OpenSslCipher::new_chacha20([3; 32]));
+    }
+}

easytier-core/src/tunnel/encrypt/ring.rs

@@ -0,0 +1,175 @@
+use rand::RngCore as _;
+use ring::aead::{self, LessSafeKey, UnboundKey};
+use zerocopy::{AsBytes as _, FromBytes as _, FromZeroes as _};
+
+use crate::packet::{StandardAeadTail, ZCPacket};
+
+use super::{Encryptor, Error};
+
+#[derive(Clone)]
+pub struct RingCipher {
+    cipher: RingCipherKind,
+}
+
+enum RingCipherKind {
+    Aes128Gcm(LessSafeKey, [u8; 16]),
+    Aes256Gcm(LessSafeKey, [u8; 32]),
+    ChaCha20Poly1305(LessSafeKey, [u8; 32]),
+}
+
+impl RingCipherKind {
+    fn key(&self) -> &LessSafeKey {
+        match self {
+            Self::Aes128Gcm(cipher, _)
+            | Self::Aes256Gcm(cipher, _)
+            | Self::ChaCha20Poly1305(cipher, _) => cipher,
+        }
+    }
+}
+
+impl Clone for RingCipherKind {
+    fn clone(&self) -> Self {
+        match self {
+            Self::Aes128Gcm(_, key) => Self::Aes128Gcm(
+                LessSafeKey::new(UnboundKey::new(&aead::AES_128_GCM, key).unwrap()),
+                *key,
+            ),
+            Self::Aes256Gcm(_, key) => Self::Aes256Gcm(
+                LessSafeKey::new(UnboundKey::new(&aead::AES_256_GCM, key).unwrap()),
+                *key,
+            ),
+            Self::ChaCha20Poly1305(_, key) => Self::ChaCha20Poly1305(
+                LessSafeKey::new(UnboundKey::new(&aead::CHACHA20_POLY1305, key).unwrap()),
+                *key,
+            ),
+        }
+    }
+}
+
+impl RingCipher {
+    pub fn new_aes128_gcm(key: [u8; 16]) -> Self {
+        Self {
+            cipher: RingCipherKind::Aes128Gcm(
+                LessSafeKey::new(UnboundKey::new(&aead::AES_128_GCM, &key).unwrap()),
+                key,
+            ),
+        }
+    }
+
+    pub fn new_aes256_gcm(key: [u8; 32]) -> Self {
+        Self {
+            cipher: RingCipherKind::Aes256Gcm(
+                LessSafeKey::new(UnboundKey::new(&aead::AES_256_GCM, &key).unwrap()),
+                key,
+            ),
+        }
+    }
+
+    pub fn new_chacha20(key: [u8; 32]) -> Self {
+        Self {
+            cipher: RingCipherKind::ChaCha20Poly1305(
+                LessSafeKey::new(UnboundKey::new(&aead::CHACHA20_POLY1305, &key).unwrap()),
+                key,
+            ),
+        }
+    }
+}
+
+impl Encryptor for RingCipher {
+    fn decrypt(&self, packet: &mut ZCPacket) -> Result<(), Error> {
+        let header = packet.peer_manager_header().unwrap();
+        if !header.is_encrypted() {
+            return Ok(());
+        }
+
+        let payload_len = packet.payload().len();
+        if payload_len < StandardAeadTail::SIZE {
+            return Err(Error::PacketTooShort(payload_len));
+        }
+
+        let text_and_tag_len = payload_len - StandardAeadTail::SIZE + StandardAeadTail::TAG_SIZE;
+        let tail = StandardAeadTail::ref_from_suffix(packet.payload()).unwrap();
+        let nonce = aead::Nonce::assume_unique_for_key(tail.nonce);
+
+        self.cipher
+            .key()
+            .open_in_place(
+                nonce,
+                aead::Aad::empty(),
+                &mut packet.mut_payload()[..text_and_tag_len],
+            )
+            .map_err(|_| Error::DecryptionFailed)?;
+
+        packet
+            .mut_peer_manager_header()
+            .unwrap()
+            .set_encrypted(false);
+        let old_len = packet.buf_len();
+        packet
+            .mut_inner()
+            .truncate(old_len - StandardAeadTail::SIZE);
+        Ok(())
+    }
+
+    fn encrypt(&self, packet: &mut ZCPacket) -> Result<(), Error> {
+        self.encrypt_with_nonce(packet, None)
+    }
+
+    fn encrypt_with_nonce(&self, packet: &mut ZCPacket, nonce: Option<&[u8]>) -> Result<(), Error> {
+        let header = packet.peer_manager_header().unwrap();
+        if header.is_encrypted() {
+            tracing::warn!(?packet, "packet is already encrypted");
+            return Ok(());
+        }
+
+        let mut tail = StandardAeadTail::new_zeroed();
+        match nonce {
+            Some(nonce) => {
+                tail.nonce = nonce.try_into().map_err(|_| Error::EncryptionFailed)?;
+            }
+            None => rand::thread_rng().fill_bytes(&mut tail.nonce),
+        }
+
+        let nonce = aead::Nonce::assume_unique_for_key(tail.nonce);
+        let tag = self
+            .cipher
+            .key()
+            .seal_in_place_separate_tag(nonce, aead::Aad::empty(), packet.mut_payload())
+            .map_err(|_| Error::EncryptionFailed)?;
+        tail.tag.copy_from_slice(tag.as_ref());
+
+        packet
+            .mut_peer_manager_header()
+            .unwrap()
+            .set_encrypted(true);
+        packet.mut_inner().extend_from_slice(tail.as_bytes());
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn round_trip(cipher: RingCipher) {
+        let plaintext = b"ring accelerated packet";
+        let mut packet = ZCPacket::new_with_payload(plaintext);
+        packet.fill_peer_manager_hdr(1, 2, 1);
+
+        cipher
+            .encrypt_with_nonce(&mut packet, Some(&[3; StandardAeadTail::NONCE_SIZE]))
+            .unwrap();
+        assert!(packet.peer_manager_header().unwrap().is_encrypted());
+
+        cipher.decrypt(&mut packet).unwrap();
+        assert_eq!(packet.payload(), plaintext);
+        assert!(!packet.peer_manager_header().unwrap().is_encrypted());
+    }
+
+    #[test]
+    fn ring_algorithms_round_trip() {
+        round_trip(RingCipher::new_aes128_gcm([1; 16]));
+        round_trip(RingCipher::new_aes256_gcm([2; 32]));
+        round_trip(RingCipher::new_chacha20([3; 32]));
+    }
+}

easytier-core/src/tunnel/secure_datagram.rs

@@ -755,7 +755,18 @@ fn now_ms() -> u64 {
 #[cfg(test)]
 mod tests {
     use super::*;
-    #[cfg(all(feature = "aes-gcm", feature = "chacha20"))]
+    #[cfg(all(
+        any(
+            feature = "aes-gcm",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        ),
+        any(
+            feature = "chacha20",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        )
+    ))]
     use crate::packet::PacketType;
 
     impl SecureDatagramSession {
@@ -775,7 +786,18 @@ mod tests {
     }
 
     #[test]
-    #[cfg(all(feature = "aes-gcm", feature = "chacha20"))]
+    #[cfg(all(
+        any(
+            feature = "aes-gcm",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        ),
+        any(
+            feature = "chacha20",
+            feature = "openssl-crypto",
+            feature = "ring-crypto"
+        )
+    ))]
     fn secure_datagram_supports_asymmetric_algorithms() {
         let root_key = SecureDatagramSession::new_root_key();
         let generation = 1u32;
@@ -837,7 +859,11 @@ mod tests {
     }
 
     #[test]
-    #[cfg(feature = "aes-gcm")]
+    #[cfg(any(
+        feature = "aes-gcm",
+        feature = "openssl-crypto",
+        feature = "ring-crypto"
+    ))]
     fn failed_decrypt_does_not_poison_replay_window() {
         use crate::packet::PacketType;
 

easytier-core/src/tunnel/tcp.rs

@@ -34,7 +34,7 @@ where
             .unwrap()
             .take()
             .expect("TcpTunnel can only be split once");
-        let (reader, writer) = tokio::io::split(socket);
+        let (reader, writer) = socket.into_split();
         (
             Box::pin(FramedReader::new(reader, self.max_packet_size)),
             Box::pin(FramedWriter::new(writer)),

easytier-core/src/tunnel/udp.rs

@@ -6,7 +6,7 @@ use std::{
 
 use bytes::BytesMut;
 use futures::{Sink, Stream};
-use tokio::sync::{oneshot, watch};
+use tokio::sync::watch;
 
 use crate::{
     packet::{UDP_TUNNEL_HEADER_SIZE, UdpPacketType, ZCPacket, ZCPacketType},
@@ -76,7 +76,10 @@ impl Stream for UdpTunnelStream {
         Poll::Ready(ret.map(|payload| {
             payload
                 .map_err(ring_socket_error_to_tunnel)
-                .and_then(|datagram| zcpacket_from_udp_session_payload(&datagram.payload))
+                .and_then(|datagram| match datagram.into_tunnel_packet() {
+                    Ok(packet) => Ok(packet),
+                    Err(payload) => zcpacket_from_udp_session_payload(&payload),
+                })
         }))
     }
 }
@@ -114,15 +117,10 @@ impl Sink<SinkItem> for UdpTunnelSink {
         }
 
         let packet = item.convert_type(ZCPacketType::UDP);
-        let payload = BytesMut::from(packet.udp_payload());
         this.codec
-            .validate_payload(&payload)
+            .validate_payload(packet.udp_payload())
             .map_err(TunnelError::IOError)?;
-        let (completion, _sent) = oneshot::channel();
-        let outbound = UdpSessionOutbound {
-            payload,
-            completion,
-        };
+        let outbound = UdpSessionOutbound::TunnelPacket(packet);
         this.session_send_tx
             .force_send(outbound)
             .map_err(ring_send_error_to_tunnel)

easytier-core/src/tunnel/web_security.rs

@@ -158,7 +158,7 @@ fn decode_noise_payload(payload: &[u8]) -> Option<&[u8]> {
 }
 
 pub fn web_secure_tunnel_supported() -> bool {
-    cfg!(feature = "aes-gcm")
+    crate::tunnel::encrypt::algorithm_is_available(crate::config::EncryptionAlgorithm::AesGcm)
 }
 
 fn web_secure_cipher_algorithm() -> Result<&'static str, TunnelError> {

easytier-core/src/wasi/runtime.rs

@@ -34,7 +34,11 @@ pub(super) fn new_wasi_core_runtime(
 ) -> anyhow::Result<WasiCoreRuntime> {
     use std::sync::Arc;
 
-    use crate::host::{dns::HostDnsResolver, packet::HostPacketSink, socket::HostSocketRuntime};
+    use crate::host::{
+        dns::HostDnsResolver,
+        packet::{HostPacket, HostPacketSink},
+        socket::HostSocketRuntime,
+    };
     use crate::{
         connectivity::connector_host::new_connector_host,
         instance::{CoreHostAdapters, CoreInstance},
@@ -402,7 +406,10 @@ mod abi {
         fn send_packet(&self, packet: Vec<u8>) {
             let packet_plane = self.core.core().packet_plane();
             self.execution.lock().unwrap().runtime.spawn(async move {
-                if let Err(error) = packet_plane.send_ip_packet(packet).await {
+                if let Err(error) = packet_plane
+                    .send_ip_packet(HostPacket::copy_from_payload(&packet))
+                    .await
+                {
                     tracing::warn!(?error, "host packet ingress failed");
                 }
             });