847 lines
32 KiB
Rust
847 lines
32 KiB
Rust
//! The network engine: endpoint lifecycle, connection registry and messaging.
|
|
|
|
use std::collections::HashMap;
|
|
use std::fmt;
|
|
use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
|
|
use std::sync::{Arc, Mutex, MutexGuard, PoisonError, Weak};
|
|
use std::time::Duration;
|
|
|
|
use iroh::endpoint::{Connection, RecvStream, SendStream, VarInt, presets};
|
|
use iroh::protocol::{AcceptError, ProtocolHandler, Router};
|
|
use iroh::{Endpoint, EndpointAddr, EndpointId};
|
|
use serde::Serialize;
|
|
use serde::de::DeserializeOwned;
|
|
use tokio::sync::{Semaphore, mpsc, watch};
|
|
use tokio::task::JoinSet;
|
|
use tokio::time::timeout;
|
|
use tracing::{debug, info, warn};
|
|
|
|
use crate::config::NetworkConfig;
|
|
use crate::error::{NetworkError, Result};
|
|
use crate::event::{ConnectionDirection, NetworkEvent, NetworkEventReceiver};
|
|
use crate::identity;
|
|
use crate::protocol::{
|
|
ALPN, Handshake, HandshakeAck, HandshakeErrorCode, MAX_HANDSHAKE_FRAME_SIZE,
|
|
MessageRejectReason, MessageRequest, MessageResponse, NetworkId, PROTOCOL_VERSION, SchemaId,
|
|
};
|
|
use crate::rendezvous::{RendezvousClient, RendezvousConfig, now_ms};
|
|
use crate::ticket::{PeerTicket, TICKET_VERSION};
|
|
use crate::wire;
|
|
|
|
/// Close code for a graceful, error-free close.
|
|
const CLOSE_CODE_OK: VarInt = VarInt::from_u32(0);
|
|
/// Close code used when the handshake was rejected by either side.
|
|
const CLOSE_CODE_HANDSHAKE_REJECTED: VarInt = VarInt::from_u32(1);
|
|
/// Close code used when the engine shuts down.
|
|
const CLOSE_CODE_SHUTDOWN: VarInt = VarInt::from_u32(2);
|
|
/// Close code used when a connection is replaced by a newer one.
|
|
const CLOSE_CODE_REPLACED: VarInt = VarInt::from_u32(3);
|
|
|
|
/// Extra bytes allowed on top of `max_message_size` for message framing
|
|
/// (request id, length varints and enum tags).
|
|
const MESSAGE_FRAME_OVERHEAD: usize = 64;
|
|
/// Maximum size of an encoded [`MessageResponse`] frame.
|
|
const MAX_RESPONSE_FRAME_SIZE: usize = 1024;
|
|
/// How long the accepting side keeps a rejected connection open so the
|
|
/// rejection ack reaches the initiator before the connection is torn down.
|
|
const REJECT_LINGER: Duration = Duration::from_secs(3);
|
|
/// How long [`NetworkEngine::shutdown`] waits for background tasks before
|
|
/// aborting them.
|
|
const SHUTDOWN_TASK_GRACE: Duration = Duration::from_secs(5);
|
|
/// Upper bound of the rendezvous round delay while no peer is connected yet.
|
|
const RENDEZVOUS_LONELY_INTERVAL: Duration = Duration::from_secs(15);
|
|
|
|
/// Bounds required of a domain message type.
|
|
///
|
|
/// This trait is implemented automatically for every type that satisfies the
|
|
/// bounds; applications never implement it by hand.
|
|
pub trait Message: Serialize + DeserializeOwned + Send + Sync + 'static {}
|
|
|
|
impl<T> Message for T where T: Serialize + DeserializeOwned + Send + Sync + 'static {}
|
|
|
|
/// Locks a mutex, recovering the guard if the mutex was poisoned.
|
|
///
|
|
/// The engine never panics while holding one of its locks, so poisoning can
|
|
/// only originate from a panic in unrelated user code; recovering is safe.
|
|
fn lock<T>(mutex: &Mutex<T>) -> MutexGuard<'_, T> {
|
|
mutex.lock().unwrap_or_else(PoisonError::into_inner)
|
|
}
|
|
|
|
/// State kept for every active peer connection.
|
|
struct PeerState {
|
|
connection: Connection,
|
|
direction: ConnectionDirection,
|
|
semaphore: Arc<Semaphore>,
|
|
/// Monotonic registration counter, used so a stale connection's cleanup
|
|
/// never removes a newer connection to the same peer.
|
|
generation: u64,
|
|
}
|
|
|
|
/// State shared between the engine handles, the protocol handler and all
|
|
/// background tasks.
|
|
struct Shared<M> {
|
|
config: NetworkConfig,
|
|
endpoint: Endpoint,
|
|
router: Mutex<Option<Router>>,
|
|
peers: Mutex<HashMap<EndpointId, PeerState>>,
|
|
events: Mutex<Option<mpsc::Sender<NetworkEvent<M>>>>,
|
|
tasks: Mutex<JoinSet<()>>,
|
|
next_generation: AtomicU64,
|
|
shutting_down: AtomicBool,
|
|
/// Broadcasts the start of the shutdown to long-running background
|
|
/// loops so they can stop promptly instead of being aborted.
|
|
shutdown_signal: watch::Sender<bool>,
|
|
}
|
|
|
|
impl<M: Message> Shared<M> {
|
|
fn is_shutting_down(&self) -> bool {
|
|
self.shutting_down.load(Ordering::SeqCst)
|
|
}
|
|
|
|
fn ensure_running(&self) -> Result<()> {
|
|
if self.is_shutting_down() {
|
|
Err(NetworkError::ShuttingDown)
|
|
} else {
|
|
Ok(())
|
|
}
|
|
}
|
|
|
|
/// Delivers an event to the application.
|
|
///
|
|
/// The channel is bounded; if it is full this awaits until the
|
|
/// application consumes events, applying natural back-pressure without
|
|
/// unbounded buffering.
|
|
async fn emit(&self, event: NetworkEvent<M>) {
|
|
let sender = lock(&self.events).clone();
|
|
if let Some(sender) = sender
|
|
&& sender.send(event).await.is_err()
|
|
{
|
|
debug!("event receiver dropped; event discarded");
|
|
}
|
|
}
|
|
|
|
/// Spawns a background task tracked until shutdown.
|
|
fn spawn_task<F>(&self, future: F)
|
|
where
|
|
F: Future<Output = ()> + Send + 'static,
|
|
{
|
|
let mut tasks = lock(&self.tasks);
|
|
// Reap already-finished tasks so the set does not grow unboundedly.
|
|
while tasks.try_join_next().is_some() {}
|
|
tasks.spawn(future);
|
|
}
|
|
|
|
/// Adds a connection to the registry, replacing (and closing) any previous
|
|
/// connection to the same peer. Returns the registration generation.
|
|
fn register_connection(
|
|
&self,
|
|
peer_id: EndpointId,
|
|
connection: Connection,
|
|
direction: ConnectionDirection,
|
|
) -> u64 {
|
|
let generation = self.next_generation.fetch_add(1, Ordering::Relaxed);
|
|
let state = PeerState {
|
|
connection,
|
|
direction,
|
|
semaphore: Arc::new(Semaphore::new(self.config.max_concurrent_streams_per_peer)),
|
|
generation,
|
|
};
|
|
let replaced = lock(&self.peers).insert(peer_id, state);
|
|
if let Some(old) = replaced {
|
|
debug!(peer = %peer_id, "replacing existing connection");
|
|
old.connection
|
|
.close(CLOSE_CODE_REPLACED, b"replaced by a new connection");
|
|
}
|
|
generation
|
|
}
|
|
|
|
/// Removes the connection from the registry and reports the disconnect,
|
|
/// unless a newer connection to the same peer took its place.
|
|
async fn cleanup_connection(
|
|
&self,
|
|
peer_id: EndpointId,
|
|
generation: u64,
|
|
reason: Option<String>,
|
|
) {
|
|
let removed = {
|
|
let mut peers = lock(&self.peers);
|
|
match peers.get(&peer_id) {
|
|
Some(state) if state.generation == generation => peers.remove(&peer_id),
|
|
_ => None,
|
|
}
|
|
};
|
|
if let Some(state) = removed {
|
|
info!(
|
|
peer = %peer_id,
|
|
direction = ?state.direction,
|
|
reason = reason.as_deref().unwrap_or("unknown"),
|
|
"peer disconnected"
|
|
);
|
|
self.emit(NetworkEvent::PeerDisconnected { peer_id, reason })
|
|
.await;
|
|
}
|
|
}
|
|
|
|
/// Validates a remote handshake against the local configuration.
|
|
fn validate_handshake(&self, handshake: &Handshake) -> Option<HandshakeErrorCode> {
|
|
if handshake.protocol_version != PROTOCOL_VERSION {
|
|
Some(HandshakeErrorCode::UnsupportedProtocolVersion)
|
|
} else if handshake.network_id != self.config.network_id {
|
|
Some(HandshakeErrorCode::NetworkMismatch)
|
|
} else if handshake.schema_id != self.config.schema_id {
|
|
Some(HandshakeErrorCode::SchemaMismatch)
|
|
} else {
|
|
None
|
|
}
|
|
}
|
|
|
|
/// Drives an established connection: accepts message streams until the
|
|
/// connection closes, then cleans up the registry entry.
|
|
async fn connection_loop(
|
|
self: Arc<Self>,
|
|
peer_id: EndpointId,
|
|
connection: Connection,
|
|
generation: u64,
|
|
) {
|
|
let semaphore = {
|
|
let peers = lock(&self.peers);
|
|
match peers.get(&peer_id) {
|
|
Some(state) if state.generation == generation => state.semaphore.clone(),
|
|
// The connection was already replaced; nothing to drive.
|
|
_ => return,
|
|
}
|
|
};
|
|
loop {
|
|
let (send, recv) = match connection.accept_bi().await {
|
|
Ok(streams) => streams,
|
|
Err(_) => break,
|
|
};
|
|
let permit = match semaphore.clone().acquire_owned().await {
|
|
Ok(permit) => permit,
|
|
Err(_) => break,
|
|
};
|
|
let shared = self.clone();
|
|
self.spawn_task(async move {
|
|
shared
|
|
.handle_message_stream(peer_id, send, recv, permit)
|
|
.await;
|
|
// The permit is released when the task finishes.
|
|
});
|
|
}
|
|
let reason = connection.close_reason().map(|err| err.to_string());
|
|
self.cleanup_connection(peer_id, generation, reason).await;
|
|
}
|
|
|
|
/// Handles a single incoming message stream: decodes the request, hands
|
|
/// the message to the application and acknowledges it.
|
|
async fn handle_message_stream(
|
|
&self,
|
|
peer_id: EndpointId,
|
|
mut send: SendStream,
|
|
mut recv: RecvStream,
|
|
_permit: tokio::sync::OwnedSemaphorePermit,
|
|
) {
|
|
let request_timeout = self.config.request_timeout;
|
|
let max_frame = self.config.max_message_size + MESSAGE_FRAME_OVERHEAD;
|
|
let request: MessageRequest =
|
|
match timeout(request_timeout, wire::read_frame(&mut recv, max_frame)).await {
|
|
Ok(Ok(request)) => request,
|
|
Ok(Err(err)) => {
|
|
self.report_protocol_error(Some(peer_id), format!("bad message frame: {err}"))
|
|
.await;
|
|
return;
|
|
}
|
|
Err(_) => {
|
|
self.report_protocol_error(
|
|
Some(peer_id),
|
|
"timed out reading message frame".to_string(),
|
|
)
|
|
.await;
|
|
return;
|
|
}
|
|
};
|
|
let request_id = request.request_id;
|
|
let response = if self.is_shutting_down() {
|
|
MessageResponse::Rejected {
|
|
request_id,
|
|
reason: MessageRejectReason::ShuttingDown,
|
|
}
|
|
} else {
|
|
match wire::decode::<M>(&request.payload) {
|
|
Ok(message) => {
|
|
self.emit(NetworkEvent::MessageReceived { peer_id, message })
|
|
.await;
|
|
MessageResponse::Accepted { request_id }
|
|
}
|
|
Err(err) => {
|
|
self.report_protocol_error(
|
|
Some(peer_id),
|
|
format!("failed to decode domain message: {err}"),
|
|
)
|
|
.await;
|
|
MessageResponse::Rejected {
|
|
request_id,
|
|
reason: MessageRejectReason::MalformedPayload,
|
|
}
|
|
}
|
|
}
|
|
};
|
|
let write = timeout(
|
|
request_timeout,
|
|
wire::write_frame(&mut send, &response, MAX_RESPONSE_FRAME_SIZE),
|
|
)
|
|
.await;
|
|
match write {
|
|
Ok(Ok(())) => {
|
|
let _ = send.finish();
|
|
}
|
|
Ok(Err(err)) => {
|
|
debug!(peer = %peer_id, error = %err, "failed to send message response");
|
|
}
|
|
Err(_) => {
|
|
debug!(peer = %peer_id, "timed out sending message response");
|
|
}
|
|
}
|
|
}
|
|
|
|
async fn report_protocol_error(&self, peer_id: Option<EndpointId>, error: String) {
|
|
match peer_id {
|
|
Some(peer) => warn!(peer = %peer, error = %error, "protocol error"),
|
|
None => warn!(error = %error, "protocol error"),
|
|
}
|
|
self.emit(NetworkEvent::ProtocolError { peer_id, error })
|
|
.await;
|
|
}
|
|
|
|
/// Handles a freshly accepted incoming connection: performs the handshake
|
|
/// and, if accepted, drives the connection until it closes.
|
|
async fn handle_incoming(self: &Arc<Self>, connection: &Connection) -> Result<()> {
|
|
let peer_id = connection.remote_id();
|
|
if self.is_shutting_down() {
|
|
connection.close(CLOSE_CODE_SHUTDOWN, b"engine is shutting down");
|
|
return Err(NetworkError::ShuttingDown);
|
|
}
|
|
debug!(peer = %peer_id, "incoming connection");
|
|
let request_timeout = self.config.request_timeout;
|
|
|
|
let (mut send, mut recv) = timeout(request_timeout, connection.accept_bi())
|
|
.await
|
|
.map_err(|_| NetworkError::Timeout)
|
|
.and_then(|res| {
|
|
res.map_err(|err| {
|
|
NetworkError::Transport(format!("failed to accept handshake stream: {err}"))
|
|
})
|
|
})
|
|
.inspect_err(|_| {
|
|
connection.close(CLOSE_CODE_HANDSHAKE_REJECTED, b"handshake failed");
|
|
})?;
|
|
|
|
let handshake: Result<Handshake> = timeout(
|
|
request_timeout,
|
|
wire::read_frame(&mut recv, MAX_HANDSHAKE_FRAME_SIZE),
|
|
)
|
|
.await
|
|
.map_err(|_| NetworkError::Timeout)
|
|
.and_then(|res| res);
|
|
|
|
let verdict = match &handshake {
|
|
Ok(handshake) => self.validate_handshake(handshake),
|
|
Err(_) => Some(HandshakeErrorCode::InvalidHandshake),
|
|
};
|
|
|
|
if let Some(code) = verdict {
|
|
warn!(peer = %peer_id, code = ?code, "handshake rejected");
|
|
let ack = HandshakeAck {
|
|
accepted: false,
|
|
error: Some(code),
|
|
};
|
|
let sent = timeout(
|
|
request_timeout,
|
|
wire::write_frame(&mut send, &ack, MAX_HANDSHAKE_FRAME_SIZE),
|
|
)
|
|
.await;
|
|
if matches!(sent, Ok(Ok(()))) {
|
|
let _ = send.finish();
|
|
// Keep the connection open briefly so QUIC delivers the ack;
|
|
// the initiator closes as soon as it has read the rejection.
|
|
let _ = timeout(REJECT_LINGER, connection.closed()).await;
|
|
}
|
|
connection.close(CLOSE_CODE_HANDSHAKE_REJECTED, b"handshake rejected");
|
|
return Err(handshake_code_to_error(code));
|
|
}
|
|
|
|
let ack = HandshakeAck {
|
|
accepted: true,
|
|
error: None,
|
|
};
|
|
timeout(
|
|
request_timeout,
|
|
wire::write_frame(&mut send, &ack, MAX_HANDSHAKE_FRAME_SIZE),
|
|
)
|
|
.await
|
|
.map_err(|_| NetworkError::Timeout)
|
|
.and_then(|res| res)
|
|
.inspect_err(|_| {
|
|
connection.close(CLOSE_CODE_HANDSHAKE_REJECTED, b"handshake failed");
|
|
})?;
|
|
let _ = send.finish();
|
|
|
|
let generation =
|
|
self.register_connection(peer_id, connection.clone(), ConnectionDirection::Incoming);
|
|
info!(peer = %peer_id, "peer connected (incoming)");
|
|
self.emit(NetworkEvent::PeerConnected {
|
|
peer_id,
|
|
direction: ConnectionDirection::Incoming,
|
|
})
|
|
.await;
|
|
// Drive the connection inside the router's accept task; it runs for
|
|
// as long as the connection lives.
|
|
self.clone()
|
|
.connection_loop(peer_id, connection.clone(), generation)
|
|
.await;
|
|
Ok(())
|
|
}
|
|
|
|
/// Dials `addr`, runs the client handshake and registers the connection.
|
|
///
|
|
/// An existing healthy connection to the same peer is reused. The remote
|
|
/// identity is taken from the authenticated Iroh connection, never from
|
|
/// `addr` itself.
|
|
async fn connect_to_addr(self: &Arc<Self>, addr: EndpointAddr) -> Result<EndpointId> {
|
|
self.ensure_running()?;
|
|
let target = addr.id;
|
|
if lock(&self.peers).contains_key(&target) {
|
|
debug!(peer = %target, "reusing existing connection");
|
|
return Ok(target);
|
|
}
|
|
|
|
let request_timeout = self.config.request_timeout;
|
|
let connection = timeout(request_timeout, self.endpoint.connect(addr, ALPN))
|
|
.await
|
|
.map_err(|_| NetworkError::Timeout)?
|
|
.map_err(|err| NetworkError::Transport(format!("failed to connect: {err}")))?;
|
|
let peer_id = connection.remote_id();
|
|
|
|
let handshake = timeout(request_timeout, self.run_client_handshake(&connection))
|
|
.await
|
|
.map_err(|_| NetworkError::Timeout)
|
|
.and_then(|res| res);
|
|
if let Err(err) = handshake {
|
|
connection.close(CLOSE_CODE_HANDSHAKE_REJECTED, b"handshake failed");
|
|
return Err(err);
|
|
}
|
|
|
|
let generation =
|
|
self.register_connection(peer_id, connection.clone(), ConnectionDirection::Outgoing);
|
|
info!(peer = %peer_id, "peer connected (outgoing)");
|
|
self.emit(NetworkEvent::PeerConnected {
|
|
peer_id,
|
|
direction: ConnectionDirection::Outgoing,
|
|
})
|
|
.await;
|
|
let loop_shared = self.clone();
|
|
self.spawn_task(async move {
|
|
loop_shared
|
|
.connection_loop(peer_id, connection, generation)
|
|
.await;
|
|
});
|
|
Ok(peer_id)
|
|
}
|
|
|
|
/// Runs the client side of the handshake on a fresh outgoing connection.
|
|
async fn run_client_handshake(&self, connection: &Connection) -> Result<()> {
|
|
let (mut send, mut recv) = connection.open_bi().await.map_err(|err| {
|
|
NetworkError::Transport(format!("failed to open handshake stream: {err}"))
|
|
})?;
|
|
let handshake = Handshake {
|
|
protocol_version: PROTOCOL_VERSION,
|
|
network_id: self.config.network_id,
|
|
schema_id: self.config.schema_id,
|
|
};
|
|
wire::write_frame(&mut send, &handshake, MAX_HANDSHAKE_FRAME_SIZE).await?;
|
|
let _ = send.finish();
|
|
let ack: HandshakeAck = wire::read_frame(&mut recv, MAX_HANDSHAKE_FRAME_SIZE).await?;
|
|
if ack.accepted {
|
|
Ok(())
|
|
} else {
|
|
let code = ack.error.unwrap_or(HandshakeErrorCode::InvalidHandshake);
|
|
Err(handshake_code_to_error(code))
|
|
}
|
|
}
|
|
}
|
|
|
|
/// Periodically publishes this peer to the network's rendezvous record and
|
|
/// dials every other peer found there.
|
|
///
|
|
/// Failures of a single round or dial are logged and retried on the next
|
|
/// round; the loop only ends when the engine shuts down (the task is
|
|
/// aborted).
|
|
async fn rendezvous_loop<M: Message>(
|
|
shared: Arc<Shared<M>>,
|
|
client: RendezvousClient,
|
|
config: RendezvousConfig,
|
|
) {
|
|
let mut shutdown = shared.shutdown_signal.subscribe();
|
|
// Give the endpoint a moment to learn its relay and direct addresses so
|
|
// the very first published record is already dialable.
|
|
let _ = timeout(shared.config.request_timeout, shared.endpoint.online()).await;
|
|
let self_id = shared.endpoint.id();
|
|
loop {
|
|
if shared.is_shutting_down() {
|
|
return;
|
|
}
|
|
let round = async {
|
|
let addr = shared.endpoint.addr();
|
|
let self_addr = (!addr.is_empty()).then_some(addr);
|
|
match client.round(self_addr, now_ms()).await {
|
|
Ok(peers) => {
|
|
for peer_addr in peers {
|
|
let peer = peer_addr.id;
|
|
if peer == self_id || lock(&shared.peers).contains_key(&peer) {
|
|
continue;
|
|
}
|
|
debug!(peer = %peer, "rendezvous discovered a peer; connecting");
|
|
if let Err(err) = shared.connect_to_addr(peer_addr).await {
|
|
// Stale entries (peers that left) fail here; they
|
|
// age out of the record by TTL.
|
|
debug!(peer = %peer, error = %err, "rendezvous connect attempt failed");
|
|
}
|
|
}
|
|
}
|
|
Err(err) => debug!(error = %err, "rendezvous round failed"),
|
|
}
|
|
};
|
|
tokio::select! {
|
|
_ = round => {}
|
|
_ = shutdown.changed() => return,
|
|
}
|
|
// Poll faster while the peer is still alone: joining a network
|
|
// should not have to wait a full interval for the next round.
|
|
let delay = if lock(&shared.peers).is_empty() {
|
|
config.interval.min(RENDEZVOUS_LONELY_INTERVAL)
|
|
} else {
|
|
config.interval
|
|
};
|
|
tokio::select! {
|
|
_ = tokio::time::sleep(delay) => {}
|
|
_ = shutdown.changed() => return,
|
|
}
|
|
}
|
|
}
|
|
|
|
fn handshake_code_to_error(code: HandshakeErrorCode) -> NetworkError {
|
|
match code {
|
|
HandshakeErrorCode::UnsupportedProtocolVersion => NetworkError::UnsupportedProtocolVersion,
|
|
HandshakeErrorCode::NetworkMismatch => NetworkError::NetworkMismatch,
|
|
HandshakeErrorCode::SchemaMismatch => NetworkError::SchemaMismatch,
|
|
HandshakeErrorCode::InvalidHandshake => NetworkError::HandshakeRejected,
|
|
}
|
|
}
|
|
|
|
/// Protocol handler registered with the Iroh [`Router`] for [`ALPN`].
|
|
///
|
|
/// Holds only a weak reference to the shared state so the router does not
|
|
/// keep the engine alive after all engine handles are dropped.
|
|
struct FederationProtocol<M> {
|
|
shared: Weak<Shared<M>>,
|
|
}
|
|
|
|
impl<M> fmt::Debug for FederationProtocol<M> {
|
|
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
|
|
f.debug_struct("FederationProtocol").finish_non_exhaustive()
|
|
}
|
|
}
|
|
|
|
impl<M: Message> ProtocolHandler for FederationProtocol<M> {
|
|
async fn accept(&self, connection: Connection) -> Result<(), AcceptError> {
|
|
let Some(shared) = self.shared.upgrade() else {
|
|
connection.close(CLOSE_CODE_SHUTDOWN, b"engine is gone");
|
|
return Ok(());
|
|
};
|
|
let peer_id = connection.remote_id();
|
|
if let Err(err) = shared.handle_incoming(&connection).await {
|
|
shared
|
|
.report_protocol_error(Some(peer_id), format!("incoming connection failed: {err}"))
|
|
.await;
|
|
}
|
|
// Errors of a single connection never bring down the router; the
|
|
// connection has already been closed with an appropriate code.
|
|
Ok(())
|
|
}
|
|
}
|
|
|
|
/// A generic peer-to-peer network engine on top of Iroh.
|
|
///
|
|
/// `M` is the application-defined domain message type; the engine treats it
|
|
/// as an opaque, postcard-serializable payload. The engine is cheaply
|
|
/// clonable: all clones share the same endpoint and connection registry.
|
|
pub struct NetworkEngine<M> {
|
|
shared: Arc<Shared<M>>,
|
|
}
|
|
|
|
impl<M> Clone for NetworkEngine<M> {
|
|
fn clone(&self) -> Self {
|
|
Self {
|
|
shared: self.shared.clone(),
|
|
}
|
|
}
|
|
}
|
|
|
|
impl<M> fmt::Debug for NetworkEngine<M> {
|
|
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
|
|
f.debug_struct("NetworkEngine").finish_non_exhaustive()
|
|
}
|
|
}
|
|
|
|
impl<M: Message> NetworkEngine<M> {
|
|
/// Starts the engine: loads (or creates) the persistent identity, binds
|
|
/// the Iroh endpoint and starts accepting connections.
|
|
///
|
|
/// Returns the engine handle together with the single receiver for
|
|
/// [`NetworkEvent`]s.
|
|
pub async fn start(config: NetworkConfig) -> Result<(Self, NetworkEventReceiver<M>)> {
|
|
let secret_key = identity::load_or_create(&config.data_dir).await?;
|
|
let endpoint = Endpoint::builder(presets::N0)
|
|
.secret_key(secret_key)
|
|
.bind()
|
|
.await
|
|
.map_err(|err| NetworkError::Transport(format!("failed to bind endpoint: {err}")))?;
|
|
let endpoint_id = endpoint.id();
|
|
let (sender, receiver) = mpsc::channel(config.event_channel_capacity);
|
|
let shared = Arc::new(Shared {
|
|
config,
|
|
endpoint: endpoint.clone(),
|
|
router: Mutex::new(None),
|
|
peers: Mutex::new(HashMap::new()),
|
|
events: Mutex::new(Some(sender)),
|
|
tasks: Mutex::new(JoinSet::new()),
|
|
next_generation: AtomicU64::new(0),
|
|
shutting_down: AtomicBool::new(false),
|
|
shutdown_signal: watch::Sender::new(false),
|
|
});
|
|
let handler = FederationProtocol {
|
|
shared: Arc::downgrade(&shared),
|
|
};
|
|
let router = Router::builder(endpoint).accept(ALPN, handler).spawn();
|
|
*lock(&shared.router) = Some(router);
|
|
if let Some(rendezvous) = shared.config.rendezvous.clone() {
|
|
match RendezvousClient::new(shared.config.network_id, &rendezvous) {
|
|
Ok(client) => {
|
|
let loop_shared = shared.clone();
|
|
shared.spawn_task(async move {
|
|
rendezvous_loop(loop_shared, client, rendezvous).await;
|
|
});
|
|
}
|
|
// Rendezvous is a convenience; the engine stays usable via
|
|
// tickets even when the DHT client cannot start.
|
|
Err(err) => warn!(error = %err, "peer rendezvous disabled"),
|
|
}
|
|
}
|
|
info!(
|
|
endpoint_id = %endpoint_id,
|
|
network_id = %shared.config.network_id,
|
|
schema_id = %shared.config.schema_id,
|
|
"network engine started"
|
|
);
|
|
Ok((Self { shared }, NetworkEventReceiver::new(receiver)))
|
|
}
|
|
|
|
/// Returns the stable identifier of this peer.
|
|
pub fn endpoint_id(&self) -> EndpointId {
|
|
self.shared.endpoint.id()
|
|
}
|
|
|
|
/// Returns the network this engine participates in.
|
|
pub fn network_id(&self) -> NetworkId {
|
|
self.shared.config.network_id
|
|
}
|
|
|
|
/// Returns the message schema this engine uses.
|
|
pub fn schema_id(&self) -> SchemaId {
|
|
self.shared.config.schema_id
|
|
}
|
|
|
|
/// Creates a shareable [`PeerTicket`] for this peer.
|
|
///
|
|
/// Waits (bounded by the configured request timeout) for the endpoint to
|
|
/// come online so the ticket contains relay information whenever
|
|
/// possible; on an isolated network the ticket falls back to the locally
|
|
/// known direct addresses.
|
|
pub async fn ticket(&self) -> Result<PeerTicket> {
|
|
self.shared.ensure_running()?;
|
|
let endpoint = &self.shared.endpoint;
|
|
let has_relay = endpoint.addr().relay_urls().next().is_some();
|
|
if !has_relay {
|
|
let _ = timeout(self.shared.config.request_timeout, endpoint.online()).await;
|
|
}
|
|
let endpoint_addr = endpoint.addr();
|
|
if endpoint_addr.is_empty() {
|
|
return Err(NetworkError::Transport(
|
|
"endpoint has no reachable addresses yet".to_string(),
|
|
));
|
|
}
|
|
Ok(PeerTicket {
|
|
ticket_version: TICKET_VERSION,
|
|
protocol_version: PROTOCOL_VERSION,
|
|
network_id: self.shared.config.network_id,
|
|
schema_id: self.shared.config.schema_id,
|
|
endpoint_addr,
|
|
})
|
|
}
|
|
|
|
/// Connects to the peer described by `ticket` and performs the handshake.
|
|
///
|
|
/// Incompatible tickets (different network, schema or protocol version)
|
|
/// are rejected before any connection is attempted. An existing healthy
|
|
/// connection to the same peer is reused.
|
|
pub async fn connect(&self, ticket: PeerTicket) -> Result<EndpointId> {
|
|
let shared = &self.shared;
|
|
shared.ensure_running()?;
|
|
if ticket.ticket_version != TICKET_VERSION {
|
|
return Err(NetworkError::InvalidTicket(format!(
|
|
"unsupported ticket version {}",
|
|
ticket.ticket_version
|
|
)));
|
|
}
|
|
if ticket.protocol_version != PROTOCOL_VERSION {
|
|
return Err(NetworkError::UnsupportedProtocolVersion);
|
|
}
|
|
if ticket.network_id != shared.config.network_id {
|
|
return Err(NetworkError::NetworkMismatch);
|
|
}
|
|
if ticket.schema_id != shared.config.schema_id {
|
|
return Err(NetworkError::SchemaMismatch);
|
|
}
|
|
if ticket.endpoint_id() == self.endpoint_id() {
|
|
return Err(NetworkError::InvalidTicket(
|
|
"the ticket points to this peer itself".to_string(),
|
|
));
|
|
}
|
|
shared.connect_to_addr(ticket.endpoint_addr).await
|
|
}
|
|
|
|
/// Sends a domain message to a connected peer.
|
|
///
|
|
/// Returns only after the peer acknowledged the message. Fails with
|
|
/// [`NetworkError::PeerNotConnected`] if there is no active connection to
|
|
/// the peer; no implicit dialing or peer lookup is performed.
|
|
pub async fn send(&self, peer: EndpointId, message: &M) -> Result<()> {
|
|
let shared = &self.shared;
|
|
shared.ensure_running()?;
|
|
let connection = lock(&shared.peers)
|
|
.get(&peer)
|
|
.map(|state| state.connection.clone())
|
|
.ok_or(NetworkError::PeerNotConnected(peer))?;
|
|
let payload = wire::encode(message, shared.config.max_message_size)?;
|
|
let request_id: [u8; 16] = rand::random();
|
|
let request = MessageRequest {
|
|
request_id,
|
|
payload,
|
|
};
|
|
let max_frame = shared.config.max_message_size + MESSAGE_FRAME_OVERHEAD;
|
|
timeout(shared.config.request_timeout, async move {
|
|
let (mut send, mut recv) = connection.open_bi().await.map_err(|err| {
|
|
NetworkError::Transport(format!("failed to open message stream: {err}"))
|
|
})?;
|
|
wire::write_frame(&mut send, &request, max_frame).await?;
|
|
let _ = send.finish();
|
|
let response: MessageResponse =
|
|
wire::read_frame(&mut recv, MAX_RESPONSE_FRAME_SIZE).await?;
|
|
match response {
|
|
MessageResponse::Accepted { request_id: id } if id == request_id => Ok(()),
|
|
MessageResponse::Rejected {
|
|
request_id: id,
|
|
reason,
|
|
} if id == request_id => Err(NetworkError::MessageRejected(reason.to_string())),
|
|
_ => Err(NetworkError::Transport(
|
|
"peer responded to a different request".to_string(),
|
|
)),
|
|
}
|
|
})
|
|
.await
|
|
.map_err(|_| NetworkError::Timeout)?
|
|
}
|
|
|
|
/// Returns the ids of all currently connected peers.
|
|
pub fn connected_peers(&self) -> Vec<EndpointId> {
|
|
lock(&self.shared.peers).keys().copied().collect()
|
|
}
|
|
|
|
/// Returns `true` if there is an active connection to `peer`.
|
|
pub fn is_connected(&self, peer: EndpointId) -> bool {
|
|
lock(&self.shared.peers).contains_key(&peer)
|
|
}
|
|
|
|
/// Closes the connection to `peer` and removes it from the registry.
|
|
pub async fn disconnect(&self, peer: EndpointId) -> Result<()> {
|
|
let state = lock(&self.shared.peers)
|
|
.remove(&peer)
|
|
.ok_or(NetworkError::PeerNotConnected(peer))?;
|
|
state
|
|
.connection
|
|
.close(CLOSE_CODE_OK, b"disconnected by local peer");
|
|
info!(peer = %peer, "peer disconnected (local request)");
|
|
self.shared
|
|
.emit(NetworkEvent::PeerDisconnected {
|
|
peer_id: peer,
|
|
reason: Some("disconnected by local peer".to_string()),
|
|
})
|
|
.await;
|
|
Ok(())
|
|
}
|
|
|
|
/// Shuts the engine down gracefully.
|
|
///
|
|
/// Rejects new operations, closes all connections, stops the protocol
|
|
/// router, closes the endpoint, waits for background tasks and finally
|
|
/// closes the event channel. Calling `shutdown` on several clones of the
|
|
/// same engine is safe; only the first call does the work.
|
|
pub async fn shutdown(self) -> Result<()> {
|
|
let shared = &self.shared;
|
|
if shared.shutting_down.swap(true, Ordering::SeqCst) {
|
|
return Ok(());
|
|
}
|
|
info!("network engine shutting down");
|
|
let _ = shared.shutdown_signal.send(true);
|
|
|
|
// Close all active connections.
|
|
let states: Vec<PeerState> = lock(&shared.peers)
|
|
.drain()
|
|
.map(|(_, state)| state)
|
|
.collect();
|
|
for state in &states {
|
|
state
|
|
.connection
|
|
.close(CLOSE_CODE_SHUTDOWN, b"engine is shutting down");
|
|
}
|
|
drop(states);
|
|
|
|
// Shut down the router; this stops the protocol handler and closes
|
|
// the endpoint.
|
|
let router = lock(&shared.router).take();
|
|
if let Some(router) = router
|
|
&& let Err(err) = router.shutdown().await
|
|
{
|
|
warn!(error = %err, "router shutdown reported an error");
|
|
}
|
|
shared.endpoint.close().await;
|
|
|
|
// Wait for the remaining background tasks; abort stragglers.
|
|
let mut tasks = std::mem::take(&mut *lock(&shared.tasks));
|
|
let drained = timeout(SHUTDOWN_TASK_GRACE, async {
|
|
while tasks.join_next().await.is_some() {}
|
|
})
|
|
.await;
|
|
if drained.is_err() {
|
|
warn!("background tasks did not finish in time; aborting them");
|
|
tasks.abort_all();
|
|
}
|
|
|
|
// Close the event channel: dropping the last sender makes
|
|
// `NetworkEventReceiver::recv` return `None` once drained.
|
|
*lock(&shared.events) = None;
|
|
info!("network engine shut down");
|
|
Ok(())
|
|
}
|
|
}
|