X-Git-Url: http://git.bitcoin.ninja/index.cgi?a=blobdiff_plain;f=lightning-net-tokio%2Fsrc%2Flib.rs;h=2ac10762b04eaf23f602bc3d8dcc987438542cae;hb=ab20284e2673e946ebdffcd5be7388de917a168f;hp=d102778a460f20d202fe772920d1c0ee3418e40b;hpb=a1f95de7342b32a097cb1375a6e48b262fc3730a;p=rust-lightning diff --git a/lightning-net-tokio/src/lib.rs b/lightning-net-tokio/src/lib.rs index d102778a..2ac10762 100644 --- a/lightning-net-tokio/src/lib.rs +++ b/lightning-net-tokio/src/lib.rs @@ -23,7 +23,7 @@ //! # Example //! ``` //! use std::net::TcpStream; -//! use bitcoin::secp256k1::key::PublicKey; +//! use bitcoin::secp256k1::PublicKey; //! use lightning::util::events::{Event, EventHandler, EventsProvider}; //! use std::net::SocketAddr; //! use std::sync::Arc; @@ -34,7 +34,7 @@ //! type Logger = dyn lightning::util::logger::Logger + Send + Sync; //! type ChainAccess = dyn lightning::chain::Access + Send + Sync; //! type ChainFilter = dyn lightning::chain::Filter + Send + Sync; -//! type DataPersister = dyn lightning::chain::channelmonitor::Persist + Send + Sync; +//! type DataPersister = dyn lightning::chain::chainmonitor::Persist + Send + Sync; //! type ChainMonitor = lightning::chain::chainmonitor::ChainMonitor, Arc, Arc, Arc, Arc>; //! type ChannelManager = Arc>; //! type PeerManager = Arc>; @@ -43,13 +43,12 @@ //! async fn connect_to_node(peer_manager: PeerManager, chain_monitor: Arc, channel_manager: ChannelManager, their_node_id: PublicKey, addr: SocketAddr) { //! lightning_net_tokio::connect_outbound(peer_manager, their_node_id, addr).await; //! loop { -//! channel_manager.await_persistable_update(); -//! channel_manager.process_pending_events(&|event| { -//! // Handle the event! -//! }); -//! chain_monitor.process_pending_events(&|event| { +//! let event_handler = |event: &Event| { //! // Handle the event! -//! }); +//! }; +//! channel_manager.await_persistable_update(); +//! channel_manager.process_pending_events(&event_handler); +//! chain_monitor.process_pending_events(&event_handler); //! } //! } //! @@ -57,13 +56,12 @@ //! async fn accept_socket(peer_manager: PeerManager, chain_monitor: Arc, channel_manager: ChannelManager, socket: TcpStream) { //! lightning_net_tokio::setup_inbound(peer_manager, socket); //! loop { -//! channel_manager.await_persistable_update(); -//! channel_manager.process_pending_events(&|event| { +//! let event_handler = |event: &Event| { //! // Handle the event! -//! }); -//! chain_monitor.process_pending_events(&|event| { -//! // Handle the event! -//! }); +//! }; +//! channel_manager.await_persistable_update(); +//! channel_manager.process_pending_events(&event_handler); +//! chain_monitor.process_pending_events(&event_handler); //! } //! } //! ``` @@ -71,7 +69,9 @@ #![deny(broken_intra_doc_links)] #![deny(missing_docs)] -use bitcoin::secp256k1::key::PublicKey; +#![cfg_attr(docsrs, feature(doc_auto_cfg))] + +use bitcoin::secp256k1::PublicKey; use tokio::net::TcpStream; use tokio::{io, time}; @@ -80,10 +80,11 @@ use tokio::io::{AsyncReadExt, AsyncWrite, AsyncWriteExt}; use lightning::ln::peer_handler; use lightning::ln::peer_handler::SocketDescriptor as LnSocketTrait; -use lightning::ln::msgs::{ChannelMessageHandler, RoutingMessageHandler}; +use lightning::ln::peer_handler::CustomMessageHandler; +use lightning::ln::msgs::{ChannelMessageHandler, RoutingMessageHandler, NetAddress}; use lightning::util::logger::Logger; -use std::{task, thread}; +use std::task; use std::net::SocketAddr; use std::net::TcpStream as StdTcpStream; use std::sync::{Arc, Mutex}; @@ -114,20 +115,33 @@ struct Connection { // socket. To wake it up (without otherwise changing its state, we can push a value into this // Sender. read_waker: mpsc::Sender<()>, - // When we are told by rust-lightning to disconnect, we can't return to rust-lightning until we - // are sure we won't call any more read/write PeerManager functions with the same connection. - // This is set to true if we're in such a condition (with disconnect checked before with the - // top-level mutex held) and false when we can return. - block_disconnect_socket: bool, read_paused: bool, rl_requested_disconnect: bool, id: u64, } impl Connection { - async fn schedule_read(peer_manager: Arc, Arc, Arc>>, us: Arc>, mut reader: io::ReadHalf, mut read_wake_receiver: mpsc::Receiver<()>, mut write_avail_receiver: mpsc::Receiver<()>) where - CMH: ChannelMessageHandler + 'static, - RMH: RoutingMessageHandler + 'static, - L: Logger + 'static + ?Sized { + async fn poll_event_process(peer_manager: Arc, Arc, Arc, Arc>>, mut event_receiver: mpsc::Receiver<()>) where + CMH: ChannelMessageHandler + 'static + Send + Sync, + RMH: RoutingMessageHandler + 'static + Send + Sync, + L: Logger + 'static + ?Sized + Send + Sync, + UMH: CustomMessageHandler + 'static + Send + Sync { + loop { + if event_receiver.recv().await.is_none() { + return; + } + peer_manager.process_events(); + } + } + + async fn schedule_read(peer_manager: Arc, Arc, Arc, Arc>>, us: Arc>, mut reader: io::ReadHalf, mut read_wake_receiver: mpsc::Receiver<()>, mut write_avail_receiver: mpsc::Receiver<()>) where + CMH: ChannelMessageHandler + 'static + Send + Sync, + RMH: RoutingMessageHandler + 'static + Send + Sync, + L: Logger + 'static + ?Sized + Send + Sync, + UMH: CustomMessageHandler + 'static + Send + Sync { + // Create a waker to wake up poll_event_process, above + let (event_waker, event_receiver) = mpsc::channel(1); + tokio::spawn(Self::poll_event_process(Arc::clone(&peer_manager), event_receiver)); + // 8KB is nice and big but also should never cause any issues with stack overflowing. let mut buf = [0; 8192]; @@ -146,38 +160,24 @@ impl Connection { PeerDisconnected } let disconnect_type = loop { - macro_rules! shutdown_socket { - ($err: expr, $need_disconnect: expr) => { { - println!("Disconnecting peer due to {}!", $err); - break $need_disconnect; - } } - } - - macro_rules! prepare_read_write_call { - () => { { - let mut us_lock = us.lock().unwrap(); - if us_lock.rl_requested_disconnect { - shutdown_socket!("disconnect_socket() call from RL", Disconnect::CloseConnection); - } - us_lock.block_disconnect_socket = true; - } } - } - - let read_paused = us.lock().unwrap().read_paused; + let read_paused = { + let us_lock = us.lock().unwrap(); + if us_lock.rl_requested_disconnect { + break Disconnect::CloseConnection; + } + us_lock.read_paused + }; tokio::select! { v = write_avail_receiver.recv() => { assert!(v.is_some()); // We can't have dropped the sending end, its in the us Arc! - prepare_read_write_call!(); - if let Err(e) = peer_manager.write_buffer_space_avail(&mut our_descriptor) { - shutdown_socket!(e, Disconnect::CloseConnection); + if let Err(_) = peer_manager.write_buffer_space_avail(&mut our_descriptor) { + break Disconnect::CloseConnection; } - us.lock().unwrap().block_disconnect_socket = false; }, _ = read_wake_receiver.recv() => {}, read = reader.read(&mut buf), if !read_paused => match read { - Ok(0) => shutdown_socket!("Connection closed", Disconnect::PeerDisconnected), + Ok(0) => break Disconnect::PeerDisconnected, Ok(len) => { - prepare_read_write_call!(); let read_res = peer_manager.read_event(&mut our_descriptor, &buf[0..len]); let mut us_lock = us.lock().unwrap(); match read_res { @@ -186,14 +186,20 @@ impl Connection { us_lock.read_paused = true; } }, - Err(e) => shutdown_socket!(e, Disconnect::CloseConnection), + Err(_) => break Disconnect::CloseConnection, } - us_lock.block_disconnect_socket = false; }, - Err(e) => shutdown_socket!(e, Disconnect::PeerDisconnected), + Err(_) => break Disconnect::PeerDisconnected, }, } - peer_manager.process_events(); + let _ = event_waker.try_send(()); + + // At this point we've processed a message or two, and reset the ping timer for this + // peer, at least in the "are we still receiving messages" context, if we don't give up + // our timeslice to another task we may just spin on this peer, starving other peers + // and eventually disconnecting them for ping timeouts. Instead, we explicitly yield + // here. + tokio::task::yield_now().await; }; let writer_option = us.lock().unwrap().writer.take(); if let Some(mut writer) = writer_option { @@ -223,27 +229,43 @@ impl Connection { (reader, write_receiver, read_receiver, Arc::new(Mutex::new(Self { writer: Some(writer), write_avail, read_waker, read_paused: false, - block_disconnect_socket: false, rl_requested_disconnect: false, + rl_requested_disconnect: false, id: ID_COUNTER.fetch_add(1, Ordering::AcqRel) }))) } } +fn get_addr_from_stream(stream: &StdTcpStream) -> Option { + match stream.peer_addr() { + Ok(SocketAddr::V4(sockaddr)) => Some(NetAddress::IPv4 { + addr: sockaddr.ip().octets(), + port: sockaddr.port(), + }), + Ok(SocketAddr::V6(sockaddr)) => Some(NetAddress::IPv6 { + addr: sockaddr.ip().octets(), + port: sockaddr.port(), + }), + Err(_) => None, + } +} + /// Process incoming messages and feed outgoing messages on the provided socket generated by /// accepting an incoming connection. /// /// The returned future will complete when the peer is disconnected and associated handling /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do /// not need to poll the provided future in order to make progress. -pub fn setup_inbound(peer_manager: Arc, Arc, Arc>>, stream: StdTcpStream) -> impl std::future::Future where +pub fn setup_inbound(peer_manager: Arc, Arc, Arc, Arc>>, stream: StdTcpStream) -> impl std::future::Future where CMH: ChannelMessageHandler + 'static + Send + Sync, RMH: RoutingMessageHandler + 'static + Send + Sync, - L: Logger + 'static + ?Sized + Send + Sync { + L: Logger + 'static + ?Sized + Send + Sync, + UMH: CustomMessageHandler + 'static + Send + Sync { + let remote_addr = get_addr_from_stream(&stream); let (reader, write_receiver, read_receiver, us) = Connection::new(stream); #[cfg(debug_assertions)] let last_us = Arc::clone(&us); - let handle_opt = if let Ok(_) = peer_manager.new_inbound_connection(SocketDescriptor::new(us.clone())) { + let handle_opt = if let Ok(_) = peer_manager.new_inbound_connection(SocketDescriptor::new(us.clone()), remote_addr) { Some(tokio::spawn(Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver))) } else { // Note that we will skip socket_disconnected here, in accordance with the PeerManager @@ -275,15 +297,16 @@ pub fn setup_inbound(peer_manager: Arc(peer_manager: Arc, Arc, Arc>>, their_node_id: PublicKey, stream: StdTcpStream) -> impl std::future::Future where +pub fn setup_outbound(peer_manager: Arc, Arc, Arc, Arc>>, their_node_id: PublicKey, stream: StdTcpStream) -> impl std::future::Future where CMH: ChannelMessageHandler + 'static + Send + Sync, RMH: RoutingMessageHandler + 'static + Send + Sync, - L: Logger + 'static + ?Sized + Send + Sync { + L: Logger + 'static + ?Sized + Send + Sync, + UMH: CustomMessageHandler + 'static + Send + Sync { + let remote_addr = get_addr_from_stream(&stream); let (reader, mut write_receiver, read_receiver, us) = Connection::new(stream); #[cfg(debug_assertions)] let last_us = Arc::clone(&us); - - let handle_opt = if let Ok(initial_send) = peer_manager.new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone())) { + let handle_opt = if let Ok(initial_send) = peer_manager.new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone()), remote_addr) { Some(tokio::spawn(async move { // We should essentially always have enough room in a TCP socket buffer to send the // initial 10s of bytes. However, tokio running in single-threaded mode will always @@ -345,10 +368,11 @@ pub fn setup_outbound(peer_manager: Arc(peer_manager: Arc, Arc, Arc>>, their_node_id: PublicKey, addr: SocketAddr) -> Option> where +pub async fn connect_outbound(peer_manager: Arc, Arc, Arc, Arc>>, their_node_id: PublicKey, addr: SocketAddr) -> Option> where CMH: ChannelMessageHandler + 'static + Send + Sync, RMH: RoutingMessageHandler + 'static + Send + Sync, - L: Logger + 'static + ?Sized + Send + Sync { + L: Logger + 'static + ?Sized + Send + Sync, + UMH: CustomMessageHandler + 'static + Send + Sync { if let Ok(Ok(stream)) = time::timeout(Duration::from_secs(10), async { TcpStream::connect(&addr).await.map(|s| s.into_std().unwrap()) }).await { Some(setup_outbound(peer_manager, their_node_id, stream)) } else { None } @@ -443,6 +467,9 @@ impl peer_handler::SocketDescriptor for SocketDescriptor { // pause read given we're now waiting on the remote end to ACK (and in // accordance with the send_data() docs). us.read_paused = true; + // Further, to avoid any current pending read causing a `read_event` call, wake + // up the read_waker and restart its loop. + let _ = us.read_waker.try_send(()); return written_len; }, } @@ -450,18 +477,10 @@ impl peer_handler::SocketDescriptor for SocketDescriptor { } fn disconnect_socket(&mut self) { - { - let mut us = self.conn.lock().unwrap(); - us.rl_requested_disconnect = true; - us.read_paused = true; - // Wake up the sending thread, assuming it is still alive - let _ = us.write_avail.try_send(()); - // Happy-path return: - if !us.block_disconnect_socket { return; } - } - while self.conn.lock().unwrap().block_disconnect_socket { - thread::yield_now(); - } + let mut us = self.conn.lock().unwrap(); + us.rl_requested_disconnect = true; + // Wake up the sending thread, assuming it is still alive + let _ = us.write_avail.try_send(()); } } impl Clone for SocketDescriptor { @@ -517,10 +536,9 @@ mod tests { fn handle_node_announcement(&self, _msg: &NodeAnnouncement) -> Result { Ok(false) } fn handle_channel_announcement(&self, _msg: &ChannelAnnouncement) -> Result { Ok(false) } fn handle_channel_update(&self, _msg: &ChannelUpdate) -> Result { Ok(false) } - fn handle_htlc_fail_channel_update(&self, _update: &HTLCFailChannelUpdate) { } fn get_next_channel_announcements(&self, _starting_point: u64, _batch_amount: u8) -> Vec<(ChannelAnnouncement, Option, Option)> { Vec::new() } fn get_next_node_announcements(&self, _starting_point: Option<&PublicKey>, _batch_amount: u8) -> Vec { Vec::new() } - fn sync_routing_table(&self, _their_node_id: &PublicKey, _init_msg: &Init) { } + fn peer_connected(&self, _their_node_id: &PublicKey, _init_msg: &Init) { } fn handle_reply_channel_range(&self, _their_node_id: &PublicKey, _msg: ReplyChannelRange) -> Result<(), LightningError> { Ok(()) } fn handle_reply_short_channel_ids_end(&self, _their_node_id: &PublicKey, _msg: ReplyShortChannelIdsEnd) -> Result<(), LightningError> { Ok(()) } fn handle_query_channel_range(&self, _their_node_id: &PublicKey, _msg: QueryChannelRange) -> Result<(), LightningError> { Ok(()) } @@ -531,7 +549,7 @@ mod tests { fn handle_accept_channel(&self, _their_node_id: &PublicKey, _their_features: InitFeatures, _msg: &AcceptChannel) {} fn handle_funding_created(&self, _their_node_id: &PublicKey, _msg: &FundingCreated) {} fn handle_funding_signed(&self, _their_node_id: &PublicKey, _msg: &FundingSigned) {} - fn handle_funding_locked(&self, _their_node_id: &PublicKey, _msg: &FundingLocked) {} + fn handle_channel_ready(&self, _their_node_id: &PublicKey, _msg: &ChannelReady) {} fn handle_shutdown(&self, _their_node_id: &PublicKey, _their_features: &InitFeatures, _msg: &Shutdown) {} fn handle_closing_signed(&self, _their_node_id: &PublicKey, _msg: &ClosingSigned) {} fn handle_update_add_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateAddHTLC) {} @@ -565,6 +583,22 @@ mod tests { } } + fn make_tcp_connection() -> (std::net::TcpStream, std::net::TcpStream) { + if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") { + (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0) + } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:19735") { + (std::net::TcpStream::connect("127.0.0.1:19735").unwrap(), listener.accept().unwrap().0) + } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9997") { + (std::net::TcpStream::connect("127.0.0.1:9997").unwrap(), listener.accept().unwrap().0) + } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9998") { + (std::net::TcpStream::connect("127.0.0.1:9998").unwrap(), listener.accept().unwrap().0) + } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") { + (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0) + } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") { + (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0) + } else { panic!("Failed to bind to v4 localhost on common ports"); } + } + async fn do_basic_connection_test() { let secp_ctx = Secp256k1::new(); let a_key = SecretKey::from_slice(&[1; 32]).unwrap(); @@ -584,7 +618,7 @@ mod tests { let a_manager = Arc::new(PeerManager::new(MessageHandler { chan_handler: Arc::clone(&a_handler), route_handler: Arc::clone(&a_handler), - }, a_key.clone(), &[1; 32], Arc::new(TestLogger()))); + }, a_key.clone(), &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}))); let (b_connected_sender, mut b_connected) = mpsc::channel(1); let (b_disconnected_sender, mut b_disconnected) = mpsc::channel(1); @@ -598,19 +632,13 @@ mod tests { let b_manager = Arc::new(PeerManager::new(MessageHandler { chan_handler: Arc::clone(&b_handler), route_handler: Arc::clone(&b_handler), - }, b_key.clone(), &[2; 32], Arc::new(TestLogger()))); + }, b_key.clone(), &[2; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}))); // We bind on localhost, hoping the environment is properly configured with a local // address. This may not always be the case in containers and the like, so if this test is // failing for you check that you have a loopback interface and it is configured with // 127.0.0.1. - let (conn_a, conn_b) = if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") { - (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0) - } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") { - (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0) - } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") { - (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0) - } else { panic!("Failed to bind to v4 localhost on common ports"); }; + let (conn_a, conn_b) = make_tcp_connection(); let fut_a = super::setup_outbound(Arc::clone(&a_manager), b_pub, conn_a); let fut_b = super::setup_inbound(b_manager, conn_b); @@ -638,8 +666,53 @@ mod tests { async fn basic_threaded_connection_test() { do_basic_connection_test().await; } + #[tokio::test] async fn basic_unthreaded_connection_test() { do_basic_connection_test().await; } + + async fn race_disconnect_accept() { + // Previously, if we handed an already-disconnected socket to `setup_inbound` we'd panic. + // This attempts to find other similar races by opening connections and shutting them down + // while connecting. Sadly in testing this did *not* reproduce the previous issue. + let secp_ctx = Secp256k1::new(); + let a_key = SecretKey::from_slice(&[1; 32]).unwrap(); + let b_key = SecretKey::from_slice(&[2; 32]).unwrap(); + let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key); + + let a_manager = Arc::new(PeerManager::new(MessageHandler { + chan_handler: Arc::new(lightning::ln::peer_handler::ErroringMessageHandler::new()), + route_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}), + }, a_key, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}))); + + // Make two connections, one for an inbound and one for an outbound connection + let conn_a = { + let (conn_a, _) = make_tcp_connection(); + conn_a + }; + let conn_b = { + let (_, conn_b) = make_tcp_connection(); + conn_b + }; + + // Call connection setup inside new tokio tasks. + let manager_reference = Arc::clone(&a_manager); + tokio::spawn(async move { + super::setup_inbound(manager_reference, conn_a).await + }); + tokio::spawn(async move { + super::setup_outbound(a_manager, b_pub, conn_b).await + }); + } + + #[tokio::test(flavor = "multi_thread")] + async fn threaded_race_disconnect_accept() { + race_disconnect_accept().await; + } + + #[tokio::test] + async fn unthreaded_race_disconnect_accept() { + race_disconnect_accept().await; + } }