_ Git - rust-lightning/blob - lightning-net-tokio/src/lib.rs

   1 // This file is Copyright its original authors, visible in version control
   2 // history.
   3 //
   4 // This file is licensed under the Apache License, Version 2.0 <LICENSE-APACHE
   5 // or http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
   6 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your option.
   7 // You may not use this file except in accordance with one or both of these
   8 // licenses.
   9
  10 //! A socket handling library for those running in Tokio environments who wish to use
  11 //! rust-lightning with native TcpStreams.
  12 //!
  13 //! Designed to be as simple as possible, the high-level usage is almost as simple as "hand over a
  14 //! TcpStream and a reference to a PeerManager and the rest is handled", except for the
  15 //! [Event](../lightning/util/events/enum.Event.html) handling mechanism; see example below.
  16 //!
  17 //! The PeerHandler, due to the fire-and-forget nature of this logic, must be an Arc, and must use
  18 //! the SocketDescriptor provided here as the PeerHandler's SocketDescriptor.
  19 //!
  20 //! Three methods are exposed to register a new connection for handling in tokio::spawn calls; see
  21 //! their individual docs for details.
  22 //!
  23 //! # Example
  24 //! ```
  25 //! use std::net::TcpStream;
  26 //! use bitcoin::secp256k1::PublicKey;
  27 //! use lightning::util::events::{Event, EventHandler, EventsProvider};
  28 //! use std::net::SocketAddr;
  29 //! use std::sync::Arc;
  30 //!
  31 //! // Define concrete types for our high-level objects:
  32 //! type TxBroadcaster = dyn lightning::chain::chaininterface::BroadcasterInterface + Send + Sync;
  33 //! type FeeEstimator = dyn lightning::chain::chaininterface::FeeEstimator + Send + Sync;
  34 //! type Logger = dyn lightning::util::logger::Logger + Send + Sync;
  35 //! type NodeSigner = dyn lightning::chain::keysinterface::NodeSigner + Send + Sync;
  36 //! type UtxoLookup = dyn lightning::routing::utxo::UtxoLookup + Send + Sync;
  37 //! type ChainFilter = dyn lightning::chain::Filter + Send + Sync;
  38 //! type DataPersister = dyn lightning::chain::chainmonitor::Persist<lightning::chain::keysinterface::InMemorySigner> + Send + Sync;
  39 //! type ChainMonitor = lightning::chain::chainmonitor::ChainMonitor<lightning::chain::keysinterface::InMemorySigner, Arc<ChainFilter>, Arc<TxBroadcaster>, Arc<FeeEstimator>, Arc<Logger>, Arc<DataPersister>>;
  40 //! type ChannelManager = Arc<lightning::ln::channelmanager::SimpleArcChannelManager<ChainMonitor, TxBroadcaster, FeeEstimator, Logger>>;
  41 //! type PeerManager = Arc<lightning::ln::peer_handler::SimpleArcPeerManager<lightning_net_tokio::SocketDescriptor, ChainMonitor, TxBroadcaster, FeeEstimator, UtxoLookup, Logger>>;
  42 //!
  43 //! // Connect to node with pubkey their_node_id at addr:
  44 //! async fn connect_to_node(peer_manager: PeerManager, chain_monitor: Arc<ChainMonitor>, channel_manager: ChannelManager, their_node_id: PublicKey, addr: SocketAddr) {
  45 //!     lightning_net_tokio::connect_outbound(peer_manager, their_node_id, addr).await;
  46 //!     loop {
  47 //!             let event_handler = |event: Event| {
  48 //!                     // Handle the event!
  49 //!             };
  50 //!             channel_manager.await_persistable_update();
  51 //!             channel_manager.process_pending_events(&event_handler);
  52 //!             chain_monitor.process_pending_events(&event_handler);
  53 //!     }
  54 //! }
  55 //!
  56 //! // Begin reading from a newly accepted socket and talk to the peer:
  57 //! async fn accept_socket(peer_manager: PeerManager, chain_monitor: Arc<ChainMonitor>, channel_manager: ChannelManager, socket: TcpStream) {
  58 //!     lightning_net_tokio::setup_inbound(peer_manager, socket);
  59 //!     loop {
  60 //!             let event_handler = |event: Event| {
  61 //!                     // Handle the event!
  62 //!             };
  63 //!             channel_manager.await_persistable_update();
  64 //!             channel_manager.process_pending_events(&event_handler);
  65 //!             chain_monitor.process_pending_events(&event_handler);
  66 //!     }
  67 //! }
  68 //! ```
  69
  70 // Prefix these with `rustdoc::` when we update our MSRV to be >= 1.52 to remove warnings.
  71 #![deny(broken_intra_doc_links)]
  72 #![deny(private_intra_doc_links)]
  73
  74 #![deny(missing_docs)]
  75 #![cfg_attr(docsrs, feature(doc_auto_cfg))]
  76
  77 use bitcoin::secp256k1::PublicKey;
  78
  79 use tokio::net::TcpStream;
  80 use tokio::{io, time};
  81 use tokio::sync::mpsc;
  82 use tokio::io::{AsyncReadExt, AsyncWrite, AsyncWriteExt};
  83
  84 use lightning::chain::keysinterface::NodeSigner;
  85 use lightning::ln::peer_handler;
  86 use lightning::ln::peer_handler::SocketDescriptor as LnSocketTrait;
  87 use lightning::ln::peer_handler::CustomMessageHandler;
  88 use lightning::ln::msgs::{ChannelMessageHandler, NetAddress, OnionMessageHandler, RoutingMessageHandler};
  89 use lightning::util::logger::Logger;
  90
  91 use std::ops::Deref;
  92 use std::task;
  93 use std::net::SocketAddr;
  94 use std::net::TcpStream as StdTcpStream;
  95 use std::sync::{Arc, Mutex};
  96 use std::sync::atomic::{AtomicU64, Ordering};
  97 use std::time::Duration;
  98 use std::hash::Hash;
  99
 100 static ID_COUNTER: AtomicU64 = AtomicU64::new(0);
 101
 102 /// Connection contains all our internal state for a connection - we hold a reference to the
 103 /// Connection object (in an Arc<Mutex<>>) in each SocketDescriptor we create as well as in the
 104 /// read future (which is returned by schedule_read).
 105 struct Connection {
 106         writer: Option<io::WriteHalf<TcpStream>>,
 107         // Because our PeerManager is templated by user-provided types, and we can't (as far as I can
 108         // tell) have a const RawWakerVTable built out of templated functions, we need some indirection
 109         // between being woken up with write-ready and calling PeerManager::write_buffer_space_avail.
 110         // This provides that indirection, with a Sender which gets handed to the PeerManager Arc on
 111         // the schedule_read stack.
 112         //
 113         // An alternative (likely more effecient) approach would involve creating a RawWakerVTable at
 114         // runtime with functions templated by the Arc<PeerManager> type, calling
 115         // write_buffer_space_avail directly from tokio's write wake, however doing so would require
 116         // more unsafe voodo than I really feel like writing.
 117         write_avail: mpsc::Sender<()>,
 118         // When we are told by rust-lightning to pause read (because we have writes backing up), we do
 119         // so by setting read_paused. At that point, the read task will stop reading bytes from the
 120         // socket. To wake it up (without otherwise changing its state, we can push a value into this
 121         // Sender.
 122         read_waker: mpsc::Sender<()>,
 123         read_paused: bool,
 124         rl_requested_disconnect: bool,
 125         id: u64,
 126 }
 127 impl Connection {
 128         async fn poll_event_process<PM, CMH, RMH, OMH, L, UMH, NS>(
 129                 peer_manager: PM,
 130                 mut event_receiver: mpsc::Receiver<()>,
 131         ) where
 132                         PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH, NS>> + 'static + Send + Sync,
 133                         CMH: Deref + 'static + Send + Sync,
 134                         RMH: Deref + 'static + Send + Sync,
 135                         OMH: Deref + 'static + Send + Sync,
 136                         L: Deref + 'static + Send + Sync,
 137                         UMH: Deref + 'static + Send + Sync,
 138                         NS: Deref + 'static + Send + Sync,
 139                         CMH::Target: ChannelMessageHandler + Send + Sync,
 140                         RMH::Target: RoutingMessageHandler + Send + Sync,
 141                         OMH::Target: OnionMessageHandler + Send + Sync,
 142                         L::Target: Logger + Send + Sync,
 143                         UMH::Target: CustomMessageHandler + Send + Sync,
 144                         NS::Target: NodeSigner + Send + Sync,
 145         {
 146                 loop {
 147                         if event_receiver.recv().await.is_none() {
 148                                 return;
 149                         }
 150                         peer_manager.process_events();
 151                 }
 152         }
 153
 154         async fn schedule_read<PM, CMH, RMH, OMH, L, UMH, NS>(
 155                 peer_manager: PM,
 156                 us: Arc<Mutex<Self>>,
 157                 mut reader: io::ReadHalf<TcpStream>,
 158                 mut read_wake_receiver: mpsc::Receiver<()>,
 159                 mut write_avail_receiver: mpsc::Receiver<()>,
 160         ) where
 161                         PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH, NS>> + 'static + Send + Sync + Clone,
 162                         CMH: Deref + 'static + Send + Sync,
 163                         RMH: Deref + 'static + Send + Sync,
 164                         OMH: Deref + 'static + Send + Sync,
 165                         L: Deref + 'static + Send + Sync,
 166                         UMH: Deref + 'static + Send + Sync,
 167                         NS: Deref + 'static + Send + Sync,
 168                         CMH::Target: ChannelMessageHandler + 'static + Send + Sync,
 169                         RMH::Target: RoutingMessageHandler + 'static + Send + Sync,
 170                         OMH::Target: OnionMessageHandler + 'static + Send + Sync,
 171                         L::Target: Logger + 'static + Send + Sync,
 172                         UMH::Target: CustomMessageHandler + 'static + Send + Sync,
 173                         NS::Target: NodeSigner + 'static + Send + Sync,
 174                 {
 175                 // Create a waker to wake up poll_event_process, above
 176                 let (event_waker, event_receiver) = mpsc::channel(1);
 177                 tokio::spawn(Self::poll_event_process(peer_manager.clone(), event_receiver));
 178
 179                 // 4KiB is nice and big without handling too many messages all at once, giving other peers
 180                 // a chance to do some work.
 181                 let mut buf = [0; 4096];
 182
 183                 let mut our_descriptor = SocketDescriptor::new(us.clone());
 184                 // An enum describing why we did/are disconnecting:
 185                 enum Disconnect {
 186                         // Rust-Lightning told us to disconnect, either by returning an Err or by calling
 187                         // SocketDescriptor::disconnect_socket.
 188                         // In this case, we do not call peer_manager.socket_disconnected() as Rust-Lightning
 189                         // already knows we're disconnected.
 190                         CloseConnection,
 191                         // The connection was disconnected for some other reason, ie because the socket was
 192                         // closed.
 193                         // In this case, we do need to call peer_manager.socket_disconnected() to inform
 194                         // Rust-Lightning that the socket is gone.
 195                         PeerDisconnected
 196                 }
 197                 let disconnect_type = loop {
 198                         let read_paused = {
 199                                 let us_lock = us.lock().unwrap();
 200                                 if us_lock.rl_requested_disconnect {
 201                                         break Disconnect::CloseConnection;
 202                                 }
 203                                 us_lock.read_paused
 204                         };
 205                         tokio::select! {
 206                                 v = write_avail_receiver.recv() => {
 207                                         assert!(v.is_some()); // We can't have dropped the sending end, its in the us Arc!
 208                                         if let Err(_) = peer_manager.write_buffer_space_avail(&mut our_descriptor) {
 209                                                 break Disconnect::CloseConnection;
 210                                         }
 211                                 },
 212                                 _ = read_wake_receiver.recv() => {},
 213                                 read = reader.read(&mut buf), if !read_paused => match read {
 214                                         Ok(0) => break Disconnect::PeerDisconnected,
 215                                         Ok(len) => {
 216                                                 let read_res = peer_manager.read_event(&mut our_descriptor, &buf[0..len]);
 217                                                 let mut us_lock = us.lock().unwrap();
 218                                                 match read_res {
 219                                                         Ok(pause_read) => {
 220                                                                 if pause_read {
 221                                                                         us_lock.read_paused = true;
 222                                                                 }
 223                                                         },
 224                                                         Err(_) => break Disconnect::CloseConnection,
 225                                                 }
 226                                         },
 227                                         Err(_) => break Disconnect::PeerDisconnected,
 228                                 },
 229                         }
 230                         let _ = event_waker.try_send(());
 231
 232                         // At this point we've processed a message or two, and reset the ping timer for this
 233                         // peer, at least in the "are we still receiving messages" context, if we don't give up
 234                         // our timeslice to another task we may just spin on this peer, starving other peers
 235                         // and eventually disconnecting them for ping timeouts. Instead, we explicitly yield
 236                         // here.
 237                         tokio::task::yield_now().await;
 238                 };
 239                 let writer_option = us.lock().unwrap().writer.take();
 240                 if let Some(mut writer) = writer_option {
 241                         // If the socket is already closed, shutdown() will fail, so just ignore it.
 242                         let _ = writer.shutdown().await;
 243                 }
 244                 if let Disconnect::PeerDisconnected = disconnect_type {
 245                         peer_manager.socket_disconnected(&our_descriptor);
 246                         peer_manager.process_events();
 247                 }
 248         }
 249
 250         fn new(stream: StdTcpStream) -> (io::ReadHalf<TcpStream>, mpsc::Receiver<()>, mpsc::Receiver<()>, Arc<Mutex<Self>>) {
 251                 // We only ever need a channel of depth 1 here: if we returned a non-full write to the
 252                 // PeerManager, we will eventually get notified that there is room in the socket to write
 253                 // new bytes, which will generate an event. That event will be popped off the queue before
 254                 // we call write_buffer_space_avail, ensuring that we have room to push a new () if, during
 255                 // the write_buffer_space_avail() call, send_data() returns a non-full write.
 256                 let (write_avail, write_receiver) = mpsc::channel(1);
 257                 // Similarly here - our only goal is to make sure the reader wakes up at some point after
 258                 // we shove a value into the channel which comes after we've reset the read_paused bool to
 259                 // false.
 260                 let (read_waker, read_receiver) = mpsc::channel(1);
 261                 stream.set_nonblocking(true).unwrap();
 262                 let (reader, writer) = io::split(TcpStream::from_std(stream).unwrap());
 263
 264                 (reader, write_receiver, read_receiver,
 265                 Arc::new(Mutex::new(Self {
 266                         writer: Some(writer), write_avail, read_waker, read_paused: false,
 267                         rl_requested_disconnect: false,
 268                         id: ID_COUNTER.fetch_add(1, Ordering::AcqRel)
 269                 })))
 270         }
 271 }
 272
 273 fn get_addr_from_stream(stream: &StdTcpStream) -> Option<NetAddress> {
 274         match stream.peer_addr() {
 275                 Ok(SocketAddr::V4(sockaddr)) => Some(NetAddress::IPv4 {
 276                         addr: sockaddr.ip().octets(),
 277                         port: sockaddr.port(),
 278                 }),
 279                 Ok(SocketAddr::V6(sockaddr)) => Some(NetAddress::IPv6 {
 280                         addr: sockaddr.ip().octets(),
 281                         port: sockaddr.port(),
 282                 }),
 283                 Err(_) => None,
 284         }
 285 }
 286
 287 /// Process incoming messages and feed outgoing messages on the provided socket generated by
 288 /// accepting an incoming connection.
 289 ///
 290 /// The returned future will complete when the peer is disconnected and associated handling
 291 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
 292 /// not need to poll the provided future in order to make progress.
 293 pub fn setup_inbound<PM, CMH, RMH, OMH, L, UMH, NS>(
 294         peer_manager: PM,
 295         stream: StdTcpStream,
 296 ) -> impl std::future::Future<Output=()> where
 297                 PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH, NS>> + 'static + Send + Sync + Clone,
 298                 CMH: Deref + 'static + Send + Sync,
 299                 RMH: Deref + 'static + Send + Sync,
 300                 OMH: Deref + 'static + Send + Sync,
 301                 L: Deref + 'static + Send + Sync,
 302                 UMH: Deref + 'static + Send + Sync,
 303                 NS: Deref + 'static + Send + Sync,
 304                 CMH::Target: ChannelMessageHandler + Send + Sync,
 305                 RMH::Target: RoutingMessageHandler + Send + Sync,
 306                 OMH::Target: OnionMessageHandler + Send + Sync,
 307                 L::Target: Logger + Send + Sync,
 308                 UMH::Target: CustomMessageHandler + Send + Sync,
 309                 NS::Target: NodeSigner + Send + Sync,
 310 {
 311         let remote_addr = get_addr_from_stream(&stream);
 312         let (reader, write_receiver, read_receiver, us) = Connection::new(stream);
 313         #[cfg(test)]
 314         let last_us = Arc::clone(&us);
 315
 316         let handle_opt = if let Ok(_) = peer_manager.new_inbound_connection(SocketDescriptor::new(us.clone()), remote_addr) {
 317                 Some(tokio::spawn(Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver)))
 318         } else {
 319                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
 320                 // requirements.
 321                 None
 322         };
 323
 324         async move {
 325                 if let Some(handle) = handle_opt {
 326                         if let Err(e) = handle.await {
 327                                 assert!(e.is_cancelled());
 328                         } else {
 329                                 // This is certainly not guaranteed to always be true - the read loop may exit
 330                                 // while there are still pending write wakers that need to be woken up after the
 331                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
 332                                 // keep too many wakers around, this makes sense. The race should be rare (we do
 333                                 // some work after shutdown()) and an error would be a major memory leak.
 334                                 #[cfg(test)]
 335                                 debug_assert!(Arc::try_unwrap(last_us).is_ok());
 336                         }
 337                 }
 338         }
 339 }
 340
 341 /// Process incoming messages and feed outgoing messages on the provided socket generated by
 342 /// making an outbound connection which is expected to be accepted by a peer with the given
 343 /// public key. The relevant processing is set to run free (via tokio::spawn).
 344 ///
 345 /// The returned future will complete when the peer is disconnected and associated handling
 346 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
 347 /// not need to poll the provided future in order to make progress.
 348 pub fn setup_outbound<PM, CMH, RMH, OMH, L, UMH, NS>(
 349         peer_manager: PM,
 350         their_node_id: PublicKey,
 351         stream: StdTcpStream,
 352 ) -> impl std::future::Future<Output=()> where
 353                 PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH, NS>> + 'static + Send + Sync + Clone,
 354                 CMH: Deref + 'static + Send + Sync,
 355                 RMH: Deref + 'static + Send + Sync,
 356                 OMH: Deref + 'static + Send + Sync,
 357                 L: Deref + 'static + Send + Sync,
 358                 UMH: Deref + 'static + Send + Sync,
 359                 NS: Deref + 'static + Send + Sync,
 360                 CMH::Target: ChannelMessageHandler + Send + Sync,
 361                 RMH::Target: RoutingMessageHandler + Send + Sync,
 362                 OMH::Target: OnionMessageHandler + Send + Sync,
 363                 L::Target: Logger + Send + Sync,
 364                 UMH::Target: CustomMessageHandler + Send + Sync,
 365                 NS::Target: NodeSigner + Send + Sync,
 366 {
 367         let remote_addr = get_addr_from_stream(&stream);
 368         let (reader, mut write_receiver, read_receiver, us) = Connection::new(stream);
 369         #[cfg(test)]
 370         let last_us = Arc::clone(&us);
 371         let handle_opt = if let Ok(initial_send) = peer_manager.new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone()), remote_addr) {
 372                 Some(tokio::spawn(async move {
 373                         // We should essentially always have enough room in a TCP socket buffer to send the
 374                         // initial 10s of bytes. However, tokio running in single-threaded mode will always
 375                         // fail writes and wake us back up later to write. Thus, we handle a single
 376                         // std::task::Poll::Pending but still expect to write the full set of bytes at once
 377                         // and use a relatively tight timeout.
 378                         if let Ok(Ok(())) = tokio::time::timeout(Duration::from_millis(100), async {
 379                                 loop {
 380                                         match SocketDescriptor::new(us.clone()).send_data(&initial_send, true) {
 381                                                 v if v == initial_send.len() => break Ok(()),
 382                                                 0 => {
 383                                                         write_receiver.recv().await;
 384                                                         // In theory we could check for if we've been instructed to disconnect
 385                                                         // the peer here, but its OK to just skip it - we'll check for it in
 386                                                         // schedule_read prior to any relevant calls into RL.
 387                                                 },
 388                                                 _ => {
 389                                                         eprintln!("Failed to write first full message to socket!");
 390                                                         peer_manager.socket_disconnected(&SocketDescriptor::new(Arc::clone(&us)));
 391                                                         break Err(());
 392                                                 }
 393                                         }
 394                                 }
 395                         }).await {
 396                                 Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver).await;
 397                         }
 398                 }))
 399         } else {
 400                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
 401                 // requirements.
 402                 None
 403         };
 404
 405         async move {
 406                 if let Some(handle) = handle_opt {
 407                         if let Err(e) = handle.await {
 408                                 assert!(e.is_cancelled());
 409                         } else {
 410                                 // This is certainly not guaranteed to always be true - the read loop may exit
 411                                 // while there are still pending write wakers that need to be woken up after the
 412                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
 413                                 // keep too many wakers around, this makes sense. The race should be rare (we do
 414                                 // some work after shutdown()) and an error would be a major memory leak.
 415                                 #[cfg(test)]
 416                                 debug_assert!(Arc::try_unwrap(last_us).is_ok());
 417                         }
 418                 }
 419         }
 420 }
 421
 422 /// Process incoming messages and feed outgoing messages on a new connection made to the given
 423 /// socket address which is expected to be accepted by a peer with the given public key (by
 424 /// scheduling futures with tokio::spawn).
 425 ///
 426 /// Shorthand for TcpStream::connect(addr) with a timeout followed by setup_outbound().
 427 ///
 428 /// Returns a future (as the fn is async) which needs to be polled to complete the connection and
 429 /// connection setup. That future then returns a future which will complete when the peer is
 430 /// disconnected and associated handling futures are freed, though, because all processing in said
 431 /// futures are spawned with tokio::spawn, you do not need to poll the second future in order to
 432 /// make progress.
 433 pub async fn connect_outbound<PM, CMH, RMH, OMH, L, UMH, NS>(
 434         peer_manager: PM,
 435         their_node_id: PublicKey,
 436         addr: SocketAddr,
 437 ) -> Option<impl std::future::Future<Output=()>> where
 438                 PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH, NS>> + 'static + Send + Sync + Clone,
 439                 CMH: Deref + 'static + Send + Sync,
 440                 RMH: Deref + 'static + Send + Sync,
 441                 OMH: Deref + 'static + Send + Sync,
 442                 L: Deref + 'static + Send + Sync,
 443                 UMH: Deref + 'static + Send + Sync,
 444                 NS: Deref + 'static + Send + Sync,
 445                 CMH::Target: ChannelMessageHandler + Send + Sync,
 446                 RMH::Target: RoutingMessageHandler + Send + Sync,
 447                 OMH::Target: OnionMessageHandler + Send + Sync,
 448                 L::Target: Logger + Send + Sync,
 449                 UMH::Target: CustomMessageHandler + Send + Sync,
 450                 NS::Target: NodeSigner + Send + Sync,
 451 {
 452         if let Ok(Ok(stream)) = time::timeout(Duration::from_secs(10), async { TcpStream::connect(&addr).await.map(|s| s.into_std().unwrap()) }).await {
 453                 Some(setup_outbound(peer_manager, their_node_id, stream))
 454         } else { None }
 455 }
 456
 457 const SOCK_WAKER_VTABLE: task::RawWakerVTable =
 458         task::RawWakerVTable::new(clone_socket_waker, wake_socket_waker, wake_socket_waker_by_ref, drop_socket_waker);
 459
 460 fn clone_socket_waker(orig_ptr: *const ()) -> task::RawWaker {
 461         write_avail_to_waker(orig_ptr as *const mpsc::Sender<()>)
 462 }
 463 // When waking, an error should be fine. Most likely we got two send_datas in a row, both of which
 464 // failed to fully write, but we only need to call write_buffer_space_avail() once. Otherwise, the
 465 // sending thread may have already gone away due to a socket close, in which case there's nothing
 466 // to wake up anyway.
 467 fn wake_socket_waker(orig_ptr: *const ()) {
 468         let sender = unsafe { &mut *(orig_ptr as *mut mpsc::Sender<()>) };
 469         let _ = sender.try_send(());
 470         drop_socket_waker(orig_ptr);
 471 }
 472 fn wake_socket_waker_by_ref(orig_ptr: *const ()) {
 473         let sender_ptr = orig_ptr as *const mpsc::Sender<()>;
 474         let sender = unsafe { (*sender_ptr).clone() };
 475         let _ = sender.try_send(());
 476 }
 477 fn drop_socket_waker(orig_ptr: *const ()) {
 478         let _orig_box = unsafe { Box::from_raw(orig_ptr as *mut mpsc::Sender<()>) };
 479         // _orig_box is now dropped
 480 }
 481 fn write_avail_to_waker(sender: *const mpsc::Sender<()>) -> task::RawWaker {
 482         let new_box = Box::leak(Box::new(unsafe { (*sender).clone() }));
 483         let new_ptr = new_box as *const mpsc::Sender<()>;
 484         task::RawWaker::new(new_ptr as *const (), &SOCK_WAKER_VTABLE)
 485 }
 486
 487 /// The SocketDescriptor used to refer to sockets by a PeerHandler. This is pub only as it is a
 488 /// type in the template of PeerHandler.
 489 pub struct SocketDescriptor {
 490         conn: Arc<Mutex<Connection>>,
 491         id: u64,
 492 }
 493 impl SocketDescriptor {
 494         fn new(conn: Arc<Mutex<Connection>>) -> Self {
 495                 let id = conn.lock().unwrap().id;
 496                 Self { conn, id }
 497         }
 498 }
 499 impl peer_handler::SocketDescriptor for SocketDescriptor {
 500         fn send_data(&mut self, data: &[u8], resume_read: bool) -> usize {
 501                 // To send data, we take a lock on our Connection to access the WriteHalf of the TcpStream,
 502                 // writing to it if there's room in the kernel buffer, or otherwise create a new Waker with
 503                 // a SocketDescriptor in it which can wake up the write_avail Sender, waking up the
 504                 // processing future which will call write_buffer_space_avail and we'll end up back here.
 505                 let mut us = self.conn.lock().unwrap();
 506                 if us.writer.is_none() {
 507                         // The writer gets take()n when it is time to shut down, so just fast-return 0 here.
 508                         return 0;
 509                 }
 510
 511                 if resume_read && us.read_paused {
 512                         // The schedule_read future may go to lock up but end up getting woken up by there
 513                         // being more room in the write buffer, dropping the other end of this Sender
 514                         // before we get here, so we ignore any failures to wake it up.
 515                         us.read_paused = false;
 516                         let _ = us.read_waker.try_send(());
 517                 }
 518                 if data.is_empty() { return 0; }
 519                 let waker = unsafe { task::Waker::from_raw(write_avail_to_waker(&us.write_avail)) };
 520                 let mut ctx = task::Context::from_waker(&waker);
 521                 let mut written_len = 0;
 522                 loop {
 523                         match std::pin::Pin::new(us.writer.as_mut().unwrap()).poll_write(&mut ctx, &data[written_len..]) {
 524                                 task::Poll::Ready(Ok(res)) => {
 525                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
 526                                         // know how to handle it if it does (cause it should be a Poll::Pending
 527                                         // instead):
 528                                         assert_ne!(res, 0);
 529                                         written_len += res;
 530                                         if written_len == data.len() { return written_len; }
 531                                 },
 532                                 task::Poll::Ready(Err(e)) => {
 533                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
 534                                         // know how to handle it if it does (cause it should be a Poll::Pending
 535                                         // instead):
 536                                         assert_ne!(e.kind(), io::ErrorKind::WouldBlock);
 537                                         // Probably we've already been closed, just return what we have and let the
 538                                         // read thread handle closing logic.
 539                                         return written_len;
 540                                 },
 541                                 task::Poll::Pending => {
 542                                         // We're queued up for a write event now, but we need to make sure we also
 543                                         // pause read given we're now waiting on the remote end to ACK (and in
 544                                         // accordance with the send_data() docs).
 545                                         us.read_paused = true;
 546                                         // Further, to avoid any current pending read causing a `read_event` call, wake
 547                                         // up the read_waker and restart its loop.
 548                                         let _ = us.read_waker.try_send(());
 549                                         return written_len;
 550                                 },
 551                         }
 552                 }
 553         }
 554
 555         fn disconnect_socket(&mut self) {
 556                 let mut us = self.conn.lock().unwrap();
 557                 us.rl_requested_disconnect = true;
 558                 // Wake up the sending thread, assuming it is still alive
 559                 let _ = us.write_avail.try_send(());
 560         }
 561 }
 562 impl Clone for SocketDescriptor {
 563         fn clone(&self) -> Self {
 564                 Self {
 565                         conn: Arc::clone(&self.conn),
 566                         id: self.id,
 567                 }
 568         }
 569 }
 570 impl Eq for SocketDescriptor {}
 571 impl PartialEq for SocketDescriptor {
 572         fn eq(&self, o: &Self) -> bool {
 573                 self.id == o.id
 574         }
 575 }
 576 impl Hash for SocketDescriptor {
 577         fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
 578                 self.id.hash(state);
 579         }
 580 }
 581
 582 #[cfg(test)]
 583 mod tests {
 584         use lightning::ln::features::*;
 585         use lightning::ln::msgs::*;
 586         use lightning::ln::peer_handler::{MessageHandler, PeerManager};
 587         use lightning::ln::features::NodeFeatures;
 588         use lightning::routing::gossip::NodeId;
 589         use lightning::util::events::*;
 590         use lightning::util::test_utils::TestNodeSigner;
 591         use bitcoin::secp256k1::{Secp256k1, SecretKey, PublicKey};
 592
 593         use tokio::sync::mpsc;
 594
 595         use std::mem;
 596         use std::sync::atomic::{AtomicBool, Ordering};
 597         use std::sync::{Arc, Mutex};
 598         use std::time::Duration;
 599
 600         pub struct TestLogger();
 601         impl lightning::util::logger::Logger for TestLogger {
 602                 fn log(&self, record: &lightning::util::logger::Record) {
 603                         println!("{:<5} [{} : {}, {}] {}", record.level.to_string(), record.module_path, record.file, record.line, record.args);
 604                 }
 605         }
 606
 607         struct MsgHandler{
 608                 expected_pubkey: PublicKey,
 609                 pubkey_connected: mpsc::Sender<()>,
 610                 pubkey_disconnected: mpsc::Sender<()>,
 611                 disconnected_flag: AtomicBool,
 612                 msg_events: Mutex<Vec<MessageSendEvent>>,
 613         }
 614         impl RoutingMessageHandler for MsgHandler {
 615                 fn handle_node_announcement(&self, _msg: &NodeAnnouncement) -> Result<bool, LightningError> { Ok(false) }
 616                 fn handle_channel_announcement(&self, _msg: &ChannelAnnouncement) -> Result<bool, LightningError> { Ok(false) }
 617                 fn handle_channel_update(&self, _msg: &ChannelUpdate) -> Result<bool, LightningError> { Ok(false) }
 618                 fn get_next_channel_announcement(&self, _starting_point: u64) -> Option<(ChannelAnnouncement, Option<ChannelUpdate>, Option<ChannelUpdate>)> { None }
 619                 fn get_next_node_announcement(&self, _starting_point: Option<&NodeId>) -> Option<NodeAnnouncement> { None }
 620                 fn peer_connected(&self, _their_node_id: &PublicKey, _init_msg: &Init) -> Result<(), ()> { Ok(()) }
 621                 fn handle_reply_channel_range(&self, _their_node_id: &PublicKey, _msg: ReplyChannelRange) -> Result<(), LightningError> { Ok(()) }
 622                 fn handle_reply_short_channel_ids_end(&self, _their_node_id: &PublicKey, _msg: ReplyShortChannelIdsEnd) -> Result<(), LightningError> { Ok(()) }
 623                 fn handle_query_channel_range(&self, _their_node_id: &PublicKey, _msg: QueryChannelRange) -> Result<(), LightningError> { Ok(()) }
 624                 fn handle_query_short_channel_ids(&self, _their_node_id: &PublicKey, _msg: QueryShortChannelIds) -> Result<(), LightningError> { Ok(()) }
 625                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
 626                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
 627                 fn processing_queue_high(&self) -> bool { false }
 628         }
 629         impl ChannelMessageHandler for MsgHandler {
 630                 fn handle_open_channel(&self, _their_node_id: &PublicKey, _msg: &OpenChannel) {}
 631                 fn handle_accept_channel(&self, _their_node_id: &PublicKey, _msg: &AcceptChannel) {}
 632                 fn handle_funding_created(&self, _their_node_id: &PublicKey, _msg: &FundingCreated) {}
 633                 fn handle_funding_signed(&self, _their_node_id: &PublicKey, _msg: &FundingSigned) {}
 634                 fn handle_channel_ready(&self, _their_node_id: &PublicKey, _msg: &ChannelReady) {}
 635                 fn handle_shutdown(&self, _their_node_id: &PublicKey, _msg: &Shutdown) {}
 636                 fn handle_closing_signed(&self, _their_node_id: &PublicKey, _msg: &ClosingSigned) {}
 637                 fn handle_update_add_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateAddHTLC) {}
 638                 fn handle_update_fulfill_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFulfillHTLC) {}
 639                 fn handle_update_fail_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailHTLC) {}
 640                 fn handle_update_fail_malformed_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailMalformedHTLC) {}
 641                 fn handle_commitment_signed(&self, _their_node_id: &PublicKey, _msg: &CommitmentSigned) {}
 642                 fn handle_revoke_and_ack(&self, _their_node_id: &PublicKey, _msg: &RevokeAndACK) {}
 643                 fn handle_update_fee(&self, _their_node_id: &PublicKey, _msg: &UpdateFee) {}
 644                 fn handle_announcement_signatures(&self, _their_node_id: &PublicKey, _msg: &AnnouncementSignatures) {}
 645                 fn handle_channel_update(&self, _their_node_id: &PublicKey, _msg: &ChannelUpdate) {}
 646                 fn peer_disconnected(&self, their_node_id: &PublicKey) {
 647                         if *their_node_id == self.expected_pubkey {
 648                                 self.disconnected_flag.store(true, Ordering::SeqCst);
 649                                 self.pubkey_disconnected.clone().try_send(()).unwrap();
 650                         }
 651                 }
 652                 fn peer_connected(&self, their_node_id: &PublicKey, _init_msg: &Init) -> Result<(), ()> {
 653                         if *their_node_id == self.expected_pubkey {
 654                                 self.pubkey_connected.clone().try_send(()).unwrap();
 655                         }
 656                         Ok(())
 657                 }
 658                 fn handle_channel_reestablish(&self, _their_node_id: &PublicKey, _msg: &ChannelReestablish) {}
 659                 fn handle_error(&self, _their_node_id: &PublicKey, _msg: &ErrorMessage) {}
 660                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
 661                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
 662         }
 663         impl MessageSendEventsProvider for MsgHandler {
 664                 fn get_and_clear_pending_msg_events(&self) -> Vec<MessageSendEvent> {
 665                         let mut ret = Vec::new();
 666                         mem::swap(&mut *self.msg_events.lock().unwrap(), &mut ret);
 667                         ret
 668                 }
 669         }
 670
 671         fn make_tcp_connection() -> (std::net::TcpStream, std::net::TcpStream) {
 672                 if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") {
 673                         (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0)
 674                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:19735") {
 675                         (std::net::TcpStream::connect("127.0.0.1:19735").unwrap(), listener.accept().unwrap().0)
 676                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9997") {
 677                         (std::net::TcpStream::connect("127.0.0.1:9997").unwrap(), listener.accept().unwrap().0)
 678                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9998") {
 679                         (std::net::TcpStream::connect("127.0.0.1:9998").unwrap(), listener.accept().unwrap().0)
 680                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") {
 681                         (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0)
 682                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") {
 683                         (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0)
 684                 } else { panic!("Failed to bind to v4 localhost on common ports"); }
 685         }
 686
 687         async fn do_basic_connection_test() {
 688                 let secp_ctx = Secp256k1::new();
 689                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
 690                 let b_key = SecretKey::from_slice(&[1; 32]).unwrap();
 691                 let a_pub = PublicKey::from_secret_key(&secp_ctx, &a_key);
 692                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
 693
 694                 let (a_connected_sender, mut a_connected) = mpsc::channel(1);
 695                 let (a_disconnected_sender, mut a_disconnected) = mpsc::channel(1);
 696                 let a_handler = Arc::new(MsgHandler {
 697                         expected_pubkey: b_pub,
 698                         pubkey_connected: a_connected_sender,
 699                         pubkey_disconnected: a_disconnected_sender,
 700                         disconnected_flag: AtomicBool::new(false),
 701                         msg_events: Mutex::new(Vec::new()),
 702                 });
 703                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
 704                         chan_handler: Arc::clone(&a_handler),
 705                         route_handler: Arc::clone(&a_handler),
 706                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 707                 }, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}), Arc::new(TestNodeSigner::new(a_key))));
 708
 709                 let (b_connected_sender, mut b_connected) = mpsc::channel(1);
 710                 let (b_disconnected_sender, mut b_disconnected) = mpsc::channel(1);
 711                 let b_handler = Arc::new(MsgHandler {
 712                         expected_pubkey: a_pub,
 713                         pubkey_connected: b_connected_sender,
 714                         pubkey_disconnected: b_disconnected_sender,
 715                         disconnected_flag: AtomicBool::new(false),
 716                         msg_events: Mutex::new(Vec::new()),
 717                 });
 718                 let b_manager = Arc::new(PeerManager::new(MessageHandler {
 719                         chan_handler: Arc::clone(&b_handler),
 720                         route_handler: Arc::clone(&b_handler),
 721                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 722                 }, 0, &[2; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}), Arc::new(TestNodeSigner::new(b_key))));
 723
 724                 // We bind on localhost, hoping the environment is properly configured with a local
 725                 // address. This may not always be the case in containers and the like, so if this test is
 726                 // failing for you check that you have a loopback interface and it is configured with
 727                 // 127.0.0.1.
 728                 let (conn_a, conn_b) = make_tcp_connection();
 729
 730                 let fut_a = super::setup_outbound(Arc::clone(&a_manager), b_pub, conn_a);
 731                 let fut_b = super::setup_inbound(b_manager, conn_b);
 732
 733                 tokio::time::timeout(Duration::from_secs(10), a_connected.recv()).await.unwrap();
 734                 tokio::time::timeout(Duration::from_secs(1), b_connected.recv()).await.unwrap();
 735
 736                 a_handler.msg_events.lock().unwrap().push(MessageSendEvent::HandleError {
 737                         node_id: b_pub, action: ErrorAction::DisconnectPeer { msg: None }
 738                 });
 739                 assert!(!a_handler.disconnected_flag.load(Ordering::SeqCst));
 740                 assert!(!b_handler.disconnected_flag.load(Ordering::SeqCst));
 741
 742                 a_manager.process_events();
 743                 tokio::time::timeout(Duration::from_secs(10), a_disconnected.recv()).await.unwrap();
 744                 tokio::time::timeout(Duration::from_secs(1), b_disconnected.recv()).await.unwrap();
 745                 assert!(a_handler.disconnected_flag.load(Ordering::SeqCst));
 746                 assert!(b_handler.disconnected_flag.load(Ordering::SeqCst));
 747
 748                 fut_a.await;
 749                 fut_b.await;
 750         }
 751
 752         #[tokio::test(flavor = "multi_thread")]
 753         async fn basic_threaded_connection_test() {
 754                 do_basic_connection_test().await;
 755         }
 756
 757         #[tokio::test]
 758         async fn basic_unthreaded_connection_test() {
 759                 do_basic_connection_test().await;
 760         }
 761
 762         async fn race_disconnect_accept() {
 763                 // Previously, if we handed an already-disconnected socket to `setup_inbound` we'd panic.
 764                 // This attempts to find other similar races by opening connections and shutting them down
 765                 // while connecting. Sadly in testing this did *not* reproduce the previous issue.
 766                 let secp_ctx = Secp256k1::new();
 767                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
 768                 let b_key = SecretKey::from_slice(&[2; 32]).unwrap();
 769                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
 770
 771                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
 772                         chan_handler: Arc::new(lightning::ln::peer_handler::ErroringMessageHandler::new()),
 773                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 774                         route_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 775                 }, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}), Arc::new(TestNodeSigner::new(a_key))));
 776
 777                 // Make two connections, one for an inbound and one for an outbound connection
 778                 let conn_a = {
 779                         let (conn_a, _) = make_tcp_connection();
 780                         conn_a
 781                 };
 782                 let conn_b = {
 783                         let (_, conn_b) = make_tcp_connection();
 784                         conn_b
 785                 };
 786
 787                 // Call connection setup inside new tokio tasks.
 788                 let manager_reference = Arc::clone(&a_manager);
 789                 tokio::spawn(async move {
 790                         super::setup_inbound(manager_reference, conn_a).await
 791                 });
 792                 tokio::spawn(async move {
 793                         super::setup_outbound(a_manager, b_pub, conn_b).await
 794                 });
 795         }
 796
 797         #[tokio::test(flavor = "multi_thread")]
 798         async fn threaded_race_disconnect_accept() {
 799                 race_disconnect_accept().await;
 800         }
 801
 802         #[tokio::test]
 803         async fn unthreaded_race_disconnect_accept() {
 804                 race_disconnect_accept().await;
 805         }
 806 }