_ 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 ChainAccess = dyn lightning::chain::Access + Send + Sync;
  36 //! type ChainFilter = dyn lightning::chain::Filter + Send + Sync;
  37 //! type DataPersister = dyn lightning::chain::chainmonitor::Persist<lightning::chain::keysinterface::InMemorySigner> + Send + Sync;
  38 //! type ChainMonitor = lightning::chain::chainmonitor::ChainMonitor<lightning::chain::keysinterface::InMemorySigner, Arc<ChainFilter>, Arc<TxBroadcaster>, Arc<FeeEstimator>, Arc<Logger>, Arc<DataPersister>>;
  39 //! type ChannelManager = Arc<lightning::ln::channelmanager::SimpleArcChannelManager<ChainMonitor, TxBroadcaster, FeeEstimator, Logger>>;
  40 //! type PeerManager = Arc<lightning::ln::peer_handler::SimpleArcPeerManager<lightning_net_tokio::SocketDescriptor, ChainMonitor, TxBroadcaster, FeeEstimator, ChainAccess, Logger>>;
  41 //!
  42 //! // Connect to node with pubkey their_node_id at addr:
  43 //! async fn connect_to_node(peer_manager: PeerManager, chain_monitor: Arc<ChainMonitor>, channel_manager: ChannelManager, their_node_id: PublicKey, addr: SocketAddr) {
  44 //!     lightning_net_tokio::connect_outbound(peer_manager, their_node_id, addr).await;
  45 //!     loop {
  46 //!             let event_handler = |event: Event| {
  47 //!                     // Handle the event!
  48 //!             };
  49 //!             channel_manager.await_persistable_update();
  50 //!             channel_manager.process_pending_events(&event_handler);
  51 //!             chain_monitor.process_pending_events(&event_handler);
  52 //!     }
  53 //! }
  54 //!
  55 //! // Begin reading from a newly accepted socket and talk to the peer:
  56 //! async fn accept_socket(peer_manager: PeerManager, chain_monitor: Arc<ChainMonitor>, channel_manager: ChannelManager, socket: TcpStream) {
  57 //!     lightning_net_tokio::setup_inbound(peer_manager, socket);
  58 //!     loop {
  59 //!             let event_handler = |event: Event| {
  60 //!                     // Handle the event!
  61 //!             };
  62 //!             channel_manager.await_persistable_update();
  63 //!             channel_manager.process_pending_events(&event_handler);
  64 //!             chain_monitor.process_pending_events(&event_handler);
  65 //!     }
  66 //! }
  67 //! ```
  68
  69 // Prefix these with `rustdoc::` when we update our MSRV to be >= 1.52 to remove warnings.
  70 #![deny(broken_intra_doc_links)]
  71 #![deny(private_intra_doc_links)]
  72
  73 #![deny(missing_docs)]
  74 #![cfg_attr(docsrs, feature(doc_auto_cfg))]
  75
  76 use bitcoin::secp256k1::PublicKey;
  77
  78 use tokio::net::TcpStream;
  79 use tokio::{io, time};
  80 use tokio::sync::mpsc;
  81 use tokio::io::{AsyncReadExt, AsyncWrite, AsyncWriteExt};
  82
  83 use lightning::ln::peer_handler;
  84 use lightning::ln::peer_handler::SocketDescriptor as LnSocketTrait;
  85 use lightning::ln::peer_handler::CustomMessageHandler;
  86 use lightning::ln::msgs::{ChannelMessageHandler, NetAddress, OnionMessageHandler, RoutingMessageHandler};
  87 use lightning::util::logger::Logger;
  88
  89 use std::ops::Deref;
  90 use std::task;
  91 use std::net::SocketAddr;
  92 use std::net::TcpStream as StdTcpStream;
  93 use std::sync::{Arc, Mutex};
  94 use std::sync::atomic::{AtomicU64, Ordering};
  95 use std::time::Duration;
  96 use std::hash::Hash;
  97
  98 static ID_COUNTER: AtomicU64 = AtomicU64::new(0);
  99
 100 /// Connection contains all our internal state for a connection - we hold a reference to the
 101 /// Connection object (in an Arc<Mutex<>>) in each SocketDescriptor we create as well as in the
 102 /// read future (which is returned by schedule_read).
 103 struct Connection {
 104         writer: Option<io::WriteHalf<TcpStream>>,
 105         // Because our PeerManager is templated by user-provided types, and we can't (as far as I can
 106         // tell) have a const RawWakerVTable built out of templated functions, we need some indirection
 107         // between being woken up with write-ready and calling PeerManager::write_buffer_space_avail.
 108         // This provides that indirection, with a Sender which gets handed to the PeerManager Arc on
 109         // the schedule_read stack.
 110         //
 111         // An alternative (likely more effecient) approach would involve creating a RawWakerVTable at
 112         // runtime with functions templated by the Arc<PeerManager> type, calling
 113         // write_buffer_space_avail directly from tokio's write wake, however doing so would require
 114         // more unsafe voodo than I really feel like writing.
 115         write_avail: mpsc::Sender<()>,
 116         // When we are told by rust-lightning to pause read (because we have writes backing up), we do
 117         // so by setting read_paused. At that point, the read task will stop reading bytes from the
 118         // socket. To wake it up (without otherwise changing its state, we can push a value into this
 119         // Sender.
 120         read_waker: mpsc::Sender<()>,
 121         read_paused: bool,
 122         rl_requested_disconnect: bool,
 123         id: u64,
 124 }
 125 impl Connection {
 126         async fn poll_event_process<PM, CMH, RMH, OMH, L, UMH>(
 127                 peer_manager: PM,
 128                 mut event_receiver: mpsc::Receiver<()>,
 129         ) where
 130                         PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>> + 'static + Send + Sync,
 131                         CMH: Deref + 'static + Send + Sync,
 132                         RMH: Deref + 'static + Send + Sync,
 133                         OMH: Deref + 'static + Send + Sync,
 134                         L: Deref + 'static + Send + Sync,
 135                         UMH: Deref + 'static + Send + Sync,
 136                         CMH::Target: ChannelMessageHandler + Send + Sync,
 137                         RMH::Target: RoutingMessageHandler + Send + Sync,
 138                         OMH::Target: OnionMessageHandler + Send + Sync,
 139                         L::Target: Logger + Send + Sync,
 140                         UMH::Target: CustomMessageHandler + Send + Sync,
 141         {
 142                 loop {
 143                         if event_receiver.recv().await.is_none() {
 144                                 return;
 145                         }
 146                         peer_manager.process_events();
 147                 }
 148         }
 149
 150         async fn schedule_read<PM, CMH, RMH, OMH, L, UMH>(
 151                 peer_manager: PM,
 152                 us: Arc<Mutex<Self>>,
 153                 mut reader: io::ReadHalf<TcpStream>,
 154                 mut read_wake_receiver: mpsc::Receiver<()>,
 155                 mut write_avail_receiver: mpsc::Receiver<()>,
 156         ) where
 157                         PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>> + 'static + Send + Sync + Clone,
 158                         CMH: Deref + 'static + Send + Sync,
 159                         RMH: Deref + 'static + Send + Sync,
 160                         OMH: Deref + 'static + Send + Sync,
 161                         L: Deref + 'static + Send + Sync,
 162                         UMH: Deref + 'static + Send + Sync,
 163                         CMH::Target: ChannelMessageHandler + 'static + Send + Sync,
 164                         RMH::Target: RoutingMessageHandler + 'static + Send + Sync,
 165                         OMH::Target: OnionMessageHandler + 'static + Send + Sync,
 166                         L::Target: Logger + 'static + Send + Sync,
 167                         UMH::Target: CustomMessageHandler + 'static + Send + Sync,
 168                 {
 169                 // Create a waker to wake up poll_event_process, above
 170                 let (event_waker, event_receiver) = mpsc::channel(1);
 171                 tokio::spawn(Self::poll_event_process(peer_manager.clone(), event_receiver));
 172
 173                 // 8KB is nice and big but also should never cause any issues with stack overflowing.
 174                 let mut buf = [0; 8192];
 175
 176                 let mut our_descriptor = SocketDescriptor::new(us.clone());
 177                 // An enum describing why we did/are disconnecting:
 178                 enum Disconnect {
 179                         // Rust-Lightning told us to disconnect, either by returning an Err or by calling
 180                         // SocketDescriptor::disconnect_socket.
 181                         // In this case, we do not call peer_manager.socket_disconnected() as Rust-Lightning
 182                         // already knows we're disconnected.
 183                         CloseConnection,
 184                         // The connection was disconnected for some other reason, ie because the socket was
 185                         // closed.
 186                         // In this case, we do need to call peer_manager.socket_disconnected() to inform
 187                         // Rust-Lightning that the socket is gone.
 188                         PeerDisconnected
 189                 }
 190                 let disconnect_type = loop {
 191                         let read_paused = {
 192                                 let us_lock = us.lock().unwrap();
 193                                 if us_lock.rl_requested_disconnect {
 194                                         break Disconnect::CloseConnection;
 195                                 }
 196                                 us_lock.read_paused
 197                         };
 198                         tokio::select! {
 199                                 v = write_avail_receiver.recv() => {
 200                                         assert!(v.is_some()); // We can't have dropped the sending end, its in the us Arc!
 201                                         if let Err(_) = peer_manager.write_buffer_space_avail(&mut our_descriptor) {
 202                                                 break Disconnect::CloseConnection;
 203                                         }
 204                                 },
 205                                 _ = read_wake_receiver.recv() => {},
 206                                 read = reader.read(&mut buf), if !read_paused => match read {
 207                                         Ok(0) => break Disconnect::PeerDisconnected,
 208                                         Ok(len) => {
 209                                                 let read_res = peer_manager.read_event(&mut our_descriptor, &buf[0..len]);
 210                                                 let mut us_lock = us.lock().unwrap();
 211                                                 match read_res {
 212                                                         Ok(pause_read) => {
 213                                                                 if pause_read {
 214                                                                         us_lock.read_paused = true;
 215                                                                 }
 216                                                         },
 217                                                         Err(_) => break Disconnect::CloseConnection,
 218                                                 }
 219                                         },
 220                                         Err(_) => break Disconnect::PeerDisconnected,
 221                                 },
 222                         }
 223                         let _ = event_waker.try_send(());
 224
 225                         // At this point we've processed a message or two, and reset the ping timer for this
 226                         // peer, at least in the "are we still receiving messages" context, if we don't give up
 227                         // our timeslice to another task we may just spin on this peer, starving other peers
 228                         // and eventually disconnecting them for ping timeouts. Instead, we explicitly yield
 229                         // here.
 230                         tokio::task::yield_now().await;
 231                 };
 232                 let writer_option = us.lock().unwrap().writer.take();
 233                 if let Some(mut writer) = writer_option {
 234                         // If the socket is already closed, shutdown() will fail, so just ignore it.
 235                         let _ = writer.shutdown().await;
 236                 }
 237                 if let Disconnect::PeerDisconnected = disconnect_type {
 238                         peer_manager.socket_disconnected(&our_descriptor);
 239                         peer_manager.process_events();
 240                 }
 241         }
 242
 243         fn new(stream: StdTcpStream) -> (io::ReadHalf<TcpStream>, mpsc::Receiver<()>, mpsc::Receiver<()>, Arc<Mutex<Self>>) {
 244                 // We only ever need a channel of depth 1 here: if we returned a non-full write to the
 245                 // PeerManager, we will eventually get notified that there is room in the socket to write
 246                 // new bytes, which will generate an event. That event will be popped off the queue before
 247                 // we call write_buffer_space_avail, ensuring that we have room to push a new () if, during
 248                 // the write_buffer_space_avail() call, send_data() returns a non-full write.
 249                 let (write_avail, write_receiver) = mpsc::channel(1);
 250                 // Similarly here - our only goal is to make sure the reader wakes up at some point after
 251                 // we shove a value into the channel which comes after we've reset the read_paused bool to
 252                 // false.
 253                 let (read_waker, read_receiver) = mpsc::channel(1);
 254                 stream.set_nonblocking(true).unwrap();
 255                 let (reader, writer) = io::split(TcpStream::from_std(stream).unwrap());
 256
 257                 (reader, write_receiver, read_receiver,
 258                 Arc::new(Mutex::new(Self {
 259                         writer: Some(writer), write_avail, read_waker, read_paused: false,
 260                         rl_requested_disconnect: false,
 261                         id: ID_COUNTER.fetch_add(1, Ordering::AcqRel)
 262                 })))
 263         }
 264 }
 265
 266 fn get_addr_from_stream(stream: &StdTcpStream) -> Option<NetAddress> {
 267         match stream.peer_addr() {
 268                 Ok(SocketAddr::V4(sockaddr)) => Some(NetAddress::IPv4 {
 269                         addr: sockaddr.ip().octets(),
 270                         port: sockaddr.port(),
 271                 }),
 272                 Ok(SocketAddr::V6(sockaddr)) => Some(NetAddress::IPv6 {
 273                         addr: sockaddr.ip().octets(),
 274                         port: sockaddr.port(),
 275                 }),
 276                 Err(_) => None,
 277         }
 278 }
 279
 280 /// Process incoming messages and feed outgoing messages on the provided socket generated by
 281 /// accepting an incoming connection.
 282 ///
 283 /// The returned future will complete when the peer is disconnected and associated handling
 284 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
 285 /// not need to poll the provided future in order to make progress.
 286 pub fn setup_inbound<PM, CMH, RMH, OMH, L, UMH>(
 287         peer_manager: PM,
 288         stream: StdTcpStream,
 289 ) -> impl std::future::Future<Output=()> where
 290                 PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>> + 'static + Send + Sync + Clone,
 291                 CMH: Deref + 'static + Send + Sync,
 292                 RMH: Deref + 'static + Send + Sync,
 293                 OMH: Deref + 'static + Send + Sync,
 294                 L: Deref + 'static + Send + Sync,
 295                 UMH: Deref + 'static + Send + Sync,
 296                 CMH::Target: ChannelMessageHandler + Send + Sync,
 297                 RMH::Target: RoutingMessageHandler + Send + Sync,
 298                 OMH::Target: OnionMessageHandler + Send + Sync,
 299                 L::Target: Logger + Send + Sync,
 300                 UMH::Target: CustomMessageHandler + Send + Sync,
 301 {
 302         let remote_addr = get_addr_from_stream(&stream);
 303         let (reader, write_receiver, read_receiver, us) = Connection::new(stream);
 304         #[cfg(debug_assertions)]
 305         let last_us = Arc::clone(&us);
 306
 307         let handle_opt = if let Ok(_) = peer_manager.new_inbound_connection(SocketDescriptor::new(us.clone()), remote_addr) {
 308                 Some(tokio::spawn(Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver)))
 309         } else {
 310                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
 311                 // requirements.
 312                 None
 313         };
 314
 315         async move {
 316                 if let Some(handle) = handle_opt {
 317                         if let Err(e) = handle.await {
 318                                 assert!(e.is_cancelled());
 319                         } else {
 320                                 // This is certainly not guaranteed to always be true - the read loop may exit
 321                                 // while there are still pending write wakers that need to be woken up after the
 322                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
 323                                 // keep too many wakers around, this makes sense. The race should be rare (we do
 324                                 // some work after shutdown()) and an error would be a major memory leak.
 325                                 #[cfg(debug_assertions)]
 326                                 assert!(Arc::try_unwrap(last_us).is_ok());
 327                         }
 328                 }
 329         }
 330 }
 331
 332 /// Process incoming messages and feed outgoing messages on the provided socket generated by
 333 /// making an outbound connection which is expected to be accepted by a peer with the given
 334 /// public key. The relevant processing is set to run free (via tokio::spawn).
 335 ///
 336 /// The returned future will complete when the peer is disconnected and associated handling
 337 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
 338 /// not need to poll the provided future in order to make progress.
 339 pub fn setup_outbound<PM, CMH, RMH, OMH, L, UMH>(
 340         peer_manager: PM,
 341         their_node_id: PublicKey,
 342         stream: StdTcpStream,
 343 ) -> impl std::future::Future<Output=()> where
 344                 PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>> + 'static + Send + Sync + Clone,
 345                 CMH: Deref + 'static + Send + Sync,
 346                 RMH: Deref + 'static + Send + Sync,
 347                 OMH: Deref + 'static + Send + Sync,
 348                 L: Deref + 'static + Send + Sync,
 349                 UMH: Deref + 'static + Send + Sync,
 350                 CMH::Target: ChannelMessageHandler + Send + Sync,
 351                 RMH::Target: RoutingMessageHandler + Send + Sync,
 352                 OMH::Target: OnionMessageHandler + Send + Sync,
 353                 L::Target: Logger + Send + Sync,
 354                 UMH::Target: CustomMessageHandler + Send + Sync,
 355 {
 356         let remote_addr = get_addr_from_stream(&stream);
 357         let (reader, mut write_receiver, read_receiver, us) = Connection::new(stream);
 358         #[cfg(debug_assertions)]
 359         let last_us = Arc::clone(&us);
 360         let handle_opt = if let Ok(initial_send) = peer_manager.new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone()), remote_addr) {
 361                 Some(tokio::spawn(async move {
 362                         // We should essentially always have enough room in a TCP socket buffer to send the
 363                         // initial 10s of bytes. However, tokio running in single-threaded mode will always
 364                         // fail writes and wake us back up later to write. Thus, we handle a single
 365                         // std::task::Poll::Pending but still expect to write the full set of bytes at once
 366                         // and use a relatively tight timeout.
 367                         if let Ok(Ok(())) = tokio::time::timeout(Duration::from_millis(100), async {
 368                                 loop {
 369                                         match SocketDescriptor::new(us.clone()).send_data(&initial_send, true) {
 370                                                 v if v == initial_send.len() => break Ok(()),
 371                                                 0 => {
 372                                                         write_receiver.recv().await;
 373                                                         // In theory we could check for if we've been instructed to disconnect
 374                                                         // the peer here, but its OK to just skip it - we'll check for it in
 375                                                         // schedule_read prior to any relevant calls into RL.
 376                                                 },
 377                                                 _ => {
 378                                                         eprintln!("Failed to write first full message to socket!");
 379                                                         peer_manager.socket_disconnected(&SocketDescriptor::new(Arc::clone(&us)));
 380                                                         break Err(());
 381                                                 }
 382                                         }
 383                                 }
 384                         }).await {
 385                                 Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver).await;
 386                         }
 387                 }))
 388         } else {
 389                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
 390                 // requirements.
 391                 None
 392         };
 393
 394         async move {
 395                 if let Some(handle) = handle_opt {
 396                         if let Err(e) = handle.await {
 397                                 assert!(e.is_cancelled());
 398                         } else {
 399                                 // This is certainly not guaranteed to always be true - the read loop may exit
 400                                 // while there are still pending write wakers that need to be woken up after the
 401                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
 402                                 // keep too many wakers around, this makes sense. The race should be rare (we do
 403                                 // some work after shutdown()) and an error would be a major memory leak.
 404                                 #[cfg(debug_assertions)]
 405                                 assert!(Arc::try_unwrap(last_us).is_ok());
 406                         }
 407                 }
 408         }
 409 }
 410
 411 /// Process incoming messages and feed outgoing messages on a new connection made to the given
 412 /// socket address which is expected to be accepted by a peer with the given public key (by
 413 /// scheduling futures with tokio::spawn).
 414 ///
 415 /// Shorthand for TcpStream::connect(addr) with a timeout followed by setup_outbound().
 416 ///
 417 /// Returns a future (as the fn is async) which needs to be polled to complete the connection and
 418 /// connection setup. That future then returns a future which will complete when the peer is
 419 /// disconnected and associated handling futures are freed, though, because all processing in said
 420 /// futures are spawned with tokio::spawn, you do not need to poll the second future in order to
 421 /// make progress.
 422 pub async fn connect_outbound<PM, CMH, RMH, OMH, L, UMH>(
 423         peer_manager: PM,
 424         their_node_id: PublicKey,
 425         addr: SocketAddr,
 426 ) -> Option<impl std::future::Future<Output=()>> where
 427                 PM: Deref<Target = peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>> + 'static + Send + Sync + Clone,
 428                 CMH: Deref + 'static + Send + Sync,
 429                 RMH: Deref + 'static + Send + Sync,
 430                 OMH: Deref + 'static + Send + Sync,
 431                 L: Deref + 'static + Send + Sync,
 432                 UMH: Deref + 'static + Send + Sync,
 433                 CMH::Target: ChannelMessageHandler + Send + Sync,
 434                 RMH::Target: RoutingMessageHandler + Send + Sync,
 435                 OMH::Target: OnionMessageHandler + Send + Sync,
 436                 L::Target: Logger + Send + Sync,
 437                 UMH::Target: CustomMessageHandler + Send + Sync,
 438 {
 439         if let Ok(Ok(stream)) = time::timeout(Duration::from_secs(10), async { TcpStream::connect(&addr).await.map(|s| s.into_std().unwrap()) }).await {
 440                 Some(setup_outbound(peer_manager, their_node_id, stream))
 441         } else { None }
 442 }
 443
 444 const SOCK_WAKER_VTABLE: task::RawWakerVTable =
 445         task::RawWakerVTable::new(clone_socket_waker, wake_socket_waker, wake_socket_waker_by_ref, drop_socket_waker);
 446
 447 fn clone_socket_waker(orig_ptr: *const ()) -> task::RawWaker {
 448         write_avail_to_waker(orig_ptr as *const mpsc::Sender<()>)
 449 }
 450 // When waking, an error should be fine. Most likely we got two send_datas in a row, both of which
 451 // failed to fully write, but we only need to call write_buffer_space_avail() once. Otherwise, the
 452 // sending thread may have already gone away due to a socket close, in which case there's nothing
 453 // to wake up anyway.
 454 fn wake_socket_waker(orig_ptr: *const ()) {
 455         let sender = unsafe { &mut *(orig_ptr as *mut mpsc::Sender<()>) };
 456         let _ = sender.try_send(());
 457         drop_socket_waker(orig_ptr);
 458 }
 459 fn wake_socket_waker_by_ref(orig_ptr: *const ()) {
 460         let sender_ptr = orig_ptr as *const mpsc::Sender<()>;
 461         let sender = unsafe { (*sender_ptr).clone() };
 462         let _ = sender.try_send(());
 463 }
 464 fn drop_socket_waker(orig_ptr: *const ()) {
 465         let _orig_box = unsafe { Box::from_raw(orig_ptr as *mut mpsc::Sender<()>) };
 466         // _orig_box is now dropped
 467 }
 468 fn write_avail_to_waker(sender: *const mpsc::Sender<()>) -> task::RawWaker {
 469         let new_box = Box::leak(Box::new(unsafe { (*sender).clone() }));
 470         let new_ptr = new_box as *const mpsc::Sender<()>;
 471         task::RawWaker::new(new_ptr as *const (), &SOCK_WAKER_VTABLE)
 472 }
 473
 474 /// The SocketDescriptor used to refer to sockets by a PeerHandler. This is pub only as it is a
 475 /// type in the template of PeerHandler.
 476 pub struct SocketDescriptor {
 477         conn: Arc<Mutex<Connection>>,
 478         id: u64,
 479 }
 480 impl SocketDescriptor {
 481         fn new(conn: Arc<Mutex<Connection>>) -> Self {
 482                 let id = conn.lock().unwrap().id;
 483                 Self { conn, id }
 484         }
 485 }
 486 impl peer_handler::SocketDescriptor for SocketDescriptor {
 487         fn send_data(&mut self, data: &[u8], resume_read: bool) -> usize {
 488                 // To send data, we take a lock on our Connection to access the WriteHalf of the TcpStream,
 489                 // writing to it if there's room in the kernel buffer, or otherwise create a new Waker with
 490                 // a SocketDescriptor in it which can wake up the write_avail Sender, waking up the
 491                 // processing future which will call write_buffer_space_avail and we'll end up back here.
 492                 let mut us = self.conn.lock().unwrap();
 493                 if us.writer.is_none() {
 494                         // The writer gets take()n when it is time to shut down, so just fast-return 0 here.
 495                         return 0;
 496                 }
 497
 498                 if resume_read && us.read_paused {
 499                         // The schedule_read future may go to lock up but end up getting woken up by there
 500                         // being more room in the write buffer, dropping the other end of this Sender
 501                         // before we get here, so we ignore any failures to wake it up.
 502                         us.read_paused = false;
 503                         let _ = us.read_waker.try_send(());
 504                 }
 505                 if data.is_empty() { return 0; }
 506                 let waker = unsafe { task::Waker::from_raw(write_avail_to_waker(&us.write_avail)) };
 507                 let mut ctx = task::Context::from_waker(&waker);
 508                 let mut written_len = 0;
 509                 loop {
 510                         match std::pin::Pin::new(us.writer.as_mut().unwrap()).poll_write(&mut ctx, &data[written_len..]) {
 511                                 task::Poll::Ready(Ok(res)) => {
 512                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
 513                                         // know how to handle it if it does (cause it should be a Poll::Pending
 514                                         // instead):
 515                                         assert_ne!(res, 0);
 516                                         written_len += res;
 517                                         if written_len == data.len() { return written_len; }
 518                                 },
 519                                 task::Poll::Ready(Err(e)) => {
 520                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
 521                                         // know how to handle it if it does (cause it should be a Poll::Pending
 522                                         // instead):
 523                                         assert_ne!(e.kind(), io::ErrorKind::WouldBlock);
 524                                         // Probably we've already been closed, just return what we have and let the
 525                                         // read thread handle closing logic.
 526                                         return written_len;
 527                                 },
 528                                 task::Poll::Pending => {
 529                                         // We're queued up for a write event now, but we need to make sure we also
 530                                         // pause read given we're now waiting on the remote end to ACK (and in
 531                                         // accordance with the send_data() docs).
 532                                         us.read_paused = true;
 533                                         // Further, to avoid any current pending read causing a `read_event` call, wake
 534                                         // up the read_waker and restart its loop.
 535                                         let _ = us.read_waker.try_send(());
 536                                         return written_len;
 537                                 },
 538                         }
 539                 }
 540         }
 541
 542         fn disconnect_socket(&mut self) {
 543                 let mut us = self.conn.lock().unwrap();
 544                 us.rl_requested_disconnect = true;
 545                 // Wake up the sending thread, assuming it is still alive
 546                 let _ = us.write_avail.try_send(());
 547         }
 548 }
 549 impl Clone for SocketDescriptor {
 550         fn clone(&self) -> Self {
 551                 Self {
 552                         conn: Arc::clone(&self.conn),
 553                         id: self.id,
 554                 }
 555         }
 556 }
 557 impl Eq for SocketDescriptor {}
 558 impl PartialEq for SocketDescriptor {
 559         fn eq(&self, o: &Self) -> bool {
 560                 self.id == o.id
 561         }
 562 }
 563 impl Hash for SocketDescriptor {
 564         fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
 565                 self.id.hash(state);
 566         }
 567 }
 568
 569 #[cfg(test)]
 570 mod tests {
 571         use lightning::ln::features::*;
 572         use lightning::ln::msgs::*;
 573         use lightning::ln::peer_handler::{MessageHandler, PeerManager};
 574         use lightning::ln::features::NodeFeatures;
 575         use lightning::util::events::*;
 576         use bitcoin::secp256k1::{Secp256k1, SecretKey, PublicKey};
 577
 578         use tokio::sync::mpsc;
 579
 580         use std::mem;
 581         use std::sync::atomic::{AtomicBool, Ordering};
 582         use std::sync::{Arc, Mutex};
 583         use std::time::Duration;
 584
 585         pub struct TestLogger();
 586         impl lightning::util::logger::Logger for TestLogger {
 587                 fn log(&self, record: &lightning::util::logger::Record) {
 588                         println!("{:<5} [{} : {}, {}] {}", record.level.to_string(), record.module_path, record.file, record.line, record.args);
 589                 }
 590         }
 591
 592         struct MsgHandler{
 593                 expected_pubkey: PublicKey,
 594                 pubkey_connected: mpsc::Sender<()>,
 595                 pubkey_disconnected: mpsc::Sender<()>,
 596                 disconnected_flag: AtomicBool,
 597                 msg_events: Mutex<Vec<MessageSendEvent>>,
 598         }
 599         impl RoutingMessageHandler for MsgHandler {
 600                 fn handle_node_announcement(&self, _msg: &NodeAnnouncement) -> Result<bool, LightningError> { Ok(false) }
 601                 fn handle_channel_announcement(&self, _msg: &ChannelAnnouncement) -> Result<bool, LightningError> { Ok(false) }
 602                 fn handle_channel_update(&self, _msg: &ChannelUpdate) -> Result<bool, LightningError> { Ok(false) }
 603                 fn get_next_channel_announcement(&self, _starting_point: u64) -> Option<(ChannelAnnouncement, Option<ChannelUpdate>, Option<ChannelUpdate>)> { None }
 604                 fn get_next_node_announcement(&self, _starting_point: Option<&PublicKey>) -> Option<NodeAnnouncement> { None }
 605                 fn peer_connected(&self, _their_node_id: &PublicKey, _init_msg: &Init) -> Result<(), ()> { Ok(()) }
 606                 fn handle_reply_channel_range(&self, _their_node_id: &PublicKey, _msg: ReplyChannelRange) -> Result<(), LightningError> { Ok(()) }
 607                 fn handle_reply_short_channel_ids_end(&self, _their_node_id: &PublicKey, _msg: ReplyShortChannelIdsEnd) -> Result<(), LightningError> { Ok(()) }
 608                 fn handle_query_channel_range(&self, _their_node_id: &PublicKey, _msg: QueryChannelRange) -> Result<(), LightningError> { Ok(()) }
 609                 fn handle_query_short_channel_ids(&self, _their_node_id: &PublicKey, _msg: QueryShortChannelIds) -> Result<(), LightningError> { Ok(()) }
 610                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
 611                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
 612         }
 613         impl ChannelMessageHandler for MsgHandler {
 614                 fn handle_open_channel(&self, _their_node_id: &PublicKey, _msg: &OpenChannel) {}
 615                 fn handle_accept_channel(&self, _their_node_id: &PublicKey, _msg: &AcceptChannel) {}
 616                 fn handle_funding_created(&self, _their_node_id: &PublicKey, _msg: &FundingCreated) {}
 617                 fn handle_funding_signed(&self, _their_node_id: &PublicKey, _msg: &FundingSigned) {}
 618                 fn handle_channel_ready(&self, _their_node_id: &PublicKey, _msg: &ChannelReady) {}
 619                 fn handle_shutdown(&self, _their_node_id: &PublicKey, _msg: &Shutdown) {}
 620                 fn handle_closing_signed(&self, _their_node_id: &PublicKey, _msg: &ClosingSigned) {}
 621                 fn handle_update_add_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateAddHTLC) {}
 622                 fn handle_update_fulfill_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFulfillHTLC) {}
 623                 fn handle_update_fail_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailHTLC) {}
 624                 fn handle_update_fail_malformed_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailMalformedHTLC) {}
 625                 fn handle_commitment_signed(&self, _their_node_id: &PublicKey, _msg: &CommitmentSigned) {}
 626                 fn handle_revoke_and_ack(&self, _their_node_id: &PublicKey, _msg: &RevokeAndACK) {}
 627                 fn handle_update_fee(&self, _their_node_id: &PublicKey, _msg: &UpdateFee) {}
 628                 fn handle_announcement_signatures(&self, _their_node_id: &PublicKey, _msg: &AnnouncementSignatures) {}
 629                 fn handle_channel_update(&self, _their_node_id: &PublicKey, _msg: &ChannelUpdate) {}
 630                 fn peer_disconnected(&self, their_node_id: &PublicKey, _no_connection_possible: bool) {
 631                         if *their_node_id == self.expected_pubkey {
 632                                 self.disconnected_flag.store(true, Ordering::SeqCst);
 633                                 self.pubkey_disconnected.clone().try_send(()).unwrap();
 634                         }
 635                 }
 636                 fn peer_connected(&self, their_node_id: &PublicKey, _init_msg: &Init) -> Result<(), ()> {
 637                         if *their_node_id == self.expected_pubkey {
 638                                 self.pubkey_connected.clone().try_send(()).unwrap();
 639                         }
 640                         Ok(())
 641                 }
 642                 fn handle_channel_reestablish(&self, _their_node_id: &PublicKey, _msg: &ChannelReestablish) {}
 643                 fn handle_error(&self, _their_node_id: &PublicKey, _msg: &ErrorMessage) {}
 644                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
 645                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
 646         }
 647         impl MessageSendEventsProvider for MsgHandler {
 648                 fn get_and_clear_pending_msg_events(&self) -> Vec<MessageSendEvent> {
 649                         let mut ret = Vec::new();
 650                         mem::swap(&mut *self.msg_events.lock().unwrap(), &mut ret);
 651                         ret
 652                 }
 653         }
 654
 655         fn make_tcp_connection() -> (std::net::TcpStream, std::net::TcpStream) {
 656                 if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") {
 657                         (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0)
 658                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:19735") {
 659                         (std::net::TcpStream::connect("127.0.0.1:19735").unwrap(), listener.accept().unwrap().0)
 660                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9997") {
 661                         (std::net::TcpStream::connect("127.0.0.1:9997").unwrap(), listener.accept().unwrap().0)
 662                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9998") {
 663                         (std::net::TcpStream::connect("127.0.0.1:9998").unwrap(), listener.accept().unwrap().0)
 664                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") {
 665                         (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0)
 666                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") {
 667                         (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0)
 668                 } else { panic!("Failed to bind to v4 localhost on common ports"); }
 669         }
 670
 671         async fn do_basic_connection_test() {
 672                 let secp_ctx = Secp256k1::new();
 673                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
 674                 let b_key = SecretKey::from_slice(&[1; 32]).unwrap();
 675                 let a_pub = PublicKey::from_secret_key(&secp_ctx, &a_key);
 676                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
 677
 678                 let (a_connected_sender, mut a_connected) = mpsc::channel(1);
 679                 let (a_disconnected_sender, mut a_disconnected) = mpsc::channel(1);
 680                 let a_handler = Arc::new(MsgHandler {
 681                         expected_pubkey: b_pub,
 682                         pubkey_connected: a_connected_sender,
 683                         pubkey_disconnected: a_disconnected_sender,
 684                         disconnected_flag: AtomicBool::new(false),
 685                         msg_events: Mutex::new(Vec::new()),
 686                 });
 687                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
 688                         chan_handler: Arc::clone(&a_handler),
 689                         route_handler: Arc::clone(&a_handler),
 690                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 691                 }, a_key.clone(), 0, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
 692
 693                 let (b_connected_sender, mut b_connected) = mpsc::channel(1);
 694                 let (b_disconnected_sender, mut b_disconnected) = mpsc::channel(1);
 695                 let b_handler = Arc::new(MsgHandler {
 696                         expected_pubkey: a_pub,
 697                         pubkey_connected: b_connected_sender,
 698                         pubkey_disconnected: b_disconnected_sender,
 699                         disconnected_flag: AtomicBool::new(false),
 700                         msg_events: Mutex::new(Vec::new()),
 701                 });
 702                 let b_manager = Arc::new(PeerManager::new(MessageHandler {
 703                         chan_handler: Arc::clone(&b_handler),
 704                         route_handler: Arc::clone(&b_handler),
 705                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 706                 }, b_key.clone(), 0, &[2; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
 707
 708                 // We bind on localhost, hoping the environment is properly configured with a local
 709                 // address. This may not always be the case in containers and the like, so if this test is
 710                 // failing for you check that you have a loopback interface and it is configured with
 711                 // 127.0.0.1.
 712                 let (conn_a, conn_b) = make_tcp_connection();
 713
 714                 let fut_a = super::setup_outbound(Arc::clone(&a_manager), b_pub, conn_a);
 715                 let fut_b = super::setup_inbound(b_manager, conn_b);
 716
 717                 tokio::time::timeout(Duration::from_secs(10), a_connected.recv()).await.unwrap();
 718                 tokio::time::timeout(Duration::from_secs(1), b_connected.recv()).await.unwrap();
 719
 720                 a_handler.msg_events.lock().unwrap().push(MessageSendEvent::HandleError {
 721                         node_id: b_pub, action: ErrorAction::DisconnectPeer { msg: None }
 722                 });
 723                 assert!(!a_handler.disconnected_flag.load(Ordering::SeqCst));
 724                 assert!(!b_handler.disconnected_flag.load(Ordering::SeqCst));
 725
 726                 a_manager.process_events();
 727                 tokio::time::timeout(Duration::from_secs(10), a_disconnected.recv()).await.unwrap();
 728                 tokio::time::timeout(Duration::from_secs(1), b_disconnected.recv()).await.unwrap();
 729                 assert!(a_handler.disconnected_flag.load(Ordering::SeqCst));
 730                 assert!(b_handler.disconnected_flag.load(Ordering::SeqCst));
 731
 732                 fut_a.await;
 733                 fut_b.await;
 734         }
 735
 736         #[tokio::test(flavor = "multi_thread")]
 737         async fn basic_threaded_connection_test() {
 738                 do_basic_connection_test().await;
 739         }
 740
 741         #[tokio::test]
 742         async fn basic_unthreaded_connection_test() {
 743                 do_basic_connection_test().await;
 744         }
 745
 746         async fn race_disconnect_accept() {
 747                 // Previously, if we handed an already-disconnected socket to `setup_inbound` we'd panic.
 748                 // This attempts to find other similar races by opening connections and shutting them down
 749                 // while connecting. Sadly in testing this did *not* reproduce the previous issue.
 750                 let secp_ctx = Secp256k1::new();
 751                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
 752                 let b_key = SecretKey::from_slice(&[2; 32]).unwrap();
 753                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
 754
 755                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
 756                         chan_handler: Arc::new(lightning::ln::peer_handler::ErroringMessageHandler::new()),
 757                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 758                         route_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
 759                 }, a_key, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
 760
 761                 // Make two connections, one for an inbound and one for an outbound connection
 762                 let conn_a = {
 763                         let (conn_a, _) = make_tcp_connection();
 764                         conn_a
 765                 };
 766                 let conn_b = {
 767                         let (_, conn_b) = make_tcp_connection();
 768                         conn_b
 769                 };
 770
 771                 // Call connection setup inside new tokio tasks.
 772                 let manager_reference = Arc::clone(&a_manager);
 773                 tokio::spawn(async move {
 774                         super::setup_inbound(manager_reference, conn_a).await
 775                 });
 776                 tokio::spawn(async move {
 777                         super::setup_outbound(a_manager, b_pub, conn_b).await
 778                 });
 779         }
 780
 781         #[tokio::test(flavor = "multi_thread")]
 782         async fn threaded_race_disconnect_accept() {
 783                 race_disconnect_accept().await;
 784         }
 785
 786         #[tokio::test]
 787         async fn unthreaded_race_disconnect_accept() {
 788                 race_disconnect_accept().await;
 789         }
 790 }