43bffac6252b0543eb0e007f69e7a1ca1b726192
[rust-lightning] / 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 Router = dyn lightning::routing::router::Router + Send + Sync;
35 //! type Logger = dyn lightning::util::logger::Logger + Send + Sync;
36 //! type ChainAccess = dyn lightning::chain::Access + 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, Router, Logger>>;
41 //! type PeerManager = Arc<lightning::ln::peer_handler::SimpleArcPeerManager<lightning_net_tokio::SocketDescriptor, ChainMonitor, TxBroadcaster, FeeEstimator, ChainAccess, Router, 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::ln::peer_handler;
85 use lightning::ln::peer_handler::SocketDescriptor as LnSocketTrait;
86 use lightning::ln::peer_handler::CustomMessageHandler;
87 use lightning::ln::msgs::{ChannelMessageHandler, NetAddress, OnionMessageHandler, RoutingMessageHandler};
88 use lightning::util::logger::Logger;
89
90 use std::ops::Deref;
91 use std::task;
92 use std::net::SocketAddr;
93 use std::net::TcpStream as StdTcpStream;
94 use std::sync::{Arc, Mutex};
95 use std::sync::atomic::{AtomicU64, Ordering};
96 use std::time::Duration;
97 use std::hash::Hash;
98
99 static ID_COUNTER: AtomicU64 = AtomicU64::new(0);
100
101 /// Connection contains all our internal state for a connection - we hold a reference to the
102 /// Connection object (in an Arc<Mutex<>>) in each SocketDescriptor we create as well as in the
103 /// read future (which is returned by schedule_read).
104 struct Connection {
105         writer: Option<io::WriteHalf<TcpStream>>,
106         // Because our PeerManager is templated by user-provided types, and we can't (as far as I can
107         // tell) have a const RawWakerVTable built out of templated functions, we need some indirection
108         // between being woken up with write-ready and calling PeerManager::write_buffer_space_avail.
109         // This provides that indirection, with a Sender which gets handed to the PeerManager Arc on
110         // the schedule_read stack.
111         //
112         // An alternative (likely more effecient) approach would involve creating a RawWakerVTable at
113         // runtime with functions templated by the Arc<PeerManager> type, calling
114         // write_buffer_space_avail directly from tokio's write wake, however doing so would require
115         // more unsafe voodo than I really feel like writing.
116         write_avail: mpsc::Sender<()>,
117         // When we are told by rust-lightning to pause read (because we have writes backing up), we do
118         // so by setting read_paused. At that point, the read task will stop reading bytes from the
119         // socket. To wake it up (without otherwise changing its state, we can push a value into this
120         // Sender.
121         read_waker: mpsc::Sender<()>,
122         read_paused: bool,
123         rl_requested_disconnect: bool,
124         id: u64,
125 }
126 impl Connection {
127         async fn poll_event_process<CMH, RMH, OMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>>, mut event_receiver: mpsc::Receiver<()>) where
128                         CMH: Deref + 'static + Send + Sync,
129                         RMH: Deref + 'static + Send + Sync,
130                         OMH: Deref + 'static + Send + Sync,
131                         L: Deref + 'static + Send + Sync,
132                         UMH: Deref + 'static + Send + Sync,
133                         CMH::Target: ChannelMessageHandler + Send + Sync,
134                         RMH::Target: RoutingMessageHandler + Send + Sync,
135                         OMH::Target: OnionMessageHandler + Send + Sync,
136                         L::Target: Logger + Send + Sync,
137                         UMH::Target: CustomMessageHandler + Send + Sync,
138     {
139                 loop {
140                         if event_receiver.recv().await.is_none() {
141                                 return;
142                         }
143                         peer_manager.process_events();
144                 }
145         }
146
147         async fn schedule_read<CMH, RMH, OMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>>, us: Arc<Mutex<Self>>, mut reader: io::ReadHalf<TcpStream>, mut read_wake_receiver: mpsc::Receiver<()>, mut write_avail_receiver: mpsc::Receiver<()>) where
148                         CMH: Deref + 'static + Send + Sync,
149                         RMH: Deref + 'static + Send + Sync,
150                         OMH: Deref + 'static + Send + Sync,
151                         L: Deref + 'static + Send + Sync,
152                         UMH: Deref + 'static + Send + Sync,
153                         CMH::Target: ChannelMessageHandler + 'static + Send + Sync,
154                         RMH::Target: RoutingMessageHandler + 'static + Send + Sync,
155                         OMH::Target: OnionMessageHandler + 'static + Send + Sync,
156                         L::Target: Logger + 'static + Send + Sync,
157                         UMH::Target: CustomMessageHandler + 'static + Send + Sync,
158         {
159                 // Create a waker to wake up poll_event_process, above
160                 let (event_waker, event_receiver) = mpsc::channel(1);
161                 tokio::spawn(Self::poll_event_process(Arc::clone(&peer_manager), event_receiver));
162
163                 // 8KB is nice and big but also should never cause any issues with stack overflowing.
164                 let mut buf = [0; 8192];
165
166                 let mut our_descriptor = SocketDescriptor::new(us.clone());
167                 // An enum describing why we did/are disconnecting:
168                 enum Disconnect {
169                         // Rust-Lightning told us to disconnect, either by returning an Err or by calling
170                         // SocketDescriptor::disconnect_socket.
171                         // In this case, we do not call peer_manager.socket_disconnected() as Rust-Lightning
172                         // already knows we're disconnected.
173                         CloseConnection,
174                         // The connection was disconnected for some other reason, ie because the socket was
175                         // closed.
176                         // In this case, we do need to call peer_manager.socket_disconnected() to inform
177                         // Rust-Lightning that the socket is gone.
178                         PeerDisconnected
179                 }
180                 let disconnect_type = loop {
181                         let read_paused = {
182                                 let us_lock = us.lock().unwrap();
183                                 if us_lock.rl_requested_disconnect {
184                                         break Disconnect::CloseConnection;
185                                 }
186                                 us_lock.read_paused
187                         };
188                         tokio::select! {
189                                 v = write_avail_receiver.recv() => {
190                                         assert!(v.is_some()); // We can't have dropped the sending end, its in the us Arc!
191                                         if let Err(_) = peer_manager.write_buffer_space_avail(&mut our_descriptor) {
192                                                 break Disconnect::CloseConnection;
193                                         }
194                                 },
195                                 _ = read_wake_receiver.recv() => {},
196                                 read = reader.read(&mut buf), if !read_paused => match read {
197                                         Ok(0) => break Disconnect::PeerDisconnected,
198                                         Ok(len) => {
199                                                 let read_res = peer_manager.read_event(&mut our_descriptor, &buf[0..len]);
200                                                 let mut us_lock = us.lock().unwrap();
201                                                 match read_res {
202                                                         Ok(pause_read) => {
203                                                                 if pause_read {
204                                                                         us_lock.read_paused = true;
205                                                                 }
206                                                         },
207                                                         Err(_) => break Disconnect::CloseConnection,
208                                                 }
209                                         },
210                                         Err(_) => break Disconnect::PeerDisconnected,
211                                 },
212                         }
213                         let _ = event_waker.try_send(());
214
215                         // At this point we've processed a message or two, and reset the ping timer for this
216                         // peer, at least in the "are we still receiving messages" context, if we don't give up
217                         // our timeslice to another task we may just spin on this peer, starving other peers
218                         // and eventually disconnecting them for ping timeouts. Instead, we explicitly yield
219                         // here.
220                         tokio::task::yield_now().await;
221                 };
222                 let writer_option = us.lock().unwrap().writer.take();
223                 if let Some(mut writer) = writer_option {
224                         // If the socket is already closed, shutdown() will fail, so just ignore it.
225                         let _ = writer.shutdown().await;
226                 }
227                 if let Disconnect::PeerDisconnected = disconnect_type {
228                         peer_manager.socket_disconnected(&our_descriptor);
229                         peer_manager.process_events();
230                 }
231         }
232
233         fn new(stream: StdTcpStream) -> (io::ReadHalf<TcpStream>, mpsc::Receiver<()>, mpsc::Receiver<()>, Arc<Mutex<Self>>) {
234                 // We only ever need a channel of depth 1 here: if we returned a non-full write to the
235                 // PeerManager, we will eventually get notified that there is room in the socket to write
236                 // new bytes, which will generate an event. That event will be popped off the queue before
237                 // we call write_buffer_space_avail, ensuring that we have room to push a new () if, during
238                 // the write_buffer_space_avail() call, send_data() returns a non-full write.
239                 let (write_avail, write_receiver) = mpsc::channel(1);
240                 // Similarly here - our only goal is to make sure the reader wakes up at some point after
241                 // we shove a value into the channel which comes after we've reset the read_paused bool to
242                 // false.
243                 let (read_waker, read_receiver) = mpsc::channel(1);
244                 stream.set_nonblocking(true).unwrap();
245                 let (reader, writer) = io::split(TcpStream::from_std(stream).unwrap());
246
247                 (reader, write_receiver, read_receiver,
248                 Arc::new(Mutex::new(Self {
249                         writer: Some(writer), write_avail, read_waker, read_paused: false,
250                         rl_requested_disconnect: false,
251                         id: ID_COUNTER.fetch_add(1, Ordering::AcqRel)
252                 })))
253         }
254 }
255
256 fn get_addr_from_stream(stream: &StdTcpStream) -> Option<NetAddress> {
257         match stream.peer_addr() {
258                 Ok(SocketAddr::V4(sockaddr)) => Some(NetAddress::IPv4 {
259                         addr: sockaddr.ip().octets(),
260                         port: sockaddr.port(),
261                 }),
262                 Ok(SocketAddr::V6(sockaddr)) => Some(NetAddress::IPv6 {
263                         addr: sockaddr.ip().octets(),
264                         port: sockaddr.port(),
265                 }),
266                 Err(_) => None,
267         }
268 }
269
270 /// Process incoming messages and feed outgoing messages on the provided socket generated by
271 /// accepting an incoming connection.
272 ///
273 /// The returned future will complete when the peer is disconnected and associated handling
274 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
275 /// not need to poll the provided future in order to make progress.
276 pub fn setup_inbound<CMH, RMH, OMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>>, stream: StdTcpStream) -> impl std::future::Future<Output=()> where
277                 CMH: Deref + 'static + Send + Sync,
278                 RMH: Deref + 'static + Send + Sync,
279                 OMH: Deref + 'static + Send + Sync,
280                 L: Deref + 'static + Send + Sync,
281                 UMH: Deref + 'static + Send + Sync,
282                 CMH::Target: ChannelMessageHandler + Send + Sync,
283                 RMH::Target: RoutingMessageHandler + Send + Sync,
284                 OMH::Target: OnionMessageHandler + Send + Sync,
285                 L::Target: Logger + Send + Sync,
286                 UMH::Target: CustomMessageHandler + Send + Sync,
287 {
288         let remote_addr = get_addr_from_stream(&stream);
289         let (reader, write_receiver, read_receiver, us) = Connection::new(stream);
290         #[cfg(debug_assertions)]
291         let last_us = Arc::clone(&us);
292
293         let handle_opt = if let Ok(_) = peer_manager.new_inbound_connection(SocketDescriptor::new(us.clone()), remote_addr) {
294                 Some(tokio::spawn(Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver)))
295         } else {
296                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
297                 // requirements.
298                 None
299         };
300
301         async move {
302                 if let Some(handle) = handle_opt {
303                         if let Err(e) = handle.await {
304                                 assert!(e.is_cancelled());
305                         } else {
306                                 // This is certainly not guaranteed to always be true - the read loop may exit
307                                 // while there are still pending write wakers that need to be woken up after the
308                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
309                                 // keep too many wakers around, this makes sense. The race should be rare (we do
310                                 // some work after shutdown()) and an error would be a major memory leak.
311                                 #[cfg(debug_assertions)]
312                                 assert!(Arc::try_unwrap(last_us).is_ok());
313                         }
314                 }
315         }
316 }
317
318 /// Process incoming messages and feed outgoing messages on the provided socket generated by
319 /// making an outbound connection which is expected to be accepted by a peer with the given
320 /// public key. The relevant processing is set to run free (via tokio::spawn).
321 ///
322 /// The returned future will complete when the peer is disconnected and associated handling
323 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
324 /// not need to poll the provided future in order to make progress.
325 pub fn setup_outbound<CMH, RMH, OMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>>, their_node_id: PublicKey, stream: StdTcpStream) -> impl std::future::Future<Output=()> where
326                 CMH: Deref + 'static + Send + Sync,
327                 RMH: Deref + 'static + Send + Sync,
328                 OMH: Deref + 'static + Send + Sync,
329                 L: Deref + 'static + Send + Sync,
330                 UMH: Deref + 'static + Send + Sync,
331                 CMH::Target: ChannelMessageHandler + Send + Sync,
332                 RMH::Target: RoutingMessageHandler + Send + Sync,
333                 OMH::Target: OnionMessageHandler + Send + Sync,
334                 L::Target: Logger + Send + Sync,
335                 UMH::Target: CustomMessageHandler + Send + Sync,
336 {
337         let remote_addr = get_addr_from_stream(&stream);
338         let (reader, mut write_receiver, read_receiver, us) = Connection::new(stream);
339         #[cfg(debug_assertions)]
340         let last_us = Arc::clone(&us);
341         let handle_opt = if let Ok(initial_send) = peer_manager.new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone()), remote_addr) {
342                 Some(tokio::spawn(async move {
343                         // We should essentially always have enough room in a TCP socket buffer to send the
344                         // initial 10s of bytes. However, tokio running in single-threaded mode will always
345                         // fail writes and wake us back up later to write. Thus, we handle a single
346                         // std::task::Poll::Pending but still expect to write the full set of bytes at once
347                         // and use a relatively tight timeout.
348                         if let Ok(Ok(())) = tokio::time::timeout(Duration::from_millis(100), async {
349                                 loop {
350                                         match SocketDescriptor::new(us.clone()).send_data(&initial_send, true) {
351                                                 v if v == initial_send.len() => break Ok(()),
352                                                 0 => {
353                                                         write_receiver.recv().await;
354                                                         // In theory we could check for if we've been instructed to disconnect
355                                                         // the peer here, but its OK to just skip it - we'll check for it in
356                                                         // schedule_read prior to any relevant calls into RL.
357                                                 },
358                                                 _ => {
359                                                         eprintln!("Failed to write first full message to socket!");
360                                                         peer_manager.socket_disconnected(&SocketDescriptor::new(Arc::clone(&us)));
361                                                         break Err(());
362                                                 }
363                                         }
364                                 }
365                         }).await {
366                                 Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver).await;
367                         }
368                 }))
369         } else {
370                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
371                 // requirements.
372                 None
373         };
374
375         async move {
376                 if let Some(handle) = handle_opt {
377                         if let Err(e) = handle.await {
378                                 assert!(e.is_cancelled());
379                         } else {
380                                 // This is certainly not guaranteed to always be true - the read loop may exit
381                                 // while there are still pending write wakers that need to be woken up after the
382                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
383                                 // keep too many wakers around, this makes sense. The race should be rare (we do
384                                 // some work after shutdown()) and an error would be a major memory leak.
385                                 #[cfg(debug_assertions)]
386                                 assert!(Arc::try_unwrap(last_us).is_ok());
387                         }
388                 }
389         }
390 }
391
392 /// Process incoming messages and feed outgoing messages on a new connection made to the given
393 /// socket address which is expected to be accepted by a peer with the given public key (by
394 /// scheduling futures with tokio::spawn).
395 ///
396 /// Shorthand for TcpStream::connect(addr) with a timeout followed by setup_outbound().
397 ///
398 /// Returns a future (as the fn is async) which needs to be polled to complete the connection and
399 /// connection setup. That future then returns a future which will complete when the peer is
400 /// disconnected and associated handling futures are freed, though, because all processing in said
401 /// futures are spawned with tokio::spawn, you do not need to poll the second future in order to
402 /// make progress.
403 pub async fn connect_outbound<CMH, RMH, OMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, CMH, RMH, OMH, L, UMH>>, their_node_id: PublicKey, addr: SocketAddr) -> Option<impl std::future::Future<Output=()>> where
404                 CMH: Deref + 'static + Send + Sync,
405                 RMH: Deref + 'static + Send + Sync,
406                 OMH: Deref + 'static + Send + Sync,
407                 L: Deref + 'static + Send + Sync,
408                 UMH: Deref + 'static + Send + Sync,
409                 CMH::Target: ChannelMessageHandler + Send + Sync,
410                 RMH::Target: RoutingMessageHandler + Send + Sync,
411                 OMH::Target: OnionMessageHandler + Send + Sync,
412                 L::Target: Logger + Send + Sync,
413                 UMH::Target: CustomMessageHandler + Send + Sync,
414 {
415         if let Ok(Ok(stream)) = time::timeout(Duration::from_secs(10), async { TcpStream::connect(&addr).await.map(|s| s.into_std().unwrap()) }).await {
416                 Some(setup_outbound(peer_manager, their_node_id, stream))
417         } else { None }
418 }
419
420 const SOCK_WAKER_VTABLE: task::RawWakerVTable =
421         task::RawWakerVTable::new(clone_socket_waker, wake_socket_waker, wake_socket_waker_by_ref, drop_socket_waker);
422
423 fn clone_socket_waker(orig_ptr: *const ()) -> task::RawWaker {
424         write_avail_to_waker(orig_ptr as *const mpsc::Sender<()>)
425 }
426 // When waking, an error should be fine. Most likely we got two send_datas in a row, both of which
427 // failed to fully write, but we only need to call write_buffer_space_avail() once. Otherwise, the
428 // sending thread may have already gone away due to a socket close, in which case there's nothing
429 // to wake up anyway.
430 fn wake_socket_waker(orig_ptr: *const ()) {
431         let sender = unsafe { &mut *(orig_ptr as *mut mpsc::Sender<()>) };
432         let _ = sender.try_send(());
433         drop_socket_waker(orig_ptr);
434 }
435 fn wake_socket_waker_by_ref(orig_ptr: *const ()) {
436         let sender_ptr = orig_ptr as *const mpsc::Sender<()>;
437         let sender = unsafe { (*sender_ptr).clone() };
438         let _ = sender.try_send(());
439 }
440 fn drop_socket_waker(orig_ptr: *const ()) {
441         let _orig_box = unsafe { Box::from_raw(orig_ptr as *mut mpsc::Sender<()>) };
442         // _orig_box is now dropped
443 }
444 fn write_avail_to_waker(sender: *const mpsc::Sender<()>) -> task::RawWaker {
445         let new_box = Box::leak(Box::new(unsafe { (*sender).clone() }));
446         let new_ptr = new_box as *const mpsc::Sender<()>;
447         task::RawWaker::new(new_ptr as *const (), &SOCK_WAKER_VTABLE)
448 }
449
450 /// The SocketDescriptor used to refer to sockets by a PeerHandler. This is pub only as it is a
451 /// type in the template of PeerHandler.
452 pub struct SocketDescriptor {
453         conn: Arc<Mutex<Connection>>,
454         id: u64,
455 }
456 impl SocketDescriptor {
457         fn new(conn: Arc<Mutex<Connection>>) -> Self {
458                 let id = conn.lock().unwrap().id;
459                 Self { conn, id }
460         }
461 }
462 impl peer_handler::SocketDescriptor for SocketDescriptor {
463         fn send_data(&mut self, data: &[u8], resume_read: bool) -> usize {
464                 // To send data, we take a lock on our Connection to access the WriteHalf of the TcpStream,
465                 // writing to it if there's room in the kernel buffer, or otherwise create a new Waker with
466                 // a SocketDescriptor in it which can wake up the write_avail Sender, waking up the
467                 // processing future which will call write_buffer_space_avail and we'll end up back here.
468                 let mut us = self.conn.lock().unwrap();
469                 if us.writer.is_none() {
470                         // The writer gets take()n when it is time to shut down, so just fast-return 0 here.
471                         return 0;
472                 }
473
474                 if resume_read && us.read_paused {
475                         // The schedule_read future may go to lock up but end up getting woken up by there
476                         // being more room in the write buffer, dropping the other end of this Sender
477                         // before we get here, so we ignore any failures to wake it up.
478                         us.read_paused = false;
479                         let _ = us.read_waker.try_send(());
480                 }
481                 if data.is_empty() { return 0; }
482                 let waker = unsafe { task::Waker::from_raw(write_avail_to_waker(&us.write_avail)) };
483                 let mut ctx = task::Context::from_waker(&waker);
484                 let mut written_len = 0;
485                 loop {
486                         match std::pin::Pin::new(us.writer.as_mut().unwrap()).poll_write(&mut ctx, &data[written_len..]) {
487                                 task::Poll::Ready(Ok(res)) => {
488                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
489                                         // know how to handle it if it does (cause it should be a Poll::Pending
490                                         // instead):
491                                         assert_ne!(res, 0);
492                                         written_len += res;
493                                         if written_len == data.len() { return written_len; }
494                                 },
495                                 task::Poll::Ready(Err(e)) => {
496                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
497                                         // know how to handle it if it does (cause it should be a Poll::Pending
498                                         // instead):
499                                         assert_ne!(e.kind(), io::ErrorKind::WouldBlock);
500                                         // Probably we've already been closed, just return what we have and let the
501                                         // read thread handle closing logic.
502                                         return written_len;
503                                 },
504                                 task::Poll::Pending => {
505                                         // We're queued up for a write event now, but we need to make sure we also
506                                         // pause read given we're now waiting on the remote end to ACK (and in
507                                         // accordance with the send_data() docs).
508                                         us.read_paused = true;
509                                         // Further, to avoid any current pending read causing a `read_event` call, wake
510                                         // up the read_waker and restart its loop.
511                                         let _ = us.read_waker.try_send(());
512                                         return written_len;
513                                 },
514                         }
515                 }
516         }
517
518         fn disconnect_socket(&mut self) {
519                 let mut us = self.conn.lock().unwrap();
520                 us.rl_requested_disconnect = true;
521                 // Wake up the sending thread, assuming it is still alive
522                 let _ = us.write_avail.try_send(());
523         }
524 }
525 impl Clone for SocketDescriptor {
526         fn clone(&self) -> Self {
527                 Self {
528                         conn: Arc::clone(&self.conn),
529                         id: self.id,
530                 }
531         }
532 }
533 impl Eq for SocketDescriptor {}
534 impl PartialEq for SocketDescriptor {
535         fn eq(&self, o: &Self) -> bool {
536                 self.id == o.id
537         }
538 }
539 impl Hash for SocketDescriptor {
540         fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
541                 self.id.hash(state);
542         }
543 }
544
545 #[cfg(test)]
546 mod tests {
547         use lightning::ln::features::*;
548         use lightning::ln::msgs::*;
549         use lightning::ln::peer_handler::{MessageHandler, PeerManager};
550         use lightning::ln::features::NodeFeatures;
551         use lightning::util::events::*;
552         use bitcoin::secp256k1::{Secp256k1, SecretKey, PublicKey};
553
554         use tokio::sync::mpsc;
555
556         use std::mem;
557         use std::sync::atomic::{AtomicBool, Ordering};
558         use std::sync::{Arc, Mutex};
559         use std::time::Duration;
560
561         pub struct TestLogger();
562         impl lightning::util::logger::Logger for TestLogger {
563                 fn log(&self, record: &lightning::util::logger::Record) {
564                         println!("{:<5} [{} : {}, {}] {}", record.level.to_string(), record.module_path, record.file, record.line, record.args);
565                 }
566         }
567
568         struct MsgHandler{
569                 expected_pubkey: PublicKey,
570                 pubkey_connected: mpsc::Sender<()>,
571                 pubkey_disconnected: mpsc::Sender<()>,
572                 disconnected_flag: AtomicBool,
573                 msg_events: Mutex<Vec<MessageSendEvent>>,
574         }
575         impl RoutingMessageHandler for MsgHandler {
576                 fn handle_node_announcement(&self, _msg: &NodeAnnouncement) -> Result<bool, LightningError> { Ok(false) }
577                 fn handle_channel_announcement(&self, _msg: &ChannelAnnouncement) -> Result<bool, LightningError> { Ok(false) }
578                 fn handle_channel_update(&self, _msg: &ChannelUpdate) -> Result<bool, LightningError> { Ok(false) }
579                 fn get_next_channel_announcement(&self, _starting_point: u64) -> Option<(ChannelAnnouncement, Option<ChannelUpdate>, Option<ChannelUpdate>)> { None }
580                 fn get_next_node_announcement(&self, _starting_point: Option<&PublicKey>) -> Option<NodeAnnouncement> { None }
581                 fn peer_connected(&self, _their_node_id: &PublicKey, _init_msg: &Init) -> Result<(), ()> { Ok(()) }
582                 fn handle_reply_channel_range(&self, _their_node_id: &PublicKey, _msg: ReplyChannelRange) -> Result<(), LightningError> { Ok(()) }
583                 fn handle_reply_short_channel_ids_end(&self, _their_node_id: &PublicKey, _msg: ReplyShortChannelIdsEnd) -> Result<(), LightningError> { Ok(()) }
584                 fn handle_query_channel_range(&self, _their_node_id: &PublicKey, _msg: QueryChannelRange) -> Result<(), LightningError> { Ok(()) }
585                 fn handle_query_short_channel_ids(&self, _their_node_id: &PublicKey, _msg: QueryShortChannelIds) -> Result<(), LightningError> { Ok(()) }
586                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
587                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
588         }
589         impl ChannelMessageHandler for MsgHandler {
590                 fn handle_open_channel(&self, _their_node_id: &PublicKey, _their_features: InitFeatures, _msg: &OpenChannel) {}
591                 fn handle_accept_channel(&self, _their_node_id: &PublicKey, _their_features: InitFeatures, _msg: &AcceptChannel) {}
592                 fn handle_funding_created(&self, _their_node_id: &PublicKey, _msg: &FundingCreated) {}
593                 fn handle_funding_signed(&self, _their_node_id: &PublicKey, _msg: &FundingSigned) {}
594                 fn handle_channel_ready(&self, _their_node_id: &PublicKey, _msg: &ChannelReady) {}
595                 fn handle_shutdown(&self, _their_node_id: &PublicKey, _their_features: &InitFeatures, _msg: &Shutdown) {}
596                 fn handle_closing_signed(&self, _their_node_id: &PublicKey, _msg: &ClosingSigned) {}
597                 fn handle_update_add_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateAddHTLC) {}
598                 fn handle_update_fulfill_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFulfillHTLC) {}
599                 fn handle_update_fail_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailHTLC) {}
600                 fn handle_update_fail_malformed_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailMalformedHTLC) {}
601                 fn handle_commitment_signed(&self, _their_node_id: &PublicKey, _msg: &CommitmentSigned) {}
602                 fn handle_revoke_and_ack(&self, _their_node_id: &PublicKey, _msg: &RevokeAndACK) {}
603                 fn handle_update_fee(&self, _their_node_id: &PublicKey, _msg: &UpdateFee) {}
604                 fn handle_announcement_signatures(&self, _their_node_id: &PublicKey, _msg: &AnnouncementSignatures) {}
605                 fn handle_channel_update(&self, _their_node_id: &PublicKey, _msg: &ChannelUpdate) {}
606                 fn peer_disconnected(&self, their_node_id: &PublicKey, _no_connection_possible: bool) {
607                         if *their_node_id == self.expected_pubkey {
608                                 self.disconnected_flag.store(true, Ordering::SeqCst);
609                                 self.pubkey_disconnected.clone().try_send(()).unwrap();
610                         }
611                 }
612                 fn peer_connected(&self, their_node_id: &PublicKey, _init_msg: &Init) -> Result<(), ()> {
613                         if *their_node_id == self.expected_pubkey {
614                                 self.pubkey_connected.clone().try_send(()).unwrap();
615                         }
616                         Ok(())
617                 }
618                 fn handle_channel_reestablish(&self, _their_node_id: &PublicKey, _msg: &ChannelReestablish) {}
619                 fn handle_error(&self, _their_node_id: &PublicKey, _msg: &ErrorMessage) {}
620                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
621                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
622         }
623         impl MessageSendEventsProvider for MsgHandler {
624                 fn get_and_clear_pending_msg_events(&self) -> Vec<MessageSendEvent> {
625                         let mut ret = Vec::new();
626                         mem::swap(&mut *self.msg_events.lock().unwrap(), &mut ret);
627                         ret
628                 }
629         }
630
631         fn make_tcp_connection() -> (std::net::TcpStream, std::net::TcpStream) {
632                 if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") {
633                         (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0)
634                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:19735") {
635                         (std::net::TcpStream::connect("127.0.0.1:19735").unwrap(), listener.accept().unwrap().0)
636                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9997") {
637                         (std::net::TcpStream::connect("127.0.0.1:9997").unwrap(), listener.accept().unwrap().0)
638                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9998") {
639                         (std::net::TcpStream::connect("127.0.0.1:9998").unwrap(), listener.accept().unwrap().0)
640                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") {
641                         (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0)
642                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") {
643                         (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0)
644                 } else { panic!("Failed to bind to v4 localhost on common ports"); }
645         }
646
647         async fn do_basic_connection_test() {
648                 let secp_ctx = Secp256k1::new();
649                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
650                 let b_key = SecretKey::from_slice(&[1; 32]).unwrap();
651                 let a_pub = PublicKey::from_secret_key(&secp_ctx, &a_key);
652                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
653
654                 let (a_connected_sender, mut a_connected) = mpsc::channel(1);
655                 let (a_disconnected_sender, mut a_disconnected) = mpsc::channel(1);
656                 let a_handler = Arc::new(MsgHandler {
657                         expected_pubkey: b_pub,
658                         pubkey_connected: a_connected_sender,
659                         pubkey_disconnected: a_disconnected_sender,
660                         disconnected_flag: AtomicBool::new(false),
661                         msg_events: Mutex::new(Vec::new()),
662                 });
663                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
664                         chan_handler: Arc::clone(&a_handler),
665                         route_handler: Arc::clone(&a_handler),
666                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
667                 }, a_key.clone(), 0, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
668
669                 let (b_connected_sender, mut b_connected) = mpsc::channel(1);
670                 let (b_disconnected_sender, mut b_disconnected) = mpsc::channel(1);
671                 let b_handler = Arc::new(MsgHandler {
672                         expected_pubkey: a_pub,
673                         pubkey_connected: b_connected_sender,
674                         pubkey_disconnected: b_disconnected_sender,
675                         disconnected_flag: AtomicBool::new(false),
676                         msg_events: Mutex::new(Vec::new()),
677                 });
678                 let b_manager = Arc::new(PeerManager::new(MessageHandler {
679                         chan_handler: Arc::clone(&b_handler),
680                         route_handler: Arc::clone(&b_handler),
681                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
682                 }, b_key.clone(), 0, &[2; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
683
684                 // We bind on localhost, hoping the environment is properly configured with a local
685                 // address. This may not always be the case in containers and the like, so if this test is
686                 // failing for you check that you have a loopback interface and it is configured with
687                 // 127.0.0.1.
688                 let (conn_a, conn_b) = make_tcp_connection();
689
690                 let fut_a = super::setup_outbound(Arc::clone(&a_manager), b_pub, conn_a);
691                 let fut_b = super::setup_inbound(b_manager, conn_b);
692
693                 tokio::time::timeout(Duration::from_secs(10), a_connected.recv()).await.unwrap();
694                 tokio::time::timeout(Duration::from_secs(1), b_connected.recv()).await.unwrap();
695
696                 a_handler.msg_events.lock().unwrap().push(MessageSendEvent::HandleError {
697                         node_id: b_pub, action: ErrorAction::DisconnectPeer { msg: None }
698                 });
699                 assert!(!a_handler.disconnected_flag.load(Ordering::SeqCst));
700                 assert!(!b_handler.disconnected_flag.load(Ordering::SeqCst));
701
702                 a_manager.process_events();
703                 tokio::time::timeout(Duration::from_secs(10), a_disconnected.recv()).await.unwrap();
704                 tokio::time::timeout(Duration::from_secs(1), b_disconnected.recv()).await.unwrap();
705                 assert!(a_handler.disconnected_flag.load(Ordering::SeqCst));
706                 assert!(b_handler.disconnected_flag.load(Ordering::SeqCst));
707
708                 fut_a.await;
709                 fut_b.await;
710         }
711
712         #[tokio::test(flavor = "multi_thread")]
713         async fn basic_threaded_connection_test() {
714                 do_basic_connection_test().await;
715         }
716
717         #[tokio::test]
718         async fn basic_unthreaded_connection_test() {
719                 do_basic_connection_test().await;
720         }
721
722         async fn race_disconnect_accept() {
723                 // Previously, if we handed an already-disconnected socket to `setup_inbound` we'd panic.
724                 // This attempts to find other similar races by opening connections and shutting them down
725                 // while connecting. Sadly in testing this did *not* reproduce the previous issue.
726                 let secp_ctx = Secp256k1::new();
727                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
728                 let b_key = SecretKey::from_slice(&[2; 32]).unwrap();
729                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
730
731                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
732                         chan_handler: Arc::new(lightning::ln::peer_handler::ErroringMessageHandler::new()),
733                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
734                         route_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
735                 }, a_key, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
736
737                 // Make two connections, one for an inbound and one for an outbound connection
738                 let conn_a = {
739                         let (conn_a, _) = make_tcp_connection();
740                         conn_a
741                 };
742                 let conn_b = {
743                         let (_, conn_b) = make_tcp_connection();
744                         conn_b
745                 };
746
747                 // Call connection setup inside new tokio tasks.
748                 let manager_reference = Arc::clone(&a_manager);
749                 tokio::spawn(async move {
750                         super::setup_inbound(manager_reference, conn_a).await
751                 });
752                 tokio::spawn(async move {
753                         super::setup_outbound(a_manager, b_pub, conn_b).await
754                 });
755         }
756
757         #[tokio::test(flavor = "multi_thread")]
758         async fn threaded_race_disconnect_accept() {
759                 race_disconnect_accept().await;
760         }
761
762         #[tokio::test]
763         async fn unthreaded_race_disconnect_accept() {
764                 race_disconnect_accept().await;
765         }
766 }