2ac10762b04eaf23f602bc3d8dcc987438542cae
[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 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 #![deny(broken_intra_doc_links)]
70 #![deny(missing_docs)]
71
72 #![cfg_attr(docsrs, feature(doc_auto_cfg))]
73
74 use bitcoin::secp256k1::PublicKey;
75
76 use tokio::net::TcpStream;
77 use tokio::{io, time};
78 use tokio::sync::mpsc;
79 use tokio::io::{AsyncReadExt, AsyncWrite, AsyncWriteExt};
80
81 use lightning::ln::peer_handler;
82 use lightning::ln::peer_handler::SocketDescriptor as LnSocketTrait;
83 use lightning::ln::peer_handler::CustomMessageHandler;
84 use lightning::ln::msgs::{ChannelMessageHandler, RoutingMessageHandler, NetAddress};
85 use lightning::util::logger::Logger;
86
87 use std::task;
88 use std::net::SocketAddr;
89 use std::net::TcpStream as StdTcpStream;
90 use std::sync::{Arc, Mutex};
91 use std::sync::atomic::{AtomicU64, Ordering};
92 use std::time::Duration;
93 use std::hash::Hash;
94
95 static ID_COUNTER: AtomicU64 = AtomicU64::new(0);
96
97 /// Connection contains all our internal state for a connection - we hold a reference to the
98 /// Connection object (in an Arc<Mutex<>>) in each SocketDescriptor we create as well as in the
99 /// read future (which is returned by schedule_read).
100 struct Connection {
101         writer: Option<io::WriteHalf<TcpStream>>,
102         // Because our PeerManager is templated by user-provided types, and we can't (as far as I can
103         // tell) have a const RawWakerVTable built out of templated functions, we need some indirection
104         // between being woken up with write-ready and calling PeerManager::write_buffer_space_avail.
105         // This provides that indirection, with a Sender which gets handed to the PeerManager Arc on
106         // the schedule_read stack.
107         //
108         // An alternative (likely more effecient) approach would involve creating a RawWakerVTable at
109         // runtime with functions templated by the Arc<PeerManager> type, calling
110         // write_buffer_space_avail directly from tokio's write wake, however doing so would require
111         // more unsafe voodo than I really feel like writing.
112         write_avail: mpsc::Sender<()>,
113         // When we are told by rust-lightning to pause read (because we have writes backing up), we do
114         // so by setting read_paused. At that point, the read task will stop reading bytes from the
115         // socket. To wake it up (without otherwise changing its state, we can push a value into this
116         // Sender.
117         read_waker: mpsc::Sender<()>,
118         read_paused: bool,
119         rl_requested_disconnect: bool,
120         id: u64,
121 }
122 impl Connection {
123         async fn poll_event_process<CMH, RMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, Arc<CMH>, Arc<RMH>, Arc<L>, Arc<UMH>>>, mut event_receiver: mpsc::Receiver<()>) where
124                         CMH: ChannelMessageHandler + 'static + Send + Sync,
125                         RMH: RoutingMessageHandler + 'static + Send + Sync,
126                         L: Logger + 'static + ?Sized + Send + Sync,
127                         UMH: CustomMessageHandler + 'static + Send + Sync {
128                 loop {
129                         if event_receiver.recv().await.is_none() {
130                                 return;
131                         }
132                         peer_manager.process_events();
133                 }
134         }
135
136         async fn schedule_read<CMH, RMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, Arc<CMH>, Arc<RMH>, Arc<L>, Arc<UMH>>>, us: Arc<Mutex<Self>>, mut reader: io::ReadHalf<TcpStream>, mut read_wake_receiver: mpsc::Receiver<()>, mut write_avail_receiver: mpsc::Receiver<()>) where
137                         CMH: ChannelMessageHandler + 'static + Send + Sync,
138                         RMH: RoutingMessageHandler + 'static + Send + Sync,
139                         L: Logger + 'static + ?Sized + Send + Sync,
140                         UMH: CustomMessageHandler + 'static + Send + Sync {
141                 // Create a waker to wake up poll_event_process, above
142                 let (event_waker, event_receiver) = mpsc::channel(1);
143                 tokio::spawn(Self::poll_event_process(Arc::clone(&peer_manager), event_receiver));
144
145                 // 8KB is nice and big but also should never cause any issues with stack overflowing.
146                 let mut buf = [0; 8192];
147
148                 let mut our_descriptor = SocketDescriptor::new(us.clone());
149                 // An enum describing why we did/are disconnecting:
150                 enum Disconnect {
151                         // Rust-Lightning told us to disconnect, either by returning an Err or by calling
152                         // SocketDescriptor::disconnect_socket.
153                         // In this case, we do not call peer_manager.socket_disconnected() as Rust-Lightning
154                         // already knows we're disconnected.
155                         CloseConnection,
156                         // The connection was disconnected for some other reason, ie because the socket was
157                         // closed.
158                         // In this case, we do need to call peer_manager.socket_disconnected() to inform
159                         // Rust-Lightning that the socket is gone.
160                         PeerDisconnected
161                 }
162                 let disconnect_type = loop {
163                         let read_paused = {
164                                 let us_lock = us.lock().unwrap();
165                                 if us_lock.rl_requested_disconnect {
166                                         break Disconnect::CloseConnection;
167                                 }
168                                 us_lock.read_paused
169                         };
170                         tokio::select! {
171                                 v = write_avail_receiver.recv() => {
172                                         assert!(v.is_some()); // We can't have dropped the sending end, its in the us Arc!
173                                         if let Err(_) = peer_manager.write_buffer_space_avail(&mut our_descriptor) {
174                                                 break Disconnect::CloseConnection;
175                                         }
176                                 },
177                                 _ = read_wake_receiver.recv() => {},
178                                 read = reader.read(&mut buf), if !read_paused => match read {
179                                         Ok(0) => break Disconnect::PeerDisconnected,
180                                         Ok(len) => {
181                                                 let read_res = peer_manager.read_event(&mut our_descriptor, &buf[0..len]);
182                                                 let mut us_lock = us.lock().unwrap();
183                                                 match read_res {
184                                                         Ok(pause_read) => {
185                                                                 if pause_read {
186                                                                         us_lock.read_paused = true;
187                                                                 }
188                                                         },
189                                                         Err(_) => break Disconnect::CloseConnection,
190                                                 }
191                                         },
192                                         Err(_) => break Disconnect::PeerDisconnected,
193                                 },
194                         }
195                         let _ = event_waker.try_send(());
196
197                         // At this point we've processed a message or two, and reset the ping timer for this
198                         // peer, at least in the "are we still receiving messages" context, if we don't give up
199                         // our timeslice to another task we may just spin on this peer, starving other peers
200                         // and eventually disconnecting them for ping timeouts. Instead, we explicitly yield
201                         // here.
202                         tokio::task::yield_now().await;
203                 };
204                 let writer_option = us.lock().unwrap().writer.take();
205                 if let Some(mut writer) = writer_option {
206                         // If the socket is already closed, shutdown() will fail, so just ignore it.
207                         let _ = writer.shutdown().await;
208                 }
209                 if let Disconnect::PeerDisconnected = disconnect_type {
210                         peer_manager.socket_disconnected(&our_descriptor);
211                         peer_manager.process_events();
212                 }
213         }
214
215         fn new(stream: StdTcpStream) -> (io::ReadHalf<TcpStream>, mpsc::Receiver<()>, mpsc::Receiver<()>, Arc<Mutex<Self>>) {
216                 // We only ever need a channel of depth 1 here: if we returned a non-full write to the
217                 // PeerManager, we will eventually get notified that there is room in the socket to write
218                 // new bytes, which will generate an event. That event will be popped off the queue before
219                 // we call write_buffer_space_avail, ensuring that we have room to push a new () if, during
220                 // the write_buffer_space_avail() call, send_data() returns a non-full write.
221                 let (write_avail, write_receiver) = mpsc::channel(1);
222                 // Similarly here - our only goal is to make sure the reader wakes up at some point after
223                 // we shove a value into the channel which comes after we've reset the read_paused bool to
224                 // false.
225                 let (read_waker, read_receiver) = mpsc::channel(1);
226                 stream.set_nonblocking(true).unwrap();
227                 let (reader, writer) = io::split(TcpStream::from_std(stream).unwrap());
228
229                 (reader, write_receiver, read_receiver,
230                 Arc::new(Mutex::new(Self {
231                         writer: Some(writer), write_avail, read_waker, read_paused: false,
232                         rl_requested_disconnect: false,
233                         id: ID_COUNTER.fetch_add(1, Ordering::AcqRel)
234                 })))
235         }
236 }
237
238 fn get_addr_from_stream(stream: &StdTcpStream) -> Option<NetAddress> {
239         match stream.peer_addr() {
240                 Ok(SocketAddr::V4(sockaddr)) => Some(NetAddress::IPv4 {
241                         addr: sockaddr.ip().octets(),
242                         port: sockaddr.port(),
243                 }),
244                 Ok(SocketAddr::V6(sockaddr)) => Some(NetAddress::IPv6 {
245                         addr: sockaddr.ip().octets(),
246                         port: sockaddr.port(),
247                 }),
248                 Err(_) => None,
249         }
250 }
251
252 /// Process incoming messages and feed outgoing messages on the provided socket generated by
253 /// accepting an incoming connection.
254 ///
255 /// The returned future will complete when the peer is disconnected and associated handling
256 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
257 /// not need to poll the provided future in order to make progress.
258 pub fn setup_inbound<CMH, RMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, Arc<CMH>, Arc<RMH>, Arc<L>, Arc<UMH>>>, stream: StdTcpStream) -> impl std::future::Future<Output=()> where
259                 CMH: ChannelMessageHandler + 'static + Send + Sync,
260                 RMH: RoutingMessageHandler + 'static + Send + Sync,
261                 L: Logger + 'static + ?Sized + Send + Sync,
262                 UMH: CustomMessageHandler + 'static + Send + Sync {
263         let remote_addr = get_addr_from_stream(&stream);
264         let (reader, write_receiver, read_receiver, us) = Connection::new(stream);
265         #[cfg(debug_assertions)]
266         let last_us = Arc::clone(&us);
267
268         let handle_opt = if let Ok(_) = peer_manager.new_inbound_connection(SocketDescriptor::new(us.clone()), remote_addr) {
269                 Some(tokio::spawn(Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver)))
270         } else {
271                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
272                 // requirements.
273                 None
274         };
275
276         async move {
277                 if let Some(handle) = handle_opt {
278                         if let Err(e) = handle.await {
279                                 assert!(e.is_cancelled());
280                         } else {
281                                 // This is certainly not guaranteed to always be true - the read loop may exit
282                                 // while there are still pending write wakers that need to be woken up after the
283                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
284                                 // keep too many wakers around, this makes sense. The race should be rare (we do
285                                 // some work after shutdown()) and an error would be a major memory leak.
286                                 #[cfg(debug_assertions)]
287                                 assert!(Arc::try_unwrap(last_us).is_ok());
288                         }
289                 }
290         }
291 }
292
293 /// Process incoming messages and feed outgoing messages on the provided socket generated by
294 /// making an outbound connection which is expected to be accepted by a peer with the given
295 /// public key. The relevant processing is set to run free (via tokio::spawn).
296 ///
297 /// The returned future will complete when the peer is disconnected and associated handling
298 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
299 /// not need to poll the provided future in order to make progress.
300 pub fn setup_outbound<CMH, RMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, Arc<CMH>, Arc<RMH>, Arc<L>, Arc<UMH>>>, their_node_id: PublicKey, stream: StdTcpStream) -> impl std::future::Future<Output=()> where
301                 CMH: ChannelMessageHandler + 'static + Send + Sync,
302                 RMH: RoutingMessageHandler + 'static + Send + Sync,
303                 L: Logger + 'static + ?Sized + Send + Sync,
304                 UMH: CustomMessageHandler + 'static + Send + Sync {
305         let remote_addr = get_addr_from_stream(&stream);
306         let (reader, mut write_receiver, read_receiver, us) = Connection::new(stream);
307         #[cfg(debug_assertions)]
308         let last_us = Arc::clone(&us);
309         let handle_opt = if let Ok(initial_send) = peer_manager.new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone()), remote_addr) {
310                 Some(tokio::spawn(async move {
311                         // We should essentially always have enough room in a TCP socket buffer to send the
312                         // initial 10s of bytes. However, tokio running in single-threaded mode will always
313                         // fail writes and wake us back up later to write. Thus, we handle a single
314                         // std::task::Poll::Pending but still expect to write the full set of bytes at once
315                         // and use a relatively tight timeout.
316                         if let Ok(Ok(())) = tokio::time::timeout(Duration::from_millis(100), async {
317                                 loop {
318                                         match SocketDescriptor::new(us.clone()).send_data(&initial_send, true) {
319                                                 v if v == initial_send.len() => break Ok(()),
320                                                 0 => {
321                                                         write_receiver.recv().await;
322                                                         // In theory we could check for if we've been instructed to disconnect
323                                                         // the peer here, but its OK to just skip it - we'll check for it in
324                                                         // schedule_read prior to any relevant calls into RL.
325                                                 },
326                                                 _ => {
327                                                         eprintln!("Failed to write first full message to socket!");
328                                                         peer_manager.socket_disconnected(&SocketDescriptor::new(Arc::clone(&us)));
329                                                         break Err(());
330                                                 }
331                                         }
332                                 }
333                         }).await {
334                                 Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver).await;
335                         }
336                 }))
337         } else {
338                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
339                 // requirements.
340                 None
341         };
342
343         async move {
344                 if let Some(handle) = handle_opt {
345                         if let Err(e) = handle.await {
346                                 assert!(e.is_cancelled());
347                         } else {
348                                 // This is certainly not guaranteed to always be true - the read loop may exit
349                                 // while there are still pending write wakers that need to be woken up after the
350                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
351                                 // keep too many wakers around, this makes sense. The race should be rare (we do
352                                 // some work after shutdown()) and an error would be a major memory leak.
353                                 #[cfg(debug_assertions)]
354                                 assert!(Arc::try_unwrap(last_us).is_ok());
355                         }
356                 }
357         }
358 }
359
360 /// Process incoming messages and feed outgoing messages on a new connection made to the given
361 /// socket address which is expected to be accepted by a peer with the given public key (by
362 /// scheduling futures with tokio::spawn).
363 ///
364 /// Shorthand for TcpStream::connect(addr) with a timeout followed by setup_outbound().
365 ///
366 /// Returns a future (as the fn is async) which needs to be polled to complete the connection and
367 /// connection setup. That future then returns a future which will complete when the peer is
368 /// disconnected and associated handling futures are freed, though, because all processing in said
369 /// futures are spawned with tokio::spawn, you do not need to poll the second future in order to
370 /// make progress.
371 pub async fn connect_outbound<CMH, RMH, L, UMH>(peer_manager: Arc<peer_handler::PeerManager<SocketDescriptor, Arc<CMH>, Arc<RMH>, Arc<L>, Arc<UMH>>>, their_node_id: PublicKey, addr: SocketAddr) -> Option<impl std::future::Future<Output=()>> where
372                 CMH: ChannelMessageHandler + 'static + Send + Sync,
373                 RMH: RoutingMessageHandler + 'static + Send + Sync,
374                 L: Logger + 'static + ?Sized + Send + Sync,
375                 UMH: CustomMessageHandler + 'static + Send + Sync {
376         if let Ok(Ok(stream)) = time::timeout(Duration::from_secs(10), async { TcpStream::connect(&addr).await.map(|s| s.into_std().unwrap()) }).await {
377                 Some(setup_outbound(peer_manager, their_node_id, stream))
378         } else { None }
379 }
380
381 const SOCK_WAKER_VTABLE: task::RawWakerVTable =
382         task::RawWakerVTable::new(clone_socket_waker, wake_socket_waker, wake_socket_waker_by_ref, drop_socket_waker);
383
384 fn clone_socket_waker(orig_ptr: *const ()) -> task::RawWaker {
385         write_avail_to_waker(orig_ptr as *const mpsc::Sender<()>)
386 }
387 // When waking, an error should be fine. Most likely we got two send_datas in a row, both of which
388 // failed to fully write, but we only need to call write_buffer_space_avail() once. Otherwise, the
389 // sending thread may have already gone away due to a socket close, in which case there's nothing
390 // to wake up anyway.
391 fn wake_socket_waker(orig_ptr: *const ()) {
392         let sender = unsafe { &mut *(orig_ptr as *mut mpsc::Sender<()>) };
393         let _ = sender.try_send(());
394         drop_socket_waker(orig_ptr);
395 }
396 fn wake_socket_waker_by_ref(orig_ptr: *const ()) {
397         let sender_ptr = orig_ptr as *const mpsc::Sender<()>;
398         let sender = unsafe { (*sender_ptr).clone() };
399         let _ = sender.try_send(());
400 }
401 fn drop_socket_waker(orig_ptr: *const ()) {
402         let _orig_box = unsafe { Box::from_raw(orig_ptr as *mut mpsc::Sender<()>) };
403         // _orig_box is now dropped
404 }
405 fn write_avail_to_waker(sender: *const mpsc::Sender<()>) -> task::RawWaker {
406         let new_box = Box::leak(Box::new(unsafe { (*sender).clone() }));
407         let new_ptr = new_box as *const mpsc::Sender<()>;
408         task::RawWaker::new(new_ptr as *const (), &SOCK_WAKER_VTABLE)
409 }
410
411 /// The SocketDescriptor used to refer to sockets by a PeerHandler. This is pub only as it is a
412 /// type in the template of PeerHandler.
413 pub struct SocketDescriptor {
414         conn: Arc<Mutex<Connection>>,
415         id: u64,
416 }
417 impl SocketDescriptor {
418         fn new(conn: Arc<Mutex<Connection>>) -> Self {
419                 let id = conn.lock().unwrap().id;
420                 Self { conn, id }
421         }
422 }
423 impl peer_handler::SocketDescriptor for SocketDescriptor {
424         fn send_data(&mut self, data: &[u8], resume_read: bool) -> usize {
425                 // To send data, we take a lock on our Connection to access the WriteHalf of the TcpStream,
426                 // writing to it if there's room in the kernel buffer, or otherwise create a new Waker with
427                 // a SocketDescriptor in it which can wake up the write_avail Sender, waking up the
428                 // processing future which will call write_buffer_space_avail and we'll end up back here.
429                 let mut us = self.conn.lock().unwrap();
430                 if us.writer.is_none() {
431                         // The writer gets take()n when it is time to shut down, so just fast-return 0 here.
432                         return 0;
433                 }
434
435                 if resume_read && us.read_paused {
436                         // The schedule_read future may go to lock up but end up getting woken up by there
437                         // being more room in the write buffer, dropping the other end of this Sender
438                         // before we get here, so we ignore any failures to wake it up.
439                         us.read_paused = false;
440                         let _ = us.read_waker.try_send(());
441                 }
442                 if data.is_empty() { return 0; }
443                 let waker = unsafe { task::Waker::from_raw(write_avail_to_waker(&us.write_avail)) };
444                 let mut ctx = task::Context::from_waker(&waker);
445                 let mut written_len = 0;
446                 loop {
447                         match std::pin::Pin::new(us.writer.as_mut().unwrap()).poll_write(&mut ctx, &data[written_len..]) {
448                                 task::Poll::Ready(Ok(res)) => {
449                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
450                                         // know how to handle it if it does (cause it should be a Poll::Pending
451                                         // instead):
452                                         assert_ne!(res, 0);
453                                         written_len += res;
454                                         if written_len == data.len() { return written_len; }
455                                 },
456                                 task::Poll::Ready(Err(e)) => {
457                                         // The tokio docs *seem* to indicate this can't happen, and I certainly don't
458                                         // know how to handle it if it does (cause it should be a Poll::Pending
459                                         // instead):
460                                         assert_ne!(e.kind(), io::ErrorKind::WouldBlock);
461                                         // Probably we've already been closed, just return what we have and let the
462                                         // read thread handle closing logic.
463                                         return written_len;
464                                 },
465                                 task::Poll::Pending => {
466                                         // We're queued up for a write event now, but we need to make sure we also
467                                         // pause read given we're now waiting on the remote end to ACK (and in
468                                         // accordance with the send_data() docs).
469                                         us.read_paused = true;
470                                         // Further, to avoid any current pending read causing a `read_event` call, wake
471                                         // up the read_waker and restart its loop.
472                                         let _ = us.read_waker.try_send(());
473                                         return written_len;
474                                 },
475                         }
476                 }
477         }
478
479         fn disconnect_socket(&mut self) {
480                 let mut us = self.conn.lock().unwrap();
481                 us.rl_requested_disconnect = true;
482                 // Wake up the sending thread, assuming it is still alive
483                 let _ = us.write_avail.try_send(());
484         }
485 }
486 impl Clone for SocketDescriptor {
487         fn clone(&self) -> Self {
488                 Self {
489                         conn: Arc::clone(&self.conn),
490                         id: self.id,
491                 }
492         }
493 }
494 impl Eq for SocketDescriptor {}
495 impl PartialEq for SocketDescriptor {
496         fn eq(&self, o: &Self) -> bool {
497                 self.id == o.id
498         }
499 }
500 impl Hash for SocketDescriptor {
501         fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
502                 self.id.hash(state);
503         }
504 }
505
506 #[cfg(test)]
507 mod tests {
508         use lightning::ln::features::*;
509         use lightning::ln::msgs::*;
510         use lightning::ln::peer_handler::{MessageHandler, PeerManager};
511         use lightning::util::events::*;
512         use bitcoin::secp256k1::{Secp256k1, SecretKey, PublicKey};
513
514         use tokio::sync::mpsc;
515
516         use std::mem;
517         use std::sync::atomic::{AtomicBool, Ordering};
518         use std::sync::{Arc, Mutex};
519         use std::time::Duration;
520
521         pub struct TestLogger();
522         impl lightning::util::logger::Logger for TestLogger {
523                 fn log(&self, record: &lightning::util::logger::Record) {
524                         println!("{:<5} [{} : {}, {}] {}", record.level.to_string(), record.module_path, record.file, record.line, record.args);
525                 }
526         }
527
528         struct MsgHandler{
529                 expected_pubkey: PublicKey,
530                 pubkey_connected: mpsc::Sender<()>,
531                 pubkey_disconnected: mpsc::Sender<()>,
532                 disconnected_flag: AtomicBool,
533                 msg_events: Mutex<Vec<MessageSendEvent>>,
534         }
535         impl RoutingMessageHandler for MsgHandler {
536                 fn handle_node_announcement(&self, _msg: &NodeAnnouncement) -> Result<bool, LightningError> { Ok(false) }
537                 fn handle_channel_announcement(&self, _msg: &ChannelAnnouncement) -> Result<bool, LightningError> { Ok(false) }
538                 fn handle_channel_update(&self, _msg: &ChannelUpdate) -> Result<bool, LightningError> { Ok(false) }
539                 fn get_next_channel_announcements(&self, _starting_point: u64, _batch_amount: u8) -> Vec<(ChannelAnnouncement, Option<ChannelUpdate>, Option<ChannelUpdate>)> { Vec::new() }
540                 fn get_next_node_announcements(&self, _starting_point: Option<&PublicKey>, _batch_amount: u8) -> Vec<NodeAnnouncement> { Vec::new() }
541                 fn peer_connected(&self, _their_node_id: &PublicKey, _init_msg: &Init) { }
542                 fn handle_reply_channel_range(&self, _their_node_id: &PublicKey, _msg: ReplyChannelRange) -> Result<(), LightningError> { Ok(()) }
543                 fn handle_reply_short_channel_ids_end(&self, _their_node_id: &PublicKey, _msg: ReplyShortChannelIdsEnd) -> Result<(), LightningError> { Ok(()) }
544                 fn handle_query_channel_range(&self, _their_node_id: &PublicKey, _msg: QueryChannelRange) -> Result<(), LightningError> { Ok(()) }
545                 fn handle_query_short_channel_ids(&self, _their_node_id: &PublicKey, _msg: QueryShortChannelIds) -> Result<(), LightningError> { Ok(()) }
546         }
547         impl ChannelMessageHandler for MsgHandler {
548                 fn handle_open_channel(&self, _their_node_id: &PublicKey, _their_features: InitFeatures, _msg: &OpenChannel) {}
549                 fn handle_accept_channel(&self, _their_node_id: &PublicKey, _their_features: InitFeatures, _msg: &AcceptChannel) {}
550                 fn handle_funding_created(&self, _their_node_id: &PublicKey, _msg: &FundingCreated) {}
551                 fn handle_funding_signed(&self, _their_node_id: &PublicKey, _msg: &FundingSigned) {}
552                 fn handle_channel_ready(&self, _their_node_id: &PublicKey, _msg: &ChannelReady) {}
553                 fn handle_shutdown(&self, _their_node_id: &PublicKey, _their_features: &InitFeatures, _msg: &Shutdown) {}
554                 fn handle_closing_signed(&self, _their_node_id: &PublicKey, _msg: &ClosingSigned) {}
555                 fn handle_update_add_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateAddHTLC) {}
556                 fn handle_update_fulfill_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFulfillHTLC) {}
557                 fn handle_update_fail_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailHTLC) {}
558                 fn handle_update_fail_malformed_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailMalformedHTLC) {}
559                 fn handle_commitment_signed(&self, _their_node_id: &PublicKey, _msg: &CommitmentSigned) {}
560                 fn handle_revoke_and_ack(&self, _their_node_id: &PublicKey, _msg: &RevokeAndACK) {}
561                 fn handle_update_fee(&self, _their_node_id: &PublicKey, _msg: &UpdateFee) {}
562                 fn handle_announcement_signatures(&self, _their_node_id: &PublicKey, _msg: &AnnouncementSignatures) {}
563                 fn handle_channel_update(&self, _their_node_id: &PublicKey, _msg: &ChannelUpdate) {}
564                 fn peer_disconnected(&self, their_node_id: &PublicKey, _no_connection_possible: bool) {
565                         if *their_node_id == self.expected_pubkey {
566                                 self.disconnected_flag.store(true, Ordering::SeqCst);
567                                 self.pubkey_disconnected.clone().try_send(()).unwrap();
568                         }
569                 }
570                 fn peer_connected(&self, their_node_id: &PublicKey, _msg: &Init) {
571                         if *their_node_id == self.expected_pubkey {
572                                 self.pubkey_connected.clone().try_send(()).unwrap();
573                         }
574                 }
575                 fn handle_channel_reestablish(&self, _their_node_id: &PublicKey, _msg: &ChannelReestablish) {}
576                 fn handle_error(&self, _their_node_id: &PublicKey, _msg: &ErrorMessage) {}
577         }
578         impl MessageSendEventsProvider for MsgHandler {
579                 fn get_and_clear_pending_msg_events(&self) -> Vec<MessageSendEvent> {
580                         let mut ret = Vec::new();
581                         mem::swap(&mut *self.msg_events.lock().unwrap(), &mut ret);
582                         ret
583                 }
584         }
585
586         fn make_tcp_connection() -> (std::net::TcpStream, std::net::TcpStream) {
587                 if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") {
588                         (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0)
589                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:19735") {
590                         (std::net::TcpStream::connect("127.0.0.1:19735").unwrap(), listener.accept().unwrap().0)
591                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9997") {
592                         (std::net::TcpStream::connect("127.0.0.1:9997").unwrap(), listener.accept().unwrap().0)
593                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9998") {
594                         (std::net::TcpStream::connect("127.0.0.1:9998").unwrap(), listener.accept().unwrap().0)
595                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") {
596                         (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0)
597                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") {
598                         (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0)
599                 } else { panic!("Failed to bind to v4 localhost on common ports"); }
600         }
601
602         async fn do_basic_connection_test() {
603                 let secp_ctx = Secp256k1::new();
604                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
605                 let b_key = SecretKey::from_slice(&[1; 32]).unwrap();
606                 let a_pub = PublicKey::from_secret_key(&secp_ctx, &a_key);
607                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
608
609                 let (a_connected_sender, mut a_connected) = mpsc::channel(1);
610                 let (a_disconnected_sender, mut a_disconnected) = mpsc::channel(1);
611                 let a_handler = Arc::new(MsgHandler {
612                         expected_pubkey: b_pub,
613                         pubkey_connected: a_connected_sender,
614                         pubkey_disconnected: a_disconnected_sender,
615                         disconnected_flag: AtomicBool::new(false),
616                         msg_events: Mutex::new(Vec::new()),
617                 });
618                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
619                         chan_handler: Arc::clone(&a_handler),
620                         route_handler: Arc::clone(&a_handler),
621                 }, a_key.clone(), &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
622
623                 let (b_connected_sender, mut b_connected) = mpsc::channel(1);
624                 let (b_disconnected_sender, mut b_disconnected) = mpsc::channel(1);
625                 let b_handler = Arc::new(MsgHandler {
626                         expected_pubkey: a_pub,
627                         pubkey_connected: b_connected_sender,
628                         pubkey_disconnected: b_disconnected_sender,
629                         disconnected_flag: AtomicBool::new(false),
630                         msg_events: Mutex::new(Vec::new()),
631                 });
632                 let b_manager = Arc::new(PeerManager::new(MessageHandler {
633                         chan_handler: Arc::clone(&b_handler),
634                         route_handler: Arc::clone(&b_handler),
635                 }, b_key.clone(), &[2; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
636
637                 // We bind on localhost, hoping the environment is properly configured with a local
638                 // address. This may not always be the case in containers and the like, so if this test is
639                 // failing for you check that you have a loopback interface and it is configured with
640                 // 127.0.0.1.
641                 let (conn_a, conn_b) = make_tcp_connection();
642
643                 let fut_a = super::setup_outbound(Arc::clone(&a_manager), b_pub, conn_a);
644                 let fut_b = super::setup_inbound(b_manager, conn_b);
645
646                 tokio::time::timeout(Duration::from_secs(10), a_connected.recv()).await.unwrap();
647                 tokio::time::timeout(Duration::from_secs(1), b_connected.recv()).await.unwrap();
648
649                 a_handler.msg_events.lock().unwrap().push(MessageSendEvent::HandleError {
650                         node_id: b_pub, action: ErrorAction::DisconnectPeer { msg: None }
651                 });
652                 assert!(!a_handler.disconnected_flag.load(Ordering::SeqCst));
653                 assert!(!b_handler.disconnected_flag.load(Ordering::SeqCst));
654
655                 a_manager.process_events();
656                 tokio::time::timeout(Duration::from_secs(10), a_disconnected.recv()).await.unwrap();
657                 tokio::time::timeout(Duration::from_secs(1), b_disconnected.recv()).await.unwrap();
658                 assert!(a_handler.disconnected_flag.load(Ordering::SeqCst));
659                 assert!(b_handler.disconnected_flag.load(Ordering::SeqCst));
660
661                 fut_a.await;
662                 fut_b.await;
663         }
664
665         #[tokio::test(flavor = "multi_thread")]
666         async fn basic_threaded_connection_test() {
667                 do_basic_connection_test().await;
668         }
669
670         #[tokio::test]
671         async fn basic_unthreaded_connection_test() {
672                 do_basic_connection_test().await;
673         }
674
675         async fn race_disconnect_accept() {
676                 // Previously, if we handed an already-disconnected socket to `setup_inbound` we'd panic.
677                 // This attempts to find other similar races by opening connections and shutting them down
678                 // while connecting. Sadly in testing this did *not* reproduce the previous issue.
679                 let secp_ctx = Secp256k1::new();
680                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
681                 let b_key = SecretKey::from_slice(&[2; 32]).unwrap();
682                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
683
684                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
685                         chan_handler: Arc::new(lightning::ln::peer_handler::ErroringMessageHandler::new()),
686                         route_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
687                 }, a_key, &[1; 32], Arc::new(TestLogger()), Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{})));
688
689                 // Make two connections, one for an inbound and one for an outbound connection
690                 let conn_a = {
691                         let (conn_a, _) = make_tcp_connection();
692                         conn_a
693                 };
694                 let conn_b = {
695                         let (_, conn_b) = make_tcp_connection();
696                         conn_b
697                 };
698
699                 // Call connection setup inside new tokio tasks.
700                 let manager_reference = Arc::clone(&a_manager);
701                 tokio::spawn(async move {
702                         super::setup_inbound(manager_reference, conn_a).await
703                 });
704                 tokio::spawn(async move {
705                         super::setup_outbound(a_manager, b_pub, conn_b).await
706                 });
707         }
708
709         #[tokio::test(flavor = "multi_thread")]
710         async fn threaded_race_disconnect_accept() {
711                 race_disconnect_accept().await;
712         }
713
714         #[tokio::test]
715         async fn unthreaded_race_disconnect_accept() {
716                 race_disconnect_accept().await;
717         }
718 }