]> git.bitcoin.ninja Git - rust-lightning/blob - lightning-net-tokio/src/lib.rs
Improved error logging in ChannelManager::do_accept_inbound_channel
[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 [`TcpStream`]s.
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".
15 //!
16 //! The [`PeerManager`], due to the fire-and-forget nature of this logic, must be a reference,
17 //! (e.g. an [`Arc`]) and must use the [`SocketDescriptor`] provided here as the [`PeerManager`]'s
18 //! `SocketDescriptor` implementation.
19 //!
20 //! Three methods are exposed to register a new connection for handling in [`tokio::spawn`] calls;
21 //! see their individual docs for details.
22 //!
23 //! [`PeerManager`]: lightning::ln::peer_handler::PeerManager
24
25 #![deny(rustdoc::broken_intra_doc_links)]
26 #![deny(rustdoc::private_intra_doc_links)]
27
28 #![deny(missing_docs)]
29 #![cfg_attr(docsrs, feature(doc_auto_cfg))]
30
31 use bitcoin::secp256k1::PublicKey;
32
33 use tokio::net::TcpStream;
34 use tokio::time;
35 use tokio::sync::mpsc;
36
37 use lightning::ln::peer_handler;
38 use lightning::ln::peer_handler::SocketDescriptor as LnSocketTrait;
39 use lightning::ln::peer_handler::APeerManager;
40 use lightning::ln::msgs::SocketAddress;
41
42 use std::ops::Deref;
43 use std::task::{self, Poll};
44 use std::future::Future;
45 use std::net::SocketAddr;
46 use std::net::TcpStream as StdTcpStream;
47 use std::sync::{Arc, Mutex};
48 use std::sync::atomic::{AtomicU64, Ordering};
49 use std::time::Duration;
50 use std::pin::Pin;
51 use std::hash::Hash;
52
53 static ID_COUNTER: AtomicU64 = AtomicU64::new(0);
54
55 // We only need to select over multiple futures in one place, and taking on the full `tokio/macros`
56 // dependency tree in order to do so (which has broken our MSRV before) is excessive. Instead, we
57 // define a trivial two- and three- select macro with the specific types we need and just use that.
58
59 pub(crate) enum SelectorOutput {
60         A(Option<()>), B(Option<()>), C(tokio::io::Result<()>),
61 }
62
63 pub(crate) struct TwoSelector<
64         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin
65 > {
66         pub a: A,
67         pub b: B,
68 }
69
70 impl<
71         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin
72 > Future for TwoSelector<A, B> {
73         type Output = SelectorOutput;
74         fn poll(mut self: Pin<&mut Self>, ctx: &mut task::Context<'_>) -> Poll<SelectorOutput> {
75                 match Pin::new(&mut self.a).poll(ctx) {
76                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::A(res)); },
77                         Poll::Pending => {},
78                 }
79                 match Pin::new(&mut self.b).poll(ctx) {
80                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::B(res)); },
81                         Poll::Pending => {},
82                 }
83                 Poll::Pending
84         }
85 }
86
87 pub(crate) struct ThreeSelector<
88         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin, C: Future<Output=tokio::io::Result<()>> + Unpin
89 > {
90         pub a: A,
91         pub b: B,
92         pub c: C,
93 }
94
95 impl<
96         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin, C: Future<Output=tokio::io::Result<()>> + Unpin
97 > Future for ThreeSelector<A, B, C> {
98         type Output = SelectorOutput;
99         fn poll(mut self: Pin<&mut Self>, ctx: &mut task::Context<'_>) -> Poll<SelectorOutput> {
100                 match Pin::new(&mut self.a).poll(ctx) {
101                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::A(res)); },
102                         Poll::Pending => {},
103                 }
104                 match Pin::new(&mut self.b).poll(ctx) {
105                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::B(res)); },
106                         Poll::Pending => {},
107                 }
108                 match Pin::new(&mut self.c).poll(ctx) {
109                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::C(res)); },
110                         Poll::Pending => {},
111                 }
112                 Poll::Pending
113         }
114 }
115
116 /// Connection contains all our internal state for a connection - we hold a reference to the
117 /// Connection object (in an Arc<Mutex<>>) in each SocketDescriptor we create as well as in the
118 /// read future (which is returned by schedule_read).
119 struct Connection {
120         writer: Option<Arc<TcpStream>>,
121         // Because our PeerManager is templated by user-provided types, and we can't (as far as I can
122         // tell) have a const RawWakerVTable built out of templated functions, we need some indirection
123         // between being woken up with write-ready and calling PeerManager::write_buffer_space_avail.
124         // This provides that indirection, with a Sender which gets handed to the PeerManager Arc on
125         // the schedule_read stack.
126         //
127         // An alternative (likely more effecient) approach would involve creating a RawWakerVTable at
128         // runtime with functions templated by the Arc<PeerManager> type, calling
129         // write_buffer_space_avail directly from tokio's write wake, however doing so would require
130         // more unsafe voodo than I really feel like writing.
131         write_avail: mpsc::Sender<()>,
132         // When we are told by rust-lightning to pause read (because we have writes backing up), we do
133         // so by setting read_paused. At that point, the read task will stop reading bytes from the
134         // socket. To wake it up (without otherwise changing its state, we can push a value into this
135         // Sender.
136         read_waker: mpsc::Sender<()>,
137         read_paused: bool,
138         rl_requested_disconnect: bool,
139         id: u64,
140 }
141 impl Connection {
142         async fn poll_event_process<PM: Deref + 'static + Send + Sync>(
143                 peer_manager: PM,
144                 mut event_receiver: mpsc::Receiver<()>,
145         ) where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
146                 loop {
147                         if event_receiver.recv().await.is_none() {
148                                 return;
149                         }
150                         peer_manager.as_ref().process_events();
151                 }
152         }
153
154         async fn schedule_read<PM: Deref + 'static + Send + Sync + Clone>(
155                 peer_manager: PM,
156                 us: Arc<Mutex<Self>>,
157                 reader: Arc<TcpStream>,
158                 mut read_wake_receiver: mpsc::Receiver<()>,
159                 mut write_avail_receiver: mpsc::Receiver<()>,
160         ) where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
161                 // Create a waker to wake up poll_event_process, above
162                 let (event_waker, event_receiver) = mpsc::channel(1);
163                 tokio::spawn(Self::poll_event_process(peer_manager.clone(), event_receiver));
164
165                 // 4KiB is nice and big without handling too many messages all at once, giving other peers
166                 // a chance to do some work.
167                 let mut buf = [0; 4096];
168
169                 let mut our_descriptor = SocketDescriptor::new(us.clone());
170                 // An enum describing why we did/are disconnecting:
171                 enum Disconnect {
172                         // Rust-Lightning told us to disconnect, either by returning an Err or by calling
173                         // SocketDescriptor::disconnect_socket.
174                         // In this case, we do not call peer_manager.socket_disconnected() as Rust-Lightning
175                         // already knows we're disconnected.
176                         CloseConnection,
177                         // The connection was disconnected for some other reason, ie because the socket was
178                         // closed.
179                         // In this case, we do need to call peer_manager.socket_disconnected() to inform
180                         // Rust-Lightning that the socket is gone.
181                         PeerDisconnected
182                 }
183                 let disconnect_type = loop {
184                         let read_paused = {
185                                 let us_lock = us.lock().unwrap();
186                                 if us_lock.rl_requested_disconnect {
187                                         break Disconnect::CloseConnection;
188                                 }
189                                 us_lock.read_paused
190                         };
191                         // TODO: Drop the Box'ing of the futures once Rust has pin-on-stack support.
192                         let select_result = if read_paused {
193                                 TwoSelector {
194                                         a: Box::pin(write_avail_receiver.recv()),
195                                         b: Box::pin(read_wake_receiver.recv()),
196                                 }.await
197                         } else {
198                                 ThreeSelector {
199                                         a: Box::pin(write_avail_receiver.recv()),
200                                         b: Box::pin(read_wake_receiver.recv()),
201                                         c: Box::pin(reader.readable()),
202                                 }.await
203                         };
204                         match select_result {
205                                 SelectorOutput::A(v) => {
206                                         assert!(v.is_some()); // We can't have dropped the sending end, its in the us Arc!
207                                         if peer_manager.as_ref().write_buffer_space_avail(&mut our_descriptor).is_err() {
208                                                 break Disconnect::CloseConnection;
209                                         }
210                                 },
211                                 SelectorOutput::B(_) => {},
212                                 SelectorOutput::C(res) => {
213                                         if res.is_err() { break Disconnect::PeerDisconnected; }
214                                         match reader.try_read(&mut buf) {
215                                                 Ok(0) => break Disconnect::PeerDisconnected,
216                                                 Ok(len) => {
217                                                         let read_res = peer_manager.as_ref().read_event(&mut our_descriptor, &buf[0..len]);
218                                                         let mut us_lock = us.lock().unwrap();
219                                                         match read_res {
220                                                                 Ok(pause_read) => {
221                                                                         if pause_read {
222                                                                                 us_lock.read_paused = true;
223                                                                         }
224                                                                 },
225                                                                 Err(_) => break Disconnect::CloseConnection,
226                                                         }
227                                                 },
228                                                 Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => {
229                                                         // readable() is allowed to spuriously wake, so we have to handle
230                                                         // WouldBlock here.
231                                                 },
232                                                 Err(_) => break Disconnect::PeerDisconnected,
233                                         }
234                                 },
235                         }
236                         let _ = event_waker.try_send(());
237
238                         // At this point we've processed a message or two, and reset the ping timer for this
239                         // peer, at least in the "are we still receiving messages" context, if we don't give up
240                         // our timeslice to another task we may just spin on this peer, starving other peers
241                         // and eventually disconnecting them for ping timeouts. Instead, we explicitly yield
242                         // here.
243                         let _ = tokio::task::yield_now().await;
244                 };
245                 us.lock().unwrap().writer.take();
246                 if let Disconnect::PeerDisconnected = disconnect_type {
247                         peer_manager.as_ref().socket_disconnected(&our_descriptor);
248                         peer_manager.as_ref().process_events();
249                 }
250         }
251
252         fn new(stream: StdTcpStream) -> (Arc<TcpStream>, mpsc::Receiver<()>, mpsc::Receiver<()>, Arc<Mutex<Self>>) {
253                 // We only ever need a channel of depth 1 here: if we returned a non-full write to the
254                 // PeerManager, we will eventually get notified that there is room in the socket to write
255                 // new bytes, which will generate an event. That event will be popped off the queue before
256                 // we call write_buffer_space_avail, ensuring that we have room to push a new () if, during
257                 // the write_buffer_space_avail() call, send_data() returns a non-full write.
258                 let (write_avail, write_receiver) = mpsc::channel(1);
259                 // Similarly here - our only goal is to make sure the reader wakes up at some point after
260                 // we shove a value into the channel which comes after we've reset the read_paused bool to
261                 // false.
262                 let (read_waker, read_receiver) = mpsc::channel(1);
263                 stream.set_nonblocking(true).unwrap();
264                 let tokio_stream = Arc::new(TcpStream::from_std(stream).unwrap());
265
266                 (Arc::clone(&tokio_stream), write_receiver, read_receiver,
267                 Arc::new(Mutex::new(Self {
268                         writer: Some(tokio_stream), write_avail, read_waker, read_paused: false,
269                         rl_requested_disconnect: false,
270                         id: ID_COUNTER.fetch_add(1, Ordering::AcqRel)
271                 })))
272         }
273 }
274
275 fn get_addr_from_stream(stream: &StdTcpStream) -> Option<SocketAddress> {
276         match stream.peer_addr() {
277                 Ok(SocketAddr::V4(sockaddr)) => Some(SocketAddress::TcpIpV4 {
278                         addr: sockaddr.ip().octets(),
279                         port: sockaddr.port(),
280                 }),
281                 Ok(SocketAddr::V6(sockaddr)) => Some(SocketAddress::TcpIpV6 {
282                         addr: sockaddr.ip().octets(),
283                         port: sockaddr.port(),
284                 }),
285                 Err(_) => None,
286         }
287 }
288
289 /// Process incoming messages and feed outgoing messages on the provided socket generated by
290 /// accepting an incoming connection.
291 ///
292 /// The returned future will complete when the peer is disconnected and associated handling
293 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
294 /// not need to poll the provided future in order to make progress.
295 pub fn setup_inbound<PM: Deref + 'static + Send + Sync + Clone>(
296         peer_manager: PM,
297         stream: StdTcpStream,
298 ) -> impl std::future::Future<Output=()>
299 where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
300         let remote_addr = get_addr_from_stream(&stream);
301         let (reader, write_receiver, read_receiver, us) = Connection::new(stream);
302         #[cfg(test)]
303         let last_us = Arc::clone(&us);
304
305         let handle_opt = if peer_manager.as_ref().new_inbound_connection(SocketDescriptor::new(us.clone()), remote_addr).is_ok() {
306                 Some(tokio::spawn(Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver)))
307         } else {
308                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
309                 // requirements.
310                 None
311         };
312
313         async move {
314                 if let Some(handle) = handle_opt {
315                         if let Err(e) = handle.await {
316                                 assert!(e.is_cancelled());
317                         } else {
318                                 // This is certainly not guaranteed to always be true - the read loop may exit
319                                 // while there are still pending write wakers that need to be woken up after the
320                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
321                                 // keep too many wakers around, this makes sense. The race should be rare (we do
322                                 // some work after shutdown()) and an error would be a major memory leak.
323                                 #[cfg(test)]
324                                 debug_assert!(Arc::try_unwrap(last_us).is_ok());
325                         }
326                 }
327         }
328 }
329
330 /// Process incoming messages and feed outgoing messages on the provided socket generated by
331 /// making an outbound connection which is expected to be accepted by a peer with the given
332 /// public key. The relevant processing is set to run free (via tokio::spawn).
333 ///
334 /// The returned future will complete when the peer is disconnected and associated handling
335 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
336 /// not need to poll the provided future in order to make progress.
337 pub fn setup_outbound<PM: Deref + 'static + Send + Sync + Clone>(
338         peer_manager: PM,
339         their_node_id: PublicKey,
340         stream: StdTcpStream,
341 ) -> impl std::future::Future<Output=()>
342 where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
343         let remote_addr = get_addr_from_stream(&stream);
344         let (reader, mut write_receiver, read_receiver, us) = Connection::new(stream);
345         #[cfg(test)]
346         let last_us = Arc::clone(&us);
347         let handle_opt = if let Ok(initial_send) = peer_manager.as_ref().new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone()), remote_addr) {
348                 Some(tokio::spawn(async move {
349                         // We should essentially always have enough room in a TCP socket buffer to send the
350                         // initial 10s of bytes. However, tokio running in single-threaded mode will always
351                         // fail writes and wake us back up later to write. Thus, we handle a single
352                         // std::task::Poll::Pending but still expect to write the full set of bytes at once
353                         // and use a relatively tight timeout.
354                         if let Ok(Ok(())) = tokio::time::timeout(Duration::from_millis(100), async {
355                                 loop {
356                                         match SocketDescriptor::new(us.clone()).send_data(&initial_send, true) {
357                                                 v if v == initial_send.len() => break Ok(()),
358                                                 0 => {
359                                                         write_receiver.recv().await;
360                                                         // In theory we could check for if we've been instructed to disconnect
361                                                         // the peer here, but its OK to just skip it - we'll check for it in
362                                                         // schedule_read prior to any relevant calls into RL.
363                                                 },
364                                                 _ => {
365                                                         eprintln!("Failed to write first full message to socket!");
366                                                         peer_manager.as_ref().socket_disconnected(&SocketDescriptor::new(Arc::clone(&us)));
367                                                         break Err(());
368                                                 }
369                                         }
370                                 }
371                         }).await {
372                                 Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver).await;
373                         }
374                 }))
375         } else {
376                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
377                 // requirements.
378                 None
379         };
380
381         async move {
382                 if let Some(handle) = handle_opt {
383                         if let Err(e) = handle.await {
384                                 assert!(e.is_cancelled());
385                         } else {
386                                 // This is certainly not guaranteed to always be true - the read loop may exit
387                                 // while there are still pending write wakers that need to be woken up after the
388                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
389                                 // keep too many wakers around, this makes sense. The race should be rare (we do
390                                 // some work after shutdown()) and an error would be a major memory leak.
391                                 #[cfg(test)]
392                                 debug_assert!(Arc::try_unwrap(last_us).is_ok());
393                         }
394                 }
395         }
396 }
397
398 /// Process incoming messages and feed outgoing messages on a new connection made to the given
399 /// socket address which is expected to be accepted by a peer with the given public key (by
400 /// scheduling futures with tokio::spawn).
401 ///
402 /// Shorthand for TcpStream::connect(addr) with a timeout followed by setup_outbound().
403 ///
404 /// Returns a future (as the fn is async) which needs to be polled to complete the connection and
405 /// connection setup. That future then returns a future which will complete when the peer is
406 /// disconnected and associated handling futures are freed, though, because all processing in said
407 /// futures are spawned with tokio::spawn, you do not need to poll the second future in order to
408 /// make progress.
409 pub async fn connect_outbound<PM: Deref + 'static + Send + Sync + Clone>(
410         peer_manager: PM,
411         their_node_id: PublicKey,
412         addr: SocketAddr,
413 ) -> Option<impl std::future::Future<Output=()>>
414 where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
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         let new_waker = unsafe { Arc::from_raw(orig_ptr as *const mpsc::Sender<()>) };
425         let res = write_avail_to_waker(&new_waker);
426         // Don't decrement the refcount when dropping new_waker by turning it back `into_raw`.
427         let _ = Arc::into_raw(new_waker);
428         res
429 }
430 // When waking, an error should be fine. Most likely we got two send_datas in a row, both of which
431 // failed to fully write, but we only need to call write_buffer_space_avail() once. Otherwise, the
432 // sending thread may have already gone away due to a socket close, in which case there's nothing
433 // to wake up anyway.
434 fn wake_socket_waker(orig_ptr: *const ()) {
435         let sender = unsafe { &mut *(orig_ptr as *mut mpsc::Sender<()>) };
436         let _ = sender.try_send(());
437         drop_socket_waker(orig_ptr);
438 }
439 fn wake_socket_waker_by_ref(orig_ptr: *const ()) {
440         let sender_ptr = orig_ptr as *const mpsc::Sender<()>;
441         let sender = unsafe { &*sender_ptr };
442         let _ = sender.try_send(());
443 }
444 fn drop_socket_waker(orig_ptr: *const ()) {
445         let _orig_arc = unsafe { Arc::from_raw(orig_ptr as *mut mpsc::Sender<()>) };
446         // _orig_arc is now dropped
447 }
448 fn write_avail_to_waker(sender: &Arc<mpsc::Sender<()>>) -> task::RawWaker {
449         let new_ptr = Arc::into_raw(Arc::clone(&sender));
450         task::RawWaker::new(new_ptr as *const (), &SOCK_WAKER_VTABLE)
451 }
452
453 /// The SocketDescriptor used to refer to sockets by a PeerHandler. This is pub only as it is a
454 /// type in the template of PeerHandler.
455 pub struct SocketDescriptor {
456         conn: Arc<Mutex<Connection>>,
457         // We store a copy of the mpsc::Sender to wake the read task in an Arc here. While we can
458         // simply clone the sender and store a copy in each waker, that would require allocating for
459         // each waker. Instead, we can simply `Arc::clone`, creating a new reference and store the
460         // pointer in the waker.
461         write_avail_sender: Arc<mpsc::Sender<()>>,
462         id: u64,
463 }
464 impl SocketDescriptor {
465         fn new(conn: Arc<Mutex<Connection>>) -> Self {
466                 let (id, write_avail_sender) = {
467                         let us = conn.lock().unwrap();
468                         (us.id, Arc::new(us.write_avail.clone()))
469                 };
470                 Self { conn, id, write_avail_sender }
471         }
472 }
473 impl peer_handler::SocketDescriptor for SocketDescriptor {
474         fn send_data(&mut self, data: &[u8], resume_read: bool) -> usize {
475                 // To send data, we take a lock on our Connection to access the TcpStream, writing to it if
476                 // there's room in the kernel buffer, or otherwise create a new Waker with a
477                 // SocketDescriptor in it which can wake up the write_avail Sender, waking up the
478                 // processing future which will call write_buffer_space_avail and we'll end up back here.
479                 let mut us = self.conn.lock().unwrap();
480                 if us.writer.is_none() {
481                         // The writer gets take()n when it is time to shut down, so just fast-return 0 here.
482                         return 0;
483                 }
484
485                 if resume_read && us.read_paused {
486                         // The schedule_read future may go to lock up but end up getting woken up by there
487                         // being more room in the write buffer, dropping the other end of this Sender
488                         // before we get here, so we ignore any failures to wake it up.
489                         us.read_paused = false;
490                         let _ = us.read_waker.try_send(());
491                 }
492                 if data.is_empty() { return 0; }
493                 let waker = unsafe { task::Waker::from_raw(write_avail_to_waker(&self.write_avail_sender)) };
494                 let mut ctx = task::Context::from_waker(&waker);
495                 let mut written_len = 0;
496                 loop {
497                         match us.writer.as_ref().unwrap().poll_write_ready(&mut ctx) {
498                                 task::Poll::Ready(Ok(())) => {
499                                         match us.writer.as_ref().unwrap().try_write(&data[written_len..]) {
500                                                 Ok(res) => {
501                                                         debug_assert_ne!(res, 0);
502                                                         written_len += res;
503                                                         if written_len == data.len() { return written_len; }
504                                                 },
505                                                 Err(_) => return written_len,
506                                         }
507                                 },
508                                 task::Poll::Ready(Err(_)) => return written_len,
509                                 task::Poll::Pending => {
510                                         // We're queued up for a write event now, but we need to make sure we also
511                                         // pause read given we're now waiting on the remote end to ACK (and in
512                                         // accordance with the send_data() docs).
513                                         us.read_paused = true;
514                                         // Further, to avoid any current pending read causing a `read_event` call, wake
515                                         // up the read_waker and restart its loop.
516                                         let _ = us.read_waker.try_send(());
517                                         return written_len;
518                                 },
519                         }
520                 }
521         }
522
523         fn disconnect_socket(&mut self) {
524                 let mut us = self.conn.lock().unwrap();
525                 us.rl_requested_disconnect = true;
526                 // Wake up the sending thread, assuming it is still alive
527                 let _ = us.write_avail.try_send(());
528         }
529 }
530 impl Clone for SocketDescriptor {
531         fn clone(&self) -> Self {
532                 Self {
533                         conn: Arc::clone(&self.conn),
534                         id: self.id,
535                         write_avail_sender: Arc::clone(&self.write_avail_sender),
536                 }
537         }
538 }
539 impl Eq for SocketDescriptor {}
540 impl PartialEq for SocketDescriptor {
541         fn eq(&self, o: &Self) -> bool {
542                 self.id == o.id
543         }
544 }
545 impl Hash for SocketDescriptor {
546         fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
547                 self.id.hash(state);
548         }
549 }
550
551 #[cfg(test)]
552 mod tests {
553         use lightning::ln::features::*;
554         use lightning::ln::msgs::*;
555         use lightning::ln::peer_handler::{MessageHandler, PeerManager};
556         use lightning::ln::features::NodeFeatures;
557         use lightning::routing::gossip::NodeId;
558         use lightning::events::*;
559         use lightning::util::test_utils::TestNodeSigner;
560         use bitcoin::Network;
561         use bitcoin::blockdata::constants::ChainHash;
562         use bitcoin::secp256k1::{Secp256k1, SecretKey, PublicKey};
563
564         use tokio::sync::mpsc;
565
566         use std::mem;
567         use std::sync::atomic::{AtomicBool, Ordering};
568         use std::sync::{Arc, Mutex};
569         use std::time::Duration;
570
571         pub struct TestLogger();
572         impl lightning::util::logger::Logger for TestLogger {
573                 fn log(&self, record: lightning::util::logger::Record) {
574                         println!("{:<5} [{} : {}, {}] {}", record.level.to_string(), record.module_path, record.file, record.line, record.args);
575                 }
576         }
577
578         struct MsgHandler{
579                 expected_pubkey: PublicKey,
580                 pubkey_connected: mpsc::Sender<()>,
581                 pubkey_disconnected: mpsc::Sender<()>,
582                 disconnected_flag: AtomicBool,
583                 msg_events: Mutex<Vec<MessageSendEvent>>,
584         }
585         impl RoutingMessageHandler for MsgHandler {
586                 fn handle_node_announcement(&self, _msg: &NodeAnnouncement) -> Result<bool, LightningError> { Ok(false) }
587                 fn handle_channel_announcement(&self, _msg: &ChannelAnnouncement) -> Result<bool, LightningError> { Ok(false) }
588                 fn handle_channel_update(&self, _msg: &ChannelUpdate) -> Result<bool, LightningError> { Ok(false) }
589                 fn get_next_channel_announcement(&self, _starting_point: u64) -> Option<(ChannelAnnouncement, Option<ChannelUpdate>, Option<ChannelUpdate>)> { None }
590                 fn get_next_node_announcement(&self, _starting_point: Option<&NodeId>) -> Option<NodeAnnouncement> { None }
591                 fn peer_connected(&self, _their_node_id: &PublicKey, _init_msg: &Init, _inbound: bool) -> Result<(), ()> { Ok(()) }
592                 fn handle_reply_channel_range(&self, _their_node_id: &PublicKey, _msg: ReplyChannelRange) -> Result<(), LightningError> { Ok(()) }
593                 fn handle_reply_short_channel_ids_end(&self, _their_node_id: &PublicKey, _msg: ReplyShortChannelIdsEnd) -> Result<(), LightningError> { Ok(()) }
594                 fn handle_query_channel_range(&self, _their_node_id: &PublicKey, _msg: QueryChannelRange) -> Result<(), LightningError> { Ok(()) }
595                 fn handle_query_short_channel_ids(&self, _their_node_id: &PublicKey, _msg: QueryShortChannelIds) -> Result<(), LightningError> { Ok(()) }
596                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
597                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
598                 fn processing_queue_high(&self) -> bool { false }
599         }
600         impl ChannelMessageHandler for MsgHandler {
601                 fn handle_open_channel(&self, _their_node_id: &PublicKey, _msg: &OpenChannel) {}
602                 fn handle_accept_channel(&self, _their_node_id: &PublicKey, _msg: &AcceptChannel) {}
603                 fn handle_funding_created(&self, _their_node_id: &PublicKey, _msg: &FundingCreated) {}
604                 fn handle_funding_signed(&self, _their_node_id: &PublicKey, _msg: &FundingSigned) {}
605                 fn handle_channel_ready(&self, _their_node_id: &PublicKey, _msg: &ChannelReady) {}
606                 fn handle_shutdown(&self, _their_node_id: &PublicKey, _msg: &Shutdown) {}
607                 fn handle_closing_signed(&self, _their_node_id: &PublicKey, _msg: &ClosingSigned) {}
608                 fn handle_update_add_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateAddHTLC) {}
609                 fn handle_update_fulfill_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFulfillHTLC) {}
610                 fn handle_update_fail_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailHTLC) {}
611                 fn handle_update_fail_malformed_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailMalformedHTLC) {}
612                 fn handle_commitment_signed(&self, _their_node_id: &PublicKey, _msg: &CommitmentSigned) {}
613                 fn handle_revoke_and_ack(&self, _their_node_id: &PublicKey, _msg: &RevokeAndACK) {}
614                 fn handle_update_fee(&self, _their_node_id: &PublicKey, _msg: &UpdateFee) {}
615                 fn handle_announcement_signatures(&self, _their_node_id: &PublicKey, _msg: &AnnouncementSignatures) {}
616                 fn handle_channel_update(&self, _their_node_id: &PublicKey, _msg: &ChannelUpdate) {}
617                 fn handle_open_channel_v2(&self, _their_node_id: &PublicKey, _msg: &OpenChannelV2) {}
618                 fn handle_accept_channel_v2(&self, _their_node_id: &PublicKey, _msg: &AcceptChannelV2) {}
619                 fn handle_stfu(&self, _their_node_id: &PublicKey, _msg: &Stfu) {}
620                 fn handle_splice(&self, _their_node_id: &PublicKey, _msg: &Splice) {}
621                 fn handle_splice_ack(&self, _their_node_id: &PublicKey, _msg: &SpliceAck) {}
622                 fn handle_splice_locked(&self, _their_node_id: &PublicKey, _msg: &SpliceLocked) {}
623                 fn handle_tx_add_input(&self, _their_node_id: &PublicKey, _msg: &TxAddInput) {}
624                 fn handle_tx_add_output(&self, _their_node_id: &PublicKey, _msg: &TxAddOutput) {}
625                 fn handle_tx_remove_input(&self, _their_node_id: &PublicKey, _msg: &TxRemoveInput) {}
626                 fn handle_tx_remove_output(&self, _their_node_id: &PublicKey, _msg: &TxRemoveOutput) {}
627                 fn handle_tx_complete(&self, _their_node_id: &PublicKey, _msg: &TxComplete) {}
628                 fn handle_tx_signatures(&self, _their_node_id: &PublicKey, _msg: &TxSignatures) {}
629                 fn handle_tx_init_rbf(&self, _their_node_id: &PublicKey, _msg: &TxInitRbf) {}
630                 fn handle_tx_ack_rbf(&self, _their_node_id: &PublicKey, _msg: &TxAckRbf) {}
631                 fn handle_tx_abort(&self, _their_node_id: &PublicKey, _msg: &TxAbort) {}
632                 fn peer_disconnected(&self, their_node_id: &PublicKey) {
633                         if *their_node_id == self.expected_pubkey {
634                                 self.disconnected_flag.store(true, Ordering::SeqCst);
635                                 self.pubkey_disconnected.clone().try_send(()).unwrap();
636                         }
637                 }
638                 fn peer_connected(&self, their_node_id: &PublicKey, _init_msg: &Init, _inbound: bool) -> Result<(), ()> {
639                         if *their_node_id == self.expected_pubkey {
640                                 self.pubkey_connected.clone().try_send(()).unwrap();
641                         }
642                         Ok(())
643                 }
644                 fn handle_channel_reestablish(&self, _their_node_id: &PublicKey, _msg: &ChannelReestablish) {}
645                 fn handle_error(&self, _their_node_id: &PublicKey, _msg: &ErrorMessage) {}
646                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
647                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
648                 fn get_chain_hashes(&self) -> Option<Vec<ChainHash>> {
649                         Some(vec![ChainHash::using_genesis_block(Network::Testnet)])
650                 }
651         }
652         impl MessageSendEventsProvider for MsgHandler {
653                 fn get_and_clear_pending_msg_events(&self) -> Vec<MessageSendEvent> {
654                         let mut ret = Vec::new();
655                         mem::swap(&mut *self.msg_events.lock().unwrap(), &mut ret);
656                         ret
657                 }
658         }
659
660         fn make_tcp_connection() -> (std::net::TcpStream, std::net::TcpStream) {
661                 if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") {
662                         (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0)
663                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:19735") {
664                         (std::net::TcpStream::connect("127.0.0.1:19735").unwrap(), listener.accept().unwrap().0)
665                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9997") {
666                         (std::net::TcpStream::connect("127.0.0.1:9997").unwrap(), listener.accept().unwrap().0)
667                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9998") {
668                         (std::net::TcpStream::connect("127.0.0.1:9998").unwrap(), listener.accept().unwrap().0)
669                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") {
670                         (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0)
671                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") {
672                         (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0)
673                 } else { panic!("Failed to bind to v4 localhost on common ports"); }
674         }
675
676         async fn do_basic_connection_test() {
677                 let secp_ctx = Secp256k1::new();
678                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
679                 let b_key = SecretKey::from_slice(&[1; 32]).unwrap();
680                 let a_pub = PublicKey::from_secret_key(&secp_ctx, &a_key);
681                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
682
683                 let (a_connected_sender, mut a_connected) = mpsc::channel(1);
684                 let (a_disconnected_sender, mut a_disconnected) = mpsc::channel(1);
685                 let a_handler = Arc::new(MsgHandler {
686                         expected_pubkey: b_pub,
687                         pubkey_connected: a_connected_sender,
688                         pubkey_disconnected: a_disconnected_sender,
689                         disconnected_flag: AtomicBool::new(false),
690                         msg_events: Mutex::new(Vec::new()),
691                 });
692                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
693                         chan_handler: Arc::clone(&a_handler),
694                         route_handler: Arc::clone(&a_handler),
695                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
696                         custom_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
697                 }, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(TestNodeSigner::new(a_key))));
698
699                 let (b_connected_sender, mut b_connected) = mpsc::channel(1);
700                 let (b_disconnected_sender, mut b_disconnected) = mpsc::channel(1);
701                 let b_handler = Arc::new(MsgHandler {
702                         expected_pubkey: a_pub,
703                         pubkey_connected: b_connected_sender,
704                         pubkey_disconnected: b_disconnected_sender,
705                         disconnected_flag: AtomicBool::new(false),
706                         msg_events: Mutex::new(Vec::new()),
707                 });
708                 let b_manager = Arc::new(PeerManager::new(MessageHandler {
709                         chan_handler: Arc::clone(&b_handler),
710                         route_handler: Arc::clone(&b_handler),
711                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
712                         custom_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
713                 }, 0, &[2; 32], Arc::new(TestLogger()), Arc::new(TestNodeSigner::new(b_key))));
714
715                 // We bind on localhost, hoping the environment is properly configured with a local
716                 // address. This may not always be the case in containers and the like, so if this test is
717                 // failing for you check that you have a loopback interface and it is configured with
718                 // 127.0.0.1.
719                 let (conn_a, conn_b) = make_tcp_connection();
720
721                 let fut_a = super::setup_outbound(Arc::clone(&a_manager), b_pub, conn_a);
722                 let fut_b = super::setup_inbound(b_manager, conn_b);
723
724                 tokio::time::timeout(Duration::from_secs(10), a_connected.recv()).await.unwrap();
725                 tokio::time::timeout(Duration::from_secs(1), b_connected.recv()).await.unwrap();
726
727                 a_handler.msg_events.lock().unwrap().push(MessageSendEvent::HandleError {
728                         node_id: b_pub, action: ErrorAction::DisconnectPeer { msg: None }
729                 });
730                 assert!(!a_handler.disconnected_flag.load(Ordering::SeqCst));
731                 assert!(!b_handler.disconnected_flag.load(Ordering::SeqCst));
732
733                 a_manager.process_events();
734                 tokio::time::timeout(Duration::from_secs(10), a_disconnected.recv()).await.unwrap();
735                 tokio::time::timeout(Duration::from_secs(1), b_disconnected.recv()).await.unwrap();
736                 assert!(a_handler.disconnected_flag.load(Ordering::SeqCst));
737                 assert!(b_handler.disconnected_flag.load(Ordering::SeqCst));
738
739                 fut_a.await;
740                 fut_b.await;
741         }
742
743         #[tokio::test(flavor = "multi_thread")]
744         async fn basic_threaded_connection_test() {
745                 do_basic_connection_test().await;
746         }
747
748         #[tokio::test]
749         async fn basic_unthreaded_connection_test() {
750                 do_basic_connection_test().await;
751         }
752
753         async fn race_disconnect_accept() {
754                 // Previously, if we handed an already-disconnected socket to `setup_inbound` we'd panic.
755                 // This attempts to find other similar races by opening connections and shutting them down
756                 // while connecting. Sadly in testing this did *not* reproduce the previous issue.
757                 let secp_ctx = Secp256k1::new();
758                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
759                 let b_key = SecretKey::from_slice(&[2; 32]).unwrap();
760                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
761
762                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
763                         chan_handler: Arc::new(lightning::ln::peer_handler::ErroringMessageHandler::new()),
764                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
765                         route_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
766                         custom_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
767                 }, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(TestNodeSigner::new(a_key))));
768
769                 // Make two connections, one for an inbound and one for an outbound connection
770                 let conn_a = {
771                         let (conn_a, _) = make_tcp_connection();
772                         conn_a
773                 };
774                 let conn_b = {
775                         let (_, conn_b) = make_tcp_connection();
776                         conn_b
777                 };
778
779                 // Call connection setup inside new tokio tasks.
780                 let manager_reference = Arc::clone(&a_manager);
781                 tokio::spawn(async move {
782                         super::setup_inbound(manager_reference, conn_a).await
783                 });
784                 tokio::spawn(async move {
785                         super::setup_outbound(a_manager, b_pub, conn_b).await
786                 });
787         }
788
789         #[tokio::test(flavor = "multi_thread")]
790         async fn threaded_race_disconnect_accept() {
791                 race_disconnect_accept().await;
792         }
793
794         #[tokio::test]
795         async fn unthreaded_race_disconnect_accept() {
796                 race_disconnect_accept().await;
797         }
798 }