Merge pull request #2641 from alexanderwiederin/2585-preflight-test-coverage
[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 // Prefix these with `rustdoc::` when we update our MSRV to be >= 1.52 to remove warnings.
26 #![deny(broken_intra_doc_links)]
27 #![deny(private_intra_doc_links)]
28
29 #![deny(missing_docs)]
30 #![cfg_attr(docsrs, feature(doc_auto_cfg))]
31
32 use bitcoin::secp256k1::PublicKey;
33
34 use tokio::net::TcpStream;
35 use tokio::time;
36 use tokio::sync::mpsc;
37
38 use lightning::ln::peer_handler;
39 use lightning::ln::peer_handler::SocketDescriptor as LnSocketTrait;
40 use lightning::ln::peer_handler::APeerManager;
41 use lightning::ln::msgs::SocketAddress;
42
43 use std::ops::Deref;
44 use std::task::{self, Poll};
45 use std::future::Future;
46 use std::net::SocketAddr;
47 use std::net::TcpStream as StdTcpStream;
48 use std::sync::{Arc, Mutex};
49 use std::sync::atomic::{AtomicU64, Ordering};
50 use std::time::Duration;
51 use std::pin::Pin;
52 use std::hash::Hash;
53
54 static ID_COUNTER: AtomicU64 = AtomicU64::new(0);
55
56 // We only need to select over multiple futures in one place, and taking on the full `tokio/macros`
57 // dependency tree in order to do so (which has broken our MSRV before) is excessive. Instead, we
58 // define a trivial two- and three- select macro with the specific types we need and just use that.
59
60 pub(crate) enum SelectorOutput {
61         A(Option<()>), B(Option<()>), C(tokio::io::Result<()>),
62 }
63
64 pub(crate) struct TwoSelector<
65         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin
66 > {
67         pub a: A,
68         pub b: B,
69 }
70
71 impl<
72         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin
73 > Future for TwoSelector<A, B> {
74         type Output = SelectorOutput;
75         fn poll(mut self: Pin<&mut Self>, ctx: &mut task::Context<'_>) -> Poll<SelectorOutput> {
76                 match Pin::new(&mut self.a).poll(ctx) {
77                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::A(res)); },
78                         Poll::Pending => {},
79                 }
80                 match Pin::new(&mut self.b).poll(ctx) {
81                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::B(res)); },
82                         Poll::Pending => {},
83                 }
84                 Poll::Pending
85         }
86 }
87
88 pub(crate) struct ThreeSelector<
89         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin, C: Future<Output=tokio::io::Result<()>> + Unpin
90 > {
91         pub a: A,
92         pub b: B,
93         pub c: C,
94 }
95
96 impl<
97         A: Future<Output=Option<()>> + Unpin, B: Future<Output=Option<()>> + Unpin, C: Future<Output=tokio::io::Result<()>> + Unpin
98 > Future for ThreeSelector<A, B, C> {
99         type Output = SelectorOutput;
100         fn poll(mut self: Pin<&mut Self>, ctx: &mut task::Context<'_>) -> Poll<SelectorOutput> {
101                 match Pin::new(&mut self.a).poll(ctx) {
102                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::A(res)); },
103                         Poll::Pending => {},
104                 }
105                 match Pin::new(&mut self.b).poll(ctx) {
106                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::B(res)); },
107                         Poll::Pending => {},
108                 }
109                 match Pin::new(&mut self.c).poll(ctx) {
110                         Poll::Ready(res) => { return Poll::Ready(SelectorOutput::C(res)); },
111                         Poll::Pending => {},
112                 }
113                 Poll::Pending
114         }
115 }
116
117 /// Connection contains all our internal state for a connection - we hold a reference to the
118 /// Connection object (in an Arc<Mutex<>>) in each SocketDescriptor we create as well as in the
119 /// read future (which is returned by schedule_read).
120 struct Connection {
121         writer: Option<Arc<TcpStream>>,
122         // Because our PeerManager is templated by user-provided types, and we can't (as far as I can
123         // tell) have a const RawWakerVTable built out of templated functions, we need some indirection
124         // between being woken up with write-ready and calling PeerManager::write_buffer_space_avail.
125         // This provides that indirection, with a Sender which gets handed to the PeerManager Arc on
126         // the schedule_read stack.
127         //
128         // An alternative (likely more effecient) approach would involve creating a RawWakerVTable at
129         // runtime with functions templated by the Arc<PeerManager> type, calling
130         // write_buffer_space_avail directly from tokio's write wake, however doing so would require
131         // more unsafe voodo than I really feel like writing.
132         write_avail: mpsc::Sender<()>,
133         // When we are told by rust-lightning to pause read (because we have writes backing up), we do
134         // so by setting read_paused. At that point, the read task will stop reading bytes from the
135         // socket. To wake it up (without otherwise changing its state, we can push a value into this
136         // Sender.
137         read_waker: mpsc::Sender<()>,
138         read_paused: bool,
139         rl_requested_disconnect: bool,
140         id: u64,
141 }
142 impl Connection {
143         async fn poll_event_process<PM: Deref + 'static + Send + Sync>(
144                 peer_manager: PM,
145                 mut event_receiver: mpsc::Receiver<()>,
146         ) where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
147                 loop {
148                         if event_receiver.recv().await.is_none() {
149                                 return;
150                         }
151                         peer_manager.as_ref().process_events();
152                 }
153         }
154
155         async fn schedule_read<PM: Deref + 'static + Send + Sync + Clone>(
156                 peer_manager: PM,
157                 us: Arc<Mutex<Self>>,
158                 reader: Arc<TcpStream>,
159                 mut read_wake_receiver: mpsc::Receiver<()>,
160                 mut write_avail_receiver: mpsc::Receiver<()>,
161         ) where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
162                 // Create a waker to wake up poll_event_process, above
163                 let (event_waker, event_receiver) = mpsc::channel(1);
164                 tokio::spawn(Self::poll_event_process(peer_manager.clone(), event_receiver));
165
166                 // 4KiB is nice and big without handling too many messages all at once, giving other peers
167                 // a chance to do some work.
168                 let mut buf = [0; 4096];
169
170                 let mut our_descriptor = SocketDescriptor::new(us.clone());
171                 // An enum describing why we did/are disconnecting:
172                 enum Disconnect {
173                         // Rust-Lightning told us to disconnect, either by returning an Err or by calling
174                         // SocketDescriptor::disconnect_socket.
175                         // In this case, we do not call peer_manager.socket_disconnected() as Rust-Lightning
176                         // already knows we're disconnected.
177                         CloseConnection,
178                         // The connection was disconnected for some other reason, ie because the socket was
179                         // closed.
180                         // In this case, we do need to call peer_manager.socket_disconnected() to inform
181                         // Rust-Lightning that the socket is gone.
182                         PeerDisconnected
183                 }
184                 let disconnect_type = loop {
185                         let read_paused = {
186                                 let us_lock = us.lock().unwrap();
187                                 if us_lock.rl_requested_disconnect {
188                                         break Disconnect::CloseConnection;
189                                 }
190                                 us_lock.read_paused
191                         };
192                         // TODO: Drop the Box'ing of the futures once Rust has pin-on-stack support.
193                         let select_result = if read_paused {
194                                 TwoSelector {
195                                         a: Box::pin(write_avail_receiver.recv()),
196                                         b: Box::pin(read_wake_receiver.recv()),
197                                 }.await
198                         } else {
199                                 ThreeSelector {
200                                         a: Box::pin(write_avail_receiver.recv()),
201                                         b: Box::pin(read_wake_receiver.recv()),
202                                         c: Box::pin(reader.readable()),
203                                 }.await
204                         };
205                         match select_result {
206                                 SelectorOutput::A(v) => {
207                                         assert!(v.is_some()); // We can't have dropped the sending end, its in the us Arc!
208                                         if peer_manager.as_ref().write_buffer_space_avail(&mut our_descriptor).is_err() {
209                                                 break Disconnect::CloseConnection;
210                                         }
211                                 },
212                                 SelectorOutput::B(_) => {},
213                                 SelectorOutput::C(res) => {
214                                         if res.is_err() { break Disconnect::PeerDisconnected; }
215                                         match reader.try_read(&mut buf) {
216                                                 Ok(0) => break Disconnect::PeerDisconnected,
217                                                 Ok(len) => {
218                                                         let read_res = peer_manager.as_ref().read_event(&mut our_descriptor, &buf[0..len]);
219                                                         let mut us_lock = us.lock().unwrap();
220                                                         match read_res {
221                                                                 Ok(pause_read) => {
222                                                                         if pause_read {
223                                                                                 us_lock.read_paused = true;
224                                                                         }
225                                                                 },
226                                                                 Err(_) => break Disconnect::CloseConnection,
227                                                         }
228                                                 },
229                                                 Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => {
230                                                         // readable() is allowed to spuriously wake, so we have to handle
231                                                         // WouldBlock here.
232                                                 },
233                                                 Err(_) => break Disconnect::PeerDisconnected,
234                                         }
235                                 },
236                         }
237                         let _ = event_waker.try_send(());
238
239                         // At this point we've processed a message or two, and reset the ping timer for this
240                         // peer, at least in the "are we still receiving messages" context, if we don't give up
241                         // our timeslice to another task we may just spin on this peer, starving other peers
242                         // and eventually disconnecting them for ping timeouts. Instead, we explicitly yield
243                         // here.
244                         let _ = tokio::task::yield_now().await;
245                 };
246                 us.lock().unwrap().writer.take();
247                 if let Disconnect::PeerDisconnected = disconnect_type {
248                         peer_manager.as_ref().socket_disconnected(&our_descriptor);
249                         peer_manager.as_ref().process_events();
250                 }
251         }
252
253         fn new(stream: StdTcpStream) -> (Arc<TcpStream>, mpsc::Receiver<()>, mpsc::Receiver<()>, Arc<Mutex<Self>>) {
254                 // We only ever need a channel of depth 1 here: if we returned a non-full write to the
255                 // PeerManager, we will eventually get notified that there is room in the socket to write
256                 // new bytes, which will generate an event. That event will be popped off the queue before
257                 // we call write_buffer_space_avail, ensuring that we have room to push a new () if, during
258                 // the write_buffer_space_avail() call, send_data() returns a non-full write.
259                 let (write_avail, write_receiver) = mpsc::channel(1);
260                 // Similarly here - our only goal is to make sure the reader wakes up at some point after
261                 // we shove a value into the channel which comes after we've reset the read_paused bool to
262                 // false.
263                 let (read_waker, read_receiver) = mpsc::channel(1);
264                 stream.set_nonblocking(true).unwrap();
265                 let tokio_stream = Arc::new(TcpStream::from_std(stream).unwrap());
266
267                 (Arc::clone(&tokio_stream), write_receiver, read_receiver,
268                 Arc::new(Mutex::new(Self {
269                         writer: Some(tokio_stream), write_avail, read_waker, read_paused: false,
270                         rl_requested_disconnect: false,
271                         id: ID_COUNTER.fetch_add(1, Ordering::AcqRel)
272                 })))
273         }
274 }
275
276 fn get_addr_from_stream(stream: &StdTcpStream) -> Option<SocketAddress> {
277         match stream.peer_addr() {
278                 Ok(SocketAddr::V4(sockaddr)) => Some(SocketAddress::TcpIpV4 {
279                         addr: sockaddr.ip().octets(),
280                         port: sockaddr.port(),
281                 }),
282                 Ok(SocketAddr::V6(sockaddr)) => Some(SocketAddress::TcpIpV6 {
283                         addr: sockaddr.ip().octets(),
284                         port: sockaddr.port(),
285                 }),
286                 Err(_) => None,
287         }
288 }
289
290 /// Process incoming messages and feed outgoing messages on the provided socket generated by
291 /// accepting an incoming connection.
292 ///
293 /// The returned future will complete when the peer is disconnected and associated handling
294 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
295 /// not need to poll the provided future in order to make progress.
296 pub fn setup_inbound<PM: Deref + 'static + Send + Sync + Clone>(
297         peer_manager: PM,
298         stream: StdTcpStream,
299 ) -> impl std::future::Future<Output=()>
300 where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
301         let remote_addr = get_addr_from_stream(&stream);
302         let (reader, write_receiver, read_receiver, us) = Connection::new(stream);
303         #[cfg(test)]
304         let last_us = Arc::clone(&us);
305
306         let handle_opt = if peer_manager.as_ref().new_inbound_connection(SocketDescriptor::new(us.clone()), remote_addr).is_ok() {
307                 Some(tokio::spawn(Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver)))
308         } else {
309                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
310                 // requirements.
311                 None
312         };
313
314         async move {
315                 if let Some(handle) = handle_opt {
316                         if let Err(e) = handle.await {
317                                 assert!(e.is_cancelled());
318                         } else {
319                                 // This is certainly not guaranteed to always be true - the read loop may exit
320                                 // while there are still pending write wakers that need to be woken up after the
321                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
322                                 // keep too many wakers around, this makes sense. The race should be rare (we do
323                                 // some work after shutdown()) and an error would be a major memory leak.
324                                 #[cfg(test)]
325                                 debug_assert!(Arc::try_unwrap(last_us).is_ok());
326                         }
327                 }
328         }
329 }
330
331 /// Process incoming messages and feed outgoing messages on the provided socket generated by
332 /// making an outbound connection which is expected to be accepted by a peer with the given
333 /// public key. The relevant processing is set to run free (via tokio::spawn).
334 ///
335 /// The returned future will complete when the peer is disconnected and associated handling
336 /// futures are freed, though, because all processing futures are spawned with tokio::spawn, you do
337 /// not need to poll the provided future in order to make progress.
338 pub fn setup_outbound<PM: Deref + 'static + Send + Sync + Clone>(
339         peer_manager: PM,
340         their_node_id: PublicKey,
341         stream: StdTcpStream,
342 ) -> impl std::future::Future<Output=()>
343 where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
344         let remote_addr = get_addr_from_stream(&stream);
345         let (reader, mut write_receiver, read_receiver, us) = Connection::new(stream);
346         #[cfg(test)]
347         let last_us = Arc::clone(&us);
348         let handle_opt = if let Ok(initial_send) = peer_manager.as_ref().new_outbound_connection(their_node_id, SocketDescriptor::new(us.clone()), remote_addr) {
349                 Some(tokio::spawn(async move {
350                         // We should essentially always have enough room in a TCP socket buffer to send the
351                         // initial 10s of bytes. However, tokio running in single-threaded mode will always
352                         // fail writes and wake us back up later to write. Thus, we handle a single
353                         // std::task::Poll::Pending but still expect to write the full set of bytes at once
354                         // and use a relatively tight timeout.
355                         if let Ok(Ok(())) = tokio::time::timeout(Duration::from_millis(100), async {
356                                 loop {
357                                         match SocketDescriptor::new(us.clone()).send_data(&initial_send, true) {
358                                                 v if v == initial_send.len() => break Ok(()),
359                                                 0 => {
360                                                         write_receiver.recv().await;
361                                                         // In theory we could check for if we've been instructed to disconnect
362                                                         // the peer here, but its OK to just skip it - we'll check for it in
363                                                         // schedule_read prior to any relevant calls into RL.
364                                                 },
365                                                 _ => {
366                                                         eprintln!("Failed to write first full message to socket!");
367                                                         peer_manager.as_ref().socket_disconnected(&SocketDescriptor::new(Arc::clone(&us)));
368                                                         break Err(());
369                                                 }
370                                         }
371                                 }
372                         }).await {
373                                 Connection::schedule_read(peer_manager, us, reader, read_receiver, write_receiver).await;
374                         }
375                 }))
376         } else {
377                 // Note that we will skip socket_disconnected here, in accordance with the PeerManager
378                 // requirements.
379                 None
380         };
381
382         async move {
383                 if let Some(handle) = handle_opt {
384                         if let Err(e) = handle.await {
385                                 assert!(e.is_cancelled());
386                         } else {
387                                 // This is certainly not guaranteed to always be true - the read loop may exit
388                                 // while there are still pending write wakers that need to be woken up after the
389                                 // socket shutdown(). Still, as a check during testing, to make sure tokio doesn't
390                                 // keep too many wakers around, this makes sense. The race should be rare (we do
391                                 // some work after shutdown()) and an error would be a major memory leak.
392                                 #[cfg(test)]
393                                 debug_assert!(Arc::try_unwrap(last_us).is_ok());
394                         }
395                 }
396         }
397 }
398
399 /// Process incoming messages and feed outgoing messages on a new connection made to the given
400 /// socket address which is expected to be accepted by a peer with the given public key (by
401 /// scheduling futures with tokio::spawn).
402 ///
403 /// Shorthand for TcpStream::connect(addr) with a timeout followed by setup_outbound().
404 ///
405 /// Returns a future (as the fn is async) which needs to be polled to complete the connection and
406 /// connection setup. That future then returns a future which will complete when the peer is
407 /// disconnected and associated handling futures are freed, though, because all processing in said
408 /// futures are spawned with tokio::spawn, you do not need to poll the second future in order to
409 /// make progress.
410 pub async fn connect_outbound<PM: Deref + 'static + Send + Sync + Clone>(
411         peer_manager: PM,
412         their_node_id: PublicKey,
413         addr: SocketAddr,
414 ) -> Option<impl std::future::Future<Output=()>>
415 where PM::Target: APeerManager<Descriptor = SocketDescriptor> {
416         if let Ok(Ok(stream)) = time::timeout(Duration::from_secs(10), async { TcpStream::connect(&addr).await.map(|s| s.into_std().unwrap()) }).await {
417                 Some(setup_outbound(peer_manager, their_node_id, stream))
418         } else { None }
419 }
420
421 const SOCK_WAKER_VTABLE: task::RawWakerVTable =
422         task::RawWakerVTable::new(clone_socket_waker, wake_socket_waker, wake_socket_waker_by_ref, drop_socket_waker);
423
424 fn clone_socket_waker(orig_ptr: *const ()) -> task::RawWaker {
425         write_avail_to_waker(orig_ptr as *const mpsc::Sender<()>)
426 }
427 // When waking, an error should be fine. Most likely we got two send_datas in a row, both of which
428 // failed to fully write, but we only need to call write_buffer_space_avail() once. Otherwise, the
429 // sending thread may have already gone away due to a socket close, in which case there's nothing
430 // to wake up anyway.
431 fn wake_socket_waker(orig_ptr: *const ()) {
432         let sender = unsafe { &mut *(orig_ptr as *mut mpsc::Sender<()>) };
433         let _ = sender.try_send(());
434         drop_socket_waker(orig_ptr);
435 }
436 fn wake_socket_waker_by_ref(orig_ptr: *const ()) {
437         let sender_ptr = orig_ptr as *const mpsc::Sender<()>;
438         let sender = unsafe { (*sender_ptr).clone() };
439         let _ = sender.try_send(());
440 }
441 fn drop_socket_waker(orig_ptr: *const ()) {
442         let _orig_box = unsafe { Box::from_raw(orig_ptr as *mut mpsc::Sender<()>) };
443         // _orig_box is now dropped
444 }
445 fn write_avail_to_waker(sender: *const mpsc::Sender<()>) -> task::RawWaker {
446         let new_box = Box::leak(Box::new(unsafe { (*sender).clone() }));
447         let new_ptr = new_box as *const mpsc::Sender<()>;
448         task::RawWaker::new(new_ptr as *const (), &SOCK_WAKER_VTABLE)
449 }
450
451 /// The SocketDescriptor used to refer to sockets by a PeerHandler. This is pub only as it is a
452 /// type in the template of PeerHandler.
453 pub struct SocketDescriptor {
454         conn: Arc<Mutex<Connection>>,
455         id: u64,
456 }
457 impl SocketDescriptor {
458         fn new(conn: Arc<Mutex<Connection>>) -> Self {
459                 let id = conn.lock().unwrap().id;
460                 Self { conn, id }
461         }
462 }
463 impl peer_handler::SocketDescriptor for SocketDescriptor {
464         fn send_data(&mut self, data: &[u8], resume_read: bool) -> usize {
465                 // To send data, we take a lock on our Connection to access the TcpStream, writing to it if
466                 // there's room in the kernel buffer, or otherwise create a new Waker with a
467                 // SocketDescriptor in it which can wake up the write_avail Sender, waking up the
468                 // processing future which will call write_buffer_space_avail and we'll end up back here.
469                 let mut us = self.conn.lock().unwrap();
470                 if us.writer.is_none() {
471                         // The writer gets take()n when it is time to shut down, so just fast-return 0 here.
472                         return 0;
473                 }
474
475                 if resume_read && us.read_paused {
476                         // The schedule_read future may go to lock up but end up getting woken up by there
477                         // being more room in the write buffer, dropping the other end of this Sender
478                         // before we get here, so we ignore any failures to wake it up.
479                         us.read_paused = false;
480                         let _ = us.read_waker.try_send(());
481                 }
482                 if data.is_empty() { return 0; }
483                 let waker = unsafe { task::Waker::from_raw(write_avail_to_waker(&us.write_avail)) };
484                 let mut ctx = task::Context::from_waker(&waker);
485                 let mut written_len = 0;
486                 loop {
487                         match us.writer.as_ref().unwrap().poll_write_ready(&mut ctx) {
488                                 task::Poll::Ready(Ok(())) => {
489                                         match us.writer.as_ref().unwrap().try_write(&data[written_len..]) {
490                                                 Ok(res) => {
491                                                         debug_assert_ne!(res, 0);
492                                                         written_len += res;
493                                                         if written_len == data.len() { return written_len; }
494                                                 },
495                                                 Err(_) => return written_len,
496                                         }
497                                 },
498                                 task::Poll::Ready(Err(_)) => return written_len,
499                                 task::Poll::Pending => {
500                                         // We're queued up for a write event now, but we need to make sure we also
501                                         // pause read given we're now waiting on the remote end to ACK (and in
502                                         // accordance with the send_data() docs).
503                                         us.read_paused = true;
504                                         // Further, to avoid any current pending read causing a `read_event` call, wake
505                                         // up the read_waker and restart its loop.
506                                         let _ = us.read_waker.try_send(());
507                                         return written_len;
508                                 },
509                         }
510                 }
511         }
512
513         fn disconnect_socket(&mut self) {
514                 let mut us = self.conn.lock().unwrap();
515                 us.rl_requested_disconnect = true;
516                 // Wake up the sending thread, assuming it is still alive
517                 let _ = us.write_avail.try_send(());
518         }
519 }
520 impl Clone for SocketDescriptor {
521         fn clone(&self) -> Self {
522                 Self {
523                         conn: Arc::clone(&self.conn),
524                         id: self.id,
525                 }
526         }
527 }
528 impl Eq for SocketDescriptor {}
529 impl PartialEq for SocketDescriptor {
530         fn eq(&self, o: &Self) -> bool {
531                 self.id == o.id
532         }
533 }
534 impl Hash for SocketDescriptor {
535         fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
536                 self.id.hash(state);
537         }
538 }
539
540 #[cfg(test)]
541 mod tests {
542         use lightning::ln::features::*;
543         use lightning::ln::msgs::*;
544         use lightning::ln::peer_handler::{MessageHandler, PeerManager};
545         use lightning::ln::features::NodeFeatures;
546         use lightning::routing::gossip::NodeId;
547         use lightning::events::*;
548         use lightning::util::test_utils::TestNodeSigner;
549         use bitcoin::Network;
550         use bitcoin::blockdata::constants::ChainHash;
551         use bitcoin::secp256k1::{Secp256k1, SecretKey, PublicKey};
552
553         use tokio::sync::mpsc;
554
555         use std::mem;
556         use std::sync::atomic::{AtomicBool, Ordering};
557         use std::sync::{Arc, Mutex};
558         use std::time::Duration;
559
560         pub struct TestLogger();
561         impl lightning::util::logger::Logger for TestLogger {
562                 fn log(&self, record: &lightning::util::logger::Record) {
563                         println!("{:<5} [{} : {}, {}] {}", record.level.to_string(), record.module_path, record.file, record.line, record.args);
564                 }
565         }
566
567         struct MsgHandler{
568                 expected_pubkey: PublicKey,
569                 pubkey_connected: mpsc::Sender<()>,
570                 pubkey_disconnected: mpsc::Sender<()>,
571                 disconnected_flag: AtomicBool,
572                 msg_events: Mutex<Vec<MessageSendEvent>>,
573         }
574         impl RoutingMessageHandler for MsgHandler {
575                 fn handle_node_announcement(&self, _msg: &NodeAnnouncement) -> Result<bool, LightningError> { Ok(false) }
576                 fn handle_channel_announcement(&self, _msg: &ChannelAnnouncement) -> Result<bool, LightningError> { Ok(false) }
577                 fn handle_channel_update(&self, _msg: &ChannelUpdate) -> Result<bool, LightningError> { Ok(false) }
578                 fn get_next_channel_announcement(&self, _starting_point: u64) -> Option<(ChannelAnnouncement, Option<ChannelUpdate>, Option<ChannelUpdate>)> { None }
579                 fn get_next_node_announcement(&self, _starting_point: Option<&NodeId>) -> Option<NodeAnnouncement> { None }
580                 fn peer_connected(&self, _their_node_id: &PublicKey, _init_msg: &Init, _inbound: bool) -> Result<(), ()> { Ok(()) }
581                 fn handle_reply_channel_range(&self, _their_node_id: &PublicKey, _msg: ReplyChannelRange) -> Result<(), LightningError> { Ok(()) }
582                 fn handle_reply_short_channel_ids_end(&self, _their_node_id: &PublicKey, _msg: ReplyShortChannelIdsEnd) -> Result<(), LightningError> { Ok(()) }
583                 fn handle_query_channel_range(&self, _their_node_id: &PublicKey, _msg: QueryChannelRange) -> Result<(), LightningError> { Ok(()) }
584                 fn handle_query_short_channel_ids(&self, _their_node_id: &PublicKey, _msg: QueryShortChannelIds) -> Result<(), LightningError> { Ok(()) }
585                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
586                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
587                 fn processing_queue_high(&self) -> bool { false }
588         }
589         impl ChannelMessageHandler for MsgHandler {
590                 fn handle_open_channel(&self, _their_node_id: &PublicKey, _msg: &OpenChannel) {}
591                 fn handle_accept_channel(&self, _their_node_id: &PublicKey, _msg: &AcceptChannel) {}
592                 fn handle_funding_created(&self, _their_node_id: &PublicKey, _msg: &FundingCreated) {}
593                 fn handle_funding_signed(&self, _their_node_id: &PublicKey, _msg: &FundingSigned) {}
594                 fn handle_channel_ready(&self, _their_node_id: &PublicKey, _msg: &ChannelReady) {}
595                 fn handle_shutdown(&self, _their_node_id: &PublicKey, _msg: &Shutdown) {}
596                 fn handle_closing_signed(&self, _their_node_id: &PublicKey, _msg: &ClosingSigned) {}
597                 fn handle_update_add_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateAddHTLC) {}
598                 fn handle_update_fulfill_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFulfillHTLC) {}
599                 fn handle_update_fail_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailHTLC) {}
600                 fn handle_update_fail_malformed_htlc(&self, _their_node_id: &PublicKey, _msg: &UpdateFailMalformedHTLC) {}
601                 fn handle_commitment_signed(&self, _their_node_id: &PublicKey, _msg: &CommitmentSigned) {}
602                 fn handle_revoke_and_ack(&self, _their_node_id: &PublicKey, _msg: &RevokeAndACK) {}
603                 fn handle_update_fee(&self, _their_node_id: &PublicKey, _msg: &UpdateFee) {}
604                 fn handle_announcement_signatures(&self, _their_node_id: &PublicKey, _msg: &AnnouncementSignatures) {}
605                 fn handle_channel_update(&self, _their_node_id: &PublicKey, _msg: &ChannelUpdate) {}
606                 fn handle_open_channel_v2(&self, _their_node_id: &PublicKey, _msg: &OpenChannelV2) {}
607                 fn handle_accept_channel_v2(&self, _their_node_id: &PublicKey, _msg: &AcceptChannelV2) {}
608                 fn handle_tx_add_input(&self, _their_node_id: &PublicKey, _msg: &TxAddInput) {}
609                 fn handle_tx_add_output(&self, _their_node_id: &PublicKey, _msg: &TxAddOutput) {}
610                 fn handle_tx_remove_input(&self, _their_node_id: &PublicKey, _msg: &TxRemoveInput) {}
611                 fn handle_tx_remove_output(&self, _their_node_id: &PublicKey, _msg: &TxRemoveOutput) {}
612                 fn handle_tx_complete(&self, _their_node_id: &PublicKey, _msg: &TxComplete) {}
613                 fn handle_tx_signatures(&self, _their_node_id: &PublicKey, _msg: &TxSignatures) {}
614                 fn handle_tx_init_rbf(&self, _their_node_id: &PublicKey, _msg: &TxInitRbf) {}
615                 fn handle_tx_ack_rbf(&self, _their_node_id: &PublicKey, _msg: &TxAckRbf) {}
616                 fn handle_tx_abort(&self, _their_node_id: &PublicKey, _msg: &TxAbort) {}
617                 fn peer_disconnected(&self, their_node_id: &PublicKey) {
618                         if *their_node_id == self.expected_pubkey {
619                                 self.disconnected_flag.store(true, Ordering::SeqCst);
620                                 self.pubkey_disconnected.clone().try_send(()).unwrap();
621                         }
622                 }
623                 fn peer_connected(&self, their_node_id: &PublicKey, _init_msg: &Init, _inbound: bool) -> Result<(), ()> {
624                         if *their_node_id == self.expected_pubkey {
625                                 self.pubkey_connected.clone().try_send(()).unwrap();
626                         }
627                         Ok(())
628                 }
629                 fn handle_channel_reestablish(&self, _their_node_id: &PublicKey, _msg: &ChannelReestablish) {}
630                 fn handle_error(&self, _their_node_id: &PublicKey, _msg: &ErrorMessage) {}
631                 fn provided_node_features(&self) -> NodeFeatures { NodeFeatures::empty() }
632                 fn provided_init_features(&self, _their_node_id: &PublicKey) -> InitFeatures { InitFeatures::empty() }
633                 fn get_chain_hashes(&self) -> Option<Vec<ChainHash>> {
634                         Some(vec![ChainHash::using_genesis_block(Network::Testnet)])
635                 }
636         }
637         impl MessageSendEventsProvider for MsgHandler {
638                 fn get_and_clear_pending_msg_events(&self) -> Vec<MessageSendEvent> {
639                         let mut ret = Vec::new();
640                         mem::swap(&mut *self.msg_events.lock().unwrap(), &mut ret);
641                         ret
642                 }
643         }
644
645         fn make_tcp_connection() -> (std::net::TcpStream, std::net::TcpStream) {
646                 if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9735") {
647                         (std::net::TcpStream::connect("127.0.0.1:9735").unwrap(), listener.accept().unwrap().0)
648                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:19735") {
649                         (std::net::TcpStream::connect("127.0.0.1:19735").unwrap(), listener.accept().unwrap().0)
650                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9997") {
651                         (std::net::TcpStream::connect("127.0.0.1:9997").unwrap(), listener.accept().unwrap().0)
652                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9998") {
653                         (std::net::TcpStream::connect("127.0.0.1:9998").unwrap(), listener.accept().unwrap().0)
654                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:9999") {
655                         (std::net::TcpStream::connect("127.0.0.1:9999").unwrap(), listener.accept().unwrap().0)
656                 } else if let Ok(listener) = std::net::TcpListener::bind("127.0.0.1:46926") {
657                         (std::net::TcpStream::connect("127.0.0.1:46926").unwrap(), listener.accept().unwrap().0)
658                 } else { panic!("Failed to bind to v4 localhost on common ports"); }
659         }
660
661         async fn do_basic_connection_test() {
662                 let secp_ctx = Secp256k1::new();
663                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
664                 let b_key = SecretKey::from_slice(&[1; 32]).unwrap();
665                 let a_pub = PublicKey::from_secret_key(&secp_ctx, &a_key);
666                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
667
668                 let (a_connected_sender, mut a_connected) = mpsc::channel(1);
669                 let (a_disconnected_sender, mut a_disconnected) = mpsc::channel(1);
670                 let a_handler = Arc::new(MsgHandler {
671                         expected_pubkey: b_pub,
672                         pubkey_connected: a_connected_sender,
673                         pubkey_disconnected: a_disconnected_sender,
674                         disconnected_flag: AtomicBool::new(false),
675                         msg_events: Mutex::new(Vec::new()),
676                 });
677                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
678                         chan_handler: Arc::clone(&a_handler),
679                         route_handler: Arc::clone(&a_handler),
680                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
681                         custom_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
682                 }, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(TestNodeSigner::new(a_key))));
683
684                 let (b_connected_sender, mut b_connected) = mpsc::channel(1);
685                 let (b_disconnected_sender, mut b_disconnected) = mpsc::channel(1);
686                 let b_handler = Arc::new(MsgHandler {
687                         expected_pubkey: a_pub,
688                         pubkey_connected: b_connected_sender,
689                         pubkey_disconnected: b_disconnected_sender,
690                         disconnected_flag: AtomicBool::new(false),
691                         msg_events: Mutex::new(Vec::new()),
692                 });
693                 let b_manager = Arc::new(PeerManager::new(MessageHandler {
694                         chan_handler: Arc::clone(&b_handler),
695                         route_handler: Arc::clone(&b_handler),
696                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
697                         custom_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
698                 }, 0, &[2; 32], Arc::new(TestLogger()), Arc::new(TestNodeSigner::new(b_key))));
699
700                 // We bind on localhost, hoping the environment is properly configured with a local
701                 // address. This may not always be the case in containers and the like, so if this test is
702                 // failing for you check that you have a loopback interface and it is configured with
703                 // 127.0.0.1.
704                 let (conn_a, conn_b) = make_tcp_connection();
705
706                 let fut_a = super::setup_outbound(Arc::clone(&a_manager), b_pub, conn_a);
707                 let fut_b = super::setup_inbound(b_manager, conn_b);
708
709                 tokio::time::timeout(Duration::from_secs(10), a_connected.recv()).await.unwrap();
710                 tokio::time::timeout(Duration::from_secs(1), b_connected.recv()).await.unwrap();
711
712                 a_handler.msg_events.lock().unwrap().push(MessageSendEvent::HandleError {
713                         node_id: b_pub, action: ErrorAction::DisconnectPeer { msg: None }
714                 });
715                 assert!(!a_handler.disconnected_flag.load(Ordering::SeqCst));
716                 assert!(!b_handler.disconnected_flag.load(Ordering::SeqCst));
717
718                 a_manager.process_events();
719                 tokio::time::timeout(Duration::from_secs(10), a_disconnected.recv()).await.unwrap();
720                 tokio::time::timeout(Duration::from_secs(1), b_disconnected.recv()).await.unwrap();
721                 assert!(a_handler.disconnected_flag.load(Ordering::SeqCst));
722                 assert!(b_handler.disconnected_flag.load(Ordering::SeqCst));
723
724                 fut_a.await;
725                 fut_b.await;
726         }
727
728         #[tokio::test(flavor = "multi_thread")]
729         async fn basic_threaded_connection_test() {
730                 do_basic_connection_test().await;
731         }
732
733         #[tokio::test]
734         async fn basic_unthreaded_connection_test() {
735                 do_basic_connection_test().await;
736         }
737
738         async fn race_disconnect_accept() {
739                 // Previously, if we handed an already-disconnected socket to `setup_inbound` we'd panic.
740                 // This attempts to find other similar races by opening connections and shutting them down
741                 // while connecting. Sadly in testing this did *not* reproduce the previous issue.
742                 let secp_ctx = Secp256k1::new();
743                 let a_key = SecretKey::from_slice(&[1; 32]).unwrap();
744                 let b_key = SecretKey::from_slice(&[2; 32]).unwrap();
745                 let b_pub = PublicKey::from_secret_key(&secp_ctx, &b_key);
746
747                 let a_manager = Arc::new(PeerManager::new(MessageHandler {
748                         chan_handler: Arc::new(lightning::ln::peer_handler::ErroringMessageHandler::new()),
749                         onion_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
750                         route_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
751                         custom_message_handler: Arc::new(lightning::ln::peer_handler::IgnoringMessageHandler{}),
752                 }, 0, &[1; 32], Arc::new(TestLogger()), Arc::new(TestNodeSigner::new(a_key))));
753
754                 // Make two connections, one for an inbound and one for an outbound connection
755                 let conn_a = {
756                         let (conn_a, _) = make_tcp_connection();
757                         conn_a
758                 };
759                 let conn_b = {
760                         let (_, conn_b) = make_tcp_connection();
761                         conn_b
762                 };
763
764                 // Call connection setup inside new tokio tasks.
765                 let manager_reference = Arc::clone(&a_manager);
766                 tokio::spawn(async move {
767                         super::setup_inbound(manager_reference, conn_a).await
768                 });
769                 tokio::spawn(async move {
770                         super::setup_outbound(a_manager, b_pub, conn_b).await
771                 });
772         }
773
774         #[tokio::test(flavor = "multi_thread")]
775         async fn threaded_race_disconnect_accept() {
776                 race_disconnect_accept().await;
777         }
778
779         #[tokio::test]
780         async fn unthreaded_race_disconnect_accept() {
781                 race_disconnect_accept().await;
782         }
783 }