tet_libp2p_ping/handler.rs
1// Copyright 2019 Parity Technologies (UK) Ltd.
2//
3// Permission is hereby granted, free of charge, to any person obtaining a
4// copy of this software and associated documentation files (the "Software"),
5// to deal in the Software without restriction, including without limitation
6// the rights to use, copy, modify, merge, publish, distribute, sublicense,
7// and/or sell copies of the Software, and to permit persons to whom the
8// Software is furnished to do so, subject to the following conditions:
9//
10// The above copyright notice and this permission notice shall be included in
11// all copies or substantial portions of the Software.
12//
13// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
14// OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
18// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
19// DEALINGS IN THE SOFTWARE.
20
21use crate::protocol;
22use futures::prelude::*;
23use futures::future::BoxFuture;
24use tet_libp2p_swarm::{
25 KeepAlive,
26 NegotiatedSubstream,
27 SubstreamProtocol,
28 ProtocolsHandler,
29 ProtocolsHandlerUpgrErr,
30 ProtocolsHandlerEvent
31};
32use std::{
33 error::Error,
34 io,
35 fmt,
36 num::NonZeroU32,
37 task::{Context, Poll},
38 time::Duration
39};
40use std::collections::VecDeque;
41use wasm_timer::Delay;
42use void::Void;
43
44/// The configuration for outbound pings.
45#[derive(Clone, Debug)]
46pub struct PingConfig {
47 /// The timeout of an outbound ping.
48 timeout: Duration,
49 /// The duration between the last successful outbound or inbound ping
50 /// and the next outbound ping.
51 interval: Duration,
52 /// The maximum number of failed outbound pings before the associated
53 /// connection is deemed unhealthy, indicating to the `Swarm` that it
54 /// should be closed.
55 max_failures: NonZeroU32,
56 /// Whether the connection should generally be kept alive unless
57 /// `max_failures` occur.
58 keep_alive: bool,
59}
60
61impl PingConfig {
62 /// Creates a new `PingConfig` with the following default settings:
63 ///
64 /// * [`PingConfig::with_interval`] 15s
65 /// * [`PingConfig::with_timeout`] 20s
66 /// * [`PingConfig::with_max_failures`] 1
67 /// * [`PingConfig::with_keep_alive`] false
68 ///
69 /// These settings have the following effect:
70 ///
71 /// * A ping is sent every 15 seconds on a healthy connection.
72 /// * Every ping sent must yield a response within 20 seconds in order to
73 /// be successful.
74 /// * A single ping failure is sufficient for the connection to be subject
75 /// to being closed.
76 /// * The connection may be closed at any time as far as the ping protocol
77 /// is concerned, i.e. the ping protocol itself does not keep the
78 /// connection alive.
79 pub fn new() -> Self {
80 Self {
81 timeout: Duration::from_secs(20),
82 interval: Duration::from_secs(15),
83 max_failures: NonZeroU32::new(1).expect("1 != 0"),
84 keep_alive: false
85 }
86 }
87
88 /// Sets the ping timeout.
89 pub fn with_timeout(mut self, d: Duration) -> Self {
90 self.timeout = d;
91 self
92 }
93
94 /// Sets the ping interval.
95 pub fn with_interval(mut self, d: Duration) -> Self {
96 self.interval = d;
97 self
98 }
99
100 /// Sets the maximum number of consecutive ping failures upon which the remote
101 /// peer is considered unreachable and the connection closed.
102 pub fn with_max_failures(mut self, n: NonZeroU32) -> Self {
103 self.max_failures = n;
104 self
105 }
106
107 /// Sets whether the ping protocol itself should keep the connection alive,
108 /// apart from the maximum allowed failures.
109 ///
110 /// By default, the ping protocol itself allows the connection to be closed
111 /// at any time, i.e. in the absence of ping failures the connection lifetime
112 /// is determined by other protocol handlers.
113 ///
114 /// If the maximum number of allowed ping failures is reached, the
115 /// connection is always terminated as a result of [`ProtocolsHandler::poll`]
116 /// returning an error, regardless of the keep-alive setting.
117 pub fn with_keep_alive(mut self, b: bool) -> Self {
118 self.keep_alive = b;
119 self
120 }
121}
122
123/// The result of an inbound or outbound ping.
124pub type PingResult = Result<PingSuccess, PingFailure>;
125
126/// The successful result of processing an inbound or outbound ping.
127#[derive(Debug)]
128pub enum PingSuccess {
129 /// Received a ping and sent back a pong.
130 Pong,
131 /// Sent a ping and received back a pong.
132 ///
133 /// Includes the round-trip time.
134 Ping { rtt: Duration },
135}
136
137/// An outbound ping failure.
138#[derive(Debug)]
139pub enum PingFailure {
140 /// The ping timed out, i.e. no response was received within the
141 /// configured ping timeout.
142 Timeout,
143 /// The ping failed for reasons other than a timeout.
144 Other { error: Box<dyn std::error::Error + Send + 'static> }
145}
146
147impl fmt::Display for PingFailure {
148 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
149 match self {
150 PingFailure::Timeout => f.write_str("Ping timeout"),
151 PingFailure::Other { error } => write!(f, "Ping error: {}", error)
152 }
153 }
154}
155
156impl Error for PingFailure {
157 fn source(&self) -> Option<&(dyn Error + 'static)> {
158 match self {
159 PingFailure::Timeout => None,
160 PingFailure::Other { error } => Some(&**error)
161 }
162 }
163}
164
165/// Protocol handler that handles pinging the remote at a regular period
166/// and answering ping queries.
167///
168/// If the remote doesn't respond, produces an error that closes the connection.
169pub struct PingHandler {
170 /// Configuration options.
171 config: PingConfig,
172 /// The timer used for the delay to the next ping as well as
173 /// the ping timeout.
174 timer: Delay,
175 /// Outbound ping failures that are pending to be processed by `poll()`.
176 pending_errors: VecDeque<PingFailure>,
177 /// The number of consecutive ping failures that occurred.
178 ///
179 /// Each successful ping resets this counter to 0.
180 failures: u32,
181 /// The outbound ping state.
182 outbound: Option<PingState>,
183 /// The inbound pong handler, i.e. if there is an inbound
184 /// substream, this is always a future that waits for the
185 /// next inbound ping to be answered.
186 inbound: Option<PongFuture>,
187}
188
189impl PingHandler {
190 /// Builds a new `PingHandler` with the given configuration.
191 pub fn new(config: PingConfig) -> Self {
192 PingHandler {
193 config,
194 timer: Delay::new(Duration::new(0, 0)),
195 pending_errors: VecDeque::with_capacity(2),
196 failures: 0,
197 outbound: None,
198 inbound: None,
199 }
200 }
201}
202
203impl ProtocolsHandler for PingHandler {
204 type InEvent = Void;
205 type OutEvent = PingResult;
206 type Error = PingFailure;
207 type InboundProtocol = protocol::Ping;
208 type OutboundProtocol = protocol::Ping;
209 type OutboundOpenInfo = ();
210 type InboundOpenInfo = ();
211
212 fn listen_protocol(&self) -> SubstreamProtocol<protocol::Ping, ()> {
213 SubstreamProtocol::new(protocol::Ping, ())
214 }
215
216 fn inject_fully_negotiated_inbound(&mut self, stream: NegotiatedSubstream, (): ()) {
217 self.inbound = Some(protocol::recv_ping(stream).boxed());
218 }
219
220 fn inject_fully_negotiated_outbound(&mut self, stream: NegotiatedSubstream, (): ()) {
221 self.timer.reset(self.config.timeout);
222 self.outbound = Some(PingState::Ping(protocol::send_ping(stream).boxed()));
223 }
224
225 fn inject_event(&mut self, _: Void) {}
226
227 fn inject_dial_upgrade_error(&mut self, _info: (), error: ProtocolsHandlerUpgrErr<Void>) {
228 self.outbound = None; // Request a new substream on the next `poll`.
229 self.pending_errors.push_front(
230 match error {
231 // Note: This timeout only covers protocol negotiation.
232 ProtocolsHandlerUpgrErr::Timeout => PingFailure::Timeout,
233 e => PingFailure::Other { error: Box::new(e) },
234 })
235 }
236
237 fn connection_keep_alive(&self) -> KeepAlive {
238 if self.config.keep_alive {
239 KeepAlive::Yes
240 } else {
241 KeepAlive::No
242 }
243 }
244
245 fn poll(&mut self, cx: &mut Context<'_>) -> Poll<ProtocolsHandlerEvent<protocol::Ping, (), PingResult, Self::Error>> {
246 // Respond to inbound pings.
247 if let Some(fut) = self.inbound.as_mut() {
248 match fut.poll_unpin(cx) {
249 Poll::Pending => {},
250 Poll::Ready(Err(e)) => {
251 log::debug!("Inbound ping error: {:?}", e);
252 self.inbound = None;
253 }
254 Poll::Ready(Ok(stream)) => {
255 // A ping from a remote peer has been answered, wait for the next.
256 self.inbound = Some(protocol::recv_ping(stream).boxed());
257 return Poll::Ready(ProtocolsHandlerEvent::Custom(Ok(PingSuccess::Pong)))
258 }
259 }
260 }
261
262 loop {
263 // Check for outbound ping failures.
264 if let Some(error) = self.pending_errors.pop_back() {
265 log::debug!("Ping failure: {:?}", error);
266
267 self.failures += 1;
268
269 // Note: For backward-compatibility, with configured
270 // `max_failures == 1`, the first failure is always "free"
271 // and silent. This allows peers who still use a new substream
272 // for each ping to have successful ping exchanges with peers
273 // that use a single substream, since every successful ping
274 // resets `failures` to `0`, while at the same time emitting
275 // events only for `max_failures - 1` failures, as before.
276 if self.failures > 1 || self.config.max_failures.get() > 1 {
277 if self.failures >= self.config.max_failures.get() {
278 log::debug!("Too many failures ({}). Closing connection.", self.failures);
279 return Poll::Ready(ProtocolsHandlerEvent::Close(error))
280 }
281
282 return Poll::Ready(ProtocolsHandlerEvent::Custom(Err(error)))
283 }
284 }
285
286 // Continue outbound pings.
287 match self.outbound.take() {
288 Some(PingState::Ping(mut ping)) => match ping.poll_unpin(cx) {
289 Poll::Pending => {
290 if self.timer.poll_unpin(cx).is_ready() {
291 self.pending_errors.push_front(PingFailure::Timeout);
292 } else {
293 self.outbound = Some(PingState::Ping(ping));
294 break
295 }
296 },
297 Poll::Ready(Ok((stream, rtt))) => {
298 self.failures = 0;
299 self.timer.reset(self.config.interval);
300 self.outbound = Some(PingState::Idle(stream));
301 return Poll::Ready(
302 ProtocolsHandlerEvent::Custom(
303 Ok(PingSuccess::Ping { rtt })))
304 }
305 Poll::Ready(Err(e)) => {
306 self.pending_errors.push_front(PingFailure::Other {
307 error: Box::new(e)
308 });
309 }
310 },
311 Some(PingState::Idle(stream)) => match self.timer.poll_unpin(cx) {
312 Poll::Pending => {
313 self.outbound = Some(PingState::Idle(stream));
314 break
315 },
316 Poll::Ready(Ok(())) => {
317 self.timer.reset(self.config.timeout);
318 self.outbound = Some(PingState::Ping(protocol::send_ping(stream).boxed()));
319 },
320 Poll::Ready(Err(e)) => {
321 return Poll::Ready(ProtocolsHandlerEvent::Close(
322 PingFailure::Other {
323 error: Box::new(e)
324 }))
325 }
326 }
327 Some(PingState::OpenStream) => {
328 self.outbound = Some(PingState::OpenStream);
329 break
330 }
331 None => {
332 self.outbound = Some(PingState::OpenStream);
333 let protocol = SubstreamProtocol::new(protocol::Ping, ())
334 .with_timeout(self.config.timeout);
335 return Poll::Ready(ProtocolsHandlerEvent::OutboundSubstreamRequest {
336 protocol
337 })
338 }
339 }
340 }
341
342 Poll::Pending
343 }
344}
345
346type PingFuture = BoxFuture<'static, Result<(NegotiatedSubstream, Duration), io::Error>>;
347type PongFuture = BoxFuture<'static, Result<NegotiatedSubstream, io::Error>>;
348
349/// The current state w.r.t. outbound pings.
350enum PingState {
351 /// A new substream is being negotiated for the ping protocol.
352 OpenStream,
353 /// The substream is idle, waiting to send the next ping.
354 Idle(NegotiatedSubstream),
355 /// A ping is being sent and the response awaited.
356 Ping(PingFuture),
357}
358