291 lines
10 KiB
Rust
Raw Normal View History

2017-11-22 10:58:06 +01:00
// Copyright 2017 Parity Technologies (UK) Ltd.
//
// Permission is hereby granted, free of charge, to any person obtaining a
// copy of this software and associated documentation files (the "Software"),
// to deal in the Software without restriction, including without limitation
// the rights to use, copy, modify, merge, publish, distribute, sublicense,
// and/or sell copies of the Software, and to permit persons to whom the
// Software is furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
// OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
// FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
// DEALINGS IN THE SOFTWARE.
//! Handles the `/ipfs/ping/1.0.0` protocol. This allows pinging a remote node and waiting for an
//! answer.
//!
//! # Usage
//!
//! Create a `Ping` struct, which implements the `ConnectionUpgrade` trait. When used as a
//! connection upgrade, it will produce a tuple of type `(Pinger, impl Future<Item = ()>)` which
//! are named the *pinger* and the *ponger*.
//!
//! The *pinger* has a method named `ping` which will send a ping to the remote, while the *ponger*
//! is a future that will process the data received on the socket and will be signalled only when
//! the connection closes.
//!
//! # About timeouts
//!
//! For technical reasons, this crate doesn't handle timeouts. The action of pinging returns a
//! future that is signalled only when the remote answers. If the remote is not responsive, the
//! future will never be signalled.
//!
//! For implementation reasons, resources allocated for a ping are only ever fully reclaimed after
//! a pong has been received by the remote. Therefore if you repeatidely ping a non-responsive
//! remote you will end up using more and memory memory (albeit the amount is very very small every
//! time), even if you destroy the future returned by `ping`.
//!
//! This is probably not a problem in practice, because the nature of the ping protocol is to
//! determine whether a remote is still alive, and any reasonable user of this crate will close
//! connections to non-responsive remotes.
//!
extern crate bytes;
extern crate futures;
extern crate libp2p_swarm;
extern crate multistream_select;
extern crate parking_lot;
extern crate rand;
extern crate tokio_io;
use bytes::{Bytes, BytesMut, BufMut};
use futures::{Future, Sink, Stream};
use futures::future::{FutureResult, IntoFuture, loop_fn, Loop};
use futures::sync::{mpsc, oneshot};
use libp2p_swarm::transport::ConnectionUpgrade;
use parking_lot::Mutex;
use rand::Rand;
use rand::os::OsRng;
use std::collections::HashMap;
use std::error::Error;
use std::io::Error as IoError;
use std::iter;
use std::sync::Arc;
use tokio_io::{AsyncRead, AsyncWrite};
use tokio_io::codec::{Encoder, Decoder};
/// Represents a prototype for an upgrade to handle the ping protocol.
///
/// According to the design of libp2p, this struct would normally contain the configuration options
/// for the protocol, but in the case of `Ping` no configuration is required.
#[derive(Debug, Copy, Clone, Default)]
pub struct Ping;
impl<C> ConnectionUpgrade<C> for Ping
where C: AsyncRead + AsyncWrite + 'static
{
type NamesIter = iter::Once<(Bytes, Self::UpgradeIdentifier)>;
type UpgradeIdentifier = ();
#[inline]
fn protocol_names(&self) -> Self::NamesIter {
iter::once(("/ipfs/ping/1.0.0".into(), ()))
}
type Output = (Pinger, Box<Future<Item = (), Error = IoError>>);
type Future = FutureResult<Self::Output, IoError>;
#[inline]
fn upgrade(self, socket: C, _: Self::UpgradeIdentifier) -> Self::Future {
// # How does it work?
//
// All the actual processing is performed by the *ponger*.
// We use a channel in order to send ping requests from the pinger to the ponger.
let (tx, rx) = mpsc::channel(8);
// Ignore the errors if `tx` closed. `tx` is only ever closed if the ponger is closed,
// which means that the connection to the remote is closed. Therefore we make the `rx`
// never produce anything.
let rx = rx.then(|r| Ok(r.ok())).filter_map(|a| a);
let os_rng = match OsRng::new() {
Ok(r) => r,
Err(err) => return Err(err).into_future(),
};
let pinger = Pinger {
send: tx,
os_rng: os_rng,
};
// Hashmap that associates outgoing payloads to one-shot senders.
// TODO: can't figure out how to make it work without using an Arc/Mutex
let expected_pongs = Arc::new(Mutex::new(HashMap::with_capacity(4)));
let sink_stream = socket.framed(Codec).map(|msg| Message::Received(msg.freeze()));
let (sink, stream) = sink_stream.split();
let future = loop_fn((sink, stream.select(rx)), move |(sink, stream)| {
let expected_pongs = expected_pongs.clone();
stream.into_future().map_err(|(err, _)| err).and_then(move |(message, stream)| {
let mut expected_pongs = expected_pongs.lock();
if let Some(message) = message {
match message {
Message::Ping(payload, finished) => {
// Ping requested by the user through the `Pinger`.
expected_pongs.insert(payload.clone(), finished);
Box::new(
sink.send(payload).map(|sink| Loop::Continue((sink, stream))),
) as Box<Future<Item = _, Error = _>>
}
Message::Received(payload) => {
// Received a payload from the remote.
if let Some(fut) = expected_pongs.remove(&payload) {
// Payload was ours. Signalling future.
// Errors can happen if the user closed the receiving end of
// the future, which is fine to ignore.
let _ = fut.send(());
Box::new(Ok(Loop::Continue((sink, stream))).into_future()) as
Box<Future<Item = _, Error = _>>
} else {
// Payload was not ours. Sending it back.
Box::new(
sink.send(payload).map(|sink| Loop::Continue((sink, stream))),
) as Box<Future<Item = _, Error = _>>
}
}
}
} else {
Box::new(Ok(Loop::Break(())).into_future()) as Box<Future<Item = _, Error = _>>
}
})
});
Ok((pinger, Box::new(future) as Box<_>)).into_future()
}
}
/// Controller for the ping service. Makes it possible to send pings to the remote.
pub struct Pinger {
send: mpsc::Sender<Message>,
os_rng: OsRng,
}
impl Pinger {
/// Sends a ping. Returns a future that is signaled when a pong is received.
///
/// **Note**: Please be aware that there is no timeout on the ping. You should handle the
/// timeout yourself when you call this function.
pub fn ping(&mut self) -> Box<Future<Item = (), Error = Box<Error + Send + Sync>>> {
let (tx, rx) = oneshot::channel();
let payload: [u8; 32] = Rand::rand(&mut self.os_rng);
// Ignore errors if the ponger has been already destroyed. The returned future will never
// be signalled.
let fut = self.send.clone().send(Message::Ping(Bytes::from(payload.to_vec()), tx))
.from_err()
.and_then(|_| rx.from_err());
Box::new(fut) as Box<_>
}
}
enum Message {
Ping(Bytes, oneshot::Sender<()>),
Received(Bytes),
}
// Implementation of the `Codec` trait of tokio-io. Splits frames into groups of 32 bytes.
#[derive(Copy, Clone, Debug, Eq, PartialEq, Ord, PartialOrd, Hash)]
struct Codec;
impl Decoder for Codec {
type Item = BytesMut;
type Error = IoError;
#[inline]
fn decode(&mut self, buf: &mut BytesMut) -> Result<Option<BytesMut>, IoError> {
if buf.len() >= 32 { Ok(Some(buf.split_to(32))) } else { Ok(None) }
}
}
impl Encoder for Codec {
type Item = Bytes;
type Error = IoError;
#[inline]
fn encode(&mut self, mut data: Bytes, buf: &mut BytesMut) -> Result<(), IoError> {
if data.len() != 0 {
let split = 32 * (1 + ((data.len() - 1) / 32));
buf.put(data.split_to(split));
}
Ok(())
}
}
#[cfg(test)]
mod tests {
extern crate tokio_core;
use self::tokio_core::net::TcpListener;
use self::tokio_core::net::TcpStream;
use self::tokio_core::reactor::Core;
use super::Ping;
use futures::future::join_all;
use futures::Future;
use futures::Stream;
use libp2p_swarm::transport::ConnectionUpgrade;
#[test]
fn ping_pong() {
let mut core = Core::new().unwrap();
let listener = TcpListener::bind(&"127.0.0.1:0".parse().unwrap(), &core.handle()).unwrap();
let listener_addr = listener.local_addr().unwrap();
let server = listener.incoming()
.into_future()
.map_err(|(e, _)| e.into())
.and_then(|(c, _)| Ping.upgrade(c.unwrap().0, ()))
.and_then(|(mut pinger, service)| {
pinger.ping().map_err(|_| panic!()).select(service).map_err(|_| panic!())
});
let client = TcpStream::connect(&listener_addr, &core.handle())
.map_err(|e| e.into())
.and_then(|c| Ping.upgrade(c, ()))
.and_then(|(mut pinger, service)| {
pinger.ping().map_err(|_| panic!()).select(service).map_err(|_| panic!())
});
core.run(server.join(client)).unwrap();
}
#[test]
fn multipings() {
// Check that we can send multiple pings in a row and it will still work.
let mut core = Core::new().unwrap();
let listener = TcpListener::bind(&"127.0.0.1:0".parse().unwrap(), &core.handle()).unwrap();
let listener_addr = listener.local_addr().unwrap();
let server = listener.incoming()
.into_future()
.map_err(|(e, _)| e.into())
.and_then(|(c, _)| Ping.upgrade(c.unwrap().0, ()))
.and_then(|(_, service)| service.map_err(|_| panic!()));
let client = TcpStream::connect(&listener_addr, &core.handle())
.map_err(|e| e.into())
.and_then(|c| Ping.upgrade(c, ()))
.and_then(|(mut pinger, service)| {
let pings = (0 .. 20).map(move |_| {
pinger.ping().map_err(|_| ())
});
join_all(pings).map(|_| ()).map_err(|_| panic!())
.select(service).map(|_| ()).map_err(|_| panic!())
});
core.run(server.select(client)).unwrap_or_else(|_| panic!());
}
}