-
-
Notifications
You must be signed in to change notification settings - Fork 360
/
endpoint.rs
621 lines (570 loc) · 21.8 KB
/
endpoint.rs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
use std::{
collections::VecDeque,
future::Future,
io,
io::IoSliceMut,
mem::MaybeUninit,
net::{SocketAddr, SocketAddrV6},
pin::Pin,
str,
sync::{Arc, Mutex},
task::{Context, Poll, Waker},
time::Instant,
};
use bytes::Bytes;
use futures_channel::mpsc;
use futures_core::Stream;
use proto::{
self as proto, ClientConfig, ConnectError, ConnectionHandle, DatagramEvent, ServerConfig,
};
use rustc_hash::FxHashMap;
use tokio::sync::Notify;
use udp::{RecvMeta, UdpSocket, UdpState, BATCH_SIZE};
use crate::{
connection::Connecting, poll_fn, work_limiter::WorkLimiter, ConnectionEvent, EndpointConfig,
EndpointEvent, VarInt, IO_LOOP_BOUND, RECV_TIME_BOUND, SEND_TIME_BOUND,
};
/// A QUIC endpoint.
///
/// An endpoint corresponds to a single UDP socket, may host many connections, and may act as both
/// client and server for different connections.
///
/// May be cloned to obtain another handle to the same endpoint.
#[derive(Debug, Clone)]
pub struct Endpoint {
pub(crate) inner: EndpointRef,
pub(crate) default_client_config: Option<ClientConfig>,
}
impl Endpoint {
/// Helper to construct an endpoint for use with outgoing connections only
///
/// Must be called from within a tokio runtime context. Note that `addr` is the *local* address
/// to bind to, which should usually be a wildcard address like `0.0.0.0:0` or `[::]:0`, which
/// allow communication with any reachable IPv4 or IPv6 address respectively from an OS-assigned
/// port.
///
/// Platform defaults for dual-stack sockets vary. For example, any socket bound to a wildcard
/// IPv6 address on Windows will not by default be able to communicate with IPv4
/// addresses. Portable applications should bind an address that matches the family they wish to
/// communicate within.
#[cfg(feature = "ring")]
pub fn client(addr: SocketAddr) -> io::Result<Self> {
let socket = std::net::UdpSocket::bind(addr)?;
Ok(Self::new(EndpointConfig::default(), None, socket)?.0)
}
/// Helper to construct an endpoint for use with both incoming and outgoing connections
///
/// Must be called from within a tokio runtime context.
///
/// Platform defaults for dual-stack sockets vary. For example, any socket bound to a wildcard
/// IPv6 address on Windows will not by default be able to communicate with IPv4
/// addresses. Portable applications should bind an address that matches the family they wish to
/// communicate within.
#[cfg(feature = "ring")]
pub fn server(config: ServerConfig, addr: SocketAddr) -> io::Result<(Self, Incoming)> {
let socket = std::net::UdpSocket::bind(addr)?;
Self::new(EndpointConfig::default(), Some(config), socket)
}
/// Construct an endpoint with arbitrary configuration
///
/// Must be called from within a tokio runtime context.
pub fn new(
config: EndpointConfig,
server_config: Option<ServerConfig>,
socket: std::net::UdpSocket,
) -> io::Result<(Self, Incoming)> {
let addr = socket.local_addr()?;
let socket = UdpSocket::from_std(socket)?;
let rc = EndpointRef::new(
socket,
proto::Endpoint::new(Arc::new(config), server_config.map(Arc::new)),
addr.is_ipv6(),
);
let driver = EndpointDriver(rc.clone());
tokio::spawn(async {
if let Err(e) = driver.await {
tracing::error!("I/O error: {}", e);
}
});
Ok((
Self {
inner: rc.clone(),
default_client_config: None,
},
Incoming::new(rc),
))
}
/// Set the client configuration used by `connect`
pub fn set_default_client_config(&mut self, config: ClientConfig) {
self.default_client_config = Some(config);
}
/// Connect to a remote endpoint
///
/// `server_name` must be covered by the certificate presented by the server. This prevents a
/// connection from being intercepted by an attacker with a valid certificate for some other
/// server.
///
/// May fail immediately due to configuration errors, or in the future if the connection could
/// not be established.
pub fn connect(&self, addr: SocketAddr, server_name: &str) -> Result<Connecting, ConnectError> {
let config = match &self.default_client_config {
Some(config) => config.clone(),
None => return Err(ConnectError::NoDefaultClientConfig),
};
self.connect_with(config, addr, server_name)
}
/// Connect to a remote endpoint using a custom configuration.
///
/// See [`connect()`] for details.
///
/// [`connect()`]: Endpoint::connect
pub fn connect_with(
&self,
config: ClientConfig,
addr: SocketAddr,
server_name: &str,
) -> Result<Connecting, ConnectError> {
let mut endpoint = self.inner.lock().unwrap();
if endpoint.driver_lost {
return Err(ConnectError::EndpointStopping);
}
if addr.is_ipv6() && !endpoint.ipv6 {
return Err(ConnectError::InvalidRemoteAddress(addr));
}
let addr = if endpoint.ipv6 {
SocketAddr::V6(ensure_ipv6(addr))
} else {
addr
};
let (ch, conn) = endpoint.inner.connect(config, addr, server_name)?;
let udp_state = endpoint.udp_state.clone();
Ok(endpoint.connections.insert(ch, conn, udp_state))
}
/// Switch to a new UDP socket
///
/// Allows the endpoint's address to be updated live, affecting all active connections. Incoming
/// connections and connections to servers unreachable from the new address will be lost.
///
/// On error, the old UDP socket is retained.
pub fn rebind(&self, socket: std::net::UdpSocket) -> io::Result<()> {
let addr = socket.local_addr()?;
let socket = UdpSocket::from_std(socket)?;
let mut inner = self.inner.lock().unwrap();
inner.socket = socket;
inner.ipv6 = addr.is_ipv6();
// Generate some activity so peers notice the rebind
for sender in inner.connections.senders.values() {
// Ignoring errors from dropped connections
let _ = sender.unbounded_send(ConnectionEvent::Ping);
}
Ok(())
}
/// Replace the server configuration, affecting new incoming connections only
///
/// Useful for e.g. refreshing TLS certificates without disrupting existing connections.
pub fn set_server_config(&self, server_config: Option<ServerConfig>) {
self.inner
.lock()
.unwrap()
.inner
.set_server_config(server_config.map(Arc::new))
}
/// Get the local `SocketAddr` the underlying socket is bound to
pub fn local_addr(&self) -> io::Result<SocketAddr> {
self.inner.lock().unwrap().socket.local_addr()
}
/// Close all of this endpoint's connections immediately and cease accepting new connections.
///
/// See [`Connection::close()`] for details.
///
/// [`Connection::close()`]: crate::Connection::close
pub fn close(&self, error_code: VarInt, reason: &[u8]) {
let reason = Bytes::copy_from_slice(reason);
let mut endpoint = self.inner.lock().unwrap();
endpoint.connections.close = Some((error_code, reason.clone()));
for sender in endpoint.connections.senders.values() {
// Ignoring errors from dropped connections
let _ = sender.unbounded_send(ConnectionEvent::Close {
error_code,
reason: reason.clone(),
});
}
if let Some(task) = endpoint.incoming_reader.take() {
task.wake();
}
}
/// Wait for all connections on the endpoint to be cleanly shut down
///
/// Waiting for this condition before exiting ensures that a good-faith effort is made to notify
/// peers of recent connection closes, whereas exiting immediately could force them to wait out
/// the idle timeout period.
///
/// Does not proactively close existing connections or cause incoming connections to be
/// rejected. Consider calling [`close()`] and dropping the [`Incoming`] stream if
/// that is desired.
///
/// [`close()`]: Endpoint::close
/// [`Incoming`]: crate::Incoming
pub async fn wait_idle(&self) {
loop {
let idle;
{
let endpoint = &mut *self.inner.lock().unwrap();
if endpoint.connections.is_empty() {
break;
}
// Clone the `Arc<Notify>` so we can wait on the underlying `Notify` without holding
// the lock. Store it in the outer scope to ensure it outlives the lock guard.
idle = endpoint.idle.clone();
// Construct the future while the lock is held to ensure we can't miss a wakeup if
// the `Notify` is signaled immediately after we release the lock. `await` it after
// the lock guard is out of scope.
idle.notified()
}
.await;
}
}
}
/// A future that drives IO on an endpoint
///
/// This task functions as the switch point between the UDP socket object and the
/// `Endpoint` responsible for routing datagrams to their owning `Connection`.
/// In order to do so, it also facilitates the exchange of different types of events
/// flowing between the `Endpoint` and the tasks managing `Connection`s. As such,
/// running this task is necessary to keep the endpoint's connections running.
///
/// `EndpointDriver` futures terminate when the `Incoming` stream and all clones of the `Endpoint`
/// have been dropped, or when an I/O error occurs.
#[must_use = "endpoint drivers must be spawned for I/O to occur"]
#[derive(Debug)]
pub(crate) struct EndpointDriver(pub(crate) EndpointRef);
impl Future for EndpointDriver {
type Output = Result<(), io::Error>;
#[allow(unused_mut)] // MSRV
fn poll(mut self: Pin<&mut Self>, cx: &mut Context) -> Poll<Self::Output> {
let mut endpoint = self.0.lock().unwrap();
if endpoint.driver.is_none() {
endpoint.driver = Some(cx.waker().clone());
}
let now = Instant::now();
let mut keep_going = false;
keep_going |= endpoint.drive_recv(cx, now)?;
keep_going |= endpoint.handle_events(cx);
keep_going |= endpoint.drive_send(cx)?;
if !endpoint.incoming.is_empty() {
if let Some(task) = endpoint.incoming_reader.take() {
task.wake();
}
}
if endpoint.ref_count == 0 && endpoint.connections.is_empty() {
Poll::Ready(Ok(()))
} else {
drop(endpoint);
// If there is more work to do schedule the endpoint task again.
// `wake_by_ref()` is called outside the lock to minimize
// lock contention on a multithreaded runtime.
if keep_going {
cx.waker().wake_by_ref();
}
Poll::Pending
}
}
}
impl Drop for EndpointDriver {
fn drop(&mut self) {
let mut endpoint = self.0.lock().unwrap();
endpoint.driver_lost = true;
if let Some(task) = endpoint.incoming_reader.take() {
task.wake();
}
// Drop all outgoing channels, signaling the termination of the endpoint to the associated
// connections.
endpoint.connections.senders.clear();
}
}
#[derive(Debug)]
pub(crate) struct EndpointInner {
socket: UdpSocket,
udp_state: Arc<UdpState>,
inner: proto::Endpoint,
outgoing: VecDeque<proto::Transmit>,
incoming: VecDeque<Connecting>,
incoming_reader: Option<Waker>,
driver: Option<Waker>,
ipv6: bool,
connections: ConnectionSet,
events: mpsc::UnboundedReceiver<(ConnectionHandle, EndpointEvent)>,
/// Number of live handles that can be used to initiate or handle I/O; excludes the driver
ref_count: usize,
driver_lost: bool,
recv_limiter: WorkLimiter,
recv_buf: Box<[u8]>,
send_limiter: WorkLimiter,
idle: Arc<Notify>,
}
impl EndpointInner {
fn drive_recv<'a>(&'a mut self, cx: &mut Context, now: Instant) -> Result<bool, io::Error> {
self.recv_limiter.start_cycle();
let mut metas = [RecvMeta::default(); BATCH_SIZE];
let mut iovs = MaybeUninit::<[IoSliceMut<'a>; BATCH_SIZE]>::uninit();
self.recv_buf
.chunks_mut(self.recv_buf.len() / BATCH_SIZE)
.enumerate()
.for_each(|(i, buf)| unsafe {
iovs.as_mut_ptr()
.cast::<IoSliceMut>()
.add(i)
.write(IoSliceMut::<'a>::new(buf));
});
let mut iovs = unsafe { iovs.assume_init() };
loop {
match self.socket.poll_recv(cx, &mut iovs, &mut metas) {
Poll::Ready(Ok(msgs)) => {
self.recv_limiter.record_work(msgs);
for (meta, buf) in metas.iter().zip(iovs.iter()).take(msgs) {
let data = buf[0..meta.len].into();
match self
.inner
.handle(now, meta.addr, meta.dst_ip, meta.ecn, data)
{
Some((handle, DatagramEvent::NewConnection(conn))) => {
let conn =
self.connections
.insert(handle, conn, self.udp_state.clone());
self.incoming.push_back(conn);
}
Some((handle, DatagramEvent::ConnectionEvent(event))) => {
// Ignoring errors from dropped connections that haven't yet been cleaned up
let _ = self
.connections
.senders
.get_mut(&handle)
.unwrap()
.unbounded_send(ConnectionEvent::Proto(event));
}
None => {}
}
}
}
Poll::Pending => {
break;
}
// Ignore ECONNRESET as it's undefined in QUIC and may be injected by an
// attacker
Poll::Ready(Err(ref e)) if e.kind() == io::ErrorKind::ConnectionReset => {
continue;
}
Poll::Ready(Err(e)) => {
return Err(e);
}
}
if !self.recv_limiter.allow_work() {
self.recv_limiter.finish_cycle();
return Ok(true);
}
}
self.recv_limiter.finish_cycle();
Ok(false)
}
fn drive_send(&mut self, cx: &mut Context) -> Result<bool, io::Error> {
self.send_limiter.start_cycle();
let result = loop {
while self.outgoing.len() < BATCH_SIZE {
match self.inner.poll_transmit() {
Some(x) => self.outgoing.push_back(x),
None => break,
}
}
if self.outgoing.is_empty() {
break Ok(false);
}
if !self.send_limiter.allow_work() {
break Ok(true);
}
match self
.socket
.poll_send(&self.udp_state, cx, self.outgoing.as_slices().0)
{
Poll::Ready(Ok(n)) => {
self.outgoing.drain(..n);
// We count transmits instead of `poll_send` calls since the cost
// of a `sendmmsg` still linearily increases with number of packets.
self.send_limiter.record_work(n);
}
Poll::Pending => {
break Ok(false);
}
Poll::Ready(Err(e)) => {
break Err(e);
}
}
};
self.send_limiter.finish_cycle();
result
}
fn handle_events(&mut self, cx: &mut Context) -> bool {
use EndpointEvent::*;
for _ in 0..IO_LOOP_BOUND {
match Pin::new(&mut self.events).poll_next(cx) {
Poll::Ready(Some((ch, event))) => match event {
Proto(e) => {
if e.is_drained() {
self.connections.senders.remove(&ch);
if self.connections.is_empty() {
self.idle.notify_waiters();
}
}
if let Some(event) = self.inner.handle_event(ch, e) {
// Ignoring errors from dropped connections that haven't yet been cleaned up
let _ = self
.connections
.senders
.get_mut(&ch)
.unwrap()
.unbounded_send(ConnectionEvent::Proto(event));
}
}
Transmit(t) => self.outgoing.push_back(t),
},
Poll::Ready(None) => unreachable!("EndpointInner owns one sender"),
Poll::Pending => {
return false;
}
}
}
true
}
}
#[derive(Debug)]
struct ConnectionSet {
/// Senders for communicating with the endpoint's connections
senders: FxHashMap<ConnectionHandle, mpsc::UnboundedSender<ConnectionEvent>>,
/// Stored to give out clones to new ConnectionInners
sender: mpsc::UnboundedSender<(ConnectionHandle, EndpointEvent)>,
/// Set if the endpoint has been manually closed
close: Option<(VarInt, Bytes)>,
}
impl ConnectionSet {
fn insert(
&mut self,
handle: ConnectionHandle,
conn: proto::Connection,
udp_state: Arc<UdpState>,
) -> Connecting {
let (send, recv) = mpsc::unbounded();
if let Some((error_code, ref reason)) = self.close {
send.unbounded_send(ConnectionEvent::Close {
error_code,
reason: reason.clone(),
})
.unwrap();
}
self.senders.insert(handle, send);
Connecting::new(handle, conn, self.sender.clone(), recv, udp_state)
}
fn is_empty(&self) -> bool {
self.senders.is_empty()
}
}
fn ensure_ipv6(x: SocketAddr) -> SocketAddrV6 {
match x {
SocketAddr::V6(x) => x,
SocketAddr::V4(x) => SocketAddrV6::new(x.ip().to_ipv6_mapped(), x.port(), 0, 0),
}
}
/// Stream of incoming connections.
#[derive(Debug)]
pub struct Incoming(EndpointRef);
impl Incoming {
pub(crate) fn new(inner: EndpointRef) -> Self {
Self(inner)
}
}
impl Incoming {
/// Fetch the next incoming connection, or `None` if the endpoint has been closed
pub async fn next(&mut self) -> Option<Connecting> {
poll_fn(move |cx| Pin::new(&mut *self).poll_next(cx)).await
}
}
impl Stream for Incoming {
type Item = Connecting;
#[allow(unused_mut)] // MSRV
fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context) -> Poll<Option<Self::Item>> {
let endpoint = &mut *self.0.lock().unwrap();
if endpoint.driver_lost {
Poll::Ready(None)
} else if let Some(conn) = endpoint.incoming.pop_front() {
Poll::Ready(Some(conn))
} else if endpoint.connections.close.is_some() {
Poll::Ready(None)
} else {
endpoint.incoming_reader = Some(cx.waker().clone());
Poll::Pending
}
}
}
impl Drop for Incoming {
fn drop(&mut self) {
let endpoint = &mut *self.0.lock().unwrap();
endpoint.inner.reject_new_connections();
endpoint.incoming_reader = None;
}
}
#[derive(Debug)]
pub(crate) struct EndpointRef(Arc<Mutex<EndpointInner>>);
impl EndpointRef {
pub(crate) fn new(socket: UdpSocket, inner: proto::Endpoint, ipv6: bool) -> Self {
let recv_buf =
vec![0; inner.config().get_max_udp_payload_size().min(64 * 1024) as usize * BATCH_SIZE];
let (sender, events) = mpsc::unbounded();
Self(Arc::new(Mutex::new(EndpointInner {
socket,
udp_state: Arc::new(UdpState::new()),
inner,
ipv6,
events,
outgoing: VecDeque::new(),
incoming: VecDeque::new(),
incoming_reader: None,
driver: None,
connections: ConnectionSet {
senders: FxHashMap::default(),
sender,
close: None,
},
ref_count: 0,
driver_lost: false,
recv_buf: recv_buf.into(),
recv_limiter: WorkLimiter::new(RECV_TIME_BOUND),
send_limiter: WorkLimiter::new(SEND_TIME_BOUND),
idle: Arc::new(Notify::new()),
})))
}
}
impl Clone for EndpointRef {
fn clone(&self) -> Self {
self.0.lock().unwrap().ref_count += 1;
Self(self.0.clone())
}
}
impl Drop for EndpointRef {
fn drop(&mut self) {
let endpoint = &mut *self.0.lock().unwrap();
if let Some(x) = endpoint.ref_count.checked_sub(1) {
endpoint.ref_count = x;
if x == 0 {
// If the driver is about to be on its own, ensure it can shut down if the last
// connection is gone.
if let Some(task) = endpoint.driver.take() {
task.wake();
}
}
}
}
}
impl std::ops::Deref for EndpointRef {
type Target = Mutex<EndpointInner>;
fn deref(&self) -> &Self::Target {
&self.0
}
}