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
// Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
use crate::{
allocator::Allocator,
clock::Timer,
event, msg,
stream::{
shared::{AcceptState, ArcShared, ShutdownKind},
socket::Socket,
Actor,
},
};
use core::task::{Context, Poll};
use s2n_quic_core::{
buffer, dc::ApplicationParams, endpoint, ensure, ready, time::clock::Timer as _,
};
use std::{io, time::Duration};
use tracing::{debug, trace};
const INITIAL_TIMEOUT: Duration = Duration::from_millis(2);
mod waiting {
use s2n_quic_core::state::{event, is};
#[derive(Clone, Debug, Default, PartialEq)]
pub enum State {
PeekPacket,
EpochTimeout,
#[default]
Cooldown,
DataRecvd,
Detached,
TimeWait,
Finished,
}
impl State {
is!(is_peek_packet, PeekPacket);
is!(is_time_wait, TimeWait);
event! {
on_peek_packet(PeekPacket => EpochTimeout);
on_cooldown_elapsed(Cooldown => PeekPacket);
on_epoch_unchanged(EpochTimeout => PeekPacket);
on_application_progress(PeekPacket | EpochTimeout | Cooldown => Cooldown);
on_application_detach(PeekPacket | EpochTimeout | Cooldown => Detached);
on_data_received(PeekPacket | EpochTimeout | Cooldown => DataRecvd);
on_time_wait(Detached | DataRecvd => TimeWait);
on_finished(PeekPacket | EpochTimeout | Cooldown | Detached | DataRecvd | TimeWait => Finished);
}
}
#[test]
fn dot_test() {
insta::assert_snapshot!(State::dot());
}
}
#[repr(u8)]
pub(crate) enum ErrorCode {
/// The application dropped the stream without errors
None = 0,
/// General error code for application-level errors
Application = 1,
}
pub struct Worker<S, Sub>
where
S: Socket,
Sub: event::Subscriber,
{
shared: ArcShared<Sub>,
last_observed_epoch: u64,
send_buffer: msg::send::Message,
state: waiting::State,
peek_timer: Timer,
idle_timer: Timer,
idle_timeout_duration: Duration,
backoff: u8,
socket: S,
accept_state: AcceptState,
}
impl<S, Sub> Worker<S, Sub>
where
S: Socket,
Sub: event::Subscriber,
{
#[inline]
pub fn new(
socket: S,
shared: ArcShared<Sub>,
endpoint: endpoint::Type,
parameters: &ApplicationParams,
) -> Self {
let send_buffer = msg::send::Message::new(shared.remote_addr(), shared.gso.clone());
let idle_timeout_duration = parameters
.max_idle_timeout()
.unwrap_or_else(|| Duration::from_secs(30));
let peek_timer = Timer::new_with_timeout(&shared.clock, INITIAL_TIMEOUT);
let idle_timer = Timer::new_with_timeout(&shared.clock, idle_timeout_duration);
let state = match endpoint {
// on the client we delay before reading from the socket
endpoint::Type::Client => waiting::State::Cooldown,
// on the server we need the application to read after accepting, otherwise the peer
// won't know what our port is
endpoint::Type::Server => waiting::State::EpochTimeout,
};
Self {
shared,
last_observed_epoch: 0,
send_buffer,
state,
peek_timer,
idle_timer,
idle_timeout_duration,
backoff: 0,
socket,
accept_state: AcceptState::Waiting,
}
}
#[inline]
pub fn update_waker(&self, cx: &mut Context) {
self.shared.receiver.worker_waker.update(cx.waker());
}
#[inline]
pub fn poll(&mut self, cx: &mut Context) -> Poll<()> {
s2n_quic_core::task::waker::debug_assert_contract(cx, |cx| {
ready!(self.poll_impl(cx));
tracing::debug!("read worker shutting down");
Poll::Ready(())
})
}
#[inline]
fn poll_impl(&mut self, cx: &mut Context) -> Poll<()> {
if let Poll::Ready(Err(err)) = self.poll_flush_socket(cx) {
tracing::error!(socket_error = ?err);
// TODO should we return? if we get a send error it's most likely fatal
return Poll::Ready(());
}
if let Poll::Ready(Err(err)) = self.poll_socket(cx) {
tracing::error!(socket_error = ?err);
// TODO should we return? if we get a recv error it's most likely fatal
return Poll::Ready(());
}
// go until we get into the finished state
if let waiting::State::Finished = &self.state {
return Poll::Ready(());
}
{
let target = self.shared.last_peer_activity() + self.idle_timeout_duration;
self.idle_timer.update(target);
if self.idle_timer.poll_ready(cx).is_ready() {
return Poll::Ready(());
}
}
Poll::Pending
}
#[inline]
fn poll_socket(&mut self, cx: &mut Context) -> Poll<io::Result<()>> {
loop {
match &self.state {
waiting::State::PeekPacket => {
// check to see if the application is progressing before peeking the socket
ensure!(!self.is_application_progressing(), continue);
// check if we have something pending
ready!(self.shared.receiver.poll_peek_worker(
cx,
&self.socket,
&self.shared.clock,
&self.shared.subscriber,
));
self.arm_peek_timer();
#[expect(
clippy::unwrap_used,
reason = "we matched on the PeekPacket state above so this transition is always valid"
)]
self.state.on_peek_packet().unwrap();
continue;
}
waiting::State::EpochTimeout => {
// check to see if the application is progressing before checking the timer
ensure!(!self.is_application_progressing(), continue);
ready!(self.peek_timer.poll_ready(cx));
// the application isn't making progress so emit the timer expired event
#[expect(
clippy::unwrap_used,
reason = "we matched on the EpochTimeout state above so this transition is always valid"
)]
self.state.on_epoch_unchanged().unwrap();
// only log this message after the first observation
if self.last_observed_epoch > 0 {
debug!("application reading too slowly from socket");
}
// reset the backoff with the assumption that the application will go slow in
// the future
self.backoff = 0;
// drain the socket if the application isn't going fast enough
return self.poll_drain_recv_socket(cx);
}
waiting::State::Cooldown => {
// check to see if the application is progressing before checking the timer
ensure!(!self.is_application_progressing(), continue);
ready!(self.peek_timer.poll_ready(cx));
// go back to waiting for a packet
let _ = self.state.on_cooldown_elapsed();
continue;
}
waiting::State::Detached | waiting::State::DataRecvd => {
// check if we have any packets in the queue
let _ = self.poll_drain_recv_socket(cx);
// transition to time wait and arm the timer
ensure!(self.state.on_time_wait().is_ok(), continue);
// TODO instead of arming a timer, we should add a mode to the `stream` receiver
// that allows it to be marked as "free" for reuse while holding the last control
// packet that this worker sent. The recv socket pool would look at the credentials
// on each packet to see if it should intercept and respond with an old control packet
// in case the sender didn't see the control packet. This is similar to TCP Reuse/Recycle.
let now = self.shared.clock.get_time();
let target = now + Duration::from_millis(500);
self.peek_timer.update(target);
}
waiting::State::TimeWait => {
// check if we have any packets in the socket
let _ = self.poll_drain_recv_socket(cx);
// make sure we're still in `TimeWait` after looking at the socket
ensure!(self.state.is_time_wait(), continue);
// wait for the timer to expire
ready!(self.peek_timer.poll_ready(cx));
// after the timer expires, then transition to the finished state
let _ = self.state.on_finished();
}
waiting::State::Finished => {
// nothing left to do
return Ok(()).into();
}
}
}
}
#[inline]
fn is_application_progressing(&mut self) -> bool {
// check to see if the application shut down
if let super::shared::ApplicationState::Closed { shutdown_kind } =
self.shared.receiver.application_state()
{
if matches!(shutdown_kind, ShutdownKind::Pruned) {
// if the stream was pruned then we don't need to do anything else
let _ = self.state.on_finished();
self.peek_timer.cancel();
self.idle_timer.cancel();
return true;
}
if let Ok(Some(mut recv)) = self.shared.receiver.worker_try_lock() {
// check to see if we have anything in the reassembler as well
let is_buffer_empty = recv.payload_is_empty() && recv.reassembler.is_empty();
let error = if let Some(code) = shutdown_kind.error_code() {
code
} else if !is_buffer_empty {
// we still had data in our buffer so notify the sender
ErrorCode::Application as u8
} else {
// no error - the application is just going away
ErrorCode::None as u8
};
let publisher = self.shared.publisher();
recv.receiver.stop_sending(error.into(), &publisher);
if recv.receiver.is_finished() {
let _ = self.state.on_finished();
}
}
let _ = self.state.on_application_detach();
return true;
}
let current_epoch = self.shared.receiver.application_epoch();
// If the application incremented its epoch then it's been accepted
if current_epoch > 0 {
self.accept_state = AcceptState::Accepted;
}
// make sure the epoch has changed since we last saw it before cooling down
ensure!(self.last_observed_epoch < current_epoch, false);
// record the new observation
self.last_observed_epoch = current_epoch;
// the application is making progress since the packet is different - loop back to cooldown
trace!("application is making progress");
// delay when we read from the socket again to avoid spinning
let _ = self.state.on_application_progress();
self.arm_peek_timer();
// after successful progress from the application we want to intervene less
self.backoff = (self.backoff + 1).min(10);
true
}
#[inline]
fn poll_drain_recv_socket(&mut self, cx: &mut Context) -> Poll<io::Result<()>> {
let mut should_transmit = false;
let mut received_packets = 0;
let _res = self.process_packets(cx, &mut received_packets, &mut should_transmit);
ensure!(
should_transmit,
if received_packets == 0 {
Poll::Pending
} else {
Ok(()).into()
}
);
// send an ACK if needed
if let Some(mut recv) = self.shared.receiver.worker_try_lock()? {
// use the latest value rather than trying to transmit an old one
if !self.send_buffer.is_empty() {
let _ = self.send_buffer.drain();
}
recv.fill_transmit_queue(&self.shared, &mut self.send_buffer);
if recv.receiver.state().is_data_received() {
let _ = self.state.on_data_received();
}
if recv.receiver.is_finished() {
let _ = self.state.on_finished();
}
}
ready!(self.poll_flush_socket(cx))?;
Ok(()).into()
}
#[inline]
fn process_packets(
&mut self,
cx: &mut Context,
received_packets: &mut usize,
should_transmit: &mut bool,
) -> io::Result<()> {
// loop until we hit Pending from the socket
loop {
// try_lock the state before reading so we don't consume a packet the application is
// about to read
let Some(mut recv) = self.shared.receiver.worker_try_lock()? else {
// if the application is locking the state then we don't want to transmit, since it
// will do that for us
*should_transmit = false;
break;
};
// make sure to process any left over packets, if any
if !recv.payload_is_empty() {
*should_transmit |= recv.process_recv_buffer(
&mut buffer::writer::storage::Empty,
&self.shared,
self.socket.features(),
self.accept_state,
);
}
let res = recv.poll_fill_recv_buffer(
cx,
Actor::Worker,
&self.socket,
&self.shared.clock,
&self.shared.subscriber,
);
match res {
Poll::Pending => break,
Poll::Ready(res) => res?,
};
*received_packets += 1;
// process the packet we just received
*should_transmit |= recv.process_recv_buffer(
&mut buffer::writer::storage::Empty,
&self.shared,
self.socket.features(),
self.accept_state,
);
}
Ok(())
}
#[inline]
fn poll_flush_socket(&mut self, cx: &mut Context) -> Poll<io::Result<()>> {
while !self.send_buffer.is_empty() {
ready!(self.socket.poll_send_buffer(cx, &mut self.send_buffer))?;
}
Ok(()).into()
}
#[inline]
fn arm_peek_timer(&mut self) {
// TODO do we derive this from RTT?
let mut timeout = INITIAL_TIMEOUT;
// don't back off on packet peeks
if !self.state.is_peek_packet() {
timeout *= (self.backoff as u32) + 1;
}
let now = self.shared.clock.get_time();
let target = now + timeout;
self.peek_timer.update(target);
}
}