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
use std::fs::File;
use std::io::Read;
use std::net::ToSocketAddrs;
use std::os::fd::{AsRawFd, FromRawFd, OwnedFd};
use mio::{unix::SourceFd, Events, Interest, Token};
use parking_lot::{Mutex, RwLock};
use sharded_slab::Slab;
use core::future::poll_fn;
use core::task::Poll;
use mfio::backend::fd::FdWakerOwner;
use mfio::backend::*;
use mfio::tarc::BaseArc;
use super::unix_extra::set_nonblock;
use crate::util::Key;
use tracing::instrument::Instrument;
use file::FileInner;
pub use file::FileWrapper;
use tcp_stream::StreamInner;
pub use tcp_stream::{TcpConnectFuture, TcpStream};
use tcp_listener::ListenerInner;
pub use tcp_listener::TcpListener;
mod file;
mod tcp_listener;
mod tcp_stream;
const RW_INTERESTS: Interest = Interest::READABLE.add(Interest::WRITABLE);
#[derive(Default)]
struct BlockTrack {
cur_interests: Option<Interest>,
read_blocked: bool,
write_blocked: bool,
update_queued: bool,
}
impl BlockTrack {
pub fn expected_interests(&self) -> Option<Interest> {
let mut expected_interests = Some(RW_INTERESTS);
if !self.read_blocked {
expected_interests = expected_interests.and_then(|v| v.remove(Interest::READABLE));
}
if !self.write_blocked {
expected_interests = expected_interests.and_then(|v| v.remove(Interest::WRITABLE));
}
expected_interests
}
}
struct MioState {
poll: Mutex<mio::Poll>,
files: RwLock<Slab<Mutex<FileInner>>>,
streams: RwLock<Slab<Mutex<StreamInner>>>,
listeners: Slab<Mutex<ListenerInner>>,
opqueue: Mutex<Vec<Key>>,
}
impl MioState {
fn try_new() -> std::io::Result<Self> {
Ok(Self {
poll: mio::Poll::new()?.into(),
files: Default::default(),
streams: Default::default(),
listeners: Default::default(),
opqueue: vec![].into(),
})
}
}
pub struct Runtime {
state: BaseArc<MioState>,
backend: BackendContainer<DynBackend>,
waker: FdWakerOwner<OwnedFd>,
}
impl Runtime {
pub fn try_new() -> std::io::Result<Self> {
let state = MioState::try_new()?;
let (wake_read, wake_write) = nix::unistd::pipe()?;
set_nonblock(wake_read)?;
set_nonblock(wake_write)?;
log::trace!("{wake_read} {wake_write}");
let mut wake_read = unsafe { File::from_raw_fd(wake_read) };
let wake_write = unsafe { OwnedFd::from_raw_fd(wake_write) };
let waker = FdWakerOwner::from(wake_write);
// Register the waker in a special manner
state.poll.lock().registry().register(
&mut SourceFd(&wake_read.as_raw_fd()),
Token(usize::MAX),
Interest::READABLE,
)?;
let state = BaseArc::new(state);
let backend = {
let state = state.clone();
let waker = waker.clone();
async move {
// We could use a set here, but arguably, computing interests multiple times
// should be quicker than walking down the set. TODO: verify this
let mut interest_update_queue = vec![];
let mut events = Events::with_capacity(1024);
loop {
let ret = async {
if let Err(_e) = state
.poll
.lock()
.poll(&mut events, Some(Default::default()))
{
return false;
}
let mut observed_blocking = false;
for event in events.iter() {
let key = event.token().0;
log::trace!("Key: {key:x}");
interest_update_queue.push(Key::from(key));
// These will fail with usize::MAX key
if key == usize::MAX {
observed_blocking = true;
} else {
match Key::from(key) {
Key::File(v) => {
if let Some(file) = state.files.read().get(v) {
log::trace!(
"readable={} writeable={}",
event.is_readable(),
event.is_writable()
);
file.lock()
.do_ops(event.is_readable(), event.is_writable());
}
}
Key::Stream(v) => {
if let Some(stream) = state.streams.read().get(v) {
log::trace!(
"readable={} writeable={}",
event.is_readable(),
event.is_writable()
);
stream
.lock()
.do_ops(event.is_readable(), event.is_writable());
}
}
Key::TcpListener(v) => {
if let Some(listener) = state.listeners.get(v) {
log::trace!(
"readable={} writeable={}",
event.is_readable(),
event.is_writable()
);
listener
.lock()
.do_ops(event.is_readable(), event.is_writable());
}
}
}
}
}
// process the operation queue
for key in state.opqueue.lock().drain(..) {
interest_update_queue.push(key);
match key {
Key::File(v) => {
if let Some(file) = state.files.read().get(v) {
file.lock().on_queue();
}
}
Key::Stream(v) => {
if let Some(stream) = state.streams.read().get(v) {
stream.lock().on_queue();
}
}
Key::TcpListener(_) => (),
}
}
let poll = state.poll.lock();
// Update polling interests for any FDs that did any work.
for key in interest_update_queue.drain(..) {
let res = match key {
Key::File(v) => {
if let Some(file) = state.files.read().get(v) {
file.lock().update_interests(key.key(), poll.registry())
} else {
Ok(())
}
}
Key::Stream(v) => {
if let Some(stream) = state.streams.read().get(v) {
stream.lock().update_interests(key.key(), poll.registry())
} else {
Ok(())
}
}
Key::TcpListener(v) => {
if let Some(listener) = state.listeners.get(v) {
listener.lock().update_interests(key.key(), poll.registry())
} else {
Ok(())
}
}
};
// TODO: handle errors
match res {
// EPERM using epoll means file descriptor is always ready.
// However, this also comes with a caveat that even though the file always shows up as
// ready, it does not necessarily work in non-blocking mode.
#[cfg(target_os = "linux")]
Err(e) if e.kind() == std::io::ErrorKind::PermissionDenied => (),
Err(e) => panic!("Unable to update interests: {e}"),
Ok(_) => (),
}
}
if observed_blocking {
log::trace!("Observe block");
// Drain the waker
loop {
let mut buf = [0u8; 64];
match wake_read.read(&mut buf) {
Ok(1..) => {}
_ => break,
}
}
// Set the self wake flag here
waker.wake_by_ref();
}
true
}
.instrument(tracing::span!(tracing::Level::TRACE, "mio poll"))
.await;
if !ret {
break;
}
let mut signaled = false;
poll_fn(|_| {
if signaled {
Poll::Ready(())
} else {
signaled = true;
Poll::Pending
}
})
.await;
}
}
};
Ok(Self {
state,
backend: BackendContainer::new_dyn(backend),
waker,
})
}
pub fn cancel_all_ops(&self) {
self.state
.files
.write()
.unique_iter()
.for_each(|v| v.lock().cancel_all_ops());
self.state
.streams
.write()
.unique_iter()
.for_each(|v| v.lock().cancel_all_ops());
}
}
impl IoBackend for Runtime {
type Backend = DynBackend;
fn polling_handle(&self) -> Option<PollingHandle> {
static READ: PollingFlags = PollingFlags::new().read(true);
Some(PollingHandle {
handle: self.state.poll.lock().as_raw_fd(),
cur_flags: &READ,
max_flags: PollingFlags::new().read(true),
waker: self.waker.clone().into_waker(),
})
}
fn get_backend(&self) -> BackendHandle<Self::Backend> {
self.backend.acquire(Some(self.waker.flags()))
}
}
impl Runtime {
pub fn register_file(&self, file: File) -> FileWrapper {
let fd = file.as_raw_fd();
set_nonblock(fd).unwrap();
let files = self.state.files.read();
let entry = files.vacant_entry().unwrap();
// 2N mapping, to accomodate for streams
let key = Key::File(entry.key());
let file = FileInner::from(file);
log::trace!(
"Register file={:?} self={:?} state={:?}: key={key:?}",
file.as_raw_fd(),
self as *const _,
self.state.as_ptr()
);
entry.insert(file.into());
FileWrapper::new(key.idx(), self.state.clone())
}
pub fn register_stream(&self, stream: std::net::TcpStream) -> TcpStream {
TcpStream::register_stream(&self.state, mio::net::TcpStream::from_std(stream))
}
pub fn register_listener(&self, listener: std::net::TcpListener) -> TcpListener {
TcpListener::register_listener(&self.state, mio::net::TcpListener::from_std(listener))
}
pub fn tcp_connect<'a, A: ToSocketAddrs + Send + 'a>(
&'a self,
addrs: A,
) -> TcpConnectFuture<'a, A> {
TcpStream::tcp_connect(&self.state, addrs)
}
}
pub use core::convert::identity as map_options;