1 | //! In-process memory IO types. |
2 | |
3 | use crate::io::{AsyncRead, AsyncWrite, ReadBuf}; |
4 | use crate::loom::sync::Mutex; |
5 | |
6 | use bytes::{Buf, BytesMut}; |
7 | use std::{ |
8 | pin::Pin, |
9 | sync::Arc, |
10 | task::{self, Poll, Waker}, |
11 | }; |
12 | |
13 | /// A bidirectional pipe to read and write bytes in memory. |
14 | /// |
15 | /// A pair of `DuplexStream`s are created together, and they act as a "channel" |
16 | /// that can be used as in-memory IO types. Writing to one of the pairs will |
17 | /// allow that data to be read from the other, and vice versa. |
18 | /// |
19 | /// # Closing a `DuplexStream` |
20 | /// |
21 | /// If one end of the `DuplexStream` channel is dropped, any pending reads on |
22 | /// the other side will continue to read data until the buffer is drained, then |
23 | /// they will signal EOF by returning 0 bytes. Any writes to the other side, |
24 | /// including pending ones (that are waiting for free space in the buffer) will |
25 | /// return `Err(BrokenPipe)` immediately. |
26 | /// |
27 | /// # Example |
28 | /// |
29 | /// ``` |
30 | /// # async fn ex() -> std::io::Result<()> { |
31 | /// # use tokio::io::{AsyncReadExt, AsyncWriteExt}; |
32 | /// let (mut client, mut server) = tokio::io::duplex(64); |
33 | /// |
34 | /// client.write_all(b"ping" ).await?; |
35 | /// |
36 | /// let mut buf = [0u8; 4]; |
37 | /// server.read_exact(&mut buf).await?; |
38 | /// assert_eq!(&buf, b"ping" ); |
39 | /// |
40 | /// server.write_all(b"pong" ).await?; |
41 | /// |
42 | /// client.read_exact(&mut buf).await?; |
43 | /// assert_eq!(&buf, b"pong" ); |
44 | /// # Ok(()) |
45 | /// # } |
46 | /// ``` |
47 | #[derive (Debug)] |
48 | #[cfg_attr (docsrs, doc(cfg(feature = "io-util" )))] |
49 | pub struct DuplexStream { |
50 | read: Arc<Mutex<Pipe>>, |
51 | write: Arc<Mutex<Pipe>>, |
52 | } |
53 | |
54 | /// A unidirectional IO over a piece of memory. |
55 | /// |
56 | /// Data can be written to the pipe, and reading will return that data. |
57 | #[derive (Debug)] |
58 | struct Pipe { |
59 | /// The buffer storing the bytes written, also read from. |
60 | /// |
61 | /// Using a `BytesMut` because it has efficient `Buf` and `BufMut` |
62 | /// functionality already. Additionally, it can try to copy data in the |
63 | /// same buffer if there read index has advanced far enough. |
64 | buffer: BytesMut, |
65 | /// Determines if the write side has been closed. |
66 | is_closed: bool, |
67 | /// The maximum amount of bytes that can be written before returning |
68 | /// `Poll::Pending`. |
69 | max_buf_size: usize, |
70 | /// If the `read` side has been polled and is pending, this is the waker |
71 | /// for that parked task. |
72 | read_waker: Option<Waker>, |
73 | /// If the `write` side has filled the `max_buf_size` and returned |
74 | /// `Poll::Pending`, this is the waker for that parked task. |
75 | write_waker: Option<Waker>, |
76 | } |
77 | |
78 | // ===== impl DuplexStream ===== |
79 | |
80 | /// Create a new pair of `DuplexStream`s that act like a pair of connected sockets. |
81 | /// |
82 | /// The `max_buf_size` argument is the maximum amount of bytes that can be |
83 | /// written to a side before the write returns `Poll::Pending`. |
84 | #[cfg_attr (docsrs, doc(cfg(feature = "io-util" )))] |
85 | pub fn duplex(max_buf_size: usize) -> (DuplexStream, DuplexStream) { |
86 | let one: Arc> = Arc::new(data:Mutex::new(Pipe::new(max_buf_size))); |
87 | let two: Arc> = Arc::new(data:Mutex::new(Pipe::new(max_buf_size))); |
88 | |
89 | ( |
90 | DuplexStream { |
91 | read: one.clone(), |
92 | write: two.clone(), |
93 | }, |
94 | DuplexStream { |
95 | read: two, |
96 | write: one, |
97 | }, |
98 | ) |
99 | } |
100 | |
101 | impl AsyncRead for DuplexStream { |
102 | // Previous rustc required this `self` to be `mut`, even though newer |
103 | // versions recognize it isn't needed to call `lock()`. So for |
104 | // compatibility, we include the `mut` and `allow` the lint. |
105 | // |
106 | // See https://github.com/rust-lang/rust/issues/73592 |
107 | #[allow (unused_mut)] |
108 | fn poll_read( |
109 | mut self: Pin<&mut Self>, |
110 | cx: &mut task::Context<'_>, |
111 | buf: &mut ReadBuf<'_>, |
112 | ) -> Poll<std::io::Result<()>> { |
113 | Pin::new(&mut *self.read.lock()).poll_read(cx, buf) |
114 | } |
115 | } |
116 | |
117 | impl AsyncWrite for DuplexStream { |
118 | #[allow (unused_mut)] |
119 | fn poll_write( |
120 | mut self: Pin<&mut Self>, |
121 | cx: &mut task::Context<'_>, |
122 | buf: &[u8], |
123 | ) -> Poll<std::io::Result<usize>> { |
124 | Pin::new(&mut *self.write.lock()).poll_write(cx, buf) |
125 | } |
126 | |
127 | fn poll_write_vectored( |
128 | self: Pin<&mut Self>, |
129 | cx: &mut task::Context<'_>, |
130 | bufs: &[std::io::IoSlice<'_>], |
131 | ) -> Poll<Result<usize, std::io::Error>> { |
132 | Pin::new(&mut *self.write.lock()).poll_write_vectored(cx, bufs) |
133 | } |
134 | |
135 | fn is_write_vectored(&self) -> bool { |
136 | true |
137 | } |
138 | |
139 | #[allow (unused_mut)] |
140 | fn poll_flush( |
141 | mut self: Pin<&mut Self>, |
142 | cx: &mut task::Context<'_>, |
143 | ) -> Poll<std::io::Result<()>> { |
144 | Pin::new(&mut *self.write.lock()).poll_flush(cx) |
145 | } |
146 | |
147 | #[allow (unused_mut)] |
148 | fn poll_shutdown( |
149 | mut self: Pin<&mut Self>, |
150 | cx: &mut task::Context<'_>, |
151 | ) -> Poll<std::io::Result<()>> { |
152 | Pin::new(&mut *self.write.lock()).poll_shutdown(cx) |
153 | } |
154 | } |
155 | |
156 | impl Drop for DuplexStream { |
157 | fn drop(&mut self) { |
158 | // notify the other side of the closure |
159 | self.write.lock().close_write(); |
160 | self.read.lock().close_read(); |
161 | } |
162 | } |
163 | |
164 | // ===== impl Pipe ===== |
165 | |
166 | impl Pipe { |
167 | fn new(max_buf_size: usize) -> Self { |
168 | Pipe { |
169 | buffer: BytesMut::new(), |
170 | is_closed: false, |
171 | max_buf_size, |
172 | read_waker: None, |
173 | write_waker: None, |
174 | } |
175 | } |
176 | |
177 | fn close_write(&mut self) { |
178 | self.is_closed = true; |
179 | // needs to notify any readers that no more data will come |
180 | if let Some(waker) = self.read_waker.take() { |
181 | waker.wake(); |
182 | } |
183 | } |
184 | |
185 | fn close_read(&mut self) { |
186 | self.is_closed = true; |
187 | // needs to notify any writers that they have to abort |
188 | if let Some(waker) = self.write_waker.take() { |
189 | waker.wake(); |
190 | } |
191 | } |
192 | |
193 | fn poll_read_internal( |
194 | mut self: Pin<&mut Self>, |
195 | cx: &mut task::Context<'_>, |
196 | buf: &mut ReadBuf<'_>, |
197 | ) -> Poll<std::io::Result<()>> { |
198 | if self.buffer.has_remaining() { |
199 | let max = self.buffer.remaining().min(buf.remaining()); |
200 | buf.put_slice(&self.buffer[..max]); |
201 | self.buffer.advance(max); |
202 | if max > 0 { |
203 | // The passed `buf` might have been empty, don't wake up if |
204 | // no bytes have been moved. |
205 | if let Some(waker) = self.write_waker.take() { |
206 | waker.wake(); |
207 | } |
208 | } |
209 | Poll::Ready(Ok(())) |
210 | } else if self.is_closed { |
211 | Poll::Ready(Ok(())) |
212 | } else { |
213 | self.read_waker = Some(cx.waker().clone()); |
214 | Poll::Pending |
215 | } |
216 | } |
217 | |
218 | fn poll_write_internal( |
219 | mut self: Pin<&mut Self>, |
220 | cx: &mut task::Context<'_>, |
221 | buf: &[u8], |
222 | ) -> Poll<std::io::Result<usize>> { |
223 | if self.is_closed { |
224 | return Poll::Ready(Err(std::io::ErrorKind::BrokenPipe.into())); |
225 | } |
226 | let avail = self.max_buf_size - self.buffer.len(); |
227 | if avail == 0 { |
228 | self.write_waker = Some(cx.waker().clone()); |
229 | return Poll::Pending; |
230 | } |
231 | |
232 | let len = buf.len().min(avail); |
233 | self.buffer.extend_from_slice(&buf[..len]); |
234 | if let Some(waker) = self.read_waker.take() { |
235 | waker.wake(); |
236 | } |
237 | Poll::Ready(Ok(len)) |
238 | } |
239 | |
240 | fn poll_write_vectored_internal( |
241 | mut self: Pin<&mut Self>, |
242 | cx: &mut task::Context<'_>, |
243 | bufs: &[std::io::IoSlice<'_>], |
244 | ) -> Poll<Result<usize, std::io::Error>> { |
245 | if self.is_closed { |
246 | return Poll::Ready(Err(std::io::ErrorKind::BrokenPipe.into())); |
247 | } |
248 | let avail = self.max_buf_size - self.buffer.len(); |
249 | if avail == 0 { |
250 | self.write_waker = Some(cx.waker().clone()); |
251 | return Poll::Pending; |
252 | } |
253 | |
254 | let mut rem = avail; |
255 | for buf in bufs { |
256 | if rem == 0 { |
257 | break; |
258 | } |
259 | |
260 | let len = buf.len().min(rem); |
261 | self.buffer.extend_from_slice(&buf[..len]); |
262 | rem -= len; |
263 | } |
264 | |
265 | if let Some(waker) = self.read_waker.take() { |
266 | waker.wake(); |
267 | } |
268 | Poll::Ready(Ok(avail - rem)) |
269 | } |
270 | } |
271 | |
272 | impl AsyncRead for Pipe { |
273 | cfg_coop! { |
274 | fn poll_read( |
275 | self: Pin<&mut Self>, |
276 | cx: &mut task::Context<'_>, |
277 | buf: &mut ReadBuf<'_>, |
278 | ) -> Poll<std::io::Result<()>> { |
279 | ready!(crate::trace::trace_leaf(cx)); |
280 | let coop = ready!(crate::runtime::coop::poll_proceed(cx)); |
281 | |
282 | let ret = self.poll_read_internal(cx, buf); |
283 | if ret.is_ready() { |
284 | coop.made_progress(); |
285 | } |
286 | ret |
287 | } |
288 | } |
289 | |
290 | cfg_not_coop! { |
291 | fn poll_read( |
292 | self: Pin<&mut Self>, |
293 | cx: &mut task::Context<'_>, |
294 | buf: &mut ReadBuf<'_>, |
295 | ) -> Poll<std::io::Result<()>> { |
296 | ready!(crate::trace::trace_leaf(cx)); |
297 | self.poll_read_internal(cx, buf) |
298 | } |
299 | } |
300 | } |
301 | |
302 | impl AsyncWrite for Pipe { |
303 | cfg_coop! { |
304 | fn poll_write( |
305 | self: Pin<&mut Self>, |
306 | cx: &mut task::Context<'_>, |
307 | buf: &[u8], |
308 | ) -> Poll<std::io::Result<usize>> { |
309 | ready!(crate::trace::trace_leaf(cx)); |
310 | let coop = ready!(crate::runtime::coop::poll_proceed(cx)); |
311 | |
312 | let ret = self.poll_write_internal(cx, buf); |
313 | if ret.is_ready() { |
314 | coop.made_progress(); |
315 | } |
316 | ret |
317 | } |
318 | } |
319 | |
320 | cfg_not_coop! { |
321 | fn poll_write( |
322 | self: Pin<&mut Self>, |
323 | cx: &mut task::Context<'_>, |
324 | buf: &[u8], |
325 | ) -> Poll<std::io::Result<usize>> { |
326 | ready!(crate::trace::trace_leaf(cx)); |
327 | self.poll_write_internal(cx, buf) |
328 | } |
329 | } |
330 | |
331 | cfg_coop! { |
332 | fn poll_write_vectored( |
333 | self: Pin<&mut Self>, |
334 | cx: &mut task::Context<'_>, |
335 | bufs: &[std::io::IoSlice<'_>], |
336 | ) -> Poll<Result<usize, std::io::Error>> { |
337 | ready!(crate::trace::trace_leaf(cx)); |
338 | let coop = ready!(crate::runtime::coop::poll_proceed(cx)); |
339 | |
340 | let ret = self.poll_write_vectored_internal(cx, bufs); |
341 | if ret.is_ready() { |
342 | coop.made_progress(); |
343 | } |
344 | ret |
345 | } |
346 | } |
347 | |
348 | cfg_not_coop! { |
349 | fn poll_write_vectored( |
350 | self: Pin<&mut Self>, |
351 | cx: &mut task::Context<'_>, |
352 | bufs: &[std::io::IoSlice<'_>], |
353 | ) -> Poll<Result<usize, std::io::Error>> { |
354 | ready!(crate::trace::trace_leaf(cx)); |
355 | self.poll_write_vectored_internal(cx, bufs) |
356 | } |
357 | } |
358 | |
359 | fn is_write_vectored(&self) -> bool { |
360 | true |
361 | } |
362 | |
363 | fn poll_flush(self: Pin<&mut Self>, _: &mut task::Context<'_>) -> Poll<std::io::Result<()>> { |
364 | Poll::Ready(Ok(())) |
365 | } |
366 | |
367 | fn poll_shutdown( |
368 | mut self: Pin<&mut Self>, |
369 | _: &mut task::Context<'_>, |
370 | ) -> Poll<std::io::Result<()>> { |
371 | self.close_write(); |
372 | Poll::Ready(Ok(())) |
373 | } |
374 | } |
375 | |