1 | use crate::future::poll_fn; |
2 | use crate::io::{AsyncRead, AsyncWrite, Interest, PollEvented, ReadBuf, Ready}; |
3 | use crate::net::unix::split::{split, ReadHalf, WriteHalf}; |
4 | use crate::net::unix::split_owned::{split_owned, OwnedReadHalf, OwnedWriteHalf}; |
5 | use crate::net::unix::ucred::{self, UCred}; |
6 | use crate::net::unix::SocketAddr; |
7 | |
8 | use std::fmt; |
9 | use std::io::{self, Read, Write}; |
10 | use std::net::Shutdown; |
11 | #[cfg (not(tokio_no_as_fd))] |
12 | use std::os::unix::io::{AsFd, BorrowedFd}; |
13 | use std::os::unix::io::{AsRawFd, FromRawFd, IntoRawFd, RawFd}; |
14 | use std::os::unix::net; |
15 | use std::path::Path; |
16 | use std::pin::Pin; |
17 | use std::task::{Context, Poll}; |
18 | |
19 | cfg_io_util! { |
20 | use bytes::BufMut; |
21 | } |
22 | |
23 | cfg_net_unix! { |
24 | /// A structure representing a connected Unix socket. |
25 | /// |
26 | /// This socket can be connected directly with [`UnixStream::connect`] or accepted |
27 | /// from a listener with [`UnixListener::accept`]. Additionally, a pair of |
28 | /// anonymous Unix sockets can be created with `UnixStream::pair`. |
29 | /// |
30 | /// To shut down the stream in the write direction, you can call the |
31 | /// [`shutdown()`] method. This will cause the other peer to receive a read of |
32 | /// length 0, indicating that no more data will be sent. This only closes |
33 | /// the stream in one direction. |
34 | /// |
35 | /// [`shutdown()`]: fn@crate::io::AsyncWriteExt::shutdown |
36 | /// [`UnixListener::accept`]: crate::net::UnixListener::accept |
37 | #[cfg_attr (docsrs, doc(alias = "uds" ))] |
38 | pub struct UnixStream { |
39 | io: PollEvented<mio::net::UnixStream>, |
40 | } |
41 | } |
42 | |
43 | impl UnixStream { |
44 | /// Connects to the socket named by `path`. |
45 | /// |
46 | /// This function will create a new Unix socket and connect to the path |
47 | /// specified, associating the returned stream with the default event loop's |
48 | /// handle. |
49 | pub async fn connect<P>(path: P) -> io::Result<UnixStream> |
50 | where |
51 | P: AsRef<Path>, |
52 | { |
53 | let stream = mio::net::UnixStream::connect(path)?; |
54 | let stream = UnixStream::new(stream)?; |
55 | |
56 | poll_fn(|cx| stream.io.registration().poll_write_ready(cx)).await?; |
57 | |
58 | if let Some(e) = stream.io.take_error()? { |
59 | return Err(e); |
60 | } |
61 | |
62 | Ok(stream) |
63 | } |
64 | |
65 | /// Waits for any of the requested ready states. |
66 | /// |
67 | /// This function is usually paired with `try_read()` or `try_write()`. It |
68 | /// can be used to concurrently read / write to the same socket on a single |
69 | /// task without splitting the socket. |
70 | /// |
71 | /// The function may complete without the socket being ready. This is a |
72 | /// false-positive and attempting an operation will return with |
73 | /// `io::ErrorKind::WouldBlock`. The function can also return with an empty |
74 | /// [`Ready`] set, so you should always check the returned value and possibly |
75 | /// wait again if the requested states are not set. |
76 | /// |
77 | /// # Cancel safety |
78 | /// |
79 | /// This method is cancel safe. Once a readiness event occurs, the method |
80 | /// will continue to return immediately until the readiness event is |
81 | /// consumed by an attempt to read or write that fails with `WouldBlock` or |
82 | /// `Poll::Pending`. |
83 | /// |
84 | /// # Examples |
85 | /// |
86 | /// Concurrently read and write to the stream on the same task without |
87 | /// splitting. |
88 | /// |
89 | /// ```no_run |
90 | /// use tokio::io::Interest; |
91 | /// use tokio::net::UnixStream; |
92 | /// use std::error::Error; |
93 | /// use std::io; |
94 | /// |
95 | /// #[tokio::main] |
96 | /// async fn main() -> Result<(), Box<dyn Error>> { |
97 | /// let dir = tempfile::tempdir().unwrap(); |
98 | /// let bind_path = dir.path().join("bind_path" ); |
99 | /// let stream = UnixStream::connect(bind_path).await?; |
100 | /// |
101 | /// loop { |
102 | /// let ready = stream.ready(Interest::READABLE | Interest::WRITABLE).await?; |
103 | /// |
104 | /// if ready.is_readable() { |
105 | /// let mut data = vec![0; 1024]; |
106 | /// // Try to read data, this may still fail with `WouldBlock` |
107 | /// // if the readiness event is a false positive. |
108 | /// match stream.try_read(&mut data) { |
109 | /// Ok(n) => { |
110 | /// println!("read {} bytes" , n); |
111 | /// } |
112 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
113 | /// continue; |
114 | /// } |
115 | /// Err(e) => { |
116 | /// return Err(e.into()); |
117 | /// } |
118 | /// } |
119 | /// |
120 | /// } |
121 | /// |
122 | /// if ready.is_writable() { |
123 | /// // Try to write data, this may still fail with `WouldBlock` |
124 | /// // if the readiness event is a false positive. |
125 | /// match stream.try_write(b"hello world" ) { |
126 | /// Ok(n) => { |
127 | /// println!("write {} bytes" , n); |
128 | /// } |
129 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
130 | /// continue; |
131 | /// } |
132 | /// Err(e) => { |
133 | /// return Err(e.into()); |
134 | /// } |
135 | /// } |
136 | /// } |
137 | /// } |
138 | /// } |
139 | /// ``` |
140 | pub async fn ready(&self, interest: Interest) -> io::Result<Ready> { |
141 | let event = self.io.registration().readiness(interest).await?; |
142 | Ok(event.ready) |
143 | } |
144 | |
145 | /// Waits for the socket to become readable. |
146 | /// |
147 | /// This function is equivalent to `ready(Interest::READABLE)` and is usually |
148 | /// paired with `try_read()`. |
149 | /// |
150 | /// # Cancel safety |
151 | /// |
152 | /// This method is cancel safe. Once a readiness event occurs, the method |
153 | /// will continue to return immediately until the readiness event is |
154 | /// consumed by an attempt to read that fails with `WouldBlock` or |
155 | /// `Poll::Pending`. |
156 | /// |
157 | /// # Examples |
158 | /// |
159 | /// ```no_run |
160 | /// use tokio::net::UnixStream; |
161 | /// use std::error::Error; |
162 | /// use std::io; |
163 | /// |
164 | /// #[tokio::main] |
165 | /// async fn main() -> Result<(), Box<dyn Error>> { |
166 | /// // Connect to a peer |
167 | /// let dir = tempfile::tempdir().unwrap(); |
168 | /// let bind_path = dir.path().join("bind_path" ); |
169 | /// let stream = UnixStream::connect(bind_path).await?; |
170 | /// |
171 | /// let mut msg = vec![0; 1024]; |
172 | /// |
173 | /// loop { |
174 | /// // Wait for the socket to be readable |
175 | /// stream.readable().await?; |
176 | /// |
177 | /// // Try to read data, this may still fail with `WouldBlock` |
178 | /// // if the readiness event is a false positive. |
179 | /// match stream.try_read(&mut msg) { |
180 | /// Ok(n) => { |
181 | /// msg.truncate(n); |
182 | /// break; |
183 | /// } |
184 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
185 | /// continue; |
186 | /// } |
187 | /// Err(e) => { |
188 | /// return Err(e.into()); |
189 | /// } |
190 | /// } |
191 | /// } |
192 | /// |
193 | /// println!("GOT = {:?}" , msg); |
194 | /// Ok(()) |
195 | /// } |
196 | /// ``` |
197 | pub async fn readable(&self) -> io::Result<()> { |
198 | self.ready(Interest::READABLE).await?; |
199 | Ok(()) |
200 | } |
201 | |
202 | /// Polls for read readiness. |
203 | /// |
204 | /// If the unix stream is not currently ready for reading, this method will |
205 | /// store a clone of the `Waker` from the provided `Context`. When the unix |
206 | /// stream becomes ready for reading, `Waker::wake` will be called on the |
207 | /// waker. |
208 | /// |
209 | /// Note that on multiple calls to `poll_read_ready` or `poll_read`, only |
210 | /// the `Waker` from the `Context` passed to the most recent call is |
211 | /// scheduled to receive a wakeup. (However, `poll_write_ready` retains a |
212 | /// second, independent waker.) |
213 | /// |
214 | /// This function is intended for cases where creating and pinning a future |
215 | /// via [`readable`] is not feasible. Where possible, using [`readable`] is |
216 | /// preferred, as this supports polling from multiple tasks at once. |
217 | /// |
218 | /// # Return value |
219 | /// |
220 | /// The function returns: |
221 | /// |
222 | /// * `Poll::Pending` if the unix stream is not ready for reading. |
223 | /// * `Poll::Ready(Ok(()))` if the unix stream is ready for reading. |
224 | /// * `Poll::Ready(Err(e))` if an error is encountered. |
225 | /// |
226 | /// # Errors |
227 | /// |
228 | /// This function may encounter any standard I/O error except `WouldBlock`. |
229 | /// |
230 | /// [`readable`]: method@Self::readable |
231 | pub fn poll_read_ready(&self, cx: &mut Context<'_>) -> Poll<io::Result<()>> { |
232 | self.io.registration().poll_read_ready(cx).map_ok(|_| ()) |
233 | } |
234 | |
235 | /// Try to read data from the stream into the provided buffer, returning how |
236 | /// many bytes were read. |
237 | /// |
238 | /// Receives any pending data from the socket but does not wait for new data |
239 | /// to arrive. On success, returns the number of bytes read. Because |
240 | /// `try_read()` is non-blocking, the buffer does not have to be stored by |
241 | /// the async task and can exist entirely on the stack. |
242 | /// |
243 | /// Usually, [`readable()`] or [`ready()`] is used with this function. |
244 | /// |
245 | /// [`readable()`]: UnixStream::readable() |
246 | /// [`ready()`]: UnixStream::ready() |
247 | /// |
248 | /// # Return |
249 | /// |
250 | /// If data is successfully read, `Ok(n)` is returned, where `n` is the |
251 | /// number of bytes read. If `n` is `0`, then it can indicate one of two scenarios: |
252 | /// |
253 | /// 1. The stream's read half is closed and will no longer yield data. |
254 | /// 2. The specified buffer was 0 bytes in length. |
255 | /// |
256 | /// If the stream is not ready to read data, |
257 | /// `Err(io::ErrorKind::WouldBlock)` is returned. |
258 | /// |
259 | /// # Examples |
260 | /// |
261 | /// ```no_run |
262 | /// use tokio::net::UnixStream; |
263 | /// use std::error::Error; |
264 | /// use std::io; |
265 | /// |
266 | /// #[tokio::main] |
267 | /// async fn main() -> Result<(), Box<dyn Error>> { |
268 | /// // Connect to a peer |
269 | /// let dir = tempfile::tempdir().unwrap(); |
270 | /// let bind_path = dir.path().join("bind_path" ); |
271 | /// let stream = UnixStream::connect(bind_path).await?; |
272 | /// |
273 | /// loop { |
274 | /// // Wait for the socket to be readable |
275 | /// stream.readable().await?; |
276 | /// |
277 | /// // Creating the buffer **after** the `await` prevents it from |
278 | /// // being stored in the async task. |
279 | /// let mut buf = [0; 4096]; |
280 | /// |
281 | /// // Try to read data, this may still fail with `WouldBlock` |
282 | /// // if the readiness event is a false positive. |
283 | /// match stream.try_read(&mut buf) { |
284 | /// Ok(0) => break, |
285 | /// Ok(n) => { |
286 | /// println!("read {} bytes" , n); |
287 | /// } |
288 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
289 | /// continue; |
290 | /// } |
291 | /// Err(e) => { |
292 | /// return Err(e.into()); |
293 | /// } |
294 | /// } |
295 | /// } |
296 | /// |
297 | /// Ok(()) |
298 | /// } |
299 | /// ``` |
300 | pub fn try_read(&self, buf: &mut [u8]) -> io::Result<usize> { |
301 | self.io |
302 | .registration() |
303 | .try_io(Interest::READABLE, || (&*self.io).read(buf)) |
304 | } |
305 | |
306 | /// Tries to read data from the stream into the provided buffers, returning |
307 | /// how many bytes were read. |
308 | /// |
309 | /// Data is copied to fill each buffer in order, with the final buffer |
310 | /// written to possibly being only partially filled. This method behaves |
311 | /// equivalently to a single call to [`try_read()`] with concatenated |
312 | /// buffers. |
313 | /// |
314 | /// Receives any pending data from the socket but does not wait for new data |
315 | /// to arrive. On success, returns the number of bytes read. Because |
316 | /// `try_read_vectored()` is non-blocking, the buffer does not have to be |
317 | /// stored by the async task and can exist entirely on the stack. |
318 | /// |
319 | /// Usually, [`readable()`] or [`ready()`] is used with this function. |
320 | /// |
321 | /// [`try_read()`]: UnixStream::try_read() |
322 | /// [`readable()`]: UnixStream::readable() |
323 | /// [`ready()`]: UnixStream::ready() |
324 | /// |
325 | /// # Return |
326 | /// |
327 | /// If data is successfully read, `Ok(n)` is returned, where `n` is the |
328 | /// number of bytes read. `Ok(0)` indicates the stream's read half is closed |
329 | /// and will no longer yield data. If the stream is not ready to read data |
330 | /// `Err(io::ErrorKind::WouldBlock)` is returned. |
331 | /// |
332 | /// # Examples |
333 | /// |
334 | /// ```no_run |
335 | /// use tokio::net::UnixStream; |
336 | /// use std::error::Error; |
337 | /// use std::io::{self, IoSliceMut}; |
338 | /// |
339 | /// #[tokio::main] |
340 | /// async fn main() -> Result<(), Box<dyn Error>> { |
341 | /// // Connect to a peer |
342 | /// let dir = tempfile::tempdir().unwrap(); |
343 | /// let bind_path = dir.path().join("bind_path" ); |
344 | /// let stream = UnixStream::connect(bind_path).await?; |
345 | /// |
346 | /// loop { |
347 | /// // Wait for the socket to be readable |
348 | /// stream.readable().await?; |
349 | /// |
350 | /// // Creating the buffer **after** the `await` prevents it from |
351 | /// // being stored in the async task. |
352 | /// let mut buf_a = [0; 512]; |
353 | /// let mut buf_b = [0; 1024]; |
354 | /// let mut bufs = [ |
355 | /// IoSliceMut::new(&mut buf_a), |
356 | /// IoSliceMut::new(&mut buf_b), |
357 | /// ]; |
358 | /// |
359 | /// // Try to read data, this may still fail with `WouldBlock` |
360 | /// // if the readiness event is a false positive. |
361 | /// match stream.try_read_vectored(&mut bufs) { |
362 | /// Ok(0) => break, |
363 | /// Ok(n) => { |
364 | /// println!("read {} bytes" , n); |
365 | /// } |
366 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
367 | /// continue; |
368 | /// } |
369 | /// Err(e) => { |
370 | /// return Err(e.into()); |
371 | /// } |
372 | /// } |
373 | /// } |
374 | /// |
375 | /// Ok(()) |
376 | /// } |
377 | /// ``` |
378 | pub fn try_read_vectored(&self, bufs: &mut [io::IoSliceMut<'_>]) -> io::Result<usize> { |
379 | self.io |
380 | .registration() |
381 | .try_io(Interest::READABLE, || (&*self.io).read_vectored(bufs)) |
382 | } |
383 | |
384 | cfg_io_util! { |
385 | /// Tries to read data from the stream into the provided buffer, advancing the |
386 | /// buffer's internal cursor, returning how many bytes were read. |
387 | /// |
388 | /// Receives any pending data from the socket but does not wait for new data |
389 | /// to arrive. On success, returns the number of bytes read. Because |
390 | /// `try_read_buf()` is non-blocking, the buffer does not have to be stored by |
391 | /// the async task and can exist entirely on the stack. |
392 | /// |
393 | /// Usually, [`readable()`] or [`ready()`] is used with this function. |
394 | /// |
395 | /// [`readable()`]: UnixStream::readable() |
396 | /// [`ready()`]: UnixStream::ready() |
397 | /// |
398 | /// # Return |
399 | /// |
400 | /// If data is successfully read, `Ok(n)` is returned, where `n` is the |
401 | /// number of bytes read. `Ok(0)` indicates the stream's read half is closed |
402 | /// and will no longer yield data. If the stream is not ready to read data |
403 | /// `Err(io::ErrorKind::WouldBlock)` is returned. |
404 | /// |
405 | /// # Examples |
406 | /// |
407 | /// ```no_run |
408 | /// use tokio::net::UnixStream; |
409 | /// use std::error::Error; |
410 | /// use std::io; |
411 | /// |
412 | /// #[tokio::main] |
413 | /// async fn main() -> Result<(), Box<dyn Error>> { |
414 | /// // Connect to a peer |
415 | /// let dir = tempfile::tempdir().unwrap(); |
416 | /// let bind_path = dir.path().join("bind_path"); |
417 | /// let stream = UnixStream::connect(bind_path).await?; |
418 | /// |
419 | /// loop { |
420 | /// // Wait for the socket to be readable |
421 | /// stream.readable().await?; |
422 | /// |
423 | /// let mut buf = Vec::with_capacity(4096); |
424 | /// |
425 | /// // Try to read data, this may still fail with `WouldBlock` |
426 | /// // if the readiness event is a false positive. |
427 | /// match stream.try_read_buf(&mut buf) { |
428 | /// Ok(0) => break, |
429 | /// Ok(n) => { |
430 | /// println!("read {} bytes", n); |
431 | /// } |
432 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
433 | /// continue; |
434 | /// } |
435 | /// Err(e) => { |
436 | /// return Err(e.into()); |
437 | /// } |
438 | /// } |
439 | /// } |
440 | /// |
441 | /// Ok(()) |
442 | /// } |
443 | /// ``` |
444 | pub fn try_read_buf<B: BufMut>(&self, buf: &mut B) -> io::Result<usize> { |
445 | self.io.registration().try_io(Interest::READABLE, || { |
446 | use std::io::Read; |
447 | |
448 | let dst = buf.chunk_mut(); |
449 | let dst = |
450 | unsafe { &mut *(dst as *mut _ as *mut [std::mem::MaybeUninit<u8>] as *mut [u8]) }; |
451 | |
452 | // Safety: We trust `UnixStream::read` to have filled up `n` bytes in the |
453 | // buffer. |
454 | let n = (&*self.io).read(dst)?; |
455 | |
456 | unsafe { |
457 | buf.advance_mut(n); |
458 | } |
459 | |
460 | Ok(n) |
461 | }) |
462 | } |
463 | } |
464 | |
465 | /// Waits for the socket to become writable. |
466 | /// |
467 | /// This function is equivalent to `ready(Interest::WRITABLE)` and is usually |
468 | /// paired with `try_write()`. |
469 | /// |
470 | /// # Cancel safety |
471 | /// |
472 | /// This method is cancel safe. Once a readiness event occurs, the method |
473 | /// will continue to return immediately until the readiness event is |
474 | /// consumed by an attempt to write that fails with `WouldBlock` or |
475 | /// `Poll::Pending`. |
476 | /// |
477 | /// # Examples |
478 | /// |
479 | /// ```no_run |
480 | /// use tokio::net::UnixStream; |
481 | /// use std::error::Error; |
482 | /// use std::io; |
483 | /// |
484 | /// #[tokio::main] |
485 | /// async fn main() -> Result<(), Box<dyn Error>> { |
486 | /// // Connect to a peer |
487 | /// let dir = tempfile::tempdir().unwrap(); |
488 | /// let bind_path = dir.path().join("bind_path" ); |
489 | /// let stream = UnixStream::connect(bind_path).await?; |
490 | /// |
491 | /// loop { |
492 | /// // Wait for the socket to be writable |
493 | /// stream.writable().await?; |
494 | /// |
495 | /// // Try to write data, this may still fail with `WouldBlock` |
496 | /// // if the readiness event is a false positive. |
497 | /// match stream.try_write(b"hello world" ) { |
498 | /// Ok(n) => { |
499 | /// break; |
500 | /// } |
501 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
502 | /// continue; |
503 | /// } |
504 | /// Err(e) => { |
505 | /// return Err(e.into()); |
506 | /// } |
507 | /// } |
508 | /// } |
509 | /// |
510 | /// Ok(()) |
511 | /// } |
512 | /// ``` |
513 | pub async fn writable(&self) -> io::Result<()> { |
514 | self.ready(Interest::WRITABLE).await?; |
515 | Ok(()) |
516 | } |
517 | |
518 | /// Polls for write readiness. |
519 | /// |
520 | /// If the unix stream is not currently ready for writing, this method will |
521 | /// store a clone of the `Waker` from the provided `Context`. When the unix |
522 | /// stream becomes ready for writing, `Waker::wake` will be called on the |
523 | /// waker. |
524 | /// |
525 | /// Note that on multiple calls to `poll_write_ready` or `poll_write`, only |
526 | /// the `Waker` from the `Context` passed to the most recent call is |
527 | /// scheduled to receive a wakeup. (However, `poll_read_ready` retains a |
528 | /// second, independent waker.) |
529 | /// |
530 | /// This function is intended for cases where creating and pinning a future |
531 | /// via [`writable`] is not feasible. Where possible, using [`writable`] is |
532 | /// preferred, as this supports polling from multiple tasks at once. |
533 | /// |
534 | /// # Return value |
535 | /// |
536 | /// The function returns: |
537 | /// |
538 | /// * `Poll::Pending` if the unix stream is not ready for writing. |
539 | /// * `Poll::Ready(Ok(()))` if the unix stream is ready for writing. |
540 | /// * `Poll::Ready(Err(e))` if an error is encountered. |
541 | /// |
542 | /// # Errors |
543 | /// |
544 | /// This function may encounter any standard I/O error except `WouldBlock`. |
545 | /// |
546 | /// [`writable`]: method@Self::writable |
547 | pub fn poll_write_ready(&self, cx: &mut Context<'_>) -> Poll<io::Result<()>> { |
548 | self.io.registration().poll_write_ready(cx).map_ok(|_| ()) |
549 | } |
550 | |
551 | /// Tries to write a buffer to the stream, returning how many bytes were |
552 | /// written. |
553 | /// |
554 | /// The function will attempt to write the entire contents of `buf`, but |
555 | /// only part of the buffer may be written. |
556 | /// |
557 | /// This function is usually paired with `writable()`. |
558 | /// |
559 | /// # Return |
560 | /// |
561 | /// If data is successfully written, `Ok(n)` is returned, where `n` is the |
562 | /// number of bytes written. If the stream is not ready to write data, |
563 | /// `Err(io::ErrorKind::WouldBlock)` is returned. |
564 | /// |
565 | /// # Examples |
566 | /// |
567 | /// ```no_run |
568 | /// use tokio::net::UnixStream; |
569 | /// use std::error::Error; |
570 | /// use std::io; |
571 | /// |
572 | /// #[tokio::main] |
573 | /// async fn main() -> Result<(), Box<dyn Error>> { |
574 | /// // Connect to a peer |
575 | /// let dir = tempfile::tempdir().unwrap(); |
576 | /// let bind_path = dir.path().join("bind_path" ); |
577 | /// let stream = UnixStream::connect(bind_path).await?; |
578 | /// |
579 | /// loop { |
580 | /// // Wait for the socket to be writable |
581 | /// stream.writable().await?; |
582 | /// |
583 | /// // Try to write data, this may still fail with `WouldBlock` |
584 | /// // if the readiness event is a false positive. |
585 | /// match stream.try_write(b"hello world" ) { |
586 | /// Ok(n) => { |
587 | /// break; |
588 | /// } |
589 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
590 | /// continue; |
591 | /// } |
592 | /// Err(e) => { |
593 | /// return Err(e.into()); |
594 | /// } |
595 | /// } |
596 | /// } |
597 | /// |
598 | /// Ok(()) |
599 | /// } |
600 | /// ``` |
601 | pub fn try_write(&self, buf: &[u8]) -> io::Result<usize> { |
602 | self.io |
603 | .registration() |
604 | .try_io(Interest::WRITABLE, || (&*self.io).write(buf)) |
605 | } |
606 | |
607 | /// Tries to write several buffers to the stream, returning how many bytes |
608 | /// were written. |
609 | /// |
610 | /// Data is written from each buffer in order, with the final buffer read |
611 | /// from possible being only partially consumed. This method behaves |
612 | /// equivalently to a single call to [`try_write()`] with concatenated |
613 | /// buffers. |
614 | /// |
615 | /// This function is usually paired with `writable()`. |
616 | /// |
617 | /// [`try_write()`]: UnixStream::try_write() |
618 | /// |
619 | /// # Return |
620 | /// |
621 | /// If data is successfully written, `Ok(n)` is returned, where `n` is the |
622 | /// number of bytes written. If the stream is not ready to write data, |
623 | /// `Err(io::ErrorKind::WouldBlock)` is returned. |
624 | /// |
625 | /// # Examples |
626 | /// |
627 | /// ```no_run |
628 | /// use tokio::net::UnixStream; |
629 | /// use std::error::Error; |
630 | /// use std::io; |
631 | /// |
632 | /// #[tokio::main] |
633 | /// async fn main() -> Result<(), Box<dyn Error>> { |
634 | /// // Connect to a peer |
635 | /// let dir = tempfile::tempdir().unwrap(); |
636 | /// let bind_path = dir.path().join("bind_path" ); |
637 | /// let stream = UnixStream::connect(bind_path).await?; |
638 | /// |
639 | /// let bufs = [io::IoSlice::new(b"hello " ), io::IoSlice::new(b"world" )]; |
640 | /// |
641 | /// loop { |
642 | /// // Wait for the socket to be writable |
643 | /// stream.writable().await?; |
644 | /// |
645 | /// // Try to write data, this may still fail with `WouldBlock` |
646 | /// // if the readiness event is a false positive. |
647 | /// match stream.try_write_vectored(&bufs) { |
648 | /// Ok(n) => { |
649 | /// break; |
650 | /// } |
651 | /// Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => { |
652 | /// continue; |
653 | /// } |
654 | /// Err(e) => { |
655 | /// return Err(e.into()); |
656 | /// } |
657 | /// } |
658 | /// } |
659 | /// |
660 | /// Ok(()) |
661 | /// } |
662 | /// ``` |
663 | pub fn try_write_vectored(&self, buf: &[io::IoSlice<'_>]) -> io::Result<usize> { |
664 | self.io |
665 | .registration() |
666 | .try_io(Interest::WRITABLE, || (&*self.io).write_vectored(buf)) |
667 | } |
668 | |
669 | /// Tries to read or write from the socket using a user-provided IO operation. |
670 | /// |
671 | /// If the socket is ready, the provided closure is called. The closure |
672 | /// should attempt to perform IO operation on the socket by manually |
673 | /// calling the appropriate syscall. If the operation fails because the |
674 | /// socket is not actually ready, then the closure should return a |
675 | /// `WouldBlock` error and the readiness flag is cleared. The return value |
676 | /// of the closure is then returned by `try_io`. |
677 | /// |
678 | /// If the socket is not ready, then the closure is not called |
679 | /// and a `WouldBlock` error is returned. |
680 | /// |
681 | /// The closure should only return a `WouldBlock` error if it has performed |
682 | /// an IO operation on the socket that failed due to the socket not being |
683 | /// ready. Returning a `WouldBlock` error in any other situation will |
684 | /// incorrectly clear the readiness flag, which can cause the socket to |
685 | /// behave incorrectly. |
686 | /// |
687 | /// The closure should not perform the IO operation using any of the methods |
688 | /// defined on the Tokio `UnixStream` type, as this will mess with the |
689 | /// readiness flag and can cause the socket to behave incorrectly. |
690 | /// |
691 | /// This method is not intended to be used with combined interests. |
692 | /// The closure should perform only one type of IO operation, so it should not |
693 | /// require more than one ready state. This method may panic or sleep forever |
694 | /// if it is called with a combined interest. |
695 | /// |
696 | /// Usually, [`readable()`], [`writable()`] or [`ready()`] is used with this function. |
697 | /// |
698 | /// [`readable()`]: UnixStream::readable() |
699 | /// [`writable()`]: UnixStream::writable() |
700 | /// [`ready()`]: UnixStream::ready() |
701 | pub fn try_io<R>( |
702 | &self, |
703 | interest: Interest, |
704 | f: impl FnOnce() -> io::Result<R>, |
705 | ) -> io::Result<R> { |
706 | self.io |
707 | .registration() |
708 | .try_io(interest, || self.io.try_io(f)) |
709 | } |
710 | |
711 | /// Reads or writes from the socket using a user-provided IO operation. |
712 | /// |
713 | /// The readiness of the socket is awaited and when the socket is ready, |
714 | /// the provided closure is called. The closure should attempt to perform |
715 | /// IO operation on the socket by manually calling the appropriate syscall. |
716 | /// If the operation fails because the socket is not actually ready, |
717 | /// then the closure should return a `WouldBlock` error. In such case the |
718 | /// readiness flag is cleared and the socket readiness is awaited again. |
719 | /// This loop is repeated until the closure returns an `Ok` or an error |
720 | /// other than `WouldBlock`. |
721 | /// |
722 | /// The closure should only return a `WouldBlock` error if it has performed |
723 | /// an IO operation on the socket that failed due to the socket not being |
724 | /// ready. Returning a `WouldBlock` error in any other situation will |
725 | /// incorrectly clear the readiness flag, which can cause the socket to |
726 | /// behave incorrectly. |
727 | /// |
728 | /// The closure should not perform the IO operation using any of the methods |
729 | /// defined on the Tokio `UnixStream` type, as this will mess with the |
730 | /// readiness flag and can cause the socket to behave incorrectly. |
731 | /// |
732 | /// This method is not intended to be used with combined interests. |
733 | /// The closure should perform only one type of IO operation, so it should not |
734 | /// require more than one ready state. This method may panic or sleep forever |
735 | /// if it is called with a combined interest. |
736 | pub async fn async_io<R>( |
737 | &self, |
738 | interest: Interest, |
739 | mut f: impl FnMut() -> io::Result<R>, |
740 | ) -> io::Result<R> { |
741 | self.io |
742 | .registration() |
743 | .async_io(interest, || self.io.try_io(&mut f)) |
744 | .await |
745 | } |
746 | |
747 | /// Creates new `UnixStream` from a `std::os::unix::net::UnixStream`. |
748 | /// |
749 | /// This function is intended to be used to wrap a UnixStream from the |
750 | /// standard library in the Tokio equivalent. |
751 | /// |
752 | /// # Notes |
753 | /// |
754 | /// The caller is responsible for ensuring that the stream is in |
755 | /// non-blocking mode. Otherwise all I/O operations on the stream |
756 | /// will block the thread, which will cause unexpected behavior. |
757 | /// Non-blocking mode can be set using [`set_nonblocking`]. |
758 | /// |
759 | /// [`set_nonblocking`]: std::os::unix::net::UnixStream::set_nonblocking |
760 | /// |
761 | /// # Examples |
762 | /// |
763 | /// ```no_run |
764 | /// use tokio::net::UnixStream; |
765 | /// use std::os::unix::net::UnixStream as StdUnixStream; |
766 | /// # use std::error::Error; |
767 | /// |
768 | /// # async fn dox() -> Result<(), Box<dyn Error>> { |
769 | /// let std_stream = StdUnixStream::connect("/path/to/the/socket" )?; |
770 | /// std_stream.set_nonblocking(true)?; |
771 | /// let stream = UnixStream::from_std(std_stream)?; |
772 | /// # Ok(()) |
773 | /// # } |
774 | /// ``` |
775 | /// |
776 | /// # Panics |
777 | /// |
778 | /// This function panics if it is not called from within a runtime with |
779 | /// IO enabled. |
780 | /// |
781 | /// The runtime is usually set implicitly when this function is called |
782 | /// from a future driven by a tokio runtime, otherwise runtime can be set |
783 | /// explicitly with [`Runtime::enter`](crate::runtime::Runtime::enter) function. |
784 | #[track_caller ] |
785 | pub fn from_std(stream: net::UnixStream) -> io::Result<UnixStream> { |
786 | let stream = mio::net::UnixStream::from_std(stream); |
787 | let io = PollEvented::new(stream)?; |
788 | |
789 | Ok(UnixStream { io }) |
790 | } |
791 | |
792 | /// Turns a [`tokio::net::UnixStream`] into a [`std::os::unix::net::UnixStream`]. |
793 | /// |
794 | /// The returned [`std::os::unix::net::UnixStream`] will have nonblocking |
795 | /// mode set as `true`. Use [`set_nonblocking`] to change the blocking |
796 | /// mode if needed. |
797 | /// |
798 | /// # Examples |
799 | /// |
800 | /// ``` |
801 | /// use std::error::Error; |
802 | /// use std::io::Read; |
803 | /// use tokio::net::UnixListener; |
804 | /// # use tokio::net::UnixStream; |
805 | /// # use tokio::io::AsyncWriteExt; |
806 | /// |
807 | /// #[tokio::main] |
808 | /// async fn main() -> Result<(), Box<dyn Error>> { |
809 | /// let dir = tempfile::tempdir().unwrap(); |
810 | /// let bind_path = dir.path().join("bind_path" ); |
811 | /// |
812 | /// let mut data = [0u8; 12]; |
813 | /// let listener = UnixListener::bind(&bind_path)?; |
814 | /// # let handle = tokio::spawn(async { |
815 | /// # let mut stream = UnixStream::connect(bind_path).await.unwrap(); |
816 | /// # stream.write(b"Hello world!" ).await.unwrap(); |
817 | /// # }); |
818 | /// let (tokio_unix_stream, _) = listener.accept().await?; |
819 | /// let mut std_unix_stream = tokio_unix_stream.into_std()?; |
820 | /// # handle.await.expect("The task being joined has panicked" ); |
821 | /// std_unix_stream.set_nonblocking(false)?; |
822 | /// std_unix_stream.read_exact(&mut data)?; |
823 | /// # assert_eq!(b"Hello world!" , &data); |
824 | /// Ok(()) |
825 | /// } |
826 | /// ``` |
827 | /// [`tokio::net::UnixStream`]: UnixStream |
828 | /// [`std::os::unix::net::UnixStream`]: std::os::unix::net::UnixStream |
829 | /// [`set_nonblocking`]: fn@std::os::unix::net::UnixStream::set_nonblocking |
830 | pub fn into_std(self) -> io::Result<std::os::unix::net::UnixStream> { |
831 | self.io |
832 | .into_inner() |
833 | .map(|io| io.into_raw_fd()) |
834 | .map(|raw_fd| unsafe { std::os::unix::net::UnixStream::from_raw_fd(raw_fd) }) |
835 | } |
836 | |
837 | /// Creates an unnamed pair of connected sockets. |
838 | /// |
839 | /// This function will create a pair of interconnected Unix sockets for |
840 | /// communicating back and forth between one another. Each socket will |
841 | /// be associated with the default event loop's handle. |
842 | pub fn pair() -> io::Result<(UnixStream, UnixStream)> { |
843 | let (a, b) = mio::net::UnixStream::pair()?; |
844 | let a = UnixStream::new(a)?; |
845 | let b = UnixStream::new(b)?; |
846 | |
847 | Ok((a, b)) |
848 | } |
849 | |
850 | pub(crate) fn new(stream: mio::net::UnixStream) -> io::Result<UnixStream> { |
851 | let io = PollEvented::new(stream)?; |
852 | Ok(UnixStream { io }) |
853 | } |
854 | |
855 | /// Returns the socket address of the local half of this connection. |
856 | /// |
857 | /// # Examples |
858 | /// |
859 | /// ```no_run |
860 | /// use tokio::net::UnixStream; |
861 | /// |
862 | /// # async fn dox() -> Result<(), Box<dyn std::error::Error>> { |
863 | /// let dir = tempfile::tempdir().unwrap(); |
864 | /// let bind_path = dir.path().join("bind_path" ); |
865 | /// let stream = UnixStream::connect(bind_path).await?; |
866 | /// |
867 | /// println!("{:?}" , stream.local_addr()?); |
868 | /// # Ok(()) |
869 | /// # } |
870 | /// ``` |
871 | pub fn local_addr(&self) -> io::Result<SocketAddr> { |
872 | self.io.local_addr().map(SocketAddr) |
873 | } |
874 | |
875 | /// Returns the socket address of the remote half of this connection. |
876 | /// |
877 | /// # Examples |
878 | /// |
879 | /// ```no_run |
880 | /// use tokio::net::UnixStream; |
881 | /// |
882 | /// # async fn dox() -> Result<(), Box<dyn std::error::Error>> { |
883 | /// let dir = tempfile::tempdir().unwrap(); |
884 | /// let bind_path = dir.path().join("bind_path" ); |
885 | /// let stream = UnixStream::connect(bind_path).await?; |
886 | /// |
887 | /// println!("{:?}" , stream.peer_addr()?); |
888 | /// # Ok(()) |
889 | /// # } |
890 | /// ``` |
891 | pub fn peer_addr(&self) -> io::Result<SocketAddr> { |
892 | self.io.peer_addr().map(SocketAddr) |
893 | } |
894 | |
895 | /// Returns effective credentials of the process which called `connect` or `pair`. |
896 | pub fn peer_cred(&self) -> io::Result<UCred> { |
897 | ucred::get_peer_cred(self) |
898 | } |
899 | |
900 | /// Returns the value of the `SO_ERROR` option. |
901 | pub fn take_error(&self) -> io::Result<Option<io::Error>> { |
902 | self.io.take_error() |
903 | } |
904 | |
905 | /// Shuts down the read, write, or both halves of this connection. |
906 | /// |
907 | /// This function will cause all pending and future I/O calls on the |
908 | /// specified portions to immediately return with an appropriate value |
909 | /// (see the documentation of `Shutdown`). |
910 | pub(super) fn shutdown_std(&self, how: Shutdown) -> io::Result<()> { |
911 | self.io.shutdown(how) |
912 | } |
913 | |
914 | // These lifetime markers also appear in the generated documentation, and make |
915 | // it more clear that this is a *borrowed* split. |
916 | #[allow (clippy::needless_lifetimes)] |
917 | /// Splits a `UnixStream` into a read half and a write half, which can be used |
918 | /// to read and write the stream concurrently. |
919 | /// |
920 | /// This method is more efficient than [`into_split`], but the halves cannot be |
921 | /// moved into independently spawned tasks. |
922 | /// |
923 | /// [`into_split`]: Self::into_split() |
924 | pub fn split<'a>(&'a mut self) -> (ReadHalf<'a>, WriteHalf<'a>) { |
925 | split(self) |
926 | } |
927 | |
928 | /// Splits a `UnixStream` into a read half and a write half, which can be used |
929 | /// to read and write the stream concurrently. |
930 | /// |
931 | /// Unlike [`split`], the owned halves can be moved to separate tasks, however |
932 | /// this comes at the cost of a heap allocation. |
933 | /// |
934 | /// **Note:** Dropping the write half will shut down the write half of the |
935 | /// stream. This is equivalent to calling [`shutdown()`] on the `UnixStream`. |
936 | /// |
937 | /// [`split`]: Self::split() |
938 | /// [`shutdown()`]: fn@crate::io::AsyncWriteExt::shutdown |
939 | pub fn into_split(self) -> (OwnedReadHalf, OwnedWriteHalf) { |
940 | split_owned(self) |
941 | } |
942 | } |
943 | |
944 | impl TryFrom<net::UnixStream> for UnixStream { |
945 | type Error = io::Error; |
946 | |
947 | /// Consumes stream, returning the tokio I/O object. |
948 | /// |
949 | /// This is equivalent to |
950 | /// [`UnixStream::from_std(stream)`](UnixStream::from_std). |
951 | fn try_from(stream: net::UnixStream) -> io::Result<Self> { |
952 | Self::from_std(stream) |
953 | } |
954 | } |
955 | |
956 | impl AsyncRead for UnixStream { |
957 | fn poll_read( |
958 | self: Pin<&mut Self>, |
959 | cx: &mut Context<'_>, |
960 | buf: &mut ReadBuf<'_>, |
961 | ) -> Poll<io::Result<()>> { |
962 | self.poll_read_priv(cx, buf) |
963 | } |
964 | } |
965 | |
966 | impl AsyncWrite for UnixStream { |
967 | fn poll_write( |
968 | self: Pin<&mut Self>, |
969 | cx: &mut Context<'_>, |
970 | buf: &[u8], |
971 | ) -> Poll<io::Result<usize>> { |
972 | self.poll_write_priv(cx, buf) |
973 | } |
974 | |
975 | fn poll_write_vectored( |
976 | self: Pin<&mut Self>, |
977 | cx: &mut Context<'_>, |
978 | bufs: &[io::IoSlice<'_>], |
979 | ) -> Poll<io::Result<usize>> { |
980 | self.poll_write_vectored_priv(cx, bufs) |
981 | } |
982 | |
983 | fn is_write_vectored(&self) -> bool { |
984 | true |
985 | } |
986 | |
987 | fn poll_flush(self: Pin<&mut Self>, _: &mut Context<'_>) -> Poll<io::Result<()>> { |
988 | Poll::Ready(Ok(())) |
989 | } |
990 | |
991 | fn poll_shutdown(self: Pin<&mut Self>, _: &mut Context<'_>) -> Poll<io::Result<()>> { |
992 | self.shutdown_std(std::net::Shutdown::Write)?; |
993 | Poll::Ready(Ok(())) |
994 | } |
995 | } |
996 | |
997 | impl UnixStream { |
998 | // == Poll IO functions that takes `&self` == |
999 | // |
1000 | // To read or write without mutable access to the `UnixStream`, combine the |
1001 | // `poll_read_ready` or `poll_write_ready` methods with the `try_read` or |
1002 | // `try_write` methods. |
1003 | |
1004 | pub(crate) fn poll_read_priv( |
1005 | &self, |
1006 | cx: &mut Context<'_>, |
1007 | buf: &mut ReadBuf<'_>, |
1008 | ) -> Poll<io::Result<()>> { |
1009 | // Safety: `UnixStream::read` correctly handles reads into uninitialized memory |
1010 | unsafe { self.io.poll_read(cx, buf) } |
1011 | } |
1012 | |
1013 | pub(crate) fn poll_write_priv( |
1014 | &self, |
1015 | cx: &mut Context<'_>, |
1016 | buf: &[u8], |
1017 | ) -> Poll<io::Result<usize>> { |
1018 | self.io.poll_write(cx, buf) |
1019 | } |
1020 | |
1021 | pub(super) fn poll_write_vectored_priv( |
1022 | &self, |
1023 | cx: &mut Context<'_>, |
1024 | bufs: &[io::IoSlice<'_>], |
1025 | ) -> Poll<io::Result<usize>> { |
1026 | self.io.poll_write_vectored(cx, bufs) |
1027 | } |
1028 | } |
1029 | |
1030 | impl fmt::Debug for UnixStream { |
1031 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
1032 | self.io.fmt(f) |
1033 | } |
1034 | } |
1035 | |
1036 | impl AsRawFd for UnixStream { |
1037 | fn as_raw_fd(&self) -> RawFd { |
1038 | self.io.as_raw_fd() |
1039 | } |
1040 | } |
1041 | |
1042 | #[cfg (not(tokio_no_as_fd))] |
1043 | impl AsFd for UnixStream { |
1044 | fn as_fd(&self) -> BorrowedFd<'_> { |
1045 | unsafe { BorrowedFd::borrow_raw(self.as_raw_fd()) } |
1046 | } |
1047 | } |
1048 | |