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
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
use crate::bindings::sockets::tcp::ErrorCode;
use crate::host::network;
use crate::network::SocketAddressFamily;
use crate::runtime::{with_ambient_tokio_runtime, AbortOnDropJoinHandle};
use crate::{
    HostInputStream, HostOutputStream, InputStream, OutputStream, SocketResult, StreamError,
    Subscribe,
};
use anyhow::{Error, Result};
use cap_net_ext::AddressFamily;
use futures::Future;
use io_lifetimes::views::SocketlikeView;
use io_lifetimes::AsSocketlike;
use rustix::io::Errno;
use rustix::net::sockopt;
use std::io;
use std::mem;
use std::net::SocketAddr;
use std::pin::Pin;
use std::sync::Arc;
use std::task::Poll;

/// Value taken from rust std library.
const DEFAULT_BACKLOG: u32 = 128;

/// The state of a TCP socket.
///
/// This represents the various states a socket can be in during the
/// activities of binding, listening, accepting, and connecting.
enum TcpState {
    /// The initial state for a newly-created socket.
    Default(tokio::net::TcpSocket),

    /// Binding started via `start_bind`.
    BindStarted(tokio::net::TcpSocket),

    /// Binding finished via `finish_bind`. The socket has an address but
    /// is not yet listening for connections.
    Bound(tokio::net::TcpSocket),

    /// Listening started via `listen_start`.
    ListenStarted(tokio::net::TcpSocket),

    /// The socket is now listening and waiting for an incoming connection.
    Listening {
        listener: tokio::net::TcpListener,
        pending_accept: Option<io::Result<tokio::net::TcpStream>>,
    },

    /// An outgoing connection is started via `start_connect`.
    Connecting(Pin<Box<dyn Future<Output = io::Result<tokio::net::TcpStream>> + Send>>),

    /// An outgoing connection is ready to be established.
    ConnectReady(io::Result<tokio::net::TcpStream>),

    /// An outgoing connection has been established.
    Connected(Arc<tokio::net::TcpStream>),

    Closed,
}

impl std::fmt::Debug for TcpState {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Self::Default(_) => f.debug_tuple("Default").finish(),
            Self::BindStarted(_) => f.debug_tuple("BindStarted").finish(),
            Self::Bound(_) => f.debug_tuple("Bound").finish(),
            Self::ListenStarted(_) => f.debug_tuple("ListenStarted").finish(),
            Self::Listening { pending_accept, .. } => f
                .debug_struct("Listening")
                .field("pending_accept", pending_accept)
                .finish(),
            Self::Connecting(_) => f.debug_tuple("Connecting").finish(),
            Self::ConnectReady(_) => f.debug_tuple("ConnectReady").finish(),
            Self::Connected(_) => f.debug_tuple("Connected").finish(),
            Self::Closed => write!(f, "Closed"),
        }
    }
}

/// A host TCP socket, plus associated bookkeeping.
pub struct TcpSocket {
    /// The current state in the bind/listen/accept/connect progression.
    tcp_state: TcpState,

    /// The desired listen queue size.
    listen_backlog_size: u32,

    family: SocketAddressFamily,

    // The socket options below are not automatically inherited from the listener
    // on all platforms. So we keep track of which options have been explicitly
    // set and manually apply those values to newly accepted clients.
    #[cfg(target_os = "macos")]
    receive_buffer_size: Option<usize>,
    #[cfg(target_os = "macos")]
    send_buffer_size: Option<usize>,
    #[cfg(target_os = "macos")]
    hop_limit: Option<u8>,
    #[cfg(target_os = "macos")]
    keep_alive_idle_time: Option<std::time::Duration>,
}

impl TcpSocket {
    /// Create a new socket in the given family.
    pub fn new(family: AddressFamily) -> io::Result<Self> {
        with_ambient_tokio_runtime(|| {
            let (socket, family) = match family {
                AddressFamily::Ipv4 => {
                    let socket = tokio::net::TcpSocket::new_v4()?;
                    (socket, SocketAddressFamily::Ipv4)
                }
                AddressFamily::Ipv6 => {
                    let socket = tokio::net::TcpSocket::new_v6()?;
                    sockopt::set_ipv6_v6only(&socket, true)?;
                    (socket, SocketAddressFamily::Ipv6)
                }
            };

            Self::from_state(TcpState::Default(socket), family)
        })
    }

    /// Create a `TcpSocket` from an existing socket.
    fn from_state(state: TcpState, family: SocketAddressFamily) -> io::Result<Self> {
        Ok(Self {
            tcp_state: state,
            listen_backlog_size: DEFAULT_BACKLOG,
            family,
            #[cfg(target_os = "macos")]
            receive_buffer_size: None,
            #[cfg(target_os = "macos")]
            send_buffer_size: None,
            #[cfg(target_os = "macos")]
            hop_limit: None,
            #[cfg(target_os = "macos")]
            keep_alive_idle_time: None,
        })
    }

    fn as_std_view(&self) -> SocketResult<SocketlikeView<'_, std::net::TcpStream>> {
        use crate::bindings::sockets::network::ErrorCode;

        match &self.tcp_state {
            TcpState::Default(socket) | TcpState::Bound(socket) => {
                Ok(socket.as_socketlike_view::<std::net::TcpStream>())
            }
            TcpState::Connected(stream) => Ok(stream.as_socketlike_view::<std::net::TcpStream>()),
            TcpState::Listening { listener, .. } => {
                Ok(listener.as_socketlike_view::<std::net::TcpStream>())
            }

            TcpState::BindStarted(..)
            | TcpState::ListenStarted(..)
            | TcpState::Connecting(..)
            | TcpState::ConnectReady(..)
            | TcpState::Closed => Err(ErrorCode::InvalidState.into()),
        }
    }
}

impl TcpSocket {
    pub fn start_bind(&mut self, local_address: SocketAddr) -> io::Result<()> {
        let tokio_socket = match &self.tcp_state {
            TcpState::Default(socket) => socket,
            TcpState::BindStarted(..) => return Err(Errno::ALREADY.into()),
            _ => return Err(Errno::ISCONN.into()),
        };

        network::util::validate_unicast(&local_address)?;
        network::util::validate_address_family(&local_address, &self.family)?;

        {
            // Automatically bypass the TIME_WAIT state when the user is trying
            // to bind to a specific port:
            let reuse_addr = local_address.port() > 0;

            // Unconditionally (re)set SO_REUSEADDR, even when the value is false.
            // This ensures we're not accidentally affected by any socket option
            // state left behind by a previous failed call to this method (start_bind).
            network::util::set_tcp_reuseaddr(&tokio_socket, reuse_addr)?;

            // Perform the OS bind call.
            tokio_socket.bind(local_address).map_err(|error| {
                match Errno::from_io_error(&error) {
                    // From https://pubs.opengroup.org/onlinepubs/9699919799/functions/bind.html:
                    // > [EAFNOSUPPORT] The specified address is not a valid address for the address family of the specified socket
                    //
                    // The most common reasons for this error should have already
                    // been handled by our own validation slightly higher up in this
                    // function. This error mapping is here just in case there is
                    // an edge case we didn't catch.
                    Some(Errno::AFNOSUPPORT) =>  io::Error::new(
                        io::ErrorKind::InvalidInput,
                        "The specified address is not a valid address for the address family of the specified socket",
                    ),

                    // See: https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-bind#:~:text=WSAENOBUFS
                    // Windows returns WSAENOBUFS when the ephemeral ports have been exhausted.
                    #[cfg(windows)]
                    Some(Errno::NOBUFS) => io::Error::new(io::ErrorKind::AddrInUse, "no more free local ports"),

                    _ => error,
                }
            })?;

            self.tcp_state = match std::mem::replace(&mut self.tcp_state, TcpState::Closed) {
                TcpState::Default(socket) => TcpState::BindStarted(socket),
                _ => unreachable!(),
            };

            Ok(())
        }
    }

    pub fn finish_bind(&mut self) -> SocketResult<()> {
        match std::mem::replace(&mut self.tcp_state, TcpState::Closed) {
            TcpState::BindStarted(socket) => {
                self.tcp_state = TcpState::Bound(socket);
                Ok(())
            }
            current_state => {
                // Reset the state so that the outside world doesn't see this socket as closed
                self.tcp_state = current_state;
                Err(ErrorCode::NotInProgress.into())
            }
        }
    }

    pub fn start_connect(&mut self, remote_address: SocketAddr) -> SocketResult<()> {
        match self.tcp_state {
            TcpState::Default(..) => {}

            TcpState::Connecting(..) | TcpState::ConnectReady(..) => {
                return Err(ErrorCode::ConcurrencyConflict.into())
            }

            _ => return Err(ErrorCode::InvalidState.into()),
        };

        network::util::validate_unicast(&remote_address)?;
        network::util::validate_remote_address(&remote_address)?;
        network::util::validate_address_family(&remote_address, &self.family)?;

        let TcpState::Default(tokio_socket) =
            std::mem::replace(&mut self.tcp_state, TcpState::Closed)
        else {
            unreachable!();
        };

        let future = tokio_socket.connect(remote_address);

        self.tcp_state = TcpState::Connecting(Box::pin(future));
        Ok(())
    }

    pub fn finish_connect(&mut self) -> SocketResult<(InputStream, OutputStream)> {
        let previous_state = std::mem::replace(&mut self.tcp_state, TcpState::Closed);
        let result = match previous_state {
            TcpState::ConnectReady(result) => result,
            TcpState::Connecting(mut future) => {
                let mut cx = std::task::Context::from_waker(futures::task::noop_waker_ref());
                match with_ambient_tokio_runtime(|| future.as_mut().poll(&mut cx)) {
                    Poll::Ready(result) => result,
                    Poll::Pending => {
                        self.tcp_state = TcpState::Connecting(future);
                        return Err(ErrorCode::WouldBlock.into());
                    }
                }
            }
            previous_state => {
                self.tcp_state = previous_state;
                return Err(ErrorCode::NotInProgress.into());
            }
        };

        match result {
            Ok(stream) => {
                let stream = Arc::new(stream);
                self.tcp_state = TcpState::Connected(stream.clone());
                let input: InputStream = Box::new(TcpReadStream::new(stream.clone()));
                let output: OutputStream = Box::new(TcpWriteStream::new(stream));
                Ok((input, output))
            }
            Err(err) => {
                self.tcp_state = TcpState::Closed;
                Err(err.into())
            }
        }
    }

    pub fn start_listen(&mut self) -> SocketResult<()> {
        match std::mem::replace(&mut self.tcp_state, TcpState::Closed) {
            TcpState::Bound(tokio_socket) => {
                self.tcp_state = TcpState::ListenStarted(tokio_socket);
                Ok(())
            }
            TcpState::ListenStarted(tokio_socket) => {
                self.tcp_state = TcpState::ListenStarted(tokio_socket);
                Err(ErrorCode::ConcurrencyConflict.into())
            }
            previous_state => {
                self.tcp_state = previous_state;
                Err(ErrorCode::InvalidState.into())
            }
        }
    }

    pub fn finish_listen(&mut self) -> SocketResult<()> {
        let tokio_socket = match std::mem::replace(&mut self.tcp_state, TcpState::Closed) {
            TcpState::ListenStarted(tokio_socket) => tokio_socket,
            previous_state => {
                self.tcp_state = previous_state;
                return Err(ErrorCode::NotInProgress.into());
            }
        };

        match with_ambient_tokio_runtime(|| tokio_socket.listen(self.listen_backlog_size)) {
            Ok(listener) => {
                self.tcp_state = TcpState::Listening {
                    listener,
                    pending_accept: None,
                };
                Ok(())
            }
            Err(err) => {
                self.tcp_state = TcpState::Closed;

                Err(match Errno::from_io_error(&err) {
                    // See: https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-listen#:~:text=WSAEMFILE
                    // According to the docs, `listen` can return EMFILE on Windows.
                    // This is odd, because we're not trying to create a new socket
                    // or file descriptor of any kind. So we rewrite it to less
                    // surprising error code.
                    //
                    // At the time of writing, this behavior has never been experimentally
                    // observed by any of the wasmtime authors, so we're relying fully
                    // on Microsoft's documentation here.
                    #[cfg(windows)]
                    Some(Errno::MFILE) => Errno::NOBUFS.into(),

                    _ => err.into(),
                })
            }
        }
    }

    pub fn accept(&mut self) -> SocketResult<(Self, InputStream, OutputStream)> {
        let TcpState::Listening {
            listener,
            pending_accept,
        } = &mut self.tcp_state
        else {
            return Err(ErrorCode::InvalidState.into());
        };

        let result = match pending_accept.take() {
            Some(result) => result,
            None => {
                let mut cx = std::task::Context::from_waker(futures::task::noop_waker_ref());
                match with_ambient_tokio_runtime(|| listener.poll_accept(&mut cx))
                    .map_ok(|(stream, _)| stream)
                {
                    Poll::Ready(result) => result,
                    Poll::Pending => Err(Errno::WOULDBLOCK.into()),
                }
            }
        };

        let client = result.map_err(|err| match Errno::from_io_error(&err) {
            // From: https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-accept#:~:text=WSAEINPROGRESS
            // > WSAEINPROGRESS: A blocking Windows Sockets 1.1 call is in progress,
            // > or the service provider is still processing a callback function.
            //
            // wasi-sockets doesn't have an equivalent to the EINPROGRESS error,
            // because in POSIX this error is only returned by a non-blocking
            // `connect` and wasi-sockets has a different solution for that.
            #[cfg(windows)]
            Some(Errno::INPROGRESS) => Errno::INTR.into(),

            // Normalize Linux' non-standard behavior.
            //
            // From https://man7.org/linux/man-pages/man2/accept.2.html:
            // > Linux accept() passes already-pending network errors on the
            // > new socket as an error code from accept(). This behavior
            // > differs from other BSD socket implementations. (...)
            #[cfg(target_os = "linux")]
            Some(
                Errno::CONNRESET
                | Errno::NETRESET
                | Errno::HOSTUNREACH
                | Errno::HOSTDOWN
                | Errno::NETDOWN
                | Errno::NETUNREACH
                | Errno::PROTO
                | Errno::NOPROTOOPT
                | Errno::NONET
                | Errno::OPNOTSUPP,
            ) => Errno::CONNABORTED.into(),

            _ => err,
        })?;

        #[cfg(target_os = "macos")]
        {
            // Manually inherit socket options from listener. We only have to
            // do this on platforms that don't already do this automatically
            // and only if a specific value was explicitly set on the listener.

            if let Some(size) = self.receive_buffer_size {
                _ = network::util::set_socket_recv_buffer_size(&client, size); // Ignore potential error.
            }

            if let Some(size) = self.send_buffer_size {
                _ = network::util::set_socket_send_buffer_size(&client, size); // Ignore potential error.
            }

            // For some reason, IP_TTL is inherited, but IPV6_UNICAST_HOPS isn't.
            if let (SocketAddressFamily::Ipv6, Some(ttl)) = (self.family, self.hop_limit) {
                _ = network::util::set_ipv6_unicast_hops(&client, ttl); // Ignore potential error.
            }

            if let Some(value) = self.keep_alive_idle_time {
                _ = network::util::set_tcp_keepidle(&client, value); // Ignore potential error.
            }
        }

        let client = Arc::new(client);

        let input: InputStream = Box::new(TcpReadStream::new(client.clone()));
        let output: OutputStream = Box::new(TcpWriteStream::new(client.clone()));
        let tcp_socket = TcpSocket::from_state(TcpState::Connected(client), self.family)?;

        Ok((tcp_socket, input, output))
    }

    pub fn local_address(&self) -> SocketResult<SocketAddr> {
        let view = match self.tcp_state {
            TcpState::Default(..) => return Err(ErrorCode::InvalidState.into()),
            TcpState::BindStarted(..) => return Err(ErrorCode::ConcurrencyConflict.into()),
            _ => self.as_std_view()?,
        };

        Ok(view.local_addr()?)
    }

    pub fn remote_address(&self) -> SocketResult<SocketAddr> {
        let view = match self.tcp_state {
            TcpState::Connected(..) => self.as_std_view()?,
            TcpState::Connecting(..) | TcpState::ConnectReady(..) => {
                return Err(ErrorCode::ConcurrencyConflict.into())
            }
            _ => return Err(ErrorCode::InvalidState.into()),
        };

        Ok(view.peer_addr()?)
    }

    pub fn is_listening(&self) -> bool {
        matches!(self.tcp_state, TcpState::Listening { .. })
    }

    pub fn address_family(&self) -> SocketAddressFamily {
        self.family
    }

    pub fn set_listen_backlog_size(&mut self, value: u32) -> SocketResult<()> {
        const MIN_BACKLOG: u32 = 1;
        const MAX_BACKLOG: u32 = i32::MAX as u32; // OS'es will most likely limit it down even further.

        if value == 0 {
            return Err(ErrorCode::InvalidArgument.into());
        }

        // Silently clamp backlog size. This is OK for us to do, because operating systems do this too.
        let value = value.clamp(MIN_BACKLOG, MAX_BACKLOG);

        match &self.tcp_state {
            TcpState::Default(..) | TcpState::Bound(..) => {
                // Socket not listening yet. Stash value for first invocation to `listen`.
            }
            TcpState::Listening { listener, .. } => {
                // Try to update the backlog by calling `listen` again.
                // Not all platforms support this. We'll only update our own value if the OS supports changing the backlog size after the fact.

                rustix::net::listen(&listener, value.try_into().unwrap())
                    .map_err(|_| ErrorCode::NotSupported)?;
            }
            _ => return Err(ErrorCode::InvalidState.into()),
        }
        self.listen_backlog_size = value;

        Ok(())
    }

    pub fn keep_alive_enabled(&self) -> SocketResult<bool> {
        let view = &*self.as_std_view()?;
        Ok(sockopt::get_socket_keepalive(view)?)
    }

    pub fn set_keep_alive_enabled(&self, value: bool) -> SocketResult<()> {
        let view = &*self.as_std_view()?;
        Ok(sockopt::set_socket_keepalive(view, value)?)
    }

    pub fn keep_alive_idle_time(&self) -> SocketResult<std::time::Duration> {
        let view = &*self.as_std_view()?;
        Ok(sockopt::get_tcp_keepidle(view)?)
    }

    pub fn set_keep_alive_idle_time(&mut self, duration: std::time::Duration) -> SocketResult<()> {
        {
            let view = &*self.as_std_view()?;
            network::util::set_tcp_keepidle(view, duration)?;
        }

        #[cfg(target_os = "macos")]
        {
            self.keep_alive_idle_time = Some(duration);
        }

        Ok(())
    }

    pub fn keep_alive_interval(&self) -> SocketResult<std::time::Duration> {
        let view = &*self.as_std_view()?;
        Ok(sockopt::get_tcp_keepintvl(view)?)
    }

    pub fn set_keep_alive_interval(&self, duration: std::time::Duration) -> SocketResult<()> {
        let view = &*self.as_std_view()?;
        Ok(network::util::set_tcp_keepintvl(view, duration)?)
    }

    pub fn keep_alive_count(&self) -> SocketResult<u32> {
        let view = &*self.as_std_view()?;
        Ok(sockopt::get_tcp_keepcnt(view)?)
    }

    pub fn set_keep_alive_count(&self, value: u32) -> SocketResult<()> {
        let view = &*self.as_std_view()?;
        Ok(network::util::set_tcp_keepcnt(view, value)?)
    }

    pub fn hop_limit(&self) -> SocketResult<u8> {
        let view = &*self.as_std_view()?;

        let ttl = match self.family {
            SocketAddressFamily::Ipv4 => network::util::get_ip_ttl(view)?,
            SocketAddressFamily::Ipv6 => network::util::get_ipv6_unicast_hops(view)?,
        };

        Ok(ttl)
    }

    pub fn set_hop_limit(&mut self, value: u8) -> SocketResult<()> {
        {
            let view = &*self.as_std_view()?;

            match self.family {
                SocketAddressFamily::Ipv4 => network::util::set_ip_ttl(view, value)?,
                SocketAddressFamily::Ipv6 => network::util::set_ipv6_unicast_hops(view, value)?,
            }
        }

        #[cfg(target_os = "macos")]
        {
            self.hop_limit = Some(value);
        }

        Ok(())
    }

    pub fn receive_buffer_size(&self) -> SocketResult<usize> {
        let view = &*self.as_std_view()?;

        Ok(network::util::get_socket_recv_buffer_size(view)?)
    }

    pub fn set_receive_buffer_size(&mut self, value: usize) -> SocketResult<()> {
        {
            let view = &*self.as_std_view()?;

            network::util::set_socket_recv_buffer_size(view, value)?;
        }

        #[cfg(target_os = "macos")]
        {
            self.receive_buffer_size = Some(value);
        }

        Ok(())
    }

    pub fn send_buffer_size(&self) -> SocketResult<usize> {
        let view = &*self.as_std_view()?;

        Ok(network::util::get_socket_send_buffer_size(view)?)
    }

    pub fn set_send_buffer_size(&mut self, value: usize) -> SocketResult<()> {
        {
            let view = &*self.as_std_view()?;

            network::util::set_socket_send_buffer_size(view, value)?;
        }

        #[cfg(target_os = "macos")]
        {
            self.send_buffer_size = Some(value);
        }

        Ok(())
    }

    pub fn shutdown(&self, how: std::net::Shutdown) -> io::Result<()> {
        let stream = match &self.tcp_state {
            TcpState::Connected(stream) => stream,
            _ => {
                return Err(io::Error::new(
                    io::ErrorKind::NotConnected,
                    "socket not connected",
                ))
            }
        };

        stream
            .as_socketlike_view::<std::net::TcpStream>()
            .shutdown(how)?;
        Ok(())
    }
}

#[async_trait::async_trait]
impl Subscribe for TcpSocket {
    async fn ready(&mut self) {
        match &mut self.tcp_state {
            TcpState::Default(..)
            | TcpState::BindStarted(..)
            | TcpState::Bound(..)
            | TcpState::ListenStarted(..)
            | TcpState::ConnectReady(..)
            | TcpState::Closed
            | TcpState::Connected(..) => {
                // No async operation in progress.
            }
            TcpState::Connecting(future) => {
                self.tcp_state = TcpState::ConnectReady(future.as_mut().await);
            }
            TcpState::Listening {
                listener,
                pending_accept,
            } => match pending_accept {
                Some(_) => {}
                None => {
                    let result = futures::future::poll_fn(|cx| {
                        listener.poll_accept(cx).map_ok(|(stream, _)| stream)
                    })
                    .await;
                    *pending_accept = Some(result);
                }
            },
        }
    }
}

struct TcpReadStream {
    stream: Arc<tokio::net::TcpStream>,
    closed: bool,
}

impl TcpReadStream {
    fn new(stream: Arc<tokio::net::TcpStream>) -> Self {
        Self {
            stream,
            closed: false,
        }
    }
}

#[async_trait::async_trait]
impl HostInputStream for TcpReadStream {
    fn read(&mut self, size: usize) -> Result<bytes::Bytes, StreamError> {
        if self.closed {
            return Err(StreamError::Closed);
        }
        if size == 0 {
            return Ok(bytes::Bytes::new());
        }

        let mut buf = bytes::BytesMut::with_capacity(size);
        let n = match self.stream.try_read_buf(&mut buf) {
            // A 0-byte read indicates that the stream has closed.
            Ok(0) => {
                self.closed = true;
                return Err(StreamError::Closed);
            }
            Ok(n) => n,

            // Failing with `EWOULDBLOCK` is how we differentiate between a closed channel and no
            // data to read right now.
            Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => 0,

            Err(e) => {
                self.closed = true;
                return Err(StreamError::LastOperationFailed(e.into()));
            }
        };

        buf.truncate(n);
        Ok(buf.freeze())
    }
}

#[async_trait::async_trait]
impl Subscribe for TcpReadStream {
    async fn ready(&mut self) {
        if self.closed {
            return;
        }
        self.stream.readable().await.unwrap();
    }
}

const SOCKET_READY_SIZE: usize = 1024 * 1024 * 1024;

struct TcpWriteStream {
    stream: Arc<tokio::net::TcpStream>,
    last_write: LastWrite,
}

enum LastWrite {
    Waiting(AbortOnDropJoinHandle<Result<()>>),
    Error(Error),
    Done,
    Closed,
}

impl TcpWriteStream {
    fn new(stream: Arc<tokio::net::TcpStream>) -> Self {
        Self {
            stream,
            last_write: LastWrite::Done,
        }
    }

    fn try_write_portable(stream: &tokio::net::TcpStream, buf: &[u8]) -> io::Result<usize> {
        stream.try_write(buf).map_err(|error| {
            match Errno::from_io_error(&error) {
                // Windows returns `WSAESHUTDOWN` when writing to a shut down socket.
                // We normalize this to EPIPE, because that is what the other platforms return.
                // See: https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-send#:~:text=WSAESHUTDOWN
                #[cfg(windows)]
                Some(Errno::SHUTDOWN) => io::Error::new(io::ErrorKind::BrokenPipe, error),

                _ => error,
            }
        })
    }

    /// Write `bytes` in a background task, remembering the task handle for use in a future call to
    /// `write_ready`
    fn background_write(&mut self, mut bytes: bytes::Bytes) {
        assert!(matches!(self.last_write, LastWrite::Done));

        let stream = self.stream.clone();
        self.last_write = LastWrite::Waiting(crate::runtime::spawn(async move {
            // Note: we are not using the AsyncWrite impl here, and instead using the TcpStream
            // primitive try_write, which goes directly to attempt a write with mio. This has
            // two advantages: 1. this operation takes a &TcpStream instead of a &mut TcpStream
            // required to AsyncWrite, and 2. it eliminates any buffering in tokio we may need
            // to flush.
            while !bytes.is_empty() {
                stream.writable().await?;
                match Self::try_write_portable(&stream, &bytes) {
                    Ok(n) => {
                        let _ = bytes.split_to(n);
                    }
                    Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => continue,
                    Err(e) => return Err(e.into()),
                }
            }

            Ok(())
        }));
    }
}

#[async_trait::async_trait]
impl HostOutputStream for TcpWriteStream {
    fn write(&mut self, mut bytes: bytes::Bytes) -> Result<(), StreamError> {
        match self.last_write {
            LastWrite::Done => {}
            LastWrite::Waiting(_) | LastWrite::Error(_) | LastWrite::Closed => {
                return Err(StreamError::Trap(anyhow::anyhow!(
                    "unpermitted: must call check_write first"
                )));
            }
        }
        while !bytes.is_empty() {
            match Self::try_write_portable(&self.stream, &bytes) {
                Ok(n) => {
                    let _ = bytes.split_to(n);
                }

                Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => {
                    // As `try_write` indicated that it would have blocked, we'll perform the write
                    // in the background to allow us to return immediately.
                    self.background_write(bytes);

                    return Ok(());
                }

                Err(e) if e.kind() == std::io::ErrorKind::BrokenPipe => {
                    self.last_write = LastWrite::Closed;
                    return Err(StreamError::Closed);
                }

                Err(e) => return Err(StreamError::LastOperationFailed(e.into())),
            }
        }

        Ok(())
    }

    fn flush(&mut self) -> Result<(), StreamError> {
        // `flush` is a no-op here, as we're not managing any internal buffer. Additionally,
        // `write_ready` will join the background write task if it's active, so following `flush`
        // with `write_ready` will have the desired effect.
        match self.last_write {
            LastWrite::Done | LastWrite::Waiting(_) | LastWrite::Error(_) => Ok(()),
            LastWrite::Closed => Err(StreamError::Closed),
        }
    }

    fn check_write(&mut self) -> Result<usize, StreamError> {
        match mem::replace(&mut self.last_write, LastWrite::Closed) {
            LastWrite::Waiting(task) => {
                self.last_write = LastWrite::Waiting(task);
                return Ok(0);
            }
            LastWrite::Done => {
                self.last_write = LastWrite::Done;
            }
            LastWrite::Closed => return Err(StreamError::Closed),
            LastWrite::Error(e) => return Err(StreamError::LastOperationFailed(e.into())),
        }

        let writable = self.stream.writable();
        futures::pin_mut!(writable);
        if crate::runtime::poll_noop(writable).is_none() {
            return Ok(0);
        }
        Ok(SOCKET_READY_SIZE)
    }
    async fn cancel(&mut self) {
        match mem::replace(&mut self.last_write, LastWrite::Closed) {
            LastWrite::Waiting(task) => _ = task.abort_wait().await,
            _ => {}
        }
    }
}

#[async_trait::async_trait]
impl Subscribe for TcpWriteStream {
    async fn ready(&mut self) {
        if let LastWrite::Waiting(task) = &mut self.last_write {
            self.last_write = match task.await {
                Ok(()) => LastWrite::Done,
                Err(e) => LastWrite::Error(e),
            };
        }
        if let LastWrite::Done = self.last_write {
            self.stream.writable().await.unwrap();
        }
    }
}