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
use std::fmt;
use std::io;
use std::net::{SocketAddr, TcpListener as StdTcpListener};
use std::time::Duration;

use tokio::net::TcpListener;
use tokio::time::Sleep;
use tracing::{debug, error, trace};

use crate::common::{task, Future, Pin, Poll};

#[allow(unreachable_pub)] // https://github.com/rust-lang/rust/issues/57411
pub use self::addr_stream::AddrStream;
use super::accept::Accept;

/// A stream of connections from binding to an address.
#[must_use = "streams do nothing unless polled"]
pub struct AddrIncoming {
    addr: SocketAddr,
    listener: TcpListener,
    sleep_on_errors: bool,
    tcp_keepalive_timeout: Option<Duration>,
    tcp_nodelay: bool,
    timeout: Option<Pin<Box<Sleep>>>,
}

impl AddrIncoming {
    pub(super) fn new(addr: &SocketAddr) -> crate::Result<Self> {
        let std_listener = StdTcpListener::bind(addr).map_err(crate::Error::new_listen)?;

        AddrIncoming::from_std(std_listener)
    }

    pub(super) fn from_std(std_listener: StdTcpListener) -> crate::Result<Self> {
        // TcpListener::from_std doesn't set O_NONBLOCK
        std_listener
            .set_nonblocking(true)
            .map_err(crate::Error::new_listen)?;
        let listener = TcpListener::from_std(std_listener).map_err(crate::Error::new_listen)?;
        AddrIncoming::from_listener(listener)
    }

    /// Creates a new `AddrIncoming` binding to provided socket address.
    pub fn bind(addr: &SocketAddr) -> crate::Result<Self> {
        AddrIncoming::new(addr)
    }

    /// Creates a new `AddrIncoming` from an existing `tokio::net::TcpListener`.
    pub fn from_listener(listener: TcpListener) -> crate::Result<Self> {
        let addr = listener.local_addr().map_err(crate::Error::new_listen)?;
        Ok(AddrIncoming {
            listener,
            addr,
            sleep_on_errors: true,
            tcp_keepalive_timeout: None,
            tcp_nodelay: false,
            timeout: None,
        })
    }

    /// Get the local address bound to this listener.
    pub fn local_addr(&self) -> SocketAddr {
        self.addr
    }

    /// Set whether TCP keepalive messages are enabled on accepted connections.
    ///
    /// If `None` is specified, keepalive is disabled, otherwise the duration
    /// specified will be the time to remain idle before sending TCP keepalive
    /// probes.
    pub fn set_keepalive(&mut self, keepalive: Option<Duration>) -> &mut Self {
        self.tcp_keepalive_timeout = keepalive;
        self
    }

    /// Set the value of `TCP_NODELAY` option for accepted connections.
    pub fn set_nodelay(&mut self, enabled: bool) -> &mut Self {
        self.tcp_nodelay = enabled;
        self
    }

    /// Set whether to sleep on accept errors.
    ///
    /// A possible scenario is that the process has hit the max open files
    /// allowed, and so trying to accept a new connection will fail with
    /// `EMFILE`. In some cases, it's preferable to just wait for some time, if
    /// the application will likely close some files (or connections), and try
    /// to accept the connection again. If this option is `true`, the error
    /// will be logged at the `error` level, since it is still a big deal,
    /// and then the listener will sleep for 1 second.
    ///
    /// In other cases, hitting the max open files should be treat similarly
    /// to being out-of-memory, and simply error (and shutdown). Setting
    /// this option to `false` will allow that.
    ///
    /// Default is `true`.
    pub fn set_sleep_on_errors(&mut self, val: bool) {
        self.sleep_on_errors = val;
    }

    fn poll_next_(&mut self, cx: &mut task::Context<'_>) -> Poll<io::Result<AddrStream>> {
        // Check if a previous timeout is active that was set by IO errors.
        if let Some(ref mut to) = self.timeout {
            ready!(Pin::new(to).poll(cx));
        }
        self.timeout = None;

        loop {
            match ready!(self.listener.poll_accept(cx)) {
                Ok((socket, remote_addr)) => {
                    if let Some(dur) = self.tcp_keepalive_timeout {
                        let socket = socket2::SockRef::from(&socket);
                        let conf = socket2::TcpKeepalive::new().with_time(dur);
                        if let Err(e) = socket.set_tcp_keepalive(&conf) {
                            trace!("error trying to set TCP keepalive: {}", e);
                        }
                    }
                    if let Err(e) = socket.set_nodelay(self.tcp_nodelay) {
                        trace!("error trying to set TCP nodelay: {}", e);
                    }
                    let local_addr = socket.local_addr()?;
                    return Poll::Ready(Ok(AddrStream::new(socket, remote_addr, local_addr)));
                }
                Err(e) => {
                    // Connection errors can be ignored directly, continue by
                    // accepting the next request.
                    if is_connection_error(&e) {
                        debug!("accepted connection already errored: {}", e);
                        continue;
                    }

                    if self.sleep_on_errors {
                        error!("accept error: {}", e);

                        // Sleep 1s.
                        let mut timeout = Box::pin(tokio::time::sleep(Duration::from_secs(1)));

                        match timeout.as_mut().poll(cx) {
                            Poll::Ready(()) => {
                                // Wow, it's been a second already? Ok then...
                                continue;
                            }
                            Poll::Pending => {
                                self.timeout = Some(timeout);
                                return Poll::Pending;
                            }
                        }
                    } else {
                        return Poll::Ready(Err(e));
                    }
                }
            }
        }
    }
}

impl Accept for AddrIncoming {
    type Conn = AddrStream;
    type Error = io::Error;

    fn poll_accept(
        mut self: Pin<&mut Self>,
        cx: &mut task::Context<'_>,
    ) -> Poll<Option<Result<Self::Conn, Self::Error>>> {
        let result = ready!(self.poll_next_(cx));
        Poll::Ready(Some(result))
    }
}

/// This function defines errors that are per-connection. Which basically
/// means that if we get this error from `accept()` system call it means
/// next connection might be ready to be accepted.
///
/// All other errors will incur a timeout before next `accept()` is performed.
/// The timeout is useful to handle resource exhaustion errors like ENFILE
/// and EMFILE. Otherwise, could enter into tight loop.
fn is_connection_error(e: &io::Error) -> bool {
    matches!(
        e.kind(),
        io::ErrorKind::ConnectionRefused
            | io::ErrorKind::ConnectionAborted
            | io::ErrorKind::ConnectionReset
    )
}

impl fmt::Debug for AddrIncoming {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_struct("AddrIncoming")
            .field("addr", &self.addr)
            .field("sleep_on_errors", &self.sleep_on_errors)
            .field("tcp_keepalive_timeout", &self.tcp_keepalive_timeout)
            .field("tcp_nodelay", &self.tcp_nodelay)
            .finish()
    }
}

mod addr_stream {
    use std::io;
    use std::net::SocketAddr;
    #[cfg(unix)]
    use std::os::unix::io::{AsRawFd, RawFd};
    use tokio::io::{AsyncRead, AsyncWrite, ReadBuf};
    use tokio::net::TcpStream;

    use crate::common::{task, Pin, Poll};

    pin_project_lite::pin_project! {
        /// A transport returned yieled by `AddrIncoming`.
        #[derive(Debug)]
        pub struct AddrStream {
            #[pin]
            inner: TcpStream,
            pub(super) remote_addr: SocketAddr,
            pub(super) local_addr: SocketAddr
        }
    }

    impl AddrStream {
        pub(super) fn new(
            tcp: TcpStream,
            remote_addr: SocketAddr,
            local_addr: SocketAddr,
        ) -> AddrStream {
            AddrStream {
                inner: tcp,
                remote_addr,
                local_addr,
            }
        }

        /// Returns the remote (peer) address of this connection.
        #[inline]
        pub fn remote_addr(&self) -> SocketAddr {
            self.remote_addr
        }

        /// Returns the local address of this connection.
        #[inline]
        pub fn local_addr(&self) -> SocketAddr {
            self.local_addr
        }

        /// Consumes the AddrStream and returns the underlying IO object
        #[inline]
        pub fn into_inner(self) -> TcpStream {
            self.inner
        }

        /// Attempt to receive data on the socket, without removing that data
        /// from the queue, registering the current task for wakeup if data is
        /// not yet available.
        pub fn poll_peek(
            &mut self,
            cx: &mut task::Context<'_>,
            buf: &mut tokio::io::ReadBuf<'_>,
        ) -> Poll<io::Result<usize>> {
            self.inner.poll_peek(cx, buf)
        }
    }

    impl AsyncRead for AddrStream {
        #[inline]
        fn poll_read(
            self: Pin<&mut Self>,
            cx: &mut task::Context<'_>,
            buf: &mut ReadBuf<'_>,
        ) -> Poll<io::Result<()>> {
            self.project().inner.poll_read(cx, buf)
        }
    }

    impl AsyncWrite for AddrStream {
        #[inline]
        fn poll_write(
            self: Pin<&mut Self>,
            cx: &mut task::Context<'_>,
            buf: &[u8],
        ) -> Poll<io::Result<usize>> {
            self.project().inner.poll_write(cx, buf)
        }

        #[inline]
        fn poll_write_vectored(
            self: Pin<&mut Self>,
            cx: &mut task::Context<'_>,
            bufs: &[io::IoSlice<'_>],
        ) -> Poll<io::Result<usize>> {
            self.project().inner.poll_write_vectored(cx, bufs)
        }

        #[inline]
        fn poll_flush(self: Pin<&mut Self>, _cx: &mut task::Context<'_>) -> Poll<io::Result<()>> {
            // TCP flush is a noop
            Poll::Ready(Ok(()))
        }

        #[inline]
        fn poll_shutdown(self: Pin<&mut Self>, cx: &mut task::Context<'_>) -> Poll<io::Result<()>> {
            self.project().inner.poll_shutdown(cx)
        }

        #[inline]
        fn is_write_vectored(&self) -> bool {
            // Note that since `self.inner` is a `TcpStream`, this could
            // *probably* be hard-coded to return `true`...but it seems more
            // correct to ask it anyway (maybe we're on some platform without
            // scatter-gather IO?)
            self.inner.is_write_vectored()
        }
    }

    #[cfg(unix)]
    impl AsRawFd for AddrStream {
        fn as_raw_fd(&self) -> RawFd {
            self.inner.as_raw_fd()
        }
    }
}