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
use std::{
cell::Cell,
io,
marker::PhantomData,
os::unix::io::{AsRawFd, RawFd},
process::{Child, ChildStderr},
time::Duration,
};
#[cfg(any(
target_os = "linux",
target_os = "android",
target_vendor = "apple",
target_os = "freebsd",
target_os = "dragonfly",
target_os = "netbsd",
target_os = "openbsd"
))]
use std::os::unix::io::{FromRawFd, OwnedFd};
#[cfg(any(
target_vendor = "apple",
target_os = "freebsd",
target_os = "dragonfly",
target_os = "netbsd",
target_os = "openbsd"
))]
use std::{
mem::{self, MaybeUninit},
ptr,
};
use crate::{cell_modify, cell_push};
#[derive(Default)]
pub(crate) struct Reactor {
poll_fds: Cell<Vec<libc::pollfd>>,
/// Set once `pidfd_open` turns out not to be available, such as on Linux
/// before 5.3 or when blocked by seccomp, so it isn't tried again.
#[cfg(any(target_os = "linux", target_os = "android"))]
no_pidfd: Cell<bool>,
/// Created by the first [`Reactor::register_child`], to be told about
/// child exits. From then on it is always polled, as it is only readable
/// while it has events, which every wait consumes.
#[cfg(any(
target_vendor = "apple",
target_os = "freebsd",
target_os = "dragonfly",
target_os = "netbsd",
target_os = "openbsd"
))]
kqueue: Cell<Option<OwnedFd>>,
}
pub(crate) struct Registration<'a> {
reactor: &'a Reactor,
fd: RawFd,
}
/// Keeps a child registered with the [`Reactor`] until dropped.
pub(crate) struct ChildRegistration<'a> {
/// Its pidfd, if it could be opened, and the reactor polling it.
#[cfg(any(target_os = "linux", target_os = "android"))]
pidfd: Option<(&'a Reactor, OwnedFd)>,
_reactor: PhantomData<&'a Reactor>,
}
impl Reactor {
pub(crate) fn register_stderr(&self, stderr: &ChildStderr) -> io::Result<Registration<'_>> {
Ok(self.register_fd(stderr.as_raw_fd()))
}
pub(crate) fn register_fd(&self, fd: RawFd) -> Registration<'_> {
self.add_fd(fd);
Registration { reactor: self, fd }
}
fn add_fd(&self, fd: RawFd) {
cell_push(
&self.poll_fds,
libc::pollfd {
fd,
events: libc::POLLIN,
revents: 0,
},
);
}
/// Must be called before `fd` is closed, while its number can't have
/// been reused.
fn remove_fd(&self, fd: RawFd) {
cell_modify(&self.poll_fds, |poll_fds| {
if let Some(i) = poll_fds.iter().position(|pollfd| pollfd.fd == fd) {
poll_fds.swap_remove(i);
}
});
}
/// Wake [`Reactor::wait_with_timeout`] when `child` exits, where that can
/// be waited for. Otherwise its exit is only noticed through its stderr
/// being closed, which a process it started may keep open, or the
/// timeout.
pub(crate) fn register_child(&self, _child: &Child) -> io::Result<ChildRegistration<'_>> {
// Nothing to undo when it is dropped: the event is deleted once
// returned, or with the kqueue.
#[cfg(any(
target_vendor = "apple",
target_os = "freebsd",
target_os = "dragonfly",
target_os = "netbsd",
target_os = "openbsd"
))]
self.register_proc_kevent(_child)?;
Ok(ChildRegistration {
#[cfg(any(target_os = "linux", target_os = "android"))]
pidfd: self.register_pidfd(_child)?,
_reactor: PhantomData,
})
}
#[cfg(any(target_os = "linux", target_os = "android"))]
fn register_pidfd(&self, child: &Child) -> io::Result<Option<(&Self, OwnedFd)>> {
if !self.no_pidfd.get() {
// Not yet waited on, so the pid is still the child's.
let ret = unsafe {
libc::syscall(
libc::SYS_pidfd_open,
libc::c_long::from(child.id() as libc::pid_t),
0 as libc::c_long,
)
};
if ret >= 0 {
// Always close-on-exec.
let pidfd = unsafe { OwnedFd::from_raw_fd(ret as RawFd) };
self.add_fd(pidfd.as_raw_fd());
return Ok(Some((self, pidfd)));
}
let err = io::Error::last_os_error();
if matches!(
err.raw_os_error(),
Some(libc::ENOSYS | libc::EPERM | libc::EACCES)
) {
self.no_pidfd.set(true);
return Ok(None);
}
return Err(err);
}
Ok(None)
}
#[cfg(any(
target_vendor = "apple",
target_os = "freebsd",
target_os = "dragonfly",
target_os = "netbsd",
target_os = "openbsd"
))]
fn register_proc_kevent(&self, child: &Child) -> io::Result<()> {
let kqueue = cell_modify(&self.kqueue, |kqueue| {
if let Some(kqueue) = kqueue {
return Ok(kqueue.as_raw_fd());
}
let fd = unsafe { libc::kqueue() };
if fd < 0 {
return Err(io::Error::last_os_error());
}
*kqueue = Some(unsafe { OwnedFd::from_raw_fd(fd) });
self.add_fd(fd);
Ok(fd)
})?;
let change = libc::kevent {
ident: child.id() as _,
filter: libc::EVFILT_PROC as _,
// Deleted once returned by `kevent`, see `drain_kqueue`.
flags: (libc::EV_ADD | libc::EV_ONESHOT) as _,
fflags: libc::NOTE_EXIT as _,
// `data` and `udata` are unused, but their types differ between
// these, and FreeBSD 12+ adds `ext`, depending on how libc is
// built. All are plain integers or pointers, zero is valid.
..unsafe { mem::zeroed() }
};
// Only applies `change`: with no room for events, it doesn't wait for
// any, and a failure to apply it is returned as an error.
let ret = unsafe { libc::kevent(kqueue, &change, 1, ptr::null_mut(), 0, ptr::null()) };
if ret < 0 {
let err = io::Error::last_os_error();
// It has already exited, so there is nothing to wait for.
if err.raw_os_error() == Some(libc::ESRCH) {
return Ok(());
}
return Err(err);
}
Ok(())
}
/// Consumes the child exits so far, which are what the futures are about
/// to be polled for, so that the kqueue isn't readable for them anymore.
#[cfg(any(
target_vendor = "apple",
target_os = "freebsd",
target_os = "dragonfly",
target_os = "netbsd",
target_os = "openbsd"
))]
fn drain_kqueue(&self) -> io::Result<()> {
cell_modify(&self.kqueue, |kqueue| {
let Some(kqueue) = kqueue else {
return Ok(());
};
const N: usize = 64;
// Only written to by `kevent`, never read.
let mut events = MaybeUninit::<[libc::kevent; N]>::uninit();
let timeout = libc::timespec {
tv_sec: 0,
tv_nsec: 0,
};
loop {
// No changes, up to `N` events, without waiting.
let n = unsafe {
libc::kevent(
kqueue.as_raw_fd(),
ptr::null(),
0,
events.as_mut_ptr().cast(),
N as _,
&timeout,
)
};
if n < 0 {
return Err(io::Error::last_os_error());
}
// Fewer than asked for: there are no more.
if n < N as _ {
return Ok(());
}
}
})
}
pub(crate) fn wait_with_timeout(&self, timeout: Duration) -> io::Result<()> {
let timeout = timeout.as_millis().try_into().unwrap();
let result = cell_modify(&self.poll_fds, |poll_fds| {
let ret = unsafe {
libc::poll(
poll_fds.as_mut_ptr(),
poll_fds.len() as libc::nfds_t,
timeout,
)
};
if ret < 0 {
Err(io::Error::last_os_error())
} else {
Ok(())
}
});
#[cfg(any(
target_vendor = "apple",
target_os = "freebsd",
target_os = "dragonfly",
target_os = "netbsd",
target_os = "openbsd"
))]
let result = result.and_then(|()| self.drain_kqueue());
match result {
Err(err) if err.kind() == io::ErrorKind::Interrupted => Ok(()),
result => result,
}
}
}
impl Drop for Registration<'_> {
fn drop(&mut self) {
self.reactor.remove_fd(self.fd);
}
}
#[cfg(any(target_os = "linux", target_os = "android"))]
impl Drop for ChildRegistration<'_> {
fn drop(&mut self) {
// The pidfd itself is closed right after, as a field.
if let Some((reactor, pidfd)) = &self.pidfd {
reactor.remove_fd(pidfd.as_raw_fd());
}
}
}