1 use crate::convert::{TryFrom, TryInto};
3 use crate::io::{self, Error, ErrorKind};
5 use crate::num::NonZeroI32;
6 use crate::os::raw::NonZero_c_int;
10 use crate::sys::process::process_common::*;
12 #[cfg(target_os = "linux")]
13 use crate::os::linux::process::PidFd;
15 #[cfg(target_os = "linux")]
16 use crate::sys::weak::raw_syscall;
20 target_os = "freebsd",
21 all(target_os = "linux", target_env = "gnu"),
22 all(target_os = "linux", target_env = "musl"),
24 use crate::sys::weak::weak;
26 #[cfg(target_os = "vxworks")]
27 use libc::RTP_ID as pid_t;
29 #[cfg(not(target_os = "vxworks"))]
30 use libc::{c_int, gid_t, pid_t, uid_t};
32 ////////////////////////////////////////////////////////////////////////////////
34 ////////////////////////////////////////////////////////////////////////////////
41 ) -> io::Result<(Process, StdioPipes)> {
42 const CLOEXEC_MSG_FOOTER: [u8; 4] = *b"NOEX";
44 let envp = self.capture_env();
47 return Err(io::const_io_error!(
48 ErrorKind::InvalidInput,
49 "nul byte found in provided data",
53 let (ours, theirs) = self.setup_io(default, needs_stdin)?;
55 if let Some(ret) = self.posix_spawn(&theirs, envp.as_ref())? {
56 return Ok((ret, ours));
59 let (input, output) = sys::pipe::anon_pipe()?;
61 // Whatever happens after the fork is almost for sure going to touch or
62 // look at the environment in one way or another (PATH in `execvp` or
63 // accessing the `environ` pointer ourselves). Make sure no other thread
64 // is accessing the environment when we do the fork itself.
66 // Note that as soon as we're done with the fork there's no need to hold
67 // a lock any more because the parent won't do anything and the child is
68 // in its own process. Thus the parent drops the lock guard while the child
69 // forgets it to avoid unlocking it on a new thread, which would be invalid.
70 let env_lock = sys::os::env_read_lock();
71 let (pid, pidfd) = unsafe { self.do_fork()? };
74 crate::panic::always_abort();
75 mem::forget(env_lock);
77 let Err(err) = unsafe { self.do_exec(theirs, envp.as_ref()) };
78 let errno = err.raw_os_error().unwrap_or(libc::EINVAL) as u32;
79 let errno = errno.to_be_bytes();
85 CLOEXEC_MSG_FOOTER[0],
86 CLOEXEC_MSG_FOOTER[1],
87 CLOEXEC_MSG_FOOTER[2],
88 CLOEXEC_MSG_FOOTER[3],
90 // pipe I/O up to PIPE_BUF bytes should be atomic, and then
91 // we want to be sure we *don't* run at_exit destructors as
92 // we're being torn down regardless
93 rtassert!(output.write(&bytes).is_ok());
94 unsafe { libc::_exit(1) }
100 // Safety: We obtained the pidfd from calling `clone3` with
101 // `CLONE_PIDFD` so it's valid an otherwise unowned.
102 let mut p = unsafe { Process::new(pid, pidfd) };
103 let mut bytes = [0; 8];
105 // loop to handle EINTR
107 match input.read(&mut bytes) {
108 Ok(0) => return Ok((p, ours)),
110 let (errno, footer) = bytes.split_at(4);
112 CLOEXEC_MSG_FOOTER, footer,
113 "Validation on the CLOEXEC pipe failed: {:?}",
116 let errno = i32::from_be_bytes(errno.try_into().unwrap());
117 assert!(p.wait().is_ok(), "wait() should either return Ok or panic");
118 return Err(Error::from_raw_os_error(errno));
120 Err(ref e) if e.kind() == ErrorKind::Interrupted => {}
122 assert!(p.wait().is_ok(), "wait() should either return Ok or panic");
123 panic!("the CLOEXEC pipe failed: {:?}", e)
126 // pipe I/O up to PIPE_BUF bytes should be atomic
127 assert!(p.wait().is_ok(), "wait() should either return Ok or panic");
128 panic!("short read on the CLOEXEC pipe")
134 // Attempts to fork the process. If successful, returns Ok((0, -1))
135 // in the child, and Ok((child_pid, -1)) in the parent.
136 #[cfg(not(target_os = "linux"))]
137 unsafe fn do_fork(&mut self) -> Result<(pid_t, pid_t), io::Error> {
138 cvt(libc::fork()).map(|res| (res, -1))
141 // Attempts to fork the process. If successful, returns Ok((0, -1))
142 // in the child, and Ok((child_pid, child_pidfd)) in the parent.
143 #[cfg(target_os = "linux")]
144 unsafe fn do_fork(&mut self) -> Result<(pid_t, pid_t), io::Error> {
145 use crate::sync::atomic::{AtomicBool, Ordering};
147 static HAS_CLONE3: AtomicBool = AtomicBool::new(true);
148 const CLONE_PIDFD: u64 = 0x00001000;
166 fn clone3(cl_args: *mut clone_args, len: libc::size_t) -> libc::c_long
169 // Bypassing libc for `clone3` can make further libc calls unsafe,
170 // so we use it sparingly for now. See #89522 for details.
171 // Some tools (e.g. sandboxing tools) may also expect `fork`
172 // rather than `clone3`.
173 let want_clone3_pidfd = self.get_create_pidfd();
175 // If we fail to create a pidfd for any reason, this will
176 // stay as -1, which indicates an error.
177 let mut pidfd: pid_t = -1;
179 // Attempt to use the `clone3` syscall, which supports more arguments
180 // (in particular, the ability to create a pidfd). If this fails,
181 // we will fall through this block to a call to `fork()`
182 if want_clone3_pidfd && HAS_CLONE3.load(Ordering::Relaxed) {
183 let mut args = clone_args {
185 pidfd: &mut pidfd as *mut pid_t as u64,
188 exit_signal: libc::SIGCHLD as u64,
197 let args_ptr = &mut args as *mut clone_args;
198 let args_size = crate::mem::size_of::<clone_args>();
200 let res = cvt(clone3(args_ptr, args_size));
202 Ok(n) => return Ok((n as pid_t, pidfd)),
203 Err(e) => match e.raw_os_error() {
204 // Multiple threads can race to execute this store,
205 // but that's fine - that just means that multiple threads
206 // will have tried and failed to execute the same syscall,
207 // with no other side effects.
208 Some(libc::ENOSYS) => HAS_CLONE3.store(false, Ordering::Relaxed),
209 // Fallback to fork if `EPERM` is returned. (e.g. blocked by seccomp)
210 Some(libc::EPERM) => {}
216 // Generally, we just call `fork`. If we get here after wanting `clone3`,
217 // then the syscall does not exist or we do not have permission to call it.
218 cvt(libc::fork()).map(|res| (res, pidfd))
221 pub fn exec(&mut self, default: Stdio) -> io::Error {
222 let envp = self.capture_env();
225 return io::const_io_error!(ErrorKind::InvalidInput, "nul byte found in provided data",);
228 match self.setup_io(default, true) {
231 // Similar to when forking, we want to ensure that access to
232 // the environment is synchronized, so make sure to grab the
233 // environment lock before we try to exec.
234 let _lock = sys::os::env_read_lock();
236 let Err(e) = self.do_exec(theirs, envp.as_ref());
244 // And at this point we've reached a special time in the life of the
245 // child. The child must now be considered hamstrung and unable to
246 // do anything other than syscalls really. Consider the following
249 // 1. Thread A of process 1 grabs the malloc() mutex
250 // 2. Thread B of process 1 forks(), creating thread C
251 // 3. Thread C of process 2 then attempts to malloc()
252 // 4. The memory of process 2 is the same as the memory of
253 // process 1, so the mutex is locked.
255 // This situation looks a lot like deadlock, right? It turns out
256 // that this is what pthread_atfork() takes care of, which is
257 // presumably implemented across platforms. The first thing that
258 // threads to *before* forking is to do things like grab the malloc
259 // mutex, and then after the fork they unlock it.
261 // Despite this information, libnative's spawn has been witnessed to
262 // deadlock on both macOS and FreeBSD. I'm not entirely sure why, but
263 // all collected backtraces point at malloc/free traffic in the
264 // child spawned process.
266 // For this reason, the block of code below should contain 0
267 // invocations of either malloc of free (or their related friends).
269 // As an example of not having malloc/free traffic, we don't close
270 // this file descriptor by dropping the FileDesc (which contains an
271 // allocation). Instead we just close it manually. This will never
272 // have the drop glue anyway because this code never returns (the
273 // child will either exec() or invoke libc::exit)
277 maybe_envp: Option<&CStringArray>,
278 ) -> Result<!, io::Error> {
279 use crate::sys::{self, cvt_r};
281 if let Some(fd) = stdio.stdin.fd() {
282 cvt_r(|| libc::dup2(fd, libc::STDIN_FILENO))?;
284 if let Some(fd) = stdio.stdout.fd() {
285 cvt_r(|| libc::dup2(fd, libc::STDOUT_FILENO))?;
287 if let Some(fd) = stdio.stderr.fd() {
288 cvt_r(|| libc::dup2(fd, libc::STDERR_FILENO))?;
291 #[cfg(not(target_os = "l4re"))]
293 if let Some(_g) = self.get_groups() {
294 //FIXME: Redox kernel does not support setgroups yet
295 #[cfg(not(target_os = "redox"))]
296 cvt(libc::setgroups(_g.len().try_into().unwrap(), _g.as_ptr()))?;
298 if let Some(u) = self.get_gid() {
299 cvt(libc::setgid(u as gid_t))?;
301 if let Some(u) = self.get_uid() {
302 // When dropping privileges from root, the `setgroups` call
303 // will remove any extraneous groups. We only drop groups
304 // if the current uid is 0 and we weren't given an explicit
305 // set of groups. If we don't call this, then even though our
306 // uid has dropped, we may still have groups that enable us to
307 // do super-user things.
308 //FIXME: Redox kernel does not support setgroups yet
309 #[cfg(not(target_os = "redox"))]
310 if libc::getuid() == 0 && self.get_groups().is_none() {
311 cvt(libc::setgroups(0, ptr::null()))?;
313 cvt(libc::setuid(u as uid_t))?;
316 if let Some(ref cwd) = *self.get_cwd() {
317 cvt(libc::chdir(cwd.as_ptr()))?;
320 // emscripten has no signal support.
321 #[cfg(not(target_os = "emscripten"))]
323 use crate::mem::MaybeUninit;
324 // Reset signal handling so the child process starts in a
325 // standardized state. libstd ignores SIGPIPE, and signal-handling
326 // libraries often set a mask. Child processes inherit ignored
327 // signals and the signal mask from their parent, but most
328 // UNIX programs do not reset these things on their own, so we
329 // need to clean things up now to avoid confusing the program
330 // we're about to run.
331 let mut set = MaybeUninit::<libc::sigset_t>::uninit();
332 cvt(sigemptyset(set.as_mut_ptr()))?;
333 cvt(libc::pthread_sigmask(libc::SIG_SETMASK, set.as_ptr(), ptr::null_mut()))?;
335 #[cfg(target_os = "android")] // see issue #88585
337 let mut action: libc::sigaction = mem::zeroed();
338 action.sa_sigaction = libc::SIG_DFL;
339 cvt(libc::sigaction(libc::SIGPIPE, &action, ptr::null_mut()))?;
341 #[cfg(not(target_os = "android"))]
343 let ret = sys::signal(libc::SIGPIPE, libc::SIG_DFL);
344 if ret == libc::SIG_ERR {
345 return Err(io::Error::last_os_error());
350 for callback in self.get_closures().iter_mut() {
354 // Although we're performing an exec here we may also return with an
355 // error from this function (without actually exec'ing) in which case we
356 // want to be sure to restore the global environment back to what it
357 // once was, ensuring that our temporary override, when free'd, doesn't
358 // corrupt our process's environment.
359 let mut _reset = None;
360 if let Some(envp) = maybe_envp {
361 struct Reset(*const *const libc::c_char);
363 impl Drop for Reset {
366 *sys::os::environ() = self.0;
371 _reset = Some(Reset(*sys::os::environ()));
372 *sys::os::environ() = envp.as_ptr();
375 libc::execvp(self.get_program_cstr().as_ptr(), self.get_argv().as_ptr());
376 Err(io::Error::last_os_error())
381 target_os = "freebsd",
382 all(target_os = "linux", target_env = "gnu"),
383 all(target_os = "linux", target_env = "musl"),
388 _: Option<&CStringArray>,
389 ) -> io::Result<Option<Process>> {
393 // Only support platforms for which posix_spawn() can return ENOENT
397 target_os = "freebsd",
398 all(target_os = "linux", target_env = "gnu"),
399 all(target_os = "linux", target_env = "musl"),
404 envp: Option<&CStringArray>,
405 ) -> io::Result<Option<Process>> {
406 use crate::mem::MaybeUninit;
407 use crate::sys::{self, cvt_nz};
409 if self.get_gid().is_some()
410 || self.get_uid().is_some()
411 || (self.env_saw_path() && !self.program_is_path())
412 || !self.get_closures().is_empty()
413 || self.get_groups().is_some()
414 || self.get_create_pidfd()
419 // Only glibc 2.24+ posix_spawn() supports returning ENOENT directly.
420 #[cfg(all(target_os = "linux", target_env = "gnu"))]
422 if let Some(version) = sys::os::glibc_version() {
423 if version < (2, 24) {
431 // Solaris, glibc 2.29+, and musl 1.24+ can set a new working directory,
432 // and maybe others will gain this non-POSIX function too. We'll check
433 // for this weak symbol as soon as it's needed, so we can return early
434 // otherwise to do a manual chdir before exec.
436 fn posix_spawn_file_actions_addchdir_np(
437 *mut libc::posix_spawn_file_actions_t,
441 let addchdir = match self.get_cwd() {
443 if cfg!(target_os = "macos") {
444 // There is a bug in macOS where a relative executable
445 // path like "../myprogram" will cause `posix_spawn` to
446 // successfully launch the program, but erroneously return
447 // ENOENT when used with posix_spawn_file_actions_addchdir_np
448 // which was introduced in macOS 10.15.
451 match posix_spawn_file_actions_addchdir_np.get() {
452 Some(f) => Some((f, cwd)),
453 None => return Ok(None),
459 // Safety: -1 indicates we don't have a pidfd.
460 let mut p = unsafe { Process::new(0, -1) };
462 struct PosixSpawnFileActions<'a>(&'a mut MaybeUninit<libc::posix_spawn_file_actions_t>);
464 impl Drop for PosixSpawnFileActions<'_> {
467 libc::posix_spawn_file_actions_destroy(self.0.as_mut_ptr());
472 struct PosixSpawnattr<'a>(&'a mut MaybeUninit<libc::posix_spawnattr_t>);
474 impl Drop for PosixSpawnattr<'_> {
477 libc::posix_spawnattr_destroy(self.0.as_mut_ptr());
483 let mut attrs = MaybeUninit::uninit();
484 cvt_nz(libc::posix_spawnattr_init(attrs.as_mut_ptr()))?;
485 let attrs = PosixSpawnattr(&mut attrs);
487 let mut file_actions = MaybeUninit::uninit();
488 cvt_nz(libc::posix_spawn_file_actions_init(file_actions.as_mut_ptr()))?;
489 let file_actions = PosixSpawnFileActions(&mut file_actions);
491 if let Some(fd) = stdio.stdin.fd() {
492 cvt_nz(libc::posix_spawn_file_actions_adddup2(
493 file_actions.0.as_mut_ptr(),
498 if let Some(fd) = stdio.stdout.fd() {
499 cvt_nz(libc::posix_spawn_file_actions_adddup2(
500 file_actions.0.as_mut_ptr(),
505 if let Some(fd) = stdio.stderr.fd() {
506 cvt_nz(libc::posix_spawn_file_actions_adddup2(
507 file_actions.0.as_mut_ptr(),
512 if let Some((f, cwd)) = addchdir {
513 cvt_nz(f(file_actions.0.as_mut_ptr(), cwd.as_ptr()))?;
516 let mut set = MaybeUninit::<libc::sigset_t>::uninit();
517 cvt(sigemptyset(set.as_mut_ptr()))?;
518 cvt_nz(libc::posix_spawnattr_setsigmask(attrs.0.as_mut_ptr(), set.as_ptr()))?;
519 cvt(sigaddset(set.as_mut_ptr(), libc::SIGPIPE))?;
520 cvt_nz(libc::posix_spawnattr_setsigdefault(attrs.0.as_mut_ptr(), set.as_ptr()))?;
522 let flags = libc::POSIX_SPAWN_SETSIGDEF | libc::POSIX_SPAWN_SETSIGMASK;
523 cvt_nz(libc::posix_spawnattr_setflags(attrs.0.as_mut_ptr(), flags as _))?;
525 // Make sure we synchronize access to the global `environ` resource
526 let _env_lock = sys::os::env_read_lock();
527 let envp = envp.map(|c| c.as_ptr()).unwrap_or_else(|| *sys::os::environ() as *const _);
528 cvt_nz(libc::posix_spawnp(
530 self.get_program_cstr().as_ptr(),
531 file_actions.0.as_ptr(),
533 self.get_argv().as_ptr() as *const _,
541 ////////////////////////////////////////////////////////////////////////////////
543 ////////////////////////////////////////////////////////////////////////////////
545 /// The unique ID of the process (this should never be negative).
548 status: Option<ExitStatus>,
549 // On Linux, stores the pidfd created for this child.
550 // This is None if the user did not request pidfd creation,
551 // or if the pidfd could not be created for some reason
552 // (e.g. the `clone3` syscall was not available).
553 #[cfg(target_os = "linux")]
554 pidfd: Option<PidFd>,
558 #[cfg(target_os = "linux")]
559 unsafe fn new(pid: pid_t, pidfd: pid_t) -> Self {
560 use crate::os::unix::io::FromRawFd;
561 use crate::sys_common::FromInner;
562 // Safety: If `pidfd` is nonnegative, we assume it's valid and otherwise unowned.
563 let pidfd = (pidfd >= 0).then(|| PidFd::from_inner(sys::fd::FileDesc::from_raw_fd(pidfd)));
564 Process { pid, status: None, pidfd }
567 #[cfg(not(target_os = "linux"))]
568 unsafe fn new(pid: pid_t, _pidfd: pid_t) -> Self {
569 Process { pid, status: None }
572 pub fn id(&self) -> u32 {
576 pub fn kill(&mut self) -> io::Result<()> {
577 // If we've already waited on this process then the pid can be recycled
578 // and used for another process, and we probably shouldn't be killing
579 // random processes, so just return an error.
580 if self.status.is_some() {
581 Err(io::const_io_error!(
582 ErrorKind::InvalidInput,
583 "invalid argument: can't kill an exited process",
586 cvt(unsafe { libc::kill(self.pid, libc::SIGKILL) }).map(drop)
590 pub fn wait(&mut self) -> io::Result<ExitStatus> {
591 use crate::sys::cvt_r;
592 if let Some(status) = self.status {
595 let mut status = 0 as c_int;
596 cvt_r(|| unsafe { libc::waitpid(self.pid, &mut status, 0) })?;
597 self.status = Some(ExitStatus::new(status));
598 Ok(ExitStatus::new(status))
601 pub fn try_wait(&mut self) -> io::Result<Option<ExitStatus>> {
602 if let Some(status) = self.status {
603 return Ok(Some(status));
605 let mut status = 0 as c_int;
606 let pid = cvt(unsafe { libc::waitpid(self.pid, &mut status, libc::WNOHANG) })?;
610 self.status = Some(ExitStatus::new(status));
611 Ok(Some(ExitStatus::new(status)))
616 /// Unix exit statuses
618 // This is not actually an "exit status" in Unix terminology. Rather, it is a "wait status".
619 // See the discussion in comments and doc comments for `std::process::ExitStatus`.
620 #[derive(PartialEq, Eq, Clone, Copy)]
621 pub struct ExitStatus(c_int);
623 impl fmt::Debug for ExitStatus {
624 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
625 f.debug_tuple("unix_wait_status").field(&self.0).finish()
630 pub fn new(status: c_int) -> ExitStatus {
634 fn exited(&self) -> bool {
635 libc::WIFEXITED(self.0)
638 pub fn exit_ok(&self) -> Result<(), ExitStatusError> {
639 // This assumes that WIFEXITED(status) && WEXITSTATUS==0 corresponds to status==0. This is
640 // true on all actual versions of Unix, is widely assumed, and is specified in SuS
641 // https://pubs.opengroup.org/onlinepubs/9699919799/functions/wait.html . If it is not
642 // true for a platform pretending to be Unix, the tests (our doctests, and also
643 // procsss_unix/tests.rs) will spot it. `ExitStatusError::code` assumes this too.
644 match NonZero_c_int::try_from(self.0) {
645 /* was nonzero */ Ok(failure) => Err(ExitStatusError(failure)),
646 /* was zero, couldn't convert */ Err(_) => Ok(()),
650 pub fn code(&self) -> Option<i32> {
651 if self.exited() { Some(libc::WEXITSTATUS(self.0)) } else { None }
654 pub fn signal(&self) -> Option<i32> {
655 if libc::WIFSIGNALED(self.0) { Some(libc::WTERMSIG(self.0)) } else { None }
658 pub fn core_dumped(&self) -> bool {
659 libc::WIFSIGNALED(self.0) && libc::WCOREDUMP(self.0)
662 pub fn stopped_signal(&self) -> Option<i32> {
663 if libc::WIFSTOPPED(self.0) { Some(libc::WSTOPSIG(self.0)) } else { None }
666 pub fn continued(&self) -> bool {
667 libc::WIFCONTINUED(self.0)
670 pub fn into_raw(&self) -> c_int {
675 /// Converts a raw `c_int` to a type-safe `ExitStatus` by wrapping it without copying.
676 impl From<c_int> for ExitStatus {
677 fn from(a: c_int) -> ExitStatus {
682 impl fmt::Display for ExitStatus {
683 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
684 if let Some(code) = self.code() {
685 write!(f, "exit status: {}", code)
686 } else if let Some(signal) = self.signal() {
687 if self.core_dumped() {
688 write!(f, "signal: {} (core dumped)", signal)
690 write!(f, "signal: {}", signal)
692 } else if let Some(signal) = self.stopped_signal() {
693 write!(f, "stopped (not terminated) by signal: {}", signal)
694 } else if self.continued() {
695 write!(f, "continued (WIFCONTINUED)")
697 write!(f, "unrecognised wait status: {} {:#x}", self.0, self.0)
702 #[derive(PartialEq, Eq, Clone, Copy)]
703 pub struct ExitStatusError(NonZero_c_int);
705 impl Into<ExitStatus> for ExitStatusError {
706 fn into(self) -> ExitStatus {
707 ExitStatus(self.0.into())
711 impl fmt::Debug for ExitStatusError {
712 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
713 f.debug_tuple("unix_wait_status").field(&self.0).finish()
717 impl ExitStatusError {
718 pub fn code(self) -> Option<NonZeroI32> {
719 ExitStatus(self.0.into()).code().map(|st| st.try_into().unwrap())
723 #[cfg(target_os = "linux")]
724 #[unstable(feature = "linux_pidfd", issue = "82971")]
725 impl crate::os::linux::process::ChildExt for crate::process::Child {
726 fn pidfd(&self) -> io::Result<&PidFd> {
730 .ok_or_else(|| Error::new(ErrorKind::Other, "No pidfd was created."))
733 fn take_pidfd(&mut self) -> io::Result<PidFd> {
737 .ok_or_else(|| Error::new(ErrorKind::Other, "No pidfd was created."))
742 #[path = "process_unix/tests.rs"]