Runtime removal: refactor process

This patch continues the runtime removal by moving and refactoring the
process implementation into the new `sys` module.

Because this eliminates APIs in `libnative` and `librustrt`, it is a:

[breaking-change]

This functionality is likely to be available publicly, in some form,
from `std` in the future.

3 files changed, 999 insertions, 1 deletions

diff --git a/src/libstd/sys/unix/fs.rs b/src/libstd/sys/unix/fs.rs
new file mode 100644
index 00000000000..3dcd99859e8
--- /dev/null
+++ b/src/libstd/sys/unix/fs.rs
@@ -0,0 +1,411 @@
+// Copyright 2013-2014 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+//! Blocking posix-based file I/O
+
+use libc::{mod, c_int, c_void};
+use c_str::CString;
+use mem;
+use io;
+
+use prelude::*;
+
+use io::{FilePermission, Write, UnstableFileStat, Open, FileAccess, FileMode};
+use io::{IoResult, FileStat, SeekStyle, Reader};
+use io::{Read, Truncate, SeekCur, SeekSet, ReadWrite, SeekEnd, Append};
+use result::{Ok, Err};
+use sys::retry;
+use sys_common::{keep_going, eof, mkerr_libc};
+
+pub use path::PosixPath as Path;
+
+pub type fd_t = libc::c_int;
+
+pub struct FileDesc {
+    /// The underlying C file descriptor.
+    fd: fd_t,
+
+    /// Whether to close the file descriptor on drop.
+    close_on_drop: bool,
+}
+
+impl FileDesc {
+    pub fn new(fd: fd_t, close_on_drop: bool) -> FileDesc {
+        FileDesc { fd: fd, close_on_drop: close_on_drop }
+    }
+
+    pub fn read(&self, buf: &mut [u8]) -> IoResult<uint> {
+        let ret = retry(|| unsafe {
+            libc::read(self.fd(),
+                       buf.as_mut_ptr() as *mut libc::c_void,
+                       buf.len() as libc::size_t)
+        });
+        if ret == 0 {
+            Err(eof())
+        } else if ret < 0 {
+            Err(super::last_error())
+        } else {
+            Ok(ret as uint)
+        }
+    }
+    pub fn write(&self, buf: &[u8]) -> IoResult<()> {
+        let ret = keep_going(buf, |buf, len| {
+            unsafe {
+                libc::write(self.fd(), buf as *const libc::c_void,
+                            len as libc::size_t) as i64
+            }
+        });
+        if ret < 0 {
+            Err(super::last_error())
+        } else {
+            Ok(())
+        }
+    }
+
+    pub fn fd(&self) -> fd_t { self.fd }
+
+    pub fn seek(&self, pos: i64, whence: SeekStyle) -> IoResult<u64> {
+        let whence = match whence {
+            SeekSet => libc::SEEK_SET,
+            SeekEnd => libc::SEEK_END,
+            SeekCur => libc::SEEK_CUR,
+        };
+        let n = unsafe { libc::lseek(self.fd(), pos as libc::off_t, whence) };
+        if n < 0 {
+            Err(super::last_error())
+        } else {
+            Ok(n as u64)
+        }
+    }
+
+    pub fn tell(&self) -> IoResult<u64> {
+        let n = unsafe { libc::lseek(self.fd(), 0, libc::SEEK_CUR) };
+        if n < 0 {
+            Err(super::last_error())
+        } else {
+            Ok(n as u64)
+        }
+    }
+
+    pub fn fsync(&self) -> IoResult<()> {
+        mkerr_libc(retry(|| unsafe { libc::fsync(self.fd()) }))
+    }
+
+    pub fn datasync(&self) -> IoResult<()> {
+        return mkerr_libc(os_datasync(self.fd()));
+
+        #[cfg(any(target_os = "macos", target_os = "ios"))]
+        fn os_datasync(fd: c_int) -> c_int {
+            unsafe { libc::fcntl(fd, libc::F_FULLFSYNC) }
+        }
+        #[cfg(target_os = "linux")]
+        fn os_datasync(fd: c_int) -> c_int {
+            retry(|| unsafe { libc::fdatasync(fd) })
+        }
+        #[cfg(not(any(target_os = "macos", target_os = "ios", target_os = "linux")))]
+        fn os_datasync(fd: c_int) -> c_int {
+            retry(|| unsafe { libc::fsync(fd) })
+        }
+    }
+
+    pub fn truncate(&self, offset: i64) -> IoResult<()> {
+        mkerr_libc(retry(|| unsafe {
+            libc::ftruncate(self.fd(), offset as libc::off_t)
+        }))
+    }
+
+    pub fn fstat(&self) -> IoResult<FileStat> {
+        let mut stat: libc::stat = unsafe { mem::zeroed() };
+        match unsafe { libc::fstat(self.fd(), &mut stat) } {
+            0 => Ok(mkstat(&stat)),
+            _ => Err(super::last_error()),
+        }
+    }
+
+    /// Extract the actual filedescriptor without closing it.
+    pub fn unwrap(self) -> fd_t {
+        let fd = self.fd;
+        unsafe { mem::forget(self) };
+        fd
+    }
+}
+
+/*
+
+impl RtioTTY for FileDesc {
+    fn read(&mut self, buf: &mut [u8]) -> IoResult<uint> {
+        self.inner_read(buf)
+    }
+    fn write(&mut self, buf: &[u8]) -> IoResult<()> {
+        self.inner_write(buf)
+    }
+    fn set_raw(&mut self, _raw: bool) -> IoResult<()> {
+        Err(super::unimpl())
+    }
+    fn get_winsize(&mut self) -> IoResult<(int, int)> {
+        Err(super::unimpl())
+    }
+    fn isatty(&self) -> bool { false }
+}
+*/
+
+impl Drop for FileDesc {
+    fn drop(&mut self) {
+        // closing stdio file handles makes no sense, so never do it. Also, note
+        // that errors are ignored when closing a file descriptor. The reason
+        // for this is that if an error occurs we don't actually know if the
+        // file descriptor was closed or not, and if we retried (for something
+        // like EINTR), we might close another valid file descriptor (opened
+        // after we closed ours.
+        if self.close_on_drop && self.fd > libc::STDERR_FILENO {
+            let n = unsafe { libc::close(self.fd) };
+            if n != 0 {
+                println!("error {} when closing file descriptor {}", n, self.fd);
+            }
+        }
+    }
+}
+
+pub fn open(path: &Path, fm: FileMode, fa: FileAccess) -> IoResult<FileDesc> {
+    let flags = match fm {
+        Open => 0,
+        Append => libc::O_APPEND,
+        Truncate => libc::O_TRUNC,
+    };
+    // Opening with a write permission must silently create the file.
+    let (flags, mode) = match fa {
+        Read => (flags | libc::O_RDONLY, 0),
+        Write => (flags | libc::O_WRONLY | libc::O_CREAT,
+                        libc::S_IRUSR | libc::S_IWUSR),
+        ReadWrite => (flags | libc::O_RDWR | libc::O_CREAT,
+                            libc::S_IRUSR | libc::S_IWUSR),
+    };
+
+    let path = path.to_c_str();
+    match retry(|| unsafe { libc::open(path.as_ptr(), flags, mode) }) {
+        -1 => Err(super::last_error()),
+        fd => Ok(FileDesc::new(fd, true)),
+    }
+}
+
+pub fn mkdir(p: &Path, mode: uint) -> IoResult<()> {
+    let p = p.to_c_str();
+    mkerr_libc(unsafe { libc::mkdir(p.as_ptr(), mode as libc::mode_t) })
+}
+
+pub fn readdir(p: &Path) -> IoResult<Vec<Path>> {
+    use libc::{dirent_t};
+    use libc::{opendir, readdir_r, closedir};
+
+    fn prune(root: &CString, dirs: Vec<Path>) -> Vec<Path> {
+        let root = unsafe { CString::new(root.as_ptr(), false) };
+        let root = Path::new(root);
+
+        dirs.into_iter().filter(|path| {
+            path.as_vec() != b"." && path.as_vec() != b".."
+        }).map(|path| root.join(path)).collect()
+    }
+
+    extern {
+        fn rust_dirent_t_size() -> libc::c_int;
+        fn rust_list_dir_val(ptr: *mut dirent_t) -> *const libc::c_char;
+    }
+
+    let size = unsafe { rust_dirent_t_size() };
+    let mut buf = Vec::<u8>::with_capacity(size as uint);
+    let ptr = buf.as_mut_slice().as_mut_ptr() as *mut dirent_t;
+
+    let p = p.to_c_str();
+    let dir_ptr = unsafe {opendir(p.as_ptr())};
+
+    if dir_ptr as uint != 0 {
+        let mut paths = vec!();
+        let mut entry_ptr = 0 as *mut dirent_t;
+        while unsafe { readdir_r(dir_ptr, ptr, &mut entry_ptr) == 0 } {
+            if entry_ptr.is_null() { break }
+            let cstr = unsafe {
+                CString::new(rust_list_dir_val(entry_ptr), false)
+            };
+            paths.push(Path::new(cstr));
+        }
+        assert_eq!(unsafe { closedir(dir_ptr) }, 0);
+        Ok(prune(&p, paths))
+    } else {
+        Err(super::last_error())
+    }
+}
+
+pub fn unlink(p: &Path) -> IoResult<()> {
+    let p = p.to_c_str();
+    mkerr_libc(unsafe { libc::unlink(p.as_ptr()) })
+}
+
+pub fn rename(old: &Path, new: &Path) -> IoResult<()> {
+    let old = old.to_c_str();
+    let new = new.to_c_str();
+    mkerr_libc(unsafe {
+        libc::rename(old.as_ptr(), new.as_ptr())
+    })
+}
+
+pub fn chmod(p: &Path, mode: uint) -> IoResult<()> {
+    let p = p.to_c_str();
+    mkerr_libc(retry(|| unsafe {
+        libc::chmod(p.as_ptr(), mode as libc::mode_t)
+    }))
+}
+
+pub fn rmdir(p: &Path) -> IoResult<()> {
+    let p = p.to_c_str();
+    mkerr_libc(unsafe { libc::rmdir(p.as_ptr()) })
+}
+
+pub fn chown(p: &Path, uid: int, gid: int) -> IoResult<()> {
+    let p = p.to_c_str();
+    mkerr_libc(retry(|| unsafe {
+        libc::chown(p.as_ptr(), uid as libc::uid_t, gid as libc::gid_t)
+    }))
+}
+
+pub fn readlink(p: &Path) -> IoResult<Path> {
+    let c_path = p.to_c_str();
+    let p = c_path.as_ptr();
+    let mut len = unsafe { libc::pathconf(p as *mut _, libc::_PC_NAME_MAX) };
+    if len == -1 {
+        len = 1024; // FIXME: read PATH_MAX from C ffi?
+    }
+    let mut buf: Vec<u8> = Vec::with_capacity(len as uint);
+    match unsafe {
+        libc::readlink(p, buf.as_ptr() as *mut libc::c_char,
+                       len as libc::size_t) as libc::c_int
+    } {
+        -1 => Err(super::last_error()),
+        n => {
+            assert!(n > 0);
+            unsafe { buf.set_len(n as uint); }
+            Ok(Path::new(buf))
+        }
+    }
+}
+
+pub fn symlink(src: &Path, dst: &Path) -> IoResult<()> {
+    let src = src.to_c_str();
+    let dst = dst.to_c_str();
+    mkerr_libc(unsafe { libc::symlink(src.as_ptr(), dst.as_ptr()) })
+}
+
+pub fn link(src: &Path, dst: &Path) -> IoResult<()> {
+    let src = src.to_c_str();
+    let dst = dst.to_c_str();
+    mkerr_libc(unsafe { libc::link(src.as_ptr(), dst.as_ptr()) })
+}
+
+fn mkstat(stat: &libc::stat) -> FileStat {
+    // FileStat times are in milliseconds
+    fn mktime(secs: u64, nsecs: u64) -> u64 { secs * 1000 + nsecs / 1000000 }
+
+    #[cfg(not(any(target_os = "linux", target_os = "android")))]
+    fn flags(stat: &libc::stat) -> u64 { stat.st_flags as u64 }
+    #[cfg(any(target_os = "linux", target_os = "android"))]
+    fn flags(_stat: &libc::stat) -> u64 { 0 }
+
+    #[cfg(not(any(target_os = "linux", target_os = "android")))]
+    fn gen(stat: &libc::stat) -> u64 { stat.st_gen as u64 }
+    #[cfg(any(target_os = "linux", target_os = "android"))]
+    fn gen(_stat: &libc::stat) -> u64 { 0 }
+
+    FileStat {
+        size: stat.st_size as u64,
+        kind: match (stat.st_mode as libc::mode_t) & libc::S_IFMT {
+            libc::S_IFREG => io::TypeFile,
+            libc::S_IFDIR => io::TypeDirectory,
+            libc::S_IFIFO => io::TypeNamedPipe,
+            libc::S_IFBLK => io::TypeBlockSpecial,
+            libc::S_IFLNK => io::TypeSymlink,
+            _ => io::TypeUnknown,
+        },
+        perm: FilePermission::from_bits_truncate(stat.st_mode as u32),
+        created: mktime(stat.st_ctime as u64, stat.st_ctime_nsec as u64),
+        modified: mktime(stat.st_mtime as u64, stat.st_mtime_nsec as u64),
+        accessed: mktime(stat.st_atime as u64, stat.st_atime_nsec as u64),
+        unstable: UnstableFileStat {
+            device: stat.st_dev as u64,
+            inode: stat.st_ino as u64,
+            rdev: stat.st_rdev as u64,
+            nlink: stat.st_nlink as u64,
+            uid: stat.st_uid as u64,
+            gid: stat.st_gid as u64,
+            blksize: stat.st_blksize as u64,
+            blocks: stat.st_blocks as u64,
+            flags: flags(stat),
+            gen: gen(stat),
+        },
+    }
+}
+
+pub fn stat(p: &Path) -> IoResult<FileStat> {
+    let p = p.to_c_str();
+    let mut stat: libc::stat = unsafe { mem::zeroed() };
+    match unsafe { libc::stat(p.as_ptr(), &mut stat) } {
+        0 => Ok(mkstat(&stat)),
+        _ => Err(super::last_error()),
+    }
+}
+
+pub fn lstat(p: &Path) -> IoResult<FileStat> {
+    let p = p.to_c_str();
+    let mut stat: libc::stat = unsafe { mem::zeroed() };
+    match unsafe { libc::lstat(p.as_ptr(), &mut stat) } {
+        0 => Ok(mkstat(&stat)),
+        _ => Err(super::last_error()),
+    }
+}
+
+pub fn utime(p: &Path, atime: u64, mtime: u64) -> IoResult<()> {
+    let p = p.to_c_str();
+    let buf = libc::utimbuf {
+        actime: (atime / 1000) as libc::time_t,
+        modtime: (mtime / 1000) as libc::time_t,
+    };
+    mkerr_libc(unsafe { libc::utime(p.as_ptr(), &buf) })
+}
+
+#[cfg(test)]
+mod tests {
+    use super::FileDesc;
+    use libc;
+    use os;
+    use prelude::*;
+
+    #[cfg_attr(target_os = "freebsd", ignore)] // hmm, maybe pipes have a tiny buffer
+    #[test]
+    fn test_file_desc() {
+        // Run this test with some pipes so we don't have to mess around with
+        // opening or closing files.
+        let os::Pipe { reader, writer } = unsafe { os::pipe().unwrap() };
+        let mut reader = FileDesc::new(reader, true);
+        let mut writer = FileDesc::new(writer, true);
+
+        writer.write(b"test").ok().unwrap();
+        let mut buf = [0u8, ..4];
+        match reader.read(buf) {
+            Ok(4) => {
+                assert_eq!(buf[0], 't' as u8);
+                assert_eq!(buf[1], 'e' as u8);
+                assert_eq!(buf[2], 's' as u8);
+                assert_eq!(buf[3], 't' as u8);
+            }
+            r => panic!("invalid read: {}", r),
+        }
+
+        assert!(writer.read(buf).is_err());
+        assert!(reader.write(buf).is_err());
+    }
+}
diff --git a/src/libstd/sys/unix/mod.rs b/src/libstd/sys/unix/mod.rs
index 6295864e0e1..b404dc7fdbd 100644
--- a/src/libstd/sys/unix/mod.rs
+++ b/src/libstd/sys/unix/mod.rs
@@ -17,7 +17,6 @@ use prelude::*;
 use io::{mod, IoResult, IoError};
 use sys_common::mkerr_libc;
 
-
 macro_rules! helper_init( (static $name:ident: Helper<$m:ty>) => (
     static $name: Helper<$m> = Helper {
         lock: ::rt::mutex::NATIVE_MUTEX_INIT,
@@ -34,6 +33,7 @@ pub mod tcp;
 pub mod udp;
 pub mod pipe;
 pub mod helper_signal;
+pub mod process;
 
 pub mod addrinfo {
     pub use sys_common::net::get_host_addresses;
diff --git a/src/libstd/sys/unix/process.rs b/src/libstd/sys/unix/process.rs
new file mode 100644
index 00000000000..0965d98d9b0
--- /dev/null
+++ b/src/libstd/sys/unix/process.rs
@@ -0,0 +1,587 @@
+// Copyright 2014 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+use libc::{mod, pid_t, c_void, c_int};
+use c_str::CString;
+use io::{mod, IoResult, IoError};
+use mem;
+use os;
+use ptr;
+use prelude::*;
+use io::process::{ProcessExit, ExitStatus, ExitSignal};
+use collections;
+use path::BytesContainer;
+use hash::Hash;
+
+use sys::{mod, retry, c, wouldblock, set_nonblocking, ms_to_timeval};
+use sys::fs::FileDesc;
+use sys_common::helper_thread::Helper;
+use sys_common::{AsFileDesc, mkerr_libc, timeout};
+
+pub use sys_common::ProcessConfig;
+
+helper_init!(static HELPER: Helper<Req>)
+
+/// The unique id of the process (this should never be negative).
+pub struct Process {
+    pub pid: pid_t
+}
+
+enum Req {
+    NewChild(libc::pid_t, Sender<ProcessExit>, u64),
+}
+
+impl Process {
+    pub fn id(&self) -> pid_t {
+        self.pid
+    }
+
+    pub unsafe fn kill(&self, signal: int) -> IoResult<()> {
+        Process::killpid(self.pid, signal)
+    }
+
+    pub unsafe fn killpid(pid: pid_t, signal: int) -> IoResult<()> {
+        let r = libc::funcs::posix88::signal::kill(pid, signal as c_int);
+        mkerr_libc(r)
+    }
+
+    pub fn spawn<K, V, C, P>(cfg: &C, in_fd: Option<P>,
+                              out_fd: Option<P>, err_fd: Option<P>)
+                              -> IoResult<Process>
+        where C: ProcessConfig<K, V>, P: AsFileDesc,
+              K: BytesContainer + Eq + Hash, V: BytesContainer
+    {
+        use libc::funcs::posix88::unistd::{fork, dup2, close, chdir, execvp};
+        use libc::funcs::bsd44::getdtablesize;
+
+        mod rustrt {
+            extern {
+                pub fn rust_unset_sigprocmask();
+            }
+        }
+
+        #[cfg(target_os = "macos")]
+        unsafe fn set_environ(envp: *const c_void) {
+            extern { fn _NSGetEnviron() -> *mut *const c_void; }
+
+            *_NSGetEnviron() = envp;
+        }
+        #[cfg(not(target_os = "macos"))]
+        unsafe fn set_environ(envp: *const c_void) {
+            extern { static mut environ: *const c_void; }
+            environ = envp;
+        }
+
+        unsafe fn set_cloexec(fd: c_int) {
+            let ret = c::ioctl(fd, c::FIOCLEX);
+            assert_eq!(ret, 0);
+        }
+
+        let dirp = cfg.cwd().map(|c| c.as_ptr()).unwrap_or(ptr::null());
+
+        // temporary until unboxed closures land
+        let cfg = unsafe {
+            mem::transmute::<&ProcessConfig<K,V>,&'static ProcessConfig<K,V>>(cfg)
+        };
+
+        with_envp(cfg.env(), proc(envp) {
+            with_argv(cfg.program(), cfg.args(), proc(argv) unsafe {
+                let (input, mut output) = try!(sys::os::pipe());
+
+                // We may use this in the child, so perform allocations before the
+                // fork
+                let devnull = "/dev/null".to_c_str();
+
+                set_cloexec(output.fd());
+
+                let pid = fork();
+                if pid < 0 {
+                    return Err(super::last_error())
+                } else if pid > 0 {
+                    drop(output);
+                    let mut bytes = [0, ..4];
+                    return match input.read(bytes) {
+                        Ok(4) => {
+                            let errno = (bytes[0] as i32 << 24) |
+                                        (bytes[1] as i32 << 16) |
+                                        (bytes[2] as i32 <<  8) |
+                                        (bytes[3] as i32 <<  0);
+                            Err(super::decode_error(errno))
+                        }
+                        Err(..) => Ok(Process { pid: pid }),
+                        Ok(..) => panic!("short read on the cloexec pipe"),
+                    };
+                }
+
+                // And at this point we've reached a special time in the life of the
+                // child. The child must now be considered hamstrung and unable to
+                // do anything other than syscalls really. Consider the following
+                // scenario:
+                //
+                //      1. Thread A of process 1 grabs the malloc() mutex
+                //      2. Thread B of process 1 forks(), creating thread C
+                //      3. Thread C of process 2 then attempts to malloc()
+                //      4. The memory of process 2 is the same as the memory of
+                //         process 1, so the mutex is locked.
+                //
+                // This situation looks a lot like deadlock, right? It turns out
+                // that this is what pthread_atfork() takes care of, which is
+                // presumably implemented across platforms. The first thing that
+                // threads to *before* forking is to do things like grab the malloc
+                // mutex, and then after the fork they unlock it.
+                //
+                // Despite this information, libnative's spawn has been witnessed to
+                // deadlock on both OSX and FreeBSD. I'm not entirely sure why, but
+                // all collected backtraces point at malloc/free traffic in the
+                // child spawned process.
+                //
+                // For this reason, the block of code below should contain 0
+                // invocations of either malloc of free (or their related friends).
+                //
+                // As an example of not having malloc/free traffic, we don't close
+                // this file descriptor by dropping the FileDesc (which contains an
+                // allocation). Instead we just close it manually. This will never
+                // have the drop glue anyway because this code never returns (the
+                // child will either exec() or invoke libc::exit)
+                let _ = libc::close(input.fd());
+
+                fn fail(output: &mut FileDesc) -> ! {
+                    let errno = sys::os::errno();
+                    let bytes = [
+                        (errno >> 24) as u8,
+                        (errno >> 16) as u8,
+                        (errno >>  8) as u8,
+                        (errno >>  0) as u8,
+                    ];
+                    assert!(output.write(bytes).is_ok());
+                    unsafe { libc::_exit(1) }
+                }
+
+                rustrt::rust_unset_sigprocmask();
+
+                // If a stdio file descriptor is set to be ignored (via a -1 file
+                // descriptor), then we don't actually close it, but rather open
+                // up /dev/null into that file descriptor. Otherwise, the first file
+                // descriptor opened up in the child would be numbered as one of the
+                // stdio file descriptors, which is likely to wreak havoc.
+                let setup = |src: Option<P>, dst: c_int| {
+                    let src = match src {
+                        None => {
+                            let flags = if dst == libc::STDIN_FILENO {
+                                libc::O_RDONLY
+                            } else {
+                                libc::O_RDWR
+                            };
+                            libc::open(devnull.as_ptr(), flags, 0)
+                        }
+                        Some(obj) => {
+                            let fd = obj.as_fd().fd();
+                            // Leak the memory and the file descriptor. We're in the
+                            // child now an all our resources are going to be
+                            // cleaned up very soon
+                            mem::forget(obj);
+                            fd
+                        }
+                    };
+                    src != -1 && retry(|| dup2(src, dst)) != -1
+                };
+
+                if !setup(in_fd, libc::STDIN_FILENO) { fail(&mut output) }
+                if !setup(out_fd, libc::STDOUT_FILENO) { fail(&mut output) }
+                if !setup(err_fd, libc::STDERR_FILENO) { fail(&mut output) }
+
+                // close all other fds
+                for fd in range(3, getdtablesize()).rev() {
+                    if fd != output.fd() {
+                        let _ = close(fd as c_int);
+                    }
+                }
+
+                match cfg.gid() {
+                    Some(u) => {
+                        if libc::setgid(u as libc::gid_t) != 0 {
+                            fail(&mut output);
+                        }
+                    }
+                    None => {}
+                }
+                match cfg.uid() {
+                    Some(u) => {
+                        // When dropping privileges from root, the `setgroups` call
+                        // will remove any extraneous groups. If we don't call this,
+                        // then even though our uid has dropped, we may still have
+                        // groups that enable us to do super-user things. This will
+                        // fail if we aren't root, so don't bother checking the
+                        // return value, this is just done as an optimistic
+                        // privilege dropping function.
+                        extern {
+                            fn setgroups(ngroups: libc::c_int,
+                                         ptr: *const libc::c_void) -> libc::c_int;
+                        }
+                        let _ = setgroups(0, 0 as *const libc::c_void);
+
+                        if libc::setuid(u as libc::uid_t) != 0 {
+                            fail(&mut output);
+                        }
+                    }
+                    None => {}
+                }
+                if cfg.detach() {
+                    // Don't check the error of setsid because it fails if we're the
+                    // process leader already. We just forked so it shouldn't return
+                    // error, but ignore it anyway.
+                    let _ = libc::setsid();
+                }
+                if !dirp.is_null() && chdir(dirp) == -1 {
+                    fail(&mut output);
+                }
+                if !envp.is_null() {
+                    set_environ(envp);
+                }
+                let _ = execvp(*argv, argv as *mut _);
+                fail(&mut output);
+            })
+        })
+    }
+
+    pub fn wait(&self, deadline: u64) -> IoResult<ProcessExit> {
+        use std::cmp;
+        use std::comm;
+
+        static mut WRITE_FD: libc::c_int = 0;
+
+        let mut status = 0 as c_int;
+        if deadline == 0 {
+            return match retry(|| unsafe { c::waitpid(self.pid, &mut status, 0) }) {
+                -1 => panic!("unknown waitpid error: {}", super::last_error()),
+                _ => Ok(translate_status(status)),
+            }
+        }
+
+        // On unix, wait() and its friends have no timeout parameters, so there is
+        // no way to time out a thread in wait(). From some googling and some
+        // thinking, it appears that there are a few ways to handle timeouts in
+        // wait(), but the only real reasonable one for a multi-threaded program is
+        // to listen for SIGCHLD.
+        //
+        // With this in mind, the waiting mechanism with a timeout barely uses
+        // waitpid() at all. There are a few times that waitpid() is invoked with
+        // WNOHANG, but otherwise all the necessary blocking is done by waiting for
+        // a SIGCHLD to arrive (and that blocking has a timeout). Note, however,
+        // that waitpid() is still used to actually reap the child.
+        //
+        // Signal handling is super tricky in general, and this is no exception. Due
+        // to the async nature of SIGCHLD, we use the self-pipe trick to transmit
+        // data out of the signal handler to the rest of the application. The first
+        // idea would be to have each thread waiting with a timeout to read this
+        // output file descriptor, but a write() is akin to a signal(), not a
+        // broadcast(), so it would only wake up one thread, and possibly the wrong
+        // thread. Hence a helper thread is used.
+        //
+        // The helper thread here is responsible for farming requests for a
+        // waitpid() with a timeout, and then processing all of the wait requests.
+        // By guaranteeing that only this helper thread is reading half of the
+        // self-pipe, we're sure that we'll never lose a SIGCHLD. This helper thread
+        // is also responsible for select() to wait for incoming messages or
+        // incoming SIGCHLD messages, along with passing an appropriate timeout to
+        // select() to wake things up as necessary.
+        //
+        // The ordering of the following statements is also very purposeful. First,
+        // we must be guaranteed that the helper thread is booted and available to
+        // receive SIGCHLD signals, and then we must also ensure that we do a
+        // nonblocking waitpid() at least once before we go ask the sigchld helper.
+        // This prevents the race where the child exits, we boot the helper, and
+        // then we ask for the child's exit status (never seeing a sigchld).
+        //
+        // The actual communication between the helper thread and this thread is
+        // quite simple, just a channel moving data around.
+
+        unsafe { HELPER.boot(register_sigchld, waitpid_helper) }
+
+        match self.try_wait() {
+            Some(ret) => return Ok(ret),
+            None => {}
+        }
+
+        let (tx, rx) = channel();
+        unsafe { HELPER.send(NewChild(self.pid, tx, deadline)); }
+        return match rx.recv_opt() {
+            Ok(e) => Ok(e),
+            Err(()) => Err(timeout("wait timed out")),
+        };
+
+        // Register a new SIGCHLD handler, returning the reading half of the
+        // self-pipe plus the old handler registered (return value of sigaction).
+        //
+        // Be sure to set up the self-pipe first because as soon as we register a
+        // handler we're going to start receiving signals.
+        fn register_sigchld() -> (libc::c_int, c::sigaction) {
+            unsafe {
+                let mut pipes = [0, ..2];
+                assert_eq!(libc::pipe(pipes.as_mut_ptr()), 0);
+                set_nonblocking(pipes[0], true).ok().unwrap();
+                set_nonblocking(pipes[1], true).ok().unwrap();
+                WRITE_FD = pipes[1];
+
+                let mut old: c::sigaction = mem::zeroed();
+                let mut new: c::sigaction = mem::zeroed();
+                new.sa_handler = sigchld_handler;
+                new.sa_flags = c::SA_NOCLDSTOP;
+                assert_eq!(c::sigaction(c::SIGCHLD, &new, &mut old), 0);
+                (pipes[0], old)
+            }
+        }
+
+        // Helper thread for processing SIGCHLD messages
+        fn waitpid_helper(input: libc::c_int,
+                          messages: Receiver<Req>,
+                          (read_fd, old): (libc::c_int, c::sigaction)) {
+            set_nonblocking(input, true).ok().unwrap();
+            let mut set: c::fd_set = unsafe { mem::zeroed() };
+            let mut tv: libc::timeval;
+            let mut active = Vec::<(libc::pid_t, Sender<ProcessExit>, u64)>::new();
+            let max = cmp::max(input, read_fd) + 1;
+
+            'outer: loop {
+                // Figure out the timeout of our syscall-to-happen. If we're waiting
+                // for some processes, then they'll have a timeout, otherwise we
+                // wait indefinitely for a message to arrive.
+                //
+                // FIXME: sure would be nice to not have to scan the entire array
+                let min = active.iter().map(|a| *a.ref2()).enumerate().min_by(|p| {
+                    p.val1()
+                });
+                let (p, idx) = match min {
+                    Some((idx, deadline)) => {
+                        let now = sys::timer::now();
+                        let ms = if now < deadline {deadline - now} else {0};
+                        tv = ms_to_timeval(ms);
+                        (&mut tv as *mut _, idx)
+                    }
+                    None => (ptr::null_mut(), -1),
+                };
+
+                // Wait for something to happen
+                c::fd_set(&mut set, input);
+                c::fd_set(&mut set, read_fd);
+                match unsafe { c::select(max, &mut set, ptr::null_mut(),
+                                         ptr::null_mut(), p) } {
+                    // interrupted, retry
+                    -1 if os::errno() == libc::EINTR as uint => continue,
+
+                    // We read something, break out and process
+                    1 | 2 => {}
+
+                    // Timeout, the pending request is removed
+                    0 => {
+                        drop(active.remove(idx));
+                        continue
+                    }
+
+                    n => panic!("error in select {} ({})", os::errno(), n),
+                }
+
+                // Process any pending messages
+                if drain(input) {
+                    loop {
+                        match messages.try_recv() {
+                            Ok(NewChild(pid, tx, deadline)) => {
+                                active.push((pid, tx, deadline));
+                            }
+                            Err(comm::Disconnected) => {
+                                assert!(active.len() == 0);
+                                break 'outer;
+                            }
+                            Err(comm::Empty) => break,
+                        }
+                    }
+                }
+
+                // If a child exited (somehow received SIGCHLD), then poll all
+                // children to see if any of them exited.
+                //
+                // We also attempt to be responsible netizens when dealing with
+                // SIGCHLD by invoking any previous SIGCHLD handler instead of just
+                // ignoring any previous SIGCHLD handler. Note that we don't provide
+                // a 1:1 mapping of our handler invocations to the previous handler
+                // invocations because we drain the `read_fd` entirely. This is
+                // probably OK because the kernel is already allowed to coalesce
+                // simultaneous signals, we're just doing some extra coalescing.
+                //
+                // Another point of note is that this likely runs the signal handler
+                // on a different thread than the one that received the signal. I
+                // *think* this is ok at this time.
+                //
+                // The main reason for doing this is to allow stdtest to run native
+                // tests as well. Both libgreen and libnative are running around
+                // with process timeouts, but libgreen should get there first
+                // (currently libuv doesn't handle old signal handlers).
+                if drain(read_fd) {
+                    let i: uint = unsafe { mem::transmute(old.sa_handler) };
+                    if i != 0 {
+                        assert!(old.sa_flags & c::SA_SIGINFO == 0);
+                        (old.sa_handler)(c::SIGCHLD);
+                    }
+
+                    // FIXME: sure would be nice to not have to scan the entire
+                    //        array...
+                    active.retain(|&(pid, ref tx, _)| {
+                        let pr = Process { pid: pid };
+                        match pr.try_wait() {
+                            Some(msg) => { tx.send(msg); false }
+                            None => true,
+                        }
+                    });
+                }
+            }
+
+            // Once this helper thread is done, we re-register the old sigchld
+            // handler and close our intermediate file descriptors.
+            unsafe {
+                assert_eq!(c::sigaction(c::SIGCHLD, &old, ptr::null_mut()), 0);
+                let _ = libc::close(read_fd);
+                let _ = libc::close(WRITE_FD);
+                WRITE_FD = -1;
+            }
+        }
+
+        // Drain all pending data from the file descriptor, returning if any data
+        // could be drained. This requires that the file descriptor is in
+        // nonblocking mode.
+        fn drain(fd: libc::c_int) -> bool {
+            let mut ret = false;
+            loop {
+                let mut buf = [0u8, ..1];
+                match unsafe {
+                    libc::read(fd, buf.as_mut_ptr() as *mut libc::c_void,
+                               buf.len() as libc::size_t)
+                } {
+                    n if n > 0 => { ret = true; }
+                    0 => return true,
+                    -1 if wouldblock() => return ret,
+                    n => panic!("bad read {} ({})", os::last_os_error(), n),
+                }
+            }
+        }
+
+        // Signal handler for SIGCHLD signals, must be async-signal-safe!
+        //
+        // This function will write to the writing half of the "self pipe" to wake
+        // up the helper thread if it's waiting. Note that this write must be
+        // nonblocking because if it blocks and the reader is the thread we
+        // interrupted, then we'll deadlock.
+        //
+        // When writing, if the write returns EWOULDBLOCK then we choose to ignore
+        // it. At that point we're guaranteed that there's something in the pipe
+        // which will wake up the other end at some point, so we just allow this
+        // signal to be coalesced with the pending signals on the pipe.
+        extern fn sigchld_handler(_signum: libc::c_int) {
+            let msg = 1i;
+            match unsafe {
+                libc::write(WRITE_FD, &msg as *const _ as *const libc::c_void, 1)
+            } {
+                1 => {}
+                -1 if wouldblock() => {} // see above comments
+                n => panic!("bad error on write fd: {} {}", n, os::errno()),
+            }
+        }
+    }
+
+    pub fn try_wait(&self) -> Option<ProcessExit> {
+        let mut status = 0 as c_int;
+        match retry(|| unsafe {
+            c::waitpid(self.pid, &mut status, c::WNOHANG)
+        }) {
+            n if n == self.pid => Some(translate_status(status)),
+            0 => None,
+            n => panic!("unknown waitpid error `{}`: {}", n,
+                       super::last_error()),
+        }
+    }
+}
+
+fn with_argv<T>(prog: &CString, args: &[CString],
+                cb: proc(*const *const libc::c_char) -> T) -> T {
+    let mut ptrs: Vec<*const libc::c_char> = Vec::with_capacity(args.len()+1);
+
+    // Convert the CStrings into an array of pointers. Note: the
+    // lifetime of the various CStrings involved is guaranteed to be
+    // larger than the lifetime of our invocation of cb, but this is
+    // technically unsafe as the callback could leak these pointers
+    // out of our scope.
+    ptrs.push(prog.as_ptr());
+    ptrs.extend(args.iter().map(|tmp| tmp.as_ptr()));
+
+    // Add a terminating null pointer (required by libc).
+    ptrs.push(ptr::null());
+
+    cb(ptrs.as_ptr())
+}
+
+fn with_envp<K, V, T>(env: Option<&collections::HashMap<K, V>>,
+                      cb: proc(*const c_void) -> T) -> T
+    where K: BytesContainer + Eq + Hash, V: BytesContainer
+{
+    // On posixy systems we can pass a char** for envp, which is a
+    // null-terminated array of "k=v\0" strings. Since we must create
+    // these strings locally, yet expose a raw pointer to them, we
+    // create a temporary vector to own the CStrings that outlives the
+    // call to cb.
+    match env {
+        Some(env) => {
+            let mut tmps = Vec::with_capacity(env.len());
+
+            for pair in env.iter() {
+                let mut kv = Vec::new();
+                kv.push_all(pair.ref0().container_as_bytes());
+                kv.push('=' as u8);
+                kv.push_all(pair.ref1().container_as_bytes());
+                kv.push(0); // terminating null
+                tmps.push(kv);
+            }
+
+            // As with `with_argv`, this is unsafe, since cb could leak the pointers.
+            let mut ptrs: Vec<*const libc::c_char> =
+                tmps.iter()
+                    .map(|tmp| tmp.as_ptr() as *const libc::c_char)
+                    .collect();
+            ptrs.push(ptr::null());
+
+            cb(ptrs.as_ptr() as *const c_void)
+        }
+        _ => cb(ptr::null())
+    }
+}
+
+fn translate_status(status: c_int) -> ProcessExit {
+    #![allow(non_snake_case)]
+    #[cfg(any(target_os = "linux", target_os = "android"))]
+    mod imp {
+        pub fn WIFEXITED(status: i32) -> bool { (status & 0xff) == 0 }
+        pub fn WEXITSTATUS(status: i32) -> i32 { (status >> 8) & 0xff }
+        pub fn WTERMSIG(status: i32) -> i32 { status & 0x7f }
+    }
+
+    #[cfg(any(target_os = "macos",
+              target_os = "ios",
+              target_os = "freebsd",
+              target_os = "dragonfly"))]
+    mod imp {
+        pub fn WIFEXITED(status: i32) -> bool { (status & 0x7f) == 0 }
+        pub fn WEXITSTATUS(status: i32) -> i32 { status >> 8 }
+        pub fn WTERMSIG(status: i32) -> i32 { status & 0o177 }
+    }
+
+    if imp::WIFEXITED(status) {
+        ExitStatus(imp::WEXITSTATUS(status) as int)
+    } else {
+        ExitSignal(imp::WTERMSIG(status) as int)
+    }
+}