build: separate compiler and libs

compiler/internal/lib/syscall/env_unix.go

@@ -0,0 +1,174 @@
+// Copyright 2010 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build unix || (js && wasm) || plan9 || wasip1
+
+// Unix environment variables.
+
+package syscall
+
+import (
+	"runtime"
+	"sync"
+
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/os"
+	"github.com/goplus/llgo/c/syscall"
+)
+
+var (
+	// envOnce guards initialization by copyenv, which populates env.
+	envOnce sync.Once
+
+	// envLock guards env and envs.
+	envLock sync.RWMutex
+
+	// env maps from an environment variable to its first occurrence in envs.
+	env map[string]int
+
+	// envs is provided by the runtime. elements are expected to
+	// be of the form "key=value". An empty string means deleted
+	// (or a duplicate to be ignored).
+	envs []string = runtimeEnvs()
+)
+
+func runtimeEnvs() []string {
+	ret := make([]string, 0, 8)
+	cenvs := os.Environ
+	i := 0
+	for {
+		ce := *c.Advance(cenvs, i)
+		if ce == nil {
+			return ret
+		}
+		ret = append(ret, c.GoString(ce))
+		i++
+	}
+}
+
+func copyenv() {
+	env = make(map[string]int)
+	for i, s := range envs {
+		for j := 0; j < len(s); j++ {
+			if s[j] == '=' {
+				key := s[:j]
+				if _, ok := env[key]; !ok {
+					env[key] = i // first mention of key
+				} else {
+					// Clear duplicate keys. This permits Unsetenv to
+					// safely delete only the first item without
+					// worrying about unshadowing a later one,
+					// which might be a security problem.
+					envs[i] = ""
+				}
+				break
+			}
+		}
+	}
+}
+
+func Unsetenv(key string) error {
+	envOnce.Do(copyenv)
+
+	envLock.Lock()
+	defer envLock.Unlock()
+
+	if i, ok := env[key]; ok {
+		envs[i] = ""
+		delete(env, key)
+	}
+	os.Unsetenv(c.AllocaCStr(key))
+	return nil
+}
+
+func Getenv(key string) (value string, found bool) {
+	envOnce.Do(copyenv)
+	if len(key) == 0 {
+		return "", false
+	}
+
+	envLock.RLock()
+	defer envLock.RUnlock()
+
+	i, ok := env[key]
+	if !ok {
+		return "", false
+	}
+	s := envs[i]
+	for i := 0; i < len(s); i++ {
+		if s[i] == '=' {
+			return s[i+1:], true
+		}
+	}
+	return "", false
+}
+
+/* TODO(xsw):
+func Getenv(key string) (value string, found bool) {
+	ret := os.Getenv(c.AllocaCStr(key))
+	if ret != nil {
+		return c.GoString(ret), true
+	}
+	return "", false
+}
+*/
+
+func Setenv(key, value string) error {
+	envOnce.Do(copyenv)
+	if len(key) == 0 {
+		return Errno(syscall.EINVAL)
+	}
+	for i := 0; i < len(key); i++ {
+		if key[i] == '=' || key[i] == 0 {
+			return Errno(syscall.EINVAL)
+		}
+	}
+	// On Plan 9, null is used as a separator, eg in $path.
+	if runtime.GOOS != "plan9" {
+		for i := 0; i < len(value); i++ {
+			if value[i] == 0 {
+				return Errno(syscall.EINVAL)
+			}
+		}
+	}
+
+	envLock.Lock()
+	defer envLock.Unlock()
+
+	i, ok := env[key]
+	kv := key + "=" + value
+	if ok {
+		envs[i] = kv
+	} else {
+		i = len(envs)
+		envs = append(envs, kv)
+	}
+	env[key] = i
+	os.Setenv(c.AllocaCStr(key), c.AllocaCStr(value), 1)
+	return nil
+}
+
+func Clearenv() {
+	envOnce.Do(copyenv) // prevent copyenv in Getenv/Setenv
+
+	envLock.Lock()
+	defer envLock.Unlock()
+
+	os.Clearenv()
+	env = make(map[string]int)
+	envs = []string{}
+}
+
+func Environ() []string {
+	envOnce.Do(copyenv)
+	envLock.RLock()
+	defer envLock.RUnlock()
+	a := make([]string, 0, len(envs))
+	for _, env := range envs {
+		if env != "" {
+			a = append(a, env)
+		}
+	}
+	return a
+}

compiler/internal/lib/syscall/env_windows.go

@@ -0,0 +1,96 @@
+// Copyright 2010 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Windows environment variables.
+
+package syscall
+
+import (
+	"unsafe"
+)
+
+func Getenv(key string) (value string, found bool) {
+	keyp, err := UTF16PtrFromString(key)
+	if err != nil {
+		return "", false
+	}
+	n := uint32(100)
+	for {
+		b := make([]uint16, n)
+		n, err = GetEnvironmentVariable(keyp, &b[0], uint32(len(b)))
+		if n == 0 && err == ERROR_ENVVAR_NOT_FOUND {
+			return "", false
+		}
+		if n <= uint32(len(b)) {
+			return UTF16ToString(b[:n]), true
+		}
+	}
+}
+
+func Setenv(key, value string) error {
+	v, err := UTF16PtrFromString(value)
+	if err != nil {
+		return err
+	}
+	keyp, err := UTF16PtrFromString(key)
+	if err != nil {
+		return err
+	}
+	e := SetEnvironmentVariable(keyp, v)
+	if e != nil {
+		return e
+	}
+	runtimeSetenv(key, value)
+	return nil
+}
+
+func Unsetenv(key string) error {
+	keyp, err := UTF16PtrFromString(key)
+	if err != nil {
+		return err
+	}
+	e := SetEnvironmentVariable(keyp, nil)
+	if e != nil {
+		return e
+	}
+	runtimeUnsetenv(key)
+	return nil
+}
+
+func Clearenv() {
+	for _, s := range Environ() {
+		// Environment variables can begin with =
+		// so start looking for the separator = at j=1.
+		// https://devblogs.microsoft.com/oldnewthing/20100506-00/?p=14133
+		for j := 1; j < len(s); j++ {
+			if s[j] == '=' {
+				Unsetenv(s[0:j])
+				break
+			}
+		}
+	}
+}
+
+func Environ() []string {
+	envp, e := GetEnvironmentStrings()
+	if e != nil {
+		return nil
+	}
+	defer FreeEnvironmentStrings(envp)
+
+	r := make([]string, 0, 50) // Empty with room to grow.
+	const size = unsafe.Sizeof(*envp)
+	for *envp != 0 { // environment block ends with empty string
+		// find NUL terminator
+		end := unsafe.Pointer(envp)
+		for *(*uint16)(end) != 0 {
+			end = unsafe.Add(end, size)
+		}
+
+		entry := unsafe.Slice(envp, (uintptr(end)-uintptr(unsafe.Pointer(envp)))/size)
+		r = append(r, UTF16ToString(entry))
+		envp = (*uint16)(unsafe.Add(end, size))
+	}
+	return r
+}

compiler/internal/lib/syscall/exec_libc.go

@@ -0,0 +1,321 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build aix || solaris
+
+// This file handles forkAndExecInChild function for OS using libc syscall like AIX or Solaris.
+
+package syscall
+
+import (
+	"runtime"
+	"unsafe"
+
+	"github.com/goplus/llgo/c"
+)
+
+type SysProcAttr struct {
+	Chroot     string      // Chroot.
+	Credential *Credential // Credential.
+	Setsid     bool        // Create session.
+	// Setpgid sets the process group ID of the child to Pgid,
+	// or, if Pgid == 0, to the new child's process ID.
+	Setpgid bool
+	// Setctty sets the controlling terminal of the child to
+	// file descriptor Ctty. Ctty must be a descriptor number
+	// in the child process: an index into ProcAttr.Files.
+	// This is only meaningful if Setsid is true.
+	Setctty bool
+	Noctty  bool // Detach fd 0 from controlling terminal
+	Ctty    int  // Controlling TTY fd
+	// Foreground places the child process group in the foreground.
+	// This implies Setpgid. The Ctty field must be set to
+	// the descriptor of the controlling TTY.
+	// Unlike Setctty, in this case Ctty must be a descriptor
+	// number in the parent process.
+	Foreground bool
+	Pgid       int // Child's process group ID if Setpgid.
+}
+
+// Implemented in runtime package.
+func runtime_BeforeFork()
+func runtime_AfterFork()
+func runtime_AfterForkInChild()
+
+func chdir(path uintptr) (err Errno)
+func chroot1(path uintptr) (err Errno)
+func closeFD(fd uintptr) (err Errno)
+func dup2child(old uintptr, new uintptr) (val uintptr, err Errno)
+func execve(path uintptr, argv uintptr, envp uintptr) (err Errno)
+func exit(code uintptr)
+func fcntl1(fd uintptr, cmd uintptr, arg uintptr) (val uintptr, err Errno)
+func forkx(flags uintptr) (pid uintptr, err Errno)
+func getpid() (pid uintptr, err Errno)
+func ioctl(fd uintptr, req uintptr, arg uintptr) (err Errno)
+func setgid(gid uintptr) (err Errno)
+func setgroups1(ngid uintptr, gid uintptr) (err Errno)
+func setrlimit1(which uintptr, lim unsafe.Pointer) (err Errno)
+func setsid() (pid uintptr, err Errno)
+func setuid(uid uintptr) (err Errno)
+func setpgid(pid uintptr, pgid uintptr) (err Errno)
+func write1(fd uintptr, buf uintptr, nbyte uintptr) (n uintptr, err Errno)
+
+// syscall defines this global on our behalf to avoid a build dependency on other platforms
+func init() {
+	execveLibc = execve
+}
+
+// Fork, dup fd onto 0..len(fd), and exec(argv0, argvv, envv) in child.
+// If a dup or exec fails, write the errno error to pipe.
+// (Pipe is close-on-exec so if exec succeeds, it will be closed.)
+// In the child, this function must not acquire any locks, because
+// they might have been locked at the time of the fork. This means
+// no rescheduling, no malloc calls, and no new stack segments.
+//
+// We call hand-crafted syscalls, implemented in
+// ../runtime/syscall_solaris.go, rather than generated libc wrappers
+// because we need to avoid lazy-loading the functions (might malloc,
+// split the stack, or acquire mutexes). We can't call RawSyscall
+// because it's not safe even for BSD-subsystem calls.
+//
+// func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
+func forkAndExecInChild(argv0 *c.Char, argv, envv **c.Char, chroot, dir *c.Char, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
+	// Declare all variables at top in case any
+	// declarations require heap allocation (e.g., err1).
+	var (
+		r1              uintptr
+		err1            Errno
+		nextfd          int
+		i               int
+		pgrp            _Pid_t
+		cred            *Credential
+		ngroups, groups uintptr
+	)
+
+	rlim, rlimOK := origRlimitNofile.Load().(Rlimit)
+
+	// guard against side effects of shuffling fds below.
+	// Make sure that nextfd is beyond any currently open files so
+	// that we can't run the risk of overwriting any of them.
+	fd := make([]int, len(attr.Files))
+	nextfd = len(attr.Files)
+	for i, ufd := range attr.Files {
+		if nextfd < int(ufd) {
+			nextfd = int(ufd)
+		}
+		fd[i] = int(ufd)
+	}
+	nextfd++
+
+	// About to call fork.
+	// No more allocation or calls of non-assembly functions.
+	runtime_BeforeFork()
+	r1, err1 = forkx(0x1) // FORK_NOSIGCHLD
+	if err1 != 0 {
+		runtime_AfterFork()
+		return 0, err1
+	}
+
+	if r1 != 0 {
+		// parent; return PID
+		runtime_AfterFork()
+		return int(r1), 0
+	}
+
+	// Fork succeeded, now in child.
+
+	// Session ID
+	if sys.Setsid {
+		_, err1 = setsid()
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// Set process group
+	if sys.Setpgid || sys.Foreground {
+		// Place child in process group.
+		err1 = setpgid(0, uintptr(sys.Pgid))
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	if sys.Foreground {
+		pgrp = _Pid_t(sys.Pgid)
+		if pgrp == 0 {
+			r1, err1 = getpid()
+			if err1 != 0 {
+				goto childerror
+			}
+
+			pgrp = _Pid_t(r1)
+		}
+
+		// Place process group in foreground.
+		err1 = ioctl(uintptr(sys.Ctty), uintptr(TIOCSPGRP), uintptr(unsafe.Pointer(&pgrp)))
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// Restore the signal mask. We do this after TIOCSPGRP to avoid
+	// having the kernel send a SIGTTOU signal to the process group.
+	runtime_AfterForkInChild()
+
+	// Chroot
+	if chroot != nil {
+		err1 = chroot1(uintptr(unsafe.Pointer(chroot)))
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// User and groups
+	if cred = sys.Credential; cred != nil {
+		ngroups = uintptr(len(cred.Groups))
+		groups = uintptr(0)
+		if ngroups > 0 {
+			groups = uintptr(unsafe.Pointer(&cred.Groups[0]))
+		}
+		if !cred.NoSetGroups {
+			err1 = setgroups1(ngroups, groups)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+		err1 = setgid(uintptr(cred.Gid))
+		if err1 != 0 {
+			goto childerror
+		}
+		err1 = setuid(uintptr(cred.Uid))
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// Chdir
+	if dir != nil {
+		err1 = chdir(uintptr(unsafe.Pointer(dir)))
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// Pass 1: look for fd[i] < i and move those up above len(fd)
+	// so that pass 2 won't stomp on an fd it needs later.
+	if pipe < nextfd {
+		switch runtime.GOOS {
+		case "illumos", "solaris":
+			_, err1 = fcntl1(uintptr(pipe), _F_DUP2FD_CLOEXEC, uintptr(nextfd))
+		default:
+			_, err1 = dup2child(uintptr(pipe), uintptr(nextfd))
+			if err1 != 0 {
+				goto childerror
+			}
+			_, err1 = fcntl1(uintptr(nextfd), F_SETFD, FD_CLOEXEC)
+		}
+		if err1 != 0 {
+			goto childerror
+		}
+		pipe = nextfd
+		nextfd++
+	}
+	for i = 0; i < len(fd); i++ {
+		if fd[i] >= 0 && fd[i] < i {
+			if nextfd == pipe { // don't stomp on pipe
+				nextfd++
+			}
+			switch runtime.GOOS {
+			case "illumos", "solaris":
+				_, err1 = fcntl1(uintptr(fd[i]), _F_DUP2FD_CLOEXEC, uintptr(nextfd))
+			default:
+				_, err1 = dup2child(uintptr(fd[i]), uintptr(nextfd))
+				if err1 != 0 {
+					goto childerror
+				}
+				_, err1 = fcntl1(uintptr(nextfd), F_SETFD, FD_CLOEXEC)
+			}
+			if err1 != 0 {
+				goto childerror
+			}
+			fd[i] = nextfd
+			nextfd++
+		}
+	}
+
+	// Pass 2: dup fd[i] down onto i.
+	for i = 0; i < len(fd); i++ {
+		if fd[i] == -1 {
+			closeFD(uintptr(i))
+			continue
+		}
+		if fd[i] == i {
+			// dup2(i, i) won't clear close-on-exec flag on Linux,
+			// probably not elsewhere either.
+			_, err1 = fcntl1(uintptr(fd[i]), F_SETFD, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+			continue
+		}
+		// The new fd is created NOT close-on-exec,
+		// which is exactly what we want.
+		_, err1 = dup2child(uintptr(fd[i]), uintptr(i))
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// By convention, we don't close-on-exec the fds we are
+	// started with, so if len(fd) < 3, close 0, 1, 2 as needed.
+	// Programs that know they inherit fds >= 3 will need
+	// to set them close-on-exec.
+	for i = len(fd); i < 3; i++ {
+		closeFD(uintptr(i))
+	}
+
+	// Detach fd 0 from tty
+	if sys.Noctty {
+		err1 = ioctl(0, uintptr(TIOCNOTTY), 0)
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// Set the controlling TTY to Ctty
+	if sys.Setctty {
+		// On AIX, TIOCSCTTY is undefined
+		if TIOCSCTTY == 0 {
+			err1 = ENOSYS
+			goto childerror
+		}
+		err1 = ioctl(uintptr(sys.Ctty), uintptr(TIOCSCTTY), 0)
+		if err1 != 0 {
+			goto childerror
+		}
+	}
+
+	// Restore original rlimit.
+	if rlimOK && rlim.Cur != 0 {
+		setrlimit1(RLIMIT_NOFILE, unsafe.Pointer(&rlim))
+	}
+
+	// Time to exec.
+	err1 = execve(
+		uintptr(unsafe.Pointer(argv0)),
+		uintptr(unsafe.Pointer(&argv[0])),
+		uintptr(unsafe.Pointer(&envv[0])))
+
+childerror:
+	// send error code on pipe
+	write1(uintptr(pipe), uintptr(unsafe.Pointer(&err1)), unsafe.Sizeof(err1))
+	for {
+		exit(253)
+	}
+}
+
+func ioctlPtr(fd, req uintptr, arg unsafe.Pointer) (err Errno) {
+	return ioctl(fd, req, uintptr(arg))
+}

compiler/internal/lib/syscall/exec_libc2.go

@@ -0,0 +1,329 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build darwin || (openbsd && !mips64)
+
+package syscall
+
+import (
+	"unsafe"
+
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/os"
+	"github.com/goplus/llgo/c/syscall"
+)
+
+type SysProcAttr struct {
+	Chroot     string      // Chroot.
+	Credential *Credential // Credential.
+	Ptrace     bool        // Enable tracing.
+	Setsid     bool        // Create session.
+	// Setpgid sets the process group ID of the child to Pgid,
+	// or, if Pgid == 0, to the new child's process ID.
+	Setpgid bool
+	// Setctty sets the controlling terminal of the child to
+	// file descriptor Ctty. Ctty must be a descriptor number
+	// in the child process: an index into ProcAttr.Files.
+	// This is only meaningful if Setsid is true.
+	Setctty bool
+	Noctty  bool // Detach fd 0 from controlling terminal
+	Ctty    int  // Controlling TTY fd
+	// Foreground places the child process group in the foreground.
+	// This implies Setpgid. The Ctty field must be set to
+	// the descriptor of the controlling TTY.
+	// Unlike Setctty, in this case Ctty must be a descriptor
+	// number in the parent process.
+	Foreground bool
+	Pgid       int // Child's process group ID if Setpgid.
+}
+
+/* TODO(xsw):
+// Implemented in runtime package.
+func runtime_BeforeFork()
+func runtime_AfterFork()
+func runtime_AfterForkInChild()
+*/
+
+// Fork, dup fd onto 0..len(fd), and exec(argv0, argvv, envv) in child.
+// If a dup or exec fails, write the errno error to pipe.
+// (Pipe is close-on-exec so if exec succeeds, it will be closed.)
+// In the child, this function must not acquire any locks, because
+// they might have been locked at the time of the fork. This means
+// no rescheduling, no malloc calls, and no new stack segments.
+// For the same reason compiler does not race instrument it.
+// The calls to rawSyscall are okay because they are assembly
+// functions that do not grow the stack.
+func forkAndExecInChild(argv0 *c.Char, argv, envv **c.Char, chroot, dir *c.Char, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err1 Errno) {
+	// Declare all variables at top in case any
+	// declarations require heap allocation (e.g., err1).
+	var (
+		r1     uintptr
+		nextfd int
+		i      int
+		cred   *Credential
+		// err             error
+		// pgrp            c.Int
+		// ngroups, groups uintptr
+	)
+
+	rlim, rlimOK := origRlimitNofile.Load().(Rlimit)
+
+	// guard against side effects of shuffling fds below.
+	// Make sure that nextfd is beyond any currently open files so
+	// that we can't run the risk of overwriting any of them.
+	fd := make([]int, len(attr.Files))
+	nextfd = len(attr.Files)
+	for i, ufd := range attr.Files {
+		if nextfd < int(ufd) {
+			nextfd = int(ufd)
+		}
+		fd[i] = int(ufd)
+	}
+	nextfd++
+
+	// About to call fork.
+	// No more allocation or calls of non-assembly functions.
+	// runtime_BeforeFork()
+	r1, err1 = fork()
+	if err1 != 0 {
+		// runtime_AfterFork()
+		return 0, err1
+	}
+
+	if r1 != 0 {
+		// parent; return PID
+		// runtime_AfterFork()
+		return int(r1), 0
+	}
+
+	// Fork succeeded, now in child.
+
+	// TODO(xsw): check this
+	// Enable tracing if requested.
+	if sys.Ptrace {
+		/* TODO(xsw):
+		if err = ptrace(PTRACE_TRACEME, 0, 0, 0); err != nil {
+			err1 = err.(Errno)
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - sys.Ptrace")
+	}
+
+	// Session ID
+	if sys.Setsid {
+		/* TODO(xsw):
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_setsid_trampoline), 0, 0, 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - sys.Setsid")
+	}
+
+	// Set process group
+	if sys.Setpgid || sys.Foreground {
+		/* TODO(xsw):
+		// Place child in process group.
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_setpgid_trampoline), 0, uintptr(sys.Pgid), 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - sys.Setpgid")
+	}
+
+	if sys.Foreground {
+		/* TODO(xsw):
+		// This should really be pid_t, however _C_int (aka int32) is
+		// generally equivalent.
+		pgrp = _C_int(sys.Pgid)
+		if pgrp == 0 {
+			r1, _, err1 = rawSyscall(abi.FuncPCABI0(libc_getpid_trampoline), 0, 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+			pgrp = _C_int(r1)
+		}
+
+		// Place process group in foreground.
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_ioctl_trampoline), uintptr(sys.Ctty), uintptr(TIOCSPGRP), uintptr(unsafe.Pointer(&pgrp)))
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - sys.Foreground")
+	}
+
+	// Restore the signal mask. We do this after TIOCSPGRP to avoid
+	// having the kernel send a SIGTTOU signal to the process group.
+	// runtime_AfterForkInChild()
+
+	// Chroot
+	if chroot != nil {
+		/* TODO(xsw):
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_chroot_trampoline), uintptr(unsafe.Pointer(chroot)), 0, 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - chroot")
+	}
+
+	// User and groups
+	if cred = sys.Credential; cred != nil {
+		/* TODO(xsw):
+		ngroups = uintptr(len(cred.Groups))
+		groups = uintptr(0)
+		if ngroups > 0 {
+			groups = uintptr(unsafe.Pointer(&cred.Groups[0]))
+		}
+		if !cred.NoSetGroups {
+			_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_setgroups_trampoline), ngroups, groups, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_setgid_trampoline), uintptr(cred.Gid), 0, 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_setuid_trampoline), uintptr(cred.Uid), 0, 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - sys.Credential")
+	}
+
+	// Chdir
+	if dir != nil {
+		/* TODO(xsw):
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_chdir_trampoline), uintptr(unsafe.Pointer(dir)), 0, 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - dir")
+	}
+
+	// Pass 1: look for fd[i] < i and move those up above len(fd)
+	// so that pass 2 won't stomp on an fd it needs later.
+	if pipe < nextfd {
+		/* TODO(xsw):
+		if runtime.GOOS == "openbsd" {
+			_, _, err1 = rawSyscall(dupTrampoline, uintptr(pipe), uintptr(nextfd), O_CLOEXEC)
+		} else {
+			_, _, err1 = rawSyscall(dupTrampoline, uintptr(pipe), uintptr(nextfd), 0)
+			if err1 != 0 {
+				goto childerror
+			}
+			_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_fcntl_trampoline), uintptr(nextfd), F_SETFD, FD_CLOEXEC)
+		}
+		if err1 != 0 {
+			goto childerror
+		}
+		pipe = nextfd
+		nextfd++
+		*/
+		panic("todo: syscall.forkAndExecInChild - pipe < nextfd")
+	}
+	for i = 0; i < len(fd); i++ {
+		if fd[i] >= 0 && fd[i] < i {
+			/* TODO(xsw):
+			if nextfd == pipe { // don't stomp on pipe
+				nextfd++
+			}
+			if runtime.GOOS == "openbsd" {
+				_, _, err1 = rawSyscall(dupTrampoline, uintptr(fd[i]), uintptr(nextfd), O_CLOEXEC)
+			} else {
+				_, _, err1 = rawSyscall(dupTrampoline, uintptr(fd[i]), uintptr(nextfd), 0)
+				if err1 != 0 {
+					goto childerror
+				}
+				_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_fcntl_trampoline), uintptr(nextfd), F_SETFD, FD_CLOEXEC)
+			}
+			if err1 != 0 {
+				goto childerror
+			}
+			fd[i] = nextfd
+			nextfd++
+			*/
+			panic("todo: syscall.forkAndExecInChild - for fd")
+		}
+	}
+
+	// Pass 2: dup fd[i] down onto i.
+	for i = 0; i < len(fd); i++ {
+		if fd[i] == -1 {
+			os.Close(c.Int(i))
+			continue
+		}
+		if fd[i] == i {
+			// dup2(i, i) won't clear close-on-exec flag on Linux,
+			// probably not elsewhere either.
+			if ret := os.Fcntl(c.Int(fd[i]), syscall.F_SETFD, 0); ret < 0 {
+				err1 = Errno(os.Errno())
+				goto childerror
+			}
+			continue
+		}
+		// The new fd is created NOT close-on-exec,
+		if ret := os.Dup2(c.Int(fd[i]), c.Int(i)); ret < 0 {
+			err1 = Errno(os.Errno())
+			goto childerror
+		}
+	}
+
+	// By convention, we don't close-on-exec the fds we are
+	// started with, so if len(fd) < 3, close 0, 1, 2 as needed.
+	// Programs that know they inherit fds >= 3 will need
+	// to set them close-on-exec.
+	for i = len(fd); i < 3; i++ {
+		os.Close(c.Int(i))
+	}
+
+	// Detach fd 0 from tty
+	if sys.Noctty {
+		/* TODO(xsw):
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_ioctl_trampoline), 0, uintptr(TIOCNOTTY), 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - sys.Noctty")
+	}
+
+	// Set the controlling TTY to Ctty
+	if sys.Setctty {
+		/* TODO(xsw):
+		_, _, err1 = rawSyscall(abi.FuncPCABI0(libc_ioctl_trampoline), uintptr(sys.Ctty), uintptr(TIOCSCTTY), 0)
+		if err1 != 0 {
+			goto childerror
+		}
+		*/
+		panic("todo: syscall.forkAndExecInChild - sys.Setctty")
+	}
+
+	// Restore original rlimit.
+	if rlimOK && rlim.Cur != 0 {
+		os.Setrlimit(syscall.RLIMIT_NOFILE, (*syscall.Rlimit)(&rlim))
+	}
+
+	// Time to exec.
+	os.Execve(argv0, argv, envv)
+	/* TODO(xsw):
+	ret := os.Execve(argv0, argv, envv)
+	if ret != 0 {
+		err1 = Errno(ret)
+	}
+	*/
+
+childerror:
+	// send error code on pipe
+	os.Write(c.Int(pipe), unsafe.Pointer(&err1), unsafe.Sizeof(err1))
+	for {
+		os.Exit(253)
+	}
+}

compiler/internal/lib/syscall/exec_linux.go

@@ -0,0 +1,735 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build linux
+
+package syscall
+
+import "github.com/goplus/llgo/c"
+
+// Linux unshare/clone/clone2/clone3 flags, architecture-independent,
+// copied from linux/sched.h.
+const (
+	CLONE_VM             = 0x00000100 // set if VM shared between processes
+	CLONE_FS             = 0x00000200 // set if fs info shared between processes
+	CLONE_FILES          = 0x00000400 // set if open files shared between processes
+	CLONE_SIGHAND        = 0x00000800 // set if signal handlers and blocked signals shared
+	CLONE_PIDFD          = 0x00001000 // set if a pidfd should be placed in parent
+	CLONE_PTRACE         = 0x00002000 // set if we want to let tracing continue on the child too
+	CLONE_VFORK          = 0x00004000 // set if the parent wants the child to wake it up on mm_release
+	CLONE_PARENT         = 0x00008000 // set if we want to have the same parent as the cloner
+	CLONE_THREAD         = 0x00010000 // Same thread group?
+	CLONE_NEWNS          = 0x00020000 // New mount namespace group
+	CLONE_SYSVSEM        = 0x00040000 // share system V SEM_UNDO semantics
+	CLONE_SETTLS         = 0x00080000 // create a new TLS for the child
+	CLONE_PARENT_SETTID  = 0x00100000 // set the TID in the parent
+	CLONE_CHILD_CLEARTID = 0x00200000 // clear the TID in the child
+	CLONE_DETACHED       = 0x00400000 // Unused, ignored
+	CLONE_UNTRACED       = 0x00800000 // set if the tracing process can't force CLONE_PTRACE on this clone
+	CLONE_CHILD_SETTID   = 0x01000000 // set the TID in the child
+	CLONE_NEWCGROUP      = 0x02000000 // New cgroup namespace
+	CLONE_NEWUTS         = 0x04000000 // New utsname namespace
+	CLONE_NEWIPC         = 0x08000000 // New ipc namespace
+	CLONE_NEWUSER        = 0x10000000 // New user namespace
+	CLONE_NEWPID         = 0x20000000 // New pid namespace
+	CLONE_NEWNET         = 0x40000000 // New network namespace
+	CLONE_IO             = 0x80000000 // Clone io context
+
+	// Flags for the clone3() syscall.
+
+	CLONE_CLEAR_SIGHAND = 0x100000000 // Clear any signal handler and reset to SIG_DFL.
+	CLONE_INTO_CGROUP   = 0x200000000 // Clone into a specific cgroup given the right permissions.
+
+	// Cloning flags intersect with CSIGNAL so can be used with unshare and clone3
+	// syscalls only:
+
+	CLONE_NEWTIME = 0x00000080 // New time namespace
+)
+
+// SysProcIDMap holds Container ID to Host ID mappings used for User Namespaces in Linux.
+// See user_namespaces(7).
+type SysProcIDMap struct {
+	ContainerID int // Container ID.
+	HostID      int // Host ID.
+	Size        int // Size.
+}
+
+type SysProcAttr struct {
+	Chroot     string      // Chroot.
+	Credential *Credential // Credential.
+	// Ptrace tells the child to call ptrace(PTRACE_TRACEME).
+	// Call runtime.LockOSThread before starting a process with this set,
+	// and don't call UnlockOSThread until done with PtraceSyscall calls.
+	Ptrace bool
+	Setsid bool // Create session.
+	// Setpgid sets the process group ID of the child to Pgid,
+	// or, if Pgid == 0, to the new child's process ID.
+	Setpgid bool
+	// Setctty sets the controlling terminal of the child to
+	// file descriptor Ctty. Ctty must be a descriptor number
+	// in the child process: an index into ProcAttr.Files.
+	// This is only meaningful if Setsid is true.
+	Setctty bool
+	Noctty  bool // Detach fd 0 from controlling terminal
+	Ctty    int  // Controlling TTY fd
+	// Foreground places the child process group in the foreground.
+	// This implies Setpgid. The Ctty field must be set to
+	// the descriptor of the controlling TTY.
+	// Unlike Setctty, in this case Ctty must be a descriptor
+	// number in the parent process.
+	Foreground bool
+	Pgid       int // Child's process group ID if Setpgid.
+	// Pdeathsig, if non-zero, is a signal that the kernel will send to
+	// the child process when the creating thread dies. Note that the signal
+	// is sent on thread termination, which may happen before process termination.
+	// There are more details at https://go.dev/issue/27505.
+	Pdeathsig    Signal
+	Cloneflags   uintptr        // Flags for clone calls (Linux only)
+	Unshareflags uintptr        // Flags for unshare calls (Linux only)
+	UidMappings  []SysProcIDMap // User ID mappings for user namespaces.
+	GidMappings  []SysProcIDMap // Group ID mappings for user namespaces.
+	// GidMappingsEnableSetgroups enabling setgroups syscall.
+	// If false, then setgroups syscall will be disabled for the child process.
+	// This parameter is no-op if GidMappings == nil. Otherwise for unprivileged
+	// users this should be set to false for mappings work.
+	GidMappingsEnableSetgroups bool
+	AmbientCaps                []uintptr // Ambient capabilities (Linux only)
+	UseCgroupFD                bool      // Whether to make use of the CgroupFD field.
+	CgroupFD                   int       // File descriptor of a cgroup to put the new process into.
+}
+
+var (
+	none  = [...]byte{'n', 'o', 'n', 'e', 0}
+	slash = [...]byte{'/', 0}
+)
+
+/* TODO(xsw):
+// Implemented in runtime package.
+func runtime_BeforeFork()
+func runtime_AfterFork()
+func runtime_AfterForkInChild()
+*/
+
+// Fork, dup fd onto 0..len(fd), and exec(argv0, argvv, envv) in child.
+// If a dup or exec fails, write the errno error to pipe.
+// (Pipe is close-on-exec so if exec succeeds, it will be closed.)
+// In the child, this function must not acquire any locks, because
+// they might have been locked at the time of the fork. This means
+// no rescheduling, no malloc calls, and no new stack segments.
+// For the same reason compiler does not race instrument it.
+// The calls to RawSyscall are okay because they are assembly
+// functions that do not grow the stack.
+//
+// func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
+func forkAndExecInChild(argv0 *c.Char, argv, envv **c.Char, chroot, dir *c.Char, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
+	/* TODO(xsw):
+	// Set up and fork. This returns immediately in the parent or
+	// if there's an error.
+	upid, err, mapPipe, locked := forkAndExecInChild1(argv0, argv, envv, chroot, dir, attr, sys, pipe)
+	if locked {
+		runtime_AfterFork()
+	}
+	if err != 0 {
+		return 0, err
+	}
+
+	// parent; return PID
+	pid = int(upid)
+
+	if sys.UidMappings != nil || sys.GidMappings != nil {
+		Close(mapPipe[0])
+		var err2 Errno
+		// uid/gid mappings will be written after fork and unshare(2) for user
+		// namespaces.
+		if sys.Unshareflags&CLONE_NEWUSER == 0 {
+			if err := writeUidGidMappings(pid, sys); err != nil {
+				err2 = err.(Errno)
+			}
+		}
+		RawSyscall(SYS_WRITE, uintptr(mapPipe[1]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
+		Close(mapPipe[1])
+	}
+
+	return pid, 0
+	*/
+	panic("todo: syscall.forkAndExecInChild")
+}
+
+const _LINUX_CAPABILITY_VERSION_3 = 0x20080522
+
+type capHeader struct {
+	version uint32
+	pid     int32
+}
+
+type capData struct {
+	effective   uint32
+	permitted   uint32
+	inheritable uint32
+}
+type caps struct {
+	hdr  capHeader
+	data [2]capData
+}
+
+// See CAP_TO_INDEX in linux/capability.h:
+func capToIndex(cap uintptr) uintptr { return cap >> 5 }
+
+// See CAP_TO_MASK in linux/capability.h:
+func capToMask(cap uintptr) uint32 { return 1 << uint(cap&31) }
+
+// cloneArgs holds arguments for clone3 Linux syscall.
+type cloneArgs struct {
+	flags      uint64 // Flags bit mask
+	pidFD      uint64 // Where to store PID file descriptor (int *)
+	childTID   uint64 // Where to store child TID, in child's memory (pid_t *)
+	parentTID  uint64 // Where to store child TID, in parent's memory (pid_t *)
+	exitSignal uint64 // Signal to deliver to parent on child termination
+	stack      uint64 // Pointer to lowest byte of stack
+	stackSize  uint64 // Size of stack
+	tls        uint64 // Location of new TLS
+	setTID     uint64 // Pointer to a pid_t array (since Linux 5.5)
+	setTIDSize uint64 // Number of elements in set_tid (since Linux 5.5)
+	cgroup     uint64 // File descriptor for target cgroup of child (since Linux 5.7)
+}
+
+// forkAndExecInChild1 implements the body of forkAndExecInChild up to
+// the parent's post-fork path. This is a separate function so we can
+// separate the child's and parent's stack frames if we're using
+// vfork.
+//
+// This is go:noinline because the point is to keep the stack frames
+// of this and forkAndExecInChild separate.
+//
+//go:noinline
+//go:norace
+//go:nocheckptr
+func forkAndExecInChild1(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid uintptr, err1 Errno, mapPipe [2]int, locked bool) {
+	/* TODO(xsw):
+		// Defined in linux/prctl.h starting with Linux 4.3.
+		const (
+			PR_CAP_AMBIENT       = 0x2f
+			PR_CAP_AMBIENT_RAISE = 0x2
+		)
+
+		// vfork requires that the child not touch any of the parent's
+		// active stack frames. Hence, the child does all post-fork
+		// processing in this stack frame and never returns, while the
+		// parent returns immediately from this frame and does all
+		// post-fork processing in the outer frame.
+		//
+		// Declare all variables at top in case any
+		// declarations require heap allocation (e.g., err2).
+		// ":=" should not be used to declare any variable after
+		// the call to runtime_BeforeFork.
+		//
+		// NOTE(bcmills): The allocation behavior described in the above comment
+		// seems to lack a corresponding test, and it may be rendered invalid
+		// by an otherwise-correct change in the compiler.
+		var (
+			err2                      Errno
+			nextfd                    int
+			i                         int
+			caps                      caps
+			fd1, flags                uintptr
+			puid, psetgroups, pgid    []byte
+			uidmap, setgroups, gidmap []byte
+			clone3                    *cloneArgs
+			pgrp                      int32
+			dirfd                     int
+			cred                      *Credential
+			ngroups, groups           uintptr
+			c                         uintptr
+		)
+
+		rlim, rlimOK := origRlimitNofile.Load().(Rlimit)
+
+		if sys.UidMappings != nil {
+			puid = []byte("/proc/self/uid_map\000")
+			uidmap = formatIDMappings(sys.UidMappings)
+		}
+
+		if sys.GidMappings != nil {
+			psetgroups = []byte("/proc/self/setgroups\000")
+			pgid = []byte("/proc/self/gid_map\000")
+
+			if sys.GidMappingsEnableSetgroups {
+				setgroups = []byte("allow\000")
+			} else {
+				setgroups = []byte("deny\000")
+			}
+			gidmap = formatIDMappings(sys.GidMappings)
+		}
+
+		// Record parent PID so child can test if it has died.
+		ppid, _ := rawSyscallNoError(SYS_GETPID, 0, 0, 0)
+
+		// Guard against side effects of shuffling fds below.
+		// Make sure that nextfd is beyond any currently open files so
+		// that we can't run the risk of overwriting any of them.
+		fd := make([]int, len(attr.Files))
+		nextfd = len(attr.Files)
+		for i, ufd := range attr.Files {
+			if nextfd < int(ufd) {
+				nextfd = int(ufd)
+			}
+			fd[i] = int(ufd)
+		}
+		nextfd++
+
+		// Allocate another pipe for parent to child communication for
+		// synchronizing writing of User ID/Group ID mappings.
+		if sys.UidMappings != nil || sys.GidMappings != nil {
+			if err := forkExecPipe(mapPipe[:]); err != nil {
+				err1 = err.(Errno)
+				return
+			}
+		}
+
+		flags = sys.Cloneflags
+		if sys.Cloneflags&CLONE_NEWUSER == 0 && sys.Unshareflags&CLONE_NEWUSER == 0 {
+			flags |= CLONE_VFORK | CLONE_VM
+		}
+		// Whether to use clone3.
+		if sys.UseCgroupFD {
+			clone3 = &cloneArgs{
+				flags:      uint64(flags) | CLONE_INTO_CGROUP,
+				exitSignal: uint64(SIGCHLD),
+				cgroup:     uint64(sys.CgroupFD),
+			}
+		} else if flags&CLONE_NEWTIME != 0 {
+			clone3 = &cloneArgs{
+				flags:      uint64(flags),
+				exitSignal: uint64(SIGCHLD),
+			}
+		}
+
+		// About to call fork.
+		// No more allocation or calls of non-assembly functions.
+		runtime_BeforeFork()
+		locked = true
+		if clone3 != nil {
+			pid, err1 = rawVforkSyscall(_SYS_clone3, uintptr(unsafe.Pointer(clone3)), unsafe.Sizeof(*clone3))
+		} else {
+			flags |= uintptr(SIGCHLD)
+			if runtime.GOARCH == "s390x" {
+				// On Linux/s390, the first two arguments of clone(2) are swapped.
+				pid, err1 = rawVforkSyscall(SYS_CLONE, 0, flags)
+			} else {
+				pid, err1 = rawVforkSyscall(SYS_CLONE, flags, 0)
+			}
+		}
+		if err1 != 0 || pid != 0 {
+			// If we're in the parent, we must return immediately
+			// so we're not in the same stack frame as the child.
+			// This can at most use the return PC, which the child
+			// will not modify, and the results of
+			// rawVforkSyscall, which must have been written after
+			// the child was replaced.
+			return
+		}
+
+		// Fork succeeded, now in child.
+
+		// Enable the "keep capabilities" flag to set ambient capabilities later.
+		if len(sys.AmbientCaps) > 0 {
+			_, _, err1 = RawSyscall6(SYS_PRCTL, PR_SET_KEEPCAPS, 1, 0, 0, 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// Wait for User ID/Group ID mappings to be written.
+		if sys.UidMappings != nil || sys.GidMappings != nil {
+			if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(mapPipe[1]), 0, 0); err1 != 0 {
+				goto childerror
+			}
+			pid, _, err1 = RawSyscall(SYS_READ, uintptr(mapPipe[0]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
+			if err1 != 0 {
+				goto childerror
+			}
+			if pid != unsafe.Sizeof(err2) {
+				err1 = EINVAL
+				goto childerror
+			}
+			if err2 != 0 {
+				err1 = err2
+				goto childerror
+			}
+		}
+
+		// Session ID
+		if sys.Setsid {
+			_, _, err1 = RawSyscall(SYS_SETSID, 0, 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// Set process group
+		if sys.Setpgid || sys.Foreground {
+			// Place child in process group.
+			_, _, err1 = RawSyscall(SYS_SETPGID, 0, uintptr(sys.Pgid), 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		if sys.Foreground {
+			pgrp = int32(sys.Pgid)
+			if pgrp == 0 {
+				pid, _ = rawSyscallNoError(SYS_GETPID, 0, 0, 0)
+
+				pgrp = int32(pid)
+			}
+
+			// Place process group in foreground.
+			_, _, err1 = RawSyscall(SYS_IOCTL, uintptr(sys.Ctty), uintptr(TIOCSPGRP), uintptr(unsafe.Pointer(&pgrp)))
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// Restore the signal mask. We do this after TIOCSPGRP to avoid
+		// having the kernel send a SIGTTOU signal to the process group.
+		runtime_AfterForkInChild()
+
+		// Unshare
+		if sys.Unshareflags != 0 {
+			_, _, err1 = RawSyscall(SYS_UNSHARE, sys.Unshareflags, 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+
+			if sys.Unshareflags&CLONE_NEWUSER != 0 && sys.GidMappings != nil {
+				dirfd = int(_AT_FDCWD)
+				if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&psetgroups[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
+					goto childerror
+				}
+				pid, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&setgroups[0])), uintptr(len(setgroups)))
+				if err1 != 0 {
+					goto childerror
+				}
+				if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
+					goto childerror
+				}
+
+				if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&pgid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
+					goto childerror
+				}
+				pid, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&gidmap[0])), uintptr(len(gidmap)))
+				if err1 != 0 {
+					goto childerror
+				}
+				if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
+					goto childerror
+				}
+			}
+
+			if sys.Unshareflags&CLONE_NEWUSER != 0 && sys.UidMappings != nil {
+				dirfd = int(_AT_FDCWD)
+				if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&puid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
+					goto childerror
+				}
+				pid, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&uidmap[0])), uintptr(len(uidmap)))
+				if err1 != 0 {
+					goto childerror
+				}
+				if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
+					goto childerror
+				}
+			}
+
+			// The unshare system call in Linux doesn't unshare mount points
+			// mounted with --shared. Systemd mounts / with --shared. For a
+			// long discussion of the pros and cons of this see debian bug 739593.
+			// The Go model of unsharing is more like Plan 9, where you ask
+			// to unshare and the namespaces are unconditionally unshared.
+			// To make this model work we must further mark / as MS_PRIVATE.
+			// This is what the standard unshare command does.
+			if sys.Unshareflags&CLONE_NEWNS == CLONE_NEWNS {
+				_, _, err1 = RawSyscall6(SYS_MOUNT, uintptr(unsafe.Pointer(&none[0])), uintptr(unsafe.Pointer(&slash[0])), 0, MS_REC|MS_PRIVATE, 0, 0)
+				if err1 != 0 {
+					goto childerror
+				}
+			}
+		}
+
+		// Chroot
+		if chroot != nil {
+			_, _, err1 = RawSyscall(SYS_CHROOT, uintptr(unsafe.Pointer(chroot)), 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// User and groups
+		if cred = sys.Credential; cred != nil {
+			ngroups = uintptr(len(cred.Groups))
+			groups = uintptr(0)
+			if ngroups > 0 {
+				groups = uintptr(unsafe.Pointer(&cred.Groups[0]))
+			}
+			if !(sys.GidMappings != nil && !sys.GidMappingsEnableSetgroups && ngroups == 0) && !cred.NoSetGroups {
+				_, _, err1 = RawSyscall(_SYS_setgroups, ngroups, groups, 0)
+				if err1 != 0 {
+					goto childerror
+				}
+			}
+			_, _, err1 = RawSyscall(sys_SETGID, uintptr(cred.Gid), 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+			_, _, err1 = RawSyscall(sys_SETUID, uintptr(cred.Uid), 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		if len(sys.AmbientCaps) != 0 {
+			// Ambient capabilities were added in the 4.3 kernel,
+			// so it is safe to always use _LINUX_CAPABILITY_VERSION_3.
+			caps.hdr.version = _LINUX_CAPABILITY_VERSION_3
+
+			if _, _, err1 = RawSyscall(SYS_CAPGET, uintptr(unsafe.Pointer(&caps.hdr)), uintptr(unsafe.Pointer(&caps.data[0])), 0); err1 != 0 {
+				goto childerror
+			}
+
+			for _, c = range sys.AmbientCaps {
+				// Add the c capability to the permitted and inheritable capability mask,
+				// otherwise we will not be able to add it to the ambient capability mask.
+				caps.data[capToIndex(c)].permitted |= capToMask(c)
+				caps.data[capToIndex(c)].inheritable |= capToMask(c)
+			}
+
+			if _, _, err1 = RawSyscall(SYS_CAPSET, uintptr(unsafe.Pointer(&caps.hdr)), uintptr(unsafe.Pointer(&caps.data[0])), 0); err1 != 0 {
+				goto childerror
+			}
+
+			for _, c = range sys.AmbientCaps {
+				_, _, err1 = RawSyscall6(SYS_PRCTL, PR_CAP_AMBIENT, uintptr(PR_CAP_AMBIENT_RAISE), c, 0, 0, 0)
+				if err1 != 0 {
+					goto childerror
+				}
+			}
+		}
+
+		// Chdir
+		if dir != nil {
+			_, _, err1 = RawSyscall(SYS_CHDIR, uintptr(unsafe.Pointer(dir)), 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// Parent death signal
+		if sys.Pdeathsig != 0 {
+			_, _, err1 = RawSyscall6(SYS_PRCTL, PR_SET_PDEATHSIG, uintptr(sys.Pdeathsig), 0, 0, 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+
+			// Signal self if parent is already dead. This might cause a
+			// duplicate signal in rare cases, but it won't matter when
+			// using SIGKILL.
+			pid, _ = rawSyscallNoError(SYS_GETPPID, 0, 0, 0)
+			if pid != ppid {
+				pid, _ = rawSyscallNoError(SYS_GETPID, 0, 0, 0)
+				_, _, err1 = RawSyscall(SYS_KILL, pid, uintptr(sys.Pdeathsig), 0)
+				if err1 != 0 {
+					goto childerror
+				}
+			}
+		}
+
+		// Pass 1: look for fd[i] < i and move those up above len(fd)
+		// so that pass 2 won't stomp on an fd it needs later.
+		if pipe < nextfd {
+			_, _, err1 = RawSyscall(SYS_DUP3, uintptr(pipe), uintptr(nextfd), O_CLOEXEC)
+			if err1 != 0 {
+				goto childerror
+			}
+			pipe = nextfd
+			nextfd++
+		}
+		for i = 0; i < len(fd); i++ {
+			if fd[i] >= 0 && fd[i] < i {
+				if nextfd == pipe { // don't stomp on pipe
+					nextfd++
+				}
+				_, _, err1 = RawSyscall(SYS_DUP3, uintptr(fd[i]), uintptr(nextfd), O_CLOEXEC)
+				if err1 != 0 {
+					goto childerror
+				}
+				fd[i] = nextfd
+				nextfd++
+			}
+		}
+
+		// Pass 2: dup fd[i] down onto i.
+		for i = 0; i < len(fd); i++ {
+			if fd[i] == -1 {
+				RawSyscall(SYS_CLOSE, uintptr(i), 0, 0)
+				continue
+			}
+			if fd[i] == i {
+				// dup2(i, i) won't clear close-on-exec flag on Linux,
+				// probably not elsewhere either.
+				_, _, err1 = RawSyscall(fcntl64Syscall, uintptr(fd[i]), F_SETFD, 0)
+				if err1 != 0 {
+					goto childerror
+				}
+				continue
+			}
+			// The new fd is created NOT close-on-exec,
+			// which is exactly what we want.
+			_, _, err1 = RawSyscall(SYS_DUP3, uintptr(fd[i]), uintptr(i), 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// By convention, we don't close-on-exec the fds we are
+		// started with, so if len(fd) < 3, close 0, 1, 2 as needed.
+		// Programs that know they inherit fds >= 3 will need
+		// to set them close-on-exec.
+		for i = len(fd); i < 3; i++ {
+			RawSyscall(SYS_CLOSE, uintptr(i), 0, 0)
+		}
+
+		// Detach fd 0 from tty
+		if sys.Noctty {
+			_, _, err1 = RawSyscall(SYS_IOCTL, 0, uintptr(TIOCNOTTY), 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// Set the controlling TTY to Ctty
+		if sys.Setctty {
+			_, _, err1 = RawSyscall(SYS_IOCTL, uintptr(sys.Ctty), uintptr(TIOCSCTTY), 1)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// Restore original rlimit.
+		if rlimOK && rlim.Cur != 0 {
+			rawSetrlimit(RLIMIT_NOFILE, &rlim)
+		}
+
+		// Enable tracing if requested.
+		// Do this right before exec so that we don't unnecessarily trace the runtime
+		// setting up after the fork. See issue #21428.
+		if sys.Ptrace {
+			_, _, err1 = RawSyscall(SYS_PTRACE, uintptr(PTRACE_TRACEME), 0, 0)
+			if err1 != 0 {
+				goto childerror
+			}
+		}
+
+		// Time to exec.
+		_, _, err1 = RawSyscall(SYS_EXECVE,
+			uintptr(unsafe.Pointer(argv0)),
+			uintptr(unsafe.Pointer(&argv[0])),
+			uintptr(unsafe.Pointer(&envv[0])))
+
+	childerror:
+		// send error code on pipe
+		RawSyscall(SYS_WRITE, uintptr(pipe), uintptr(unsafe.Pointer(&err1)), unsafe.Sizeof(err1))
+		for {
+			RawSyscall(SYS_EXIT, 253, 0, 0)
+		}
+	*/
+	panic("todo: syscall.forkAndExecInChild1")
+}
+
+func formatIDMappings(idMap []SysProcIDMap) []byte {
+	/* TODO(xsw):
+	var data []byte
+	for _, im := range idMap {
+		data = append(data, itoa.Itoa(im.ContainerID)+" "+itoa.Itoa(im.HostID)+" "+itoa.Itoa(im.Size)+"\n"...)
+	}
+	return data
+	*/
+	panic("todo: syscall.formatIDMappings")
+}
+
+// writeIDMappings writes the user namespace User ID or Group ID mappings to the specified path.
+func writeIDMappings(path string, idMap []SysProcIDMap) error {
+	/* TODO(xsw):
+	fd, err := Open(path, O_RDWR, 0)
+	if err != nil {
+		return err
+	}
+
+	if _, err := Write(fd, formatIDMappings(idMap)); err != nil {
+		Close(fd)
+		return err
+	}
+
+	if err := Close(fd); err != nil {
+		return err
+	}
+
+	return nil
+	*/
+	panic("todo: syscall.writeIDMappings")
+}
+
+// writeSetgroups writes to /proc/PID/setgroups "deny" if enable is false
+// and "allow" if enable is true.
+// This is needed since kernel 3.19, because you can't write gid_map without
+// disabling setgroups() system call.
+func writeSetgroups(pid int, enable bool) error {
+	/* TODO(xsw):
+	sgf := "/proc/" + itoa.Itoa(pid) + "/setgroups"
+	fd, err := Open(sgf, O_RDWR, 0)
+	if err != nil {
+		return err
+	}
+
+	var data []byte
+	if enable {
+		data = []byte("allow")
+	} else {
+		data = []byte("deny")
+	}
+
+	if _, err := Write(fd, data); err != nil {
+		Close(fd)
+		return err
+	}
+
+	return Close(fd)
+	*/
+	panic("todo: syscall.writeSetgroups")
+}
+
+// writeUidGidMappings writes User ID and Group ID mappings for user namespaces
+// for a process and it is called from the parent process.
+func writeUidGidMappings(pid int, sys *SysProcAttr) error {
+	/* TODO(xsw):
+	if sys.UidMappings != nil {
+		uidf := "/proc/" + itoa.Itoa(pid) + "/uid_map"
+		if err := writeIDMappings(uidf, sys.UidMappings); err != nil {
+			return err
+		}
+	}
+
+	if sys.GidMappings != nil {
+		// If the kernel is too old to support /proc/PID/setgroups, writeSetGroups will return ENOENT; this is OK.
+		if err := writeSetgroups(pid, sys.GidMappingsEnableSetgroups); err != nil && err != ENOENT {
+			return err
+		}
+		gidf := "/proc/" + itoa.Itoa(pid) + "/gid_map"
+		if err := writeIDMappings(gidf, sys.GidMappings); err != nil {
+			return err
+		}
+	}
+
+	return nil
+	*/
+	panic("todo: syscall.writeUidGidMappings")
+}

compiler/internal/lib/syscall/exec_unix.go

@@ -0,0 +1,231 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build unix
+
+// Fork, exec, wait, etc.
+
+package syscall
+
+import (
+	"errors"
+	"runtime"
+	"sync"
+	"unsafe"
+
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/os"
+	"github.com/goplus/llgo/c/syscall"
+)
+
+// ForkLock is used to synchronize creation of new file descriptors
+// with fork.
+//
+// We want the child in a fork/exec sequence to inherit only the
+// file descriptors we intend. To do that, we mark all file
+// descriptors close-on-exec and then, in the child, explicitly
+// unmark the ones we want the exec'ed program to keep.
+// Unix doesn't make this easy: there is, in general, no way to
+// allocate a new file descriptor close-on-exec. Instead you
+// have to allocate the descriptor and then mark it close-on-exec.
+// If a fork happens between those two events, the child's exec
+// will inherit an unwanted file descriptor.
+//
+// This lock solves that race: the create new fd/mark close-on-exec
+// operation is done holding ForkLock for reading, and the fork itself
+// is done holding ForkLock for writing. At least, that's the idea.
+// There are some complications.
+//
+// Some system calls that create new file descriptors can block
+// for arbitrarily long times: open on a hung NFS server or named
+// pipe, accept on a socket, and so on. We can't reasonably grab
+// the lock across those operations.
+//
+// It is worse to inherit some file descriptors than others.
+// If a non-malicious child accidentally inherits an open ordinary file,
+// that's not a big deal. On the other hand, if a long-lived child
+// accidentally inherits the write end of a pipe, then the reader
+// of that pipe will not see EOF until that child exits, potentially
+// causing the parent program to hang. This is a common problem
+// in threaded C programs that use popen.
+//
+// Luckily, the file descriptors that are most important not to
+// inherit are not the ones that can take an arbitrarily long time
+// to create: pipe returns instantly, and the net package uses
+// non-blocking I/O to accept on a listening socket.
+// The rules for which file descriptor-creating operations use the
+// ForkLock are as follows:
+//
+//   - Pipe. Use pipe2 if available. Otherwise, does not block,
+//     so use ForkLock.
+//   - Socket. Use SOCK_CLOEXEC if available. Otherwise, does not
+//     block, so use ForkLock.
+//   - Open. Use O_CLOEXEC if available. Otherwise, may block,
+//     so live with the race.
+//   - Dup. Use F_DUPFD_CLOEXEC or dup3 if available. Otherwise,
+//     does not block, so use ForkLock.
+var ForkLock sync.RWMutex
+
+func CloseOnExec(fd int) {
+	os.Fcntl(c.Int(fd), syscall.F_SETFD, syscall.FD_CLOEXEC)
+}
+
+func SetNonblock(fd int, nonblocking bool) (err error) {
+	/* TODO(xsw):
+	flag, err := fcntl(fd, F_GETFL, 0)
+	if err != nil {
+		return err
+	}
+	if nonblocking {
+		flag |= O_NONBLOCK
+	} else {
+		flag &^= O_NONBLOCK
+	}
+	_, err = fcntl(fd, F_SETFL, flag)
+	return err
+	*/
+	panic("todo: syscall.SetNonblock")
+}
+
+// Credential holds user and group identities to be assumed
+// by a child process started by StartProcess.
+type Credential struct {
+	Uid         uint32   // User ID.
+	Gid         uint32   // Group ID.
+	Groups      []uint32 // Supplementary group IDs.
+	NoSetGroups bool     // If true, don't set supplementary groups
+}
+
+// ProcAttr holds attributes that will be applied to a new process started
+// by StartProcess.
+type ProcAttr struct {
+	Dir   string    // Current working directory.
+	Env   []string  // Environment.
+	Files []uintptr // File descriptors.
+	Sys   *SysProcAttr
+}
+
+var zeroProcAttr ProcAttr
+var zeroSysProcAttr SysProcAttr
+
+func forkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error) {
+	var p [2]int
+	var n int
+	var err1 Errno
+	var wstatus WaitStatus
+
+	if attr == nil {
+		attr = &zeroProcAttr
+	}
+	sys := attr.Sys
+	if sys == nil {
+		sys = &zeroSysProcAttr
+	}
+
+	// Convert args to C form.
+	argv0p := c.AllocaCStr(argv0)
+	argvp := c.AllocaCStrs(argv, true)
+	envvp := c.AllocaCStrs(attr.Env, true)
+
+	if (runtime.GOOS == "freebsd" || runtime.GOOS == "dragonfly") && len(argv) > 0 && len(argv[0]) > len(argv0) {
+		*argvp = argv0p
+	}
+
+	var chroot *c.Char
+	if sys.Chroot != "" {
+		chroot = c.AllocaCStr(sys.Chroot)
+	}
+	var dir *c.Char
+	if attr.Dir != "" {
+		dir = c.AllocaCStr(attr.Dir)
+	}
+
+	// Both Setctty and Foreground use the Ctty field,
+	// but they give it slightly different meanings.
+	if sys.Setctty && sys.Foreground {
+		return 0, errors.New("both Setctty and Foreground set in SysProcAttr")
+	}
+	if sys.Setctty && sys.Ctty >= len(attr.Files) {
+		return 0, errors.New("Setctty set but Ctty not valid in child")
+	}
+
+	acquireForkLock()
+
+	// Allocate child status pipe close on exec.
+	if err = forkExecPipe(p[:]); err != nil {
+		releaseForkLock()
+		return 0, err
+	}
+
+	// Kick off child.
+	pid, err1 = forkAndExecInChild(argv0p, argvp, envvp, chroot, dir, attr, sys, p[1])
+	if err1 != 0 {
+		Close(p[0])
+		Close(p[1])
+		releaseForkLock()
+		return 0, Errno(err1)
+	}
+	releaseForkLock()
+
+	// Read child error status from pipe.
+	Close(p[1])
+	for {
+		n, err = readlen(p[0], (*byte)(unsafe.Pointer(&err1)), int(unsafe.Sizeof(err1)))
+		if err != Errno(syscall.EINTR) {
+			break
+		}
+	}
+	Close(p[0])
+	if err != nil || n != 0 {
+		if n == int(unsafe.Sizeof(err1)) {
+			err = Errno(err1)
+		}
+		if err == nil {
+			err = Errno(syscall.EPIPE)
+		}
+
+		// Child failed; wait for it to exit, to make sure
+		// the zombies don't accumulate.
+		_, err1 := Wait4(pid, &wstatus, 0, nil)
+		for err1 == Errno(syscall.EINTR) {
+			_, err1 = Wait4(pid, &wstatus, 0, nil)
+		}
+		return 0, err
+	}
+
+	// Read got EOF, so pipe closed on exec, so exec succeeded.
+	return pid, nil
+}
+
+// Combination of fork and exec, careful to be thread safe.
+func ForkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error) {
+	return forkExec(argv0, argv, attr)
+}
+
+// StartProcess wraps ForkExec for package os.
+func StartProcess(argv0 string, argv []string, attr *ProcAttr) (pid int, handle uintptr, err error) {
+	pid, err = forkExec(argv0, argv, attr)
+	return pid, 0, err
+}
+
+/* TODO(xsw):
+// Implemented in runtime package.
+func runtime_BeforeExec()
+func runtime_AfterExec()
+
+// execveLibc is non-nil on OS using libc syscall, set to execve in exec_libc.go; this
+// avoids a build dependency for other platforms.
+var execveLibc func(path uintptr, argv uintptr, envp uintptr) Errno
+var execveDarwin func(path *byte, argv **byte, envp **byte) error
+var execveOpenBSD func(path *byte, argv **byte, envp **byte) error
+*/
+
+// Exec invokes the execve(2) system call.
+func Exec(argv0 string, argv []string, envv []string) (err error) {
+	ret := os.Execve(c.AllocaCStr(argv0), c.AllocaCStrs(argv, true), c.AllocaCStrs(envv, true))
+	if ret == 0 {
+		return nil
+	}
+	return Errno(ret)
+}

compiler/internal/lib/syscall/forkpipe.go

@@ -0,0 +1,39 @@
+// Copyright 2011 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build aix || darwin
+
+package syscall
+
+import (
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/os"
+	"github.com/goplus/llgo/c/syscall"
+)
+
+// forkExecPipe opens a pipe and non-atomically sets O_CLOEXEC on both file
+// descriptors.
+func forkExecPipe(p []int) error {
+	err := Pipe(p)
+	if err != nil {
+		return err
+	}
+	ret := os.Fcntl(c.Int(p[0]), syscall.F_SETFD, syscall.FD_CLOEXEC)
+	if ret != 0 {
+		return Errno(ret)
+	}
+	ret = os.Fcntl(c.Int(p[1]), syscall.F_SETFD, syscall.FD_CLOEXEC)
+	if ret != 0 {
+		return Errno(ret)
+	}
+	return nil
+}
+
+func acquireForkLock() {
+	ForkLock.Lock()
+}
+
+func releaseForkLock() {
+	ForkLock.Unlock()
+}

compiler/internal/lib/syscall/forkpipe2.go

@@ -0,0 +1,105 @@
+// Copyright 2017 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build dragonfly || freebsd || linux || netbsd || openbsd || solaris
+
+package syscall
+
+import (
+	"sync"
+
+	"github.com/goplus/llgo/c/syscall"
+)
+
+// forkExecPipe atomically opens a pipe with O_CLOEXEC set on both file
+// descriptors.
+func forkExecPipe(p []int) error {
+	return Pipe2(p, syscall.O_CLOEXEC)
+}
+
+var (
+	// Guard the forking variable.
+	forkingLock sync.Mutex
+	// Number of goroutines currently forking, and thus the
+	// number of goroutines holding a conceptual write lock
+	// on ForkLock.
+	forking int
+)
+
+// TODO(xsw):
+// hasWaitingReaders reports whether any goroutine is waiting
+// to acquire a read lock on rw. It is defined in the sync package.
+func hasWaitingReaders(rw *sync.RWMutex) bool {
+	panic("todo: syscall.hasWaitingReaders in sync package")
+}
+
+// acquireForkLock acquires a write lock on ForkLock.
+// ForkLock is exported and we've promised that during a fork
+// we will call ForkLock.Lock, so that no other threads create
+// new fds that are not yet close-on-exec before we fork.
+// But that forces all fork calls to be serialized, which is bad.
+// But we haven't promised that serialization, and it is essentially
+// undetectable by other users of ForkLock, which is good.
+// Avoid the serialization by ensuring that ForkLock is locked
+// at the first fork and unlocked when there are no more forks.
+func acquireForkLock() {
+	forkingLock.Lock()
+	defer forkingLock.Unlock()
+
+	if forking == 0 {
+		// There is no current write lock on ForkLock.
+		ForkLock.Lock()
+		forking++
+		return
+	}
+
+	// ForkLock is currently locked for writing.
+
+	if hasWaitingReaders(&ForkLock) {
+		// ForkLock is locked for writing, and at least one
+		// goroutine is waiting to read from it.
+		// To avoid lock starvation, allow readers to proceed.
+		// The simple way to do this is for us to acquire a
+		// read lock. That will block us until all current
+		// conceptual write locks are released.
+		//
+		// Note that this case is unusual on modern systems
+		// with O_CLOEXEC and SOCK_CLOEXEC. On those systems
+		// the standard library should never take a read
+		// lock on ForkLock.
+
+		forkingLock.Unlock()
+
+		ForkLock.RLock()
+		ForkLock.RUnlock()
+
+		forkingLock.Lock()
+
+		// Readers got a chance, so now take the write lock.
+
+		if forking == 0 {
+			ForkLock.Lock()
+		}
+	}
+
+	forking++
+}
+
+// releaseForkLock releases the conceptual write lock on ForkLock
+// acquired by acquireForkLock.
+func releaseForkLock() {
+	forkingLock.Lock()
+	defer forkingLock.Unlock()
+
+	if forking <= 0 {
+		panic("syscall.releaseForkLock: negative count")
+	}
+
+	forking--
+
+	if forking == 0 {
+		// No more conceptual write locks.
+		ForkLock.Unlock()
+	}
+}

compiler/internal/lib/syscall/rlimit.go

@@ -0,0 +1,52 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build unix
+
+package syscall
+
+import (
+	"sync/atomic"
+
+	"github.com/goplus/llgo/c/syscall"
+)
+
+// origRlimitNofile, if not {0, 0}, is the original soft RLIMIT_NOFILE.
+// When we can assume that we are bootstrapping with Go 1.19,
+// this can be atomic.Pointer[Rlimit].
+var origRlimitNofile atomic.Value // of Rlimit
+
+// Some systems set an artificially low soft limit on open file count, for compatibility
+// with code that uses select and its hard-coded maximum file descriptor
+// (limited by the size of fd_set).
+//
+// Go does not use select, so it should not be subject to these limits.
+// On some systems the limit is 256, which is very easy to run into,
+// even in simple programs like gofmt when they parallelize walking
+// a file tree.
+//
+// After a long discussion on go.dev/issue/46279, we decided the
+// best approach was for Go to raise the limit unconditionally for itself,
+// and then leave old software to set the limit back as needed.
+// Code that really wants Go to leave the limit alone can set the hard limit,
+// which Go of course has no choice but to respect.
+func init() {
+	var lim Rlimit
+	if err := Getrlimit(syscall.RLIMIT_NOFILE, &lim); err == nil && lim.Cur != lim.Max {
+		origRlimitNofile.Store(lim)
+		lim.Cur = lim.Max
+		adjustFileLimit(&lim)
+		setrlimit(syscall.RLIMIT_NOFILE, &lim)
+	}
+}
+
+func Setrlimit(resource int, rlim *Rlimit) error {
+	err := setrlimit(resource, rlim)
+	if err == nil && resource == syscall.RLIMIT_NOFILE {
+		// Store zeroes in origRlimitNofile to tell StartProcess
+		// to not adjust the rlimit in the child process.
+		origRlimitNofile.Store(Rlimit{0, 0})
+	}
+	return err
+}

compiler/internal/lib/syscall/rlimit_darwin.go

@@ -0,0 +1,20 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build darwin
+
+package syscall
+
+// adjustFileLimit adds per-OS limitations on the Rlimit used for RLIMIT_NOFILE. See rlimit.go.
+func adjustFileLimit(lim *Rlimit) {
+	// On older macOS, setrlimit(RLIMIT_NOFILE, lim) with lim.Cur = infinity fails.
+	// Set to the value of kern.maxfilesperproc instead.
+	n, err := SysctlUint32("kern.maxfilesperproc")
+	if err != nil {
+		return
+	}
+	if lim.Cur > uint64(n) {
+		lim.Cur = uint64(n)
+	}
+}

compiler/internal/lib/syscall/rlimit_stub.go

@@ -0,0 +1,10 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build aix || dragonfly || freebsd || linux || netbsd || openbsd || solaris
+
+package syscall
+
+// adjustFileLimit adds per-OS limitations on the Rlimit used for RLIMIT_NOFILE. See rlimit.go.
+func adjustFileLimit(lim *Rlimit) {}

compiler/internal/lib/syscall/syscall.go

@@ -0,0 +1,192 @@
+/*
+ * Copyright (c) 2024 The GoPlus Authors (goplus.org). All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package syscall
+
+// llgo:skipall
+import (
+	"unsafe"
+
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/os"
+	"github.com/goplus/llgo/c/syscall"
+)
+
+type Timespec syscall.Timespec
+type Timeval syscall.Timeval
+
+// Unix returns the time stored in ts as seconds plus nanoseconds.
+func (ts *Timespec) Unix() (sec int64, nsec int64) {
+	return int64(ts.Sec), int64(ts.Nsec)
+}
+
+// Unix returns the time stored in tv as seconds plus nanoseconds.
+func (tv *Timeval) Unix() (sec int64, nsec int64) {
+	return int64(tv.Sec), int64(tv.Usec) * 1000
+}
+
+// Nano returns the time stored in ts as nanoseconds.
+func (ts *Timespec) Nano() int64 {
+	return int64(ts.Sec)*1e9 + int64(ts.Nsec)
+}
+
+// Nano returns the time stored in tv as nanoseconds.
+func (tv *Timeval) Nano() int64 {
+	return int64(tv.Sec)*1e9 + int64(tv.Usec)*1000
+}
+
+func Getpagesize() int {
+	panic("todo: syscall.Getpagesize")
+}
+
+func Exit(code int) {
+	os.Exit(c.Int(code))
+}
+
+func Getcwd(buf []byte) (n int, err error) {
+	ptr := unsafe.Pointer(unsafe.SliceData(buf))
+	ret := os.Getcwd(ptr, uintptr(len(buf)))
+	if ret != nil {
+		return int(c.Strlen(ret)), nil
+	}
+	return 0, Errno(os.Errno())
+}
+
+func Getwd() (string, error) {
+	wd := os.Getcwd(c.Alloca(os.PATH_MAX), os.PATH_MAX)
+	if wd != nil {
+		return c.GoString(wd), nil
+	}
+	return "", Errno(os.Errno())
+}
+
+func Getpid() (pid int) {
+	return int(os.Getpid())
+}
+
+func Kill(pid int, signum Signal) (err error) {
+	ret := os.Kill(os.PidT(pid), c.Int(signum))
+	if ret == 0 {
+		return nil
+	}
+	return Errno(ret)
+}
+
+func fork() (uintptr, Errno) {
+	ret := os.Fork()
+	if ret >= 0 {
+		return uintptr(ret), Errno(0)
+	}
+	return 0, Errno(os.Errno())
+}
+
+func wait4(pid int, wstatus *c.Int, options int, rusage *syscall.Rusage) (wpid int, err error) {
+	ret := os.Wait4(os.PidT(pid), wstatus, c.Int(options), rusage)
+	if ret >= 0 {
+		return int(ret), nil
+	}
+	return 0, Errno(os.Errno())
+}
+
+func Open(path string, mode int, perm uint32) (fd int, err error) {
+	ret := os.Open(c.AllocaCStr(path), c.Int(mode), os.ModeT(perm))
+	if ret >= 0 {
+		return int(ret), nil
+	}
+	return 0, Errno(os.Errno())
+}
+
+func Seek(fd int, offset int64, whence int) (newoffset int64, err error) {
+	ret := os.Lseek(c.Int(fd), os.OffT(offset), c.Int(whence))
+	if ret >= 0 {
+		return int64(ret), nil
+	}
+	return -1, Errno(os.Errno())
+}
+
+func Read(fd int, p []byte) (n int, err error) {
+	ret := os.Read(c.Int(fd), unsafe.Pointer(unsafe.SliceData(p)), uintptr(len(p)))
+	if ret >= 0 {
+		return ret, nil // TODO(xsw): confirm err == nil (not io.EOF) when ret == 0
+	}
+	return 0, Errno(os.Errno())
+}
+
+func readlen(fd int, buf *byte, nbuf int) (n int, err error) {
+	ret := os.Read(c.Int(fd), unsafe.Pointer(buf), uintptr(nbuf))
+	if ret >= 0 {
+		return ret, nil // TODO(xsw): confirm err == nil (not io.EOF) when ret == 0
+	}
+	return 0, Errno(os.Errno())
+}
+
+func Close(fd int) (err error) {
+	ret := os.Close(c.Int(fd))
+	if ret == 0 {
+		return nil
+	}
+	return Errno(ret)
+}
+
+type Stat_t = syscall.Stat_t
+
+func Lstat(path string, stat *Stat_t) (err error) {
+	ret := os.Lstat(c.AllocaCStr(path), stat)
+	if ret == 0 {
+		return nil
+	}
+	return Errno(os.Errno())
+}
+
+func Stat(path string, stat *Stat_t) (err error) {
+	ret := os.Stat(c.AllocaCStr(path), stat)
+	if ret == 0 {
+		return nil
+	}
+	return Errno(os.Errno())
+}
+
+func Pipe(p []int) (err error) {
+	if len(p) != 2 {
+		return Errno(syscall.EINVAL)
+	}
+	var q [2]c.Int
+	ret := os.Pipe(&q)
+	if ret == 0 {
+		p[0] = int(q[0])
+		p[1] = int(q[1])
+		return nil
+	}
+	return Errno(ret)
+}
+
+type Rlimit syscall.Rlimit
+
+func Getrlimit(which int, lim *Rlimit) (err error) {
+	ret := os.Getrlimit(c.Int(which), (*syscall.Rlimit)(lim))
+	if ret == 0 {
+		return nil
+	}
+	return Errno(ret)
+}
+
+func setrlimit(which int, lim *Rlimit) (err error) {
+	ret := os.Setrlimit(c.Int(which), (*syscall.Rlimit)(lim))
+	if ret == 0 {
+		return nil
+	}
+	return Errno(ret)
+}

compiler/internal/lib/syscall/syscall_bsd.go

@@ -0,0 +1,525 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build darwin || dragonfly || freebsd || netbsd || openbsd
+
+// BSD system call wrappers shared by *BSD based systems
+// including OS X (Darwin) and FreeBSD.  Like the other
+// syscall_*.go files it is compiled as Go code but also
+// used as input to mksyscall which parses the //sys
+// lines and generates system call stubs.
+
+package syscall
+
+import (
+	"unsafe"
+
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/os"
+	"github.com/goplus/llgo/c/syscall"
+)
+
+func Getgroups() (gids []int, err error) {
+	/* TODO(xsw):
+	n, err := getgroups(0, nil)
+	if err != nil {
+		return nil, err
+	}
+	if n == 0 {
+		return nil, nil
+	}
+
+	// Sanity check group count. Max is 16 on BSD.
+	if n < 0 || n > 1000 {
+		return nil, EINVAL
+	}
+
+	a := make([]_Gid_t, n)
+	n, err = getgroups(n, &a[0])
+	if err != nil {
+		return nil, err
+	}
+	gids = make([]int, n)
+	for i, v := range a[0:n] {
+		gids[i] = int(v)
+	}
+	return
+	*/
+	panic("todo: syscall.Getgroups")
+}
+
+func Setgroups(gids []int) (err error) {
+	/* TODO(xsw):
+	if len(gids) == 0 {
+		return setgroups(0, nil)
+	}
+
+	a := make([]_Gid_t, len(gids))
+	for i, v := range gids {
+		a[i] = _Gid_t(v)
+	}
+	return setgroups(len(a), &a[0])
+	*/
+	panic("todo: syscall.Setgroups")
+}
+
+func ReadDirent(fd int, buf []byte) (n int, err error) {
+	/* TODO(xsw):
+	// Final argument is (basep *uintptr) and the syscall doesn't take nil.
+	// 64 bits should be enough. (32 bits isn't even on 386). Since the
+	// actual system call is getdirentries64, 64 is a good guess.
+	// TODO(rsc): Can we use a single global basep for all calls?
+	var base = (*uintptr)(unsafe.Pointer(new(uint64)))
+	return Getdirentries(fd, buf, base)
+	*/
+	panic("todo: syscall.ReadDirent")
+}
+
+// Wait status is 7 bits at bottom, either 0 (exited),
+// 0x7F (stopped), or a signal number that caused an exit.
+// The 0x80 bit is whether there was a core dump.
+// An extra number (exit code, signal causing a stop)
+// is in the high bits.
+
+type WaitStatus uint32
+
+const (
+	mask  = 0x7F
+	core  = 0x80
+	shift = 8
+
+	exited  = 0
+	stopped = 0x7F
+)
+
+func (w WaitStatus) Exited() bool { return w&mask == exited }
+
+func (w WaitStatus) ExitStatus() int {
+	if w&mask != exited {
+		return -1
+	}
+	return int(w >> shift)
+}
+
+func (w WaitStatus) Signaled() bool { return w&mask != stopped && w&mask != 0 }
+
+func (w WaitStatus) Signal() Signal {
+	sig := Signal(w & mask)
+	if sig == stopped || sig == 0 {
+		return -1
+	}
+	return sig
+}
+
+func (w WaitStatus) CoreDump() bool { return w.Signaled() && w&core != 0 }
+
+func (w WaitStatus) Stopped() bool {
+	return w&mask == stopped && Signal(w>>shift) != Signal(syscall.SIGSTOP)
+}
+
+func (w WaitStatus) Continued() bool {
+	return w&mask == stopped && Signal(w>>shift) == Signal(syscall.SIGSTOP)
+}
+
+func (w WaitStatus) StopSignal() Signal {
+	if !w.Stopped() {
+		return -1
+	}
+	return Signal(w>>shift) & 0xFF
+}
+
+func (w WaitStatus) TrapCause() int { return -1 }
+
+func Wait4(pid int, wstatus *WaitStatus, options int, rusage *syscall.Rusage) (wpid int, err error) {
+	var status c.Int
+	wpid, err = wait4(pid, &status, options, rusage)
+	if wstatus != nil {
+		*wstatus = WaitStatus(status)
+	}
+	return
+}
+
+/* TODO(xsw):
+func (sa *SockaddrInet4) sockaddr() (unsafe.Pointer, _Socklen, error) {
+	if sa.Port < 0 || sa.Port > 0xFFFF {
+		return nil, 0, EINVAL
+	}
+	sa.raw.Len = SizeofSockaddrInet4
+	sa.raw.Family = AF_INET
+	p := (*[2]byte)(unsafe.Pointer(&sa.raw.Port))
+	p[0] = byte(sa.Port >> 8)
+	p[1] = byte(sa.Port)
+	sa.raw.Addr = sa.Addr
+	return unsafe.Pointer(&sa.raw), _Socklen(sa.raw.Len), nil
+}
+
+func (sa *SockaddrInet6) sockaddr() (unsafe.Pointer, _Socklen, error) {
+	if sa.Port < 0 || sa.Port > 0xFFFF {
+		return nil, 0, EINVAL
+	}
+	sa.raw.Len = SizeofSockaddrInet6
+	sa.raw.Family = AF_INET6
+	p := (*[2]byte)(unsafe.Pointer(&sa.raw.Port))
+	p[0] = byte(sa.Port >> 8)
+	p[1] = byte(sa.Port)
+	sa.raw.Scope_id = sa.ZoneId
+	sa.raw.Addr = sa.Addr
+	return unsafe.Pointer(&sa.raw), _Socklen(sa.raw.Len), nil
+}
+
+func (sa *SockaddrUnix) sockaddr() (unsafe.Pointer, _Socklen, error) {
+	name := sa.Name
+	n := len(name)
+	if n >= len(sa.raw.Path) || n == 0 {
+		return nil, 0, EINVAL
+	}
+	sa.raw.Len = byte(3 + n) // 2 for Family, Len; 1 for NUL
+	sa.raw.Family = AF_UNIX
+	for i := 0; i < n; i++ {
+		sa.raw.Path[i] = int8(name[i])
+	}
+	return unsafe.Pointer(&sa.raw), _Socklen(sa.raw.Len), nil
+}
+
+func (sa *SockaddrDatalink) sockaddr() (unsafe.Pointer, _Socklen, error) {
+	if sa.Index == 0 {
+		return nil, 0, EINVAL
+	}
+	sa.raw.Len = sa.Len
+	sa.raw.Family = AF_LINK
+	sa.raw.Index = sa.Index
+	sa.raw.Type = sa.Type
+	sa.raw.Nlen = sa.Nlen
+	sa.raw.Alen = sa.Alen
+	sa.raw.Slen = sa.Slen
+	sa.raw.Data = sa.Data
+	return unsafe.Pointer(&sa.raw), SizeofSockaddrDatalink, nil
+}
+
+func anyToSockaddr(rsa *RawSockaddrAny) (Sockaddr, error) {
+	switch rsa.Addr.Family {
+	case AF_LINK:
+		pp := (*RawSockaddrDatalink)(unsafe.Pointer(rsa))
+		sa := new(SockaddrDatalink)
+		sa.Len = pp.Len
+		sa.Family = pp.Family
+		sa.Index = pp.Index
+		sa.Type = pp.Type
+		sa.Nlen = pp.Nlen
+		sa.Alen = pp.Alen
+		sa.Slen = pp.Slen
+		sa.Data = pp.Data
+		return sa, nil
+
+	case AF_UNIX:
+		pp := (*RawSockaddrUnix)(unsafe.Pointer(rsa))
+		if pp.Len < 2 || pp.Len > SizeofSockaddrUnix {
+			return nil, EINVAL
+		}
+		sa := new(SockaddrUnix)
+
+		// Some BSDs include the trailing NUL in the length, whereas
+		// others do not. Work around this by subtracting the leading
+		// family and len. The path is then scanned to see if a NUL
+		// terminator still exists within the length.
+		n := int(pp.Len) - 2 // subtract leading Family, Len
+		for i := 0; i < n; i++ {
+			if pp.Path[i] == 0 {
+				// found early NUL; assume Len included the NUL
+				// or was overestimating.
+				n = i
+				break
+			}
+		}
+		sa.Name = string(unsafe.Slice((*byte)(unsafe.Pointer(&pp.Path[0])), n))
+		return sa, nil
+
+	case AF_INET:
+		pp := (*RawSockaddrInet4)(unsafe.Pointer(rsa))
+		sa := new(SockaddrInet4)
+		p := (*[2]byte)(unsafe.Pointer(&pp.Port))
+		sa.Port = int(p[0])<<8 + int(p[1])
+		sa.Addr = pp.Addr
+		return sa, nil
+
+	case AF_INET6:
+		pp := (*RawSockaddrInet6)(unsafe.Pointer(rsa))
+		sa := new(SockaddrInet6)
+		p := (*[2]byte)(unsafe.Pointer(&pp.Port))
+		sa.Port = int(p[0])<<8 + int(p[1])
+		sa.ZoneId = pp.Scope_id
+		sa.Addr = pp.Addr
+		return sa, nil
+	}
+	return nil, EAFNOSUPPORT
+}
+
+func Accept(fd int) (nfd int, sa Sockaddr, err error) {
+	var rsa RawSockaddrAny
+	var len _Socklen = SizeofSockaddrAny
+	nfd, err = accept(fd, &rsa, &len)
+	if err != nil {
+		return
+	}
+	if (runtime.GOOS == "darwin" || runtime.GOOS == "ios") && len == 0 {
+		// Accepted socket has no address.
+		// This is likely due to a bug in xnu kernels,
+		// where instead of ECONNABORTED error socket
+		// is accepted, but has no address.
+		Close(nfd)
+		return 0, nil, ECONNABORTED
+	}
+	sa, err = anyToSockaddr(&rsa)
+	if err != nil {
+		Close(nfd)
+		nfd = 0
+	}
+	return
+}
+
+func Getsockname(fd int) (sa Sockaddr, err error) {
+	var rsa RawSockaddrAny
+	var len _Socklen = SizeofSockaddrAny
+	if err = getsockname(fd, &rsa, &len); err != nil {
+		return
+	}
+	// TODO(jsing): DragonFly has a "bug" (see issue 3349), which should be
+	// reported upstream.
+	if runtime.GOOS == "dragonfly" && rsa.Addr.Family == AF_UNSPEC && rsa.Addr.Len == 0 {
+		rsa.Addr.Family = AF_UNIX
+		rsa.Addr.Len = SizeofSockaddrUnix
+	}
+	return anyToSockaddr(&rsa)
+}
+
+//sysnb socketpair(domain int, typ int, proto int, fd *[2]int32) (err error)
+
+func GetsockoptByte(fd, level, opt int) (value byte, err error) {
+	var n byte
+	vallen := _Socklen(1)
+	err = getsockopt(fd, level, opt, unsafe.Pointer(&n), &vallen)
+	return n, err
+}
+
+func GetsockoptInet4Addr(fd, level, opt int) (value [4]byte, err error) {
+	vallen := _Socklen(4)
+	err = getsockopt(fd, level, opt, unsafe.Pointer(&value[0]), &vallen)
+	return value, err
+}
+
+func GetsockoptIPMreq(fd, level, opt int) (*IPMreq, error) {
+	var value IPMreq
+	vallen := _Socklen(SizeofIPMreq)
+	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
+	return &value, err
+}
+
+func GetsockoptIPv6Mreq(fd, level, opt int) (*IPv6Mreq, error) {
+	var value IPv6Mreq
+	vallen := _Socklen(SizeofIPv6Mreq)
+	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
+	return &value, err
+}
+
+func GetsockoptIPv6MTUInfo(fd, level, opt int) (*IPv6MTUInfo, error) {
+	var value IPv6MTUInfo
+	vallen := _Socklen(SizeofIPv6MTUInfo)
+	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
+	return &value, err
+}
+
+func GetsockoptICMPv6Filter(fd, level, opt int) (*ICMPv6Filter, error) {
+	var value ICMPv6Filter
+	vallen := _Socklen(SizeofICMPv6Filter)
+	err := getsockopt(fd, level, opt, unsafe.Pointer(&value), &vallen)
+	return &value, err
+}
+
+//sys   recvfrom(fd int, p []byte, flags int, from *RawSockaddrAny, fromlen *_Socklen) (n int, err error)
+//sys   sendto(s int, buf []byte, flags int, to unsafe.Pointer, addrlen _Socklen) (err error)
+//sys	recvmsg(s int, msg *Msghdr, flags int) (n int, err error)
+
+func recvmsgRaw(fd int, p, oob []byte, flags int, rsa *RawSockaddrAny) (n, oobn int, recvflags int, err error) {
+	var msg Msghdr
+	msg.Name = (*byte)(unsafe.Pointer(rsa))
+	msg.Namelen = uint32(SizeofSockaddrAny)
+	var iov Iovec
+	if len(p) > 0 {
+		iov.Base = &p[0]
+		iov.SetLen(len(p))
+	}
+	var dummy byte
+	if len(oob) > 0 {
+		// receive at least one normal byte
+		if len(p) == 0 {
+			iov.Base = &dummy
+			iov.SetLen(1)
+		}
+		msg.Control = &oob[0]
+		msg.SetControllen(len(oob))
+	}
+	msg.Iov = &iov
+	msg.Iovlen = 1
+	if n, err = recvmsg(fd, &msg, flags); err != nil {
+		return
+	}
+	oobn = int(msg.Controllen)
+	recvflags = int(msg.Flags)
+	return
+}
+
+//sys	sendmsg(s int, msg *Msghdr, flags int) (n int, err error)
+
+func sendmsgN(fd int, p, oob []byte, ptr unsafe.Pointer, salen _Socklen, flags int) (n int, err error) {
+	var msg Msghdr
+	msg.Name = (*byte)(ptr)
+	msg.Namelen = uint32(salen)
+	var iov Iovec
+	if len(p) > 0 {
+		iov.Base = &p[0]
+		iov.SetLen(len(p))
+	}
+	var dummy byte
+	if len(oob) > 0 {
+		// send at least one normal byte
+		if len(p) == 0 {
+			iov.Base = &dummy
+			iov.SetLen(1)
+		}
+		msg.Control = &oob[0]
+		msg.SetControllen(len(oob))
+	}
+	msg.Iov = &iov
+	msg.Iovlen = 1
+	if n, err = sendmsg(fd, &msg, flags); err != nil {
+		return 0, err
+	}
+	if len(oob) > 0 && len(p) == 0 {
+		n = 0
+	}
+	return n, nil
+}
+
+//sys	kevent(kq int, change unsafe.Pointer, nchange int, event unsafe.Pointer, nevent int, timeout *Timespec) (n int, err error)
+
+func Kevent(kq int, changes, events []Kevent_t, timeout *Timespec) (n int, err error) {
+	var change, event unsafe.Pointer
+	if len(changes) > 0 {
+		change = unsafe.Pointer(&changes[0])
+	}
+	if len(events) > 0 {
+		event = unsafe.Pointer(&events[0])
+	}
+	return kevent(kq, change, len(changes), event, len(events), timeout)
+}
+
+func Sysctl(name string) (value string, err error) {
+	// Translate name to mib number.
+	mib, err := nametomib(name)
+	if err != nil {
+		return "", err
+	}
+
+	// Find size.
+	n := uintptr(0)
+	if err = sysctl(mib, nil, &n, nil, 0); err != nil {
+		return "", err
+	}
+	if n == 0 {
+		return "", nil
+	}
+
+	// Read into buffer of that size.
+	buf := make([]byte, n)
+	if err = sysctl(mib, &buf[0], &n, nil, 0); err != nil {
+		return "", err
+	}
+
+	// Throw away terminating NUL.
+	if n > 0 && buf[n-1] == '\x00' {
+		n--
+	}
+	return string(buf[0:n]), nil
+}
+*/
+
+func SysctlUint32(name string) (value uint32, err error) {
+	/* TODO(xsw):
+	// Translate name to mib number.
+	mib, err := nametomib(name)
+	if err != nil {
+		return 0, err
+	}
+
+	// Read into buffer of that size.
+	n := uintptr(4)
+	buf := make([]byte, 4)
+	if err = sysctl(mib, &buf[0], &n, nil, 0); err != nil {
+		return 0, err
+	}
+	if n != 4 {
+		return 0, Errno(syscall.EIO)
+	}
+	return *(*uint32)(unsafe.Pointer(&buf[0])), nil
+	*/
+	n := uintptr(4)
+	ret := os.Sysctlbyname(c.AllocaCStr(name), unsafe.Pointer(&value), &n, nil, 0)
+	if ret != 0 {
+		err = Errno(os.Errno())
+	}
+	return
+}
+
+/*
+//sys	utimes(path string, timeval *[2]Timeval) (err error)
+
+func Utimes(path string, tv []Timeval) (err error) {
+	if len(tv) != 2 {
+		return EINVAL
+	}
+	return utimes(path, (*[2]Timeval)(unsafe.Pointer(&tv[0])))
+}
+
+func UtimesNano(path string, ts []Timespec) error {
+	if len(ts) != 2 {
+		return EINVAL
+	}
+	err := utimensat(_AT_FDCWD, path, (*[2]Timespec)(unsafe.Pointer(&ts[0])), 0)
+	if err != ENOSYS {
+		return err
+	}
+	// Not as efficient as it could be because Timespec and
+	// Timeval have different types in the different OSes
+	tv := [2]Timeval{
+		NsecToTimeval(TimespecToNsec(ts[0])),
+		NsecToTimeval(TimespecToNsec(ts[1])),
+	}
+	return utimes(path, (*[2]Timeval)(unsafe.Pointer(&tv[0])))
+}
+
+//sys	futimes(fd int, timeval *[2]Timeval) (err error)
+
+func Futimes(fd int, tv []Timeval) (err error) {
+	if len(tv) != 2 {
+		return EINVAL
+	}
+	return futimes(fd, (*[2]Timeval)(unsafe.Pointer(&tv[0])))
+}
+
+//sys	fcntl(fd int, cmd int, arg int) (val int, err error)
+
+var mapper = &mmapper{
+	active: make(map[*byte][]byte),
+	mmap:   mmap,
+	munmap: munmap,
+}
+
+func Mmap(fd int, offset int64, length int, prot int, flags int) (data []byte, err error) {
+	return mapper.Mmap(fd, offset, length, prot, flags)
+}
+
+func Munmap(b []byte) (err error) {
+	return mapper.Munmap(b)
+}
+*/

compiler/internal/lib/syscall/syscall_linux.go

@@ -0,0 +1,128 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Linux system calls.
+// This file is compiled as ordinary Go code,
+// but it is also input to mksyscall,
+// which parses the //sys lines and generates system call stubs.
+// Note that sometimes we use a lowercase //sys name and
+// wrap it in our own nicer implementation.
+
+package syscall
+
+import (
+	_ "unsafe"
+
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/os"
+	"github.com/goplus/llgo/c/syscall"
+)
+
+// -----------------------------------------------------------------------------
+
+type WaitStatus uint32
+
+// Wait status is 7 bits at bottom, either 0 (exited),
+// 0x7F (stopped), or a signal number that caused an exit.
+// The 0x80 bit is whether there was a core dump.
+// An extra number (exit code, signal causing a stop)
+// is in the high bits. At least that's the idea.
+// There are various irregularities. For example, the
+// "continued" status is 0xFFFF, distinguishing itself
+// from stopped via the core dump bit.
+
+const (
+	mask    = 0x7F
+	core    = 0x80
+	exited  = 0x00
+	stopped = 0x7F
+	shift   = 8
+)
+
+func (w WaitStatus) Exited() bool { return w&mask == exited }
+
+func (w WaitStatus) Signaled() bool { return w&mask != stopped && w&mask != exited }
+
+func (w WaitStatus) Stopped() bool { return w&0xFF == stopped }
+
+func (w WaitStatus) Continued() bool { return w == 0xFFFF }
+
+func (w WaitStatus) CoreDump() bool { return w.Signaled() && w&core != 0 }
+
+func (w WaitStatus) ExitStatus() int {
+	if !w.Exited() {
+		return -1
+	}
+	return int(w>>shift) & 0xFF
+}
+
+func (w WaitStatus) Signal() Signal {
+	if !w.Signaled() {
+		return -1
+	}
+	return Signal(w & mask)
+}
+
+func (w WaitStatus) StopSignal() Signal {
+	if !w.Stopped() {
+		return -1
+	}
+	return Signal(w>>shift) & 0xFF
+}
+
+func (w WaitStatus) TrapCause() int {
+	/*
+		if w.StopSignal() != SIGTRAP {
+			return -1
+		}
+
+		return int(w>>shift) >> 8
+	*/
+	panic("todo: syscall.WaitStatus.TrapCause")
+}
+
+func Wait4(pid int, wstatus *WaitStatus, options int, rusage *syscall.Rusage) (wpid int, err error) {
+	var status c.Int
+	wpid, err = wait4(pid, &status, options, rusage)
+	if wstatus != nil {
+		*wstatus = WaitStatus(status)
+	}
+	return
+}
+
+// -----------------------------------------------------------------------------
+
+// int pipe2(int pipefd[2], int flags);
+//
+//go:linkname pipe2 C.pipe2
+func pipe2(pipefd *[2]c.Int, flags c.Int) c.Int
+
+func Pipe2(p []int, flags int) error {
+	if len(p) != 2 {
+		return Errno(syscall.EINVAL)
+	}
+	var pp [2]c.Int
+	ret := pipe2(&pp, c.Int(flags))
+	if ret == 0 {
+		p[0] = int(pp[0])
+		p[1] = int(pp[1])
+		return nil
+	}
+	return Errno(ret)
+}
+
+// -----------------------------------------------------------------------------
+
+func Faccessat(dirfd int, path string, mode uint32, flags int) (err error) {
+	ret := faccessat(c.Int(dirfd), c.AllocaCStr(path), c.Int(mode), c.Int(flags))
+	if ret != 0 {
+		return Errno(os.Errno())
+	}
+	return nil
+}
+
+//go:linkname faccessat C.faccessat
+func faccessat(dirfd c.Int, path *c.Char, mode c.Int, flags c.Int) c.Int
+
+// -----------------------------------------------------------------------------

compiler/internal/lib/syscall/syscall_unix.go

@@ -0,0 +1,71 @@
+// Copyright 2009 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build unix
+
+package syscall
+
+import (
+	"unsafe"
+
+	"github.com/goplus/llgo/c"
+	"github.com/goplus/llgo/c/syscall"
+	"github.com/goplus/llgo/compiler/internal/lib/internal/oserror"
+)
+
+var (
+	Stdin  = 0
+	Stdout = 1
+	Stderr = 2
+)
+
+type Errno uintptr
+
+func (e Errno) Error() string {
+	ret := c.Strerror(c.Int(e))
+	return unsafe.String((*byte)(unsafe.Pointer(ret)), c.Strlen(ret))
+}
+
+func (e Errno) Is(target error) bool {
+	switch target {
+	case oserror.ErrPermission:
+		return e == Errno(syscall.EACCES) || e == Errno(syscall.EPERM)
+	case oserror.ErrExist:
+		return e == Errno(syscall.EEXIST) || e == Errno(syscall.ENOTEMPTY)
+	case oserror.ErrNotExist:
+		return e == Errno(syscall.ENOENT)
+		// TODO(xsw): go1.21
+		// case errors.ErrUnsupported:
+		//	return e == ENOSYS || e == ENOTSUP || e == EOPNOTSUPP
+	}
+	return false
+}
+
+func (e Errno) Temporary() bool {
+	return e == Errno(syscall.EINTR) || e == Errno(syscall.EMFILE) ||
+		e == Errno(syscall.ENFILE) || e.Timeout()
+}
+
+func (e Errno) Timeout() bool {
+	return e == Errno(syscall.EAGAIN) || e == Errno(syscall.EWOULDBLOCK) || e == Errno(syscall.ETIMEDOUT)
+}
+
+// A Signal is a number describing a process signal.
+// It implements the os.Signal interface.
+type Signal int
+
+func (s Signal) Signal() {}
+
+func (s Signal) String() string {
+	/*
+		if 0 <= s && int(s) < len(signals) {
+			str := signals[s]
+			if str != "" {
+				return str
+			}
+		}
+		return "signal " + itoa.Itoa(int(s))
+	*/
+	panic("todo: syscall.Signal.String")
+}