Source file src/syscall/exec_linux.go

Documentation: syscall

  // 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.
  
  // +build linux
  
  package syscall
  
  import (
  	"runtime"
  	"unsafe"
  )
  
  // 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       bool           // Enable tracing.
  	Setsid       bool           // Create session.
  	Setpgid      bool           // Set process group ID to Pgid, or, if Pgid == 0, to new pid.
  	Setctty      bool           // Set controlling terminal to fd Ctty (only meaningful if Setsid is set)
  	Noctty       bool           // Detach fd 0 from controlling terminal
  	Ctty         int            // Controlling TTY fd
  	Foreground   bool           // Place child's process group in foreground. (Implies Setpgid. Uses Ctty as fd of controlling TTY)
  	Pgid         int            // Child's process group ID if Setpgid.
  	Pdeathsig    Signal         // Signal that the process will get when its parent dies (Linux only)
  	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)
  }
  
  var (
  	none  = [...]byte{'n', 'o', 'n', 'e', 0}
  	slash = [...]byte{'/', 0}
  )
  
  // 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.
  //go:norace
  func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
  	// Set up and fork. This returns immediately in the parent or
  	// if there's an error.
  	r1, err1, p, locked := forkAndExecInChild1(argv0, argv, envv, chroot, dir, attr, sys, pipe)
  	if locked {
  		runtime_AfterFork()
  	}
  	if err1 != 0 {
  		return 0, err1
  	}
  
  	// parent; return PID
  	pid = int(r1)
  
  	if sys.UidMappings != nil || sys.GidMappings != nil {
  		Close(p[0])
  		err := writeUidGidMappings(pid, sys)
  		var err2 Errno
  		if err != nil {
  			err2 = err.(Errno)
  		}
  		RawSyscall(SYS_WRITE, uintptr(p[1]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
  		Close(p[1])
  	}
  
  	return pid, 0
  }
  
  // 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
  func forkAndExecInChild1(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (r1 uintptr, err1 Errno, p [2]int, locked bool) {
  	// 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., err1).
  	var (
  		err2   Errno
  		nextfd int
  		i      int
  	)
  
  	// Record parent PID so child can test if it has died.
  	ppid, _, _ := RawSyscall(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(p[:]); err != nil {
  			err1 = err.(Errno)
  			return
  		}
  	}
  
  	// About to call fork.
  	// No more allocation or calls of non-assembly functions.
  	runtime_BeforeFork()
  	locked = true
  	switch {
  	case runtime.GOARCH == "amd64" && sys.Cloneflags&CLONE_NEWUSER == 0:
  		r1, err1 = rawVforkSyscall(SYS_CLONE, uintptr(SIGCHLD|CLONE_VFORK|CLONE_VM)|sys.Cloneflags)
  	case runtime.GOARCH == "s390x":
  		r1, _, err1 = RawSyscall6(SYS_CLONE, 0, uintptr(SIGCHLD)|sys.Cloneflags, 0, 0, 0, 0)
  	default:
  		r1, _, err1 = RawSyscall6(SYS_CLONE, uintptr(SIGCHLD)|sys.Cloneflags, 0, 0, 0, 0, 0)
  	}
  	if err1 != 0 || r1 != 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.
  
  	runtime_AfterForkInChild()
  
  	// 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(p[1]), 0, 0); err1 != 0 {
  			goto childerror
  		}
  		r1, _, err1 = RawSyscall(SYS_READ, uintptr(p[0]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
  		if err1 != 0 {
  			goto childerror
  		}
  		if r1 != 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 {
  			r1, _, err1 = RawSyscall(SYS_GETPID, 0, 0, 0)
  			if err1 != 0 {
  				goto childerror
  			}
  
  			pgrp = int32(r1)
  		}
  
  		// Place process group in foreground.
  		_, _, err1 = RawSyscall(SYS_IOCTL, uintptr(sys.Ctty), uintptr(TIOCSPGRP), uintptr(unsafe.Pointer(&pgrp)))
  		if err1 != 0 {
  			goto childerror
  		}
  	}
  
  	// Unshare
  	if sys.Unshareflags != 0 {
  		_, _, err1 = RawSyscall(SYS_UNSHARE, sys.Unshareflags, 0, 0)
  		if 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
  		}
  	}
  
  	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.
  		r1, _, _ = RawSyscall(SYS_GETPPID, 0, 0, 0)
  		if r1 != ppid {
  			pid, _, _ := RawSyscall(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_dup, uintptr(pipe), uintptr(nextfd), 0)
  		if err1 != 0 {
  			goto childerror
  		}
  		RawSyscall(SYS_FCNTL, uintptr(nextfd), F_SETFD, FD_CLOEXEC)
  		pipe = nextfd
  		nextfd++
  	}
  	for i = 0; i < len(fd); i++ {
  		if fd[i] >= 0 && fd[i] < int(i) {
  			if nextfd == pipe { // don't stomp on pipe
  				nextfd++
  			}
  			_, _, err1 = RawSyscall(_SYS_dup, uintptr(fd[i]), uintptr(nextfd), 0)
  			if err1 != 0 {
  				goto childerror
  			}
  			RawSyscall(SYS_FCNTL, uintptr(nextfd), F_SETFD, FD_CLOEXEC)
  			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] == int(i) {
  			// dup2(i, i) won't clear close-on-exec flag on Linux,
  			// probably not elsewhere either.
  			_, _, err1 = RawSyscall(SYS_FCNTL, 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_dup, 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
  		}
  	}
  
  	// 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)
  	}
  }
  
  // Try to open a pipe with O_CLOEXEC set on both file descriptors.
  func forkExecPipe(p []int) (err error) {
  	err = Pipe2(p, O_CLOEXEC)
  	// pipe2 was added in 2.6.27 and our minimum requirement is 2.6.23, so it
  	// might not be implemented.
  	if err == ENOSYS {
  		if err = Pipe(p); err != nil {
  			return
  		}
  		if _, err = fcntl(p[0], F_SETFD, FD_CLOEXEC); err != nil {
  			return
  		}
  		_, err = fcntl(p[1], F_SETFD, FD_CLOEXEC)
  	}
  	return
  }
  
  // writeIDMappings writes the user namespace User ID or Group ID mappings to the specified path.
  func writeIDMappings(path string, idMap []SysProcIDMap) error {
  	fd, err := Open(path, O_RDWR, 0)
  	if err != nil {
  		return err
  	}
  
  	data := ""
  	for _, im := range idMap {
  		data = data + itoa(im.ContainerID) + " " + itoa(im.HostID) + " " + itoa(im.Size) + "\n"
  	}
  
  	bytes, err := ByteSliceFromString(data)
  	if err != nil {
  		Close(fd)
  		return err
  	}
  
  	if _, err := Write(fd, bytes); err != nil {
  		Close(fd)
  		return err
  	}
  
  	if err := Close(fd); err != nil {
  		return err
  	}
  
  	return nil
  }
  
  // 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 {
  	sgf := "/proc/" + 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)
  }
  
  // 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 {
  	if sys.UidMappings != nil {
  		uidf := "/proc/" + 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(pid) + "/gid_map"
  		if err := writeIDMappings(gidf, sys.GidMappings); err != nil {
  			return err
  		}
  	}
  
  	return nil
  }
  

View as plain text