exec_unix.go

Documentation: syscall

     1  // Copyright 2009 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  //go:build unix
     6  
     7  // Fork, exec, wait, etc.
     8  
     9  package syscall
    10  
    11  import (
    12  	errorspkg "errors"
    13  	"internal/bytealg"
    14  	"runtime"
    15  	"sync"
    16  	"unsafe"
    17  )
    18  
    19  // Lock synchronizing creation of new file descriptors with fork.
    20  //
    21  // We want the child in a fork/exec sequence to inherit only the
    22  // file descriptors we intend. To do that, we mark all file
    23  // descriptors close-on-exec and then, in the child, explicitly
    24  // unmark the ones we want the exec'ed program to keep.
    25  // Unix doesn't make this easy: there is, in general, no way to
    26  // allocate a new file descriptor close-on-exec. Instead you
    27  // have to allocate the descriptor and then mark it close-on-exec.
    28  // If a fork happens between those two events, the child's exec
    29  // will inherit an unwanted file descriptor.
    30  //
    31  // This lock solves that race: the create new fd/mark close-on-exec
    32  // operation is done holding ForkLock for reading, and the fork itself
    33  // is done holding ForkLock for writing. At least, that's the idea.
    34  // There are some complications.
    35  //
    36  // Some system calls that create new file descriptors can block
    37  // for arbitrarily long times: open on a hung NFS server or named
    38  // pipe, accept on a socket, and so on. We can't reasonably grab
    39  // the lock across those operations.
    40  //
    41  // It is worse to inherit some file descriptors than others.
    42  // If a non-malicious child accidentally inherits an open ordinary file,
    43  // that's not a big deal. On the other hand, if a long-lived child
    44  // accidentally inherits the write end of a pipe, then the reader
    45  // of that pipe will not see EOF until that child exits, potentially
    46  // causing the parent program to hang. This is a common problem
    47  // in threaded C programs that use popen.
    48  //
    49  // Luckily, the file descriptors that are most important not to
    50  // inherit are not the ones that can take an arbitrarily long time
    51  // to create: pipe returns instantly, and the net package uses
    52  // non-blocking I/O to accept on a listening socket.
    53  // The rules for which file descriptor-creating operations use the
    54  // ForkLock are as follows:
    55  //
    56  // 1) Pipe. Does not block. Use the ForkLock.
    57  // 2) Socket. Does not block. Use the ForkLock.
    58  // 3) Accept. If using non-blocking mode, use the ForkLock.
    59  //             Otherwise, live with the race.
    60  // 4) Open. Can block. Use O_CLOEXEC if available (Linux).
    61  //             Otherwise, live with the race.
    62  // 5) Dup. Does not block. Use the ForkLock.
    63  //             On Linux, could use fcntl F_DUPFD_CLOEXEC
    64  //             instead of the ForkLock, but only for dup(fd, -1).
    65  
    66  var ForkLock sync.RWMutex
    67  
    68  // StringSlicePtr converts a slice of strings to a slice of pointers
    69  // to NUL-terminated byte arrays. If any string contains a NUL byte
    70  // this function panics instead of returning an error.
    71  //
    72  // Deprecated: Use SlicePtrFromStrings instead.
    73  func StringSlicePtr(ss []string) []*byte {
    74  	bb := make([]*byte, len(ss)+1)
    75  	for i := 0; i < len(ss); i++ {
    76  		bb[i] = StringBytePtr(ss[i])
    77  	}
    78  	bb[len(ss)] = nil
    79  	return bb
    80  }
    81  
    82  // SlicePtrFromStrings converts a slice of strings to a slice of
    83  // pointers to NUL-terminated byte arrays. If any string contains
    84  // a NUL byte, it returns (nil, EINVAL).
    85  func SlicePtrFromStrings(ss []string) ([]*byte, error) {
    86  	n := 0
    87  	for _, s := range ss {
    88  		if bytealg.IndexByteString(s, 0) != -1 {
    89  			return nil, EINVAL
    90  		}
    91  		n += len(s) + 1 // +1 for NUL
    92  	}
    93  	bb := make([]*byte, len(ss)+1)
    94  	b := make([]byte, n)
    95  	n = 0
    96  	for i, s := range ss {
    97  		bb[i] = &b[n]
    98  		copy(b[n:], s)
    99  		n += len(s) + 1
   100  	}
   101  	return bb, nil
   102  }
   103  
   104  func CloseOnExec(fd int) { fcntl(fd, F_SETFD, FD_CLOEXEC) }
   105  
   106  func SetNonblock(fd int, nonblocking bool) (err error) {
   107  	flag, err := fcntl(fd, F_GETFL, 0)
   108  	if err != nil {
   109  		return err
   110  	}
   111  	if nonblocking {
   112  		flag |= O_NONBLOCK
   113  	} else {
   114  		flag &^= O_NONBLOCK
   115  	}
   116  	_, err = fcntl(fd, F_SETFL, flag)
   117  	return err
   118  }
   119  
   120  // Credential holds user and group identities to be assumed
   121  // by a child process started by StartProcess.
   122  type Credential struct {
   123  	Uid         uint32   // User ID.
   124  	Gid         uint32   // Group ID.
   125  	Groups      []uint32 // Supplementary group IDs.
   126  	NoSetGroups bool     // If true, don't set supplementary groups
   127  }
   128  
   129  // ProcAttr holds attributes that will be applied to a new process started
   130  // by StartProcess.
   131  type ProcAttr struct {
   132  	Dir   string    // Current working directory.
   133  	Env   []string  // Environment.
   134  	Files []uintptr // File descriptors.
   135  	Sys   *SysProcAttr
   136  }
   137  
   138  var zeroProcAttr ProcAttr
   139  var zeroSysProcAttr SysProcAttr
   140  
   141  func forkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error) {
   142  	var p [2]int
   143  	var n int
   144  	var err1 Errno
   145  	var wstatus WaitStatus
   146  
   147  	if attr == nil {
   148  		attr = &zeroProcAttr
   149  	}
   150  	sys := attr.Sys
   151  	if sys == nil {
   152  		sys = &zeroSysProcAttr
   153  	}
   154  
   155  	// Convert args to C form.
   156  	argv0p, err := BytePtrFromString(argv0)
   157  	if err != nil {
   158  		return 0, err
   159  	}
   160  	argvp, err := SlicePtrFromStrings(argv)
   161  	if err != nil {
   162  		return 0, err
   163  	}
   164  	envvp, err := SlicePtrFromStrings(attr.Env)
   165  	if err != nil {
   166  		return 0, err
   167  	}
   168  
   169  	if (runtime.GOOS == "freebsd" || runtime.GOOS == "dragonfly") && len(argv[0]) > len(argv0) {
   170  		argvp[0] = argv0p
   171  	}
   172  
   173  	var chroot *byte
   174  	if sys.Chroot != "" {
   175  		chroot, err = BytePtrFromString(sys.Chroot)
   176  		if err != nil {
   177  			return 0, err
   178  		}
   179  	}
   180  	var dir *byte
   181  	if attr.Dir != "" {
   182  		dir, err = BytePtrFromString(attr.Dir)
   183  		if err != nil {
   184  			return 0, err
   185  		}
   186  	}
   187  
   188  	// Both Setctty and Foreground use the Ctty field,
   189  	// but they give it slightly different meanings.
   190  	if sys.Setctty && sys.Foreground {
   191  		return 0, errorspkg.New("both Setctty and Foreground set in SysProcAttr")
   192  	}
   193  	if sys.Setctty && sys.Ctty >= len(attr.Files) {
   194  		return 0, errorspkg.New("Setctty set but Ctty not valid in child")
   195  	}
   196  
   197  	// Acquire the fork lock so that no other threads
   198  	// create new fds that are not yet close-on-exec
   199  	// before we fork.
   200  	ForkLock.Lock()
   201  
   202  	// Allocate child status pipe close on exec.
   203  	if err = forkExecPipe(p[:]); err != nil {
   204  		ForkLock.Unlock()
   205  		return 0, err
   206  	}
   207  
   208  	// Kick off child.
   209  	pid, err1 = forkAndExecInChild(argv0p, argvp, envvp, chroot, dir, attr, sys, p[1])
   210  	if err1 != 0 {
   211  		Close(p[0])
   212  		Close(p[1])
   213  		ForkLock.Unlock()
   214  		return 0, Errno(err1)
   215  	}
   216  	ForkLock.Unlock()
   217  
   218  	// Read child error status from pipe.
   219  	Close(p[1])
   220  	for {
   221  		n, err = readlen(p[0], (*byte)(unsafe.Pointer(&err1)), int(unsafe.Sizeof(err1)))
   222  		if err != EINTR {
   223  			break
   224  		}
   225  	}
   226  	Close(p[0])
   227  	if err != nil || n != 0 {
   228  		if n == int(unsafe.Sizeof(err1)) {
   229  			err = Errno(err1)
   230  		}
   231  		if err == nil {
   232  			err = EPIPE
   233  		}
   234  
   235  		// Child failed; wait for it to exit, to make sure
   236  		// the zombies don't accumulate.
   237  		_, err1 := Wait4(pid, &wstatus, 0, nil)
   238  		for err1 == EINTR {
   239  			_, err1 = Wait4(pid, &wstatus, 0, nil)
   240  		}
   241  		return 0, err
   242  	}
   243  
   244  	// Read got EOF, so pipe closed on exec, so exec succeeded.
   245  	return pid, nil
   246  }
   247  
   248  // Combination of fork and exec, careful to be thread safe.
   249  func ForkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error) {
   250  	return forkExec(argv0, argv, attr)
   251  }
   252  
   253  // StartProcess wraps ForkExec for package os.
   254  func StartProcess(argv0 string, argv []string, attr *ProcAttr) (pid int, handle uintptr, err error) {
   255  	pid, err = forkExec(argv0, argv, attr)
   256  	return pid, 0, err
   257  }
   258  
   259  // Implemented in runtime package.
   260  func runtime_BeforeExec()
   261  func runtime_AfterExec()
   262  
   263  // execveLibc is non-nil on OS using libc syscall, set to execve in exec_libc.go; this
   264  // avoids a build dependency for other platforms.
   265  var execveLibc func(path uintptr, argv uintptr, envp uintptr) Errno
   266  var execveDarwin func(path *byte, argv **byte, envp **byte) error
   267  var execveOpenBSD func(path *byte, argv **byte, envp **byte) error
   268  
   269  // Exec invokes the execve(2) system call.
   270  func Exec(argv0 string, argv []string, envv []string) (err error) {
   271  	argv0p, err := BytePtrFromString(argv0)
   272  	if err != nil {
   273  		return err
   274  	}
   275  	argvp, err := SlicePtrFromStrings(argv)
   276  	if err != nil {
   277  		return err
   278  	}
   279  	envvp, err := SlicePtrFromStrings(envv)
   280  	if err != nil {
   281  		return err
   282  	}
   283  	runtime_BeforeExec()
   284  
   285  	var err1 error
   286  	if runtime.GOOS == "solaris" || runtime.GOOS == "illumos" || runtime.GOOS == "aix" {
   287  		// RawSyscall should never be used on Solaris, illumos, or AIX.
   288  		err1 = execveLibc(
   289  			uintptr(unsafe.Pointer(argv0p)),
   290  			uintptr(unsafe.Pointer(&argvp[0])),
   291  			uintptr(unsafe.Pointer(&envvp[0])))
   292  	} else if runtime.GOOS == "darwin" || runtime.GOOS == "ios" {
   293  		// Similarly on Darwin.
   294  		err1 = execveDarwin(argv0p, &argvp[0], &envvp[0])
   295  	} else if runtime.GOOS == "openbsd" && (runtime.GOARCH == "386" || runtime.GOARCH == "amd64" || runtime.GOARCH == "arm" || runtime.GOARCH == "arm64") {
   296  		// Similarly on OpenBSD.
   297  		err1 = execveOpenBSD(argv0p, &argvp[0], &envvp[0])
   298  	} else {
   299  		_, _, err1 = RawSyscall(SYS_EXECVE,
   300  			uintptr(unsafe.Pointer(argv0p)),
   301  			uintptr(unsafe.Pointer(&argvp[0])),
   302  			uintptr(unsafe.Pointer(&envvp[0])))
   303  	}
   304  	runtime_AfterExec()
   305  	return err1
   306  }
   307
View as plain text