Black Lives Matter. Support the Equal Justice Initiative.

Source file src/syscall/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 aix || darwin || dragonfly || freebsd || linux || netbsd || openbsd || solaris
     6  // +build aix darwin dragonfly freebsd linux netbsd openbsd solaris
     7  
     8  // Fork, exec, wait, etc.
     9  
    10  package syscall
    11  
    12  import (
    13  	errorspkg "errors"
    14  	"internal/bytealg"
    15  	"runtime"
    16  	"sync"
    17  	"unsafe"
    18  )
    19  
    20  // Lock synchronizing creation of new file descriptors with fork.
    21  //
    22  // We want the child in a fork/exec sequence to inherit only the
    23  // file descriptors we intend. To do that, we mark all file
    24  // descriptors close-on-exec and then, in the child, explicitly
    25  // unmark the ones we want the exec'ed program to keep.
    26  // Unix doesn't make this easy: there is, in general, no way to
    27  // allocate a new file descriptor close-on-exec. Instead you
    28  // have to allocate the descriptor and then mark it close-on-exec.
    29  // If a fork happens between those two events, the child's exec
    30  // will inherit an unwanted file descriptor.
    31  //
    32  // This lock solves that race: the create new fd/mark close-on-exec
    33  // operation is done holding ForkLock for reading, and the fork itself
    34  // is done holding ForkLock for writing. At least, that's the idea.
    35  // There are some complications.
    36  //
    37  // Some system calls that create new file descriptors can block
    38  // for arbitrarily long times: open on a hung NFS server or named
    39  // pipe, accept on a socket, and so on. We can't reasonably grab
    40  // the lock across those operations.
    41  //
    42  // It is worse to inherit some file descriptors than others.
    43  // If a non-malicious child accidentally inherits an open ordinary file,
    44  // that's not a big deal. On the other hand, if a long-lived child
    45  // accidentally inherits the write end of a pipe, then the reader
    46  // of that pipe will not see EOF until that child exits, potentially
    47  // causing the parent program to hang. This is a common problem
    48  // in threaded C programs that use popen.
    49  //
    50  // Luckily, the file descriptors that are most important not to
    51  // inherit are not the ones that can take an arbitrarily long time
    52  // to create: pipe returns instantly, and the net package uses
    53  // non-blocking I/O to accept on a listening socket.
    54  // The rules for which file descriptor-creating operations use the
    55  // ForkLock are as follows:
    56  //
    57  // 1) Pipe. Does not block. Use the ForkLock.
    58  // 2) Socket. Does not block. Use the ForkLock.
    59  // 3) Accept. If using non-blocking mode, use the ForkLock.
    60  //             Otherwise, live with the race.
    61  // 4) Open. Can block. Use O_CLOEXEC if available (Linux).
    62  //             Otherwise, live with the race.
    63  // 5) Dup. Does not block. Use the ForkLock.
    64  //             On Linux, could use fcntl F_DUPFD_CLOEXEC
    65  //             instead of the ForkLock, but only for dup(fd, -1).
    66  
    67  var ForkLock sync.RWMutex
    68  
    69  // StringSlicePtr converts a slice of strings to a slice of pointers
    70  // to NUL-terminated byte arrays. If any string contains a NUL byte
    71  // this function panics instead of returning an error.
    72  //
    73  // Deprecated: Use SlicePtrFromStrings instead.
    74  func StringSlicePtr(ss []string) []*byte {
    75  	bb := make([]*byte, len(ss)+1)
    76  	for i := 0; i < len(ss); i++ {
    77  		bb[i] = StringBytePtr(ss[i])
    78  	}
    79  	bb[len(ss)] = nil
    80  	return bb
    81  }
    82  
    83  // SlicePtrFromStrings converts a slice of strings to a slice of
    84  // pointers to NUL-terminated byte arrays. If any string contains
    85  // a NUL byte, it returns (nil, EINVAL).
    86  func SlicePtrFromStrings(ss []string) ([]*byte, error) {
    87  	n := 0
    88  	for _, s := range ss {
    89  		if bytealg.IndexByteString(s, 0) != -1 {
    90  			return nil, EINVAL
    91  		}
    92  		n += len(s) + 1 // +1 for NUL
    93  	}
    94  	bb := make([]*byte, len(ss)+1)
    95  	b := make([]byte, n)
    96  	n = 0
    97  	for i, s := range ss {
    98  		bb[i] = &b[n]
    99  		copy(b[n:], s)
   100  		n += len(s) + 1
   101  	}
   102  	return bb, nil
   103  }
   104  
   105  func CloseOnExec(fd int) { fcntl(fd, F_SETFD, FD_CLOEXEC) }
   106  
   107  func SetNonblock(fd int, nonblocking bool) (err error) {
   108  	flag, err := fcntl(fd, F_GETFL, 0)
   109  	if err != nil {
   110  		return err
   111  	}
   112  	if nonblocking {
   113  		flag |= O_NONBLOCK
   114  	} else {
   115  		flag &^= O_NONBLOCK
   116  	}
   117  	_, err = fcntl(fd, F_SETFL, flag)
   118  	return err
   119  }
   120  
   121  // Credential holds user and group identities to be assumed
   122  // by a child process started by StartProcess.
   123  type Credential struct {
   124  	Uid         uint32   // User ID.
   125  	Gid         uint32   // Group ID.
   126  	Groups      []uint32 // Supplementary group IDs.
   127  	NoSetGroups bool     // If true, don't set supplementary groups
   128  }
   129  
   130  // ProcAttr holds attributes that will be applied to a new process started
   131  // by StartProcess.
   132  type ProcAttr struct {
   133  	Dir   string    // Current working directory.
   134  	Env   []string  // Environment.
   135  	Files []uintptr // File descriptors.
   136  	Sys   *SysProcAttr
   137  }
   138  
   139  var zeroProcAttr ProcAttr
   140  var zeroSysProcAttr SysProcAttr
   141  
   142  func forkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error) {
   143  	var p [2]int
   144  	var n int
   145  	var err1 Errno
   146  	var wstatus WaitStatus
   147  
   148  	if attr == nil {
   149  		attr = &zeroProcAttr
   150  	}
   151  	sys := attr.Sys
   152  	if sys == nil {
   153  		sys = &zeroSysProcAttr
   154  	}
   155  
   156  	p[0] = -1
   157  	p[1] = -1
   158  
   159  	// Convert args to C form.
   160  	argv0p, err := BytePtrFromString(argv0)
   161  	if err != nil {
   162  		return 0, err
   163  	}
   164  	argvp, err := SlicePtrFromStrings(argv)
   165  	if err != nil {
   166  		return 0, err
   167  	}
   168  	envvp, err := SlicePtrFromStrings(attr.Env)
   169  	if err != nil {
   170  		return 0, err
   171  	}
   172  
   173  	if (runtime.GOOS == "freebsd" || runtime.GOOS == "dragonfly") && len(argv[0]) > len(argv0) {
   174  		argvp[0] = argv0p
   175  	}
   176  
   177  	var chroot *byte
   178  	if sys.Chroot != "" {
   179  		chroot, err = BytePtrFromString(sys.Chroot)
   180  		if err != nil {
   181  			return 0, err
   182  		}
   183  	}
   184  	var dir *byte
   185  	if attr.Dir != "" {
   186  		dir, err = BytePtrFromString(attr.Dir)
   187  		if err != nil {
   188  			return 0, err
   189  		}
   190  	}
   191  
   192  	// Both Setctty and Foreground use the Ctty field,
   193  	// but they give it slightly different meanings.
   194  	if sys.Setctty && sys.Foreground {
   195  		return 0, errorspkg.New("both Setctty and Foreground set in SysProcAttr")
   196  	}
   197  	if sys.Setctty && sys.Ctty >= len(attr.Files) {
   198  		return 0, errorspkg.New("Setctty set but Ctty not valid in child")
   199  	}
   200  
   201  	// Acquire the fork lock so that no other threads
   202  	// create new fds that are not yet close-on-exec
   203  	// before we fork.
   204  	ForkLock.Lock()
   205  
   206  	// Allocate child status pipe close on exec.
   207  	if err = forkExecPipe(p[:]); err != nil {
   208  		goto error
   209  	}
   210  
   211  	// Kick off child.
   212  	pid, err1 = forkAndExecInChild(argv0p, argvp, envvp, chroot, dir, attr, sys, p[1])
   213  	if err1 != 0 {
   214  		err = Errno(err1)
   215  		goto error
   216  	}
   217  	ForkLock.Unlock()
   218  
   219  	// Read child error status from pipe.
   220  	Close(p[1])
   221  	for {
   222  		n, err = readlen(p[0], (*byte)(unsafe.Pointer(&err1)), int(unsafe.Sizeof(err1)))
   223  		if err != EINTR {
   224  			break
   225  		}
   226  	}
   227  	Close(p[0])
   228  	if err != nil || n != 0 {
   229  		if n == int(unsafe.Sizeof(err1)) {
   230  			err = Errno(err1)
   231  		}
   232  		if err == nil {
   233  			err = EPIPE
   234  		}
   235  
   236  		// Child failed; wait for it to exit, to make sure
   237  		// the zombies don't accumulate.
   238  		_, err1 := Wait4(pid, &wstatus, 0, nil)
   239  		for err1 == EINTR {
   240  			_, err1 = Wait4(pid, &wstatus, 0, nil)
   241  		}
   242  		return 0, err
   243  	}
   244  
   245  	// Read got EOF, so pipe closed on exec, so exec succeeded.
   246  	return pid, nil
   247  
   248  error:
   249  	if p[0] >= 0 {
   250  		Close(p[0])
   251  		Close(p[1])
   252  	}
   253  	ForkLock.Unlock()
   254  	return 0, err
   255  }
   256  
   257  // Combination of fork and exec, careful to be thread safe.
   258  func ForkExec(argv0 string, argv []string, attr *ProcAttr) (pid int, err error) {
   259  	return forkExec(argv0, argv, attr)
   260  }
   261  
   262  // StartProcess wraps ForkExec for package os.
   263  func StartProcess(argv0 string, argv []string, attr *ProcAttr) (pid int, handle uintptr, err error) {
   264  	pid, err = forkExec(argv0, argv, attr)
   265  	return pid, 0, err
   266  }
   267  
   268  // Implemented in runtime package.
   269  func runtime_BeforeExec()
   270  func runtime_AfterExec()
   271  
   272  // execveLibc is non-nil on OS using libc syscall, set to execve in exec_libc.go; this
   273  // avoids a build dependency for other platforms.
   274  var execveLibc func(path uintptr, argv uintptr, envp uintptr) Errno
   275  var execveDarwin func(path *byte, argv **byte, envp **byte) error
   276  var execveOpenBSD func(path *byte, argv **byte, envp **byte) error
   277  
   278  // Exec invokes the execve(2) system call.
   279  func Exec(argv0 string, argv []string, envv []string) (err error) {
   280  	argv0p, err := BytePtrFromString(argv0)
   281  	if err != nil {
   282  		return err
   283  	}
   284  	argvp, err := SlicePtrFromStrings(argv)
   285  	if err != nil {
   286  		return err
   287  	}
   288  	envvp, err := SlicePtrFromStrings(envv)
   289  	if err != nil {
   290  		return err
   291  	}
   292  	runtime_BeforeExec()
   293  
   294  	var err1 error
   295  	if runtime.GOOS == "solaris" || runtime.GOOS == "illumos" || runtime.GOOS == "aix" {
   296  		// RawSyscall should never be used on Solaris, illumos, or AIX.
   297  		err1 = execveLibc(
   298  			uintptr(unsafe.Pointer(argv0p)),
   299  			uintptr(unsafe.Pointer(&argvp[0])),
   300  			uintptr(unsafe.Pointer(&envvp[0])))
   301  	} else if runtime.GOOS == "darwin" || runtime.GOOS == "ios" {
   302  		// Similarly on Darwin.
   303  		err1 = execveDarwin(argv0p, &argvp[0], &envvp[0])
   304  	} else if runtime.GOOS == "openbsd" && (runtime.GOARCH == "386" || runtime.GOARCH == "amd64" || runtime.GOARCH == "arm" || runtime.GOARCH == "arm64") {
   305  		// Similarly on OpenBSD.
   306  		err1 = execveOpenBSD(argv0p, &argvp[0], &envvp[0])
   307  	} else {
   308  		_, _, err1 = RawSyscall(SYS_EXECVE,
   309  			uintptr(unsafe.Pointer(argv0p)),
   310  			uintptr(unsafe.Pointer(&argvp[0])),
   311  			uintptr(unsafe.Pointer(&envvp[0])))
   312  	}
   313  	runtime_AfterExec()
   314  	return err1
   315  }
   316  

View as plain text