Source file src/syscall/exec_linux.go

     1  // Copyright 2011 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 linux
     6  // +build linux
     7  
     8  package syscall
     9  
    10  import (
    11  	"internal/itoa"
    12  	"runtime"
    13  	"unsafe"
    14  )
    15  
    16  // SysProcIDMap holds Container ID to Host ID mappings used for User Namespaces in Linux.
    17  // See user_namespaces(7).
    18  type SysProcIDMap struct {
    19  	ContainerID int // Container ID.
    20  	HostID      int // Host ID.
    21  	Size        int // Size.
    22  }
    23  
    24  type SysProcAttr struct {
    25  	Chroot     string      // Chroot.
    26  	Credential *Credential // Credential.
    27  	// Ptrace tells the child to call ptrace(PTRACE_TRACEME).
    28  	// Call runtime.LockOSThread before starting a process with this set,
    29  	// and don't call UnlockOSThread until done with PtraceSyscall calls.
    30  	Ptrace bool
    31  	Setsid bool // Create session.
    32  	// Setpgid sets the process group ID of the child to Pgid,
    33  	// or, if Pgid == 0, to the new child's process ID.
    34  	Setpgid bool
    35  	// Setctty sets the controlling terminal of the child to
    36  	// file descriptor Ctty. Ctty must be a descriptor number
    37  	// in the child process: an index into ProcAttr.Files.
    38  	// This is only meaningful if Setsid is true.
    39  	Setctty bool
    40  	Noctty  bool // Detach fd 0 from controlling terminal
    41  	Ctty    int  // Controlling TTY fd
    42  	// Foreground places the child process group in the foreground.
    43  	// This implies Setpgid. The Ctty field must be set to
    44  	// the descriptor of the controlling TTY.
    45  	// Unlike Setctty, in this case Ctty must be a descriptor
    46  	// number in the parent process.
    47  	Foreground   bool
    48  	Pgid         int            // Child's process group ID if Setpgid.
    49  	Pdeathsig    Signal         // Signal that the process will get when its parent dies (Linux only)
    50  	Cloneflags   uintptr        // Flags for clone calls (Linux only)
    51  	Unshareflags uintptr        // Flags for unshare calls (Linux only)
    52  	UidMappings  []SysProcIDMap // User ID mappings for user namespaces.
    53  	GidMappings  []SysProcIDMap // Group ID mappings for user namespaces.
    54  	// GidMappingsEnableSetgroups enabling setgroups syscall.
    55  	// If false, then setgroups syscall will be disabled for the child process.
    56  	// This parameter is no-op if GidMappings == nil. Otherwise for unprivileged
    57  	// users this should be set to false for mappings work.
    58  	GidMappingsEnableSetgroups bool
    59  	AmbientCaps                []uintptr // Ambient capabilities (Linux only)
    60  }
    61  
    62  var (
    63  	none  = [...]byte{'n', 'o', 'n', 'e', 0}
    64  	slash = [...]byte{'/', 0}
    65  )
    66  
    67  // Implemented in runtime package.
    68  func runtime_BeforeFork()
    69  func runtime_AfterFork()
    70  func runtime_AfterForkInChild()
    71  
    72  // Fork, dup fd onto 0..len(fd), and exec(argv0, argvv, envv) in child.
    73  // If a dup or exec fails, write the errno error to pipe.
    74  // (Pipe is close-on-exec so if exec succeeds, it will be closed.)
    75  // In the child, this function must not acquire any locks, because
    76  // they might have been locked at the time of the fork. This means
    77  // no rescheduling, no malloc calls, and no new stack segments.
    78  // For the same reason compiler does not race instrument it.
    79  // The calls to RawSyscall are okay because they are assembly
    80  // functions that do not grow the stack.
    81  //go:norace
    82  func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
    83  	// Set up and fork. This returns immediately in the parent or
    84  	// if there's an error.
    85  	r1, err1, p, locked := forkAndExecInChild1(argv0, argv, envv, chroot, dir, attr, sys, pipe)
    86  	if locked {
    87  		runtime_AfterFork()
    88  	}
    89  	if err1 != 0 {
    90  		return 0, err1
    91  	}
    92  
    93  	// parent; return PID
    94  	pid = int(r1)
    95  
    96  	if sys.UidMappings != nil || sys.GidMappings != nil {
    97  		Close(p[0])
    98  		var err2 Errno
    99  		// uid/gid mappings will be written after fork and unshare(2) for user
   100  		// namespaces.
   101  		if sys.Unshareflags&CLONE_NEWUSER == 0 {
   102  			if err := writeUidGidMappings(pid, sys); err != nil {
   103  				err2 = err.(Errno)
   104  			}
   105  		}
   106  		RawSyscall(SYS_WRITE, uintptr(p[1]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
   107  		Close(p[1])
   108  	}
   109  
   110  	return pid, 0
   111  }
   112  
   113  const _LINUX_CAPABILITY_VERSION_3 = 0x20080522
   114  
   115  type capHeader struct {
   116  	version uint32
   117  	pid     int32
   118  }
   119  
   120  type capData struct {
   121  	effective   uint32
   122  	permitted   uint32
   123  	inheritable uint32
   124  }
   125  type caps struct {
   126  	hdr  capHeader
   127  	data [2]capData
   128  }
   129  
   130  // See CAP_TO_INDEX in linux/capability.h:
   131  func capToIndex(cap uintptr) uintptr { return cap >> 5 }
   132  
   133  // See CAP_TO_MASK in linux/capability.h:
   134  func capToMask(cap uintptr) uint32 { return 1 << uint(cap&31) }
   135  
   136  // forkAndExecInChild1 implements the body of forkAndExecInChild up to
   137  // the parent's post-fork path. This is a separate function so we can
   138  // separate the child's and parent's stack frames if we're using
   139  // vfork.
   140  //
   141  // This is go:noinline because the point is to keep the stack frames
   142  // of this and forkAndExecInChild separate.
   143  //
   144  //go:noinline
   145  //go:norace
   146  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) {
   147  	// Defined in linux/prctl.h starting with Linux 4.3.
   148  	const (
   149  		PR_CAP_AMBIENT       = 0x2f
   150  		PR_CAP_AMBIENT_RAISE = 0x2
   151  	)
   152  
   153  	// vfork requires that the child not touch any of the parent's
   154  	// active stack frames. Hence, the child does all post-fork
   155  	// processing in this stack frame and never returns, while the
   156  	// parent returns immediately from this frame and does all
   157  	// post-fork processing in the outer frame.
   158  	// Declare all variables at top in case any
   159  	// declarations require heap allocation (e.g., err1).
   160  	var (
   161  		err2                      Errno
   162  		nextfd                    int
   163  		i                         int
   164  		caps                      caps
   165  		fd1                       uintptr
   166  		puid, psetgroups, pgid    []byte
   167  		uidmap, setgroups, gidmap []byte
   168  	)
   169  
   170  	if sys.UidMappings != nil {
   171  		puid = []byte("/proc/self/uid_map\000")
   172  		uidmap = formatIDMappings(sys.UidMappings)
   173  	}
   174  
   175  	if sys.GidMappings != nil {
   176  		psetgroups = []byte("/proc/self/setgroups\000")
   177  		pgid = []byte("/proc/self/gid_map\000")
   178  
   179  		if sys.GidMappingsEnableSetgroups {
   180  			setgroups = []byte("allow\000")
   181  		} else {
   182  			setgroups = []byte("deny\000")
   183  		}
   184  		gidmap = formatIDMappings(sys.GidMappings)
   185  	}
   186  
   187  	// Record parent PID so child can test if it has died.
   188  	ppid, _ := rawSyscallNoError(SYS_GETPID, 0, 0, 0)
   189  
   190  	// Guard against side effects of shuffling fds below.
   191  	// Make sure that nextfd is beyond any currently open files so
   192  	// that we can't run the risk of overwriting any of them.
   193  	fd := make([]int, len(attr.Files))
   194  	nextfd = len(attr.Files)
   195  	for i, ufd := range attr.Files {
   196  		if nextfd < int(ufd) {
   197  			nextfd = int(ufd)
   198  		}
   199  		fd[i] = int(ufd)
   200  	}
   201  	nextfd++
   202  
   203  	// Allocate another pipe for parent to child communication for
   204  	// synchronizing writing of User ID/Group ID mappings.
   205  	if sys.UidMappings != nil || sys.GidMappings != nil {
   206  		if err := forkExecPipe(p[:]); err != nil {
   207  			err1 = err.(Errno)
   208  			return
   209  		}
   210  	}
   211  
   212  	// About to call fork.
   213  	// No more allocation or calls of non-assembly functions.
   214  	runtime_BeforeFork()
   215  	locked = true
   216  	switch {
   217  	case sys.Cloneflags&CLONE_NEWUSER == 0 && sys.Unshareflags&CLONE_NEWUSER == 0:
   218  		r1, err1 = rawVforkSyscall(SYS_CLONE, uintptr(SIGCHLD|CLONE_VFORK|CLONE_VM)|sys.Cloneflags)
   219  	case runtime.GOARCH == "s390x":
   220  		r1, _, err1 = RawSyscall6(SYS_CLONE, 0, uintptr(SIGCHLD)|sys.Cloneflags, 0, 0, 0, 0)
   221  	default:
   222  		r1, _, err1 = RawSyscall6(SYS_CLONE, uintptr(SIGCHLD)|sys.Cloneflags, 0, 0, 0, 0, 0)
   223  	}
   224  	if err1 != 0 || r1 != 0 {
   225  		// If we're in the parent, we must return immediately
   226  		// so we're not in the same stack frame as the child.
   227  		// This can at most use the return PC, which the child
   228  		// will not modify, and the results of
   229  		// rawVforkSyscall, which must have been written after
   230  		// the child was replaced.
   231  		return
   232  	}
   233  
   234  	// Fork succeeded, now in child.
   235  
   236  	// Enable the "keep capabilities" flag to set ambient capabilities later.
   237  	if len(sys.AmbientCaps) > 0 {
   238  		_, _, err1 = RawSyscall6(SYS_PRCTL, PR_SET_KEEPCAPS, 1, 0, 0, 0, 0)
   239  		if err1 != 0 {
   240  			goto childerror
   241  		}
   242  	}
   243  
   244  	// Wait for User ID/Group ID mappings to be written.
   245  	if sys.UidMappings != nil || sys.GidMappings != nil {
   246  		if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(p[1]), 0, 0); err1 != 0 {
   247  			goto childerror
   248  		}
   249  		r1, _, err1 = RawSyscall(SYS_READ, uintptr(p[0]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
   250  		if err1 != 0 {
   251  			goto childerror
   252  		}
   253  		if r1 != unsafe.Sizeof(err2) {
   254  			err1 = EINVAL
   255  			goto childerror
   256  		}
   257  		if err2 != 0 {
   258  			err1 = err2
   259  			goto childerror
   260  		}
   261  	}
   262  
   263  	// Session ID
   264  	if sys.Setsid {
   265  		_, _, err1 = RawSyscall(SYS_SETSID, 0, 0, 0)
   266  		if err1 != 0 {
   267  			goto childerror
   268  		}
   269  	}
   270  
   271  	// Set process group
   272  	if sys.Setpgid || sys.Foreground {
   273  		// Place child in process group.
   274  		_, _, err1 = RawSyscall(SYS_SETPGID, 0, uintptr(sys.Pgid), 0)
   275  		if err1 != 0 {
   276  			goto childerror
   277  		}
   278  	}
   279  
   280  	if sys.Foreground {
   281  		pgrp := int32(sys.Pgid)
   282  		if pgrp == 0 {
   283  			r1, _ = rawSyscallNoError(SYS_GETPID, 0, 0, 0)
   284  
   285  			pgrp = int32(r1)
   286  		}
   287  
   288  		// Place process group in foreground.
   289  		_, _, err1 = RawSyscall(SYS_IOCTL, uintptr(sys.Ctty), uintptr(TIOCSPGRP), uintptr(unsafe.Pointer(&pgrp)))
   290  		if err1 != 0 {
   291  			goto childerror
   292  		}
   293  	}
   294  
   295  	// Restore the signal mask. We do this after TIOCSPGRP to avoid
   296  	// having the kernel send a SIGTTOU signal to the process group.
   297  	runtime_AfterForkInChild()
   298  
   299  	// Unshare
   300  	if sys.Unshareflags != 0 {
   301  		_, _, err1 = RawSyscall(SYS_UNSHARE, sys.Unshareflags, 0, 0)
   302  		if err1 != 0 {
   303  			goto childerror
   304  		}
   305  
   306  		if sys.Unshareflags&CLONE_NEWUSER != 0 && sys.GidMappings != nil {
   307  			dirfd := int(_AT_FDCWD)
   308  			if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&psetgroups[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
   309  				goto childerror
   310  			}
   311  			r1, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&setgroups[0])), uintptr(len(setgroups)))
   312  			if err1 != 0 {
   313  				goto childerror
   314  			}
   315  			if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
   316  				goto childerror
   317  			}
   318  
   319  			if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&pgid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
   320  				goto childerror
   321  			}
   322  			r1, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&gidmap[0])), uintptr(len(gidmap)))
   323  			if err1 != 0 {
   324  				goto childerror
   325  			}
   326  			if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
   327  				goto childerror
   328  			}
   329  		}
   330  
   331  		if sys.Unshareflags&CLONE_NEWUSER != 0 && sys.UidMappings != nil {
   332  			dirfd := int(_AT_FDCWD)
   333  			if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&puid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
   334  				goto childerror
   335  			}
   336  			r1, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&uidmap[0])), uintptr(len(uidmap)))
   337  			if err1 != 0 {
   338  				goto childerror
   339  			}
   340  			if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
   341  				goto childerror
   342  			}
   343  		}
   344  
   345  		// The unshare system call in Linux doesn't unshare mount points
   346  		// mounted with --shared. Systemd mounts / with --shared. For a
   347  		// long discussion of the pros and cons of this see debian bug 739593.
   348  		// The Go model of unsharing is more like Plan 9, where you ask
   349  		// to unshare and the namespaces are unconditionally unshared.
   350  		// To make this model work we must further mark / as MS_PRIVATE.
   351  		// This is what the standard unshare command does.
   352  		if sys.Unshareflags&CLONE_NEWNS == CLONE_NEWNS {
   353  			_, _, err1 = RawSyscall6(SYS_MOUNT, uintptr(unsafe.Pointer(&none[0])), uintptr(unsafe.Pointer(&slash[0])), 0, MS_REC|MS_PRIVATE, 0, 0)
   354  			if err1 != 0 {
   355  				goto childerror
   356  			}
   357  		}
   358  	}
   359  
   360  	// Chroot
   361  	if chroot != nil {
   362  		_, _, err1 = RawSyscall(SYS_CHROOT, uintptr(unsafe.Pointer(chroot)), 0, 0)
   363  		if err1 != 0 {
   364  			goto childerror
   365  		}
   366  	}
   367  
   368  	// User and groups
   369  	if cred := sys.Credential; cred != nil {
   370  		ngroups := uintptr(len(cred.Groups))
   371  		groups := uintptr(0)
   372  		if ngroups > 0 {
   373  			groups = uintptr(unsafe.Pointer(&cred.Groups[0]))
   374  		}
   375  		if !(sys.GidMappings != nil && !sys.GidMappingsEnableSetgroups && ngroups == 0) && !cred.NoSetGroups {
   376  			_, _, err1 = RawSyscall(_SYS_setgroups, ngroups, groups, 0)
   377  			if err1 != 0 {
   378  				goto childerror
   379  			}
   380  		}
   381  		_, _, err1 = RawSyscall(sys_SETGID, uintptr(cred.Gid), 0, 0)
   382  		if err1 != 0 {
   383  			goto childerror
   384  		}
   385  		_, _, err1 = RawSyscall(sys_SETUID, uintptr(cred.Uid), 0, 0)
   386  		if err1 != 0 {
   387  			goto childerror
   388  		}
   389  	}
   390  
   391  	if len(sys.AmbientCaps) != 0 {
   392  		// Ambient capabilities were added in the 4.3 kernel,
   393  		// so it is safe to always use _LINUX_CAPABILITY_VERSION_3.
   394  		caps.hdr.version = _LINUX_CAPABILITY_VERSION_3
   395  
   396  		if _, _, err1 := RawSyscall(SYS_CAPGET, uintptr(unsafe.Pointer(&caps.hdr)), uintptr(unsafe.Pointer(&caps.data[0])), 0); err1 != 0 {
   397  			goto childerror
   398  		}
   399  
   400  		for _, c := range sys.AmbientCaps {
   401  			// Add the c capability to the permitted and inheritable capability mask,
   402  			// otherwise we will not be able to add it to the ambient capability mask.
   403  			caps.data[capToIndex(c)].permitted |= capToMask(c)
   404  			caps.data[capToIndex(c)].inheritable |= capToMask(c)
   405  		}
   406  
   407  		if _, _, err1 := RawSyscall(SYS_CAPSET, uintptr(unsafe.Pointer(&caps.hdr)), uintptr(unsafe.Pointer(&caps.data[0])), 0); err1 != 0 {
   408  			goto childerror
   409  		}
   410  
   411  		for _, c := range sys.AmbientCaps {
   412  			_, _, err1 = RawSyscall6(SYS_PRCTL, PR_CAP_AMBIENT, uintptr(PR_CAP_AMBIENT_RAISE), c, 0, 0, 0)
   413  			if err1 != 0 {
   414  				goto childerror
   415  			}
   416  		}
   417  	}
   418  
   419  	// Chdir
   420  	if dir != nil {
   421  		_, _, err1 = RawSyscall(SYS_CHDIR, uintptr(unsafe.Pointer(dir)), 0, 0)
   422  		if err1 != 0 {
   423  			goto childerror
   424  		}
   425  	}
   426  
   427  	// Parent death signal
   428  	if sys.Pdeathsig != 0 {
   429  		_, _, err1 = RawSyscall6(SYS_PRCTL, PR_SET_PDEATHSIG, uintptr(sys.Pdeathsig), 0, 0, 0, 0)
   430  		if err1 != 0 {
   431  			goto childerror
   432  		}
   433  
   434  		// Signal self if parent is already dead. This might cause a
   435  		// duplicate signal in rare cases, but it won't matter when
   436  		// using SIGKILL.
   437  		r1, _ = rawSyscallNoError(SYS_GETPPID, 0, 0, 0)
   438  		if r1 != ppid {
   439  			pid, _ := rawSyscallNoError(SYS_GETPID, 0, 0, 0)
   440  			_, _, err1 := RawSyscall(SYS_KILL, pid, uintptr(sys.Pdeathsig), 0)
   441  			if err1 != 0 {
   442  				goto childerror
   443  			}
   444  		}
   445  	}
   446  
   447  	// Pass 1: look for fd[i] < i and move those up above len(fd)
   448  	// so that pass 2 won't stomp on an fd it needs later.
   449  	if pipe < nextfd {
   450  		_, _, err1 = RawSyscall(SYS_DUP3, uintptr(pipe), uintptr(nextfd), O_CLOEXEC)
   451  		if _SYS_dup != SYS_DUP3 && err1 == ENOSYS {
   452  			_, _, err1 = RawSyscall(_SYS_dup, uintptr(pipe), uintptr(nextfd), 0)
   453  			if err1 != 0 {
   454  				goto childerror
   455  			}
   456  			RawSyscall(fcntl64Syscall, uintptr(nextfd), F_SETFD, FD_CLOEXEC)
   457  		} else if err1 != 0 {
   458  			goto childerror
   459  		}
   460  		pipe = nextfd
   461  		nextfd++
   462  	}
   463  	for i = 0; i < len(fd); i++ {
   464  		if fd[i] >= 0 && fd[i] < int(i) {
   465  			if nextfd == pipe { // don't stomp on pipe
   466  				nextfd++
   467  			}
   468  			_, _, err1 = RawSyscall(SYS_DUP3, uintptr(fd[i]), uintptr(nextfd), O_CLOEXEC)
   469  			if _SYS_dup != SYS_DUP3 && err1 == ENOSYS {
   470  				_, _, err1 = RawSyscall(_SYS_dup, uintptr(fd[i]), uintptr(nextfd), 0)
   471  				if err1 != 0 {
   472  					goto childerror
   473  				}
   474  				RawSyscall(fcntl64Syscall, uintptr(nextfd), F_SETFD, FD_CLOEXEC)
   475  			} else if err1 != 0 {
   476  				goto childerror
   477  			}
   478  			fd[i] = nextfd
   479  			nextfd++
   480  		}
   481  	}
   482  
   483  	// Pass 2: dup fd[i] down onto i.
   484  	for i = 0; i < len(fd); i++ {
   485  		if fd[i] == -1 {
   486  			RawSyscall(SYS_CLOSE, uintptr(i), 0, 0)
   487  			continue
   488  		}
   489  		if fd[i] == int(i) {
   490  			// dup2(i, i) won't clear close-on-exec flag on Linux,
   491  			// probably not elsewhere either.
   492  			_, _, err1 = RawSyscall(fcntl64Syscall, uintptr(fd[i]), F_SETFD, 0)
   493  			if err1 != 0 {
   494  				goto childerror
   495  			}
   496  			continue
   497  		}
   498  		// The new fd is created NOT close-on-exec,
   499  		// which is exactly what we want.
   500  		_, _, err1 = RawSyscall(_SYS_dup, uintptr(fd[i]), uintptr(i), 0)
   501  		if err1 != 0 {
   502  			goto childerror
   503  		}
   504  	}
   505  
   506  	// By convention, we don't close-on-exec the fds we are
   507  	// started with, so if len(fd) < 3, close 0, 1, 2 as needed.
   508  	// Programs that know they inherit fds >= 3 will need
   509  	// to set them close-on-exec.
   510  	for i = len(fd); i < 3; i++ {
   511  		RawSyscall(SYS_CLOSE, uintptr(i), 0, 0)
   512  	}
   513  
   514  	// Detach fd 0 from tty
   515  	if sys.Noctty {
   516  		_, _, err1 = RawSyscall(SYS_IOCTL, 0, uintptr(TIOCNOTTY), 0)
   517  		if err1 != 0 {
   518  			goto childerror
   519  		}
   520  	}
   521  
   522  	// Set the controlling TTY to Ctty
   523  	if sys.Setctty {
   524  		_, _, err1 = RawSyscall(SYS_IOCTL, uintptr(sys.Ctty), uintptr(TIOCSCTTY), 1)
   525  		if err1 != 0 {
   526  			goto childerror
   527  		}
   528  	}
   529  
   530  	// Enable tracing if requested.
   531  	// Do this right before exec so that we don't unnecessarily trace the runtime
   532  	// setting up after the fork. See issue #21428.
   533  	if sys.Ptrace {
   534  		_, _, err1 = RawSyscall(SYS_PTRACE, uintptr(PTRACE_TRACEME), 0, 0)
   535  		if err1 != 0 {
   536  			goto childerror
   537  		}
   538  	}
   539  
   540  	// Time to exec.
   541  	_, _, err1 = RawSyscall(SYS_EXECVE,
   542  		uintptr(unsafe.Pointer(argv0)),
   543  		uintptr(unsafe.Pointer(&argv[0])),
   544  		uintptr(unsafe.Pointer(&envv[0])))
   545  
   546  childerror:
   547  	// send error code on pipe
   548  	RawSyscall(SYS_WRITE, uintptr(pipe), uintptr(unsafe.Pointer(&err1)), unsafe.Sizeof(err1))
   549  	for {
   550  		RawSyscall(SYS_EXIT, 253, 0, 0)
   551  	}
   552  }
   553  
   554  // Try to open a pipe with O_CLOEXEC set on both file descriptors.
   555  func forkExecPipe(p []int) (err error) {
   556  	err = Pipe2(p, O_CLOEXEC)
   557  	// pipe2 was added in 2.6.27 and our minimum requirement is 2.6.23, so it
   558  	// might not be implemented.
   559  	if err == ENOSYS {
   560  		if err = Pipe(p); err != nil {
   561  			return
   562  		}
   563  		if _, err = fcntl(p[0], F_SETFD, FD_CLOEXEC); err != nil {
   564  			return
   565  		}
   566  		_, err = fcntl(p[1], F_SETFD, FD_CLOEXEC)
   567  	}
   568  	return
   569  }
   570  
   571  func formatIDMappings(idMap []SysProcIDMap) []byte {
   572  	var data []byte
   573  	for _, im := range idMap {
   574  		data = append(data, []byte(itoa.Itoa(im.ContainerID)+" "+itoa.Itoa(im.HostID)+" "+itoa.Itoa(im.Size)+"\n")...)
   575  	}
   576  	return data
   577  }
   578  
   579  // writeIDMappings writes the user namespace User ID or Group ID mappings to the specified path.
   580  func writeIDMappings(path string, idMap []SysProcIDMap) error {
   581  	fd, err := Open(path, O_RDWR, 0)
   582  	if err != nil {
   583  		return err
   584  	}
   585  
   586  	if _, err := Write(fd, formatIDMappings(idMap)); err != nil {
   587  		Close(fd)
   588  		return err
   589  	}
   590  
   591  	if err := Close(fd); err != nil {
   592  		return err
   593  	}
   594  
   595  	return nil
   596  }
   597  
   598  // writeSetgroups writes to /proc/PID/setgroups "deny" if enable is false
   599  // and "allow" if enable is true.
   600  // This is needed since kernel 3.19, because you can't write gid_map without
   601  // disabling setgroups() system call.
   602  func writeSetgroups(pid int, enable bool) error {
   603  	sgf := "/proc/" + itoa.Itoa(pid) + "/setgroups"
   604  	fd, err := Open(sgf, O_RDWR, 0)
   605  	if err != nil {
   606  		return err
   607  	}
   608  
   609  	var data []byte
   610  	if enable {
   611  		data = []byte("allow")
   612  	} else {
   613  		data = []byte("deny")
   614  	}
   615  
   616  	if _, err := Write(fd, data); err != nil {
   617  		Close(fd)
   618  		return err
   619  	}
   620  
   621  	return Close(fd)
   622  }
   623  
   624  // writeUidGidMappings writes User ID and Group ID mappings for user namespaces
   625  // for a process and it is called from the parent process.
   626  func writeUidGidMappings(pid int, sys *SysProcAttr) error {
   627  	if sys.UidMappings != nil {
   628  		uidf := "/proc/" + itoa.Itoa(pid) + "/uid_map"
   629  		if err := writeIDMappings(uidf, sys.UidMappings); err != nil {
   630  			return err
   631  		}
   632  	}
   633  
   634  	if sys.GidMappings != nil {
   635  		// If the kernel is too old to support /proc/PID/setgroups, writeSetGroups will return ENOENT; this is OK.
   636  		if err := writeSetgroups(pid, sys.GidMappingsEnableSetgroups); err != nil && err != ENOENT {
   637  			return err
   638  		}
   639  		gidf := "/proc/" + itoa.Itoa(pid) + "/gid_map"
   640  		if err := writeIDMappings(gidf, sys.GidMappings); err != nil {
   641  			return err
   642  		}
   643  	}
   644  
   645  	return nil
   646  }
   647  

View as plain text