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  
     7  package syscall
     8  
     9  import (
    10  	"internal/itoa"
    11  	"runtime"
    12  	"unsafe"
    13  )
    14  
    15  // Linux unshare/clone/clone2/clone3 flags, architecture-independent,
    16  // copied from linux/sched.h.
    17  const (
    18  	CLONE_VM             = 0x00000100 // set if VM shared between processes
    19  	CLONE_FS             = 0x00000200 // set if fs info shared between processes
    20  	CLONE_FILES          = 0x00000400 // set if open files shared between processes
    21  	CLONE_SIGHAND        = 0x00000800 // set if signal handlers and blocked signals shared
    22  	CLONE_PIDFD          = 0x00001000 // set if a pidfd should be placed in parent
    23  	CLONE_PTRACE         = 0x00002000 // set if we want to let tracing continue on the child too
    24  	CLONE_VFORK          = 0x00004000 // set if the parent wants the child to wake it up on mm_release
    25  	CLONE_PARENT         = 0x00008000 // set if we want to have the same parent as the cloner
    26  	CLONE_THREAD         = 0x00010000 // Same thread group?
    27  	CLONE_NEWNS          = 0x00020000 // New mount namespace group
    28  	CLONE_SYSVSEM        = 0x00040000 // share system V SEM_UNDO semantics
    29  	CLONE_SETTLS         = 0x00080000 // create a new TLS for the child
    30  	CLONE_PARENT_SETTID  = 0x00100000 // set the TID in the parent
    31  	CLONE_CHILD_CLEARTID = 0x00200000 // clear the TID in the child
    32  	CLONE_DETACHED       = 0x00400000 // Unused, ignored
    33  	CLONE_UNTRACED       = 0x00800000 // set if the tracing process can't force CLONE_PTRACE on this clone
    34  	CLONE_CHILD_SETTID   = 0x01000000 // set the TID in the child
    35  	CLONE_NEWCGROUP      = 0x02000000 // New cgroup namespace
    36  	CLONE_NEWUTS         = 0x04000000 // New utsname namespace
    37  	CLONE_NEWIPC         = 0x08000000 // New ipc namespace
    38  	CLONE_NEWUSER        = 0x10000000 // New user namespace
    39  	CLONE_NEWPID         = 0x20000000 // New pid namespace
    40  	CLONE_NEWNET         = 0x40000000 // New network namespace
    41  	CLONE_IO             = 0x80000000 // Clone io context
    42  
    43  	// Flags for the clone3() syscall.
    44  
    45  	CLONE_CLEAR_SIGHAND = 0x100000000 // Clear any signal handler and reset to SIG_DFL.
    46  	CLONE_INTO_CGROUP   = 0x200000000 // Clone into a specific cgroup given the right permissions.
    47  
    48  	// Cloning flags intersect with CSIGNAL so can be used with unshare and clone3
    49  	// syscalls only:
    50  
    51  	CLONE_NEWTIME = 0x00000080 // New time namespace
    52  )
    53  
    54  // SysProcIDMap holds Container ID to Host ID mappings used for User Namespaces in Linux.
    55  // See user_namespaces(7).
    56  type SysProcIDMap struct {
    57  	ContainerID int // Container ID.
    58  	HostID      int // Host ID.
    59  	Size        int // Size.
    60  }
    61  
    62  type SysProcAttr struct {
    63  	Chroot     string      // Chroot.
    64  	Credential *Credential // Credential.
    65  	// Ptrace tells the child to call ptrace(PTRACE_TRACEME).
    66  	// Call runtime.LockOSThread before starting a process with this set,
    67  	// and don't call UnlockOSThread until done with PtraceSyscall calls.
    68  	Ptrace bool
    69  	Setsid bool // Create session.
    70  	// Setpgid sets the process group ID of the child to Pgid,
    71  	// or, if Pgid == 0, to the new child's process ID.
    72  	Setpgid bool
    73  	// Setctty sets the controlling terminal of the child to
    74  	// file descriptor Ctty. Ctty must be a descriptor number
    75  	// in the child process: an index into ProcAttr.Files.
    76  	// This is only meaningful if Setsid is true.
    77  	Setctty bool
    78  	Noctty  bool // Detach fd 0 from controlling terminal
    79  	Ctty    int  // Controlling TTY fd
    80  	// Foreground places the child process group in the foreground.
    81  	// This implies Setpgid. The Ctty field must be set to
    82  	// the descriptor of the controlling TTY.
    83  	// Unlike Setctty, in this case Ctty must be a descriptor
    84  	// number in the parent process.
    85  	Foreground bool
    86  	Pgid       int // Child's process group ID if Setpgid.
    87  	// Pdeathsig, if non-zero, is a signal that the kernel will send to
    88  	// the child process when the creating thread dies. Note that the signal
    89  	// is sent on thread termination, which may happen before process termination.
    90  	// There are more details at https://go.dev/issue/27505.
    91  	Pdeathsig    Signal
    92  	Cloneflags   uintptr        // Flags for clone calls (Linux only)
    93  	Unshareflags uintptr        // Flags for unshare calls (Linux only)
    94  	UidMappings  []SysProcIDMap // User ID mappings for user namespaces.
    95  	GidMappings  []SysProcIDMap // Group ID mappings for user namespaces.
    96  	// GidMappingsEnableSetgroups enabling setgroups syscall.
    97  	// If false, then setgroups syscall will be disabled for the child process.
    98  	// This parameter is no-op if GidMappings == nil. Otherwise for unprivileged
    99  	// users this should be set to false for mappings work.
   100  	GidMappingsEnableSetgroups bool
   101  	AmbientCaps                []uintptr // Ambient capabilities (Linux only)
   102  	UseCgroupFD                bool      // Whether to make use of the CgroupFD field.
   103  	CgroupFD                   int       // File descriptor of a cgroup to put the new process into.
   104  }
   105  
   106  var (
   107  	none  = [...]byte{'n', 'o', 'n', 'e', 0}
   108  	slash = [...]byte{'/', 0}
   109  )
   110  
   111  // Implemented in runtime package.
   112  func runtime_BeforeFork()
   113  func runtime_AfterFork()
   114  func runtime_AfterForkInChild()
   115  
   116  // Fork, dup fd onto 0..len(fd), and exec(argv0, argvv, envv) in child.
   117  // If a dup or exec fails, write the errno error to pipe.
   118  // (Pipe is close-on-exec so if exec succeeds, it will be closed.)
   119  // In the child, this function must not acquire any locks, because
   120  // they might have been locked at the time of the fork. This means
   121  // no rescheduling, no malloc calls, and no new stack segments.
   122  // For the same reason compiler does not race instrument it.
   123  // The calls to RawSyscall are okay because they are assembly
   124  // functions that do not grow the stack.
   125  //
   126  //go:norace
   127  func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr *ProcAttr, sys *SysProcAttr, pipe int) (pid int, err Errno) {
   128  	// Set up and fork. This returns immediately in the parent or
   129  	// if there's an error.
   130  	r1, err1, p, locked := forkAndExecInChild1(argv0, argv, envv, chroot, dir, attr, sys, pipe)
   131  	if locked {
   132  		runtime_AfterFork()
   133  	}
   134  	if err1 != 0 {
   135  		return 0, err1
   136  	}
   137  
   138  	// parent; return PID
   139  	pid = int(r1)
   140  
   141  	if sys.UidMappings != nil || sys.GidMappings != nil {
   142  		Close(p[0])
   143  		var err2 Errno
   144  		// uid/gid mappings will be written after fork and unshare(2) for user
   145  		// namespaces.
   146  		if sys.Unshareflags&CLONE_NEWUSER == 0 {
   147  			if err := writeUidGidMappings(pid, sys); err != nil {
   148  				err2 = err.(Errno)
   149  			}
   150  		}
   151  		RawSyscall(SYS_WRITE, uintptr(p[1]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
   152  		Close(p[1])
   153  	}
   154  
   155  	return pid, 0
   156  }
   157  
   158  const _LINUX_CAPABILITY_VERSION_3 = 0x20080522
   159  
   160  type capHeader struct {
   161  	version uint32
   162  	pid     int32
   163  }
   164  
   165  type capData struct {
   166  	effective   uint32
   167  	permitted   uint32
   168  	inheritable uint32
   169  }
   170  type caps struct {
   171  	hdr  capHeader
   172  	data [2]capData
   173  }
   174  
   175  // See CAP_TO_INDEX in linux/capability.h:
   176  func capToIndex(cap uintptr) uintptr { return cap >> 5 }
   177  
   178  // See CAP_TO_MASK in linux/capability.h:
   179  func capToMask(cap uintptr) uint32 { return 1 << uint(cap&31) }
   180  
   181  // cloneArgs holds arguments for clone3 Linux syscall.
   182  type cloneArgs struct {
   183  	flags      uint64 // Flags bit mask
   184  	pidFD      uint64 // Where to store PID file descriptor (int *)
   185  	childTID   uint64 // Where to store child TID, in child's memory (pid_t *)
   186  	parentTID  uint64 // Where to store child TID, in parent's memory (pid_t *)
   187  	exitSignal uint64 // Signal to deliver to parent on child termination
   188  	stack      uint64 // Pointer to lowest byte of stack
   189  	stackSize  uint64 // Size of stack
   190  	tls        uint64 // Location of new TLS
   191  	setTID     uint64 // Pointer to a pid_t array (since Linux 5.5)
   192  	setTIDSize uint64 // Number of elements in set_tid (since Linux 5.5)
   193  	cgroup     uint64 // File descriptor for target cgroup of child (since Linux 5.7)
   194  }
   195  
   196  // forkAndExecInChild1 implements the body of forkAndExecInChild up to
   197  // the parent's post-fork path. This is a separate function so we can
   198  // separate the child's and parent's stack frames if we're using
   199  // vfork.
   200  //
   201  // This is go:noinline because the point is to keep the stack frames
   202  // of this and forkAndExecInChild separate.
   203  //
   204  //go:noinline
   205  //go:norace
   206  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) {
   207  	// Defined in linux/prctl.h starting with Linux 4.3.
   208  	const (
   209  		PR_CAP_AMBIENT       = 0x2f
   210  		PR_CAP_AMBIENT_RAISE = 0x2
   211  	)
   212  
   213  	// vfork requires that the child not touch any of the parent's
   214  	// active stack frames. Hence, the child does all post-fork
   215  	// processing in this stack frame and never returns, while the
   216  	// parent returns immediately from this frame and does all
   217  	// post-fork processing in the outer frame.
   218  	// Declare all variables at top in case any
   219  	// declarations require heap allocation (e.g., err1).
   220  	var (
   221  		err2                      Errno
   222  		nextfd                    int
   223  		i                         int
   224  		caps                      caps
   225  		fd1, flags                uintptr
   226  		puid, psetgroups, pgid    []byte
   227  		uidmap, setgroups, gidmap []byte
   228  		clone3                    *cloneArgs
   229  	)
   230  
   231  	if sys.UidMappings != nil {
   232  		puid = []byte("/proc/self/uid_map\000")
   233  		uidmap = formatIDMappings(sys.UidMappings)
   234  	}
   235  
   236  	if sys.GidMappings != nil {
   237  		psetgroups = []byte("/proc/self/setgroups\000")
   238  		pgid = []byte("/proc/self/gid_map\000")
   239  
   240  		if sys.GidMappingsEnableSetgroups {
   241  			setgroups = []byte("allow\000")
   242  		} else {
   243  			setgroups = []byte("deny\000")
   244  		}
   245  		gidmap = formatIDMappings(sys.GidMappings)
   246  	}
   247  
   248  	// Record parent PID so child can test if it has died.
   249  	ppid, _ := rawSyscallNoError(SYS_GETPID, 0, 0, 0)
   250  
   251  	// Guard against side effects of shuffling fds below.
   252  	// Make sure that nextfd is beyond any currently open files so
   253  	// that we can't run the risk of overwriting any of them.
   254  	fd := make([]int, len(attr.Files))
   255  	nextfd = len(attr.Files)
   256  	for i, ufd := range attr.Files {
   257  		if nextfd < int(ufd) {
   258  			nextfd = int(ufd)
   259  		}
   260  		fd[i] = int(ufd)
   261  	}
   262  	nextfd++
   263  
   264  	// Allocate another pipe for parent to child communication for
   265  	// synchronizing writing of User ID/Group ID mappings.
   266  	if sys.UidMappings != nil || sys.GidMappings != nil {
   267  		if err := forkExecPipe(p[:]); err != nil {
   268  			err1 = err.(Errno)
   269  			return
   270  		}
   271  	}
   272  
   273  	flags = sys.Cloneflags
   274  	if sys.Cloneflags&CLONE_NEWUSER == 0 && sys.Unshareflags&CLONE_NEWUSER == 0 {
   275  		flags |= CLONE_VFORK | CLONE_VM
   276  	}
   277  	// Whether to use clone3.
   278  	if sys.UseCgroupFD {
   279  		clone3 = &cloneArgs{
   280  			flags:      uint64(flags) | CLONE_INTO_CGROUP,
   281  			exitSignal: uint64(SIGCHLD),
   282  			cgroup:     uint64(sys.CgroupFD),
   283  		}
   284  	}
   285  
   286  	// About to call fork.
   287  	// No more allocation or calls of non-assembly functions.
   288  	runtime_BeforeFork()
   289  	locked = true
   290  	if clone3 != nil {
   291  		r1, err1 = rawVforkSyscall(_SYS_clone3, uintptr(unsafe.Pointer(clone3)), unsafe.Sizeof(*clone3))
   292  	} else {
   293  		flags |= uintptr(SIGCHLD)
   294  		if runtime.GOARCH == "s390x" {
   295  			// On Linux/s390, the first two arguments of clone(2) are swapped.
   296  			r1, err1 = rawVforkSyscall(SYS_CLONE, 0, flags)
   297  		} else {
   298  			r1, err1 = rawVforkSyscall(SYS_CLONE, flags, 0)
   299  		}
   300  	}
   301  	if err1 != 0 || r1 != 0 {
   302  		// If we're in the parent, we must return immediately
   303  		// so we're not in the same stack frame as the child.
   304  		// This can at most use the return PC, which the child
   305  		// will not modify, and the results of
   306  		// rawVforkSyscall, which must have been written after
   307  		// the child was replaced.
   308  		return
   309  	}
   310  
   311  	// Fork succeeded, now in child.
   312  
   313  	// Enable the "keep capabilities" flag to set ambient capabilities later.
   314  	if len(sys.AmbientCaps) > 0 {
   315  		_, _, err1 = RawSyscall6(SYS_PRCTL, PR_SET_KEEPCAPS, 1, 0, 0, 0, 0)
   316  		if err1 != 0 {
   317  			goto childerror
   318  		}
   319  	}
   320  
   321  	// Wait for User ID/Group ID mappings to be written.
   322  	if sys.UidMappings != nil || sys.GidMappings != nil {
   323  		if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(p[1]), 0, 0); err1 != 0 {
   324  			goto childerror
   325  		}
   326  		r1, _, err1 = RawSyscall(SYS_READ, uintptr(p[0]), uintptr(unsafe.Pointer(&err2)), unsafe.Sizeof(err2))
   327  		if err1 != 0 {
   328  			goto childerror
   329  		}
   330  		if r1 != unsafe.Sizeof(err2) {
   331  			err1 = EINVAL
   332  			goto childerror
   333  		}
   334  		if err2 != 0 {
   335  			err1 = err2
   336  			goto childerror
   337  		}
   338  	}
   339  
   340  	// Session ID
   341  	if sys.Setsid {
   342  		_, _, err1 = RawSyscall(SYS_SETSID, 0, 0, 0)
   343  		if err1 != 0 {
   344  			goto childerror
   345  		}
   346  	}
   347  
   348  	// Set process group
   349  	if sys.Setpgid || sys.Foreground {
   350  		// Place child in process group.
   351  		_, _, err1 = RawSyscall(SYS_SETPGID, 0, uintptr(sys.Pgid), 0)
   352  		if err1 != 0 {
   353  			goto childerror
   354  		}
   355  	}
   356  
   357  	if sys.Foreground {
   358  		pgrp := int32(sys.Pgid)
   359  		if pgrp == 0 {
   360  			r1, _ = rawSyscallNoError(SYS_GETPID, 0, 0, 0)
   361  
   362  			pgrp = int32(r1)
   363  		}
   364  
   365  		// Place process group in foreground.
   366  		_, _, err1 = RawSyscall(SYS_IOCTL, uintptr(sys.Ctty), uintptr(TIOCSPGRP), uintptr(unsafe.Pointer(&pgrp)))
   367  		if err1 != 0 {
   368  			goto childerror
   369  		}
   370  	}
   371  
   372  	// Restore the signal mask. We do this after TIOCSPGRP to avoid
   373  	// having the kernel send a SIGTTOU signal to the process group.
   374  	runtime_AfterForkInChild()
   375  
   376  	// Unshare
   377  	if sys.Unshareflags != 0 {
   378  		_, _, err1 = RawSyscall(SYS_UNSHARE, sys.Unshareflags, 0, 0)
   379  		if err1 != 0 {
   380  			goto childerror
   381  		}
   382  
   383  		if sys.Unshareflags&CLONE_NEWUSER != 0 && sys.GidMappings != nil {
   384  			dirfd := int(_AT_FDCWD)
   385  			if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&psetgroups[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
   386  				goto childerror
   387  			}
   388  			r1, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&setgroups[0])), uintptr(len(setgroups)))
   389  			if err1 != 0 {
   390  				goto childerror
   391  			}
   392  			if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
   393  				goto childerror
   394  			}
   395  
   396  			if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&pgid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
   397  				goto childerror
   398  			}
   399  			r1, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&gidmap[0])), uintptr(len(gidmap)))
   400  			if err1 != 0 {
   401  				goto childerror
   402  			}
   403  			if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
   404  				goto childerror
   405  			}
   406  		}
   407  
   408  		if sys.Unshareflags&CLONE_NEWUSER != 0 && sys.UidMappings != nil {
   409  			dirfd := int(_AT_FDCWD)
   410  			if fd1, _, err1 = RawSyscall6(SYS_OPENAT, uintptr(dirfd), uintptr(unsafe.Pointer(&puid[0])), uintptr(O_WRONLY), 0, 0, 0); err1 != 0 {
   411  				goto childerror
   412  			}
   413  			r1, _, err1 = RawSyscall(SYS_WRITE, uintptr(fd1), uintptr(unsafe.Pointer(&uidmap[0])), uintptr(len(uidmap)))
   414  			if err1 != 0 {
   415  				goto childerror
   416  			}
   417  			if _, _, err1 = RawSyscall(SYS_CLOSE, uintptr(fd1), 0, 0); err1 != 0 {
   418  				goto childerror
   419  			}
   420  		}
   421  
   422  		// The unshare system call in Linux doesn't unshare mount points
   423  		// mounted with --shared. Systemd mounts / with --shared. For a
   424  		// long discussion of the pros and cons of this see debian bug 739593.
   425  		// The Go model of unsharing is more like Plan 9, where you ask
   426  		// to unshare and the namespaces are unconditionally unshared.
   427  		// To make this model work we must further mark / as MS_PRIVATE.
   428  		// This is what the standard unshare command does.
   429  		if sys.Unshareflags&CLONE_NEWNS == CLONE_NEWNS {
   430  			_, _, err1 = RawSyscall6(SYS_MOUNT, uintptr(unsafe.Pointer(&none[0])), uintptr(unsafe.Pointer(&slash[0])), 0, MS_REC|MS_PRIVATE, 0, 0)
   431  			if err1 != 0 {
   432  				goto childerror
   433  			}
   434  		}
   435  	}
   436  
   437  	// Chroot
   438  	if chroot != nil {
   439  		_, _, err1 = RawSyscall(SYS_CHROOT, uintptr(unsafe.Pointer(chroot)), 0, 0)
   440  		if err1 != 0 {
   441  			goto childerror
   442  		}
   443  	}
   444  
   445  	// User and groups
   446  	if cred := sys.Credential; cred != nil {
   447  		ngroups := uintptr(len(cred.Groups))
   448  		groups := uintptr(0)
   449  		if ngroups > 0 {
   450  			groups = uintptr(unsafe.Pointer(&cred.Groups[0]))
   451  		}
   452  		if !(sys.GidMappings != nil && !sys.GidMappingsEnableSetgroups && ngroups == 0) && !cred.NoSetGroups {
   453  			_, _, err1 = RawSyscall(_SYS_setgroups, ngroups, groups, 0)
   454  			if err1 != 0 {
   455  				goto childerror
   456  			}
   457  		}
   458  		_, _, err1 = RawSyscall(sys_SETGID, uintptr(cred.Gid), 0, 0)
   459  		if err1 != 0 {
   460  			goto childerror
   461  		}
   462  		_, _, err1 = RawSyscall(sys_SETUID, uintptr(cred.Uid), 0, 0)
   463  		if err1 != 0 {
   464  			goto childerror
   465  		}
   466  	}
   467  
   468  	if len(sys.AmbientCaps) != 0 {
   469  		// Ambient capabilities were added in the 4.3 kernel,
   470  		// so it is safe to always use _LINUX_CAPABILITY_VERSION_3.
   471  		caps.hdr.version = _LINUX_CAPABILITY_VERSION_3
   472  
   473  		if _, _, err1 = RawSyscall(SYS_CAPGET, uintptr(unsafe.Pointer(&caps.hdr)), uintptr(unsafe.Pointer(&caps.data[0])), 0); err1 != 0 {
   474  			goto childerror
   475  		}
   476  
   477  		for _, c := range sys.AmbientCaps {
   478  			// Add the c capability to the permitted and inheritable capability mask,
   479  			// otherwise we will not be able to add it to the ambient capability mask.
   480  			caps.data[capToIndex(c)].permitted |= capToMask(c)
   481  			caps.data[capToIndex(c)].inheritable |= capToMask(c)
   482  		}
   483  
   484  		if _, _, err1 = RawSyscall(SYS_CAPSET, uintptr(unsafe.Pointer(&caps.hdr)), uintptr(unsafe.Pointer(&caps.data[0])), 0); err1 != 0 {
   485  			goto childerror
   486  		}
   487  
   488  		for _, c := range sys.AmbientCaps {
   489  			_, _, err1 = RawSyscall6(SYS_PRCTL, PR_CAP_AMBIENT, uintptr(PR_CAP_AMBIENT_RAISE), c, 0, 0, 0)
   490  			if err1 != 0 {
   491  				goto childerror
   492  			}
   493  		}
   494  	}
   495  
   496  	// Chdir
   497  	if dir != nil {
   498  		_, _, err1 = RawSyscall(SYS_CHDIR, uintptr(unsafe.Pointer(dir)), 0, 0)
   499  		if err1 != 0 {
   500  			goto childerror
   501  		}
   502  	}
   503  
   504  	// Parent death signal
   505  	if sys.Pdeathsig != 0 {
   506  		_, _, err1 = RawSyscall6(SYS_PRCTL, PR_SET_PDEATHSIG, uintptr(sys.Pdeathsig), 0, 0, 0, 0)
   507  		if err1 != 0 {
   508  			goto childerror
   509  		}
   510  
   511  		// Signal self if parent is already dead. This might cause a
   512  		// duplicate signal in rare cases, but it won't matter when
   513  		// using SIGKILL.
   514  		r1, _ = rawSyscallNoError(SYS_GETPPID, 0, 0, 0)
   515  		if r1 != ppid {
   516  			pid, _ := rawSyscallNoError(SYS_GETPID, 0, 0, 0)
   517  			_, _, err1 = RawSyscall(SYS_KILL, pid, uintptr(sys.Pdeathsig), 0)
   518  			if err1 != 0 {
   519  				goto childerror
   520  			}
   521  		}
   522  	}
   523  
   524  	// Pass 1: look for fd[i] < i and move those up above len(fd)
   525  	// so that pass 2 won't stomp on an fd it needs later.
   526  	if pipe < nextfd {
   527  		_, _, err1 = RawSyscall(SYS_DUP3, uintptr(pipe), uintptr(nextfd), O_CLOEXEC)
   528  		if err1 != 0 {
   529  			goto childerror
   530  		}
   531  		pipe = nextfd
   532  		nextfd++
   533  	}
   534  	for i = 0; i < len(fd); i++ {
   535  		if fd[i] >= 0 && fd[i] < i {
   536  			if nextfd == pipe { // don't stomp on pipe
   537  				nextfd++
   538  			}
   539  			_, _, err1 = RawSyscall(SYS_DUP3, uintptr(fd[i]), uintptr(nextfd), O_CLOEXEC)
   540  			if err1 != 0 {
   541  				goto childerror
   542  			}
   543  			fd[i] = nextfd
   544  			nextfd++
   545  		}
   546  	}
   547  
   548  	// Pass 2: dup fd[i] down onto i.
   549  	for i = 0; i < len(fd); i++ {
   550  		if fd[i] == -1 {
   551  			RawSyscall(SYS_CLOSE, uintptr(i), 0, 0)
   552  			continue
   553  		}
   554  		if fd[i] == i {
   555  			// dup2(i, i) won't clear close-on-exec flag on Linux,
   556  			// probably not elsewhere either.
   557  			_, _, err1 = RawSyscall(fcntl64Syscall, uintptr(fd[i]), F_SETFD, 0)
   558  			if err1 != 0 {
   559  				goto childerror
   560  			}
   561  			continue
   562  		}
   563  		// The new fd is created NOT close-on-exec,
   564  		// which is exactly what we want.
   565  		_, _, err1 = RawSyscall(SYS_DUP3, uintptr(fd[i]), uintptr(i), 0)
   566  		if err1 != 0 {
   567  			goto childerror
   568  		}
   569  	}
   570  
   571  	// By convention, we don't close-on-exec the fds we are
   572  	// started with, so if len(fd) < 3, close 0, 1, 2 as needed.
   573  	// Programs that know they inherit fds >= 3 will need
   574  	// to set them close-on-exec.
   575  	for i = len(fd); i < 3; i++ {
   576  		RawSyscall(SYS_CLOSE, uintptr(i), 0, 0)
   577  	}
   578  
   579  	// Detach fd 0 from tty
   580  	if sys.Noctty {
   581  		_, _, err1 = RawSyscall(SYS_IOCTL, 0, uintptr(TIOCNOTTY), 0)
   582  		if err1 != 0 {
   583  			goto childerror
   584  		}
   585  	}
   586  
   587  	// Set the controlling TTY to Ctty
   588  	if sys.Setctty {
   589  		_, _, err1 = RawSyscall(SYS_IOCTL, uintptr(sys.Ctty), uintptr(TIOCSCTTY), 1)
   590  		if err1 != 0 {
   591  			goto childerror
   592  		}
   593  	}
   594  
   595  	// Enable tracing if requested.
   596  	// Do this right before exec so that we don't unnecessarily trace the runtime
   597  	// setting up after the fork. See issue #21428.
   598  	if sys.Ptrace {
   599  		_, _, err1 = RawSyscall(SYS_PTRACE, uintptr(PTRACE_TRACEME), 0, 0)
   600  		if err1 != 0 {
   601  			goto childerror
   602  		}
   603  	}
   604  
   605  	// Time to exec.
   606  	_, _, err1 = RawSyscall(SYS_EXECVE,
   607  		uintptr(unsafe.Pointer(argv0)),
   608  		uintptr(unsafe.Pointer(&argv[0])),
   609  		uintptr(unsafe.Pointer(&envv[0])))
   610  
   611  childerror:
   612  	// send error code on pipe
   613  	RawSyscall(SYS_WRITE, uintptr(pipe), uintptr(unsafe.Pointer(&err1)), unsafe.Sizeof(err1))
   614  	for {
   615  		RawSyscall(SYS_EXIT, 253, 0, 0)
   616  	}
   617  }
   618  
   619  // Try to open a pipe with O_CLOEXEC set on both file descriptors.
   620  func forkExecPipe(p []int) (err error) {
   621  	return Pipe2(p, O_CLOEXEC)
   622  }
   623  
   624  func formatIDMappings(idMap []SysProcIDMap) []byte {
   625  	var data []byte
   626  	for _, im := range idMap {
   627  		data = append(data, itoa.Itoa(im.ContainerID)+" "+itoa.Itoa(im.HostID)+" "+itoa.Itoa(im.Size)+"\n"...)
   628  	}
   629  	return data
   630  }
   631  
   632  // writeIDMappings writes the user namespace User ID or Group ID mappings to the specified path.
   633  func writeIDMappings(path string, idMap []SysProcIDMap) error {
   634  	fd, err := Open(path, O_RDWR, 0)
   635  	if err != nil {
   636  		return err
   637  	}
   638  
   639  	if _, err := Write(fd, formatIDMappings(idMap)); err != nil {
   640  		Close(fd)
   641  		return err
   642  	}
   643  
   644  	if err := Close(fd); err != nil {
   645  		return err
   646  	}
   647  
   648  	return nil
   649  }
   650  
   651  // writeSetgroups writes to /proc/PID/setgroups "deny" if enable is false
   652  // and "allow" if enable is true.
   653  // This is needed since kernel 3.19, because you can't write gid_map without
   654  // disabling setgroups() system call.
   655  func writeSetgroups(pid int, enable bool) error {
   656  	sgf := "/proc/" + itoa.Itoa(pid) + "/setgroups"
   657  	fd, err := Open(sgf, O_RDWR, 0)
   658  	if err != nil {
   659  		return err
   660  	}
   661  
   662  	var data []byte
   663  	if enable {
   664  		data = []byte("allow")
   665  	} else {
   666  		data = []byte("deny")
   667  	}
   668  
   669  	if _, err := Write(fd, data); err != nil {
   670  		Close(fd)
   671  		return err
   672  	}
   673  
   674  	return Close(fd)
   675  }
   676  
   677  // writeUidGidMappings writes User ID and Group ID mappings for user namespaces
   678  // for a process and it is called from the parent process.
   679  func writeUidGidMappings(pid int, sys *SysProcAttr) error {
   680  	if sys.UidMappings != nil {
   681  		uidf := "/proc/" + itoa.Itoa(pid) + "/uid_map"
   682  		if err := writeIDMappings(uidf, sys.UidMappings); err != nil {
   683  			return err
   684  		}
   685  	}
   686  
   687  	if sys.GidMappings != nil {
   688  		// If the kernel is too old to support /proc/PID/setgroups, writeSetGroups will return ENOENT; this is OK.
   689  		if err := writeSetgroups(pid, sys.GidMappingsEnableSetgroups); err != nil && err != ENOENT {
   690  			return err
   691  		}
   692  		gidf := "/proc/" + itoa.Itoa(pid) + "/gid_map"
   693  		if err := writeIDMappings(gidf, sys.GidMappings); err != nil {
   694  			return err
   695  		}
   696  	}
   697  
   698  	return nil
   699  }
   700  

View as plain text