github.com/zhuohuang-hust/src-cbuild@v0.0.0-20230105071821-c7aab3e7c840/mergeCode/runc/libcontainer/rootfs_linux.go (about)

     1  // +build linux
     2  
     3  package libcontainer
     4  
     5  import (
     6  	"fmt"
     7  	"io"
     8  	"io/ioutil"
     9  	"os"
    10  	"os/exec"
    11  	"path"
    12  	"path/filepath"
    13  	"strings"
    14  	"syscall"
    15  	"time"
    16  
    17  	"github.com/docker/docker/pkg/mount"
    18  	"github.com/docker/docker/pkg/symlink"
    19  	"github.com/mrunalp/fileutils"
    20  	"github.com/opencontainers/runc/libcontainer/cgroups"
    21  	"github.com/opencontainers/runc/libcontainer/configs"
    22  	"github.com/opencontainers/runc/libcontainer/label"
    23  	"github.com/opencontainers/runc/libcontainer/system"
    24  	libcontainerUtils "github.com/opencontainers/runc/libcontainer/utils"
    25  )
    26  
    27  const defaultMountFlags = syscall.MS_NOEXEC | syscall.MS_NOSUID | syscall.MS_NODEV
    28  
    29  // needsSetupDev returns true if /dev needs to be set up.
    30  func needsSetupDev(config *configs.Config) bool {
    31  	for _, m := range config.Mounts {
    32  		if m.Device == "bind" && libcontainerUtils.CleanPath(m.Destination) == "/dev" {
    33  			return false
    34  		}
    35  	}
    36  	return true
    37  }
    38  
    39  // setupRootfs sets up the devices, mount points, and filesystems for use inside a
    40  // new mount namespace.
    41  func setupRootfs(config *configs.Config, console *linuxConsole, pipe io.ReadWriter) (err error) {
    42  	if err := prepareRoot(config); err != nil {
    43  		return newSystemErrorWithCause(err, "preparing rootfs")
    44  	}
    45  
    46  	setupDev := needsSetupDev(config)
    47  	for _, m := range config.Mounts {
    48  		for _, precmd := range m.PremountCmds {
    49  			if err := mountCmd(precmd); err != nil {
    50  				return newSystemErrorWithCause(err, "running premount command")
    51  			}
    52  		}
    53  		if err := mountToRootfs(m, config.Rootfs, config.MountLabel); err != nil {
    54  			return newSystemErrorWithCausef(err, "mounting %q to rootfs %q at %q", m.Source, config.Rootfs, m.Destination)
    55  		}
    56  
    57  		for _, postcmd := range m.PostmountCmds {
    58  			if err := mountCmd(postcmd); err != nil {
    59  				return newSystemErrorWithCause(err, "running postmount command")
    60  			}
    61  		}
    62  	}
    63  	if setupDev {
    64  		if err := createDevices(config); err != nil {
    65  			return newSystemErrorWithCause(err, "creating device nodes")
    66  		}
    67  		if err := setupPtmx(config, console); err != nil {
    68  			return newSystemErrorWithCause(err, "setting up ptmx")
    69  		}
    70  		if err := setupDevSymlinks(config.Rootfs); err != nil {
    71  			return newSystemErrorWithCause(err, "setting up /dev symlinks")
    72  		}
    73  	}
    74  	// Signal the parent to run the pre-start hooks.
    75  	// The hooks are run after the mounts are setup, but before we switch to the new
    76  	// root, so that the old root is still available in the hooks for any mount
    77  	// manipulations.
    78  	if err := syncParentHooks(pipe); err != nil {
    79  		return err
    80  	}
    81  	if err := syscall.Chdir(config.Rootfs); err != nil {
    82  		return newSystemErrorWithCausef(err, "changing dir to %q", config.Rootfs)
    83  	}
    84  	if config.NoPivotRoot {
    85  		err = msMoveRoot(config.Rootfs)
    86  	} else {
    87  		err = pivotRoot(config.Rootfs)
    88  	}
    89  	if err != nil {
    90  		return newSystemErrorWithCause(err, "jailing process inside rootfs")
    91  	}
    92  	if setupDev {
    93  		if err := reOpenDevNull(); err != nil {
    94  			return newSystemErrorWithCause(err, "reopening /dev/null inside container")
    95  		}
    96  	}
    97  	// remount dev as ro if specified
    98  	for _, m := range config.Mounts {
    99  		if libcontainerUtils.CleanPath(m.Destination) == "/dev" {
   100  			if m.Flags&syscall.MS_RDONLY != 0 {
   101  				if err := remountReadonly(m.Destination); err != nil {
   102  					return newSystemErrorWithCausef(err, "remounting %q as readonly", m.Destination)
   103  				}
   104  			}
   105  			break
   106  		}
   107  	}
   108  	// set rootfs ( / ) as readonly
   109  	if config.Readonlyfs {
   110  		if err := setReadonly(); err != nil {
   111  			return newSystemErrorWithCause(err, "setting rootfs as readonly")
   112  		}
   113  	}
   114  	syscall.Umask(0022)
   115  	return nil
   116  }
   117  
   118  func mountCmd(cmd configs.Command) error {
   119  	command := exec.Command(cmd.Path, cmd.Args[:]...)
   120  	command.Env = cmd.Env
   121  	command.Dir = cmd.Dir
   122  	if out, err := command.CombinedOutput(); err != nil {
   123  		return fmt.Errorf("%#v failed: %s: %v", cmd, string(out), err)
   124  	}
   125  	return nil
   126  }
   127  
   128  func mountToRootfs(m *configs.Mount, rootfs, mountLabel string) error {
   129  	var (
   130  		dest = m.Destination
   131  	)
   132  	if !strings.HasPrefix(dest, rootfs) {
   133  		dest = filepath.Join(rootfs, dest)
   134  	}
   135  
   136  	switch m.Device {
   137  	case "proc", "sysfs":
   138  		if err := os.MkdirAll(dest, 0755); err != nil {
   139  			return err
   140  		}
   141  		// Selinux kernels do not support labeling of /proc or /sys
   142  		return mountPropagate(m, rootfs, "")
   143  	case "mqueue":
   144  		if err := os.MkdirAll(dest, 0755); err != nil {
   145  			return err
   146  		}
   147  		if err := mountPropagate(m, rootfs, mountLabel); err != nil {
   148  			// older kernels do not support labeling of /dev/mqueue
   149  			if err := mountPropagate(m, rootfs, ""); err != nil {
   150  				return err
   151  			}
   152  			return label.SetFileLabel(dest, mountLabel)
   153  		}
   154  		return nil
   155  	case "tmpfs":
   156  		copyUp := m.Extensions&configs.EXT_COPYUP == configs.EXT_COPYUP
   157  		tmpDir := ""
   158  		stat, err := os.Stat(dest)
   159  		if err != nil {
   160  			if err := os.MkdirAll(dest, 0755); err != nil {
   161  				return err
   162  			}
   163  		}
   164  		if copyUp {
   165  			tmpDir, err = ioutil.TempDir("/tmp", "runctmpdir")
   166  			if err != nil {
   167  				return newSystemErrorWithCause(err, "tmpcopyup: failed to create tmpdir")
   168  			}
   169  			defer os.RemoveAll(tmpDir)
   170  			m.Destination = tmpDir
   171  		}
   172  		if err := mountPropagate(m, rootfs, mountLabel); err != nil {
   173  			return err
   174  		}
   175  		if copyUp {
   176  			if err := fileutils.CopyDirectory(dest, tmpDir); err != nil {
   177  				errMsg := fmt.Errorf("tmpcopyup: failed to copy %s to %s: %v", dest, tmpDir, err)
   178  				if err1 := syscall.Unmount(tmpDir, syscall.MNT_DETACH); err1 != nil {
   179  					return newSystemErrorWithCausef(err1, "tmpcopyup: %v: failed to unmount", errMsg)
   180  				}
   181  				return errMsg
   182  			}
   183  			if err := syscall.Mount(tmpDir, dest, "", syscall.MS_MOVE, ""); err != nil {
   184  				errMsg := fmt.Errorf("tmpcopyup: failed to move mount %s to %s: %v", tmpDir, dest, err)
   185  				if err1 := syscall.Unmount(tmpDir, syscall.MNT_DETACH); err1 != nil {
   186  					return newSystemErrorWithCausef(err1, "tmpcopyup: %v: failed to unmount", errMsg)
   187  				}
   188  				return errMsg
   189  			}
   190  		}
   191  		if stat != nil {
   192  			if err = os.Chmod(dest, stat.Mode()); err != nil {
   193  				return err
   194  			}
   195  		}
   196  		return nil
   197  	case "bind":
   198  		stat, err := os.Stat(m.Source)
   199  		if err != nil {
   200  			// error out if the source of a bind mount does not exist as we will be
   201  			// unable to bind anything to it.
   202  			return err
   203  		}
   204  		// ensure that the destination of the bind mount is resolved of symlinks at mount time because
   205  		// any previous mounts can invalidate the next mount's destination.
   206  		// this can happen when a user specifies mounts within other mounts to cause breakouts or other
   207  		// evil stuff to try to escape the container's rootfs.
   208  		if dest, err = symlink.FollowSymlinkInScope(filepath.Join(rootfs, m.Destination), rootfs); err != nil {
   209  			return err
   210  		}
   211  		if err := checkMountDestination(rootfs, dest); err != nil {
   212  			return err
   213  		}
   214  		// update the mount with the correct dest after symlinks are resolved.
   215  		m.Destination = dest
   216  		if err := createIfNotExists(dest, stat.IsDir()); err != nil {
   217  			return err
   218  		}
   219  		if err := mountPropagate(m, rootfs, mountLabel); err != nil {
   220  			return err
   221  		}
   222  		// bind mount won't change mount options, we need remount to make mount options effective.
   223  		// first check that we have non-default options required before attempting a remount
   224  		if m.Flags&^(syscall.MS_REC|syscall.MS_REMOUNT|syscall.MS_BIND) != 0 {
   225  			// only remount if unique mount options are set
   226  			if err := remount(m, rootfs); err != nil {
   227  				return err
   228  			}
   229  		}
   230  
   231  		if m.Relabel != "" {
   232  			if err := label.Validate(m.Relabel); err != nil {
   233  				return err
   234  			}
   235  			shared := label.IsShared(m.Relabel)
   236  			if err := label.Relabel(m.Source, mountLabel, shared); err != nil {
   237  				return err
   238  			}
   239  		}
   240  	case "cgroup":
   241  		binds, err := getCgroupMounts(m)
   242  		if err != nil {
   243  			return err
   244  		}
   245  		var merged []string
   246  		for _, b := range binds {
   247  			ss := filepath.Base(b.Destination)
   248  			if strings.Contains(ss, ",") {
   249  				merged = append(merged, ss)
   250  			}
   251  		}
   252  		tmpfs := &configs.Mount{
   253  			Source:           "tmpfs",
   254  			Device:           "tmpfs",
   255  			Destination:      m.Destination,
   256  			Flags:            defaultMountFlags,
   257  			Data:             "mode=755",
   258  			PropagationFlags: m.PropagationFlags,
   259  		}
   260  		if err := mountToRootfs(tmpfs, rootfs, mountLabel); err != nil {
   261  			return err
   262  		}
   263  		for _, b := range binds {
   264  			if err := mountToRootfs(b, rootfs, mountLabel); err != nil {
   265  				return err
   266  			}
   267  		}
   268  		for _, mc := range merged {
   269  			for _, ss := range strings.Split(mc, ",") {
   270  				// symlink(2) is very dumb, it will just shove the path into
   271  				// the link and doesn't do any checks or relative path
   272  				// conversion. Also, don't error out if the cgroup already exists.
   273  				if err := os.Symlink(mc, filepath.Join(rootfs, m.Destination, ss)); err != nil && !os.IsExist(err) {
   274  					return err
   275  				}
   276  			}
   277  		}
   278  		if m.Flags&syscall.MS_RDONLY != 0 {
   279  			// remount cgroup root as readonly
   280  			mcgrouproot := &configs.Mount{
   281  				Source:      m.Destination,
   282  				Device:      "bind",
   283  				Destination: m.Destination,
   284  				Flags:       defaultMountFlags | syscall.MS_RDONLY | syscall.MS_BIND,
   285  			}
   286  			if err := remount(mcgrouproot, rootfs); err != nil {
   287  				return err
   288  			}
   289  		}
   290  	default:
   291  		if err := os.MkdirAll(dest, 0755); err != nil {
   292  			return err
   293  		}
   294  		return mountPropagate(m, rootfs, mountLabel)
   295  	}
   296  	return nil
   297  }
   298  
   299  func getCgroupMounts(m *configs.Mount) ([]*configs.Mount, error) {
   300  	mounts, err := cgroups.GetCgroupMounts(false)
   301  	if err != nil {
   302  		return nil, err
   303  	}
   304  
   305  	cgroupPaths, err := cgroups.ParseCgroupFile("/proc/self/cgroup")
   306  	if err != nil {
   307  		return nil, err
   308  	}
   309  
   310  	var binds []*configs.Mount
   311  
   312  	for _, mm := range mounts {
   313  		dir, err := mm.GetThisCgroupDir(cgroupPaths)
   314  		if err != nil {
   315  			return nil, err
   316  		}
   317  		relDir, err := filepath.Rel(mm.Root, dir)
   318  		if err != nil {
   319  			return nil, err
   320  		}
   321  		binds = append(binds, &configs.Mount{
   322  			Device:           "bind",
   323  			Source:           filepath.Join(mm.Mountpoint, relDir),
   324  			Destination:      filepath.Join(m.Destination, strings.Join(mm.Subsystems, ",")),
   325  			Flags:            syscall.MS_BIND | syscall.MS_REC | m.Flags,
   326  			PropagationFlags: m.PropagationFlags,
   327  		})
   328  	}
   329  
   330  	return binds, nil
   331  }
   332  
   333  // checkMountDestination checks to ensure that the mount destination is not over the top of /proc.
   334  // dest is required to be an abs path and have any symlinks resolved before calling this function.
   335  func checkMountDestination(rootfs, dest string) error {
   336  	invalidDestinations := []string{
   337  		"/proc",
   338  	}
   339  	// White list, it should be sub directories of invalid destinations
   340  	validDestinations := []string{
   341  		// These entries can be bind mounted by files emulated by fuse,
   342  		// so commands like top, free displays stats in container.
   343  		"/proc/cpuinfo",
   344  		"/proc/diskstats",
   345  		"/proc/meminfo",
   346  		"/proc/stat",
   347  		"/proc/swaps",
   348  		"/proc/uptime",
   349  		"/proc/net/dev",
   350  	}
   351  	for _, valid := range validDestinations {
   352  		path, err := filepath.Rel(filepath.Join(rootfs, valid), dest)
   353  		if err != nil {
   354  			return err
   355  		}
   356  		if path == "." {
   357  			return nil
   358  		}
   359  	}
   360  	for _, invalid := range invalidDestinations {
   361  		path, err := filepath.Rel(filepath.Join(rootfs, invalid), dest)
   362  		if err != nil {
   363  			return err
   364  		}
   365  		if path == "." || !strings.HasPrefix(path, "..") {
   366  			return fmt.Errorf("%q cannot be mounted because it is located inside %q", dest, invalid)
   367  		}
   368  	}
   369  	return nil
   370  }
   371  
   372  func setupDevSymlinks(rootfs string) error {
   373  	var links = [][2]string{
   374  		{"/proc/self/fd", "/dev/fd"},
   375  		{"/proc/self/fd/0", "/dev/stdin"},
   376  		{"/proc/self/fd/1", "/dev/stdout"},
   377  		{"/proc/self/fd/2", "/dev/stderr"},
   378  	}
   379  	// kcore support can be toggled with CONFIG_PROC_KCORE; only create a symlink
   380  	// in /dev if it exists in /proc.
   381  	if _, err := os.Stat("/proc/kcore"); err == nil {
   382  		links = append(links, [2]string{"/proc/kcore", "/dev/core"})
   383  	}
   384  	for _, link := range links {
   385  		var (
   386  			src = link[0]
   387  			dst = filepath.Join(rootfs, link[1])
   388  		)
   389  		if err := os.Symlink(src, dst); err != nil && !os.IsExist(err) {
   390  			return fmt.Errorf("symlink %s %s %s", src, dst, err)
   391  		}
   392  	}
   393  	return nil
   394  }
   395  
   396  // If stdin, stdout, and/or stderr are pointing to `/dev/null` in the parent's rootfs
   397  // this method will make them point to `/dev/null` in this container's rootfs.  This
   398  // needs to be called after we chroot/pivot into the container's rootfs so that any
   399  // symlinks are resolved locally.
   400  func reOpenDevNull() error {
   401  	var stat, devNullStat syscall.Stat_t
   402  	file, err := os.OpenFile("/dev/null", os.O_RDWR, 0)
   403  	if err != nil {
   404  		return fmt.Errorf("Failed to open /dev/null - %s", err)
   405  	}
   406  	defer file.Close()
   407  	if err := syscall.Fstat(int(file.Fd()), &devNullStat); err != nil {
   408  		return err
   409  	}
   410  	for fd := 0; fd < 3; fd++ {
   411  		if err := syscall.Fstat(fd, &stat); err != nil {
   412  			return err
   413  		}
   414  		if stat.Rdev == devNullStat.Rdev {
   415  			// Close and re-open the fd.
   416  			if err := syscall.Dup3(int(file.Fd()), fd, 0); err != nil {
   417  				return err
   418  			}
   419  		}
   420  	}
   421  	return nil
   422  }
   423  
   424  // Create the device nodes in the container.
   425  func createDevices(config *configs.Config) error {
   426  	useBindMount := system.RunningInUserNS() || config.Namespaces.Contains(configs.NEWUSER)
   427  	oldMask := syscall.Umask(0000)
   428  	for _, node := range config.Devices {
   429  		// containers running in a user namespace are not allowed to mknod
   430  		// devices so we can just bind mount it from the host.
   431  		if err := createDeviceNode(config.Rootfs, node, useBindMount); err != nil {
   432  			syscall.Umask(oldMask)
   433  			return err
   434  		}
   435  	}
   436  	syscall.Umask(oldMask)
   437  	return nil
   438  }
   439  
   440  func bindMountDeviceNode(dest string, node *configs.Device) error {
   441  	f, err := os.Create(dest)
   442  	if err != nil && !os.IsExist(err) {
   443  		return err
   444  	}
   445  	if f != nil {
   446  		f.Close()
   447  	}
   448  	return syscall.Mount(node.Path, dest, "bind", syscall.MS_BIND, "")
   449  }
   450  
   451  // Creates the device node in the rootfs of the container.
   452  func createDeviceNode(rootfs string, node *configs.Device, bind bool) error {
   453  	dest := filepath.Join(rootfs, node.Path)
   454  	if err := os.MkdirAll(filepath.Dir(dest), 0755); err != nil {
   455  		return err
   456  	}
   457  
   458  	if bind {
   459  		return bindMountDeviceNode(dest, node)
   460  	}
   461  	if err := mknodDevice(dest, node); err != nil {
   462  		if os.IsExist(err) {
   463  			return nil
   464  		} else if os.IsPermission(err) {
   465  			return bindMountDeviceNode(dest, node)
   466  		}
   467  		return err
   468  	}
   469  	return nil
   470  }
   471  
   472  func mknodDevice(dest string, node *configs.Device) error {
   473  	fileMode := node.FileMode
   474  	switch node.Type {
   475  	case 'c':
   476  		fileMode |= syscall.S_IFCHR
   477  	case 'b':
   478  		fileMode |= syscall.S_IFBLK
   479  	default:
   480  		return fmt.Errorf("%c is not a valid device type for device %s", node.Type, node.Path)
   481  	}
   482  	if err := syscall.Mknod(dest, uint32(fileMode), node.Mkdev()); err != nil {
   483  		return err
   484  	}
   485  	return syscall.Chown(dest, int(node.Uid), int(node.Gid))
   486  }
   487  
   488  func getMountInfo(mountinfo []*mount.Info, dir string) *mount.Info {
   489  	for _, m := range mountinfo {
   490  		if m.Mountpoint == dir {
   491  			return m
   492  		}
   493  	}
   494  	return nil
   495  }
   496  
   497  // Get the parent mount point of directory passed in as argument. Also return
   498  // optional fields.
   499  func getParentMount(rootfs string) (string, string, error) {
   500  	var path string
   501  
   502  	mountinfos, err := mount.GetMounts()
   503  	if err != nil {
   504  		return "", "", err
   505  	}
   506  
   507  	mountinfo := getMountInfo(mountinfos, rootfs)
   508  	if mountinfo != nil {
   509  		return rootfs, mountinfo.Optional, nil
   510  	}
   511  
   512  	path = rootfs
   513  	for {
   514  		path = filepath.Dir(path)
   515  
   516  		mountinfo = getMountInfo(mountinfos, path)
   517  		if mountinfo != nil {
   518  			return path, mountinfo.Optional, nil
   519  		}
   520  
   521  		if path == "/" {
   522  			break
   523  		}
   524  	}
   525  
   526  	// If we are here, we did not find parent mount. Something is wrong.
   527  	return "", "", fmt.Errorf("Could not find parent mount of %s", rootfs)
   528  }
   529  
   530  // Make parent mount private if it was shared
   531  func rootfsParentMountPrivate(rootfs string) error {
   532  	sharedMount := false
   533  
   534  	parentMount, optionalOpts, err := getParentMount(rootfs)
   535  	if err != nil {
   536  		return err
   537  	}
   538  
   539  	optsSplit := strings.Split(optionalOpts, " ")
   540  	for _, opt := range optsSplit {
   541  		if strings.HasPrefix(opt, "shared:") {
   542  			sharedMount = true
   543  			break
   544  		}
   545  	}
   546  
   547  	// Make parent mount PRIVATE if it was shared. It is needed for two
   548  	// reasons. First of all pivot_root() will fail if parent mount is
   549  	// shared. Secondly when we bind mount rootfs it will propagate to
   550  	// parent namespace and we don't want that to happen.
   551  	if sharedMount {
   552  		return syscall.Mount("", parentMount, "", syscall.MS_PRIVATE, "")
   553  	}
   554  
   555  	return nil
   556  }
   557  
   558  func prepareRoot(config *configs.Config) error {
   559  	flag := syscall.MS_SLAVE | syscall.MS_REC
   560  	if config.RootPropagation != 0 {
   561  		flag = config.RootPropagation
   562  	}
   563  	if err := syscall.Mount("", "/", "", uintptr(flag), ""); err != nil {
   564  		return err
   565  	}
   566  
   567  	// Make parent mount private to make sure following bind mount does
   568  	// not propagate in other namespaces. Also it will help with kernel
   569  	// check pass in pivot_root. (IS_SHARED(new_mnt->mnt_parent))
   570  	if err := rootfsParentMountPrivate(config.Rootfs); err != nil {
   571  		return err
   572  	}
   573  
   574  	return syscall.Mount(config.Rootfs, config.Rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, "")
   575  }
   576  
   577  func setReadonly() error {
   578  	return syscall.Mount("/", "/", "bind", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC, "")
   579  }
   580  
   581  func setupPtmx(config *configs.Config, console *linuxConsole) error {
   582  	ptmx := filepath.Join(config.Rootfs, "dev/ptmx")
   583  	if err := os.Remove(ptmx); err != nil && !os.IsNotExist(err) {
   584  		return err
   585  	}
   586  	if err := os.Symlink("pts/ptmx", ptmx); err != nil {
   587  		return fmt.Errorf("symlink dev ptmx %s", err)
   588  	}
   589  	if console != nil {
   590  		return console.mount(config.Rootfs, config.MountLabel)
   591  	}
   592  	return nil
   593  }
   594  
   595  // pivotRoot will call pivot_root such that rootfs becomes the new root
   596  // filesystem, and everything else is cleaned up.
   597  func pivotRoot(rootfs string) error {
   598  	// While the documentation may claim otherwise, pivot_root(".", ".") is
   599  	// actually valid. What this results in is / being the new root but
   600  	// /proc/self/cwd being the old root. Since we can play around with the cwd
   601  	// with pivot_root this allows us to pivot without creating directories in
   602  	// the rootfs. Shout-outs to the LXC developers for giving us this idea.
   603  
   604  	oldroot, err := syscall.Open("/", syscall.O_DIRECTORY|syscall.O_RDONLY, 0)
   605  	if err != nil {
   606  		return err
   607  	}
   608  	defer syscall.Close(oldroot)
   609  
   610  	newroot, err := syscall.Open(rootfs, syscall.O_DIRECTORY|syscall.O_RDONLY, 0)
   611  	if err != nil {
   612  		return err
   613  	}
   614  	defer syscall.Close(newroot)
   615  
   616  	// Change to the new root so that the pivot_root actually acts on it.
   617  	if err := syscall.Fchdir(newroot); err != nil {
   618  		return err
   619  	}
   620  
   621  	if err := syscall.PivotRoot(".", "."); err != nil {
   622  		return fmt.Errorf("pivot_root %s", err)
   623  	}
   624  
   625  	// Currently our "." is oldroot (according to the current kernel code).
   626  	// However, purely for safety, we will fchdir(oldroot) since there isn't
   627  	// really any guarantee from the kernel what /proc/self/cwd will be after a
   628  	// pivot_root(2).
   629  
   630  	if err := syscall.Fchdir(oldroot); err != nil {
   631  		return err
   632  	}
   633  
   634  	// Make oldroot rprivate to make sure our unmounts don't propagate to the
   635  	// host (and thus bork the machine).
   636  	if err := syscall.Mount("", ".", "", syscall.MS_PRIVATE|syscall.MS_REC, ""); err != nil {
   637  		return err
   638  	}
   639  	// Preform the unmount. MNT_DETACH allows us to unmount /proc/self/cwd.
   640  	if err := syscall.Unmount(".", syscall.MNT_DETACH); err != nil {
   641  		return err
   642  	}
   643  
   644  	// Switch back to our shiny new root.
   645  	if err := syscall.Chdir("/"); err != nil {
   646  		return fmt.Errorf("chdir / %s", err)
   647  	}
   648  	return nil
   649  }
   650  
   651  func msMoveRoot(rootfs string) error {
   652  	if err := syscall.Mount(rootfs, "/", "", syscall.MS_MOVE, ""); err != nil {
   653  		return err
   654  	}
   655  	if err := syscall.Chroot("."); err != nil {
   656  		return err
   657  	}
   658  	return syscall.Chdir("/")
   659  }
   660  
   661  // createIfNotExists creates a file or a directory only if it does not already exist.
   662  func createIfNotExists(path string, isDir bool) error {
   663  	if _, err := os.Stat(path); err != nil {
   664  		if os.IsNotExist(err) {
   665  			if isDir {
   666  				return os.MkdirAll(path, 0755)
   667  			}
   668  			if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
   669  				return err
   670  			}
   671  			f, err := os.OpenFile(path, os.O_CREATE, 0755)
   672  			if err != nil {
   673  				return err
   674  			}
   675  			f.Close()
   676  		}
   677  	}
   678  	return nil
   679  }
   680  
   681  // remountReadonly will bind over the top of an existing path and ensure that it is read-only.
   682  func remountReadonly(path string) error {
   683  	for i := 0; i < 5; i++ {
   684  		if err := syscall.Mount("", path, "", syscall.MS_REMOUNT|syscall.MS_RDONLY, ""); err != nil && !os.IsNotExist(err) {
   685  			switch err {
   686  			case syscall.EINVAL:
   687  				// Probably not a mountpoint, use bind-mount
   688  				if err := syscall.Mount(path, path, "", syscall.MS_BIND, ""); err != nil {
   689  					return err
   690  				}
   691  				return syscall.Mount(path, path, "", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC|defaultMountFlags, "")
   692  			case syscall.EBUSY:
   693  				time.Sleep(100 * time.Millisecond)
   694  				continue
   695  			default:
   696  				return err
   697  			}
   698  		}
   699  		return nil
   700  	}
   701  	return fmt.Errorf("unable to mount %s as readonly max retries reached", path)
   702  }
   703  
   704  // maskPath masks the top of the specified path inside a container to avoid
   705  // security issues from processes reading information from non-namespace aware
   706  // mounts ( proc/kcore ).
   707  // For files, maskPath bind mounts /dev/null over the top of the specified path.
   708  // For directories, maskPath mounts read-only tmpfs over the top of the specified path.
   709  func maskPath(path string) error {
   710  	if err := syscall.Mount("/dev/null", path, "", syscall.MS_BIND, ""); err != nil && !os.IsNotExist(err) {
   711  		if err == syscall.ENOTDIR {
   712  			return syscall.Mount("tmpfs", path, "tmpfs", syscall.MS_RDONLY, "")
   713  		}
   714  		return err
   715  	}
   716  	return nil
   717  }
   718  
   719  // writeSystemProperty writes the value to a path under /proc/sys as determined from the key.
   720  // For e.g. net.ipv4.ip_forward translated to /proc/sys/net/ipv4/ip_forward.
   721  func writeSystemProperty(key, value string) error {
   722  	keyPath := strings.Replace(key, ".", "/", -1)
   723  	return ioutil.WriteFile(path.Join("/proc/sys", keyPath), []byte(value), 0644)
   724  }
   725  
   726  func remount(m *configs.Mount, rootfs string) error {
   727  	var (
   728  		dest = m.Destination
   729  	)
   730  	if !strings.HasPrefix(dest, rootfs) {
   731  		dest = filepath.Join(rootfs, dest)
   732  	}
   733  	if err := syscall.Mount(m.Source, dest, m.Device, uintptr(m.Flags|syscall.MS_REMOUNT), ""); err != nil {
   734  		return err
   735  	}
   736  	return nil
   737  }
   738  
   739  // Do the mount operation followed by additional mounts required to take care
   740  // of propagation flags.
   741  func mountPropagate(m *configs.Mount, rootfs string, mountLabel string) error {
   742  	var (
   743  		dest  = m.Destination
   744  		data  = label.FormatMountLabel(m.Data, mountLabel)
   745  		flags = m.Flags
   746  	)
   747  	if libcontainerUtils.CleanPath(dest) == "/dev" {
   748  		flags &= ^syscall.MS_RDONLY
   749  	}
   750  
   751  	copyUp := m.Extensions&configs.EXT_COPYUP == configs.EXT_COPYUP
   752  	if !(copyUp || strings.HasPrefix(dest, rootfs)) {
   753  		dest = filepath.Join(rootfs, dest)
   754  	}
   755  
   756  	if err := syscall.Mount(m.Source, dest, m.Device, uintptr(flags), data); err != nil {
   757  		return err
   758  	}
   759  
   760  	for _, pflag := range m.PropagationFlags {
   761  		if err := syscall.Mount("", dest, "", uintptr(pflag), ""); err != nil {
   762  			return err
   763  		}
   764  	}
   765  	return nil
   766  }