github.com/rhatdan/docker@v0.7.7-0.20180119204836-47a0dcbcd20a/daemon/graphdriver/overlay2/overlay.go (about)

     1  // +build linux
     2  
     3  package overlay2
     4  
     5  import (
     6  	"bufio"
     7  	"errors"
     8  	"fmt"
     9  	"io"
    10  	"io/ioutil"
    11  	"os"
    12  	"os/exec"
    13  	"path"
    14  	"path/filepath"
    15  	"strconv"
    16  	"strings"
    17  	"sync"
    18  
    19  	"github.com/docker/docker/daemon/graphdriver"
    20  	"github.com/docker/docker/daemon/graphdriver/overlayutils"
    21  	"github.com/docker/docker/daemon/graphdriver/quota"
    22  	"github.com/docker/docker/pkg/archive"
    23  	"github.com/docker/docker/pkg/chrootarchive"
    24  	"github.com/docker/docker/pkg/containerfs"
    25  	"github.com/docker/docker/pkg/directory"
    26  	"github.com/docker/docker/pkg/fsutils"
    27  	"github.com/docker/docker/pkg/idtools"
    28  	"github.com/docker/docker/pkg/locker"
    29  	"github.com/docker/docker/pkg/mount"
    30  	"github.com/docker/docker/pkg/parsers"
    31  	"github.com/docker/docker/pkg/parsers/kernel"
    32  	"github.com/docker/docker/pkg/system"
    33  	"github.com/docker/go-units"
    34  	rsystem "github.com/opencontainers/runc/libcontainer/system"
    35  	"github.com/opencontainers/selinux/go-selinux/label"
    36  	"github.com/sirupsen/logrus"
    37  	"golang.org/x/sys/unix"
    38  )
    39  
    40  var (
    41  	// untar defines the untar method
    42  	untar = chrootarchive.UntarUncompressed
    43  )
    44  
    45  // This backend uses the overlay union filesystem for containers
    46  // with diff directories for each layer.
    47  
    48  // This version of the overlay driver requires at least kernel
    49  // 4.0.0 in order to support mounting multiple diff directories.
    50  
    51  // Each container/image has at least a "diff" directory and "link" file.
    52  // If there is also a "lower" file when there are diff layers
    53  // below as well as "merged" and "work" directories. The "diff" directory
    54  // has the upper layer of the overlay and is used to capture any
    55  // changes to the layer. The "lower" file contains all the lower layer
    56  // mounts separated by ":" and ordered from uppermost to lowermost
    57  // layers. The overlay itself is mounted in the "merged" directory,
    58  // and the "work" dir is needed for overlay to work.
    59  
    60  // The "link" file for each layer contains a unique string for the layer.
    61  // Under the "l" directory at the root there will be a symbolic link
    62  // with that unique string pointing the "diff" directory for the layer.
    63  // The symbolic links are used to reference lower layers in the "lower"
    64  // file and on mount. The links are used to shorten the total length
    65  // of a layer reference without requiring changes to the layer identifier
    66  // or root directory. Mounts are always done relative to root and
    67  // referencing the symbolic links in order to ensure the number of
    68  // lower directories can fit in a single page for making the mount
    69  // syscall. A hard upper limit of 128 lower layers is enforced to ensure
    70  // that mounts do not fail due to length.
    71  
    72  const (
    73  	driverName = "overlay2"
    74  	linkDir    = "l"
    75  	lowerFile  = "lower"
    76  	maxDepth   = 128
    77  
    78  	// idLength represents the number of random characters
    79  	// which can be used to create the unique link identifier
    80  	// for every layer. If this value is too long then the
    81  	// page size limit for the mount command may be exceeded.
    82  	// The idLength should be selected such that following equation
    83  	// is true (512 is a buffer for label metadata).
    84  	// ((idLength + len(linkDir) + 1) * maxDepth) <= (pageSize - 512)
    85  	idLength = 26
    86  )
    87  
    88  type overlayOptions struct {
    89  	overrideKernelCheck bool
    90  	quota               quota.Quota
    91  }
    92  
    93  // Driver contains information about the home directory and the list of active
    94  // mounts that are created using this driver.
    95  type Driver struct {
    96  	home          string
    97  	uidMaps       []idtools.IDMap
    98  	gidMaps       []idtools.IDMap
    99  	ctr           *graphdriver.RefCounter
   100  	quotaCtl      *quota.Control
   101  	options       overlayOptions
   102  	naiveDiff     graphdriver.DiffDriver
   103  	supportsDType bool
   104  	locker        *locker.Locker
   105  }
   106  
   107  var (
   108  	backingFs             = "<unknown>"
   109  	projectQuotaSupported = false
   110  
   111  	useNaiveDiffLock sync.Once
   112  	useNaiveDiffOnly bool
   113  )
   114  
   115  func init() {
   116  	graphdriver.Register(driverName, Init)
   117  }
   118  
   119  // Init returns the native diff driver for overlay filesystem.
   120  // If overlay filesystem is not supported on the host, the error
   121  // graphdriver.ErrNotSupported is returned.
   122  // If an overlay filesystem is not supported over an existing filesystem then
   123  // the error graphdriver.ErrIncompatibleFS is returned.
   124  func Init(home string, options []string, uidMaps, gidMaps []idtools.IDMap) (graphdriver.Driver, error) {
   125  	opts, err := parseOptions(options)
   126  	if err != nil {
   127  		return nil, err
   128  	}
   129  
   130  	if err := supportsOverlay(); err != nil {
   131  		return nil, graphdriver.ErrNotSupported
   132  	}
   133  
   134  	// require kernel 4.0.0 to ensure multiple lower dirs are supported
   135  	v, err := kernel.GetKernelVersion()
   136  	if err != nil {
   137  		return nil, err
   138  	}
   139  
   140  	// Perform feature detection on /var/lib/docker/overlay2 if it's an existing directory.
   141  	// This covers situations where /var/lib/docker/overlay2 is a mount, and on a different
   142  	// filesystem than /var/lib/docker.
   143  	// If the path does not exist, fall back to using /var/lib/docker for feature detection.
   144  	testdir := home
   145  	if _, err := os.Stat(testdir); os.IsNotExist(err) {
   146  		testdir = filepath.Dir(testdir)
   147  	}
   148  
   149  	fsMagic, err := graphdriver.GetFSMagic(testdir)
   150  	if err != nil {
   151  		return nil, err
   152  	}
   153  	if fsName, ok := graphdriver.FsNames[fsMagic]; ok {
   154  		backingFs = fsName
   155  	}
   156  
   157  	switch fsMagic {
   158  	case graphdriver.FsMagicAufs, graphdriver.FsMagicEcryptfs, graphdriver.FsMagicNfsFs, graphdriver.FsMagicOverlay, graphdriver.FsMagicZfs:
   159  		logrus.Errorf("'overlay2' is not supported over %s", backingFs)
   160  		return nil, graphdriver.ErrIncompatibleFS
   161  	case graphdriver.FsMagicBtrfs:
   162  		// Support for OverlayFS on BTRFS was added in kernel 4.7
   163  		// See https://btrfs.wiki.kernel.org/index.php/Changelog
   164  		if kernel.CompareKernelVersion(*v, kernel.VersionInfo{Kernel: 4, Major: 7, Minor: 0}) < 0 {
   165  			if !opts.overrideKernelCheck {
   166  				logrus.Errorf("'overlay2' requires kernel 4.7 to use on %s", backingFs)
   167  				return nil, graphdriver.ErrIncompatibleFS
   168  			}
   169  			logrus.Warn("Using pre-4.7.0 kernel for overlay2 on btrfs, may require kernel update")
   170  		}
   171  	}
   172  
   173  	if kernel.CompareKernelVersion(*v, kernel.VersionInfo{Kernel: 4, Major: 0, Minor: 0}) < 0 {
   174  		if opts.overrideKernelCheck {
   175  			logrus.Warn("Using pre-4.0.0 kernel for overlay2, mount failures may require kernel update")
   176  		} else {
   177  			if err := supportsMultipleLowerDir(testdir); err != nil {
   178  				logrus.Debugf("Multiple lower dirs not supported: %v", err)
   179  				return nil, graphdriver.ErrNotSupported
   180  			}
   181  		}
   182  	}
   183  	supportsDType, err := fsutils.SupportsDType(testdir)
   184  	if err != nil {
   185  		return nil, err
   186  	}
   187  	if !supportsDType {
   188  		if !graphdriver.IsInitialized(home) {
   189  			return nil, overlayutils.ErrDTypeNotSupported("overlay2", backingFs)
   190  		}
   191  		// allow running without d_type only for existing setups (#27443)
   192  		logrus.Warn(overlayutils.ErrDTypeNotSupported("overlay2", backingFs))
   193  	}
   194  
   195  	rootUID, rootGID, err := idtools.GetRootUIDGID(uidMaps, gidMaps)
   196  	if err != nil {
   197  		return nil, err
   198  	}
   199  	// Create the driver home dir
   200  	if err := idtools.MkdirAllAndChown(path.Join(home, linkDir), 0700, idtools.IDPair{rootUID, rootGID}); err != nil {
   201  		return nil, err
   202  	}
   203  
   204  	if err := mount.MakePrivate(home); err != nil {
   205  		return nil, err
   206  	}
   207  
   208  	d := &Driver{
   209  		home:          home,
   210  		uidMaps:       uidMaps,
   211  		gidMaps:       gidMaps,
   212  		ctr:           graphdriver.NewRefCounter(graphdriver.NewFsChecker(graphdriver.FsMagicOverlay)),
   213  		supportsDType: supportsDType,
   214  		locker:        locker.New(),
   215  		options:       *opts,
   216  	}
   217  
   218  	d.naiveDiff = graphdriver.NewNaiveDiffDriver(d, uidMaps, gidMaps)
   219  
   220  	if backingFs == "xfs" {
   221  		// Try to enable project quota support over xfs.
   222  		if d.quotaCtl, err = quota.NewControl(home); err == nil {
   223  			projectQuotaSupported = true
   224  		} else if opts.quota.Size > 0 {
   225  			return nil, fmt.Errorf("Storage option overlay2.size not supported. Filesystem does not support Project Quota: %v", err)
   226  		}
   227  	} else if opts.quota.Size > 0 {
   228  		// if xfs is not the backing fs then error out if the storage-opt overlay2.size is used.
   229  		return nil, fmt.Errorf("Storage Option overlay2.size only supported for backingFS XFS. Found %v", backingFs)
   230  	}
   231  
   232  	logrus.Debugf("backingFs=%s,  projectQuotaSupported=%v", backingFs, projectQuotaSupported)
   233  
   234  	return d, nil
   235  }
   236  
   237  func parseOptions(options []string) (*overlayOptions, error) {
   238  	o := &overlayOptions{}
   239  	for _, option := range options {
   240  		key, val, err := parsers.ParseKeyValueOpt(option)
   241  		if err != nil {
   242  			return nil, err
   243  		}
   244  		key = strings.ToLower(key)
   245  		switch key {
   246  		case "overlay2.override_kernel_check":
   247  			o.overrideKernelCheck, err = strconv.ParseBool(val)
   248  			if err != nil {
   249  				return nil, err
   250  			}
   251  		case "overlay2.size":
   252  			size, err := units.RAMInBytes(val)
   253  			if err != nil {
   254  				return nil, err
   255  			}
   256  			o.quota.Size = uint64(size)
   257  		default:
   258  			return nil, fmt.Errorf("overlay2: unknown option %s", key)
   259  		}
   260  	}
   261  	return o, nil
   262  }
   263  
   264  func supportsOverlay() error {
   265  	// We can try to modprobe overlay first before looking at
   266  	// proc/filesystems for when overlay is supported
   267  	exec.Command("modprobe", "overlay").Run()
   268  
   269  	f, err := os.Open("/proc/filesystems")
   270  	if err != nil {
   271  		return err
   272  	}
   273  	defer f.Close()
   274  
   275  	s := bufio.NewScanner(f)
   276  	for s.Scan() {
   277  		if s.Text() == "nodev\toverlay" {
   278  			return nil
   279  		}
   280  	}
   281  	logrus.Error("'overlay' not found as a supported filesystem on this host. Please ensure kernel is new enough and has overlay support loaded.")
   282  	return graphdriver.ErrNotSupported
   283  }
   284  
   285  func useNaiveDiff(home string) bool {
   286  	useNaiveDiffLock.Do(func() {
   287  		if err := doesSupportNativeDiff(home); err != nil {
   288  			logrus.Warnf("Not using native diff for overlay2, this may cause degraded performance for building images: %v", err)
   289  			useNaiveDiffOnly = true
   290  		}
   291  	})
   292  	return useNaiveDiffOnly
   293  }
   294  
   295  func (d *Driver) String() string {
   296  	return driverName
   297  }
   298  
   299  // Status returns current driver information in a two dimensional string array.
   300  // Output contains "Backing Filesystem" used in this implementation.
   301  func (d *Driver) Status() [][2]string {
   302  	return [][2]string{
   303  		{"Backing Filesystem", backingFs},
   304  		{"Supports d_type", strconv.FormatBool(d.supportsDType)},
   305  		{"Native Overlay Diff", strconv.FormatBool(!useNaiveDiff(d.home))},
   306  	}
   307  }
   308  
   309  // GetMetadata returns metadata about the overlay driver such as the LowerDir,
   310  // UpperDir, WorkDir, and MergeDir used to store data.
   311  func (d *Driver) GetMetadata(id string) (map[string]string, error) {
   312  	dir := d.dir(id)
   313  	if _, err := os.Stat(dir); err != nil {
   314  		return nil, err
   315  	}
   316  
   317  	metadata := map[string]string{
   318  		"WorkDir":   path.Join(dir, "work"),
   319  		"MergedDir": path.Join(dir, "merged"),
   320  		"UpperDir":  path.Join(dir, "diff"),
   321  	}
   322  
   323  	lowerDirs, err := d.getLowerDirs(id)
   324  	if err != nil {
   325  		return nil, err
   326  	}
   327  	if len(lowerDirs) > 0 {
   328  		metadata["LowerDir"] = strings.Join(lowerDirs, ":")
   329  	}
   330  
   331  	return metadata, nil
   332  }
   333  
   334  // Cleanup any state created by overlay which should be cleaned when daemon
   335  // is being shutdown. For now, we just have to unmount the bind mounted
   336  // we had created.
   337  func (d *Driver) Cleanup() error {
   338  	return mount.Unmount(d.home)
   339  }
   340  
   341  // CreateReadWrite creates a layer that is writable for use as a container
   342  // file system.
   343  func (d *Driver) CreateReadWrite(id, parent string, opts *graphdriver.CreateOpts) error {
   344  	if opts != nil && len(opts.StorageOpt) != 0 && !projectQuotaSupported {
   345  		return fmt.Errorf("--storage-opt is supported only for overlay over xfs with 'pquota' mount option")
   346  	}
   347  
   348  	if opts == nil {
   349  		opts = &graphdriver.CreateOpts{
   350  			StorageOpt: map[string]string{},
   351  		}
   352  	}
   353  
   354  	if _, ok := opts.StorageOpt["size"]; !ok {
   355  		if opts.StorageOpt == nil {
   356  			opts.StorageOpt = map[string]string{}
   357  		}
   358  		opts.StorageOpt["size"] = strconv.FormatUint(d.options.quota.Size, 10)
   359  	}
   360  
   361  	return d.create(id, parent, opts)
   362  }
   363  
   364  // Create is used to create the upper, lower, and merge directories required for overlay fs for a given id.
   365  // The parent filesystem is used to configure these directories for the overlay.
   366  func (d *Driver) Create(id, parent string, opts *graphdriver.CreateOpts) (retErr error) {
   367  	if opts != nil && len(opts.StorageOpt) != 0 {
   368  		if _, ok := opts.StorageOpt["size"]; ok {
   369  			return fmt.Errorf("--storage-opt size is only supported for ReadWrite Layers")
   370  		}
   371  	}
   372  	return d.create(id, parent, opts)
   373  }
   374  
   375  func (d *Driver) create(id, parent string, opts *graphdriver.CreateOpts) (retErr error) {
   376  	dir := d.dir(id)
   377  
   378  	rootUID, rootGID, err := idtools.GetRootUIDGID(d.uidMaps, d.gidMaps)
   379  	if err != nil {
   380  		return err
   381  	}
   382  	root := idtools.IDPair{UID: rootUID, GID: rootGID}
   383  
   384  	if err := idtools.MkdirAllAndChown(path.Dir(dir), 0700, root); err != nil {
   385  		return err
   386  	}
   387  	if err := idtools.MkdirAndChown(dir, 0700, root); err != nil {
   388  		return err
   389  	}
   390  
   391  	defer func() {
   392  		// Clean up on failure
   393  		if retErr != nil {
   394  			os.RemoveAll(dir)
   395  		}
   396  	}()
   397  
   398  	if opts != nil && len(opts.StorageOpt) > 0 {
   399  		driver := &Driver{}
   400  		if err := d.parseStorageOpt(opts.StorageOpt, driver); err != nil {
   401  			return err
   402  		}
   403  
   404  		if driver.options.quota.Size > 0 {
   405  			// Set container disk quota limit
   406  			if err := d.quotaCtl.SetQuota(dir, driver.options.quota); err != nil {
   407  				return err
   408  			}
   409  		}
   410  	}
   411  
   412  	if err := idtools.MkdirAndChown(path.Join(dir, "diff"), 0755, root); err != nil {
   413  		return err
   414  	}
   415  
   416  	lid := generateID(idLength)
   417  	if err := os.Symlink(path.Join("..", id, "diff"), path.Join(d.home, linkDir, lid)); err != nil {
   418  		return err
   419  	}
   420  
   421  	// Write link id to link file
   422  	if err := ioutil.WriteFile(path.Join(dir, "link"), []byte(lid), 0644); err != nil {
   423  		return err
   424  	}
   425  
   426  	// if no parent directory, done
   427  	if parent == "" {
   428  		return nil
   429  	}
   430  
   431  	if err := idtools.MkdirAndChown(path.Join(dir, "work"), 0700, root); err != nil {
   432  		return err
   433  	}
   434  
   435  	lower, err := d.getLower(parent)
   436  	if err != nil {
   437  		return err
   438  	}
   439  	if lower != "" {
   440  		if err := ioutil.WriteFile(path.Join(dir, lowerFile), []byte(lower), 0666); err != nil {
   441  			return err
   442  		}
   443  	}
   444  
   445  	return nil
   446  }
   447  
   448  // Parse overlay storage options
   449  func (d *Driver) parseStorageOpt(storageOpt map[string]string, driver *Driver) error {
   450  	// Read size to set the disk project quota per container
   451  	for key, val := range storageOpt {
   452  		key := strings.ToLower(key)
   453  		switch key {
   454  		case "size":
   455  			size, err := units.RAMInBytes(val)
   456  			if err != nil {
   457  				return err
   458  			}
   459  			driver.options.quota.Size = uint64(size)
   460  		default:
   461  			return fmt.Errorf("Unknown option %s", key)
   462  		}
   463  	}
   464  
   465  	return nil
   466  }
   467  
   468  func (d *Driver) getLower(parent string) (string, error) {
   469  	parentDir := d.dir(parent)
   470  
   471  	// Ensure parent exists
   472  	if _, err := os.Lstat(parentDir); err != nil {
   473  		return "", err
   474  	}
   475  
   476  	// Read Parent link fileA
   477  	parentLink, err := ioutil.ReadFile(path.Join(parentDir, "link"))
   478  	if err != nil {
   479  		return "", err
   480  	}
   481  	lowers := []string{path.Join(linkDir, string(parentLink))}
   482  
   483  	parentLower, err := ioutil.ReadFile(path.Join(parentDir, lowerFile))
   484  	if err == nil {
   485  		parentLowers := strings.Split(string(parentLower), ":")
   486  		lowers = append(lowers, parentLowers...)
   487  	}
   488  	if len(lowers) > maxDepth {
   489  		return "", errors.New("max depth exceeded")
   490  	}
   491  	return strings.Join(lowers, ":"), nil
   492  }
   493  
   494  func (d *Driver) dir(id string) string {
   495  	return path.Join(d.home, id)
   496  }
   497  
   498  func (d *Driver) getLowerDirs(id string) ([]string, error) {
   499  	var lowersArray []string
   500  	lowers, err := ioutil.ReadFile(path.Join(d.dir(id), lowerFile))
   501  	if err == nil {
   502  		for _, s := range strings.Split(string(lowers), ":") {
   503  			lp, err := os.Readlink(path.Join(d.home, s))
   504  			if err != nil {
   505  				return nil, err
   506  			}
   507  			lowersArray = append(lowersArray, path.Clean(path.Join(d.home, linkDir, lp)))
   508  		}
   509  	} else if !os.IsNotExist(err) {
   510  		return nil, err
   511  	}
   512  	return lowersArray, nil
   513  }
   514  
   515  // Remove cleans the directories that are created for this id.
   516  func (d *Driver) Remove(id string) error {
   517  	d.locker.Lock(id)
   518  	defer d.locker.Unlock(id)
   519  	dir := d.dir(id)
   520  	lid, err := ioutil.ReadFile(path.Join(dir, "link"))
   521  	if err == nil {
   522  		if err := os.RemoveAll(path.Join(d.home, linkDir, string(lid))); err != nil {
   523  			logrus.Debugf("Failed to remove link: %v", err)
   524  		}
   525  	}
   526  
   527  	if err := system.EnsureRemoveAll(dir); err != nil && !os.IsNotExist(err) {
   528  		return err
   529  	}
   530  	return nil
   531  }
   532  
   533  // Get creates and mounts the required file system for the given id and returns the mount path.
   534  func (d *Driver) Get(id, mountLabel string) (_ containerfs.ContainerFS, retErr error) {
   535  	d.locker.Lock(id)
   536  	defer d.locker.Unlock(id)
   537  	dir := d.dir(id)
   538  	if _, err := os.Stat(dir); err != nil {
   539  		return nil, err
   540  	}
   541  
   542  	diffDir := path.Join(dir, "diff")
   543  	lowers, err := ioutil.ReadFile(path.Join(dir, lowerFile))
   544  	if err != nil {
   545  		// If no lower, just return diff directory
   546  		if os.IsNotExist(err) {
   547  			return containerfs.NewLocalContainerFS(diffDir), nil
   548  		}
   549  		return nil, err
   550  	}
   551  
   552  	mergedDir := path.Join(dir, "merged")
   553  	if count := d.ctr.Increment(mergedDir); count > 1 {
   554  		return containerfs.NewLocalContainerFS(mergedDir), nil
   555  	}
   556  	defer func() {
   557  		if retErr != nil {
   558  			if c := d.ctr.Decrement(mergedDir); c <= 0 {
   559  				if mntErr := unix.Unmount(mergedDir, 0); mntErr != nil {
   560  					logrus.Errorf("error unmounting %v: %v", mergedDir, mntErr)
   561  				}
   562  				// Cleanup the created merged directory; see the comment in Put's rmdir
   563  				if rmErr := unix.Rmdir(mergedDir); rmErr != nil && !os.IsNotExist(rmErr) {
   564  					logrus.Debugf("Failed to remove %s: %v: %v", id, rmErr, err)
   565  				}
   566  			}
   567  		}
   568  	}()
   569  
   570  	workDir := path.Join(dir, "work")
   571  	splitLowers := strings.Split(string(lowers), ":")
   572  	absLowers := make([]string, len(splitLowers))
   573  	for i, s := range splitLowers {
   574  		absLowers[i] = path.Join(d.home, s)
   575  	}
   576  	opts := fmt.Sprintf("lowerdir=%s,upperdir=%s,workdir=%s", strings.Join(absLowers, ":"), path.Join(dir, "diff"), path.Join(dir, "work"))
   577  	mountData := label.FormatMountLabel(opts, mountLabel)
   578  	mount := unix.Mount
   579  	mountTarget := mergedDir
   580  
   581  	rootUID, rootGID, err := idtools.GetRootUIDGID(d.uidMaps, d.gidMaps)
   582  	if err != nil {
   583  		return nil, err
   584  	}
   585  	if err := idtools.MkdirAndChown(mergedDir, 0700, idtools.IDPair{rootUID, rootGID}); err != nil {
   586  		return nil, err
   587  	}
   588  
   589  	pageSize := unix.Getpagesize()
   590  
   591  	// Go can return a larger page size than supported by the system
   592  	// as of go 1.7. This will be fixed in 1.8 and this block can be
   593  	// removed when building with 1.8.
   594  	// See https://github.com/golang/go/commit/1b9499b06989d2831e5b156161d6c07642926ee1
   595  	// See https://github.com/docker/docker/issues/27384
   596  	if pageSize > 4096 {
   597  		pageSize = 4096
   598  	}
   599  
   600  	// Use relative paths and mountFrom when the mount data has exceeded
   601  	// the page size. The mount syscall fails if the mount data cannot
   602  	// fit within a page and relative links make the mount data much
   603  	// smaller at the expense of requiring a fork exec to chroot.
   604  	if len(mountData) > pageSize {
   605  		opts = fmt.Sprintf("lowerdir=%s,upperdir=%s,workdir=%s", string(lowers), path.Join(id, "diff"), path.Join(id, "work"))
   606  		mountData = label.FormatMountLabel(opts, mountLabel)
   607  		if len(mountData) > pageSize {
   608  			return nil, fmt.Errorf("cannot mount layer, mount label too large %d", len(mountData))
   609  		}
   610  
   611  		mount = func(source string, target string, mType string, flags uintptr, label string) error {
   612  			return mountFrom(d.home, source, target, mType, flags, label)
   613  		}
   614  		mountTarget = path.Join(id, "merged")
   615  	}
   616  
   617  	if err := mount("overlay", mountTarget, "overlay", 0, mountData); err != nil {
   618  		return nil, fmt.Errorf("error creating overlay mount to %s: %v", mergedDir, err)
   619  	}
   620  
   621  	// chown "workdir/work" to the remapped root UID/GID. Overlay fs inside a
   622  	// user namespace requires this to move a directory from lower to upper.
   623  	if err := os.Chown(path.Join(workDir, "work"), rootUID, rootGID); err != nil {
   624  		return nil, err
   625  	}
   626  
   627  	return containerfs.NewLocalContainerFS(mergedDir), nil
   628  }
   629  
   630  // Put unmounts the mount path created for the give id.
   631  // It also removes the 'merged' directory to force the kernel to unmount the
   632  // overlay mount in other namespaces.
   633  func (d *Driver) Put(id string) error {
   634  	d.locker.Lock(id)
   635  	defer d.locker.Unlock(id)
   636  	dir := d.dir(id)
   637  	_, err := ioutil.ReadFile(path.Join(dir, lowerFile))
   638  	if err != nil {
   639  		// If no lower, no mount happened and just return directly
   640  		if os.IsNotExist(err) {
   641  			return nil
   642  		}
   643  		return err
   644  	}
   645  
   646  	mountpoint := path.Join(dir, "merged")
   647  	if count := d.ctr.Decrement(mountpoint); count > 0 {
   648  		return nil
   649  	}
   650  	if err := unix.Unmount(mountpoint, unix.MNT_DETACH); err != nil {
   651  		logrus.Debugf("Failed to unmount %s overlay: %s - %v", id, mountpoint, err)
   652  	}
   653  	// Remove the mountpoint here. Removing the mountpoint (in newer kernels)
   654  	// will cause all other instances of this mount in other mount namespaces
   655  	// to be unmounted. This is necessary to avoid cases where an overlay mount
   656  	// that is present in another namespace will cause subsequent mounts
   657  	// operations to fail with ebusy.  We ignore any errors here because this may
   658  	// fail on older kernels which don't have
   659  	// torvalds/linux@8ed936b5671bfb33d89bc60bdcc7cf0470ba52fe applied.
   660  	if err := unix.Rmdir(mountpoint); err != nil && !os.IsNotExist(err) {
   661  		logrus.Debugf("Failed to remove %s overlay: %v", id, err)
   662  	}
   663  	return nil
   664  }
   665  
   666  // Exists checks to see if the id is already mounted.
   667  func (d *Driver) Exists(id string) bool {
   668  	_, err := os.Stat(d.dir(id))
   669  	return err == nil
   670  }
   671  
   672  // isParent determines whether the given parent is the direct parent of the
   673  // given layer id
   674  func (d *Driver) isParent(id, parent string) bool {
   675  	lowers, err := d.getLowerDirs(id)
   676  	if err != nil {
   677  		return false
   678  	}
   679  	if parent == "" && len(lowers) > 0 {
   680  		return false
   681  	}
   682  
   683  	parentDir := d.dir(parent)
   684  	var ld string
   685  	if len(lowers) > 0 {
   686  		ld = filepath.Dir(lowers[0])
   687  	}
   688  	if ld == "" && parent == "" {
   689  		return true
   690  	}
   691  	return ld == parentDir
   692  }
   693  
   694  // ApplyDiff applies the new layer into a root
   695  func (d *Driver) ApplyDiff(id string, parent string, diff io.Reader) (size int64, err error) {
   696  	if !d.isParent(id, parent) {
   697  		return d.naiveDiff.ApplyDiff(id, parent, diff)
   698  	}
   699  
   700  	applyDir := d.getDiffPath(id)
   701  
   702  	logrus.Debugf("Applying tar in %s", applyDir)
   703  	// Overlay doesn't need the parent id to apply the diff
   704  	if err := untar(diff, applyDir, &archive.TarOptions{
   705  		UIDMaps:        d.uidMaps,
   706  		GIDMaps:        d.gidMaps,
   707  		WhiteoutFormat: archive.OverlayWhiteoutFormat,
   708  		InUserNS:       rsystem.RunningInUserNS(),
   709  	}); err != nil {
   710  		return 0, err
   711  	}
   712  
   713  	return directory.Size(applyDir)
   714  }
   715  
   716  func (d *Driver) getDiffPath(id string) string {
   717  	dir := d.dir(id)
   718  
   719  	return path.Join(dir, "diff")
   720  }
   721  
   722  // DiffSize calculates the changes between the specified id
   723  // and its parent and returns the size in bytes of the changes
   724  // relative to its base filesystem directory.
   725  func (d *Driver) DiffSize(id, parent string) (size int64, err error) {
   726  	if useNaiveDiff(d.home) || !d.isParent(id, parent) {
   727  		return d.naiveDiff.DiffSize(id, parent)
   728  	}
   729  	return directory.Size(d.getDiffPath(id))
   730  }
   731  
   732  // Diff produces an archive of the changes between the specified
   733  // layer and its parent layer which may be "".
   734  func (d *Driver) Diff(id, parent string) (io.ReadCloser, error) {
   735  	if useNaiveDiff(d.home) || !d.isParent(id, parent) {
   736  		return d.naiveDiff.Diff(id, parent)
   737  	}
   738  
   739  	diffPath := d.getDiffPath(id)
   740  	logrus.Debugf("Tar with options on %s", diffPath)
   741  	return archive.TarWithOptions(diffPath, &archive.TarOptions{
   742  		Compression:    archive.Uncompressed,
   743  		UIDMaps:        d.uidMaps,
   744  		GIDMaps:        d.gidMaps,
   745  		WhiteoutFormat: archive.OverlayWhiteoutFormat,
   746  	})
   747  }
   748  
   749  // Changes produces a list of changes between the specified layer and its
   750  // parent layer. If parent is "", then all changes will be ADD changes.
   751  func (d *Driver) Changes(id, parent string) ([]archive.Change, error) {
   752  	if useNaiveDiff(d.home) || !d.isParent(id, parent) {
   753  		return d.naiveDiff.Changes(id, parent)
   754  	}
   755  	// Overlay doesn't have snapshots, so we need to get changes from all parent
   756  	// layers.
   757  	diffPath := d.getDiffPath(id)
   758  	layers, err := d.getLowerDirs(id)
   759  	if err != nil {
   760  		return nil, err
   761  	}
   762  
   763  	return archive.OverlayChanges(layers, diffPath)
   764  }