container driver: copy ca and user tls registries certs

Signed-off-by: CrazyMax <crazy-max@users.noreply.github.com>

vendor/github.com/docker/docker/pkg/archive/README.md

@@ -0,0 +1 @@
+This code provides helper functions for dealing with archive files.

vendor/github.com/docker/docker/pkg/archive/archive_linux.go

@@ -0,0 +1,100 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"archive/tar"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/docker/docker/pkg/system"
+	"github.com/pkg/errors"
+	"golang.org/x/sys/unix"
+)
+
+func getWhiteoutConverter(format WhiteoutFormat, inUserNS bool) (tarWhiteoutConverter, error) {
+	if format == OverlayWhiteoutFormat {
+		if inUserNS {
+			return nil, errors.New("specifying OverlayWhiteoutFormat is not allowed in userns")
+		}
+		return overlayWhiteoutConverter{}, nil
+	}
+	return nil, nil
+}
+
+type overlayWhiteoutConverter struct {
+}
+
+func (overlayWhiteoutConverter) ConvertWrite(hdr *tar.Header, path string, fi os.FileInfo) (wo *tar.Header, err error) {
+	// convert whiteouts to AUFS format
+	if fi.Mode()&os.ModeCharDevice != 0 && hdr.Devmajor == 0 && hdr.Devminor == 0 {
+		// we just rename the file and make it normal
+		dir, filename := filepath.Split(hdr.Name)
+		hdr.Name = filepath.Join(dir, WhiteoutPrefix+filename)
+		hdr.Mode = 0600
+		hdr.Typeflag = tar.TypeReg
+		hdr.Size = 0
+	}
+
+	if fi.Mode()&os.ModeDir != 0 {
+		// convert opaque dirs to AUFS format by writing an empty file with the prefix
+		opaque, err := system.Lgetxattr(path, "trusted.overlay.opaque")
+		if err != nil {
+			return nil, err
+		}
+		if len(opaque) == 1 && opaque[0] == 'y' {
+			if hdr.Xattrs != nil {
+				delete(hdr.Xattrs, "trusted.overlay.opaque")
+			}
+
+			// create a header for the whiteout file
+			// it should inherit some properties from the parent, but be a regular file
+			wo = &tar.Header{
+				Typeflag:   tar.TypeReg,
+				Mode:       hdr.Mode & int64(os.ModePerm),
+				Name:       filepath.Join(hdr.Name, WhiteoutOpaqueDir),
+				Size:       0,
+				Uid:        hdr.Uid,
+				Uname:      hdr.Uname,
+				Gid:        hdr.Gid,
+				Gname:      hdr.Gname,
+				AccessTime: hdr.AccessTime,
+				ChangeTime: hdr.ChangeTime,
+			}
+		}
+	}
+
+	return
+}
+
+func (c overlayWhiteoutConverter) ConvertRead(hdr *tar.Header, path string) (bool, error) {
+	base := filepath.Base(path)
+	dir := filepath.Dir(path)
+
+	// if a directory is marked as opaque by the AUFS special file, we need to translate that to overlay
+	if base == WhiteoutOpaqueDir {
+		err := unix.Setxattr(dir, "trusted.overlay.opaque", []byte{'y'}, 0)
+		if err != nil {
+			return false, errors.Wrapf(err, "setxattr(%q, trusted.overlay.opaque=y)", dir)
+		}
+		// don't write the file itself
+		return false, err
+	}
+
+	// if a file was deleted and we are using overlay, we need to create a character device
+	if strings.HasPrefix(base, WhiteoutPrefix) {
+		originalBase := base[len(WhiteoutPrefix):]
+		originalPath := filepath.Join(dir, originalBase)
+
+		if err := unix.Mknod(originalPath, unix.S_IFCHR, 0); err != nil {
+			return false, errors.Wrapf(err, "failed to mknod(%q, S_IFCHR, 0)", originalPath)
+		}
+		if err := os.Chown(originalPath, hdr.Uid, hdr.Gid); err != nil {
+			return false, err
+		}
+
+		// don't write the file itself
+		return false, nil
+	}
+
+	return true, nil
+}

vendor/github.com/docker/docker/pkg/archive/archive_other.go

@@ -0,0 +1,7 @@
+// +build !linux
+
+package archive // import "github.com/docker/docker/pkg/archive"
+
+func getWhiteoutConverter(format WhiteoutFormat, inUserNS bool) (tarWhiteoutConverter, error) {
+	return nil, nil
+}

vendor/github.com/docker/docker/pkg/archive/archive_unix.go

@@ -0,0 +1,115 @@
+// +build !windows
+
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"archive/tar"
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+	"syscall"
+
+	"github.com/containerd/containerd/pkg/userns"
+	"github.com/docker/docker/pkg/idtools"
+	"github.com/docker/docker/pkg/system"
+	"golang.org/x/sys/unix"
+)
+
+// fixVolumePathPrefix does platform specific processing to ensure that if
+// the path being passed in is not in a volume path format, convert it to one.
+func fixVolumePathPrefix(srcPath string) string {
+	return srcPath
+}
+
+// getWalkRoot calculates the root path when performing a TarWithOptions.
+// We use a separate function as this is platform specific. On Linux, we
+// can't use filepath.Join(srcPath,include) because this will clean away
+// a trailing "." or "/" which may be important.
+func getWalkRoot(srcPath string, include string) string {
+	return strings.TrimSuffix(srcPath, string(filepath.Separator)) + string(filepath.Separator) + include
+}
+
+// CanonicalTarNameForPath returns platform-specific filepath
+// to canonical posix-style path for tar archival. p is relative
+// path.
+func CanonicalTarNameForPath(p string) string {
+	return p // already unix-style
+}
+
+// chmodTarEntry is used to adjust the file permissions used in tar header based
+// on the platform the archival is done.
+
+func chmodTarEntry(perm os.FileMode) os.FileMode {
+	return perm // noop for unix as golang APIs provide perm bits correctly
+}
+
+func setHeaderForSpecialDevice(hdr *tar.Header, name string, stat interface{}) (err error) {
+	s, ok := stat.(*syscall.Stat_t)
+
+	if ok {
+		// Currently go does not fill in the major/minors
+		if s.Mode&unix.S_IFBLK != 0 ||
+			s.Mode&unix.S_IFCHR != 0 {
+			hdr.Devmajor = int64(unix.Major(uint64(s.Rdev))) //nolint: unconvert
+			hdr.Devminor = int64(unix.Minor(uint64(s.Rdev))) //nolint: unconvert
+		}
+	}
+
+	return
+}
+
+func getInodeFromStat(stat interface{}) (inode uint64, err error) {
+	s, ok := stat.(*syscall.Stat_t)
+
+	if ok {
+		inode = s.Ino
+	}
+
+	return
+}
+
+func getFileUIDGID(stat interface{}) (idtools.Identity, error) {
+	s, ok := stat.(*syscall.Stat_t)
+
+	if !ok {
+		return idtools.Identity{}, errors.New("cannot convert stat value to syscall.Stat_t")
+	}
+	return idtools.Identity{UID: int(s.Uid), GID: int(s.Gid)}, nil
+}
+
+// handleTarTypeBlockCharFifo is an OS-specific helper function used by
+// createTarFile to handle the following types of header: Block; Char; Fifo
+func handleTarTypeBlockCharFifo(hdr *tar.Header, path string) error {
+	mode := uint32(hdr.Mode & 07777)
+	switch hdr.Typeflag {
+	case tar.TypeBlock:
+		mode |= unix.S_IFBLK
+	case tar.TypeChar:
+		mode |= unix.S_IFCHR
+	case tar.TypeFifo:
+		mode |= unix.S_IFIFO
+	}
+
+	err := system.Mknod(path, mode, int(system.Mkdev(hdr.Devmajor, hdr.Devminor)))
+	if errors.Is(err, syscall.EPERM) && userns.RunningInUserNS() {
+		// In most cases, cannot create a device if running in user namespace
+		err = nil
+	}
+	return err
+}
+
+func handleLChmod(hdr *tar.Header, path string, hdrInfo os.FileInfo) error {
+	if hdr.Typeflag == tar.TypeLink {
+		if fi, err := os.Lstat(hdr.Linkname); err == nil && (fi.Mode()&os.ModeSymlink == 0) {
+			if err := os.Chmod(path, hdrInfo.Mode()); err != nil {
+				return err
+			}
+		}
+	} else if hdr.Typeflag != tar.TypeSymlink {
+		if err := os.Chmod(path, hdrInfo.Mode()); err != nil {
+			return err
+		}
+	}
+	return nil
+}

vendor/github.com/docker/docker/pkg/archive/archive_windows.go

@@ -0,0 +1,67 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"archive/tar"
+	"os"
+	"path/filepath"
+
+	"github.com/docker/docker/pkg/idtools"
+	"github.com/docker/docker/pkg/longpath"
+)
+
+// fixVolumePathPrefix does platform specific processing to ensure that if
+// the path being passed in is not in a volume path format, convert it to one.
+func fixVolumePathPrefix(srcPath string) string {
+	return longpath.AddPrefix(srcPath)
+}
+
+// getWalkRoot calculates the root path when performing a TarWithOptions.
+// We use a separate function as this is platform specific.
+func getWalkRoot(srcPath string, include string) string {
+	return filepath.Join(srcPath, include)
+}
+
+// CanonicalTarNameForPath returns platform-specific filepath
+// to canonical posix-style path for tar archival. p is relative
+// path.
+func CanonicalTarNameForPath(p string) string {
+	return filepath.ToSlash(p)
+}
+
+// chmodTarEntry is used to adjust the file permissions used in tar header based
+// on the platform the archival is done.
+func chmodTarEntry(perm os.FileMode) os.FileMode {
+	// perm &= 0755 // this 0-ed out tar flags (like link, regular file, directory marker etc.)
+	permPart := perm & os.ModePerm
+	noPermPart := perm &^ os.ModePerm
+	// Add the x bit: make everything +x from windows
+	permPart |= 0111
+	permPart &= 0755
+
+	return noPermPart | permPart
+}
+
+func setHeaderForSpecialDevice(hdr *tar.Header, name string, stat interface{}) (err error) {
+	// do nothing. no notion of Rdev, Nlink in stat on Windows
+	return
+}
+
+func getInodeFromStat(stat interface{}) (inode uint64, err error) {
+	// do nothing. no notion of Inode in stat on Windows
+	return
+}
+
+// handleTarTypeBlockCharFifo is an OS-specific helper function used by
+// createTarFile to handle the following types of header: Block; Char; Fifo
+func handleTarTypeBlockCharFifo(hdr *tar.Header, path string) error {
+	return nil
+}
+
+func handleLChmod(hdr *tar.Header, path string, hdrInfo os.FileInfo) error {
+	return nil
+}
+
+func getFileUIDGID(stat interface{}) (idtools.Identity, error) {
+	// no notion of file ownership mapping yet on Windows
+	return idtools.Identity{UID: 0, GID: 0}, nil
+}

vendor/github.com/docker/docker/pkg/archive/changes.go

@@ -0,0 +1,445 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"archive/tar"
+	"bytes"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"sort"
+	"strings"
+	"syscall"
+	"time"
+
+	"github.com/docker/docker/pkg/idtools"
+	"github.com/docker/docker/pkg/pools"
+	"github.com/docker/docker/pkg/system"
+	"github.com/sirupsen/logrus"
+)
+
+// ChangeType represents the change type.
+type ChangeType int
+
+const (
+	// ChangeModify represents the modify operation.
+	ChangeModify = iota
+	// ChangeAdd represents the add operation.
+	ChangeAdd
+	// ChangeDelete represents the delete operation.
+	ChangeDelete
+)
+
+func (c ChangeType) String() string {
+	switch c {
+	case ChangeModify:
+		return "C"
+	case ChangeAdd:
+		return "A"
+	case ChangeDelete:
+		return "D"
+	}
+	return ""
+}
+
+// Change represents a change, it wraps the change type and path.
+// It describes changes of the files in the path respect to the
+// parent layers. The change could be modify, add, delete.
+// This is used for layer diff.
+type Change struct {
+	Path string
+	Kind ChangeType
+}
+
+func (change *Change) String() string {
+	return fmt.Sprintf("%s %s", change.Kind, change.Path)
+}
+
+// for sort.Sort
+type changesByPath []Change
+
+func (c changesByPath) Less(i, j int) bool { return c[i].Path < c[j].Path }
+func (c changesByPath) Len() int           { return len(c) }
+func (c changesByPath) Swap(i, j int)      { c[j], c[i] = c[i], c[j] }
+
+// Gnu tar doesn't have sub-second mtime precision. The go tar
+// writer (1.10+) does when using PAX format, but we round times to seconds
+// to ensure archives have the same hashes for backwards compatibility.
+// See https://github.com/moby/moby/pull/35739/commits/fb170206ba12752214630b269a40ac7be6115ed4.
+//
+// Non-sub-second is problematic when we apply changes via tar
+// files. We handle this by comparing for exact times, *or* same
+// second count and either a or b having exactly 0 nanoseconds
+func sameFsTime(a, b time.Time) bool {
+	return a.Equal(b) ||
+		(a.Unix() == b.Unix() &&
+			(a.Nanosecond() == 0 || b.Nanosecond() == 0))
+}
+
+func sameFsTimeSpec(a, b syscall.Timespec) bool {
+	return a.Sec == b.Sec &&
+		(a.Nsec == b.Nsec || a.Nsec == 0 || b.Nsec == 0)
+}
+
+// Changes walks the path rw and determines changes for the files in the path,
+// with respect to the parent layers
+func Changes(layers []string, rw string) ([]Change, error) {
+	return changes(layers, rw, aufsDeletedFile, aufsMetadataSkip)
+}
+
+func aufsMetadataSkip(path string) (skip bool, err error) {
+	skip, err = filepath.Match(string(os.PathSeparator)+WhiteoutMetaPrefix+"*", path)
+	if err != nil {
+		skip = true
+	}
+	return
+}
+
+func aufsDeletedFile(root, path string, fi os.FileInfo) (string, error) {
+	f := filepath.Base(path)
+
+	// If there is a whiteout, then the file was removed
+	if strings.HasPrefix(f, WhiteoutPrefix) {
+		originalFile := f[len(WhiteoutPrefix):]
+		return filepath.Join(filepath.Dir(path), originalFile), nil
+	}
+
+	return "", nil
+}
+
+type skipChange func(string) (bool, error)
+type deleteChange func(string, string, os.FileInfo) (string, error)
+
+func changes(layers []string, rw string, dc deleteChange, sc skipChange) ([]Change, error) {
+	var (
+		changes     []Change
+		changedDirs = make(map[string]struct{})
+	)
+
+	err := filepath.Walk(rw, func(path string, f os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+
+		// Rebase path
+		path, err = filepath.Rel(rw, path)
+		if err != nil {
+			return err
+		}
+
+		// As this runs on the daemon side, file paths are OS specific.
+		path = filepath.Join(string(os.PathSeparator), path)
+
+		// Skip root
+		if path == string(os.PathSeparator) {
+			return nil
+		}
+
+		if sc != nil {
+			if skip, err := sc(path); skip {
+				return err
+			}
+		}
+
+		change := Change{
+			Path: path,
+		}
+
+		deletedFile, err := dc(rw, path, f)
+		if err != nil {
+			return err
+		}
+
+		// Find out what kind of modification happened
+		if deletedFile != "" {
+			change.Path = deletedFile
+			change.Kind = ChangeDelete
+		} else {
+			// Otherwise, the file was added
+			change.Kind = ChangeAdd
+
+			// ...Unless it already existed in a top layer, in which case, it's a modification
+			for _, layer := range layers {
+				stat, err := os.Stat(filepath.Join(layer, path))
+				if err != nil && !os.IsNotExist(err) {
+					return err
+				}
+				if err == nil {
+					// The file existed in the top layer, so that's a modification
+
+					// However, if it's a directory, maybe it wasn't actually modified.
+					// If you modify /foo/bar/baz, then /foo will be part of the changed files only because it's the parent of bar
+					if stat.IsDir() && f.IsDir() {
+						if f.Size() == stat.Size() && f.Mode() == stat.Mode() && sameFsTime(f.ModTime(), stat.ModTime()) {
+							// Both directories are the same, don't record the change
+							return nil
+						}
+					}
+					change.Kind = ChangeModify
+					break
+				}
+			}
+		}
+
+		// If /foo/bar/file.txt is modified, then /foo/bar must be part of the changed files.
+		// This block is here to ensure the change is recorded even if the
+		// modify time, mode and size of the parent directory in the rw and ro layers are all equal.
+		// Check https://github.com/docker/docker/pull/13590 for details.
+		if f.IsDir() {
+			changedDirs[path] = struct{}{}
+		}
+		if change.Kind == ChangeAdd || change.Kind == ChangeDelete {
+			parent := filepath.Dir(path)
+			if _, ok := changedDirs[parent]; !ok && parent != "/" {
+				changes = append(changes, Change{Path: parent, Kind: ChangeModify})
+				changedDirs[parent] = struct{}{}
+			}
+		}
+
+		// Record change
+		changes = append(changes, change)
+		return nil
+	})
+	if err != nil && !os.IsNotExist(err) {
+		return nil, err
+	}
+	return changes, nil
+}
+
+// FileInfo describes the information of a file.
+type FileInfo struct {
+	parent     *FileInfo
+	name       string
+	stat       *system.StatT
+	children   map[string]*FileInfo
+	capability []byte
+	added      bool
+}
+
+// LookUp looks up the file information of a file.
+func (info *FileInfo) LookUp(path string) *FileInfo {
+	// As this runs on the daemon side, file paths are OS specific.
+	parent := info
+	if path == string(os.PathSeparator) {
+		return info
+	}
+
+	pathElements := strings.Split(path, string(os.PathSeparator))
+	for _, elem := range pathElements {
+		if elem != "" {
+			child := parent.children[elem]
+			if child == nil {
+				return nil
+			}
+			parent = child
+		}
+	}
+	return parent
+}
+
+func (info *FileInfo) path() string {
+	if info.parent == nil {
+		// As this runs on the daemon side, file paths are OS specific.
+		return string(os.PathSeparator)
+	}
+	return filepath.Join(info.parent.path(), info.name)
+}
+
+func (info *FileInfo) addChanges(oldInfo *FileInfo, changes *[]Change) {
+
+	sizeAtEntry := len(*changes)
+
+	if oldInfo == nil {
+		// add
+		change := Change{
+			Path: info.path(),
+			Kind: ChangeAdd,
+		}
+		*changes = append(*changes, change)
+		info.added = true
+	}
+
+	// We make a copy so we can modify it to detect additions
+	// also, we only recurse on the old dir if the new info is a directory
+	// otherwise any previous delete/change is considered recursive
+	oldChildren := make(map[string]*FileInfo)
+	if oldInfo != nil && info.isDir() {
+		for k, v := range oldInfo.children {
+			oldChildren[k] = v
+		}
+	}
+
+	for name, newChild := range info.children {
+		oldChild := oldChildren[name]
+		if oldChild != nil {
+			// change?
+			oldStat := oldChild.stat
+			newStat := newChild.stat
+			// Note: We can't compare inode or ctime or blocksize here, because these change
+			// when copying a file into a container. However, that is not generally a problem
+			// because any content change will change mtime, and any status change should
+			// be visible when actually comparing the stat fields. The only time this
+			// breaks down is if some code intentionally hides a change by setting
+			// back mtime
+			if statDifferent(oldStat, newStat) ||
+				!bytes.Equal(oldChild.capability, newChild.capability) {
+				change := Change{
+					Path: newChild.path(),
+					Kind: ChangeModify,
+				}
+				*changes = append(*changes, change)
+				newChild.added = true
+			}
+
+			// Remove from copy so we can detect deletions
+			delete(oldChildren, name)
+		}
+
+		newChild.addChanges(oldChild, changes)
+	}
+	for _, oldChild := range oldChildren {
+		// delete
+		change := Change{
+			Path: oldChild.path(),
+			Kind: ChangeDelete,
+		}
+		*changes = append(*changes, change)
+	}
+
+	// If there were changes inside this directory, we need to add it, even if the directory
+	// itself wasn't changed. This is needed to properly save and restore filesystem permissions.
+	// As this runs on the daemon side, file paths are OS specific.
+	if len(*changes) > sizeAtEntry && info.isDir() && !info.added && info.path() != string(os.PathSeparator) {
+		change := Change{
+			Path: info.path(),
+			Kind: ChangeModify,
+		}
+		// Let's insert the directory entry before the recently added entries located inside this dir
+		*changes = append(*changes, change) // just to resize the slice, will be overwritten
+		copy((*changes)[sizeAtEntry+1:], (*changes)[sizeAtEntry:])
+		(*changes)[sizeAtEntry] = change
+	}
+
+}
+
+// Changes add changes to file information.
+func (info *FileInfo) Changes(oldInfo *FileInfo) []Change {
+	var changes []Change
+
+	info.addChanges(oldInfo, &changes)
+
+	return changes
+}
+
+func newRootFileInfo() *FileInfo {
+	// As this runs on the daemon side, file paths are OS specific.
+	root := &FileInfo{
+		name:     string(os.PathSeparator),
+		children: make(map[string]*FileInfo),
+	}
+	return root
+}
+
+// ChangesDirs compares two directories and generates an array of Change objects describing the changes.
+// If oldDir is "", then all files in newDir will be Add-Changes.
+func ChangesDirs(newDir, oldDir string) ([]Change, error) {
+	var (
+		oldRoot, newRoot *FileInfo
+	)
+	if oldDir == "" {
+		emptyDir, err := ioutil.TempDir("", "empty")
+		if err != nil {
+			return nil, err
+		}
+		defer os.Remove(emptyDir)
+		oldDir = emptyDir
+	}
+	oldRoot, newRoot, err := collectFileInfoForChanges(oldDir, newDir)
+	if err != nil {
+		return nil, err
+	}
+
+	return newRoot.Changes(oldRoot), nil
+}
+
+// ChangesSize calculates the size in bytes of the provided changes, based on newDir.
+func ChangesSize(newDir string, changes []Change) int64 {
+	var (
+		size int64
+		sf   = make(map[uint64]struct{})
+	)
+	for _, change := range changes {
+		if change.Kind == ChangeModify || change.Kind == ChangeAdd {
+			file := filepath.Join(newDir, change.Path)
+			fileInfo, err := os.Lstat(file)
+			if err != nil {
+				logrus.Errorf("Can not stat %q: %s", file, err)
+				continue
+			}
+
+			if fileInfo != nil && !fileInfo.IsDir() {
+				if hasHardlinks(fileInfo) {
+					inode := getIno(fileInfo)
+					if _, ok := sf[inode]; !ok {
+						size += fileInfo.Size()
+						sf[inode] = struct{}{}
+					}
+				} else {
+					size += fileInfo.Size()
+				}
+			}
+		}
+	}
+	return size
+}
+
+// ExportChanges produces an Archive from the provided changes, relative to dir.
+func ExportChanges(dir string, changes []Change, uidMaps, gidMaps []idtools.IDMap) (io.ReadCloser, error) {
+	reader, writer := io.Pipe()
+	go func() {
+		ta := newTarAppender(idtools.NewIDMappingsFromMaps(uidMaps, gidMaps), writer, nil)
+
+		// this buffer is needed for the duration of this piped stream
+		defer pools.BufioWriter32KPool.Put(ta.Buffer)
+
+		sort.Sort(changesByPath(changes))
+
+		// In general we log errors here but ignore them because
+		// during e.g. a diff operation the container can continue
+		// mutating the filesystem and we can see transient errors
+		// from this
+		for _, change := range changes {
+			if change.Kind == ChangeDelete {
+				whiteOutDir := filepath.Dir(change.Path)
+				whiteOutBase := filepath.Base(change.Path)
+				whiteOut := filepath.Join(whiteOutDir, WhiteoutPrefix+whiteOutBase)
+				timestamp := time.Now()
+				hdr := &tar.Header{
+					Name:       whiteOut[1:],
+					Size:       0,
+					ModTime:    timestamp,
+					AccessTime: timestamp,
+					ChangeTime: timestamp,
+				}
+				if err := ta.TarWriter.WriteHeader(hdr); err != nil {
+					logrus.Debugf("Can't write whiteout header: %s", err)
+				}
+			} else {
+				path := filepath.Join(dir, change.Path)
+				if err := ta.addTarFile(path, change.Path[1:]); err != nil {
+					logrus.Debugf("Can't add file %s to tar: %s", path, err)
+				}
+			}
+		}
+
+		// Make sure to check the error on Close.
+		if err := ta.TarWriter.Close(); err != nil {
+			logrus.Debugf("Can't close layer: %s", err)
+		}
+		if err := writer.Close(); err != nil {
+			logrus.Debugf("failed close Changes writer: %s", err)
+		}
+	}()
+	return reader, nil
+}

vendor/github.com/docker/docker/pkg/archive/changes_linux.go

@@ -0,0 +1,286 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"bytes"
+	"fmt"
+	"os"
+	"path/filepath"
+	"sort"
+	"syscall"
+	"unsafe"
+
+	"github.com/docker/docker/pkg/system"
+	"golang.org/x/sys/unix"
+)
+
+// walker is used to implement collectFileInfoForChanges on linux. Where this
+// method in general returns the entire contents of two directory trees, we
+// optimize some FS calls out on linux. In particular, we take advantage of the
+// fact that getdents(2) returns the inode of each file in the directory being
+// walked, which, when walking two trees in parallel to generate a list of
+// changes, can be used to prune subtrees without ever having to lstat(2) them
+// directly. Eliminating stat calls in this way can save up to seconds on large
+// images.
+type walker struct {
+	dir1  string
+	dir2  string
+	root1 *FileInfo
+	root2 *FileInfo
+}
+
+// collectFileInfoForChanges returns a complete representation of the trees
+// rooted at dir1 and dir2, with one important exception: any subtree or
+// leaf where the inode and device numbers are an exact match between dir1
+// and dir2 will be pruned from the results. This method is *only* to be used
+// to generating a list of changes between the two directories, as it does not
+// reflect the full contents.
+func collectFileInfoForChanges(dir1, dir2 string) (*FileInfo, *FileInfo, error) {
+	w := &walker{
+		dir1:  dir1,
+		dir2:  dir2,
+		root1: newRootFileInfo(),
+		root2: newRootFileInfo(),
+	}
+
+	i1, err := os.Lstat(w.dir1)
+	if err != nil {
+		return nil, nil, err
+	}
+	i2, err := os.Lstat(w.dir2)
+	if err != nil {
+		return nil, nil, err
+	}
+
+	if err := w.walk("/", i1, i2); err != nil {
+		return nil, nil, err
+	}
+
+	return w.root1, w.root2, nil
+}
+
+// Given a FileInfo, its path info, and a reference to the root of the tree
+// being constructed, register this file with the tree.
+func walkchunk(path string, fi os.FileInfo, dir string, root *FileInfo) error {
+	if fi == nil {
+		return nil
+	}
+	parent := root.LookUp(filepath.Dir(path))
+	if parent == nil {
+		return fmt.Errorf("walkchunk: Unexpectedly no parent for %s", path)
+	}
+	info := &FileInfo{
+		name:     filepath.Base(path),
+		children: make(map[string]*FileInfo),
+		parent:   parent,
+	}
+	cpath := filepath.Join(dir, path)
+	stat, err := system.FromStatT(fi.Sys().(*syscall.Stat_t))
+	if err != nil {
+		return err
+	}
+	info.stat = stat
+	info.capability, _ = system.Lgetxattr(cpath, "security.capability") // lgetxattr(2): fs access
+	parent.children[info.name] = info
+	return nil
+}
+
+// Walk a subtree rooted at the same path in both trees being iterated. For
+// example, /docker/overlay/1234/a/b/c/d and /docker/overlay/8888/a/b/c/d
+func (w *walker) walk(path string, i1, i2 os.FileInfo) (err error) {
+	// Register these nodes with the return trees, unless we're still at the
+	// (already-created) roots:
+	if path != "/" {
+		if err := walkchunk(path, i1, w.dir1, w.root1); err != nil {
+			return err
+		}
+		if err := walkchunk(path, i2, w.dir2, w.root2); err != nil {
+			return err
+		}
+	}
+
+	is1Dir := i1 != nil && i1.IsDir()
+	is2Dir := i2 != nil && i2.IsDir()
+
+	sameDevice := false
+	if i1 != nil && i2 != nil {
+		si1 := i1.Sys().(*syscall.Stat_t)
+		si2 := i2.Sys().(*syscall.Stat_t)
+		if si1.Dev == si2.Dev {
+			sameDevice = true
+		}
+	}
+
+	// If these files are both non-existent, or leaves (non-dirs), we are done.
+	if !is1Dir && !is2Dir {
+		return nil
+	}
+
+	// Fetch the names of all the files contained in both directories being walked:
+	var names1, names2 []nameIno
+	if is1Dir {
+		names1, err = readdirnames(filepath.Join(w.dir1, path)) // getdents(2): fs access
+		if err != nil {
+			return err
+		}
+	}
+	if is2Dir {
+		names2, err = readdirnames(filepath.Join(w.dir2, path)) // getdents(2): fs access
+		if err != nil {
+			return err
+		}
+	}
+
+	// We have lists of the files contained in both parallel directories, sorted
+	// in the same order. Walk them in parallel, generating a unique merged list
+	// of all items present in either or both directories.
+	var names []string
+	ix1 := 0
+	ix2 := 0
+
+	for {
+		if ix1 >= len(names1) {
+			break
+		}
+		if ix2 >= len(names2) {
+			break
+		}
+
+		ni1 := names1[ix1]
+		ni2 := names2[ix2]
+
+		switch bytes.Compare([]byte(ni1.name), []byte(ni2.name)) {
+		case -1: // ni1 < ni2 -- advance ni1
+			// we will not encounter ni1 in names2
+			names = append(names, ni1.name)
+			ix1++
+		case 0: // ni1 == ni2
+			if ni1.ino != ni2.ino || !sameDevice {
+				names = append(names, ni1.name)
+			}
+			ix1++
+			ix2++
+		case 1: // ni1 > ni2 -- advance ni2
+			// we will not encounter ni2 in names1
+			names = append(names, ni2.name)
+			ix2++
+		}
+	}
+	for ix1 < len(names1) {
+		names = append(names, names1[ix1].name)
+		ix1++
+	}
+	for ix2 < len(names2) {
+		names = append(names, names2[ix2].name)
+		ix2++
+	}
+
+	// For each of the names present in either or both of the directories being
+	// iterated, stat the name under each root, and recurse the pair of them:
+	for _, name := range names {
+		fname := filepath.Join(path, name)
+		var cInfo1, cInfo2 os.FileInfo
+		if is1Dir {
+			cInfo1, err = os.Lstat(filepath.Join(w.dir1, fname)) // lstat(2): fs access
+			if err != nil && !os.IsNotExist(err) {
+				return err
+			}
+		}
+		if is2Dir {
+			cInfo2, err = os.Lstat(filepath.Join(w.dir2, fname)) // lstat(2): fs access
+			if err != nil && !os.IsNotExist(err) {
+				return err
+			}
+		}
+		if err = w.walk(fname, cInfo1, cInfo2); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// {name,inode} pairs used to support the early-pruning logic of the walker type
+type nameIno struct {
+	name string
+	ino  uint64
+}
+
+type nameInoSlice []nameIno
+
+func (s nameInoSlice) Len() int           { return len(s) }
+func (s nameInoSlice) Swap(i, j int)      { s[i], s[j] = s[j], s[i] }
+func (s nameInoSlice) Less(i, j int) bool { return s[i].name < s[j].name }
+
+// readdirnames is a hacked-apart version of the Go stdlib code, exposing inode
+// numbers further up the stack when reading directory contents. Unlike
+// os.Readdirnames, which returns a list of filenames, this function returns a
+// list of {filename,inode} pairs.
+func readdirnames(dirname string) (names []nameIno, err error) {
+	var (
+		size = 100
+		buf  = make([]byte, 4096)
+		nbuf int
+		bufp int
+		nb   int
+	)
+
+	f, err := os.Open(dirname)
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	names = make([]nameIno, 0, size) // Empty with room to grow.
+	for {
+		// Refill the buffer if necessary
+		if bufp >= nbuf {
+			bufp = 0
+			nbuf, err = unix.ReadDirent(int(f.Fd()), buf) // getdents on linux
+			if nbuf < 0 {
+				nbuf = 0
+			}
+			if err != nil {
+				return nil, os.NewSyscallError("readdirent", err)
+			}
+			if nbuf <= 0 {
+				break // EOF
+			}
+		}
+
+		// Drain the buffer
+		nb, names = parseDirent(buf[bufp:nbuf], names)
+		bufp += nb
+	}
+
+	sl := nameInoSlice(names)
+	sort.Sort(sl)
+	return sl, nil
+}
+
+// parseDirent is a minor modification of unix.ParseDirent (linux version)
+// which returns {name,inode} pairs instead of just names.
+func parseDirent(buf []byte, names []nameIno) (consumed int, newnames []nameIno) {
+	origlen := len(buf)
+	for len(buf) > 0 {
+		dirent := (*unix.Dirent)(unsafe.Pointer(&buf[0]))
+		buf = buf[dirent.Reclen:]
+		if dirent.Ino == 0 { // File absent in directory.
+			continue
+		}
+		bytes := (*[10000]byte)(unsafe.Pointer(&dirent.Name[0]))
+		var name = string(bytes[0:clen(bytes[:])])
+		if name == "." || name == ".." { // Useless names
+			continue
+		}
+		names = append(names, nameIno{name, dirent.Ino})
+	}
+	return origlen - len(buf), names
+}
+
+func clen(n []byte) int {
+	for i := 0; i < len(n); i++ {
+		if n[i] == 0 {
+			return i
+		}
+	}
+	return len(n)
+}

vendor/github.com/docker/docker/pkg/archive/changes_other.go

@@ -0,0 +1,97 @@
+// +build !linux
+
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+
+	"github.com/docker/docker/pkg/system"
+)
+
+func collectFileInfoForChanges(oldDir, newDir string) (*FileInfo, *FileInfo, error) {
+	var (
+		oldRoot, newRoot *FileInfo
+		err1, err2       error
+		errs             = make(chan error, 2)
+	)
+	go func() {
+		oldRoot, err1 = collectFileInfo(oldDir)
+		errs <- err1
+	}()
+	go func() {
+		newRoot, err2 = collectFileInfo(newDir)
+		errs <- err2
+	}()
+
+	// block until both routines have returned
+	for i := 0; i < 2; i++ {
+		if err := <-errs; err != nil {
+			return nil, nil, err
+		}
+	}
+
+	return oldRoot, newRoot, nil
+}
+
+func collectFileInfo(sourceDir string) (*FileInfo, error) {
+	root := newRootFileInfo()
+
+	err := filepath.Walk(sourceDir, func(path string, f os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+
+		// Rebase path
+		relPath, err := filepath.Rel(sourceDir, path)
+		if err != nil {
+			return err
+		}
+
+		// As this runs on the daemon side, file paths are OS specific.
+		relPath = filepath.Join(string(os.PathSeparator), relPath)
+
+		// See https://github.com/golang/go/issues/9168 - bug in filepath.Join.
+		// Temporary workaround. If the returned path starts with two backslashes,
+		// trim it down to a single backslash. Only relevant on Windows.
+		if runtime.GOOS == "windows" {
+			if strings.HasPrefix(relPath, `\\`) {
+				relPath = relPath[1:]
+			}
+		}
+
+		if relPath == string(os.PathSeparator) {
+			return nil
+		}
+
+		parent := root.LookUp(filepath.Dir(relPath))
+		if parent == nil {
+			return fmt.Errorf("collectFileInfo: Unexpectedly no parent for %s", relPath)
+		}
+
+		info := &FileInfo{
+			name:     filepath.Base(relPath),
+			children: make(map[string]*FileInfo),
+			parent:   parent,
+		}
+
+		s, err := system.Lstat(path)
+		if err != nil {
+			return err
+		}
+		info.stat = s
+
+		info.capability, _ = system.Lgetxattr(path, "security.capability")
+
+		parent.children[info.name] = info
+
+		return nil
+	})
+	if err != nil {
+		return nil, err
+	}
+	return root, nil
+}

vendor/github.com/docker/docker/pkg/archive/changes_unix.go

@@ -0,0 +1,43 @@
+// +build !windows
+
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"os"
+	"syscall"
+
+	"github.com/docker/docker/pkg/system"
+	"golang.org/x/sys/unix"
+)
+
+func statDifferent(oldStat *system.StatT, newStat *system.StatT) bool {
+	// Don't look at size for dirs, its not a good measure of change
+	if oldStat.Mode() != newStat.Mode() ||
+		oldStat.UID() != newStat.UID() ||
+		oldStat.GID() != newStat.GID() ||
+		oldStat.Rdev() != newStat.Rdev() ||
+		// Don't look at size or modification time for dirs, its not a good
+		// measure of change. See https://github.com/moby/moby/issues/9874
+		// for a description of the issue with modification time, and
+		// https://github.com/moby/moby/pull/11422 for the change.
+		// (Note that in the Windows implementation of this function,
+		// modification time IS taken as a change). See
+		// https://github.com/moby/moby/pull/37982 for more information.
+		(oldStat.Mode()&unix.S_IFDIR != unix.S_IFDIR &&
+			(!sameFsTimeSpec(oldStat.Mtim(), newStat.Mtim()) || (oldStat.Size() != newStat.Size()))) {
+		return true
+	}
+	return false
+}
+
+func (info *FileInfo) isDir() bool {
+	return info.parent == nil || info.stat.Mode()&unix.S_IFDIR != 0
+}
+
+func getIno(fi os.FileInfo) uint64 {
+	return fi.Sys().(*syscall.Stat_t).Ino
+}
+
+func hasHardlinks(fi os.FileInfo) bool {
+	return fi.Sys().(*syscall.Stat_t).Nlink > 1
+}

vendor/github.com/docker/docker/pkg/archive/changes_windows.go

@@ -0,0 +1,34 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"os"
+
+	"github.com/docker/docker/pkg/system"
+)
+
+func statDifferent(oldStat *system.StatT, newStat *system.StatT) bool {
+	// Note there is slight difference between the Linux and Windows
+	// implementations here. Due to https://github.com/moby/moby/issues/9874,
+	// and the fix at https://github.com/moby/moby/pull/11422, Linux does not
+	// consider a change to the directory time as a change. Windows on NTFS
+	// does. See https://github.com/moby/moby/pull/37982 for more information.
+
+	if !sameFsTime(oldStat.Mtim(), newStat.Mtim()) ||
+		oldStat.Mode() != newStat.Mode() ||
+		oldStat.Size() != newStat.Size() && !oldStat.Mode().IsDir() {
+		return true
+	}
+	return false
+}
+
+func (info *FileInfo) isDir() bool {
+	return info.parent == nil || info.stat.Mode().IsDir()
+}
+
+func getIno(fi os.FileInfo) (inode uint64) {
+	return
+}
+
+func hasHardlinks(fi os.FileInfo) bool {
+	return false
+}

vendor/github.com/docker/docker/pkg/archive/copy.go

@@ -0,0 +1,487 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"archive/tar"
+	"errors"
+	"io"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/docker/docker/pkg/system"
+	"github.com/sirupsen/logrus"
+)
+
+// Errors used or returned by this file.
+var (
+	ErrNotDirectory      = errors.New("not a directory")
+	ErrDirNotExists      = errors.New("no such directory")
+	ErrCannotCopyDir     = errors.New("cannot copy directory")
+	ErrInvalidCopySource = errors.New("invalid copy source content")
+)
+
+// PreserveTrailingDotOrSeparator returns the given cleaned path (after
+// processing using any utility functions from the path or filepath stdlib
+// packages) and appends a trailing `/.` or `/` if its corresponding  original
+// path (from before being processed by utility functions from the path or
+// filepath stdlib packages) ends with a trailing `/.` or `/`. If the cleaned
+// path already ends in a `.` path segment, then another is not added. If the
+// clean path already ends in the separator, then another is not added.
+func PreserveTrailingDotOrSeparator(cleanedPath string, originalPath string, sep byte) string {
+	// Ensure paths are in platform semantics
+	cleanedPath = strings.Replace(cleanedPath, "/", string(sep), -1)
+	originalPath = strings.Replace(originalPath, "/", string(sep), -1)
+
+	if !specifiesCurrentDir(cleanedPath) && specifiesCurrentDir(originalPath) {
+		if !hasTrailingPathSeparator(cleanedPath, sep) {
+			// Add a separator if it doesn't already end with one (a cleaned
+			// path would only end in a separator if it is the root).
+			cleanedPath += string(sep)
+		}
+		cleanedPath += "."
+	}
+
+	if !hasTrailingPathSeparator(cleanedPath, sep) && hasTrailingPathSeparator(originalPath, sep) {
+		cleanedPath += string(sep)
+	}
+
+	return cleanedPath
+}
+
+// assertsDirectory returns whether the given path is
+// asserted to be a directory, i.e., the path ends with
+// a trailing '/' or `/.`, assuming a path separator of `/`.
+func assertsDirectory(path string, sep byte) bool {
+	return hasTrailingPathSeparator(path, sep) || specifiesCurrentDir(path)
+}
+
+// hasTrailingPathSeparator returns whether the given
+// path ends with the system's path separator character.
+func hasTrailingPathSeparator(path string, sep byte) bool {
+	return len(path) > 0 && path[len(path)-1] == sep
+}
+
+// specifiesCurrentDir returns whether the given path specifies
+// a "current directory", i.e., the last path segment is `.`.
+func specifiesCurrentDir(path string) bool {
+	return filepath.Base(path) == "."
+}
+
+// SplitPathDirEntry splits the given path between its directory name and its
+// basename by first cleaning the path but preserves a trailing "." if the
+// original path specified the current directory.
+func SplitPathDirEntry(path string) (dir, base string) {
+	cleanedPath := filepath.Clean(filepath.FromSlash(path))
+
+	if specifiesCurrentDir(path) {
+		cleanedPath += string(os.PathSeparator) + "."
+	}
+
+	return filepath.Dir(cleanedPath), filepath.Base(cleanedPath)
+}
+
+// TarResource archives the resource described by the given CopyInfo to a Tar
+// archive. A non-nil error is returned if sourcePath does not exist or is
+// asserted to be a directory but exists as another type of file.
+//
+// This function acts as a convenient wrapper around TarWithOptions, which
+// requires a directory as the source path. TarResource accepts either a
+// directory or a file path and correctly sets the Tar options.
+func TarResource(sourceInfo CopyInfo) (content io.ReadCloser, err error) {
+	return TarResourceRebase(sourceInfo.Path, sourceInfo.RebaseName)
+}
+
+// TarResourceRebase is like TarResource but renames the first path element of
+// items in the resulting tar archive to match the given rebaseName if not "".
+func TarResourceRebase(sourcePath, rebaseName string) (content io.ReadCloser, err error) {
+	sourcePath = normalizePath(sourcePath)
+	if _, err = os.Lstat(sourcePath); err != nil {
+		// Catches the case where the source does not exist or is not a
+		// directory if asserted to be a directory, as this also causes an
+		// error.
+		return
+	}
+
+	// Separate the source path between its directory and
+	// the entry in that directory which we are archiving.
+	sourceDir, sourceBase := SplitPathDirEntry(sourcePath)
+	opts := TarResourceRebaseOpts(sourceBase, rebaseName)
+
+	logrus.Debugf("copying %q from %q", sourceBase, sourceDir)
+	return TarWithOptions(sourceDir, opts)
+}
+
+// TarResourceRebaseOpts does not preform the Tar, but instead just creates the rebase
+// parameters to be sent to TarWithOptions (the TarOptions struct)
+func TarResourceRebaseOpts(sourceBase string, rebaseName string) *TarOptions {
+	filter := []string{sourceBase}
+	return &TarOptions{
+		Compression:      Uncompressed,
+		IncludeFiles:     filter,
+		IncludeSourceDir: true,
+		RebaseNames: map[string]string{
+			sourceBase: rebaseName,
+		},
+	}
+}
+
+// CopyInfo holds basic info about the source
+// or destination path of a copy operation.
+type CopyInfo struct {
+	Path       string
+	Exists     bool
+	IsDir      bool
+	RebaseName string
+}
+
+// CopyInfoSourcePath stats the given path to create a CopyInfo
+// struct representing that resource for the source of an archive copy
+// operation. The given path should be an absolute local path. A source path
+// has all symlinks evaluated that appear before the last path separator ("/"
+// on Unix). As it is to be a copy source, the path must exist.
+func CopyInfoSourcePath(path string, followLink bool) (CopyInfo, error) {
+	// normalize the file path and then evaluate the symbol link
+	// we will use the target file instead of the symbol link if
+	// followLink is set
+	path = normalizePath(path)
+
+	resolvedPath, rebaseName, err := ResolveHostSourcePath(path, followLink)
+	if err != nil {
+		return CopyInfo{}, err
+	}
+
+	stat, err := os.Lstat(resolvedPath)
+	if err != nil {
+		return CopyInfo{}, err
+	}
+
+	return CopyInfo{
+		Path:       resolvedPath,
+		Exists:     true,
+		IsDir:      stat.IsDir(),
+		RebaseName: rebaseName,
+	}, nil
+}
+
+// CopyInfoDestinationPath stats the given path to create a CopyInfo
+// struct representing that resource for the destination of an archive copy
+// operation. The given path should be an absolute local path.
+func CopyInfoDestinationPath(path string) (info CopyInfo, err error) {
+	maxSymlinkIter := 10 // filepath.EvalSymlinks uses 255, but 10 already seems like a lot.
+	path = normalizePath(path)
+	originalPath := path
+
+	stat, err := os.Lstat(path)
+
+	if err == nil && stat.Mode()&os.ModeSymlink == 0 {
+		// The path exists and is not a symlink.
+		return CopyInfo{
+			Path:   path,
+			Exists: true,
+			IsDir:  stat.IsDir(),
+		}, nil
+	}
+
+	// While the path is a symlink.
+	for n := 0; err == nil && stat.Mode()&os.ModeSymlink != 0; n++ {
+		if n > maxSymlinkIter {
+			// Don't follow symlinks more than this arbitrary number of times.
+			return CopyInfo{}, errors.New("too many symlinks in " + originalPath)
+		}
+
+		// The path is a symbolic link. We need to evaluate it so that the
+		// destination of the copy operation is the link target and not the
+		// link itself. This is notably different than CopyInfoSourcePath which
+		// only evaluates symlinks before the last appearing path separator.
+		// Also note that it is okay if the last path element is a broken
+		// symlink as the copy operation should create the target.
+		var linkTarget string
+
+		linkTarget, err = os.Readlink(path)
+		if err != nil {
+			return CopyInfo{}, err
+		}
+
+		if !system.IsAbs(linkTarget) {
+			// Join with the parent directory.
+			dstParent, _ := SplitPathDirEntry(path)
+			linkTarget = filepath.Join(dstParent, linkTarget)
+		}
+
+		path = linkTarget
+		stat, err = os.Lstat(path)
+	}
+
+	if err != nil {
+		// It's okay if the destination path doesn't exist. We can still
+		// continue the copy operation if the parent directory exists.
+		if !os.IsNotExist(err) {
+			return CopyInfo{}, err
+		}
+
+		// Ensure destination parent dir exists.
+		dstParent, _ := SplitPathDirEntry(path)
+
+		parentDirStat, err := os.Stat(dstParent)
+		if err != nil {
+			return CopyInfo{}, err
+		}
+		if !parentDirStat.IsDir() {
+			return CopyInfo{}, ErrNotDirectory
+		}
+
+		return CopyInfo{Path: path}, nil
+	}
+
+	// The path exists after resolving symlinks.
+	return CopyInfo{
+		Path:   path,
+		Exists: true,
+		IsDir:  stat.IsDir(),
+	}, nil
+}
+
+// PrepareArchiveCopy prepares the given srcContent archive, which should
+// contain the archived resource described by srcInfo, to the destination
+// described by dstInfo. Returns the possibly modified content archive along
+// with the path to the destination directory which it should be extracted to.
+func PrepareArchiveCopy(srcContent io.Reader, srcInfo, dstInfo CopyInfo) (dstDir string, content io.ReadCloser, err error) {
+	// Ensure in platform semantics
+	srcInfo.Path = normalizePath(srcInfo.Path)
+	dstInfo.Path = normalizePath(dstInfo.Path)
+
+	// Separate the destination path between its directory and base
+	// components in case the source archive contents need to be rebased.
+	dstDir, dstBase := SplitPathDirEntry(dstInfo.Path)
+	_, srcBase := SplitPathDirEntry(srcInfo.Path)
+
+	switch {
+	case dstInfo.Exists && dstInfo.IsDir:
+		// The destination exists as a directory. No alteration
+		// to srcContent is needed as its contents can be
+		// simply extracted to the destination directory.
+		return dstInfo.Path, ioutil.NopCloser(srcContent), nil
+	case dstInfo.Exists && srcInfo.IsDir:
+		// The destination exists as some type of file and the source
+		// content is a directory. This is an error condition since
+		// you cannot copy a directory to an existing file location.
+		return "", nil, ErrCannotCopyDir
+	case dstInfo.Exists:
+		// The destination exists as some type of file and the source content
+		// is also a file. The source content entry will have to be renamed to
+		// have a basename which matches the destination path's basename.
+		if len(srcInfo.RebaseName) != 0 {
+			srcBase = srcInfo.RebaseName
+		}
+		return dstDir, RebaseArchiveEntries(srcContent, srcBase, dstBase), nil
+	case srcInfo.IsDir:
+		// The destination does not exist and the source content is an archive
+		// of a directory. The archive should be extracted to the parent of
+		// the destination path instead, and when it is, the directory that is
+		// created as a result should take the name of the destination path.
+		// The source content entries will have to be renamed to have a
+		// basename which matches the destination path's basename.
+		if len(srcInfo.RebaseName) != 0 {
+			srcBase = srcInfo.RebaseName
+		}
+		return dstDir, RebaseArchiveEntries(srcContent, srcBase, dstBase), nil
+	case assertsDirectory(dstInfo.Path, os.PathSeparator):
+		// The destination does not exist and is asserted to be created as a
+		// directory, but the source content is not a directory. This is an
+		// error condition since you cannot create a directory from a file
+		// source.
+		return "", nil, ErrDirNotExists
+	default:
+		// The last remaining case is when the destination does not exist, is
+		// not asserted to be a directory, and the source content is not an
+		// archive of a directory. It this case, the destination file will need
+		// to be created when the archive is extracted and the source content
+		// entry will have to be renamed to have a basename which matches the
+		// destination path's basename.
+		if len(srcInfo.RebaseName) != 0 {
+			srcBase = srcInfo.RebaseName
+		}
+		return dstDir, RebaseArchiveEntries(srcContent, srcBase, dstBase), nil
+	}
+
+}
+
+// RebaseArchiveEntries rewrites the given srcContent archive replacing
+// an occurrence of oldBase with newBase at the beginning of entry names.
+func RebaseArchiveEntries(srcContent io.Reader, oldBase, newBase string) io.ReadCloser {
+	if oldBase == string(os.PathSeparator) {
+		// If oldBase specifies the root directory, use an empty string as
+		// oldBase instead so that newBase doesn't replace the path separator
+		// that all paths will start with.
+		oldBase = ""
+	}
+
+	rebased, w := io.Pipe()
+
+	go func() {
+		srcTar := tar.NewReader(srcContent)
+		rebasedTar := tar.NewWriter(w)
+
+		for {
+			hdr, err := srcTar.Next()
+			if err == io.EOF {
+				// Signals end of archive.
+				rebasedTar.Close()
+				w.Close()
+				return
+			}
+			if err != nil {
+				w.CloseWithError(err)
+				return
+			}
+
+			// srcContent tar stream, as served by TarWithOptions(), is
+			// definitely in PAX format, but tar.Next() mistakenly guesses it
+			// as USTAR, which creates a problem: if the newBase is >100
+			// characters long, WriteHeader() returns an error like
+			// "archive/tar: cannot encode header: Format specifies USTAR; and USTAR cannot encode Name=...".
+			//
+			// To fix, set the format to PAX here. See docker/for-linux issue #484.
+			hdr.Format = tar.FormatPAX
+			hdr.Name = strings.Replace(hdr.Name, oldBase, newBase, 1)
+			if hdr.Typeflag == tar.TypeLink {
+				hdr.Linkname = strings.Replace(hdr.Linkname, oldBase, newBase, 1)
+			}
+
+			if err = rebasedTar.WriteHeader(hdr); err != nil {
+				w.CloseWithError(err)
+				return
+			}
+
+			// Ignoring GoSec G110. See https://github.com/securego/gosec/pull/433
+			// and https://cure53.de/pentest-report_opa.pdf, which recommends to
+			// replace io.Copy with io.CopyN7. The latter allows to specify the
+			// maximum number of bytes that should be read. By properly defining
+			// the limit, it can be assured that a GZip compression bomb cannot
+			// easily cause a Denial-of-Service.
+			// After reviewing with @tonistiigi and @cpuguy83, this should not
+			// affect us, because here we do not read into memory, hence should
+			// not be vulnerable to this code consuming memory.
+			//nolint:gosec // G110: Potential DoS vulnerability via decompression bomb (gosec)
+			if _, err = io.Copy(rebasedTar, srcTar); err != nil {
+				w.CloseWithError(err)
+				return
+			}
+		}
+	}()
+
+	return rebased
+}
+
+// CopyResource performs an archive copy from the given source path to the
+// given destination path. The source path MUST exist and the destination
+// path's parent directory must exist.
+func CopyResource(srcPath, dstPath string, followLink bool) error {
+	var (
+		srcInfo CopyInfo
+		err     error
+	)
+
+	// Ensure in platform semantics
+	srcPath = normalizePath(srcPath)
+	dstPath = normalizePath(dstPath)
+
+	// Clean the source and destination paths.
+	srcPath = PreserveTrailingDotOrSeparator(filepath.Clean(srcPath), srcPath, os.PathSeparator)
+	dstPath = PreserveTrailingDotOrSeparator(filepath.Clean(dstPath), dstPath, os.PathSeparator)
+
+	if srcInfo, err = CopyInfoSourcePath(srcPath, followLink); err != nil {
+		return err
+	}
+
+	content, err := TarResource(srcInfo)
+	if err != nil {
+		return err
+	}
+	defer content.Close()
+
+	return CopyTo(content, srcInfo, dstPath)
+}
+
+// CopyTo handles extracting the given content whose
+// entries should be sourced from srcInfo to dstPath.
+func CopyTo(content io.Reader, srcInfo CopyInfo, dstPath string) error {
+	// The destination path need not exist, but CopyInfoDestinationPath will
+	// ensure that at least the parent directory exists.
+	dstInfo, err := CopyInfoDestinationPath(normalizePath(dstPath))
+	if err != nil {
+		return err
+	}
+
+	dstDir, copyArchive, err := PrepareArchiveCopy(content, srcInfo, dstInfo)
+	if err != nil {
+		return err
+	}
+	defer copyArchive.Close()
+
+	options := &TarOptions{
+		NoLchown:             true,
+		NoOverwriteDirNonDir: true,
+	}
+
+	return Untar(copyArchive, dstDir, options)
+}
+
+// ResolveHostSourcePath decides real path need to be copied with parameters such as
+// whether to follow symbol link or not, if followLink is true, resolvedPath will return
+// link target of any symbol link file, else it will only resolve symlink of directory
+// but return symbol link file itself without resolving.
+func ResolveHostSourcePath(path string, followLink bool) (resolvedPath, rebaseName string, err error) {
+	if followLink {
+		resolvedPath, err = filepath.EvalSymlinks(path)
+		if err != nil {
+			return
+		}
+
+		resolvedPath, rebaseName = GetRebaseName(path, resolvedPath)
+	} else {
+		dirPath, basePath := filepath.Split(path)
+
+		// if not follow symbol link, then resolve symbol link of parent dir
+		var resolvedDirPath string
+		resolvedDirPath, err = filepath.EvalSymlinks(dirPath)
+		if err != nil {
+			return
+		}
+		// resolvedDirPath will have been cleaned (no trailing path separators) so
+		// we can manually join it with the base path element.
+		resolvedPath = resolvedDirPath + string(filepath.Separator) + basePath
+		if hasTrailingPathSeparator(path, os.PathSeparator) &&
+			filepath.Base(path) != filepath.Base(resolvedPath) {
+			rebaseName = filepath.Base(path)
+		}
+	}
+	return resolvedPath, rebaseName, nil
+}
+
+// GetRebaseName normalizes and compares path and resolvedPath,
+// return completed resolved path and rebased file name
+func GetRebaseName(path, resolvedPath string) (string, string) {
+	// linkTarget will have been cleaned (no trailing path separators and dot) so
+	// we can manually join it with them
+	var rebaseName string
+	if specifiesCurrentDir(path) &&
+		!specifiesCurrentDir(resolvedPath) {
+		resolvedPath += string(filepath.Separator) + "."
+	}
+
+	if hasTrailingPathSeparator(path, os.PathSeparator) &&
+		!hasTrailingPathSeparator(resolvedPath, os.PathSeparator) {
+		resolvedPath += string(filepath.Separator)
+	}
+
+	if filepath.Base(path) != filepath.Base(resolvedPath) {
+		// In the case where the path had a trailing separator and a symlink
+		// evaluation has changed the last path component, we will need to
+		// rebase the name in the archive that is being copied to match the
+		// originally requested name.
+		rebaseName = filepath.Base(path)
+	}
+	return resolvedPath, rebaseName
+}

vendor/github.com/docker/docker/pkg/archive/copy_unix.go

@@ -0,0 +1,11 @@
+// +build !windows
+
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"path/filepath"
+)
+
+func normalizePath(path string) string {
+	return filepath.ToSlash(path)
+}

vendor/github.com/docker/docker/pkg/archive/copy_windows.go

@@ -0,0 +1,9 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"path/filepath"
+)
+
+func normalizePath(path string) string {
+	return filepath.FromSlash(path)
+}

vendor/github.com/docker/docker/pkg/archive/diff.go

@@ -0,0 +1,260 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"archive/tar"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strings"
+
+	"github.com/docker/docker/pkg/idtools"
+	"github.com/docker/docker/pkg/pools"
+	"github.com/docker/docker/pkg/system"
+	"github.com/sirupsen/logrus"
+)
+
+// UnpackLayer unpack `layer` to a `dest`. The stream `layer` can be
+// compressed or uncompressed.
+// Returns the size in bytes of the contents of the layer.
+func UnpackLayer(dest string, layer io.Reader, options *TarOptions) (size int64, err error) {
+	tr := tar.NewReader(layer)
+	trBuf := pools.BufioReader32KPool.Get(tr)
+	defer pools.BufioReader32KPool.Put(trBuf)
+
+	var dirs []*tar.Header
+	unpackedPaths := make(map[string]struct{})
+
+	if options == nil {
+		options = &TarOptions{}
+	}
+	if options.ExcludePatterns == nil {
+		options.ExcludePatterns = []string{}
+	}
+	idMapping := idtools.NewIDMappingsFromMaps(options.UIDMaps, options.GIDMaps)
+
+	aufsTempdir := ""
+	aufsHardlinks := make(map[string]*tar.Header)
+
+	// Iterate through the files in the archive.
+	for {
+		hdr, err := tr.Next()
+		if err == io.EOF {
+			// end of tar archive
+			break
+		}
+		if err != nil {
+			return 0, err
+		}
+
+		size += hdr.Size
+
+		// Normalize name, for safety and for a simple is-root check
+		hdr.Name = filepath.Clean(hdr.Name)
+
+		// Windows does not support filenames with colons in them. Ignore
+		// these files. This is not a problem though (although it might
+		// appear that it is). Let's suppose a client is running docker pull.
+		// The daemon it points to is Windows. Would it make sense for the
+		// client to be doing a docker pull Ubuntu for example (which has files
+		// with colons in the name under /usr/share/man/man3)? No, absolutely
+		// not as it would really only make sense that they were pulling a
+		// Windows image. However, for development, it is necessary to be able
+		// to pull Linux images which are in the repository.
+		//
+		// TODO Windows. Once the registry is aware of what images are Windows-
+		// specific or Linux-specific, this warning should be changed to an error
+		// to cater for the situation where someone does manage to upload a Linux
+		// image but have it tagged as Windows inadvertently.
+		if runtime.GOOS == "windows" {
+			if strings.Contains(hdr.Name, ":") {
+				logrus.Warnf("Windows: Ignoring %s (is this a Linux image?)", hdr.Name)
+				continue
+			}
+		}
+
+		// Note as these operations are platform specific, so must the slash be.
+		if !strings.HasSuffix(hdr.Name, string(os.PathSeparator)) {
+			// Not the root directory, ensure that the parent directory exists.
+			// This happened in some tests where an image had a tarfile without any
+			// parent directories.
+			parent := filepath.Dir(hdr.Name)
+			parentPath := filepath.Join(dest, parent)
+
+			if _, err := os.Lstat(parentPath); err != nil && os.IsNotExist(err) {
+				err = system.MkdirAll(parentPath, 0600)
+				if err != nil {
+					return 0, err
+				}
+			}
+		}
+
+		// Skip AUFS metadata dirs
+		if strings.HasPrefix(hdr.Name, WhiteoutMetaPrefix) {
+			// Regular files inside /.wh..wh.plnk can be used as hardlink targets
+			// We don't want this directory, but we need the files in them so that
+			// such hardlinks can be resolved.
+			if strings.HasPrefix(hdr.Name, WhiteoutLinkDir) && hdr.Typeflag == tar.TypeReg {
+				basename := filepath.Base(hdr.Name)
+				aufsHardlinks[basename] = hdr
+				if aufsTempdir == "" {
+					if aufsTempdir, err = ioutil.TempDir("", "dockerplnk"); err != nil {
+						return 0, err
+					}
+					defer os.RemoveAll(aufsTempdir)
+				}
+				if err := createTarFile(filepath.Join(aufsTempdir, basename), dest, hdr, tr, true, nil, options.InUserNS); err != nil {
+					return 0, err
+				}
+			}
+
+			if hdr.Name != WhiteoutOpaqueDir {
+				continue
+			}
+		}
+		path := filepath.Join(dest, hdr.Name)
+		rel, err := filepath.Rel(dest, path)
+		if err != nil {
+			return 0, err
+		}
+
+		// Note as these operations are platform specific, so must the slash be.
+		if strings.HasPrefix(rel, ".."+string(os.PathSeparator)) {
+			return 0, breakoutError(fmt.Errorf("%q is outside of %q", hdr.Name, dest))
+		}
+		base := filepath.Base(path)
+
+		if strings.HasPrefix(base, WhiteoutPrefix) {
+			dir := filepath.Dir(path)
+			if base == WhiteoutOpaqueDir {
+				_, err := os.Lstat(dir)
+				if err != nil {
+					return 0, err
+				}
+				err = filepath.Walk(dir, func(path string, info os.FileInfo, err error) error {
+					if err != nil {
+						if os.IsNotExist(err) {
+							err = nil // parent was deleted
+						}
+						return err
+					}
+					if path == dir {
+						return nil
+					}
+					if _, exists := unpackedPaths[path]; !exists {
+						err := os.RemoveAll(path)
+						return err
+					}
+					return nil
+				})
+				if err != nil {
+					return 0, err
+				}
+			} else {
+				originalBase := base[len(WhiteoutPrefix):]
+				originalPath := filepath.Join(dir, originalBase)
+				if err := os.RemoveAll(originalPath); err != nil {
+					return 0, err
+				}
+			}
+		} else {
+			// If path exits we almost always just want to remove and replace it.
+			// The only exception is when it is a directory *and* the file from
+			// the layer is also a directory. Then we want to merge them (i.e.
+			// just apply the metadata from the layer).
+			if fi, err := os.Lstat(path); err == nil {
+				if !(fi.IsDir() && hdr.Typeflag == tar.TypeDir) {
+					if err := os.RemoveAll(path); err != nil {
+						return 0, err
+					}
+				}
+			}
+
+			trBuf.Reset(tr)
+			srcData := io.Reader(trBuf)
+			srcHdr := hdr
+
+			// Hard links into /.wh..wh.plnk don't work, as we don't extract that directory, so
+			// we manually retarget these into the temporary files we extracted them into
+			if hdr.Typeflag == tar.TypeLink && strings.HasPrefix(filepath.Clean(hdr.Linkname), WhiteoutLinkDir) {
+				linkBasename := filepath.Base(hdr.Linkname)
+				srcHdr = aufsHardlinks[linkBasename]
+				if srcHdr == nil {
+					return 0, fmt.Errorf("Invalid aufs hardlink")
+				}
+				tmpFile, err := os.Open(filepath.Join(aufsTempdir, linkBasename))
+				if err != nil {
+					return 0, err
+				}
+				defer tmpFile.Close()
+				srcData = tmpFile
+			}
+
+			if err := remapIDs(idMapping, srcHdr); err != nil {
+				return 0, err
+			}
+
+			if err := createTarFile(path, dest, srcHdr, srcData, !options.NoLchown, nil, options.InUserNS); err != nil {
+				return 0, err
+			}
+
+			// Directory mtimes must be handled at the end to avoid further
+			// file creation in them to modify the directory mtime
+			if hdr.Typeflag == tar.TypeDir {
+				dirs = append(dirs, hdr)
+			}
+			unpackedPaths[path] = struct{}{}
+		}
+	}
+
+	for _, hdr := range dirs {
+		path := filepath.Join(dest, hdr.Name)
+		if err := system.Chtimes(path, hdr.AccessTime, hdr.ModTime); err != nil {
+			return 0, err
+		}
+	}
+
+	return size, nil
+}
+
+// ApplyLayer parses a diff in the standard layer format from `layer`,
+// and applies it to the directory `dest`. The stream `layer` can be
+// compressed or uncompressed.
+// Returns the size in bytes of the contents of the layer.
+func ApplyLayer(dest string, layer io.Reader) (int64, error) {
+	return applyLayerHandler(dest, layer, &TarOptions{}, true)
+}
+
+// ApplyUncompressedLayer parses a diff in the standard layer format from
+// `layer`, and applies it to the directory `dest`. The stream `layer`
+// can only be uncompressed.
+// Returns the size in bytes of the contents of the layer.
+func ApplyUncompressedLayer(dest string, layer io.Reader, options *TarOptions) (int64, error) {
+	return applyLayerHandler(dest, layer, options, false)
+}
+
+// do the bulk load of ApplyLayer, but allow for not calling DecompressStream
+func applyLayerHandler(dest string, layer io.Reader, options *TarOptions, decompress bool) (int64, error) {
+	dest = filepath.Clean(dest)
+
+	// We need to be able to set any perms
+	if runtime.GOOS != "windows" {
+		oldmask, err := system.Umask(0)
+		if err != nil {
+			return 0, err
+		}
+		defer system.Umask(oldmask)
+	}
+
+	if decompress {
+		decompLayer, err := DecompressStream(layer)
+		if err != nil {
+			return 0, err
+		}
+		defer decompLayer.Close()
+		layer = decompLayer
+	}
+	return UnpackLayer(dest, layer, options)
+}

vendor/github.com/docker/docker/pkg/archive/time_linux.go

@@ -0,0 +1,16 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"syscall"
+	"time"
+)
+
+func timeToTimespec(time time.Time) (ts syscall.Timespec) {
+	if time.IsZero() {
+		// Return UTIME_OMIT special value
+		ts.Sec = 0
+		ts.Nsec = (1 << 30) - 2
+		return
+	}
+	return syscall.NsecToTimespec(time.UnixNano())
+}

vendor/github.com/docker/docker/pkg/archive/time_unsupported.go

@@ -0,0 +1,16 @@
+// +build !linux
+
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"syscall"
+	"time"
+)
+
+func timeToTimespec(time time.Time) (ts syscall.Timespec) {
+	nsec := int64(0)
+	if !time.IsZero() {
+		nsec = time.UnixNano()
+	}
+	return syscall.NsecToTimespec(nsec)
+}

vendor/github.com/docker/docker/pkg/archive/whiteouts.go

@@ -0,0 +1,23 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+// Whiteouts are files with a special meaning for the layered filesystem.
+// Docker uses AUFS whiteout files inside exported archives. In other
+// filesystems these files are generated/handled on tar creation/extraction.
+
+// WhiteoutPrefix prefix means file is a whiteout. If this is followed by a
+// filename this means that file has been removed from the base layer.
+const WhiteoutPrefix = ".wh."
+
+// WhiteoutMetaPrefix prefix means whiteout has a special meaning and is not
+// for removing an actual file. Normally these files are excluded from exported
+// archives.
+const WhiteoutMetaPrefix = WhiteoutPrefix + WhiteoutPrefix
+
+// WhiteoutLinkDir is a directory AUFS uses for storing hardlink links to other
+// layers. Normally these should not go into exported archives and all changed
+// hardlinks should be copied to the top layer.
+const WhiteoutLinkDir = WhiteoutMetaPrefix + "plnk"
+
+// WhiteoutOpaqueDir file means directory has been made opaque - meaning
+// readdir calls to this directory do not follow to lower layers.
+const WhiteoutOpaqueDir = WhiteoutMetaPrefix + ".opq"

vendor/github.com/docker/docker/pkg/archive/wrap.go

@@ -0,0 +1,59 @@
+package archive // import "github.com/docker/docker/pkg/archive"
+
+import (
+	"archive/tar"
+	"bytes"
+	"io"
+)
+
+// Generate generates a new archive from the content provided
+// as input.
+//
+// `files` is a sequence of path/content pairs. A new file is
+// added to the archive for each pair.
+// If the last pair is incomplete, the file is created with an
+// empty content. For example:
+//
+// Generate("foo.txt", "hello world", "emptyfile")
+//
+// The above call will return an archive with 2 files:
+//  * ./foo.txt with content "hello world"
+//  * ./empty with empty content
+//
+// FIXME: stream content instead of buffering
+// FIXME: specify permissions and other archive metadata
+func Generate(input ...string) (io.Reader, error) {
+	files := parseStringPairs(input...)
+	buf := new(bytes.Buffer)
+	tw := tar.NewWriter(buf)
+	for _, file := range files {
+		name, content := file[0], file[1]
+		hdr := &tar.Header{
+			Name: name,
+			Size: int64(len(content)),
+		}
+		if err := tw.WriteHeader(hdr); err != nil {
+			return nil, err
+		}
+		if _, err := tw.Write([]byte(content)); err != nil {
+			return nil, err
+		}
+	}
+	if err := tw.Close(); err != nil {
+		return nil, err
+	}
+	return buf, nil
+}
+
+func parseStringPairs(input ...string) (output [][2]string) {
+	output = make([][2]string, 0, len(input)/2+1)
+	for i := 0; i < len(input); i += 2 {
+		var pair [2]string
+		pair[0] = input[i]
+		if i+1 < len(input) {
+			pair[1] = input[i+1]
+		}
+		output = append(output, pair)
+	}
+	return
+}

vendor/github.com/docker/docker/pkg/idtools/idtools.go

@@ -0,0 +1,241 @@
+package idtools // import "github.com/docker/docker/pkg/idtools"
+
+import (
+	"bufio"
+	"fmt"
+	"os"
+	"strconv"
+	"strings"
+)
+
+// IDMap contains a single entry for user namespace range remapping. An array
+// of IDMap entries represents the structure that will be provided to the Linux
+// kernel for creating a user namespace.
+type IDMap struct {
+	ContainerID int `json:"container_id"`
+	HostID      int `json:"host_id"`
+	Size        int `json:"size"`
+}
+
+type subIDRange struct {
+	Start  int
+	Length int
+}
+
+type ranges []subIDRange
+
+func (e ranges) Len() int           { return len(e) }
+func (e ranges) Swap(i, j int)      { e[i], e[j] = e[j], e[i] }
+func (e ranges) Less(i, j int) bool { return e[i].Start < e[j].Start }
+
+const (
+	subuidFileName = "/etc/subuid"
+	subgidFileName = "/etc/subgid"
+)
+
+// MkdirAllAndChown creates a directory (include any along the path) and then modifies
+// ownership to the requested uid/gid.  If the directory already exists, this
+// function will still change ownership and permissions.
+func MkdirAllAndChown(path string, mode os.FileMode, owner Identity) error {
+	return mkdirAs(path, mode, owner, true, true)
+}
+
+// MkdirAndChown creates a directory and then modifies ownership to the requested uid/gid.
+// If the directory already exists, this function still changes ownership and permissions.
+// Note that unlike os.Mkdir(), this function does not return IsExist error
+// in case path already exists.
+func MkdirAndChown(path string, mode os.FileMode, owner Identity) error {
+	return mkdirAs(path, mode, owner, false, true)
+}
+
+// MkdirAllAndChownNew creates a directory (include any along the path) and then modifies
+// ownership ONLY of newly created directories to the requested uid/gid. If the
+// directories along the path exist, no change of ownership or permissions will be performed
+func MkdirAllAndChownNew(path string, mode os.FileMode, owner Identity) error {
+	return mkdirAs(path, mode, owner, true, false)
+}
+
+// GetRootUIDGID retrieves the remapped root uid/gid pair from the set of maps.
+// If the maps are empty, then the root uid/gid will default to "real" 0/0
+func GetRootUIDGID(uidMap, gidMap []IDMap) (int, int, error) {
+	uid, err := toHost(0, uidMap)
+	if err != nil {
+		return -1, -1, err
+	}
+	gid, err := toHost(0, gidMap)
+	if err != nil {
+		return -1, -1, err
+	}
+	return uid, gid, nil
+}
+
+// toContainer takes an id mapping, and uses it to translate a
+// host ID to the remapped ID. If no map is provided, then the translation
+// assumes a 1-to-1 mapping and returns the passed in id
+func toContainer(hostID int, idMap []IDMap) (int, error) {
+	if idMap == nil {
+		return hostID, nil
+	}
+	for _, m := range idMap {
+		if (hostID >= m.HostID) && (hostID <= (m.HostID + m.Size - 1)) {
+			contID := m.ContainerID + (hostID - m.HostID)
+			return contID, nil
+		}
+	}
+	return -1, fmt.Errorf("Host ID %d cannot be mapped to a container ID", hostID)
+}
+
+// toHost takes an id mapping and a remapped ID, and translates the
+// ID to the mapped host ID. If no map is provided, then the translation
+// assumes a 1-to-1 mapping and returns the passed in id #
+func toHost(contID int, idMap []IDMap) (int, error) {
+	if idMap == nil {
+		return contID, nil
+	}
+	for _, m := range idMap {
+		if (contID >= m.ContainerID) && (contID <= (m.ContainerID + m.Size - 1)) {
+			hostID := m.HostID + (contID - m.ContainerID)
+			return hostID, nil
+		}
+	}
+	return -1, fmt.Errorf("Container ID %d cannot be mapped to a host ID", contID)
+}
+
+// Identity is either a UID and GID pair or a SID (but not both)
+type Identity struct {
+	UID int
+	GID int
+	SID string
+}
+
+// IdentityMapping contains a mappings of UIDs and GIDs
+type IdentityMapping struct {
+	uids []IDMap
+	gids []IDMap
+}
+
+// NewIDMappingsFromMaps creates a new mapping from two slices
+// Deprecated: this is a temporary shim while transitioning to IDMapping
+func NewIDMappingsFromMaps(uids []IDMap, gids []IDMap) *IdentityMapping {
+	return &IdentityMapping{uids: uids, gids: gids}
+}
+
+// RootPair returns a uid and gid pair for the root user. The error is ignored
+// because a root user always exists, and the defaults are correct when the uid
+// and gid maps are empty.
+func (i *IdentityMapping) RootPair() Identity {
+	uid, gid, _ := GetRootUIDGID(i.uids, i.gids)
+	return Identity{UID: uid, GID: gid}
+}
+
+// ToHost returns the host UID and GID for the container uid, gid.
+// Remapping is only performed if the ids aren't already the remapped root ids
+func (i *IdentityMapping) ToHost(pair Identity) (Identity, error) {
+	var err error
+	target := i.RootPair()
+
+	if pair.UID != target.UID {
+		target.UID, err = toHost(pair.UID, i.uids)
+		if err != nil {
+			return target, err
+		}
+	}
+
+	if pair.GID != target.GID {
+		target.GID, err = toHost(pair.GID, i.gids)
+	}
+	return target, err
+}
+
+// ToContainer returns the container UID and GID for the host uid and gid
+func (i *IdentityMapping) ToContainer(pair Identity) (int, int, error) {
+	uid, err := toContainer(pair.UID, i.uids)
+	if err != nil {
+		return -1, -1, err
+	}
+	gid, err := toContainer(pair.GID, i.gids)
+	return uid, gid, err
+}
+
+// Empty returns true if there are no id mappings
+func (i *IdentityMapping) Empty() bool {
+	return len(i.uids) == 0 && len(i.gids) == 0
+}
+
+// UIDs return the UID mapping
+// TODO: remove this once everything has been refactored to use pairs
+func (i *IdentityMapping) UIDs() []IDMap {
+	return i.uids
+}
+
+// GIDs return the UID mapping
+// TODO: remove this once everything has been refactored to use pairs
+func (i *IdentityMapping) GIDs() []IDMap {
+	return i.gids
+}
+
+func createIDMap(subidRanges ranges) []IDMap {
+	idMap := []IDMap{}
+
+	containerID := 0
+	for _, idrange := range subidRanges {
+		idMap = append(idMap, IDMap{
+			ContainerID: containerID,
+			HostID:      idrange.Start,
+			Size:        idrange.Length,
+		})
+		containerID = containerID + idrange.Length
+	}
+	return idMap
+}
+
+func parseSubuid(username string) (ranges, error) {
+	return parseSubidFile(subuidFileName, username)
+}
+
+func parseSubgid(username string) (ranges, error) {
+	return parseSubidFile(subgidFileName, username)
+}
+
+// parseSubidFile will read the appropriate file (/etc/subuid or /etc/subgid)
+// and return all found ranges for a specified username. If the special value
+// "ALL" is supplied for username, then all ranges in the file will be returned
+func parseSubidFile(path, username string) (ranges, error) {
+	var rangeList ranges
+
+	subidFile, err := os.Open(path)
+	if err != nil {
+		return rangeList, err
+	}
+	defer subidFile.Close()
+
+	s := bufio.NewScanner(subidFile)
+	for s.Scan() {
+		text := strings.TrimSpace(s.Text())
+		if text == "" || strings.HasPrefix(text, "#") {
+			continue
+		}
+		parts := strings.Split(text, ":")
+		if len(parts) != 3 {
+			return rangeList, fmt.Errorf("Cannot parse subuid/gid information: Format not correct for %s file", path)
+		}
+		if parts[0] == username || username == "ALL" {
+			startid, err := strconv.Atoi(parts[1])
+			if err != nil {
+				return rangeList, fmt.Errorf("String to int conversion failed during subuid/gid parsing of %s: %v", path, err)
+			}
+			length, err := strconv.Atoi(parts[2])
+			if err != nil {
+				return rangeList, fmt.Errorf("String to int conversion failed during subuid/gid parsing of %s: %v", path, err)
+			}
+			rangeList = append(rangeList, subIDRange{startid, length})
+		}
+	}
+
+	return rangeList, s.Err()
+}
+
+// CurrentIdentity returns the identity of the current process
+func CurrentIdentity() Identity {
+	return Identity{UID: os.Getuid(), GID: os.Getegid()}
+}

vendor/github.com/docker/docker/pkg/idtools/idtools_unix.go

@@ -0,0 +1,295 @@
+// +build !windows
+
+package idtools // import "github.com/docker/docker/pkg/idtools"
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strconv"
+	"sync"
+	"syscall"
+
+	"github.com/docker/docker/pkg/system"
+	"github.com/opencontainers/runc/libcontainer/user"
+	"github.com/pkg/errors"
+)
+
+var (
+	entOnce   sync.Once
+	getentCmd string
+)
+
+func mkdirAs(path string, mode os.FileMode, owner Identity, mkAll, chownExisting bool) error {
+	// make an array containing the original path asked for, plus (for mkAll == true)
+	// all path components leading up to the complete path that don't exist before we MkdirAll
+	// so that we can chown all of them properly at the end.  If chownExisting is false, we won't
+	// chown the full directory path if it exists
+
+	var paths []string
+
+	stat, err := system.Stat(path)
+	if err == nil {
+		if !stat.IsDir() {
+			return &os.PathError{Op: "mkdir", Path: path, Err: syscall.ENOTDIR}
+		}
+		if !chownExisting {
+			return nil
+		}
+
+		// short-circuit--we were called with an existing directory and chown was requested
+		return setPermissions(path, mode, owner.UID, owner.GID, stat)
+	}
+
+	if os.IsNotExist(err) {
+		paths = []string{path}
+	}
+
+	if mkAll {
+		// walk back to "/" looking for directories which do not exist
+		// and add them to the paths array for chown after creation
+		dirPath := path
+		for {
+			dirPath = filepath.Dir(dirPath)
+			if dirPath == "/" {
+				break
+			}
+			if _, err := os.Stat(dirPath); err != nil && os.IsNotExist(err) {
+				paths = append(paths, dirPath)
+			}
+		}
+		if err := system.MkdirAll(path, mode); err != nil {
+			return err
+		}
+	} else {
+		if err := os.Mkdir(path, mode); err != nil && !os.IsExist(err) {
+			return err
+		}
+	}
+	// even if it existed, we will chown the requested path + any subpaths that
+	// didn't exist when we called MkdirAll
+	for _, pathComponent := range paths {
+		if err := setPermissions(pathComponent, mode, owner.UID, owner.GID, nil); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// CanAccess takes a valid (existing) directory and a uid, gid pair and determines
+// if that uid, gid pair has access (execute bit) to the directory
+func CanAccess(path string, pair Identity) bool {
+	statInfo, err := system.Stat(path)
+	if err != nil {
+		return false
+	}
+	fileMode := os.FileMode(statInfo.Mode())
+	permBits := fileMode.Perm()
+	return accessible(statInfo.UID() == uint32(pair.UID),
+		statInfo.GID() == uint32(pair.GID), permBits)
+}
+
+func accessible(isOwner, isGroup bool, perms os.FileMode) bool {
+	if isOwner && (perms&0100 == 0100) {
+		return true
+	}
+	if isGroup && (perms&0010 == 0010) {
+		return true
+	}
+	if perms&0001 == 0001 {
+		return true
+	}
+	return false
+}
+
+// LookupUser uses traditional local system files lookup (from libcontainer/user) on a username,
+// followed by a call to `getent` for supporting host configured non-files passwd and group dbs
+func LookupUser(name string) (user.User, error) {
+	// first try a local system files lookup using existing capabilities
+	usr, err := user.LookupUser(name)
+	if err == nil {
+		return usr, nil
+	}
+	// local files lookup failed; attempt to call `getent` to query configured passwd dbs
+	usr, err = getentUser(name)
+	if err != nil {
+		return user.User{}, err
+	}
+	return usr, nil
+}
+
+// LookupUID uses traditional local system files lookup (from libcontainer/user) on a uid,
+// followed by a call to `getent` for supporting host configured non-files passwd and group dbs
+func LookupUID(uid int) (user.User, error) {
+	// first try a local system files lookup using existing capabilities
+	usr, err := user.LookupUid(uid)
+	if err == nil {
+		return usr, nil
+	}
+	// local files lookup failed; attempt to call `getent` to query configured passwd dbs
+	return getentUser(strconv.Itoa(uid))
+}
+
+func getentUser(name string) (user.User, error) {
+	reader, err := callGetent("passwd", name)
+	if err != nil {
+		return user.User{}, err
+	}
+	users, err := user.ParsePasswd(reader)
+	if err != nil {
+		return user.User{}, err
+	}
+	if len(users) == 0 {
+		return user.User{}, fmt.Errorf("getent failed to find passwd entry for %q", name)
+	}
+	return users[0], nil
+}
+
+// LookupGroup uses traditional local system files lookup (from libcontainer/user) on a group name,
+// followed by a call to `getent` for supporting host configured non-files passwd and group dbs
+func LookupGroup(name string) (user.Group, error) {
+	// first try a local system files lookup using existing capabilities
+	group, err := user.LookupGroup(name)
+	if err == nil {
+		return group, nil
+	}
+	// local files lookup failed; attempt to call `getent` to query configured group dbs
+	return getentGroup(name)
+}
+
+// LookupGID uses traditional local system files lookup (from libcontainer/user) on a group ID,
+// followed by a call to `getent` for supporting host configured non-files passwd and group dbs
+func LookupGID(gid int) (user.Group, error) {
+	// first try a local system files lookup using existing capabilities
+	group, err := user.LookupGid(gid)
+	if err == nil {
+		return group, nil
+	}
+	// local files lookup failed; attempt to call `getent` to query configured group dbs
+	return getentGroup(strconv.Itoa(gid))
+}
+
+func getentGroup(name string) (user.Group, error) {
+	reader, err := callGetent("group", name)
+	if err != nil {
+		return user.Group{}, err
+	}
+	groups, err := user.ParseGroup(reader)
+	if err != nil {
+		return user.Group{}, err
+	}
+	if len(groups) == 0 {
+		return user.Group{}, fmt.Errorf("getent failed to find groups entry for %q", name)
+	}
+	return groups[0], nil
+}
+
+func callGetent(database, key string) (io.Reader, error) {
+	entOnce.Do(func() { getentCmd, _ = resolveBinary("getent") })
+	// if no `getent` command on host, can't do anything else
+	if getentCmd == "" {
+		return nil, fmt.Errorf("unable to find getent command")
+	}
+	out, err := execCmd(getentCmd, database, key)
+	if err != nil {
+		exitCode, errC := system.GetExitCode(err)
+		if errC != nil {
+			return nil, err
+		}
+		switch exitCode {
+		case 1:
+			return nil, fmt.Errorf("getent reported invalid parameters/database unknown")
+		case 2:
+			return nil, fmt.Errorf("getent unable to find entry %q in %s database", key, database)
+		case 3:
+			return nil, fmt.Errorf("getent database doesn't support enumeration")
+		default:
+			return nil, err
+		}
+
+	}
+	return bytes.NewReader(out), nil
+}
+
+// setPermissions performs a chown/chmod only if the uid/gid don't match what's requested
+// Normally a Chown is a no-op if uid/gid match, but in some cases this can still cause an error, e.g. if the
+// dir is on an NFS share, so don't call chown unless we absolutely must.
+// Likewise for setting permissions.
+func setPermissions(p string, mode os.FileMode, uid, gid int, stat *system.StatT) error {
+	if stat == nil {
+		var err error
+		stat, err = system.Stat(p)
+		if err != nil {
+			return err
+		}
+	}
+	if os.FileMode(stat.Mode()).Perm() != mode.Perm() {
+		if err := os.Chmod(p, mode.Perm()); err != nil {
+			return err
+		}
+	}
+	if stat.UID() == uint32(uid) && stat.GID() == uint32(gid) {
+		return nil
+	}
+	return os.Chown(p, uid, gid)
+}
+
+// NewIdentityMapping takes a requested username and
+// using the data from /etc/sub{uid,gid} ranges, creates the
+// proper uid and gid remapping ranges for that user/group pair
+func NewIdentityMapping(name string) (*IdentityMapping, error) {
+	usr, err := LookupUser(name)
+	if err != nil {
+		return nil, fmt.Errorf("Could not get user for username %s: %v", name, err)
+	}
+
+	subuidRanges, err := lookupSubUIDRanges(usr)
+	if err != nil {
+		return nil, err
+	}
+	subgidRanges, err := lookupSubGIDRanges(usr)
+	if err != nil {
+		return nil, err
+	}
+
+	return &IdentityMapping{
+		uids: subuidRanges,
+		gids: subgidRanges,
+	}, nil
+}
+
+func lookupSubUIDRanges(usr user.User) ([]IDMap, error) {
+	rangeList, err := parseSubuid(strconv.Itoa(usr.Uid))
+	if err != nil {
+		return nil, err
+	}
+	if len(rangeList) == 0 {
+		rangeList, err = parseSubuid(usr.Name)
+		if err != nil {
+			return nil, err
+		}
+	}
+	if len(rangeList) == 0 {
+		return nil, errors.Errorf("no subuid ranges found for user %q", usr.Name)
+	}
+	return createIDMap(rangeList), nil
+}
+
+func lookupSubGIDRanges(usr user.User) ([]IDMap, error) {
+	rangeList, err := parseSubgid(strconv.Itoa(usr.Uid))
+	if err != nil {
+		return nil, err
+	}
+	if len(rangeList) == 0 {
+		rangeList, err = parseSubgid(usr.Name)
+		if err != nil {
+			return nil, err
+		}
+	}
+	if len(rangeList) == 0 {
+		return nil, errors.Errorf("no subgid ranges found for user %q", usr.Name)
+	}
+	return createIDMap(rangeList), nil
+}

vendor/github.com/docker/docker/pkg/idtools/idtools_windows.go

@@ -0,0 +1,34 @@
+package idtools // import "github.com/docker/docker/pkg/idtools"
+
+import (
+	"os"
+
+	"github.com/docker/docker/pkg/system"
+)
+
+const (
+	SeTakeOwnershipPrivilege = "SeTakeOwnershipPrivilege"
+)
+
+const (
+	ContainerAdministratorSidString = "S-1-5-93-2-1"
+	ContainerUserSidString          = "S-1-5-93-2-2"
+)
+
+// This is currently a wrapper around MkdirAll, however, since currently
+// permissions aren't set through this path, the identity isn't utilized.
+// Ownership is handled elsewhere, but in the future could be support here
+// too.
+func mkdirAs(path string, mode os.FileMode, owner Identity, mkAll, chownExisting bool) error {
+	if err := system.MkdirAll(path, mode); err != nil {
+		return err
+	}
+	return nil
+}
+
+// CanAccess takes a valid (existing) directory and a uid, gid pair and determines
+// if that uid, gid pair has access (execute bit) to the directory
+// Windows does not require/support this function, so always return true
+func CanAccess(path string, identity Identity) bool {
+	return true
+}

vendor/github.com/docker/docker/pkg/idtools/usergroupadd_linux.go

@@ -0,0 +1,164 @@
+package idtools // import "github.com/docker/docker/pkg/idtools"
+
+import (
+	"fmt"
+	"regexp"
+	"sort"
+	"strconv"
+	"strings"
+	"sync"
+)
+
+// add a user and/or group to Linux /etc/passwd, /etc/group using standard
+// Linux distribution commands:
+// adduser --system --shell /bin/false --disabled-login --disabled-password --no-create-home --group <username>
+// useradd -r -s /bin/false <username>
+
+var (
+	once        sync.Once
+	userCommand string
+	idOutRegexp = regexp.MustCompile(`uid=([0-9]+).*gid=([0-9]+)`)
+)
+
+const (
+	// default length for a UID/GID subordinate range
+	defaultRangeLen   = 65536
+	defaultRangeStart = 100000
+)
+
+// AddNamespaceRangesUser takes a username and uses the standard system
+// utility to create a system user/group pair used to hold the
+// /etc/sub{uid,gid} ranges which will be used for user namespace
+// mapping ranges in containers.
+func AddNamespaceRangesUser(name string) (int, int, error) {
+	if err := addUser(name); err != nil {
+		return -1, -1, fmt.Errorf("Error adding user %q: %v", name, err)
+	}
+
+	// Query the system for the created uid and gid pair
+	out, err := execCmd("id", name)
+	if err != nil {
+		return -1, -1, fmt.Errorf("Error trying to find uid/gid for new user %q: %v", name, err)
+	}
+	matches := idOutRegexp.FindStringSubmatch(strings.TrimSpace(string(out)))
+	if len(matches) != 3 {
+		return -1, -1, fmt.Errorf("Can't find uid, gid from `id` output: %q", string(out))
+	}
+	uid, err := strconv.Atoi(matches[1])
+	if err != nil {
+		return -1, -1, fmt.Errorf("Can't convert found uid (%s) to int: %v", matches[1], err)
+	}
+	gid, err := strconv.Atoi(matches[2])
+	if err != nil {
+		return -1, -1, fmt.Errorf("Can't convert found gid (%s) to int: %v", matches[2], err)
+	}
+
+	// Now we need to create the subuid/subgid ranges for our new user/group (system users
+	// do not get auto-created ranges in subuid/subgid)
+
+	if err := createSubordinateRanges(name); err != nil {
+		return -1, -1, fmt.Errorf("Couldn't create subordinate ID ranges: %v", err)
+	}
+	return uid, gid, nil
+}
+
+func addUser(name string) error {
+	once.Do(func() {
+		// set up which commands are used for adding users/groups dependent on distro
+		if _, err := resolveBinary("adduser"); err == nil {
+			userCommand = "adduser"
+		} else if _, err := resolveBinary("useradd"); err == nil {
+			userCommand = "useradd"
+		}
+	})
+	var args []string
+	switch userCommand {
+	case "adduser":
+		args = []string{"--system", "--shell", "/bin/false", "--no-create-home", "--disabled-login", "--disabled-password", "--group", name}
+	case "useradd":
+		args = []string{"-r", "-s", "/bin/false", name}
+	default:
+		return fmt.Errorf("cannot add user; no useradd/adduser binary found")
+	}
+
+	if out, err := execCmd(userCommand, args...); err != nil {
+		return fmt.Errorf("failed to add user with error: %v; output: %q", err, string(out))
+	}
+	return nil
+}
+
+func createSubordinateRanges(name string) error {
+
+	// first, we should verify that ranges weren't automatically created
+	// by the distro tooling
+	ranges, err := parseSubuid(name)
+	if err != nil {
+		return fmt.Errorf("Error while looking for subuid ranges for user %q: %v", name, err)
+	}
+	if len(ranges) == 0 {
+		// no UID ranges; let's create one
+		startID, err := findNextUIDRange()
+		if err != nil {
+			return fmt.Errorf("Can't find available subuid range: %v", err)
+		}
+		out, err := execCmd("usermod", "-v", fmt.Sprintf("%d-%d", startID, startID+defaultRangeLen-1), name)
+		if err != nil {
+			return fmt.Errorf("Unable to add subuid range to user: %q; output: %s, err: %v", name, out, err)
+		}
+	}
+
+	ranges, err = parseSubgid(name)
+	if err != nil {
+		return fmt.Errorf("Error while looking for subgid ranges for user %q: %v", name, err)
+	}
+	if len(ranges) == 0 {
+		// no GID ranges; let's create one
+		startID, err := findNextGIDRange()
+		if err != nil {
+			return fmt.Errorf("Can't find available subgid range: %v", err)
+		}
+		out, err := execCmd("usermod", "-w", fmt.Sprintf("%d-%d", startID, startID+defaultRangeLen-1), name)
+		if err != nil {
+			return fmt.Errorf("Unable to add subgid range to user: %q; output: %s, err: %v", name, out, err)
+		}
+	}
+	return nil
+}
+
+func findNextUIDRange() (int, error) {
+	ranges, err := parseSubuid("ALL")
+	if err != nil {
+		return -1, fmt.Errorf("Couldn't parse all ranges in /etc/subuid file: %v", err)
+	}
+	sort.Sort(ranges)
+	return findNextRangeStart(ranges)
+}
+
+func findNextGIDRange() (int, error) {
+	ranges, err := parseSubgid("ALL")
+	if err != nil {
+		return -1, fmt.Errorf("Couldn't parse all ranges in /etc/subgid file: %v", err)
+	}
+	sort.Sort(ranges)
+	return findNextRangeStart(ranges)
+}
+
+func findNextRangeStart(rangeList ranges) (int, error) {
+	startID := defaultRangeStart
+	for _, arange := range rangeList {
+		if wouldOverlap(arange, startID) {
+			startID = arange.Start + arange.Length
+		}
+	}
+	return startID, nil
+}
+
+func wouldOverlap(arange subIDRange, ID int) bool {
+	low := ID
+	high := ID + defaultRangeLen
+	if (low >= arange.Start && low <= arange.Start+arange.Length) ||
+		(high <= arange.Start+arange.Length && high >= arange.Start) {
+		return true
+	}
+	return false
+}

vendor/github.com/docker/docker/pkg/idtools/usergroupadd_unsupported.go

@@ -0,0 +1,12 @@
+// +build !linux
+
+package idtools // import "github.com/docker/docker/pkg/idtools"
+
+import "fmt"
+
+// AddNamespaceRangesUser takes a name and finds an unused uid, gid pair
+// and calls the appropriate helper function to add the group and then
+// the user to the group in /etc/group and /etc/passwd respectively.
+func AddNamespaceRangesUser(name string) (int, int, error) {
+	return -1, -1, fmt.Errorf("No support for adding users or groups on this OS")
+}

vendor/github.com/docker/docker/pkg/idtools/utils_unix.go

@@ -0,0 +1,31 @@
+// +build !windows
+
+package idtools // import "github.com/docker/docker/pkg/idtools"
+
+import (
+	"fmt"
+	"os/exec"
+	"path/filepath"
+)
+
+func resolveBinary(binname string) (string, error) {
+	binaryPath, err := exec.LookPath(binname)
+	if err != nil {
+		return "", err
+	}
+	resolvedPath, err := filepath.EvalSymlinks(binaryPath)
+	if err != nil {
+		return "", err
+	}
+	// only return no error if the final resolved binary basename
+	// matches what was searched for
+	if filepath.Base(resolvedPath) == binname {
+		return resolvedPath, nil
+	}
+	return "", fmt.Errorf("Binary %q does not resolve to a binary of that name in $PATH (%q)", binname, resolvedPath)
+}
+
+func execCmd(cmd string, arg ...string) ([]byte, error) {
+	execCmd := exec.Command(cmd, arg...)
+	return execCmd.CombinedOutput()
+}

vendor/github.com/docker/docker/pkg/pools/pools.go

@@ -0,0 +1,137 @@
+// Package pools provides a collection of pools which provide various
+// data types with buffers. These can be used to lower the number of
+// memory allocations and reuse buffers.
+//
+// New pools should be added to this package to allow them to be
+// shared across packages.
+//
+// Utility functions which operate on pools should be added to this
+// package to allow them to be reused.
+package pools // import "github.com/docker/docker/pkg/pools"
+
+import (
+	"bufio"
+	"io"
+	"sync"
+
+	"github.com/docker/docker/pkg/ioutils"
+)
+
+const buffer32K = 32 * 1024
+
+var (
+	// BufioReader32KPool is a pool which returns bufio.Reader with a 32K buffer.
+	BufioReader32KPool = newBufioReaderPoolWithSize(buffer32K)
+	// BufioWriter32KPool is a pool which returns bufio.Writer with a 32K buffer.
+	BufioWriter32KPool = newBufioWriterPoolWithSize(buffer32K)
+	buffer32KPool      = newBufferPoolWithSize(buffer32K)
+)
+
+// BufioReaderPool is a bufio reader that uses sync.Pool.
+type BufioReaderPool struct {
+	pool sync.Pool
+}
+
+// newBufioReaderPoolWithSize is unexported because new pools should be
+// added here to be shared where required.
+func newBufioReaderPoolWithSize(size int) *BufioReaderPool {
+	return &BufioReaderPool{
+		pool: sync.Pool{
+			New: func() interface{} { return bufio.NewReaderSize(nil, size) },
+		},
+	}
+}
+
+// Get returns a bufio.Reader which reads from r. The buffer size is that of the pool.
+func (bufPool *BufioReaderPool) Get(r io.Reader) *bufio.Reader {
+	buf := bufPool.pool.Get().(*bufio.Reader)
+	buf.Reset(r)
+	return buf
+}
+
+// Put puts the bufio.Reader back into the pool.
+func (bufPool *BufioReaderPool) Put(b *bufio.Reader) {
+	b.Reset(nil)
+	bufPool.pool.Put(b)
+}
+
+type bufferPool struct {
+	pool sync.Pool
+}
+
+func newBufferPoolWithSize(size int) *bufferPool {
+	return &bufferPool{
+		pool: sync.Pool{
+			New: func() interface{} { s := make([]byte, size); return &s },
+		},
+	}
+}
+
+func (bp *bufferPool) Get() *[]byte {
+	return bp.pool.Get().(*[]byte)
+}
+
+func (bp *bufferPool) Put(b *[]byte) {
+	bp.pool.Put(b)
+}
+
+// Copy is a convenience wrapper which uses a buffer to avoid allocation in io.Copy.
+func Copy(dst io.Writer, src io.Reader) (written int64, err error) {
+	buf := buffer32KPool.Get()
+	written, err = io.CopyBuffer(dst, src, *buf)
+	buffer32KPool.Put(buf)
+	return
+}
+
+// NewReadCloserWrapper returns a wrapper which puts the bufio.Reader back
+// into the pool and closes the reader if it's an io.ReadCloser.
+func (bufPool *BufioReaderPool) NewReadCloserWrapper(buf *bufio.Reader, r io.Reader) io.ReadCloser {
+	return ioutils.NewReadCloserWrapper(r, func() error {
+		if readCloser, ok := r.(io.ReadCloser); ok {
+			readCloser.Close()
+		}
+		bufPool.Put(buf)
+		return nil
+	})
+}
+
+// BufioWriterPool is a bufio writer that uses sync.Pool.
+type BufioWriterPool struct {
+	pool sync.Pool
+}
+
+// newBufioWriterPoolWithSize is unexported because new pools should be
+// added here to be shared where required.
+func newBufioWriterPoolWithSize(size int) *BufioWriterPool {
+	return &BufioWriterPool{
+		pool: sync.Pool{
+			New: func() interface{} { return bufio.NewWriterSize(nil, size) },
+		},
+	}
+}
+
+// Get returns a bufio.Writer which writes to w. The buffer size is that of the pool.
+func (bufPool *BufioWriterPool) Get(w io.Writer) *bufio.Writer {
+	buf := bufPool.pool.Get().(*bufio.Writer)
+	buf.Reset(w)
+	return buf
+}
+
+// Put puts the bufio.Writer back into the pool.
+func (bufPool *BufioWriterPool) Put(b *bufio.Writer) {
+	b.Reset(nil)
+	bufPool.pool.Put(b)
+}
+
+// NewWriteCloserWrapper returns a wrapper which puts the bufio.Writer back
+// into the pool and closes the writer if it's an io.Writecloser.
+func (bufPool *BufioWriterPool) NewWriteCloserWrapper(buf *bufio.Writer, w io.Writer) io.WriteCloser {
+	return ioutils.NewWriteCloserWrapper(w, func() error {
+		buf.Flush()
+		if writeCloser, ok := w.(io.WriteCloser); ok {
+			writeCloser.Close()
+		}
+		bufPool.Put(buf)
+		return nil
+	})
+}