Add support for COPY from previous rootfs

Signed-off-by: Tonis Tiigi <tonistiigi@gmail.com>

builder/builder.go

@@ -146,6 +146,9 @@ type Backend interface {
 
 	// SquashImage squashes the fs layers from the provided image down to the specified `to` image
 	SquashImage(from string, to string) (string, error)
+
+	// MountImage returns mounted path with rootfs of an image.
+	MountImage(name string) (string, func() error, error)
 }
 
 // Image represents a Docker image used by the builder.

builder/dockerfile/builder.go

@@ -69,7 +69,8 @@ type Builder struct {
 	runConfig        *container.Config // runconfig for cmd, run, entrypoint etc.
 	flags            *BFlags
 	tmpContainers    map[string]struct{}
-	image            string // imageID
+	image            string         // imageID
+	imageContexts    *imageContexts // helper for storing contexts from builds
 	noBaseImage      bool
 	maintainer       string
 	cmdSet           bool
@@ -88,12 +89,13 @@ type Builder struct {
 
 // BuildManager implements builder.Backend and is shared across all Builder objects.
 type BuildManager struct {
-	backend builder.Backend
+	backend   builder.Backend
+	pathCache *pathCache // TODO: make this persistent
 }
 
 // NewBuildManager creates a BuildManager.
 func NewBuildManager(b builder.Backend) (bm *BuildManager) {
-	return &BuildManager{backend: b}
+	return &BuildManager{backend: b, pathCache: &pathCache{}}
 }
 
 // BuildFromContext builds a new image from a given context.
@@ -118,6 +120,7 @@ func (bm *BuildManager) BuildFromContext(ctx context.Context, src io.ReadCloser,
 	if err != nil {
 		return "", err
 	}
+	b.imageContexts.cache = bm.pathCache
 	return b.build(pg.StdoutFormatter, pg.StderrFormatter, pg.Output)
 }
 
@@ -147,6 +150,7 @@ func NewBuilder(clientCtx context.Context, config *types.ImageBuildOptions, back
 			LookingForDirectives: true,
 		},
 	}
+	b.imageContexts = &imageContexts{b: b}
 
 	parser.SetEscapeToken(parser.DefaultEscapeToken, &b.directive) // Assume the default token for escape
 
@@ -239,6 +243,8 @@ func (b *Builder) processLabels() error {
 // * Print a happy message and return the image ID.
 //
 func (b *Builder) build(stdout io.Writer, stderr io.Writer, out io.Writer) (string, error) {
+	defer b.imageContexts.unmount()
+
 	b.Stdout = stdout
 	b.Stderr = stderr
 	b.Output = out

builder/dockerfile/dispatchers.go

@@ -8,7 +8,6 @@ package dockerfile
 // package.
 
 import (
-	"errors"
 	"fmt"
 	"regexp"
 	"runtime"
@@ -25,6 +24,7 @@ import (
 	"github.com/docker/docker/builder"
 	"github.com/docker/docker/pkg/signal"
 	"github.com/docker/go-connections/nat"
+	"github.com/pkg/errors"
 )
 
 // ENV foo bar
@@ -169,7 +169,7 @@ func add(b *Builder, args []string, attributes map[string]bool, original string)
 		return err
 	}
 
-	return b.runContextCommand(args, true, true, "ADD")
+	return b.runContextCommand(args, true, true, "ADD", nil)
 }
 
 // COPY foo /path
@@ -181,11 +181,26 @@ func dispatchCopy(b *Builder, args []string, attributes map[string]bool, origina
 		return errAtLeastTwoArguments("COPY")
 	}
 
+	flFrom := b.flags.AddString("from", "")
+
 	if err := b.flags.Parse(); err != nil {
 		return err
 	}
 
-	return b.runContextCommand(args, false, false, "COPY")
+	var contextID *int
+	if flFrom.IsUsed() {
+		var err error
+		context, err := strconv.Atoi(flFrom.Value)
+		if err != nil {
+			return errors.Wrap(err, "from expects an integer value corresponding to the context number")
+		}
+		if err := b.imageContexts.validate(context); err != nil {
+			return err
+		}
+		contextID = &context
+	}
+
+	return b.runContextCommand(args, false, false, "COPY", contextID)
 }
 
 // FROM imagename
@@ -206,6 +221,7 @@ func from(b *Builder, args []string, attributes map[string]bool, original string
 	var image builder.Image
 
 	b.resetImageCache()
+	b.imageContexts.new()
 
 	// Windows cannot support a container with no base image.
 	if name == api.NoBaseImageSpecifier {
@@ -227,6 +243,7 @@ func from(b *Builder, args []string, attributes map[string]bool, original string
 				return err
 			}
 		}
+		b.imageContexts.update(image.ImageID())
 	}
 	b.from = image
 

builder/dockerfile/dispatchers_test.go

@@ -191,6 +191,7 @@ func TestLabel(t *testing.T) {
 
 func TestFrom(t *testing.T) {
 	b := &Builder{flags: &BFlags{}, runConfig: &container.Config{}, disableCommit: true}
+	b.imageContexts = &imageContexts{b: b}
 
 	err := from(b, []string{"scratch"}, nil, "")
 

builder/dockerfile/imagecontext.go

@@ -0,0 +1,121 @@
+package dockerfile
+
+import (
+	"sync"
+
+	"github.com/Sirupsen/logrus"
+	"github.com/docker/docker/builder"
+	"github.com/docker/docker/builder/remotecontext"
+	"github.com/pkg/errors"
+)
+
+// imageContexts is a helper for stacking up built image rootfs and reusing
+// them as contexts
+type imageContexts struct {
+	b     *Builder
+	list  []*imageMount
+	cache *pathCache
+}
+
+type imageMount struct {
+	id      string
+	ctx     builder.Context
+	release func() error
+}
+
+func (ic *imageContexts) new() {
+	ic.list = append(ic.list, &imageMount{})
+}
+
+func (ic *imageContexts) update(imageID string) {
+	ic.list[len(ic.list)-1].id = imageID
+}
+
+func (ic *imageContexts) validate(i int) error {
+	if i < 0 || i >= len(ic.list)-1 {
+		var extraMsg string
+		if i == len(ic.list)-1 {
+			extraMsg = " refers current build block"
+		}
+		return errors.Errorf("invalid from flag value %d%s", i, extraMsg)
+	}
+	return nil
+}
+
+func (ic *imageContexts) context(i int) (builder.Context, error) {
+	if err := ic.validate(i); err != nil {
+		return nil, err
+	}
+	im := ic.list[i]
+	if im.ctx == nil {
+		if im.id == "" {
+			return nil, errors.Errorf("could not copy from empty context")
+		}
+		p, release, err := ic.b.docker.MountImage(im.id)
+		if err != nil {
+			return nil, errors.Wrapf(err, "failed to mount %s", im.id)
+		}
+		ctx, err := remotecontext.NewLazyContext(p)
+		if err != nil {
+			return nil, errors.Wrapf(err, "failed to create lazycontext for %s", p)
+		}
+		logrus.Debugf("mounted image: %s %s", im.id, p)
+		im.release = release
+		im.ctx = ctx
+	}
+	return im.ctx, nil
+}
+
+func (ic *imageContexts) unmount() (retErr error) {
+	for _, im := range ic.list {
+		if im.release != nil {
+			if err := im.release(); err != nil {
+				logrus.Error(errors.Wrapf(err, "failed to unmount previous build image"))
+				retErr = err
+			}
+		}
+	}
+	return
+}
+
+func (ic *imageContexts) getCache(i int, path string) (interface{}, bool) {
+	if ic.cache != nil {
+		im := ic.list[i]
+		if im.id == "" {
+			return nil, false
+		}
+		return ic.cache.get(im.id + path)
+	}
+	return nil, false
+}
+
+func (ic *imageContexts) setCache(i int, path string, v interface{}) {
+	if ic.cache != nil {
+		ic.cache.set(ic.list[i].id+path, v)
+	}
+}
+
+type pathCache struct {
+	mu    sync.Mutex
+	items map[string]interface{}
+}
+
+func (c *pathCache) set(k string, v interface{}) {
+	c.mu.Lock()
+	if c.items == nil {
+		c.items = make(map[string]interface{})
+	}
+	c.items[k] = v
+	c.mu.Unlock()
+}
+
+func (c *pathCache) get(k string) (interface{}, bool) {
+	c.mu.Lock()
+	if c.items == nil {
+		c.mu.Unlock()
+		return nil, false
+	}
+	v, ok := c.items[k]
+	c.mu.Unlock()
+	return v, ok
+}

builder/dockerfile/internals.go

@@ -6,7 +6,6 @@ package dockerfile
 import (
 	"crypto/sha256"
 	"encoding/hex"
-	"errors"
 	"fmt"
 	"io"
 	"io/ioutil"
@@ -14,6 +13,8 @@ import (
 	"net/url"
 	"os"
 	"path/filepath"
+	"regexp"
+	"runtime"
 	"sort"
 	"strings"
 	"time"
@@ -36,6 +37,7 @@ import (
 	"github.com/docker/docker/pkg/tarsum"
 	"github.com/docker/docker/pkg/urlutil"
 	"github.com/docker/docker/runconfig/opts"
+	"github.com/pkg/errors"
 )
 
 func (b *Builder) commit(id string, autoCmd strslice.StrSlice, comment string) error {
@@ -83,6 +85,7 @@ func (b *Builder) commit(id string, autoCmd strslice.StrSlice, comment string) e
 	}
 
 	b.image = imageID
+	b.imageContexts.update(imageID)
 	return nil
 }
 
@@ -91,11 +94,7 @@ type copyInfo struct {
 	decompress bool
 }
 
-func (b *Builder) runContextCommand(args []string, allowRemote bool, allowLocalDecompression bool, cmdName string) error {
-	if b.context == nil {
-		return fmt.Errorf("No context given. Impossible to use %s", cmdName)
-	}
-
+func (b *Builder) runContextCommand(args []string, allowRemote bool, allowLocalDecompression bool, cmdName string, contextID *int) error {
 	if len(args) < 2 {
 		return fmt.Errorf("Invalid %s format - at least two arguments required", cmdName)
 	}
@@ -129,7 +128,7 @@ func (b *Builder) runContextCommand(args []string, allowRemote bool, allowLocalD
 			continue
 		}
 		// not a URL
-		subInfos, err := b.calcCopyInfo(cmdName, orig, allowLocalDecompression, true)
+		subInfos, err := b.calcCopyInfo(cmdName, orig, allowLocalDecompression, true, contextID)
 		if err != nil {
 			return err
 		}
@@ -299,20 +298,41 @@ func (b *Builder) download(srcURL string) (fi builder.FileInfo, err error) {
 	return &builder.HashedFileInfo{FileInfo: builder.PathFileInfo{FileInfo: tmpFileSt, FilePath: tmpFileName}, FileHash: hash}, nil
 }
 
-func (b *Builder) calcCopyInfo(cmdName, origPath string, allowLocalDecompression, allowWildcards bool) ([]copyInfo, error) {
+func (b *Builder) calcCopyInfo(cmdName, origPath string, allowLocalDecompression, allowWildcards bool, contextID *int) ([]copyInfo, error) {
 
 	// Work in daemon-specific OS filepath semantics
 	origPath = filepath.FromSlash(origPath)
 
+	// validate windows paths from other images
+	if contextID != nil && runtime.GOOS == "windows" {
+		forbid := regexp.MustCompile("(?i)^" + string(os.PathSeparator) + "?(windows(" + string(os.PathSeparator) + ".+)?)?$")
+		if p := filepath.Clean(origPath); p == "." || forbid.MatchString(p) {
+			return nil, errors.Errorf("copy from %s is not allowed on windows", origPath)
+		}
+	}
+
 	if origPath != "" && origPath[0] == os.PathSeparator && len(origPath) > 1 {
 		origPath = origPath[1:]
 	}
 	origPath = strings.TrimPrefix(origPath, "."+string(os.PathSeparator))
 
+	context := b.context
+	var err error
+	if contextID != nil {
+		context, err = b.imageContexts.context(*contextID)
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	if context == nil {
+		return nil, errors.Errorf("No context given. Impossible to use %s", cmdName)
+	}
+
 	// Deal with wildcards
 	if allowWildcards && containsWildcards(origPath) {
 		var copyInfos []copyInfo
-		if err := b.context.Walk("", func(path string, info builder.FileInfo, err error) error {
+		if err := context.Walk("", func(path string, info builder.FileInfo, err error) error {
 			if err != nil {
 				return err
 			}
@@ -326,7 +346,7 @@ func (b *Builder) calcCopyInfo(cmdName, origPath string, allowLocalDecompression
 
 			// Note we set allowWildcards to false in case the name has
 			// a * in it
-			subInfos, err := b.calcCopyInfo(cmdName, path, allowLocalDecompression, false)
+			subInfos, err := b.calcCopyInfo(cmdName, path, allowLocalDecompression, false, contextID)
 			if err != nil {
 				return err
 			}
@@ -339,8 +359,7 @@ func (b *Builder) calcCopyInfo(cmdName, origPath string, allowLocalDecompression
 	}
 
 	// Must be a dir or a file
-
-	statPath, fi, err := b.context.Stat(origPath)
+	statPath, fi, err := context.Stat(origPath)
 	if err != nil {
 		return nil, err
 	}
@@ -351,6 +370,13 @@ func (b *Builder) calcCopyInfo(cmdName, origPath string, allowLocalDecompression
 	if !handleHash {
 		return copyInfos, nil
 	}
+	if contextID != nil {
+		// fast-cache based on imageID
+		if h, ok := b.imageContexts.getCache(*contextID, origPath); ok {
+			hfi.SetHash(h.(string))
+			return copyInfos, nil
+		}
+	}
 
 	// Deal with the single file case
 	if !fi.IsDir() {
@@ -359,7 +385,7 @@ func (b *Builder) calcCopyInfo(cmdName, origPath string, allowLocalDecompression
 	}
 	// Must be a dir
 	var subfiles []string
-	err = b.context.Walk(statPath, func(path string, info builder.FileInfo, err error) error {
+	err = context.Walk(statPath, func(path string, info builder.FileInfo, err error) error {
 		if err != nil {
 			return err
 		}
@@ -375,6 +401,9 @@ func (b *Builder) calcCopyInfo(cmdName, origPath string, allowLocalDecompression
 	hasher := sha256.New()
 	hasher.Write([]byte(strings.Join(subfiles, ",")))
 	hfi.SetHash("dir:" + hex.EncodeToString(hasher.Sum(nil)))
+	if contextID != nil {
+		b.imageContexts.setCache(*contextID, origPath, hfi.Hash())
+	}
 
 	return copyInfos, nil
 }
@@ -468,6 +497,7 @@ func (b *Builder) probeCache() (bool, error) {
 	fmt.Fprint(b.Stdout, " ---> Using cache\n")
 	logrus.Debugf("[BUILDER] Use cached version: %s", b.runConfig.Cmd)
 	b.image = string(cache)
+	b.imageContexts.update(b.image)
 
 	return true, nil
 }

builder/remotecontext/filehash.go

@@ -0,0 +1,34 @@
+package remotecontext
+
+import (
+	"archive/tar"
+	"crypto/sha256"
+	"hash"
+	"os"
+
+	"github.com/docker/docker/pkg/archive"
+	"github.com/docker/docker/pkg/tarsum"
+)
+
+// NewFileHash returns new hash that is used for the builder cache keys
+func NewFileHash(path, name string, fi os.FileInfo) (hash.Hash, error) {
+	hdr, err := archive.FileInfoHeader(path, name, fi)
+	if err != nil {
+		return nil, err
+	}
+	tsh := &tarsumHash{hdr: hdr, Hash: sha256.New()}
+	tsh.Reset() // initialize header
+	return tsh, nil
+}
+
+type tarsumHash struct {
+	hash.Hash
+	hdr *tar.Header
+}
+
+// Reset resets the Hash to its initial state.
+func (tsh *tarsumHash) Reset() {
+	// comply with hash.Hash and reset to the state hash had before any writes
+	tsh.Hash.Reset()
+	tarsum.WriteV1Header(tsh.hdr, tsh.Hash)
+}

builder/remotecontext/lazycontext.go

@@ -0,0 +1,149 @@
+package remotecontext
+
+import (
+	"encoding/hex"
+	"io"
+	"os"
+	"path/filepath"
+
+	"github.com/docker/docker/builder"
+	"github.com/docker/docker/pkg/pools"
+	"github.com/docker/docker/pkg/symlink"
+	"github.com/pkg/errors"
+)
+
+// NewLazyContext creates a new LazyContext. LazyContext defines a hashed build
+// context based on a root directory. Individual files are hashed first time
+// they are asked. It is not safe to call methods of LazyContext concurrently.
+func NewLazyContext(root string) (builder.Context, error) {
+	return &lazyContext{
+		root: root,
+		sums: make(map[string]string),
+	}, nil
+}
+
+type lazyContext struct {
+	root string
+	sums map[string]string
+}
+
+func (c *lazyContext) Close() error {
+	return nil
+}
+
+func (c *lazyContext) Open(path string) (io.ReadCloser, error) {
+	cleanPath, fullPath, err := c.normalize(path)
+	if err != nil {
+		return nil, err
+	}
+
+	r, err := os.Open(fullPath)
+	if err != nil {
+		return nil, errors.WithStack(convertPathError(err, cleanPath))
+	}
+	return r, nil
+}
+
+func (c *lazyContext) Stat(path string) (string, builder.FileInfo, error) {
+	// TODO: although stat returns builder.FileInfo it builder.Context actually requires Hashed
+	cleanPath, fullPath, err := c.normalize(path)
+	if err != nil {
+		return "", nil, err
+	}
+
+	st, err := os.Lstat(fullPath)
+	if err != nil {
+		return "", nil, errors.WithStack(convertPathError(err, cleanPath))
+	}
+
+	relPath, err := filepath.Rel(c.root, fullPath)
+	if err != nil {
+		return "", nil, errors.WithStack(convertPathError(err, cleanPath))
+	}
+
+	sum, ok := c.sums[relPath]
+	if !ok {
+		sum, err = c.prepareHash(relPath, st)
+		if err != nil {
+			return "", nil, err
+		}
+	}
+
+	fi := &builder.HashedFileInfo{
+		builder.PathFileInfo{st, fullPath, filepath.Base(cleanPath)},
+		sum,
+	}
+	return relPath, fi, nil
+}
+
+func (c *lazyContext) Walk(root string, walkFn builder.WalkFunc) error {
+	_, fullPath, err := c.normalize(root)
+	if err != nil {
+		return err
+	}
+	return filepath.Walk(fullPath, func(fullPath string, fi os.FileInfo, err error) error {
+		relPath, err := filepath.Rel(c.root, fullPath)
+		if err != nil {
+			return errors.WithStack(err)
+		}
+		if relPath == "." {
+			return nil
+		}
+
+		sum, ok := c.sums[relPath]
+		if !ok {
+			sum, err = c.prepareHash(relPath, fi)
+			if err != nil {
+				return err
+			}
+		}
+
+		hfi := &builder.HashedFileInfo{
+			builder.PathFileInfo{FileInfo: fi, FilePath: fullPath},
+			sum,
+		}
+		if err := walkFn(relPath, hfi, nil); err != nil {
+			return err
+		}
+		return nil
+	})
+}
+
+func (c *lazyContext) prepareHash(relPath string, fi os.FileInfo) (string, error) {
+	p := filepath.Join(c.root, relPath)
+	h, err := NewFileHash(p, relPath, fi)
+	if err != nil {
+		return "", errors.Wrapf(err, "failed to create hash for %s", relPath)
+	}
+	if fi.Mode().IsRegular() && fi.Size() > 0 {
+		f, err := os.Open(p)
+		if err != nil {
+			return "", errors.Wrapf(err, "failed to open %s", relPath)
+		}
+		defer f.Close()
+		if _, err := pools.Copy(h, f); err != nil {
+			return "", errors.Wrapf(err, "failed to copy file data for %s", relPath)
+		}
+	}
+	sum := hex.EncodeToString(h.Sum(nil))
+	c.sums[relPath] = sum
+	return sum, nil
+}
+
+func (c *lazyContext) normalize(path string) (cleanPath, fullPath string, err error) {
+	// todo: combine these helpers with tarsum after they are moved to same package
+	cleanPath = filepath.Clean(string(os.PathSeparator) + path)[1:]
+	fullPath, err = symlink.FollowSymlinkInScope(filepath.Join(c.root, path), c.root)
+	if err != nil {
+		return "", "", errors.Wrapf(err, "forbidden path outside the build context: %s (%s)", path, fullPath)
+	}
+	return
+}
+
+func convertPathError(err error, cleanpath string) error {
+	if err, ok := err.(*os.PathError); ok {
+		err.Path = cleanpath
+		return err
+	}
+	return err
+}

daemon/archive.go

@@ -1,7 +1,6 @@
 package daemon
 
 import (
-	"errors"
 	"io"
 	"os"
 	"path/filepath"
@@ -10,11 +9,14 @@ import (
 	"github.com/docker/docker/api/types"
 	"github.com/docker/docker/builder"
 	"github.com/docker/docker/container"
+	"github.com/docker/docker/layer"
 	"github.com/docker/docker/pkg/archive"
 	"github.com/docker/docker/pkg/chrootarchive"
 	"github.com/docker/docker/pkg/idtools"
 	"github.com/docker/docker/pkg/ioutils"
+	"github.com/docker/docker/pkg/stringid"
 	"github.com/docker/docker/pkg/system"
+	"github.com/pkg/errors"
 )
 
 // ErrExtractPointNotDirectory is used to convey that the operation to extract
@@ -454,3 +456,34 @@ func (daemon *Daemon) CopyOnBuild(cID string, destPath string, src builder.FileI
 
 	return fixPermissions(srcPath, destPath, rootUID, rootGID, destExists)
 }
+
+// MountImage returns mounted path with rootfs of an image.
+func (daemon *Daemon) MountImage(name string) (string, func() error, error) {
+	img, err := daemon.GetImage(name)
+	if err != nil {
+		return "", nil, errors.Wrapf(err, "no such image: %s", name)
+	}
+
+	mountID := stringid.GenerateRandomID()
+	rwLayer, err := daemon.layerStore.CreateRWLayer(mountID, img.RootFS.ChainID(), nil)
+	if err != nil {
+		return "", nil, errors.Wrap(err, "failed to create rwlayer")
+	}
+
+	mountPath, err := rwLayer.Mount("")
+	if err != nil {
+		metadata, releaseErr := daemon.layerStore.ReleaseRWLayer(rwLayer)
+		if releaseErr != nil {
+			err = errors.Wrapf(err, "failed to release rwlayer: %s", releaseErr.Error())
+		}
+		layer.LogReleaseMetadata(metadata)
+		return "", nil, errors.Wrap(err, "failed to mount rwlayer")
+	}
+
+	return mountPath, func() error {
+		rwLayer.Unmount()
+		metadata, err := daemon.layerStore.ReleaseRWLayer(rwLayer)
+		layer.LogReleaseMetadata(metadata)
+		return err
+	}, nil
+}

integration-cli/docker_cli_build_test.go

@@ -5755,6 +5755,138 @@ func (s *DockerSuite) TestBuildContChar(c *check.C) {
 	c.Assert(result.Combined(), checker.Contains, "Step 2/2 : RUN echo hi \\\\\n")
 }
 
+func (s *DockerSuite) TestBuildCopyFromPreviousRootFS(c *check.C) {
+	dockerfile := `
+		FROM busybox
+		COPY foo bar
+		FROM busybox
+    %s
+    COPY baz baz
+    RUN echo mno > baz/cc
+		FROM busybox
+    COPY bar /
+    COPY --from=1 baz sub/
+    COPY --from=0 bar baz
+    COPY --from=0 bar bay`
+	ctx := fakeContext(c, fmt.Sprintf(dockerfile, ""), map[string]string{
+		"Dockerfile": dockerfile,
+		"foo":        "abc",
+		"bar":        "def",
+		"baz/aa":     "ghi",
+		"baz/bb":     "jkl",
+	})
+	defer ctx.Close()
+
+	result := buildImage("build1", withExternalBuildContext(ctx))
+	result.Assert(c, icmd.Success)
+
+	out, _ := dockerCmd(c, "run", "build1", "cat", "bar")
+	c.Assert(strings.TrimSpace(out), check.Equals, "def")
+	out, _ = dockerCmd(c, "run", "build1", "cat", "sub/aa")
+	c.Assert(strings.TrimSpace(out), check.Equals, "ghi")
+	out, _ = dockerCmd(c, "run", "build1", "cat", "sub/cc")
+	c.Assert(strings.TrimSpace(out), check.Equals, "mno")
+	out, _ = dockerCmd(c, "run", "build1", "cat", "baz")
+	c.Assert(strings.TrimSpace(out), check.Equals, "abc")
+	out, _ = dockerCmd(c, "run", "build1", "cat", "bay")
+	c.Assert(strings.TrimSpace(out), check.Equals, "abc")
+
+	result = buildImage("build2", withExternalBuildContext(ctx))
+	result.Assert(c, icmd.Success)
+
+	// all commands should be cached
+	c.Assert(strings.Count(result.Combined(), "Using cache"), checker.Equals, 7)
+
+	err := ioutil.WriteFile(filepath.Join(ctx.Dir, "Dockerfile"), []byte(fmt.Sprintf(dockerfile, "COPY baz/aa foo")), 0644)
+	c.Assert(err, checker.IsNil)
+
+	// changing file in parent block should not affect last block
+	result = buildImage("build3", withExternalBuildContext(ctx))
+	result.Assert(c, icmd.Success)
+
+	c.Assert(strings.Count(result.Combined(), "Using cache"), checker.Equals, 5)
+
+	c.Assert(getIDByName(c, "build1"), checker.Equals, getIDByName(c, "build2"))
+
+	err = ioutil.WriteFile(filepath.Join(ctx.Dir, "foo"), []byte("pqr"), 0644)
+	c.Assert(err, checker.IsNil)
+
+	// changing file in parent block should affect both first and last block
+	result = buildImage("build4", withExternalBuildContext(ctx))
+	result.Assert(c, icmd.Success)
+	c.Assert(strings.Count(result.Combined(), "Using cache"), checker.Equals, 5)
+
+	out, _ = dockerCmd(c, "run", "build4", "cat", "bay")
+	c.Assert(strings.TrimSpace(out), check.Equals, "pqr")
+	out, _ = dockerCmd(c, "run", "build4", "cat", "baz")
+	c.Assert(strings.TrimSpace(out), check.Equals, "pqr")
+}
+
+func (s *DockerSuite) TestBuildCopyFromPreviousRootFSErrors(c *check.C) {
+	dockerfile := `
+		FROM busybox
+		COPY --from=foo foo bar`
+
+	ctx := fakeContext(c, dockerfile, map[string]string{
+		"Dockerfile": dockerfile,
+		"foo":        "abc",
+	})
+	defer ctx.Close()
+
+	buildImage("build1", withExternalBuildContext(ctx)).Assert(c, icmd.Expected{
+		ExitCode: 1,
+		Err:      "from expects an integer value corresponding to the context number",
+	})
+
+	dockerfile = `
+		FROM busybox
+		COPY --from=0 foo bar`
+
+	ctx = fakeContext(c, dockerfile, map[string]string{
+		"Dockerfile": dockerfile,
+		"foo":        "abc",
+	})
+	defer ctx.Close()
+
+	buildImage("build1", withExternalBuildContext(ctx)).Assert(c, icmd.Expected{
+		ExitCode: 1,
+		Err:      "invalid from flag value 0 refers current build block",
+	})
+}
+
+func (s *DockerSuite) TestBuildCopyFromPreviousFrom(c *check.C) {
+	dockerfile := `
+		FROM busybox
+		COPY foo bar`
+	ctx := fakeContext(c, dockerfile, map[string]string{
+		"Dockerfile": dockerfile,
+		"foo":        "abc",
+	})
+	defer ctx.Close()
+
+	result := buildImage("build1", withExternalBuildContext(ctx))
+	result.Assert(c, icmd.Success)
+
+	dockerfile = `
+		FROM build1:latest
+    FROM busybox
+		COPY --from=0 bar /
+		COPY foo /`
+	ctx = fakeContext(c, dockerfile, map[string]string{
+		"Dockerfile": dockerfile,
+		"foo":        "def",
+	})
+	defer ctx.Close()
+
+	result = buildImage("build2", withExternalBuildContext(ctx))
+	result.Assert(c, icmd.Success)
+
+	out, _ := dockerCmd(c, "run", "build2", "cat", "bar")
+	c.Assert(strings.TrimSpace(out), check.Equals, "abc")
+	out, _ = dockerCmd(c, "run", "build2", "cat", "foo")
+	c.Assert(strings.TrimSpace(out), check.Equals, "def")
+}
+
 // TestBuildOpaqueDirectory tests that a build succeeds which
 // creates opaque directories.
 // See https://github.com/docker/docker/issues/25244

pkg/archive/archive.go

@@ -240,6 +240,38 @@ func (compression *Compression) Extension() string {
 	return ""
 }
 
+// FileInfoHeader creates a populated Header from fi.
+// Compared to archive pkg this function fills in more information.
+func FileInfoHeader(path, name string, fi os.FileInfo) (*tar.Header, error) {
+	var link string
+	if fi.Mode()&os.ModeSymlink != 0 {
+		var err error
+		link, err = os.Readlink(path)
+		if err != nil {
+			return nil, err
+		}
+	}
+	hdr, err := tar.FileInfoHeader(fi, link)
+	if err != nil {
+		return nil, err
+	}
+	hdr.Mode = int64(chmodTarEntry(os.FileMode(hdr.Mode)))
+	name, err = canonicalTarName(name, fi.IsDir())
+	if err != nil {
+		return nil, fmt.Errorf("tar: cannot canonicalize path: %v", err)
+	}
+	hdr.Name = name
+	if err := setHeaderForSpecialDevice(hdr, name, fi.Sys()); err != nil {
+		return nil, err
+	}
+	capability, _ := system.Lgetxattr(path, "security.capability")
+	if capability != nil {
+		hdr.Xattrs = make(map[string]string)
+		hdr.Xattrs["security.capability"] = string(capability)
+	}
+	return hdr, nil
+}
+
 type tarWhiteoutConverter interface {
 	ConvertWrite(*tar.Header, string, os.FileInfo) (*tar.Header, error)
 	ConvertRead(*tar.Header, string) (bool, error)
@@ -283,26 +315,7 @@ func (ta *tarAppender) addTarFile(path, name string) error {
 		return err
 	}
 
-	link := ""
-	if fi.Mode()&os.ModeSymlink != 0 {
-		if link, err = os.Readlink(path); err != nil {
-			return err
-		}
-	}
-
-	hdr, err := tar.FileInfoHeader(fi, link)
-	if err != nil {
-		return err
-	}
-	hdr.Mode = int64(chmodTarEntry(os.FileMode(hdr.Mode)))
-
-	name, err = canonicalTarName(name, fi.IsDir())
-	if err != nil {
-		return fmt.Errorf("tar: cannot canonicalize path: %v", err)
-	}
-	hdr.Name = name
-
-	inode, err := setHeaderForSpecialDevice(hdr, ta, name, fi.Sys())
+	hdr, err := FileInfoHeader(path, name, fi)
 	if err != nil {
 		return err
 	}
@@ -310,6 +323,10 @@ func (ta *tarAppender) addTarFile(path, name string) error {
 	// if it's not a directory and has more than 1 link,
 	// it's hard linked, so set the type flag accordingly
 	if !fi.IsDir() && hasHardlinks(fi) {
+		inode, err := getInodeFromStat(fi.Sys())
+		if err != nil {
+			return err
+		}
 		// a link should have a name that it links too
 		// and that linked name should be first in the tar archive
 		if oldpath, ok := ta.SeenFiles[inode]; ok {
@@ -321,12 +338,6 @@ func (ta *tarAppender) addTarFile(path, name string) error {
 		}
 	}
 
-	capability, _ := system.Lgetxattr(path, "security.capability")
-	if capability != nil {
-		hdr.Xattrs = make(map[string]string)
-		hdr.Xattrs["security.capability"] = string(capability)
-	}
-
 	//handle re-mapping container ID mappings back to host ID mappings before
 	//writing tar headers/files. We skip whiteout files because they were written
 	//by the kernel and already have proper ownership relative to the host

pkg/archive/archive_unix.go

@@ -41,7 +41,7 @@ func chmodTarEntry(perm os.FileMode) os.FileMode {
 	return perm // noop for unix as golang APIs provide perm bits correctly
 }
 
-func setHeaderForSpecialDevice(hdr *tar.Header, ta *tarAppender, name string, stat interface{}) (inode uint64, err error) {
+func setHeaderForSpecialDevice(hdr *tar.Header, name string, stat interface{}) (err error) {
 	s, ok := stat.(*syscall.Stat_t)
 
 	if !ok {
@@ -49,8 +49,6 @@ func setHeaderForSpecialDevice(hdr *tar.Header, ta *tarAppender, name string, st
 		return
 	}
 
-	inode = uint64(s.Ino)
-
 	// Currently go does not fill in the major/minors
 	if s.Mode&syscall.S_IFBLK != 0 ||
 		s.Mode&syscall.S_IFCHR != 0 {
@@ -61,6 +59,19 @@ func setHeaderForSpecialDevice(hdr *tar.Header, ta *tarAppender, name string, st
 	return
 }
 
+func getInodeFromStat(stat interface{}) (inode uint64, err error) {
+	s, ok := stat.(*syscall.Stat_t)
+
+	if !ok {
+		err = errors.New("cannot convert stat value to syscall.Stat_t")
+		return
+	}
+
+	inode = uint64(s.Ino)
+
+	return
+}
+
 func getFileUIDGID(stat interface{}) (int, int, error) {
 	s, ok := stat.(*syscall.Stat_t)
 

pkg/archive/archive_windows.go

@@ -49,8 +49,13 @@ func chmodTarEntry(perm os.FileMode) os.FileMode {
 	return perm
 }
 
-func setHeaderForSpecialDevice(hdr *tar.Header, ta *tarAppender, name string, stat interface{}) (inode uint64, err error) {
-	// do nothing. no notion of Rdev, Inode, Nlink in stat on Windows
+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
 }
 

pkg/tarsum/versioning.go

@@ -3,6 +3,7 @@ package tarsum
 import (
 	"archive/tar"
 	"errors"
+	"io"
 	"sort"
 	"strconv"
 	"strings"
@@ -21,6 +22,13 @@ const (
 	VersionDev
 )
 
+// WriteV1Header writes a tar header to a writer in V1 tarsum format.
+func WriteV1Header(h *tar.Header, w io.Writer) {
+	for _, elem := range v1TarHeaderSelect(h) {
+		w.Write([]byte(elem[0] + elem[1]))
+	}
+}
+
 // VersionLabelForChecksum returns the label for the given tarsum
 // checksum, i.e., everything before the first `+` character in
 // the string or an empty string if no label separator is found.