Cherry-pick moby/moby #39292 for CVE-2018-15664 fixes

Signed-off-by: Matthew Heon <matthew.heon@pm.me>

pkg/chrootarchive/archive.go

@@ -1,7 +1,7 @@
 package chrootarchive
 
 import (
-	"archive/tar"
+	stdtar "archive/tar"
 	"fmt"
 	"io"
 	"io/ioutil"
@@ -34,18 +34,34 @@ func NewArchiverWithChown(tarIDMappings *idtools.IDMappings, chownOpts *idtools.
 // The archive may be compressed with one of the following algorithms:
 //  identity (uncompressed), gzip, bzip2, xz.
 func Untar(tarArchive io.Reader, dest string, options *archive.TarOptions) error {
-	return untarHandler(tarArchive, dest, options, true)
+	return untarHandler(tarArchive, dest, options, true, dest)
+}
+
+// UntarWithRoot is the same as `Untar`, but allows you to pass in a root directory
+// The root directory is the directory that will be chrooted to.
+// `dest` must be a path within `root`, if it is not an error will be returned.
+//
+// `root` should set to a directory which is not controlled by any potentially
+// malicious process.
+//
+// This should be used to prevent a potential attacker from manipulating `dest`
+// such that it would provide access to files outside of `dest` through things
+// like symlinks. Normally `ResolveSymlinksInScope` would handle this, however
+// sanitizing symlinks in this manner is inherrently racey:
+// ref: CVE-2018-15664
+func UntarWithRoot(tarArchive io.Reader, dest string, options *archive.TarOptions, root string) error {
+	return untarHandler(tarArchive, dest, options, true, root)
 }
 
 // UntarUncompressed reads a stream of bytes from `archive`, parses it as a tar archive,
 // and unpacks it into the directory at `dest`.
 // The archive must be an uncompressed stream.
 func UntarUncompressed(tarArchive io.Reader, dest string, options *archive.TarOptions) error {
-	return untarHandler(tarArchive, dest, options, false)
+	return untarHandler(tarArchive, dest, options, false, dest)
 }
 
 // Handler for teasing out the automatic decompression
-func untarHandler(tarArchive io.Reader, dest string, options *archive.TarOptions, decompress bool) error {
+func untarHandler(tarArchive io.Reader, dest string, options *archive.TarOptions, decompress bool, root string) error {
 	if tarArchive == nil {
 		return fmt.Errorf("Empty archive")
 	}
@@ -77,7 +93,15 @@ func untarHandler(tarArchive io.Reader, dest string, options *archive.TarOptions
 		r = decompressedArchive
 	}
 
-	return invokeUnpack(r, dest, options)
+	return invokeUnpack(r, dest, options, root)
+}
+
+// Tar tars the requested path while chrooted to the specified root.
+func Tar(srcPath string, options *archive.TarOptions, root string) (io.ReadCloser, error) {
+	if options == nil {
+		options = &archive.TarOptions{}
+	}
+	return invokePack(srcPath, options, root)
 }
 
 // CopyFileWithTarAndChown returns a function which copies a single file from outside
@@ -99,7 +123,7 @@ func CopyFileWithTarAndChown(chownOpts *idtools.IDPair, hasher io.Writer, uidmap
 			var hashWorker sync.WaitGroup
 			hashWorker.Add(1)
 			go func() {
-				t := tar.NewReader(contentReader)
+				t := stdtar.NewReader(contentReader)
 				_, err := t.Next()
 				if err != nil {
 					hashError = err

pkg/chrootarchive/archive_unix.go

@@ -10,10 +10,13 @@ import (
 	"io"
 	"io/ioutil"
 	"os"
+	"path/filepath"
 	"runtime"
+	"strings"
 
 	"github.com/containers/storage/pkg/archive"
 	"github.com/containers/storage/pkg/reexec"
+	"github.com/pkg/errors"
 )
 
 // untar is the entry-point for storage-untar on re-exec. This is not used on
@@ -23,18 +26,28 @@ func untar() {
 	runtime.LockOSThread()
 	flag.Parse()
 
-	var options *archive.TarOptions
+	var options archive.TarOptions
 
 	//read the options from the pipe "ExtraFiles"
 	if err := json.NewDecoder(os.NewFile(3, "options")).Decode(&options); err != nil {
 		fatal(err)
 	}
 
-	if err := chroot(flag.Arg(0)); err != nil {
+	dst := flag.Arg(0)
+	var root string
+	if len(flag.Args()) > 1 {
+		root = flag.Arg(1)
+	}
+
+	if root == "" {
+		root = dst
+	}
+
+	if err := chroot(root); err != nil {
 		fatal(err)
 	}
 
-	if err := archive.Unpack(os.Stdin, "/", options); err != nil {
+	if err := archive.Unpack(os.Stdin, dst, &options); err != nil {
 		fatal(err)
 	}
 	// fully consume stdin in case it is zero padded
@@ -45,7 +58,10 @@ func untar() {
 	os.Exit(0)
 }
 
-func invokeUnpack(decompressedArchive io.Reader, dest string, options *archive.TarOptions) error {
+func invokeUnpack(decompressedArchive io.Reader, dest string, options *archive.TarOptions, root string) error {
+	if root == "" {
+		return errors.New("must specify a root to chroot to")
+	}
 
 	// We can't pass a potentially large exclude list directly via cmd line
 	// because we easily overrun the kernel's max argument/environment size
@@ -57,7 +73,21 @@ func invokeUnpack(decompressedArchive io.Reader, dest string, options *archive.T
 		return fmt.Errorf("Untar pipe failure: %v", err)
 	}
 
-	cmd := reexec.Command("storage-untar", dest)
+	if root != "" {
+		relDest, err := filepath.Rel(root, dest)
+		if err != nil {
+			return err
+		}
+		if relDest == "." {
+			relDest = "/"
+		}
+		if relDest[0] != '/' {
+			relDest = "/" + relDest
+		}
+		dest = relDest
+	}
+
+	cmd := reexec.Command("storage-untar", dest, root)
 	cmd.Stdin = decompressedArchive
 
 	cmd.ExtraFiles = append(cmd.ExtraFiles, r)
@@ -68,6 +98,7 @@ func invokeUnpack(decompressedArchive io.Reader, dest string, options *archive.T
 	if err := cmd.Start(); err != nil {
 		return fmt.Errorf("Untar error on re-exec cmd: %v", err)
 	}
+
 	//write the options to the pipe for the untar exec to read
 	if err := json.NewEncoder(w).Encode(options); err != nil {
 		return fmt.Errorf("Untar json encode to pipe failed: %v", err)
@@ -84,3 +115,92 @@ func invokeUnpack(decompressedArchive io.Reader, dest string, options *archive.T
 	}
 	return nil
 }
+
+func tar() {
+	runtime.LockOSThread()
+	flag.Parse()
+
+	src := flag.Arg(0)
+	var root string
+	if len(flag.Args()) > 1 {
+		root = flag.Arg(1)
+	}
+
+	if root == "" {
+		root = src
+	}
+
+	if err := realChroot(root); err != nil {
+		fatal(err)
+	}
+
+	var options archive.TarOptions
+	if err := json.NewDecoder(os.Stdin).Decode(&options); err != nil {
+		fatal(err)
+	}
+
+	rdr, err := archive.TarWithOptions(src, &options)
+	if err != nil {
+		fatal(err)
+	}
+	defer rdr.Close()
+
+	if _, err := io.Copy(os.Stdout, rdr); err != nil {
+		fatal(err)
+	}
+
+	os.Exit(0)
+}
+
+func invokePack(srcPath string, options *archive.TarOptions, root string) (io.ReadCloser, error) {
+	if root == "" {
+		return nil, errors.New("root path must not be empty")
+	}
+
+	relSrc, err := filepath.Rel(root, srcPath)
+	if err != nil {
+		return nil, err
+	}
+	if relSrc == "." {
+		relSrc = "/"
+	}
+	if relSrc[0] != '/' {
+		relSrc = "/" + relSrc
+	}
+
+	// make sure we didn't trim a trailing slash with the call to `Rel`
+	if strings.HasSuffix(srcPath, "/") && !strings.HasSuffix(relSrc, "/") {
+		relSrc += "/"
+	}
+
+	cmd := reexec.Command("storage-tar", relSrc, root)
+
+	errBuff := bytes.NewBuffer(nil)
+	cmd.Stderr = errBuff
+
+	tarR, tarW := io.Pipe()
+	cmd.Stdout = tarW
+
+	stdin, err := cmd.StdinPipe()
+	if err != nil {
+		return nil, errors.Wrap(err, "error getting options pipe for tar process")
+	}
+
+	if err := cmd.Start(); err != nil {
+		return nil, errors.Wrap(err, "tar error on re-exec cmd")
+	}
+
+	go func() {
+		err := cmd.Wait()
+		err = errors.Wrapf(err, "error processing tar file: %s", errBuff)
+		tarW.CloseWithError(err)
+	}()
+
+	if err := json.NewEncoder(stdin).Encode(options); err != nil {
+		stdin.Close()
+		return nil, errors.Wrap(err, "tar json encode to pipe failed")
+	}
+	stdin.Close()
+
+	return tarR, nil
+}

pkg/chrootarchive/archive_unix_test.go

@@ -0,0 +1,171 @@
+// +build !windows
+
+package chrootarchive
+
+import (
+	gotar "archive/tar"
+	"bytes"
+	"io"
+	"io/ioutil"
+	"os"
+	"path"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/containers/storage/pkg/archive"
+	"golang.org/x/sys/unix"
+	"gotest.tools/assert"
+)
+
+// Test for CVE-2018-15664
+// Assures that in the case where an "attacker" controlled path is a symlink to
+// some path outside of a container's rootfs that we do not copy data to a
+// container path that will actually overwrite data on the host
+func TestUntarWithMaliciousSymlinks(t *testing.T) {
+	dir, err := ioutil.TempDir("", t.Name())
+	assert.NilError(t, err)
+	defer os.RemoveAll(dir)
+
+	root := filepath.Join(dir, "root")
+
+	err = os.MkdirAll(root, 0755)
+	assert.NilError(t, err)
+
+	// Add a file into a directory above root
+	// Ensure that we can't access this file while tarring.
+	err = ioutil.WriteFile(filepath.Join(dir, "host-file"), []byte("I am a host file"), 0644)
+	assert.NilError(t, err)
+
+	// Create some data which which will be copied into the "container" root into
+	// the symlinked path.
+	// Before this change, the copy would overwrite the "host" content.
+	// With this change it should not.
+	data := filepath.Join(dir, "data")
+	err = os.MkdirAll(data, 0755)
+	assert.NilError(t, err)
+	err = ioutil.WriteFile(filepath.Join(data, "local-file"), []byte("pwn3d"), 0644)
+	assert.NilError(t, err)
+
+	safe := filepath.Join(root, "safe")
+	err = unix.Symlink(dir, safe)
+	assert.NilError(t, err)
+
+	rdr, err := archive.TarWithOptions(data, &archive.TarOptions{IncludeFiles: []string{"local-file"}, RebaseNames: map[string]string{"local-file": "host-file"}})
+	assert.NilError(t, err)
+
+	// Use tee to test both the good case and the bad case w/o recreating the archive
+	bufRdr := bytes.NewBuffer(nil)
+	tee := io.TeeReader(rdr, bufRdr)
+
+	err = UntarWithRoot(tee, safe, nil, root)
+	assert.Assert(t, err != nil)
+	assert.ErrorContains(t, err, "open /safe/host-file: no such file or directory")
+
+	// Make sure the "host" file is still in tact
+	// Before the fix the host file would be overwritten
+	hostData, err := ioutil.ReadFile(filepath.Join(dir, "host-file"))
+	assert.NilError(t, err)
+	assert.Equal(t, string(hostData), "I am a host file")
+
+	// Now test by chrooting to an attacker controlled path
+	// This should succeed as is and overwrite a "host" file
+	// Note that this would be a mis-use of this function.
+	err = UntarWithRoot(bufRdr, safe, nil, safe)
+	assert.NilError(t, err)
+
+	hostData, err = ioutil.ReadFile(filepath.Join(dir, "host-file"))
+	assert.NilError(t, err)
+	assert.Equal(t, string(hostData), "pwn3d")
+}
+
+// Test for CVE-2018-15664
+// Assures that in the case where an "attacker" controlled path is a symlink to
+// some path outside of a container's rootfs that we do not unwittingly leak
+// host data into the archive.
+func TestTarWithMaliciousSymlinks(t *testing.T) {
+	dir, err := ioutil.TempDir("", t.Name())
+	assert.NilError(t, err)
+	// defer os.RemoveAll(dir)
+	t.Log(dir)
+
+	root := filepath.Join(dir, "root")
+
+	err = os.MkdirAll(root, 0755)
+	assert.NilError(t, err)
+
+	hostFileData := []byte("I am a host file")
+
+	// Add a file into a directory above root
+	// Ensure that we can't access this file while tarring.
+	err = ioutil.WriteFile(filepath.Join(dir, "host-file"), hostFileData, 0644)
+	assert.NilError(t, err)
+
+	safe := filepath.Join(root, "safe")
+	err = unix.Symlink(dir, safe)
+	assert.NilError(t, err)
+
+	data := filepath.Join(dir, "data")
+	err = os.MkdirAll(data, 0755)
+	assert.NilError(t, err)
+
+	type testCase struct {
+		p        string
+		includes []string
+	}
+
+	cases := []testCase{
+		{p: safe, includes: []string{"host-file"}},
+		{p: safe + "/", includes: []string{"host-file"}},
+		{p: safe, includes: nil},
+		{p: safe + "/", includes: nil},
+		{p: root, includes: []string{"safe/host-file"}},
+		{p: root, includes: []string{"/safe/host-file"}},
+		{p: root, includes: nil},
+	}
+
+	maxBytes := len(hostFileData)
+
+	for _, tc := range cases {
+		t.Run(path.Join(tc.p+"_"+strings.Join(tc.includes, "_")), func(t *testing.T) {
+			// Here if we use archive.TarWithOptions directly or change the "root" parameter
+			// to be the same as "safe", data from the host will be leaked into the archive
+			var opts *archive.TarOptions
+			if tc.includes != nil {
+				opts = &archive.TarOptions{
+					IncludeFiles: tc.includes,
+				}
+			}
+			rdr, err := Tar(tc.p, opts, root)
+			assert.NilError(t, err)
+			defer rdr.Close()
+
+			tr := gotar.NewReader(rdr)
+			assert.Assert(t, !isDataInTar(t, tr, hostFileData, int64(maxBytes)), "host data leaked to archive")
+		})
+	}
+}
+
+func isDataInTar(t *testing.T, tr *gotar.Reader, compare []byte, maxBytes int64) bool {
+	for {
+		h, err := tr.Next()
+		if err == io.EOF {
+			break
+		}
+		assert.NilError(t, err)
+
+		if h.Size == 0 {
+			continue
+		}
+		assert.Assert(t, h.Size <= maxBytes, "%s: file size exceeds max expected size %d: %d", h.Name, maxBytes, h.Size)
+
+		data := make([]byte, int(h.Size))
+		_, err = io.ReadFull(tr, data)
+		assert.NilError(t, err)
+		if bytes.Contains(data, compare) {
+			return true
+		}
+	}
+
+	return false
+}

pkg/chrootarchive/archive_windows.go

@@ -14,9 +14,16 @@ func chroot(path string) error {
 
 func invokeUnpack(decompressedArchive io.ReadCloser,
 	dest string,
-	options *archive.TarOptions) error {
+	options *archive.TarOptions, root string) error {
 	// Windows is different to Linux here because Windows does not support
 	// chroot. Hence there is no point sandboxing a chrooted process to
 	// do the unpack. We call inline instead within the daemon process.
 	return archive.Unpack(decompressedArchive, longpath.AddPrefix(dest), options)
 }
+
+func invokePack(srcPath string, options *archive.TarOptions, root string) (io.ReadCloser, error) {
+	// Windows is different to Linux here because Windows does not support
+	// chroot. Hence there is no point sandboxing a chrooted process to
+	// do the pack. We call inline instead within the daemon process.
+	return archive.TarWithOptions(srcPath, options)
+}

pkg/chrootarchive/init_unix.go

@@ -14,6 +14,7 @@ import (
 func init() {
 	reexec.Register("storage-applyLayer", applyLayer)
 	reexec.Register("storage-untar", untar)
+	reexec.Register("storage-tar", tar)
 }
 
 func fatal(err error) {