diff --git a/client_test.go b/client_test.go
index db80a9c30..eb6d8539a 100644
--- a/client_test.go
+++ b/client_test.go
@@ -9,7 +9,6 @@ import (
 	golog "log"
 	"os"
 	"os/exec"
-	"runtime"
 	"testing"
 	"time"
 
@@ -17,6 +16,7 @@ import (
 
 	"github.com/containerd/containerd/log"
 	"github.com/containerd/containerd/namespaces"
+	"github.com/containerd/containerd/sys"
 	"github.com/containerd/containerd/testutil"
 	"github.com/sirupsen/logrus"
 )
@@ -62,7 +62,7 @@ func TestMain(m *testing.M) {
 	defer cancel()
 
 	if !noDaemon {
-		os.RemoveAll(defaultRoot)
+		sys.ForceRemoveAll(defaultRoot)
 
 		err := ctrd.start("containerd", address, []string{
 			"--root", defaultRoot,
@@ -99,17 +99,10 @@ func TestMain(m *testing.M) {
 	}).Info("running tests against containerd")
 
 	// pull a seed image
-	if runtime.GOOS != "windows" { // TODO: remove once pull is supported on windows
-		if _, err = client.Pull(ctx, testImage, WithPullUnpack); err != nil {
-			ctrd.Stop()
-			ctrd.Wait()
-			fmt.Fprintf(os.Stderr, "%s: %s\n", err, buf.String())
-			os.Exit(1)
-		}
-	}
-
-	if err := platformTestSetup(client); err != nil {
-		fmt.Fprintln(os.Stderr, "platform test setup failed", err)
+	if _, err = client.Pull(ctx, testImage, WithPullUnpack); err != nil {
+		ctrd.Stop()
+		ctrd.Wait()
+		fmt.Fprintf(os.Stderr, "%s: %s\n", err, buf.String())
 		os.Exit(1)
 	}
 
@@ -132,7 +125,7 @@ func TestMain(m *testing.M) {
 				fmt.Fprintln(os.Stderr, "failed to wait for containerd", err)
 			}
 		}
-		if err := os.RemoveAll(defaultRoot); err != nil {
+		if err := sys.ForceRemoveAll(defaultRoot); err != nil {
 			fmt.Fprintln(os.Stderr, "failed to remove test root dir", err)
 			os.Exit(1)
 		}
@@ -169,11 +162,6 @@ func TestNewClient(t *testing.T) {
 
 // All the container's tests depends on this, we need it to run first.
 func TestImagePull(t *testing.T) {
-	if runtime.GOOS == "windows" {
-		// TODO: remove once Windows has a snapshotter
-		t.Skip("Windows does not have a snapshotter yet")
-	}
-
 	client, err := newClient(t, address)
 	if err != nil {
 		t.Fatal(err)
diff --git a/client_unix_test.go b/client_unix_test.go
index 27fe7fab1..62cce45ab 100644
--- a/client_unix_test.go
+++ b/client_unix_test.go
@@ -16,10 +16,6 @@ var (
 	testImage string
 )
 
-func platformTestSetup(client *Client) error {
-	return nil
-}
-
 func init() {
 	switch runtime.GOARCH {
 	case "386":
diff --git a/client_windows_test.go b/client_windows_test.go
index 115d71497..4cb3e580f 100644
--- a/client_windows_test.go
+++ b/client_windows_test.go
@@ -1,88 +1,16 @@
 package containerd
 
 import (
-	"encoding/json"
-	"fmt"
 	"os"
 	"path/filepath"
-
-	"github.com/pkg/errors"
 )
 
 const (
 	defaultAddress = `\\.\pipe\containerd-containerd-test`
-	testImage      = "docker.io/library/go:nanoserver"
+	testImage      = "docker.io/microsoft/nanoserver:latest"
 )
 
 var (
-	dockerLayerFolders []string
-
 	defaultRoot  = filepath.Join(os.Getenv("programfiles"), "containerd", "root-test")
 	defaultState = filepath.Join(os.Getenv("programfiles"), "containerd", "state-test")
 )
-
-func platformTestSetup(client *Client) error {
-	var (
-		roots       []string
-		layerChains = make(map[string]string)
-	)
-	// Since we can't pull images yet, we'll piggyback on the default
-	// docker's images
-	wfPath := `C:\ProgramData\docker\windowsfilter`
-	wf, err := os.Open(wfPath)
-	if err != nil {
-		return errors.Wrapf(err, "failed to access docker layers @ %s", wfPath)
-	}
-	defer wf.Close()
-	entries, err := wf.Readdirnames(0)
-	if err != nil {
-		return errors.Wrapf(err, "failed to read %s entries", wfPath)
-	}
-
-	for _, fn := range entries {
-		layerChainPath := filepath.Join(wfPath, fn, "layerchain.json")
-		lfi, err := os.Stat(layerChainPath)
-		switch {
-		case err == nil && lfi.Mode().IsRegular():
-			f, err := os.OpenFile(layerChainPath, os.O_RDONLY, 0660)
-			if err != nil {
-				fmt.Fprintln(os.Stderr,
-					errors.Wrapf(err, "failed to open %s", layerChainPath))
-				continue
-			}
-			defer f.Close()
-			l := make([]string, 0)
-			if err := json.NewDecoder(f).Decode(&l); err != nil {
-				fmt.Fprintln(os.Stderr,
-					errors.Wrapf(err, "failed to decode %s", layerChainPath))
-				continue
-			}
-			switch {
-			case len(l) == 1:
-				layerChains[l[0]] = filepath.Join(wfPath, fn)
-			case len(l) > 1:
-				fmt.Fprintf(os.Stderr, "Too many entries in %s: %d", layerChainPath, len(l))
-			case len(l) == 0:
-				roots = append(roots, filepath.Join(wfPath, fn))
-			}
-		case os.IsNotExist(err):
-			// keep on going
-		default:
-			return errors.Wrapf(err, "error trying to access %s", layerChainPath)
-		}
-	}
-
-	// They'll be 2 roots, just take the first one
-	l := roots[0]
-	dockerLayerFolders = append(dockerLayerFolders, l)
-	for {
-		l = layerChains[l]
-		if l == "" {
-			break
-		}
-
-		dockerLayerFolders = append([]string{l}, dockerLayerFolders...)
-	}
-
-	return nil
-}
diff --git a/cmd/containerd/builtins_windows.go b/cmd/containerd/builtins_windows.go
index 381489170..95c9fb2f7 100644
--- a/cmd/containerd/builtins_windows.go
+++ b/cmd/containerd/builtins_windows.go
@@ -1,6 +1,7 @@
 package main
 
 import (
+	_ "github.com/containerd/containerd/diff/windows"
 	_ "github.com/containerd/containerd/snapshots/windows"
 	_ "github.com/containerd/containerd/windows"
 )
diff --git a/cmd/ctr/commands/run/run_windows.go b/cmd/ctr/commands/run/run_windows.go
index c33f2302b..82fca433a 100644
--- a/cmd/ctr/commands/run/run_windows.go
+++ b/cmd/ctr/commands/run/run_windows.go
@@ -7,32 +7,12 @@ import (
 	"github.com/containerd/containerd"
 	"github.com/containerd/containerd/cmd/ctr/commands"
 	"github.com/containerd/containerd/containers"
-	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/oci"
 	specs "github.com/opencontainers/runtime-spec/specs-go"
-	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
 	"github.com/urfave/cli"
 )
 
-func init() {
-	Command.Flags = append(Command.Flags, cli.StringSliceFlag{
-		Name:  "layer",
-		Usage: "HCSSHIM Layers to be used",
-	})
-}
-
-func withLayers(context *cli.Context) oci.SpecOpts {
-	return func(ctx gocontext.Context, client oci.Client, c *containers.Container, s *specs.Spec) error {
-		l := context.StringSlice("layer")
-		if l == nil {
-			return errors.Wrap(errdefs.ErrInvalidArgument, "base layers must be specified with `--layer`")
-		}
-		s.Windows.LayerFolders = l
-		return nil
-	}
-}
-
 func withTTY(terminal bool) oci.SpecOpts {
 	if !terminal {
 		return func(ctx gocontext.Context, client oci.Client, c *containers.Container, s *specs.Spec) error {
@@ -51,36 +31,39 @@ func withTTY(terminal bool) oci.SpecOpts {
 
 func newContainer(ctx gocontext.Context, client *containerd.Client, context *cli.Context) (containerd.Container, error) {
 	var (
-		// ref          = context.Args().First()
-		id           = context.Args().Get(1)
-		args         = context.Args()[2:]
-		tty          = context.Bool("tty")
-		labelStrings = context.StringSlice("label")
+		ref  = context.Args().First()
+		id   = context.Args().Get(1)
+		args = context.Args()[2:]
 	)
 
-	labels := commands.LabelArgs(labelStrings)
-
-	// TODO(mlaventure): get base image once we have a snapshotter
-
-	opts := []oci.SpecOpts{
-		// TODO(mlaventure): use oci.WithImageConfig once we have a snapshotter
-		withLayers(context),
-		oci.WithEnv(context.StringSlice("env")),
-		withMounts(context),
-		withTTY(tty),
+	image, err := client.GetImage(ctx, ref)
+	if err != nil {
+		return nil, err
 	}
+
+	var (
+		opts  []oci.SpecOpts
+		cOpts []containerd.NewContainerOpts
+	)
+	opts = append(opts, oci.WithImageConfig(image))
+	opts = append(opts, oci.WithEnv(context.StringSlice("env")))
+	opts = append(opts, withMounts(context))
 	if len(args) > 0 {
 		opts = append(opts, oci.WithProcessArgs(args...))
 	}
 	if cwd := context.String("cwd"); cwd != "" {
 		opts = append(opts, oci.WithProcessCwd(cwd))
 	}
-	return client.NewContainer(ctx, id,
-		containerd.WithNewSpec(opts...),
-		containerd.WithContainerLabels(labels),
-		containerd.WithRuntime(context.String("runtime"), nil),
-		// TODO(mlaventure): containerd.WithImage(image),
-	)
+	opts = append(opts, withTTY(context.Bool("tty")))
+
+	cOpts = append(cOpts, containerd.WithContainerLabels(commands.LabelArgs(context.StringSlice("label"))))
+	cOpts = append(cOpts, containerd.WithImage(image))
+	cOpts = append(cOpts, containerd.WithSnapshotter(context.String("snapshotter")))
+	cOpts = append(cOpts, containerd.WithNewSnapshot(id, image))
+	cOpts = append(cOpts, containerd.WithRuntime(context.String("runtime"), nil))
+
+	cOpts = append([]containerd.NewContainerOpts{containerd.WithNewSpec(opts...)}, cOpts...)
+	return client.NewContainer(ctx, id, cOpts...)
 }
 
 func getNewTaskOpts(_ *cli.Context) []containerd.NewTaskOpts {
diff --git a/container_linux_test.go b/container_linux_test.go
index 82e3a9a1a..6dc6b6484 100644
--- a/container_linux_test.go
+++ b/container_linux_test.go
@@ -187,7 +187,7 @@ func TestDaemonRestart(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "30")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "30")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -266,7 +266,7 @@ func TestContainerAttach(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withCat()), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withCat()), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -427,8 +427,8 @@ func TestContainerUsername(t *testing.T) {
 
 	// squid user in the alpine image has a uid of 31
 	container, err := client.NewContainer(ctx, id,
-		withNewSnapshot(id, image),
-		WithNewSpec(withImageConfig(image), oci.WithUsername("squid"), oci.WithProcessArgs("id", "-u")),
+		WithNewSnapshot(id, image),
+		WithNewSpec(oci.WithImageConfig(image), oci.WithUsername("squid"), oci.WithProcessArgs("id", "-u")),
 	)
 	if err != nil {
 		t.Fatal(err)
@@ -487,7 +487,7 @@ func TestContainerAttachProcess(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -620,8 +620,8 @@ func TestContainerUserID(t *testing.T) {
 
 	// adm user in the alpine image has a uid of 3 and gid of 4.
 	container, err := client.NewContainer(ctx, id,
-		withNewSnapshot(id, image),
-		WithNewSpec(withImageConfig(image), oci.WithUserID(3), oci.WithProcessArgs("sh", "-c", "echo $(id -u):$(id -g)")),
+		WithNewSnapshot(id, image),
+		WithNewSpec(oci.WithImageConfig(image), oci.WithUserID(3), oci.WithProcessArgs("sh", "-c", "echo $(id -u):$(id -g)")),
 	)
 	if err != nil {
 		t.Fatal(err)
@@ -674,8 +674,8 @@ func TestContainerKillAll(t *testing.T) {
 	}
 
 	container, err := client.NewContainer(ctx, id,
-		withNewSnapshot(id, image),
-		WithNewSpec(withImageConfig(image),
+		WithNewSnapshot(id, image),
+		WithNewSpec(oci.WithImageConfig(image),
 			withProcessArgs("sh", "-c", "top"),
 			oci.WithHostNamespace(specs.PIDNamespace),
 		),
@@ -730,7 +730,7 @@ func TestShimSigkilled(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image)), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image)), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -793,7 +793,7 @@ func TestDaemonRestartWithRunningShim(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), oci.WithProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), oci.WithProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -877,8 +877,8 @@ func TestContainerRuntimeOptions(t *testing.T) {
 
 	container, err := client.NewContainer(
 		ctx, id,
-		WithNewSpec(withImageConfig(image), withExitStatus(7)),
-		withNewSnapshot(id, image),
+		WithNewSpec(oci.WithImageConfig(image), withExitStatus(7)),
+		WithNewSnapshot(id, image),
 		WithRuntime("io.containerd.runtime.v1.linux", &runctypes.RuncOptions{Runtime: "no-runc"}),
 	)
 	if err != nil {
@@ -917,8 +917,8 @@ func TestContainerKillInitPidHost(t *testing.T) {
 	}
 
 	container, err := client.NewContainer(ctx, id,
-		withNewSnapshot(id, image),
-		WithNewSpec(withImageConfig(image),
+		WithNewSnapshot(id, image),
+		WithNewSpec(oci.WithImageConfig(image),
 			withProcessArgs("sh", "-c", "sleep 42; echo hi"),
 			oci.WithHostNamespace(specs.PIDNamespace),
 		),
@@ -1007,7 +1007,7 @@ func testUserNamespaces(t *testing.T, readonlyRootFS bool) {
 		t.Fatal(err)
 	}
 
-	opts := []NewContainerOpts{WithNewSpec(withImageConfig(image),
+	opts := []NewContainerOpts{WithNewSpec(oci.WithImageConfig(image),
 		withExitStatus(7),
 		oci.WithUserNamespace(0, 1000, 10000),
 	)}
@@ -1081,13 +1081,11 @@ func TestTaskResize(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withExitStatus(7)), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withExitStatus(7)), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
diff --git a/container_test.go b/container_test.go
index d3ed29ee7..257bcd9fb 100644
--- a/container_test.go
+++ b/container_test.go
@@ -97,13 +97,11 @@ func TestContainerStart(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withExitStatus(7)), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withExitStatus(7)), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -163,13 +161,11 @@ func TestContainerOutput(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("echo", expected)), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("echo", expected)), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -233,14 +229,12 @@ func TestContainerExec(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -320,14 +314,12 @@ func TestContainerLargeExecArgs(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -398,14 +390,12 @@ func TestContainerPids(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -477,14 +467,12 @@ func TestContainerCloseIO(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withCat()), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withCat()), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -535,14 +523,12 @@ func TestDeleteRunningContainer(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -592,14 +578,12 @@ func TestContainerKill(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "10")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "10")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -649,16 +633,14 @@ func TestContainerNoBinaryExists(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
 	container, err := client.NewContainer(ctx, id,
-		WithNewSpec(withImageConfig(image), oci.WithProcessArgs("nothing")),
-		withNewSnapshot(id, image))
+		WithNewSpec(oci.WithImageConfig(image), oci.WithProcessArgs("nothing")),
+		WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -698,14 +680,12 @@ func TestContainerExecNoBinaryExists(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -765,14 +745,12 @@ func TestWaitStoppedTask(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withExitStatus(7)), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withExitStatus(7)), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -830,14 +808,12 @@ func TestWaitStoppedProcess(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -920,13 +896,11 @@ func TestTaskForceDelete(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "30")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "30")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -963,13 +937,11 @@ func TestProcessForceDelete(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "30")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "30")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -1034,18 +1006,16 @@ func TestContainerHostname(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image),
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image),
 		withProcessArgs("hostname"),
 		oci.WithHostname(expected),
 	),
-		withNewSnapshot(id, image))
+		WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -1105,14 +1075,12 @@ func TestContainerExitedAtSet(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withTrue()), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withTrue()), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -1167,14 +1135,12 @@ func TestDeleteContainerExecCreated(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 
-	container, err := client.NewContainer(ctx, id, WithNewSpec(withImageConfig(image), withProcessArgs("sleep", "100")), withNewSnapshot(id, image))
+	container, err := client.NewContainer(ctx, id, WithNewSpec(oci.WithImageConfig(image), withProcessArgs("sleep", "100")), WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -1239,15 +1205,13 @@ func TestContainerMetrics(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 	container, err := client.NewContainer(ctx, id,
-		WithNewSpec(withImageConfig(image), oci.WithProcessArgs("sleep", "30")),
-		withNewSnapshot(id, image))
+		WithNewSpec(oci.WithImageConfig(image), oci.WithProcessArgs("sleep", "30")),
+		WithNewSnapshot(id, image))
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -1297,15 +1261,13 @@ func TestDeletedContainerMetrics(t *testing.T) {
 	)
 	defer cancel()
 
-	if runtime.GOOS != "windows" {
-		image, err = client.GetImage(ctx, testImage)
-		if err != nil {
-			t.Fatal(err)
-		}
+	image, err = client.GetImage(ctx, testImage)
+	if err != nil {
+		t.Fatal(err)
 	}
 	container, err := client.NewContainer(ctx, id,
-		WithNewSpec(withImageConfig(image), withExitStatus(0)),
-		withNewSnapshot(id, image),
+		WithNewSpec(oci.WithImageConfig(image), withExitStatus(0)),
+		WithNewSnapshot(id, image),
 	)
 	if err != nil {
 		t.Fatal(err)
diff --git a/diff/walking/differ.go b/diff/walking/differ.go
index eb6c2df70..f1dbf43f5 100644
--- a/diff/walking/differ.go
+++ b/diff/walking/differ.go
@@ -6,7 +6,6 @@ import (
 	"fmt"
 	"io"
 	"io/ioutil"
-	"strings"
 	"time"
 
 	"github.com/containerd/containerd/archive"
@@ -75,18 +74,10 @@ func (s *walkingDiff) Apply(ctx context.Context, desc ocispec.Descriptor, mounts
 			}).Debugf("diff applied")
 		}
 	}()
-	var isCompressed bool
-	switch desc.MediaType {
-	case ocispec.MediaTypeImageLayer, images.MediaTypeDockerSchema2Layer:
-	case ocispec.MediaTypeImageLayerGzip, images.MediaTypeDockerSchema2LayerGzip:
-		isCompressed = true
-	default:
-		// Still apply all generic media types *.tar[.+]gzip and *.tar
-		if strings.HasSuffix(desc.MediaType, ".tar.gzip") || strings.HasSuffix(desc.MediaType, ".tar+gzip") {
-			isCompressed = true
-		} else if !strings.HasSuffix(desc.MediaType, ".tar") {
-			return emptyDesc, errors.Wrapf(errdefs.ErrNotImplemented, "unsupported diff media type: %v", desc.MediaType)
-		}
+
+	isCompressed, err := images.IsCompressedDiff(ctx, desc.MediaType)
+	if err != nil {
+		return emptyDesc, errors.Wrapf(errdefs.ErrNotImplemented, "unsupported diff media type: %v", desc.MediaType)
 	}
 
 	var ocidesc ocispec.Descriptor
diff --git a/diff/windows/windows.go b/diff/windows/windows.go
new file mode 100644
index 000000000..0bb79e15e
--- /dev/null
+++ b/diff/windows/windows.go
@@ -0,0 +1,160 @@
+// +build windows
+
+package windows
+
+import (
+	"io"
+	"io/ioutil"
+	"time"
+
+	winio "github.com/Microsoft/go-winio"
+	"github.com/containerd/containerd/archive"
+	"github.com/containerd/containerd/archive/compression"
+	"github.com/containerd/containerd/content"
+	"github.com/containerd/containerd/diff"
+	"github.com/containerd/containerd/errdefs"
+	"github.com/containerd/containerd/images"
+	"github.com/containerd/containerd/log"
+	"github.com/containerd/containerd/metadata"
+	"github.com/containerd/containerd/mount"
+	"github.com/containerd/containerd/platforms"
+	"github.com/containerd/containerd/plugin"
+	digest "github.com/opencontainers/go-digest"
+	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
+	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
+	"golang.org/x/net/context"
+)
+
+func init() {
+	plugin.Register(&plugin.Registration{
+		Type: plugin.DiffPlugin,
+		ID:   "windows",
+		Requires: []plugin.Type{
+			plugin.MetadataPlugin,
+		},
+		InitFn: func(ic *plugin.InitContext) (interface{}, error) {
+			md, err := ic.Get(plugin.MetadataPlugin)
+			if err != nil {
+				return nil, err
+			}
+
+			ic.Meta.Platforms = append(ic.Meta.Platforms, platforms.DefaultSpec())
+			return NewWindowsDiff(md.(*metadata.DB).ContentStore())
+		},
+	})
+}
+
+type windowsDiff struct {
+	store content.Store
+}
+
+var emptyDesc = ocispec.Descriptor{}
+
+// NewWindowsDiff is the Windows container layer implementation of diff.Differ.
+func NewWindowsDiff(store content.Store) (diff.Differ, error) {
+	return &windowsDiff{
+		store: store,
+	}, nil
+}
+
+// Apply applies the content associated with the provided digests onto the
+// provided mounts. Archive content will be extracted and decompressed if
+// necessary.
+func (s *windowsDiff) Apply(ctx context.Context, desc ocispec.Descriptor, mounts []mount.Mount) (d ocispec.Descriptor, err error) {
+	t1 := time.Now()
+	defer func() {
+		if err == nil {
+			log.G(ctx).WithFields(logrus.Fields{
+				"d":     time.Now().Sub(t1),
+				"dgst":  desc.Digest,
+				"size":  desc.Size,
+				"media": desc.MediaType,
+			}).Debugf("diff applied")
+		}
+	}()
+
+	isCompressed, err := images.IsCompressedDiff(ctx, desc.MediaType)
+	if err != nil {
+		return emptyDesc, errors.Wrapf(errdefs.ErrNotImplemented, "unsupported diff media type: %v", desc.MediaType)
+	}
+
+	ra, err := s.store.ReaderAt(ctx, desc.Digest)
+	if err != nil {
+		return emptyDesc, errors.Wrap(err, "failed to get reader from content store")
+	}
+	defer ra.Close()
+
+	r := content.NewReader(ra)
+	if isCompressed {
+		ds, err := compression.DecompressStream(r)
+		if err != nil {
+			return emptyDesc, err
+		}
+		defer ds.Close()
+		r = ds
+	}
+
+	digester := digest.Canonical.Digester()
+	rc := &readCounter{
+		r: io.TeeReader(r, digester.Hash()),
+	}
+
+	layer, parentLayerPaths, err := mountsToLayerAndParents(mounts)
+	if err != nil {
+		return emptyDesc, err
+	}
+
+	// TODO darrenstahlmsft: When this is done isolated, we should disable these.
+	// it currently cannot be disabled, unless we add ref counting. Since this is
+	// temporary, leaving it enabled is OK for now.
+	if err := winio.EnableProcessPrivileges([]string{winio.SeBackupPrivilege, winio.SeRestorePrivilege}); err != nil {
+		return emptyDesc, err
+	}
+
+	if _, err := archive.Apply(ctx, layer, rc, archive.WithParentLayers(parentLayerPaths), archive.AsWindowsContainerLayer()); err != nil {
+		return emptyDesc, err
+	}
+
+	// Read any trailing data
+	if _, err := io.Copy(ioutil.Discard, rc); err != nil {
+		return emptyDesc, err
+	}
+
+	return ocispec.Descriptor{
+		MediaType: ocispec.MediaTypeImageLayer,
+		Size:      rc.c,
+		Digest:    digester.Digest(),
+	}, nil
+}
+
+// DiffMounts creates a diff between the given mounts and uploads the result
+// to the content store.
+func (s *windowsDiff) DiffMounts(ctx context.Context, lower, upper []mount.Mount, opts ...diff.Opt) (d ocispec.Descriptor, err error) {
+	return emptyDesc, errdefs.ErrNotImplemented
+}
+
+type readCounter struct {
+	r io.Reader
+	c int64
+}
+
+func (rc *readCounter) Read(p []byte) (n int, err error) {
+	n, err = rc.r.Read(p)
+	rc.c += int64(n)
+	return
+}
+
+func mountsToLayerAndParents(mounts []mount.Mount) (string, []string, error) {
+	if len(mounts) != 1 {
+		return "", nil, errors.Wrap(errdefs.ErrInvalidArgument, "number of mounts should always be 1 for Windows layers")
+	}
+	layer := mounts[0].Source
+
+	parentLayerPaths, err := mounts[0].GetParentPaths()
+	if err != nil {
+		return "", nil, err
+	}
+
+	return layer, parentLayerPaths, nil
+}
diff --git a/helpers_unix_test.go b/helpers_unix_test.go
index 51c034e10..8b028f9a1 100644
--- a/helpers_unix_test.go
+++ b/helpers_unix_test.go
@@ -39,8 +39,3 @@ func withExecExitStatus(s *specs.Process, es int) {
 func withExecArgs(s *specs.Process, args ...string) {
 	s.Args = args
 }
-
-var (
-	withNewSnapshot = WithNewSnapshot
-	withImageConfig = oci.WithImageConfig
-)
diff --git a/helpers_windows_test.go b/helpers_windows_test.go
index a3b4768f6..83ecb7602 100644
--- a/helpers_windows_test.go
+++ b/helpers_windows_test.go
@@ -39,17 +39,3 @@ func withExecExitStatus(s *specs.Process, es int) {
 func withExecArgs(s *specs.Process, args ...string) {
 	s.Args = append([]string{"powershell", "-noprofile"}, args...)
 }
-
-func withImageConfig(i Image) oci.SpecOpts {
-	return func(_ context.Context, _ oci.Client, _ *containers.Container, s *specs.Spec) error {
-		s.Windows.LayerFolders = dockerLayerFolders
-		return nil
-	}
-}
-
-func withNewSnapshot(id string, i Image) NewContainerOpts {
-	// TODO: when windows has a snapshotter remove the withNewSnapshot helper
-	return func(ctx context.Context, client *Client, c *containers.Container) error {
-		return nil
-	}
-}
diff --git a/images/image.go b/images/image.go
index 3f58e0620..5fea0dcfc 100644
--- a/images/image.go
+++ b/images/image.go
@@ -3,6 +3,7 @@ package images
 import (
 	"context"
 	"encoding/json"
+	"strings"
 	"time"
 
 	"github.com/containerd/containerd/content"
@@ -359,3 +360,22 @@ func RootFS(ctx context.Context, provider content.Provider, configDesc ocispec.D
 	}
 	return config.RootFS.DiffIDs, nil
 }
+
+// IsCompressedDiff returns true if mediaType is a known compressed diff media type.
+// It returns false if the media type is a diff, but not compressed. If the media type
+// is not a known diff type, it returns errdefs.ErrNotImplemented
+func IsCompressedDiff(ctx context.Context, mediaType string) (bool, error) {
+	switch mediaType {
+	case ocispec.MediaTypeImageLayer, MediaTypeDockerSchema2Layer:
+	case ocispec.MediaTypeImageLayerGzip, MediaTypeDockerSchema2LayerGzip:
+		return true, nil
+	default:
+		// Still apply all generic media types *.tar[.+]gzip and *.tar
+		if strings.HasSuffix(mediaType, ".tar.gzip") || strings.HasSuffix(mediaType, ".tar+gzip") {
+			return true, nil
+		} else if !strings.HasSuffix(mediaType, ".tar") {
+			return false, errdefs.ErrNotImplemented
+		}
+	}
+	return false, nil
+}
diff --git a/mount/mount_windows.go b/mount/mount_windows.go
index 8ad7eab12..6d37fda2a 100644
--- a/mount/mount_windows.go
+++ b/mount/mount_windows.go
@@ -1,6 +1,13 @@
 package mount
 
-import "github.com/pkg/errors"
+import (
+	"encoding/json"
+	"path/filepath"
+	"strings"
+
+	"github.com/Microsoft/hcsshim"
+	"github.com/pkg/errors"
+)
 
 var (
 	// ErrNotImplementOnWindows is returned when an action is not implemented for windows
@@ -9,15 +16,70 @@ var (
 
 // Mount to the provided target
 func (m *Mount) Mount(target string) error {
-	return ErrNotImplementOnWindows
+	home, layerID := filepath.Split(m.Source)
+
+	parentLayerPaths, err := m.GetParentPaths()
+	if err != nil {
+		return err
+	}
+
+	var di = hcsshim.DriverInfo{
+		HomeDir: home,
+	}
+
+	if err = hcsshim.ActivateLayer(di, layerID); err != nil {
+		return errors.Wrapf(err, "failed to activate layer %s", m.Source)
+	}
+	defer func() {
+		if err != nil {
+			hcsshim.DeactivateLayer(di, layerID)
+		}
+	}()
+
+	if err = hcsshim.PrepareLayer(di, layerID, parentLayerPaths); err != nil {
+		return errors.Wrapf(err, "failed to prepare layer %s", m.Source)
+	}
+	return nil
+}
+
+// ParentLayerPathsFlag is the options flag used to represent the JSON encoded
+// list of parent layers required to use the layer
+const ParentLayerPathsFlag = "parentLayerPaths="
+
+// GetParentPaths of the mount
+func (m *Mount) GetParentPaths() ([]string, error) {
+	var parentLayerPaths []string
+	for _, option := range m.Options {
+		if strings.HasPrefix(option, ParentLayerPathsFlag) {
+			err := json.Unmarshal([]byte(option[len(ParentLayerPathsFlag):]), &parentLayerPaths)
+			if err != nil {
+				return nil, errors.Wrap(err, "failed to unmarshal parent layer paths from mount")
+			}
+		}
+	}
+	return parentLayerPaths, nil
 }
 
 // Unmount the mount at the provided path
 func Unmount(mount string, flags int) error {
-	return ErrNotImplementOnWindows
+	var (
+		home, layerID = filepath.Split(mount)
+		di            = hcsshim.DriverInfo{
+			HomeDir: home,
+		}
+	)
+
+	if err := hcsshim.UnprepareLayer(di, layerID); err != nil {
+		return errors.Wrapf(err, "failed to unprepare layer %s", mount)
+	}
+	if err := hcsshim.DeactivateLayer(di, layerID); err != nil {
+		return errors.Wrapf(err, "failed to deactivate layer %s", mount)
+	}
+
+	return nil
 }
 
-// UnmountAll mounts at the provided path
+// UnmountAll unmounts from the provided path
 func UnmountAll(mount string, flags int) error {
-	return ErrNotImplementOnWindows
+	return Unmount(mount, flags)
 }
diff --git a/services/diff/service.go b/services/diff/service.go
index f847693a4..4e1c9fcc6 100644
--- a/services/diff/service.go
+++ b/services/diff/service.go
@@ -30,9 +30,7 @@ func init() {
 		Requires: []plugin.Type{
 			plugin.DiffPlugin,
 		},
-		Config: &config{
-			Order: []string{"walking"},
-		},
+		Config: defaultDifferConfig,
 		InitFn: func(ic *plugin.InitContext) (interface{}, error) {
 			differs, err := ic.GetByType(plugin.DiffPlugin)
 			if err != nil {
diff --git a/services/diff/service_unix.go b/services/diff/service_unix.go
new file mode 100644
index 000000000..39d88bbbb
--- /dev/null
+++ b/services/diff/service_unix.go
@@ -0,0 +1,7 @@
+// +build !windows
+
+package diff
+
+var defaultDifferConfig = &config{
+	Order: []string{"walking"},
+}
diff --git a/services/diff/service_windows.go b/services/diff/service_windows.go
new file mode 100644
index 000000000..afe193287
--- /dev/null
+++ b/services/diff/service_windows.go
@@ -0,0 +1,7 @@
+// +build windows
+
+package diff
+
+var defaultDifferConfig = &config{
+	Order: []string{"windows"},
+}
diff --git a/snapshots/windows/windows.go b/snapshots/windows/windows.go
index 603428e17..58469dd4f 100644
--- a/snapshots/windows/windows.go
+++ b/snapshots/windows/windows.go
@@ -4,16 +4,23 @@ package windows
 
 import (
 	"context"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"syscall"
+	"unsafe"
 
+	"github.com/Microsoft/hcsshim"
+	"github.com/containerd/containerd/errdefs"
+	"github.com/containerd/containerd/fs"
+	"github.com/containerd/containerd/log"
 	"github.com/containerd/containerd/mount"
 	"github.com/containerd/containerd/plugin"
 	"github.com/containerd/containerd/snapshots"
+	"github.com/containerd/containerd/snapshots/storage"
 	"github.com/pkg/errors"
-)
-
-var (
-	// ErrNotImplemented is returned when an action is not implemented
-	ErrNotImplemented = errors.New("not implemented")
+	"golang.org/x/sys/windows"
 )
 
 func init() {
@@ -28,12 +35,38 @@ func init() {
 
 type snapshotter struct {
 	root string
+	info hcsshim.DriverInfo
+	ms   *storage.MetaStore
 }
 
 // NewSnapshotter returns a new windows snapshotter
 func NewSnapshotter(root string) (snapshots.Snapshotter, error) {
+	fsType, err := getFileSystemType(root)
+	if err != nil {
+		return nil, err
+	}
+	if strings.ToLower(fsType) != "ntfs" {
+		return nil, errors.Wrapf(errdefs.ErrInvalidArgument, "%s is not on an NTFS volume - only NTFS volumes are supported", root)
+	}
+
+	if err := os.MkdirAll(root, 0700); err != nil {
+		return nil, err
+	}
+	ms, err := storage.NewMetaStore(filepath.Join(root, "metadata.db"))
+	if err != nil {
+		return nil, err
+	}
+
+	if err := os.Mkdir(filepath.Join(root, "snapshots"), 0700); err != nil && !os.IsExist(err) {
+		return nil, err
+	}
+
 	return &snapshotter{
+		info: hcsshim.DriverInfo{
+			HomeDir: filepath.Join(root, "snapshots"),
+		},
 		root: root,
+		ms:   ms,
 	}, nil
 }
 
@@ -42,50 +75,270 @@ func NewSnapshotter(root string) (snapshots.Snapshotter, error) {
 //
 // Should be used for parent resolution, existence checks and to discern
 // the kind of snapshot.
-func (o *snapshotter) Stat(ctx context.Context, key string) (snapshots.Info, error) {
-	panic("not implemented")
+func (s *snapshotter) Stat(ctx context.Context, key string) (snapshots.Info, error) {
+	ctx, t, err := s.ms.TransactionContext(ctx, false)
+	if err != nil {
+		return snapshots.Info{}, err
+	}
+	defer t.Rollback()
+
+	_, info, _, err := storage.GetInfo(ctx, key)
+	return info, err
 }
 
-func (o *snapshotter) Update(ctx context.Context, info snapshots.Info, fieldpaths ...string) (snapshots.Info, error) {
-	panic("not implemented")
+func (s *snapshotter) Update(ctx context.Context, info snapshots.Info, fieldpaths ...string) (snapshots.Info, error) {
+	ctx, t, err := s.ms.TransactionContext(ctx, true)
+	if err != nil {
+		return snapshots.Info{}, err
+	}
+	defer t.Rollback()
+
+	info, err = storage.UpdateInfo(ctx, info, fieldpaths...)
+	if err != nil {
+		return snapshots.Info{}, err
+	}
+
+	if err := t.Commit(); err != nil {
+		return snapshots.Info{}, err
+	}
+
+	return info, nil
 }
 
-func (o *snapshotter) Usage(ctx context.Context, key string) (snapshots.Usage, error) {
-	panic("not implemented")
+func (s *snapshotter) Usage(ctx context.Context, key string) (snapshots.Usage, error) {
+	ctx, t, err := s.ms.TransactionContext(ctx, false)
+	if err != nil {
+		return snapshots.Usage{}, err
+	}
+	defer t.Rollback()
+
+	_, info, usage, err := storage.GetInfo(ctx, key)
+	if err != nil {
+		return snapshots.Usage{}, err
+	}
+
+	if info.Kind == snapshots.KindActive {
+		du := fs.Usage{
+			Size: 0,
+		}
+		usage = snapshots.Usage(du)
+	}
+
+	return usage, nil
 }
 
-func (o *snapshotter) Prepare(ctx context.Context, key, parent string, opts ...snapshots.Opt) ([]mount.Mount, error) {
-	panic("not implemented")
+func (s *snapshotter) Prepare(ctx context.Context, key, parent string, opts ...snapshots.Opt) ([]mount.Mount, error) {
+	return s.createSnapshot(ctx, snapshots.KindActive, key, parent, opts)
 }
 
-func (o *snapshotter) View(ctx context.Context, key, parent string, opts ...snapshots.Opt) ([]mount.Mount, error) {
-	panic("not implemented")
+func (s *snapshotter) View(ctx context.Context, key, parent string, opts ...snapshots.Opt) ([]mount.Mount, error) {
+	return s.createSnapshot(ctx, snapshots.KindView, key, parent, opts)
 }
 
 // Mounts returns the mounts for the transaction identified by key. Can be
 // called on an read-write or readonly transaction.
 //
 // This can be used to recover mounts after calling View or Prepare.
-func (o *snapshotter) Mounts(ctx context.Context, key string) ([]mount.Mount, error) {
-	panic("not implemented")
+func (s *snapshotter) Mounts(ctx context.Context, key string) ([]mount.Mount, error) {
+	ctx, t, err := s.ms.TransactionContext(ctx, false)
+	if err != nil {
+		return nil, err
+	}
+	defer t.Rollback()
+
+	snapshot, err := storage.GetSnapshot(ctx, key)
+	if err != nil {
+		return nil, errors.Wrap(err, "failed to get snapshot mount")
+	}
+	return s.mounts(snapshot), nil
 }
 
-func (o *snapshotter) Commit(ctx context.Context, name, key string, opts ...snapshots.Opt) error {
-	panic("not implemented")
+func (s *snapshotter) Commit(ctx context.Context, name, key string, opts ...snapshots.Opt) error {
+	ctx, t, err := s.ms.TransactionContext(ctx, true)
+	if err != nil {
+		return err
+	}
+	defer t.Rollback()
+
+	usage := fs.Usage{
+		Size: 0,
+	}
+
+	if _, err = storage.CommitActive(ctx, key, name, snapshots.Usage(usage), opts...); err != nil {
+		return errors.Wrap(err, "failed to commit snapshot")
+	}
+
+	if err := t.Commit(); err != nil {
+		return err
+	}
+	return nil
 }
 
 // Remove abandons the transaction identified by key. All resources
 // associated with the key will be removed.
-func (o *snapshotter) Remove(ctx context.Context, key string) error {
-	panic("not implemented")
+func (s *snapshotter) Remove(ctx context.Context, key string) error {
+	ctx, t, err := s.ms.TransactionContext(ctx, true)
+	if err != nil {
+		return err
+	}
+	defer t.Rollback()
+
+	id, _, err := storage.Remove(ctx, key)
+	if err != nil {
+		return errors.Wrap(err, "failed to remove")
+	}
+
+	path := s.getSnapshotDir(id)
+	renamedID := "rm-" + id
+	renamed := filepath.Join(s.root, "snapshots", "rm-"+id)
+	if err := os.Rename(path, renamed); err != nil && !os.IsNotExist(err) {
+		return err
+	}
+
+	if err := t.Commit(); err != nil {
+		if err1 := os.Rename(renamed, path); err1 != nil {
+			// May cause inconsistent data on disk
+			log.G(ctx).WithError(err1).WithField("path", renamed).Errorf("Failed to rename after failed commit")
+		}
+		return errors.Wrap(err, "failed to commit")
+	}
+
+	if err := hcsshim.DestroyLayer(s.info, renamedID); err != nil {
+		// Must be cleaned up, any "rm-*" could be removed if no active transactions
+		log.G(ctx).WithError(err).WithField("path", renamed).Warnf("Failed to remove root filesystem")
+	}
+
+	return nil
 }
 
 // Walk the committed snapshots.
-func (o *snapshotter) Walk(ctx context.Context, fn func(context.Context, snapshots.Info) error) error {
-	panic("not implemented")
+func (s *snapshotter) Walk(ctx context.Context, fn func(context.Context, snapshots.Info) error) error {
+	ctx, t, err := s.ms.TransactionContext(ctx, false)
+	if err != nil {
+		return err
+	}
+	defer t.Rollback()
+
+	return storage.WalkInfo(ctx, fn)
 }
 
 // Close closes the snapshotter
-func (o *snapshotter) Close() error {
-	panic("not implemented")
+func (s *snapshotter) Close() error {
+	return s.ms.Close()
+}
+
+func (s *snapshotter) mounts(sn storage.Snapshot) []mount.Mount {
+	var (
+		roFlag           string
+		source           string
+		parentLayerPaths []string
+	)
+
+	if sn.Kind == snapshots.KindView {
+		roFlag = "ro"
+	} else {
+		roFlag = "rw"
+	}
+
+	if len(sn.ParentIDs) == 0 || sn.Kind == snapshots.KindActive {
+		source = s.getSnapshotDir(sn.ID)
+		parentLayerPaths = s.parentIDsToParentPaths(sn.ParentIDs)
+	} else {
+		source = s.getSnapshotDir(sn.ParentIDs[0])
+		parentLayerPaths = s.parentIDsToParentPaths(sn.ParentIDs[1:])
+	}
+
+	// error is not checked here, as a string array will never fail to Marshal
+	parentLayersJSON, _ := json.Marshal(parentLayerPaths)
+	parentLayersOption := mount.ParentLayerPathsFlag + string(parentLayersJSON)
+
+	var mounts []mount.Mount
+	mounts = append(mounts, mount.Mount{
+		Source: source,
+		Type:   "windows-layer",
+		Options: []string{
+			roFlag,
+			parentLayersOption,
+		},
+	})
+
+	return mounts
+}
+
+func (s *snapshotter) getSnapshotDir(id string) string {
+	return filepath.Join(s.root, "snapshots", id)
+}
+
+func (s *snapshotter) createSnapshot(ctx context.Context, kind snapshots.Kind, key, parent string, opts []snapshots.Opt) ([]mount.Mount, error) {
+	ctx, t, err := s.ms.TransactionContext(ctx, true)
+	if err != nil {
+		return nil, err
+	}
+	defer t.Rollback()
+
+	newSnapshot, err := storage.CreateSnapshot(ctx, kind, key, parent, opts...)
+	if err != nil {
+		return nil, errors.Wrap(err, "failed to create snapshot")
+	}
+
+	if kind == snapshots.KindActive {
+		parentLayerPaths := s.parentIDsToParentPaths(newSnapshot.ParentIDs)
+
+		var parentPath string
+		if len(parentLayerPaths) != 0 {
+			parentPath = parentLayerPaths[0]
+		}
+
+		if err := hcsshim.CreateSandboxLayer(s.info, newSnapshot.ID, parentPath, parentLayerPaths); err != nil {
+			return nil, errors.Wrap(err, "failed to create sandbox layer")
+		}
+
+		// TODO(darrenstahlmsft): Allow changing sandbox size
+	}
+
+	if err := t.Commit(); err != nil {
+		return nil, errors.Wrap(err, "commit failed")
+	}
+
+	return s.mounts(newSnapshot), nil
+}
+
+func (s *snapshotter) parentIDsToParentPaths(parentIDs []string) []string {
+	var parentLayerPaths []string
+	for _, ID := range parentIDs {
+		parentLayerPaths = append(parentLayerPaths, s.getSnapshotDir(ID))
+	}
+	return parentLayerPaths
+}
+
+// getFileSystemType obtains the type of a file system through GetVolumeInformation
+// https://msdn.microsoft.com/en-us/library/windows/desktop/aa364993(v=vs.85).aspx
+func getFileSystemType(path string) (fsType string, hr error) {
+	drive := filepath.VolumeName(path)
+	if len(drive) != 2 {
+		return "", errors.New("getFileSystemType path must start with a drive letter")
+	}
+
+	var (
+		modkernel32              = windows.NewLazySystemDLL("kernel32.dll")
+		procGetVolumeInformation = modkernel32.NewProc("GetVolumeInformationW")
+		buf                      = make([]uint16, 255)
+		size                     = windows.MAX_PATH + 1
+	)
+	drive += `\`
+	n := uintptr(unsafe.Pointer(nil))
+	r0, _, _ := syscall.Syscall9(procGetVolumeInformation.Addr(), 8, uintptr(unsafe.Pointer(windows.StringToUTF16Ptr(drive))), n, n, n, n, n, uintptr(unsafe.Pointer(&buf[0])), uintptr(size), 0)
+	if int32(r0) < 0 {
+		hr = syscall.Errno(win32FromHresult(r0))
+	}
+	fsType = windows.UTF16ToString(buf)
+	return
+}
+
+// win32FromHresult is a helper function to get the win32 error code from an HRESULT
+func win32FromHresult(hr uintptr) uintptr {
+	if hr&0x1fff0000 == 0x00070000 {
+		return hr & 0xffff
+	}
+	return hr
 }
diff --git a/sys/filesys_unix.go b/sys/filesys_unix.go
new file mode 100644
index 000000000..ed23609c5
--- /dev/null
+++ b/sys/filesys_unix.go
@@ -0,0 +1,10 @@
+// +build !windows
+
+package sys
+
+import "os"
+
+// ForceRemoveAll on unix is just a wrapper for os.RemoveAll
+func ForceRemoveAll(path string) error {
+	return os.RemoveAll(path)
+}
diff --git a/sys/filesys_windows.go b/sys/filesys_windows.go
index b5ce13579..36395c556 100644
--- a/sys/filesys_windows.go
+++ b/sys/filesys_windows.go
@@ -11,6 +11,7 @@ import (
 	"unsafe"
 
 	winio "github.com/Microsoft/go-winio"
+	"github.com/Microsoft/hcsshim"
 )
 
 // MkdirAllWithACL is a wrapper for MkdirAll that creates a directory
@@ -234,3 +235,13 @@ func syscallOpenSequential(path string, mode int, _ uint32) (fd syscall.Handle,
 	h, e := syscall.CreateFile(pathp, access, sharemode, sa, createmode, fileFlagSequentialScan, 0)
 	return h, e
 }
+
+// ForceRemoveAll is the same as os.RemoveAll, but uses hcsshim.DestroyLayer in order
+// to delete container layers.
+func ForceRemoveAll(path string) error {
+	info := hcsshim.DriverInfo{
+		HomeDir: filepath.Dir(path),
+	}
+
+	return hcsshim.DestroyLayer(info, filepath.Base(path))
+}
diff --git a/windows/hcsshim.go b/windows/hcsshim.go
index 7801da749..8d9501361 100644
--- a/windows/hcsshim.go
+++ b/windows/hcsshim.go
@@ -4,14 +4,12 @@ package windows
 
 import (
 	"context"
-	"fmt"
 	"os"
 	"path/filepath"
 	"strings"
 
 	"github.com/Microsoft/hcsshim"
 	"github.com/containerd/containerd/errdefs"
-	"github.com/containerd/containerd/log"
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/pkg/errors"
 )
@@ -49,14 +47,14 @@ func newWindowsContainerConfig(ctx context.Context, owner, id string, spec *spec
 	}
 	conf.IgnoreFlushesDuringBoot = spec.Windows.IgnoreFlushesDuringBoot
 
-	if len(spec.Windows.LayerFolders) < 1 {
+	if len(spec.Windows.LayerFolders) < 2 {
 		return nil, errors.Wrap(errdefs.ErrInvalidArgument,
-			"spec.Windows.LayerFolders must have at least 1 layers")
+			"spec.Windows.LayerFolders must have at least 2 layers")
 	}
 	var (
-		layerFolders    = spec.Windows.LayerFolders
-		homeDir         = filepath.Dir(layerFolders[0])
-		layerFolderPath = filepath.Join(homeDir, id)
+		layerFolderPath = spec.Windows.LayerFolders[0]
+		layerFolders    = spec.Windows.LayerFolders[1:]
+		layerID         = filepath.Base(layerFolderPath)
 	)
 
 	// TODO: use the create request Mount for those
@@ -71,39 +69,12 @@ func newWindowsContainerConfig(ctx context.Context, owner, id string, spec *spec
 			Path: layerPath,
 		})
 	}
-
-	var (
-		di = hcsshim.DriverInfo{
-			Flavour: 1, // filter driver
-			HomeDir: homeDir,
-		}
-	)
 	conf.LayerFolderPath = layerFolderPath
 
-	// TODO: Once there is a snapshotter for windows, this can be deleted.
-	// The R/W Layer should come from the Rootfs Mounts provided
-	//
-	// Windows doesn't support creating a container with a readonly
-	// filesystem, so always create a RW one
-	if err = hcsshim.CreateSandboxLayer(di, id, layerFolders[0], layerFolders); err != nil {
-		return nil, errors.Wrapf(err, "failed to create sandbox layer for %s: layers: %#v, driverInfo: %#v",
-			id, layerFolders, di)
+	var di = hcsshim.DriverInfo{
+		HomeDir: filepath.Dir(layerFolderPath),
 	}
-	defer func() {
-		if err != nil {
-			removeLayer(ctx, conf.LayerFolderPath)
-		}
-	}()
-
-	if err = hcsshim.ActivateLayer(di, id); err != nil {
-		return nil, errors.Wrapf(err, "failed to activate layer %s", conf.LayerFolderPath)
-	}
-
-	if err = hcsshim.PrepareLayer(di, id, layerFolders); err != nil {
-		return nil, errors.Wrapf(err, "failed to prepare layer %s", conf.LayerFolderPath)
-	}
-
-	conf.VolumePath, err = hcsshim.GetLayerMountPath(di, id)
+	conf.VolumePath, err = hcsshim.GetLayerMountPath(di, layerID)
 	if err != nil {
 		return nil, errors.Wrapf(err, "failed to getmount path for layer %s: driverInfo: %#v", id, di)
 	}
@@ -146,41 +117,6 @@ func newWindowsContainerConfig(ctx context.Context, owner, id string, spec *spec
 	return conf, nil
 }
 
-// removeLayer deletes the given layer, all associated containers must have
-// been shutdown for this to succeed.
-func removeLayer(ctx context.Context, path string) error {
-	var (
-		err        error
-		layerID    = filepath.Base(path)
-		parentPath = filepath.Dir(path)
-		di         = hcsshim.DriverInfo{
-			Flavour: 1, // filter driver
-			HomeDir: parentPath,
-		}
-	)
-
-	if err = hcsshim.UnprepareLayer(di, layerID); err != nil {
-		log.G(ctx).WithError(err).Warnf("failed to unprepare layer %s for removal", path)
-	}
-
-	if err = hcsshim.DeactivateLayer(di, layerID); err != nil {
-		log.G(ctx).WithError(err).Warnf("failed to deactivate layer %s for removal", path)
-	}
-
-	removePath := filepath.Join(parentPath, fmt.Sprintf("%s-removing", layerID))
-	if err = os.Rename(path, removePath); err != nil {
-		log.G(ctx).WithError(err).Warnf("failed to rename container layer %s for removal", path)
-		removePath = path
-	}
-
-	if err = hcsshim.DestroyLayer(di, removePath); err != nil {
-		log.G(ctx).WithError(err).Errorf("failed to remove container layer %s", removePath)
-		return err
-	}
-
-	return nil
-}
-
 func newProcessConfig(processSpec *specs.Process, pset *pipeSet) *hcsshim.ProcessConfig {
 	conf := &hcsshim.ProcessConfig{
 		EmulateConsole:   pset.src.Terminal,
diff --git a/windows/meta.go b/windows/meta.go
deleted file mode 100644
index 81ab9245e..000000000
--- a/windows/meta.go
+++ /dev/null
@@ -1,54 +0,0 @@
-// +build windows
-
-package windows
-
-// TODO: remove this file (i.e. meta.go) once we have a snapshotter
-
-import (
-	"github.com/boltdb/bolt"
-	"github.com/containerd/containerd/errdefs"
-	"github.com/pkg/errors"
-)
-
-func newLayerFolderStore(tx *bolt.Tx) *layerFolderStore {
-	return &layerFolderStore{tx}
-}
-
-type layerFolderStore struct {
-	tx *bolt.Tx
-}
-
-func (s *layerFolderStore) Create(id, layer string) error {
-	bkt, err := s.tx.CreateBucketIfNotExists([]byte(pluginID))
-	if err != nil {
-		return errors.Wrapf(err, "failed to create bucket %s", pluginID)
-	}
-	err = bkt.Put([]byte(id), []byte(layer))
-	if err != nil {
-		return errors.Wrapf(err, "failed to store entry %s:%s", id, layer)
-	}
-
-	return nil
-}
-
-func (s *layerFolderStore) Get(id string) (string, error) {
-	bkt := s.tx.Bucket([]byte(pluginID))
-	if bkt == nil {
-		return "", errors.Wrapf(errdefs.ErrNotFound, "bucket %s", pluginID)
-	}
-
-	return string(bkt.Get([]byte(id))), nil
-}
-
-func (s *layerFolderStore) Delete(id string) error {
-	bkt := s.tx.Bucket([]byte(pluginID))
-	if bkt == nil {
-		return errors.Wrapf(errdefs.ErrNotFound, "bucket %s", pluginID)
-	}
-
-	if err := bkt.Delete([]byte(id)); err != nil {
-		return errors.Wrapf(err, "failed to delete entry %s", id)
-	}
-
-	return nil
-}
diff --git a/windows/runtime.go b/windows/runtime.go
index fd26b270c..542c97363 100644
--- a/windows/runtime.go
+++ b/windows/runtime.go
@@ -10,13 +10,12 @@ import (
 	"time"
 
 	"github.com/Microsoft/hcsshim"
-	"github.com/boltdb/bolt"
 	eventstypes "github.com/containerd/containerd/api/events"
-	containerdtypes "github.com/containerd/containerd/api/types"
+	"github.com/containerd/containerd/api/types"
 	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/events"
 	"github.com/containerd/containerd/log"
-	"github.com/containerd/containerd/metadata"
+	"github.com/containerd/containerd/mount"
 	"github.com/containerd/containerd/namespaces"
 	"github.com/containerd/containerd/plugin"
 	"github.com/containerd/containerd/runtime"
@@ -55,12 +54,6 @@ func New(ic *plugin.InitContext) (interface{}, error) {
 	if err := os.MkdirAll(ic.Root, 0700); err != nil {
 		return nil, errors.Wrapf(err, "could not create state directory at %s", ic.Root)
 	}
-
-	m, err := ic.Get(plugin.MetadataPlugin)
-	if err != nil {
-		return nil, err
-	}
-
 	r := &windowsRuntime{
 		root:    ic.Root,
 		pidPool: newPidPool(),
@@ -70,7 +63,6 @@ func New(ic *plugin.InitContext) (interface{}, error) {
 		// TODO(mlaventure): windows needs a stat monitor
 		monitor: nil,
 		tasks:   runtime.NewTaskList(),
-		db:      m.(*metadata.DB),
 	}
 
 	// Load our existing containers and kill/delete them. We don't support
@@ -91,7 +83,6 @@ type windowsRuntime struct {
 
 	monitor runtime.TaskMonitor
 	tasks   *runtime.TaskList
-	db      *metadata.DB
 }
 
 func (r *windowsRuntime) ID() string {
@@ -125,7 +116,17 @@ func (r *windowsRuntime) Create(ctx context.Context, id string, opts runtime.Cre
 		createOpts.TerminateDuration = defaultTerminateDuration
 	}
 
-	return r.newTask(ctx, namespace, id, spec, opts.IO, createOpts)
+	if len(opts.Rootfs) == 0 {
+		return nil, errors.Wrap(errdefs.ErrInvalidArgument, "rootfs was not provided to container create")
+	}
+	spec.Windows.LayerFolders = append(spec.Windows.LayerFolders, opts.Rootfs[0].Source)
+	parentLayerPaths, err := opts.Rootfs[0].GetParentPaths()
+	if err != nil {
+		return nil, err
+	}
+	spec.Windows.LayerFolders = append(spec.Windows.LayerFolders, parentLayerPaths...)
+
+	return r.newTask(ctx, namespace, id, opts.Rootfs, spec, opts.IO, createOpts)
 }
 
 func (r *windowsRuntime) Get(ctx context.Context, id string) (runtime.Task, error) {
@@ -209,14 +210,19 @@ func (r *windowsRuntime) Delete(ctx context.Context, t runtime.Task) (*runtime.E
 		ns, _ := namespaces.Namespace(ctx)
 		serviceCtx := log.WithLogger(context.Background(), log.GetLogger(ctx))
 		serviceCtx = namespaces.WithNamespace(serviceCtx, ns)
-		r.serviceTask(serviceCtx, ns, wt.id+"_servicing", wt.spec)
+		r.serviceTask(serviceCtx, ns, wt.id+"_servicing", wt.rootfs, wt.spec)
+	}
+
+	if err := mount.UnmountAll(wt.rootfs[0].Source, 0); err != nil {
+		log.G(ctx).WithError(err).WithField("path", wt.rootfs[0].Source).
+			Warn("failed to unmount rootfs on failure")
 	}
 
 	// We were never started, return failure
 	return rtExit, nil
 }
 
-func (r *windowsRuntime) newTask(ctx context.Context, namespace, id string, spec *specs.Spec, io runtime.IO, createOpts *hcsshimtypes.CreateOptions) (*task, error) {
+func (r *windowsRuntime) newTask(ctx context.Context, namespace, id string, rootfs []mount.Mount, spec *specs.Spec, io runtime.IO, createOpts *hcsshimtypes.CreateOptions) (*task, error) {
 	var (
 		err  error
 		pset *pipeSet
@@ -241,6 +247,18 @@ func (r *windowsRuntime) newTask(ctx context.Context, namespace, id string, spec
 		}
 	}()
 
+	if err := mount.All(rootfs, ""); err != nil {
+		return nil, errors.Wrap(err, "failed to mount rootfs")
+	}
+	defer func() {
+		if err != nil {
+			if err := mount.UnmountAll(rootfs[0].Source, 0); err != nil {
+				log.G(ctx).WithError(err).WithField("path", rootfs[0].Source).
+					Warn("failed to unmount rootfs on failure")
+			}
+		}
+	}()
+
 	var (
 		conf *hcsshim.ContainerConfig
 		nsid = namespace + "-" + id
@@ -248,31 +266,6 @@ func (r *windowsRuntime) newTask(ctx context.Context, namespace, id string, spec
 	if conf, err = newWindowsContainerConfig(ctx, hcsshimOwner, nsid, spec); err != nil {
 		return nil, err
 	}
-	defer func() {
-		if err != nil {
-			removeLayer(ctx, conf.LayerFolderPath)
-		}
-	}()
-
-	// TODO: remove this once we have a windows snapshotter
-	// Store the LayerFolder in the db so we can clean it if we die
-	if err = r.db.Update(func(tx *bolt.Tx) error {
-		s := newLayerFolderStore(tx)
-		return s.Create(nsid, conf.LayerFolderPath)
-	}); err != nil {
-		return nil, err
-	}
-	defer func() {
-		if err != nil {
-			if dbErr := r.db.Update(func(tx *bolt.Tx) error {
-				s := newLayerFolderStore(tx)
-				return s.Delete(nsid)
-			}); dbErr != nil {
-				log.G(ctx).WithField("id", id).
-					Error("failed to remove key from metadata")
-			}
-		}
-	}()
 
 	ctr, err := hcsshim.CreateContainer(nsid, conf)
 	if err != nil {
@@ -301,6 +294,7 @@ func (r *windowsRuntime) newTask(ctx context.Context, namespace, id string, spec
 		hyperV:            spec.Windows.HyperV != nil,
 		publisher:         r.publisher,
 		rwLayer:           conf.LayerFolderPath,
+		rootfs:            rootfs,
 		pidPool:           r.pidPool,
 		hcsContainer:      ctr,
 		terminateDuration: createOpts.TerminateDuration,
@@ -312,11 +306,12 @@ func (r *windowsRuntime) newTask(ctx context.Context, namespace, id string, spec
 	}
 	r.tasks.Add(ctx, t)
 
-	var rootfs []*containerdtypes.Mount
-	for _, l := range append([]string{t.rwLayer}, spec.Windows.LayerFolders...) {
-		rootfs = append(rootfs, &containerdtypes.Mount{
-			Type:   "windows-layer",
-			Source: l,
+	var eventRootfs []*types.Mount
+	for _, m := range rootfs {
+		eventRootfs = append(eventRootfs, &types.Mount{
+			Type:    m.Type,
+			Source:  m.Source,
+			Options: m.Options,
 		})
 	}
 
@@ -331,7 +326,7 @@ func (r *windowsRuntime) newTask(ctx context.Context, namespace, id string, spec
 				Terminal: io.Terminal,
 			},
 			Pid:    t.pid,
-			Rootfs: rootfs,
+			Rootfs: eventRootfs,
 			// TODO: what should be in Bundle for windows?
 		})
 
@@ -360,34 +355,10 @@ func (r *windowsRuntime) cleanup(ctx context.Context) {
 			container.Wait()
 		}
 		container.Close()
-
-		// TODO: remove this once we have a windows snapshotter
-		var layerFolderPath string
-		if err := r.db.View(func(tx *bolt.Tx) error {
-			s := newLayerFolderStore(tx)
-			l, e := s.Get(p.ID)
-			if err == nil {
-				layerFolderPath = l
-			}
-			return e
-		}); err == nil && layerFolderPath != "" {
-			removeLayer(ctx, layerFolderPath)
-			if dbErr := r.db.Update(func(tx *bolt.Tx) error {
-				s := newLayerFolderStore(tx)
-				return s.Delete(p.ID)
-			}); dbErr != nil {
-				log.G(ctx).WithField("id", p.ID).
-					Error("failed to remove key from metadata")
-			}
-		} else {
-			log.G(ctx).WithField("id", p.ID).
-				Debug("key not found in metadata, R/W layer may be leaked")
-		}
-
 	}
 }
 
-func (r *windowsRuntime) serviceTask(ctx context.Context, namespace, id string, spec *specs.Spec) {
+func (r *windowsRuntime) serviceTask(ctx context.Context, namespace, id string, rootfs []mount.Mount, spec *specs.Spec) {
 	var (
 		err        error
 		t          *task
@@ -397,7 +368,7 @@ func (r *windowsRuntime) serviceTask(ctx context.Context, namespace, id string,
 		}
 	)
 
-	t, err = r.newTask(ctx, namespace, id, spec, io, createOpts)
+	t, err = r.newTask(ctx, namespace, id, rootfs, spec, io, createOpts)
 	if err != nil {
 		log.G(ctx).WithError(err).WithField("id", id).
 			Warn("failed to created servicing task")
diff --git a/windows/task.go b/windows/task.go
index 89bce1694..950f2e966 100644
--- a/windows/task.go
+++ b/windows/task.go
@@ -11,6 +11,7 @@ import (
 	eventstypes "github.com/containerd/containerd/api/events"
 	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/events"
+	"github.com/containerd/containerd/mount"
 	"github.com/containerd/containerd/runtime"
 	"github.com/containerd/containerd/windows/hcsshimtypes"
 	"github.com/containerd/typeurl"
@@ -33,6 +34,7 @@ type task struct {
 
 	publisher events.Publisher
 	rwLayer   string
+	rootfs    []mount.Mount
 
 	pidPool           *pidPool
 	hcsContainer      hcsshim.Container
@@ -406,6 +408,5 @@ func (t *task) cleanup() {
 	for _, p := range t.processes {
 		t.removeProcessNL(p.id)
 	}
-	removeLayer(context.Background(), t.rwLayer)
 	t.Unlock()
 }