Update containerd to 66b984ee33

Signed-off-by: Lantao Liu <lantaol@google.com>

vendor/github.com/containerd/console/console_windows.go

@@ -17,6 +17,7 @@
 package console
 
 import (
+	"fmt"
 	"os"
 
 	"github.com/pkg/errors"
@@ -28,55 +29,90 @@ var (
 	ErrNotImplemented = errors.New("not implemented")
 )
 
-func (m *master) init() {
-	m.h = windows.Handle(m.f.Fd())
-	if err := windows.GetConsoleMode(m.h, &m.mode); err == nil {
-		if m.f == os.Stdin {
-			// Validate that windows.ENABLE_VIRTUAL_TERMINAL_INPUT is supported, but do not set it.
-			if err = windows.SetConsoleMode(m.h, m.mode|windows.ENABLE_VIRTUAL_TERMINAL_INPUT); err == nil {
-				vtInputSupported = true
-			}
-			// Unconditionally set the console mode back even on failure because SetConsoleMode
-			// remembers invalid bits on input handles.
-			windows.SetConsoleMode(m.h, m.mode)
-		} else if err := windows.SetConsoleMode(m.h, m.mode|windows.ENABLE_VIRTUAL_TERMINAL_PROCESSING); err == nil {
-			m.mode |= windows.ENABLE_VIRTUAL_TERMINAL_PROCESSING
-		} else {
-			windows.SetConsoleMode(m.h, m.mode)
+func (m *master) initStdios() {
+	m.in = windows.Handle(os.Stdin.Fd())
+	if err := windows.GetConsoleMode(m.in, &m.inMode); err == nil {
+		// Validate that windows.ENABLE_VIRTUAL_TERMINAL_INPUT is supported, but do not set it.
+		if err = windows.SetConsoleMode(m.in, m.inMode|windows.ENABLE_VIRTUAL_TERMINAL_INPUT); err == nil {
+			vtInputSupported = true
 		}
+		// Unconditionally set the console mode back even on failure because SetConsoleMode
+		// remembers invalid bits on input handles.
+		windows.SetConsoleMode(m.in, m.inMode)
+	} else {
+		fmt.Printf("failed to get console mode for stdin: %v\n", err)
+	}
+
+	m.out = windows.Handle(os.Stdout.Fd())
+	if err := windows.GetConsoleMode(m.out, &m.outMode); err == nil {
+		if err := windows.SetConsoleMode(m.out, m.outMode|windows.ENABLE_VIRTUAL_TERMINAL_PROCESSING); err == nil {
+			m.outMode |= windows.ENABLE_VIRTUAL_TERMINAL_PROCESSING
+		} else {
+			windows.SetConsoleMode(m.out, m.outMode)
+		}
+	} else {
+		fmt.Printf("failed to get console mode for stdout: %v\n", err)
+	}
+
+	m.err = windows.Handle(os.Stderr.Fd())
+	if err := windows.GetConsoleMode(m.err, &m.errMode); err == nil {
+		if err := windows.SetConsoleMode(m.err, m.errMode|windows.ENABLE_VIRTUAL_TERMINAL_PROCESSING); err == nil {
+			m.errMode |= windows.ENABLE_VIRTUAL_TERMINAL_PROCESSING
+		} else {
+			windows.SetConsoleMode(m.err, m.errMode)
+		}
+	} else {
+		fmt.Printf("failed to get console mode for stderr: %v\n", err)
 	}
 }
 
 type master struct {
-	h    windows.Handle
-	mode uint32
-	f    *os.File
+	in     windows.Handle
+	inMode uint32
+
+	out     windows.Handle
+	outMode uint32
+
+	err     windows.Handle
+	errMode uint32
 }
 
 func (m *master) SetRaw() error {
-	if m.f == os.Stdin {
-		if err := makeInputRaw(m.h, m.mode); err != nil {
-			return err
-		}
-	} else {
-		// Set StdOut and StdErr to raw mode, we ignore failures since
-		// windows.DISABLE_NEWLINE_AUTO_RETURN might not be supported on this version of
-		// Windows.
-		windows.SetConsoleMode(m.h, m.mode|windows.DISABLE_NEWLINE_AUTO_RETURN)
+	if err := makeInputRaw(m.in, m.inMode); err != nil {
+		return err
 	}
+
+	// Set StdOut and StdErr to raw mode, we ignore failures since
+	// windows.DISABLE_NEWLINE_AUTO_RETURN might not be supported on this version of
+	// Windows.
+
+	windows.SetConsoleMode(m.out, m.outMode|windows.DISABLE_NEWLINE_AUTO_RETURN)
+
+	windows.SetConsoleMode(m.err, m.errMode|windows.DISABLE_NEWLINE_AUTO_RETURN)
+
 	return nil
 }
 
 func (m *master) Reset() error {
-	if err := windows.SetConsoleMode(m.h, m.mode); err != nil {
-		return errors.Wrap(err, "unable to restore console mode")
+	for _, s := range []struct {
+		fd   windows.Handle
+		mode uint32
+	}{
+		{m.in, m.inMode},
+		{m.out, m.outMode},
+		{m.err, m.errMode},
+	} {
+		if err := windows.SetConsoleMode(s.fd, s.mode); err != nil {
+			return errors.Wrap(err, "unable to restore console mode")
+		}
 	}
+
 	return nil
 }
 
 func (m *master) Size() (WinSize, error) {
 	var info windows.ConsoleScreenBufferInfo
-	err := windows.GetConsoleScreenBufferInfo(m.h, &info)
+	err := windows.GetConsoleScreenBufferInfo(m.out, &info)
 	if err != nil {
 		return WinSize{}, errors.Wrap(err, "unable to get console info")
 	}
@@ -98,11 +134,11 @@ func (m *master) ResizeFrom(c Console) error {
 }
 
 func (m *master) DisableEcho() error {
-	mode := m.mode &^ windows.ENABLE_ECHO_INPUT
+	mode := m.inMode &^ windows.ENABLE_ECHO_INPUT
 	mode |= windows.ENABLE_PROCESSED_INPUT
 	mode |= windows.ENABLE_LINE_INPUT
 
-	if err := windows.SetConsoleMode(m.h, mode); err != nil {
+	if err := windows.SetConsoleMode(m.in, mode); err != nil {
 		return errors.Wrap(err, "unable to set console to disable echo")
 	}
 
@@ -114,15 +150,15 @@ func (m *master) Close() error {
 }
 
 func (m *master) Read(b []byte) (int, error) {
-	return m.f.Read(b)
+	return os.Stdin.Read(b)
 }
 
 func (m *master) Write(b []byte) (int, error) {
-	return m.f.Write(b)
+	return os.Stdout.Write(b)
 }
 
 func (m *master) Fd() uintptr {
-	return uintptr(m.h)
+	return uintptr(m.in)
 }
 
 // on windows, console can only be made from os.Std{in,out,err}, hence there
@@ -174,7 +210,7 @@ func newMaster(f *os.File) (Console, error) {
 	if f != os.Stdin && f != os.Stdout && f != os.Stderr {
 		return nil, errors.New("creating a console from a file is not supported on windows")
 	}
-	m := &master{f: f}
-	m.init()
+	m := &master{}
+	m.initStdios()
 	return m, nil
 }

vendor/github.com/containerd/containerd/README.md

@@ -1,7 +1,8 @@
-![banner](/docs/static/img/containerd-dark.png?raw=true)
+![banner](https://github.com/containerd/containerd.io/blob/master/static/img/containerd-dark.png?raw=true)
 
 [![GoDoc](https://godoc.org/github.com/containerd/containerd?status.svg)](https://godoc.org/github.com/containerd/containerd)
 [![Build Status](https://travis-ci.org/containerd/containerd.svg?branch=master)](https://travis-ci.org/containerd/containerd)
+[![Windows Build Status](https://ci.appveyor.com/api/projects/status/github/containerd/containerd?branch=master&svg=true)](https://ci.appveyor.com/project/mlaventure/containerd-3g73f?branch=master)
 [![FOSSA Status](https://app.fossa.io/api/projects/git%2Bhttps%3A%2F%2Fgithub.com%2Fcontainerd%2Fcontainerd.svg?type=shield)](https://app.fossa.io/projects/git%2Bhttps%3A%2F%2Fgithub.com%2Fcontainerd%2Fcontainerd?ref=badge_shield)
 [![Go Report Card](https://goreportcard.com/badge/github.com/containerd/containerd)](https://goreportcard.com/report/github.com/containerd/containerd)
 [![CII Best Practices](https://bestpractices.coreinfrastructure.org/projects/1271/badge)](https://bestpractices.coreinfrastructure.org/projects/1271)

vendor/github.com/containerd/containerd/api/services/events/v1/events.pb.go

@@ -141,7 +141,7 @@ type EventsClient interface {
 	// Forward sends an event that has already been packaged into an envelope
 	// with a timestamp and namespace.
 	//
-	// This is useful if earlier timestamping is required or when fowarding on
+	// This is useful if earlier timestamping is required or when forwarding on
 	// behalf of another component, namespace or publisher.
 	Forward(ctx context.Context, in *ForwardRequest, opts ...grpc.CallOption) (*google_protobuf2.Empty, error)
 	// Subscribe to a stream of events, possibly returning only that match any
@@ -223,7 +223,7 @@ type EventsServer interface {
 	// Forward sends an event that has already been packaged into an envelope
 	// with a timestamp and namespace.
 	//
-	// This is useful if earlier timestamping is required or when fowarding on
+	// This is useful if earlier timestamping is required or when forwarding on
 	// behalf of another component, namespace or publisher.
 	Forward(context.Context, *ForwardRequest) (*google_protobuf2.Empty, error)
 	// Subscribe to a stream of events, possibly returning only that match any

vendor/github.com/containerd/containerd/api/services/events/v1/events.proto

@@ -20,7 +20,7 @@ service Events {
 	// Forward sends an event that has already been packaged into an envelope
 	// with a timestamp and namespace.
 	//
-	// This is useful if earlier timestamping is required or when fowarding on
+	// This is useful if earlier timestamping is required or when forwarding on
 	// behalf of another component, namespace or publisher.
 	rpc Forward(ForwardRequest) returns (google.protobuf.Empty);
 

vendor/github.com/containerd/containerd/archive/tar_windows.go

@@ -259,7 +259,7 @@ func fileInfoFromHeader(hdr *tar.Header) (name string, size int64, fileInfo *win
 		if err != nil {
 			return "", 0, nil, err
 		}
-		fileInfo.FileAttributes = uintptr(attr)
+		fileInfo.FileAttributes = uint32(attr)
 	} else {
 		if hdr.Typeflag == tar.TypeDir {
 			fileInfo.FileAttributes |= syscall.FILE_ATTRIBUTE_DIRECTORY

vendor/github.com/containerd/containerd/cio/io.go

@@ -141,6 +141,15 @@ func NewCreator(opts ...Opt) Creator {
 		if err != nil {
 			return nil, err
 		}
+		if streams.Stdin == nil {
+			fifos.Stdin = ""
+		}
+		if streams.Stdout == nil {
+			fifos.Stdout = ""
+		}
+		if streams.Stderr == nil {
+			fifos.Stderr = ""
+		}
 		return copyIO(fifos, streams)
 	}
 }

vendor/github.com/containerd/containerd/cio/io_windows.go

@@ -74,7 +74,7 @@ func copyIO(fifos *FIFOSet, ioset *Streams) (*cio, error) {
 	if fifos.Stdout != "" {
 		l, err := winio.ListenPipe(fifos.Stdout, nil)
 		if err != nil {
-			return nil, errors.Wrapf(err, "failed to create stdin pipe %s", fifos.Stdout)
+			return nil, errors.Wrapf(err, "failed to create stdout pipe %s", fifos.Stdout)
 		}
 		defer func(l net.Listener) {
 			if err != nil {
@@ -99,7 +99,7 @@ func copyIO(fifos *FIFOSet, ioset *Streams) (*cio, error) {
 		}()
 	}
 
-	if !fifos.Terminal && fifos.Stderr != "" {
+	if fifos.Stderr != "" {
 		l, err := winio.ListenPipe(fifos.Stderr, nil)
 		if err != nil {
 			return nil, errors.Wrapf(err, "failed to create stderr pipe %s", fifos.Stderr)

vendor/github.com/containerd/containerd/client.go

@@ -82,6 +82,9 @@ func New(address string, opts ...ClientOpt) (*Client, error) {
 			return nil, err
 		}
 	}
+	if copts.timeout == 0 {
+		copts.timeout = 10 * time.Second
+	}
 	rt := fmt.Sprintf("%s.%s", plugin.RuntimePlugin, runtime.GOOS)
 	if copts.defaultRuntime != "" {
 		rt = copts.defaultRuntime
@@ -115,7 +118,7 @@ func New(address string, opts ...ClientOpt) (*Client, error) {
 			)
 		}
 		connector := func() (*grpc.ClientConn, error) {
-			ctx, cancel := context.WithTimeout(context.Background(), 60*time.Second)
+			ctx, cancel := context.WithTimeout(context.Background(), copts.timeout)
 			defer cancel()
 			conn, err := grpc.DialContext(ctx, dialer.DialAddress(address), gopts...)
 			if err != nil {
@@ -256,9 +259,10 @@ type RemoteContext struct {
 	// If no resolver is provided, defaults to Docker registry resolver.
 	Resolver remotes.Resolver
 
-	// Platforms defines which platforms to handle when doing the image operation.
-	// If this field is empty, content for all platforms will be pulled.
-	Platforms []string
+	// PlatformMatcher is used to match the platforms for an image
+	// operation and define the preference when a single match is required
+	// from multiple platforms.
+	PlatformMatcher platforms.MatchComparer
 
 	// Unpack is done after an image is pulled to extract into a snapshotter.
 	// If an image is not unpacked on pull, it can be unpacked any time
@@ -280,6 +284,12 @@ type RemoteContext struct {
 	// manifests. If this option is false then any image which resolves
 	// to schema 1 will return an error since schema 1 is not supported.
 	ConvertSchema1 bool
+
+	// Platforms defines which platforms to handle when doing the image operation.
+	// Platforms is ignored when a PlatformMatcher is set, otherwise the
+	// platforms will be used to create a PlatformMatcher with no ordering
+	// preference.
+	Platforms []string
 }
 
 func defaultRemoteContext() *RemoteContext {
@@ -305,13 +315,30 @@ func (c *Client) Fetch(ctx context.Context, ref string, opts ...RemoteOpt) (imag
 		return images.Image{}, errors.New("unpack on fetch not supported, try pull")
 	}
 
+	if fetchCtx.PlatformMatcher == nil {
+		if len(fetchCtx.Platforms) == 0 {
+			fetchCtx.PlatformMatcher = platforms.All
+		} else {
+			var ps []ocispec.Platform
+			for _, s := range fetchCtx.Platforms {
+				p, err := platforms.Parse(s)
+				if err != nil {
+					return images.Image{}, errors.Wrapf(err, "invalid platform %s", s)
+				}
+				ps = append(ps, p)
+			}
+
+			fetchCtx.PlatformMatcher = platforms.Any(ps...)
+		}
+	}
+
 	ctx, done, err := c.WithLease(ctx)
 	if err != nil {
 		return images.Image{}, err
 	}
 	defer done(ctx)
 
-	return c.fetch(ctx, fetchCtx, ref)
+	return c.fetch(ctx, fetchCtx, ref, 0)
 }
 
 // Pull downloads the provided content into containerd's content store
@@ -324,10 +351,19 @@ func (c *Client) Pull(ctx context.Context, ref string, opts ...RemoteOpt) (Image
 		}
 	}
 
-	if len(pullCtx.Platforms) > 1 {
-		return nil, errors.New("cannot pull multiplatform image locally, try Fetch")
-	} else if len(pullCtx.Platforms) == 0 {
-		pullCtx.Platforms = []string{platforms.Default()}
+	if pullCtx.PlatformMatcher == nil {
+		if len(pullCtx.Platforms) > 1 {
+			return nil, errors.New("cannot pull multiplatform image locally, try Fetch")
+		} else if len(pullCtx.Platforms) == 0 {
+			pullCtx.PlatformMatcher = platforms.Default()
+		} else {
+			p, err := platforms.Parse(pullCtx.Platforms[0])
+			if err != nil {
+				return nil, errors.Wrapf(err, "invalid platform %s", pullCtx.Platforms[0])
+			}
+
+			pullCtx.PlatformMatcher = platforms.Only(p)
+		}
 	}
 
 	ctx, done, err := c.WithLease(ctx)
@@ -336,12 +372,12 @@ func (c *Client) Pull(ctx context.Context, ref string, opts ...RemoteOpt) (Image
 	}
 	defer done(ctx)
 
-	img, err := c.fetch(ctx, pullCtx, ref)
+	img, err := c.fetch(ctx, pullCtx, ref, 1)
 	if err != nil {
 		return nil, err
 	}
 
-	i := NewImageWithPlatform(c, img, pullCtx.Platforms[0])
+	i := NewImageWithPlatform(c, img, pullCtx.PlatformMatcher)
 
 	if pullCtx.Unpack {
 		if err := i.Unpack(ctx, pullCtx.Snapshotter); err != nil {
@@ -352,7 +388,7 @@ func (c *Client) Pull(ctx context.Context, ref string, opts ...RemoteOpt) (Image
 	return i, nil
 }
 
-func (c *Client) fetch(ctx context.Context, rCtx *RemoteContext, ref string) (images.Image, error) {
+func (c *Client) fetch(ctx context.Context, rCtx *RemoteContext, ref string, limit int) (images.Image, error) {
 	store := c.ContentStore()
 	name, desc, err := rCtx.Resolver.Resolve(ctx, ref)
 	if err != nil {
@@ -377,7 +413,11 @@ func (c *Client) fetch(ctx context.Context, rCtx *RemoteContext, ref string) (im
 		// Set any children labels for that content
 		childrenHandler = images.SetChildrenLabels(store, childrenHandler)
 		// Filter children by platforms
-		childrenHandler = images.FilterPlatforms(childrenHandler, rCtx.Platforms...)
+		childrenHandler = images.FilterPlatforms(childrenHandler, rCtx.PlatformMatcher)
+		// Sort and limit manifests if a finite number is needed
+		if limit > 0 {
+			childrenHandler = images.LimitManifests(childrenHandler, rCtx.PlatformMatcher, limit)
+		}
 
 		handler = images.Handlers(append(rCtx.BaseHandlers,
 			remotes.FetchHandler(store, fetcher),
@@ -434,13 +474,28 @@ func (c *Client) Push(ctx context.Context, ref string, desc ocispec.Descriptor,
 			return err
 		}
 	}
+	if pushCtx.PlatformMatcher == nil {
+		if len(pushCtx.Platforms) > 0 {
+			var ps []ocispec.Platform
+			for _, platform := range pushCtx.Platforms {
+				p, err := platforms.Parse(platform)
+				if err != nil {
+					return errors.Wrapf(err, "invalid platform %s", platform)
+				}
+				ps = append(ps, p)
+			}
+			pushCtx.PlatformMatcher = platforms.Any(ps...)
+		} else {
+			pushCtx.PlatformMatcher = platforms.All
+		}
+	}
 
 	pusher, err := pushCtx.Resolver.Pusher(ctx, ref)
 	if err != nil {
 		return err
 	}
 
-	return remotes.PushContent(ctx, pusher, desc, c.ContentStore(), pushCtx.Platforms, pushCtx.BaseHandlers...)
+	return remotes.PushContent(ctx, pusher, desc, c.ContentStore(), pushCtx.PlatformMatcher, pushCtx.BaseHandlers...)
 }
 
 // GetImage returns an existing image

vendor/github.com/containerd/containerd/client_opts.go

@@ -17,6 +17,8 @@
 package containerd
 
 import (
+	"time"
+
 	"github.com/containerd/containerd/images"
 	"github.com/containerd/containerd/platforms"
 	"github.com/containerd/containerd/remotes"
@@ -28,6 +30,7 @@ type clientOpts struct {
 	defaultRuntime string
 	services       *services
 	dialOptions    []grpc.DialOption
+	timeout        time.Duration
 }
 
 // ClientOpt allows callers to set options on the containerd client
@@ -71,6 +74,14 @@ func WithServices(opts ...ServicesOpt) ClientOpt {
 	}
 }
 
+// WithTimeout sets the connection timeout for the client
+func WithTimeout(d time.Duration) ClientOpt {
+	return func(c *clientOpts) error {
+		c.timeout = d
+		return nil
+	}
+}
+
 // RemoteOpt allows the caller to set distribution options for a remote
 type RemoteOpt func(*Client, *RemoteContext) error
 
@@ -78,7 +89,7 @@ type RemoteOpt func(*Client, *RemoteContext) error
 // content for
 func WithPlatform(platform string) RemoteOpt {
 	if platform == "" {
-		platform = platforms.Default()
+		platform = platforms.DefaultString()
 	}
 	return func(_ *Client, c *RemoteContext) error {
 		for _, p := range c.Platforms {
@@ -92,6 +103,16 @@ func WithPlatform(platform string) RemoteOpt {
 	}
 }
 
+// WithPlatformMatcher specifies the matcher to use for
+// determining which platforms to pull content for.
+// This value supersedes anything set with `WithPlatform`.
+func WithPlatformMatcher(m platforms.MatchComparer) RemoteOpt {
+	return func(_ *Client, c *RemoteContext) error {
+		c.PlatformMatcher = m
+		return nil
+	}
+}
+
 // WithPullUnpack is used to unpack an image after pull. This
 // uses the snapshotter, content store, and diff service
 // configured for the client.

vendor/github.com/containerd/containerd/cmd/ctr/commands/containers/containers.go

@@ -53,11 +53,22 @@ var createCommand = cli.Command{
 	Flags:     append(commands.SnapshotterFlags, commands.ContainerFlags...),
 	Action: func(context *cli.Context) error {
 		var (
-			id  = context.Args().Get(1)
-			ref = context.Args().First()
+			id     string
+			ref    string
+			config = context.IsSet("config")
 		)
-		if ref == "" {
-			return errors.New("image ref must be provided")
+
+		if config {
+			id = context.Args().First()
+			if context.NArg() > 1 {
+				return errors.New("with spec config file, only container id should be provided")
+			}
+		} else {
+			id = context.Args().Get(1)
+			ref = context.Args().First()
+			if ref == "" {
+				return errors.New("image ref must be provided")
+			}
 		}
 		if id == "" {
 			return errors.New("container id must be provided")

vendor/github.com/containerd/containerd/cmd/ctr/commands/content/fetch.go

@@ -76,28 +76,61 @@ Most of this is experimental and there are few leaps to make this work.`,
 			return err
 		}
 		defer cancel()
-
-		_, err = Fetch(ctx, client, ref, clicontext)
+		config, err := NewFetchConfig(ctx, clicontext)
+		if err != nil {
+			return err
+		}
+		_, err = Fetch(ctx, client, ref, config)
 		return err
 	},
 }
 
-// Fetch loads all resources into the content store and returns the image
-func Fetch(ctx context.Context, client *containerd.Client, ref string, cliContext *cli.Context) (images.Image, error) {
-	resolver, err := commands.GetResolver(ctx, cliContext)
-	if err != nil {
-		return images.Image{}, err
-	}
+// FetchConfig for content fetch
+type FetchConfig struct {
+	// Resolver
+	Resolver remotes.Resolver
+	// ProgressOutput to display progress
+	ProgressOutput io.Writer
+	// Labels to set on the content
+	Labels []string
+	// Platforms to fetch
+	Platforms []string
+}
 
+// NewFetchConfig returns the default FetchConfig from cli flags
+func NewFetchConfig(ctx context.Context, clicontext *cli.Context) (*FetchConfig, error) {
+	resolver, err := commands.GetResolver(ctx, clicontext)
+	if err != nil {
+		return nil, err
+	}
+	config := &FetchConfig{
+		Resolver: resolver,
+		Labels:   clicontext.StringSlice("label"),
+	}
+	if !clicontext.GlobalBool("debug") {
+		config.ProgressOutput = os.Stdout
+	}
+	if !clicontext.Bool("all-platforms") {
+		p := clicontext.StringSlice("platform")
+		if len(p) == 0 {
+			p = append(p, platforms.DefaultString())
+		}
+		config.Platforms = p
+	}
+	return config, nil
+}
+
+// Fetch loads all resources into the content store and returns the image
+func Fetch(ctx context.Context, client *containerd.Client, ref string, config *FetchConfig) (images.Image, error) {
 	ongoing := newJobs(ref)
 
 	pctx, stopProgress := context.WithCancel(ctx)
 	progress := make(chan struct{})
 
 	go func() {
-		if !cliContext.GlobalBool("debug") {
+		if config.ProgressOutput != nil {
 			// no progress bar, because it hides some debug logs
-			showProgress(pctx, ongoing, client.ContentStore(), os.Stdout)
+			showProgress(pctx, ongoing, client.ContentStore(), config.ProgressOutput)
 		}
 		close(progress)
 	}()
@@ -110,24 +143,16 @@ func Fetch(ctx context.Context, client *containerd.Client, ref string, cliContex
 	})
 
 	log.G(pctx).WithField("image", ref).Debug("fetching")
-	labels := commands.LabelArgs(cliContext.StringSlice("label"))
+	labels := commands.LabelArgs(config.Labels)
 	opts := []containerd.RemoteOpt{
 		containerd.WithPullLabels(labels),
-		containerd.WithResolver(resolver),
+		containerd.WithResolver(config.Resolver),
 		containerd.WithImageHandler(h),
 		containerd.WithSchema1Conversion,
 	}
-
-	if !cliContext.Bool("all-platforms") {
-		p := cliContext.StringSlice("platform")
-		if len(p) == 0 {
-			p = append(p, platforms.Default())
-		}
-		for _, platform := range p {
-			opts = append(opts, containerd.WithPlatform(platform))
-		}
+	for _, platform := range config.Platforms {
+		opts = append(opts, containerd.WithPlatform(platform))
 	}
-
 	img, err := client.Fetch(pctx, ref, opts...)
 	stopProgress()
 	if err != nil {

vendor/github.com/containerd/containerd/cmd/ctr/commands/images/pull.go

@@ -25,6 +25,7 @@ import (
 	"github.com/containerd/containerd/images"
 	"github.com/containerd/containerd/log"
 	"github.com/containerd/containerd/platforms"
+	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 	"github.com/pkg/errors"
 	"github.com/urfave/cli"
 )
@@ -73,7 +74,11 @@ command. As part of this process, we do the following:
 		}
 		defer done(ctx)
 
-		img, err := content.Fetch(ctx, client, ref, context)
+		config, err := content.NewFetchConfig(ctx, context)
+		if err != nil {
+			return err
+		}
+		img, err := content.Fetch(ctx, client, ref, config)
 		if err != nil {
 			return err
 		}
@@ -82,26 +87,28 @@ command. As part of this process, we do the following:
 
 		// TODO: Show unpack status
 
-		var p []string
+		var p []ocispec.Platform
 		if context.Bool("all-platforms") {
-			all, err := images.Platforms(ctx, client.ContentStore(), img.Target)
+			p, err = images.Platforms(ctx, client.ContentStore(), img.Target)
 			if err != nil {
 				return errors.Wrap(err, "unable to resolve image platforms")
 			}
-			p = make([]string, len(all))
-			for i := range all {
-				p[i] = platforms.Format(all[i])
-			}
 		} else {
-			p = context.StringSlice("platform")
+			for _, s := range context.StringSlice("platform") {
+				ps, err := platforms.Parse(s)
+				if err != nil {
+					return errors.Wrapf(err, "unable to parse platform %s", s)
+				}
+				p = append(p, ps)
+			}
 		}
 		if len(p) == 0 {
-			p = append(p, platforms.Default())
+			p = append(p, platforms.DefaultSpec())
 		}
 
 		for _, platform := range p {
-			fmt.Printf("unpacking %s %s...\n", platform, img.Target.Digest)
-			i := containerd.NewImageWithPlatform(client, img, platform)
+			fmt.Printf("unpacking %s %s...\n", platforms.Format(platform), img.Target.Digest)
+			i := containerd.NewImageWithPlatform(client, img, platforms.Only(platform))
 			err = i.Unpack(ctx, context.String("snapshotter"))
 			if err != nil {
 				return err

vendor/github.com/containerd/containerd/cmd/ctr/commands/install/install.go

@@ -37,6 +37,10 @@ var Command = cli.Command{
 			Name:  "replace,r",
 			Usage: "replace any binaries or libs in the opt directory",
 		},
+		cli.StringFlag{
+			Name:  "path",
+			Usage: "set an optional install path other than the managed opt directory",
+		},
 	},
 	Action: func(context *cli.Context) error {
 		client, ctx, cancel, err := commands.NewClient(context)
@@ -56,6 +60,9 @@ var Command = cli.Command{
 		if context.Bool("replace") {
 			opts = append(opts, containerd.WithInstallReplace)
 		}
+		if path := context.String("path"); path != "" {
+			opts = append(opts, containerd.WithInstallPath(path))
+		}
 		return client.Install(ctx, image, opts...)
 	},
 }

vendor/github.com/containerd/containerd/cmd/ctr/commands/run/run.go

@@ -110,15 +110,26 @@ var Command = cli.Command{
 	Action: func(context *cli.Context) error {
 		var (
 			err error
+			id  string
+			ref string
 
-			id     = context.Args().Get(1)
-			ref    = context.Args().First()
 			tty    = context.Bool("tty")
 			detach = context.Bool("detach")
+			config = context.IsSet("config")
 		)
 
-		if ref == "" {
-			return errors.New("image ref must be provided")
+		if config {
+			id = context.Args().First()
+			if context.NArg() > 1 {
+				return errors.New("with spec config file, only container id should be provided")
+			}
+		} else {
+			id = context.Args().Get(1)
+			ref = context.Args().First()
+
+			if ref == "" {
+				return errors.New("image ref must be provided")
+			}
 		}
 		if id == "" {
 			return errors.New("container id must be provided")
@@ -135,9 +146,17 @@ var Command = cli.Command{
 		if context.Bool("rm") && !detach {
 			defer container.Delete(ctx, containerd.WithSnapshotCleanup)
 		}
+		var con console.Console
+		if tty {
+			con = console.Current()
+			defer con.Reset()
+			if err := con.SetRaw(); err != nil {
+				return err
+			}
+		}
 		opts := getNewTaskOpts(context)
 		ioOpts := []cio.Opt{cio.WithFIFODir(context.String("fifo-dir"))}
-		task, err := tasks.NewTask(ctx, client, container, context.String("checkpoint"), tty, context.Bool("null-io"), ioOpts, opts...)
+		task, err := tasks.NewTask(ctx, client, container, context.String("checkpoint"), con, context.Bool("null-io"), ioOpts, opts...)
 		if err != nil {
 			return err
 		}
@@ -153,14 +172,6 @@ var Command = cli.Command{
 				return err
 			}
 		}
-		var con console.Console
-		if tty {
-			con = console.Current()
-			defer con.Reset()
-			if err := con.SetRaw(); err != nil {
-				return err
-			}
-		}
 		if err := task.Start(ctx); err != nil {
 			return err
 		}

vendor/github.com/containerd/containerd/cmd/ctr/commands/run/run_unix.go

@@ -20,6 +20,7 @@ package run
 
 import (
 	gocontext "context"
+	"path/filepath"
 	"strings"
 
 	"github.com/containerd/containerd"
@@ -34,10 +35,14 @@ import (
 // NewContainer creates a new container
 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:]
+		id     string
+		config = context.IsSet("config")
 	)
+	if config {
+		id = context.Args().First()
+	} else {
+		id = context.Args().Get(1)
+	}
 
 	if raw := context.String("checkpoint"); raw != "" {
 		im, err := client.GetImage(ctx, raw)
@@ -53,78 +58,89 @@ func NewContainer(ctx gocontext.Context, client *containerd.Client, context *cli
 		spec  containerd.NewContainerOpts
 	)
 
-	if context.IsSet("config") {
+	if config {
 		opts = append(opts, oci.WithSpecFromFile(context.String("config")))
 	} else {
-		opts = append(opts, oci.WithDefaultSpec())
-	}
+		var (
+			ref = context.Args().First()
+			//for container's id is Args[1]
+			args = context.Args()[2:]
+		)
+		opts = append(opts, oci.WithDefaultSpec(), oci.WithDefaultUnixDevices)
+		opts = append(opts, oci.WithEnv(context.StringSlice("env")))
+		opts = append(opts, withMounts(context))
 
-	opts = append(opts, oci.WithEnv(context.StringSlice("env")))
-	opts = append(opts, withMounts(context))
-	cOpts = append(cOpts, containerd.WithContainerLabels(commands.LabelArgs(context.StringSlice("label"))))
-	cOpts = append(cOpts, containerd.WithRuntime(context.String("runtime"), nil))
-	if context.Bool("rootfs") {
-		opts = append(opts, oci.WithRootFSPath(ref))
-	} else {
-		snapshotter := context.String("snapshotter")
-		image, err := client.GetImage(ctx, ref)
-		if err != nil {
-			return nil, err
-		}
-		unpacked, err := image.IsUnpacked(ctx, snapshotter)
-		if err != nil {
-			return nil, err
-		}
-		if !unpacked {
-			if err := image.Unpack(ctx, snapshotter); err != nil {
+		if context.Bool("rootfs") {
+			rootfs, err := filepath.Abs(ref)
+			if err != nil {
 				return nil, err
 			}
+			opts = append(opts, oci.WithRootFSPath(rootfs))
+		} else {
+			snapshotter := context.String("snapshotter")
+			image, err := client.GetImage(ctx, ref)
+			if err != nil {
+				return nil, err
+			}
+			unpacked, err := image.IsUnpacked(ctx, snapshotter)
+			if err != nil {
+				return nil, err
+			}
+			if !unpacked {
+				if err := image.Unpack(ctx, snapshotter); err != nil {
+					return nil, err
+				}
+			}
+			opts = append(opts, oci.WithImageConfig(image))
+			cOpts = append(cOpts,
+				containerd.WithImage(image),
+				containerd.WithSnapshotter(snapshotter),
+				// Even when "readonly" is set, we don't use KindView snapshot here. (#1495)
+				// We pass writable snapshot to the OCI runtime, and the runtime remounts it as read-only,
+				// after creating some mount points on demand.
+				containerd.WithNewSnapshot(id, image))
 		}
-		opts = append(opts, oci.WithImageConfig(image))
-		cOpts = append(cOpts,
-			containerd.WithImage(image),
-			containerd.WithSnapshotter(snapshotter),
-			// Even when "readonly" is set, we don't use KindView snapshot here. (#1495)
-			// We pass writable snapshot to the OCI runtime, and the runtime remounts it as read-only,
-			// after creating some mount points on demand.
-			containerd.WithNewSnapshot(id, image))
-	}
-	if context.Bool("readonly") {
-		opts = append(opts, oci.WithRootFSReadonly())
-	}
-	if len(args) > 0 {
-		opts = append(opts, oci.WithProcessArgs(args...))
-	}
-	if cwd := context.String("cwd"); cwd != "" {
-		opts = append(opts, oci.WithProcessCwd(cwd))
-	}
-	if context.Bool("tty") {
-		opts = append(opts, oci.WithTTY)
-	}
-	if context.Bool("privileged") {
-		opts = append(opts, oci.WithPrivileged)
-	}
-	if context.Bool("net-host") {
-		opts = append(opts, oci.WithHostNamespace(specs.NetworkNamespace), oci.WithHostHostsFile, oci.WithHostResolvconf)
-	}
-	joinNs := context.StringSlice("with-ns")
-	for _, ns := range joinNs {
-		parts := strings.Split(ns, ":")
-		if len(parts) != 2 {
-			return nil, errors.New("joining a Linux namespace using --with-ns requires the format 'nstype:path'")
+		if context.Bool("readonly") {
+			opts = append(opts, oci.WithRootFSReadonly())
 		}
-		if !validNamespace(parts[0]) {
-			return nil, errors.New("the Linux namespace type specified in --with-ns is not valid: " + parts[0])
+		if len(args) > 0 {
+			opts = append(opts, oci.WithProcessArgs(args...))
+		}
+		if cwd := context.String("cwd"); cwd != "" {
+			opts = append(opts, oci.WithProcessCwd(cwd))
+		}
+		if context.Bool("tty") {
+			opts = append(opts, oci.WithTTY)
+		}
+		if context.Bool("privileged") {
+			opts = append(opts, oci.WithPrivileged)
+		}
+		if context.Bool("net-host") {
+			opts = append(opts, oci.WithHostNamespace(specs.NetworkNamespace), oci.WithHostHostsFile, oci.WithHostResolvconf)
+		}
+
+		joinNs := context.StringSlice("with-ns")
+		for _, ns := range joinNs {
+			parts := strings.Split(ns, ":")
+			if len(parts) != 2 {
+				return nil, errors.New("joining a Linux namespace using --with-ns requires the format 'nstype:path'")
+			}
+			if !validNamespace(parts[0]) {
+				return nil, errors.New("the Linux namespace type specified in --with-ns is not valid: " + parts[0])
+			}
+			opts = append(opts, oci.WithLinuxNamespace(specs.LinuxNamespace{
+				Type: specs.LinuxNamespaceType(parts[0]),
+				Path: parts[1],
+			}))
+		}
+		if context.IsSet("gpus") {
+			opts = append(opts, nvidia.WithGPUs(nvidia.WithDevices(context.Int("gpus")), nvidia.WithAllCapabilities))
 		}
-		opts = append(opts, oci.WithLinuxNamespace(specs.LinuxNamespace{
-			Type: specs.LinuxNamespaceType(parts[0]),
-			Path: parts[1],
-		}))
-	}
-	if context.IsSet("gpus") {
-		opts = append(opts, nvidia.WithGPUs(nvidia.WithDevices(context.Int("gpus")), nvidia.WithAllCapabilities))
 	}
 
+	cOpts = append(cOpts, containerd.WithContainerLabels(commands.LabelArgs(context.StringSlice("label"))))
+	cOpts = append(cOpts, containerd.WithRuntime(context.String("runtime"), nil))
+
 	var s specs.Spec
 	spec = containerd.WithSpec(&s, opts...)
 

vendor/github.com/containerd/containerd/cmd/ctr/commands/run/run_windows.go

@@ -22,29 +22,12 @@ import (
 	"github.com/containerd/console"
 	"github.com/containerd/containerd"
 	"github.com/containerd/containerd/cmd/ctr/commands"
-	"github.com/containerd/containerd/containers"
 	"github.com/containerd/containerd/oci"
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/sirupsen/logrus"
 	"github.com/urfave/cli"
 )
 
-func withTTY(terminal bool) oci.SpecOpts {
-	if !terminal {
-		return func(ctx gocontext.Context, client oci.Client, c *containers.Container, s *specs.Spec) error {
-			s.Process.Terminal = false
-			return nil
-		}
-	}
-
-	con := console.Current()
-	size, err := con.Size()
-	if err != nil {
-		logrus.WithError(err).Error("console size")
-	}
-	return oci.WithTTY(int(size.Width), int(size.Height))
-}
-
 // NewContainer creates a new container
 func NewContainer(ctx gocontext.Context, client *containerd.Client, context *cli.Context) (containerd.Container, error) {
 	var (
@@ -73,7 +56,17 @@ func NewContainer(ctx gocontext.Context, client *containerd.Client, context *cli
 	opts = append(opts, oci.WithImageConfig(image))
 	opts = append(opts, oci.WithEnv(context.StringSlice("env")))
 	opts = append(opts, withMounts(context))
-	opts = append(opts, withTTY(context.Bool("tty")))
+	if context.Bool("tty") {
+		opts = append(opts, oci.WithTTY)
+
+		con := console.Current()
+		size, err := con.Size()
+		if err != nil {
+			logrus.WithError(err).Error("console size")
+		}
+		opts = append(opts, oci.WithTTYSize(int(size.Width), int(size.Height)))
+	}
+
 	if len(args) > 0 {
 		opts = append(opts, oci.WithProcessArgs(args...))
 	}

vendor/github.com/containerd/containerd/cmd/ctr/commands/tasks/start.go

@@ -18,6 +18,7 @@ package tasks
 
 import (
 	"github.com/containerd/console"
+	"github.com/containerd/containerd"
 	"github.com/containerd/containerd/cio"
 	"github.com/containerd/containerd/cmd/ctr/commands"
 	"github.com/pkg/errors"
@@ -42,11 +43,16 @@ var startCommand = cli.Command{
 			Name:  "pid-file",
 			Usage: "file path to write the task's pid",
 		},
+		cli.BoolFlag{
+			Name:  "detach,d",
+			Usage: "detach from the task after it has started execution",
+		},
 	},
 	Action: func(context *cli.Context) error {
 		var (
-			err error
-			id  = context.Args().Get(0)
+			err    error
+			id     = context.Args().Get(0)
+			detach = context.Bool("detach")
 		)
 		if id == "" {
 			return errors.New("container id must be provided")
@@ -65,27 +71,11 @@ var startCommand = cli.Command{
 		if err != nil {
 			return err
 		}
-
 		var (
 			tty    = spec.Process.Terminal
 			opts   = getNewTaskOpts(context)
 			ioOpts = []cio.Opt{cio.WithFIFODir(context.String("fifo-dir"))}
 		)
-		task, err := NewTask(ctx, client, container, "", tty, context.Bool("null-io"), ioOpts, opts...)
-		if err != nil {
-			return err
-		}
-		defer task.Delete(ctx)
-		if context.IsSet("pid-file") {
-			if err := commands.WritePidFile(context.String("pid-file"), int(task.Pid())); err != nil {
-				return err
-			}
-		}
-		statusC, err := task.Wait(ctx)
-		if err != nil {
-			return err
-		}
-
 		var con console.Console
 		if tty {
 			con = console.Current()
@@ -94,9 +84,30 @@ var startCommand = cli.Command{
 				return err
 			}
 		}
+
+		task, err := NewTask(ctx, client, container, "", con, context.Bool("null-io"), ioOpts, opts...)
+		if err != nil {
+			return err
+		}
+		var statusC <-chan containerd.ExitStatus
+		if !detach {
+			defer task.Delete(ctx)
+			if statusC, err = task.Wait(ctx); err != nil {
+				return err
+			}
+		}
+		if context.IsSet("pid-file") {
+			if err := commands.WritePidFile(context.String("pid-file"), int(task.Pid())); err != nil {
+				return err
+			}
+		}
+
 		if err := task.Start(ctx); err != nil {
 			return err
 		}
+		if detach {
+			return nil
+		}
 		if tty {
 			if err := HandleConsoleResize(ctx, task, con); err != nil {
 				logrus.WithError(err).Error("console resize")

vendor/github.com/containerd/containerd/cmd/ctr/commands/tasks/tasks_unix.go

@@ -67,7 +67,7 @@ func HandleConsoleResize(ctx gocontext.Context, task resizer, con console.Consol
 }
 
 // NewTask creates a new task
-func NewTask(ctx gocontext.Context, client *containerd.Client, container containerd.Container, checkpoint string, tty, nullIO bool, ioOpts []cio.Opt, opts ...containerd.NewTaskOpts) (containerd.Task, error) {
+func NewTask(ctx gocontext.Context, client *containerd.Client, container containerd.Container, checkpoint string, con console.Console, nullIO bool, ioOpts []cio.Opt, opts ...containerd.NewTaskOpts) (containerd.Task, error) {
 	stdio := cio.NewCreator(append([]cio.Opt{cio.WithStdio}, ioOpts...)...)
 	if checkpoint != "" {
 		im, err := client.GetImage(ctx, checkpoint)
@@ -77,11 +77,11 @@ func NewTask(ctx gocontext.Context, client *containerd.Client, container contain
 		opts = append(opts, containerd.WithTaskCheckpoint(im))
 	}
 	ioCreator := stdio
-	if tty {
-		ioCreator = cio.NewCreator(append([]cio.Opt{cio.WithStdio, cio.WithTerminal}, ioOpts...)...)
+	if con != nil {
+		ioCreator = cio.NewCreator(append([]cio.Opt{cio.WithStreams(con, con, nil), cio.WithTerminal}, ioOpts...)...)
 	}
 	if nullIO {
-		if tty {
+		if con != nil {
 			return nil, errors.New("tty and null-io cannot be used together")
 		}
 		ioCreator = cio.NullIO

vendor/github.com/containerd/containerd/cmd/ctr/commands/tasks/tasks_windows.go

@@ -58,13 +58,13 @@ func HandleConsoleResize(ctx gocontext.Context, task resizer, con console.Consol
 }
 
 // NewTask creates a new task
-func NewTask(ctx gocontext.Context, client *containerd.Client, container containerd.Container, _ string, tty, nullIO bool, ioOpts []cio.Opt, opts ...containerd.NewTaskOpts) (containerd.Task, error) {
+func NewTask(ctx gocontext.Context, client *containerd.Client, container containerd.Container, _ string, con console.Console, nullIO bool, ioOpts []cio.Opt, opts ...containerd.NewTaskOpts) (containerd.Task, error) {
 	var ioCreator cio.Creator
-	if tty {
+	if con != nil {
 		if nullIO {
 			return nil, errors.New("tty and null-io cannot be used together")
 		}
-		ioCreator = cio.NewCreator(append([]cio.Opt{cio.WithStdio, cio.WithTerminal}, ioOpts...)...)
+		ioCreator = cio.NewCreator(append([]cio.Opt{cio.WithStreams(con, con, con), cio.WithTerminal}, ioOpts...)...)
 	} else if nullIO {
 		ioCreator = cio.NullIO
 	} else {

vendor/github.com/containerd/containerd/container_opts_unix.go

@@ -20,25 +20,21 @@ package containerd
 
 import (
 	"context"
-	"encoding/json"
 	"fmt"
 	"os"
 	"path/filepath"
 	"syscall"
 
-	"github.com/containerd/containerd/api/types"
 	"github.com/containerd/containerd/containers"
 	"github.com/containerd/containerd/content"
 	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/images"
 	"github.com/containerd/containerd/mount"
 	"github.com/containerd/containerd/platforms"
-	"github.com/containerd/containerd/runtime/linux/runctypes"
 	"github.com/gogo/protobuf/proto"
 	protobuf "github.com/gogo/protobuf/types"
 	"github.com/opencontainers/image-spec/identity"
 	"github.com/opencontainers/image-spec/specs-go/v1"
-	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 	"github.com/pkg/errors"
 )
 
@@ -105,44 +101,6 @@ func WithCheckpoint(im Image, snapshotKey string) NewContainerOpts {
 	}
 }
 
-// WithTaskCheckpoint allows a task to be created with live runtime and memory data from a
-// previous checkpoint. Additional software such as CRIU may be required to
-// restore a task from a checkpoint
-func WithTaskCheckpoint(im Image) NewTaskOpts {
-	return func(ctx context.Context, c *Client, info *TaskInfo) error {
-		desc := im.Target()
-		id := desc.Digest
-		index, err := decodeIndex(ctx, c.ContentStore(), desc)
-		if err != nil {
-			return err
-		}
-		for _, m := range index.Manifests {
-			if m.MediaType == images.MediaTypeContainerd1Checkpoint {
-				info.Checkpoint = &types.Descriptor{
-					MediaType: m.MediaType,
-					Size_:     m.Size,
-					Digest:    m.Digest,
-				}
-				return nil
-			}
-		}
-		return fmt.Errorf("checkpoint not found in index %s", id)
-	}
-}
-
-func decodeIndex(ctx context.Context, store content.Provider, desc ocispec.Descriptor) (*v1.Index, error) {
-	var index v1.Index
-	p, err := content.ReadBlob(ctx, store, desc)
-	if err != nil {
-		return nil, err
-	}
-	if err := json.Unmarshal(p, &index); err != nil {
-		return nil, err
-	}
-
-	return &index, nil
-}
-
 // WithRemappedSnapshot creates a new snapshot and remaps the uid/gid for the
 // filesystem to be used by a container with user namespaces
 func WithRemappedSnapshot(id string, i Image, uid, gid uint32) NewContainerOpts {
@@ -221,19 +179,3 @@ func incrementFS(root string, uidInc, gidInc uint32) filepath.WalkFunc {
 		return os.Lchown(path, u, g)
 	}
 }
-
-// WithNoPivotRoot instructs the runtime not to you pivot_root
-func WithNoPivotRoot(_ context.Context, _ *Client, info *TaskInfo) error {
-	if info.Options == nil {
-		info.Options = &runctypes.CreateOptions{
-			NoPivotRoot: true,
-		}
-		return nil
-	}
-	copts, ok := info.Options.(*runctypes.CreateOptions)
-	if !ok {
-		return errors.New("invalid options type, expected runctypes.CreateOptions")
-	}
-	copts.NoPivotRoot = true
-	return nil
-}

vendor/github.com/containerd/containerd/containers/containers.go

@@ -40,7 +40,7 @@ type Container struct {
 
 	// Image specifies the image reference used for a container.
 	//
-	// This property is optional but immutable.
+	// This property is optional and mutable.
 	Image string
 
 	// Runtime specifies which runtime should be used when launching container
@@ -60,7 +60,7 @@ type Container struct {
 	// look up the mounts from the snapshot service and include those on the
 	// task create request.
 	//
-	// This field is not required but immutable.
+	// This field is not required but mutable.
 	SnapshotKey string
 
 	// Snapshotter specifies the snapshotter name used for rootfs

vendor/github.com/containerd/containerd/content/local/store.go

@@ -33,6 +33,8 @@ import (
 	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/filters"
 	"github.com/containerd/containerd/log"
+
+	"github.com/containerd/continuity"
 	digest "github.com/opencontainers/go-digest"
 	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 	"github.com/pkg/errors"
@@ -651,5 +653,5 @@ func writeTimestampFile(p string, t time.Time) error {
 		return err
 	}
 
-	return ioutil.WriteFile(p, b, 0666)
+	return continuity.AtomicWriteFile(p, b, 0666)
 }

vendor/github.com/containerd/containerd/contrib/nvidia/nvidia.go

@@ -32,7 +32,7 @@ import (
 const nvidiaCLI = "nvidia-container-cli"
 
 // Capability specifies capabilities for the gpu inside the container
-// Detailed explaination of options can be found:
+// Detailed explanation of options can be found:
 // https://github.com/nvidia/nvidia-container-runtime#supported-driver-capabilities
 type Capability string
 

vendor/github.com/containerd/containerd/contrib/seccomp/seccomp.go

@@ -30,7 +30,7 @@ import (
 )
 
 // WithProfile receives the name of a file stored on disk comprising a json
-// formated seccomp profile, as specified by the opencontainers/runtime-spec.
+// formatted seccomp profile, as specified by the opencontainers/runtime-spec.
 // The profile is read from the file, unmarshaled, and set to the spec.
 func WithProfile(profile string) oci.SpecOpts {
 	return func(_ context.Context, _ oci.Client, _ *containers.Container, s *specs.Spec) error {

vendor/github.com/containerd/containerd/events/exchange/exchange.go

@@ -52,7 +52,7 @@ var _ events.Subscriber = &Exchange{}
 
 // Forward accepts an envelope to be direcly distributed on the exchange.
 //
-// This is useful when an event is forwaded on behalf of another namespace or
+// This is useful when an event is forwarded on behalf of another namespace or
 // when the event is propagated on behalf of another publisher.
 func (e *Exchange) Forward(ctx context.Context, envelope *events.Envelope) (err error) {
 	if err := validateEnvelope(envelope); err != nil {

vendor/github.com/containerd/containerd/export.go

@@ -0,0 +1,57 @@
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package containerd
+
+import (
+	"context"
+	"io"
+
+	"github.com/containerd/containerd/images"
+	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
+)
+
+type exportOpts struct {
+}
+
+// ExportOpt allows the caller to specify export-specific options
+type ExportOpt func(c *exportOpts) error
+
+func resolveExportOpt(opts ...ExportOpt) (exportOpts, error) {
+	var eopts exportOpts
+	for _, o := range opts {
+		if err := o(&eopts); err != nil {
+			return eopts, err
+		}
+	}
+	return eopts, nil
+}
+
+// Export exports an image to a Tar stream.
+// OCI format is used by default.
+// It is up to caller to put "org.opencontainers.image.ref.name" annotation to desc.
+// TODO(AkihiroSuda): support exporting multiple descriptors at once to a single archive stream.
+func (c *Client) Export(ctx context.Context, exporter images.Exporter, desc ocispec.Descriptor, opts ...ExportOpt) (io.ReadCloser, error) {
+	_, err := resolveExportOpt(opts...) // unused now
+	if err != nil {
+		return nil, err
+	}
+	pr, pw := io.Pipe()
+	go func() {
+		pw.CloseWithError(exporter.Export(ctx, c.ContentStore(), desc, pw))
+	}()
+	return pr, nil
+}

vendor/github.com/containerd/containerd/image.go

@@ -63,7 +63,7 @@ func NewImage(client *Client, i images.Image) Image {
 }
 
 // NewImageWithPlatform returns a client image object from the metadata image
-func NewImageWithPlatform(client *Client, i images.Image, platform string) Image {
+func NewImageWithPlatform(client *Client, i images.Image, platform platforms.MatchComparer) Image {
 	return &image{
 		client:   client,
 		i:        i,
@@ -75,7 +75,7 @@ type image struct {
 	client *Client
 
 	i        images.Image
-	platform string
+	platform platforms.MatchComparer
 }
 
 func (i *image) Name() string {
@@ -186,7 +186,7 @@ func (i *image) Unpack(ctx context.Context, snapshotterName string) error {
 	return nil
 }
 
-func (i *image) getLayers(ctx context.Context, platform string) ([]rootfs.Layer, error) {
+func (i *image) getLayers(ctx context.Context, platform platforms.MatchComparer) ([]rootfs.Layer, error) {
 	cs := i.client.ContentStore()
 
 	manifest, err := images.Manifest(ctx, cs, i.i.Target, platform)

vendor/github.com/containerd/containerd/images/handlers.go

@@ -19,6 +19,7 @@ package images
 import (
 	"context"
 	"fmt"
+	"sort"
 
 	"github.com/containerd/containerd/content"
 	"github.com/containerd/containerd/platforms"
@@ -183,8 +184,8 @@ func SetChildrenLabels(manager content.Manager, f HandlerFunc) HandlerFunc {
 }
 
 // FilterPlatforms is a handler wrapper which limits the descriptors returned
-// by a handler to the specified platforms.
-func FilterPlatforms(f HandlerFunc, platformList ...string) HandlerFunc {
+// based on matching the specified platform matcher.
+func FilterPlatforms(f HandlerFunc, m platforms.Matcher) HandlerFunc {
 	return func(ctx context.Context, desc ocispec.Descriptor) ([]ocispec.Descriptor, error) {
 		children, err := f(ctx, desc)
 		if err != nil {
@@ -193,20 +194,12 @@ func FilterPlatforms(f HandlerFunc, platformList ...string) HandlerFunc {
 
 		var descs []ocispec.Descriptor
 
-		if len(platformList) == 0 {
+		if m == nil {
 			descs = children
 		} else {
-			for _, platform := range platformList {
-				p, err := platforms.Parse(platform)
-				if err != nil {
-					return nil, err
-				}
-				matcher := platforms.NewMatcher(p)
-
-				for _, d := range children {
-					if d.Platform == nil || matcher.Match(*d.Platform) {
-						descs = append(descs, d)
-					}
+			for _, d := range children {
+				if d.Platform == nil || m.Match(*d.Platform) {
+					descs = append(descs, d)
 				}
 			}
 		}
@@ -214,3 +207,37 @@ func FilterPlatforms(f HandlerFunc, platformList ...string) HandlerFunc {
 		return descs, nil
 	}
 }
+
+// LimitManifests is a handler wrapper which filters the manifest descriptors
+// returned using the provided platform.
+// The results will be ordered according to the comparison operator and
+// use the ordering in the manifests for equal matches.
+// A limit of 0 or less is considered no limit.
+func LimitManifests(f HandlerFunc, m platforms.MatchComparer, n int) HandlerFunc {
+	return func(ctx context.Context, desc ocispec.Descriptor) ([]ocispec.Descriptor, error) {
+		children, err := f(ctx, desc)
+		if err != nil {
+			return children, err
+		}
+
+		switch desc.MediaType {
+		case ocispec.MediaTypeImageIndex, MediaTypeDockerSchema2ManifestList:
+			sort.SliceStable(children, func(i, j int) bool {
+				if children[i].Platform == nil {
+					return false
+				}
+				if children[j].Platform == nil {
+					return true
+				}
+				return m.Less(*children[i].Platform, *children[j].Platform)
+			})
+
+			if n > 0 && len(children) > n {
+				children = children[:n]
+			}
+		default:
+			// only limit manifests from an index
+		}
+		return children, nil
+	}
+}

vendor/github.com/containerd/containerd/images/image.go

@@ -19,6 +19,7 @@ package images
 import (
 	"context"
 	"encoding/json"
+	"sort"
 	"strings"
 	"time"
 
@@ -93,7 +94,7 @@ type Store interface {
 //
 // The caller can then use the descriptor to resolve and process the
 // configuration of the image.
-func (image *Image) Config(ctx context.Context, provider content.Provider, platform string) (ocispec.Descriptor, error) {
+func (image *Image) Config(ctx context.Context, provider content.Provider, platform platforms.MatchComparer) (ocispec.Descriptor, error) {
 	return Config(ctx, provider, image.Target, platform)
 }
 
@@ -101,7 +102,7 @@ func (image *Image) Config(ctx context.Context, provider content.Provider, platf
 //
 // These are used to verify that a set of layers unpacked to the expected
 // values.
-func (image *Image) RootFS(ctx context.Context, provider content.Provider, platform string) ([]digest.Digest, error) {
+func (image *Image) RootFS(ctx context.Context, provider content.Provider, platform platforms.MatchComparer) ([]digest.Digest, error) {
 	desc, err := image.Config(ctx, provider, platform)
 	if err != nil {
 		return nil, err
@@ -110,7 +111,7 @@ func (image *Image) RootFS(ctx context.Context, provider content.Provider, platf
 }
 
 // Size returns the total size of an image's packed resources.
-func (image *Image) Size(ctx context.Context, provider content.Provider, platform string) (int64, error) {
+func (image *Image) Size(ctx context.Context, provider content.Provider, platform platforms.MatchComparer) (int64, error) {
 	var size int64
 	return size, Walk(ctx, Handlers(HandlerFunc(func(ctx context.Context, desc ocispec.Descriptor) ([]ocispec.Descriptor, error) {
 		if desc.Size < 0 {
@@ -121,27 +122,22 @@ func (image *Image) Size(ctx context.Context, provider content.Provider, platfor
 	}), FilterPlatforms(ChildrenHandler(provider), platform)), image.Target)
 }
 
+type platformManifest struct {
+	p *ocispec.Platform
+	m *ocispec.Manifest
+}
+
 // Manifest resolves a manifest from the image for the given platform.
 //
 // TODO(stevvooe): This violates the current platform agnostic approach to this
 // package by returning a specific manifest type. We'll need to refactor this
 // to return a manifest descriptor or decide that we want to bring the API in
 // this direction because this abstraction is not needed.`
-func Manifest(ctx context.Context, provider content.Provider, image ocispec.Descriptor, platform string) (ocispec.Manifest, error) {
+func Manifest(ctx context.Context, provider content.Provider, image ocispec.Descriptor, platform platforms.MatchComparer) (ocispec.Manifest, error) {
 	var (
-		matcher  platforms.Matcher
-		m        *ocispec.Manifest
-		p        ocispec.Platform
+		m        []platformManifest
 		wasIndex bool
 	)
-	if platform != "" {
-		var err error
-		p, err = platforms.Parse(platform)
-		if err != nil {
-			return ocispec.Manifest{}, err
-		}
-		matcher = platforms.NewMatcher(p)
-	}
 
 	if err := Walk(ctx, HandlerFunc(func(ctx context.Context, desc ocispec.Descriptor) ([]ocispec.Descriptor, error) {
 		switch desc.MediaType {
@@ -156,8 +152,8 @@ func Manifest(ctx context.Context, provider content.Provider, image ocispec.Desc
 				return nil, err
 			}
 
-			if platform != "" {
-				if desc.Platform != nil && !matcher.Match(*desc.Platform) {
+			if platform != nil {
+				if desc.Platform != nil && !platform.Match(*desc.Platform) {
 					return nil, nil
 				}
 
@@ -172,14 +168,17 @@ func Manifest(ctx context.Context, provider content.Provider, image ocispec.Desc
 						return nil, err
 					}
 
-					if !matcher.Match(platforms.Normalize(ocispec.Platform{OS: image.OS, Architecture: image.Architecture})) {
+					if !platform.Match(platforms.Normalize(ocispec.Platform{OS: image.OS, Architecture: image.Architecture})) {
 						return nil, nil
 					}
 
 				}
 			}
 
-			m = &manifest
+			m = append(m, platformManifest{
+				p: desc.Platform,
+				m: &manifest,
+			})
 
 			return nil, nil
 		case MediaTypeDockerSchema2ManifestList, ocispec.MediaTypeImageIndex:
@@ -193,13 +192,13 @@ func Manifest(ctx context.Context, provider content.Provider, image ocispec.Desc
 				return nil, err
 			}
 
-			if platform == "" {
+			if platform == nil {
 				return idx.Manifests, nil
 			}
 
 			var descs []ocispec.Descriptor
 			for _, d := range idx.Manifests {
-				if d.Platform == nil || matcher.Match(*d.Platform) {
+				if d.Platform == nil || platform.Match(*d.Platform) {
 					descs = append(descs, d)
 				}
 			}
@@ -214,15 +213,25 @@ func Manifest(ctx context.Context, provider content.Provider, image ocispec.Desc
 		return ocispec.Manifest{}, err
 	}
 
-	if m == nil {
+	if len(m) == 0 {
 		err := errors.Wrapf(errdefs.ErrNotFound, "manifest %v", image.Digest)
 		if wasIndex {
-			err = errors.Wrapf(errdefs.ErrNotFound, "no match for current platform %s in manifest %v", platforms.Format(p), image.Digest)
+			err = errors.Wrapf(errdefs.ErrNotFound, "no match for platform in manifest %v", image.Digest)
 		}
 		return ocispec.Manifest{}, err
 	}
 
-	return *m, nil
+	sort.SliceStable(m, func(i, j int) bool {
+		if m[i].p == nil {
+			return false
+		}
+		if m[j].p == nil {
+			return true
+		}
+		return platform.Less(*m[i].p, *m[j].p)
+	})
+
+	return *m[0].m, nil
 }
 
 // Config resolves the image configuration descriptor using a content provided
@@ -230,7 +239,7 @@ func Manifest(ctx context.Context, provider content.Provider, image ocispec.Desc
 //
 // The caller can then use the descriptor to resolve and process the
 // configuration of the image.
-func Config(ctx context.Context, provider content.Provider, image ocispec.Descriptor, platform string) (ocispec.Descriptor, error) {
+func Config(ctx context.Context, provider content.Provider, image ocispec.Descriptor, platform platforms.MatchComparer) (ocispec.Descriptor, error) {
 	manifest, err := Manifest(ctx, provider, image, platform)
 	if err != nil {
 		return ocispec.Descriptor{}, err
@@ -276,7 +285,7 @@ func Platforms(ctx context.Context, provider content.Provider, image ocispec.Des
 // in the provider.
 //
 // If there is a problem resolving content, an error will be returned.
-func Check(ctx context.Context, provider content.Provider, image ocispec.Descriptor, platform string) (available bool, required, present, missing []ocispec.Descriptor, err error) {
+func Check(ctx context.Context, provider content.Provider, image ocispec.Descriptor, platform platforms.MatchComparer) (available bool, required, present, missing []ocispec.Descriptor, err error) {
 	mfst, err := Manifest(ctx, provider, image, platform)
 	if err != nil {
 		if errdefs.IsNotFound(err) {

vendor/github.com/containerd/containerd/import.go

@@ -22,7 +22,6 @@ import (
 
 	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/images"
-	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 )
 
 type importOpts struct {
@@ -84,35 +83,3 @@ func (c *Client) Import(ctx context.Context, importer images.Importer, reader io
 	}
 	return images, nil
 }
-
-type exportOpts struct {
-}
-
-// ExportOpt allows the caller to specify export-specific options
-type ExportOpt func(c *exportOpts) error
-
-func resolveExportOpt(opts ...ExportOpt) (exportOpts, error) {
-	var eopts exportOpts
-	for _, o := range opts {
-		if err := o(&eopts); err != nil {
-			return eopts, err
-		}
-	}
-	return eopts, nil
-}
-
-// Export exports an image to a Tar stream.
-// OCI format is used by default.
-// It is up to caller to put "org.opencontainers.image.ref.name" annotation to desc.
-// TODO(AkihiroSuda): support exporting multiple descriptors at once to a single archive stream.
-func (c *Client) Export(ctx context.Context, exporter images.Exporter, desc ocispec.Descriptor, opts ...ExportOpt) (io.ReadCloser, error) {
-	_, err := resolveExportOpt(opts...) // unused now
-	if err != nil {
-		return nil, err
-	}
-	pr, pw := io.Pipe()
-	go func() {
-		pw.CloseWithError(exporter.Export(ctx, c.ContentStore(), desc, pw))
-	}()
-	return pr, nil
-}

vendor/github.com/containerd/containerd/install.go

@@ -33,25 +33,14 @@ import (
 
 // Install a binary image into the opt service
 func (c *Client) Install(ctx context.Context, image Image, opts ...InstallOpts) error {
-	resp, err := c.IntrospectionService().Plugins(ctx, &introspectionapi.PluginsRequest{
-		Filters: []string{
-			"id==opt",
-		},
-	})
-	if err != nil {
-		return err
-	}
-	if len(resp.Plugins) != 1 {
-		return errors.New("opt service not enabled")
-	}
-	path := resp.Plugins[0].Exports["path"]
-	if path == "" {
-		return errors.New("opt path not exported")
-	}
 	var config InstallConfig
 	for _, o := range opts {
 		o(&config)
 	}
+	path, err := c.getInstallPath(ctx, config)
+	if err != nil {
+		return err
+	}
 	var (
 		cs       = image.ContentStore()
 		platform = platforms.Default()
@@ -89,3 +78,25 @@ func (c *Client) Install(ctx context.Context, image Image, opts ...InstallOpts)
 	}
 	return nil
 }
+
+func (c *Client) getInstallPath(ctx context.Context, config InstallConfig) (string, error) {
+	if config.Path != "" {
+		return config.Path, nil
+	}
+	resp, err := c.IntrospectionService().Plugins(ctx, &introspectionapi.PluginsRequest{
+		Filters: []string{
+			"id==opt",
+		},
+	})
+	if err != nil {
+		return "", err
+	}
+	if len(resp.Plugins) != 1 {
+		return "", errors.New("opt service not enabled")
+	}
+	path := resp.Plugins[0].Exports["path"]
+	if path == "" {
+		return "", errors.New("opt path not exported")
+	}
+	return path, nil
+}

vendor/github.com/containerd/containerd/install_opts.go

@@ -25,6 +25,8 @@ type InstallConfig struct {
 	Libs bool
 	// Replace will overwrite existing binaries or libs in the opt directory
 	Replace bool
+	// Path to install libs and binaries to
+	Path string
 }
 
 // WithInstallLibs installs libs from the image
@@ -36,3 +38,10 @@ func WithInstallLibs(c *InstallConfig) {
 func WithInstallReplace(c *InstallConfig) {
 	c.Replace = true
 }
+
+// WithInstallPath sets the optional install path
+func WithInstallPath(path string) InstallOpts {
+	return func(c *InstallConfig) {
+		c.Path = path
+	}
+}

vendor/github.com/containerd/containerd/metadata/buckets.go

@@ -21,7 +21,7 @@ import (
 	digest "github.com/opencontainers/go-digest"
 )
 
-// The layout where a "/" delineates a bucket is desribed in the following
+// The layout where a "/" delineates a bucket is described in the following
 // section. Please try to follow this as closely as possible when adding
 // functionality. We can bolster this with helpers and more structure if that
 // becomes an issue.

vendor/github.com/containerd/containerd/mount/mount_linux.go

@@ -17,16 +17,41 @@
 package mount
 
 import (
+	"fmt"
+	"os"
+	"path"
 	"strings"
 	"time"
 
+	"github.com/containerd/containerd/sys"
 	"github.com/pkg/errors"
 	"golang.org/x/sys/unix"
 )
 
+var pagesize = 4096
+
+func init() {
+	pagesize = os.Getpagesize()
+}
+
 // Mount to the provided target path
 func (m *Mount) Mount(target string) error {
-	flags, data := parseMountOptions(m.Options)
+	var (
+		chdir   string
+		options = m.Options
+	)
+
+	// avoid hitting one page limit of mount argument buffer
+	//
+	// NOTE: 512 is a buffer during pagesize check.
+	if m.Type == "overlay" && optionsSize(options) >= pagesize-512 {
+		chdir, options = compactLowerdirOption(options)
+	}
+
+	flags, data := parseMountOptions(options)
+	if len(data) > pagesize {
+		return errors.Errorf("mount options is too long")
+	}
 
 	// propagation types.
 	const ptypes = unix.MS_SHARED | unix.MS_PRIVATE | unix.MS_SLAVE | unix.MS_UNBINDABLE
@@ -38,7 +63,7 @@ func (m *Mount) Mount(target string) error {
 	if flags&unix.MS_REMOUNT == 0 || data != "" {
 		// Initial call applying all non-propagation flags for mount
 		// or remount with changed data
-		if err := unix.Mount(m.Source, target, m.Type, uintptr(oflags), data); err != nil {
+		if err := mountAt(chdir, m.Source, target, m.Type, uintptr(oflags), data); err != nil {
 			return err
 		}
 	}
@@ -155,3 +180,129 @@ func parseMountOptions(options []string) (int, string) {
 	}
 	return flag, strings.Join(data, ",")
 }
+
+// compactLowerdirOption updates overlay lowdir option and returns the common
+// dir among all the lowdirs.
+func compactLowerdirOption(opts []string) (string, []string) {
+	idx, dirs := findOverlayLowerdirs(opts)
+	if idx == -1 || len(dirs) == 1 {
+		// no need to compact if there is only one lowerdir
+		return "", opts
+	}
+
+	// find out common dir
+	commondir := longestCommonPrefix(dirs)
+	if commondir == "" {
+		return "", opts
+	}
+
+	// NOTE: the snapshot id is based on digits.
+	// in order to avoid to get snapshots/x, should be back to parent dir.
+	// however, there is assumption that the common dir is ${root}/io.containerd.v1.overlayfs/snapshots.
+	commondir = path.Dir(commondir)
+	if commondir == "/" {
+		return "", opts
+	}
+	commondir = commondir + "/"
+
+	newdirs := make([]string, 0, len(dirs))
+	for _, dir := range dirs {
+		newdirs = append(newdirs, dir[len(commondir):])
+	}
+
+	newopts := copyOptions(opts)
+	newopts = append(newopts[:idx], newopts[idx+1:]...)
+	newopts = append(newopts, fmt.Sprintf("lowerdir=%s", strings.Join(newdirs, ":")))
+	return commondir, newopts
+}
+
+// findOverlayLowerdirs returns the index of lowerdir in mount's options and
+// all the lowerdir target.
+func findOverlayLowerdirs(opts []string) (int, []string) {
+	var (
+		idx    = -1
+		prefix = "lowerdir="
+	)
+
+	for i, opt := range opts {
+		if strings.HasPrefix(opt, prefix) {
+			idx = i
+			break
+		}
+	}
+
+	if idx == -1 {
+		return -1, nil
+	}
+	return idx, strings.Split(opts[idx][len(prefix):], ":")
+}
+
+// longestCommonPrefix finds the longest common prefix in the string slice.
+func longestCommonPrefix(strs []string) string {
+	if len(strs) == 0 {
+		return ""
+	} else if len(strs) == 1 {
+		return strs[0]
+	}
+
+	// find out the min/max value by alphabetical order
+	min, max := strs[0], strs[0]
+	for _, str := range strs[1:] {
+		if min > str {
+			min = str
+		}
+		if max < str {
+			max = str
+		}
+	}
+
+	// find out the common part between min and max
+	for i := 0; i < len(min) && i < len(max); i++ {
+		if min[i] != max[i] {
+			return min[:i]
+		}
+	}
+	return min
+}
+
+// copyOptions copies the options.
+func copyOptions(opts []string) []string {
+	if len(opts) == 0 {
+		return nil
+	}
+
+	acopy := make([]string, len(opts))
+	copy(acopy, opts)
+	return acopy
+}
+
+// optionsSize returns the byte size of options of mount.
+func optionsSize(opts []string) int {
+	size := 0
+	for _, opt := range opts {
+		size += len(opt)
+	}
+	return size
+}
+
+func mountAt(chdir string, source, target, fstype string, flags uintptr, data string) error {
+	if chdir == "" {
+		return unix.Mount(source, target, fstype, flags, data)
+	}
+
+	f, err := os.Open(chdir)
+	if err != nil {
+		return errors.Wrap(err, "failed to mountat")
+	}
+	defer f.Close()
+
+	fs, err := f.Stat()
+	if err != nil {
+		return errors.Wrap(err, "failed to mountat")
+	}
+
+	if !fs.IsDir() {
+		return errors.Wrap(errors.Errorf("%s is not dir", chdir), "failed to mountat")
+	}
+	return errors.Wrap(sys.FMountat(f.Fd(), source, target, fstype, flags, data), "failed to mountat")
+}

vendor/github.com/containerd/containerd/mount/mount_windows.go

@@ -32,6 +32,10 @@ var (
 
 // Mount to the provided target
 func (m *Mount) Mount(target string) error {
+	if m.Type != "windows-layer" {
+		return errors.Errorf("invalid windows mount type: '%s'", m.Type)
+	}
+
 	home, layerID := filepath.Split(m.Source)
 
 	parentLayerPaths, err := m.GetParentPaths()

vendor/github.com/containerd/containerd/oci/spec.go

@@ -18,11 +18,27 @@ package oci
 
 import (
 	"context"
+	"path/filepath"
+	"runtime"
+
+	"github.com/containerd/containerd/namespaces"
+	"github.com/containerd/containerd/platforms"
 
 	"github.com/containerd/containerd/containers"
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 )
 
+const (
+	rwm               = "rwm"
+	defaultRootfsPath = "rootfs"
+)
+
+var (
+	defaultUnixEnv = []string{
+		"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
+	}
+)
+
 // Spec is a type alias to the OCI runtime spec to allow third part SpecOpts
 // to be created without the "issues" with go vendoring and package imports
 type Spec = specs.Spec
@@ -30,12 +46,36 @@ type Spec = specs.Spec
 // GenerateSpec will generate a default spec from the provided image
 // for use as a containerd container
 func GenerateSpec(ctx context.Context, client Client, c *containers.Container, opts ...SpecOpts) (*Spec, error) {
-	s, err := createDefaultSpec(ctx, c.ID)
-	if err != nil {
+	return GenerateSpecWithPlatform(ctx, client, platforms.DefaultString(), c, opts...)
+}
+
+// GenerateSpecWithPlatform will generate a default spec from the provided image
+// for use as a containerd container in the platform requested.
+func GenerateSpecWithPlatform(ctx context.Context, client Client, platform string, c *containers.Container, opts ...SpecOpts) (*Spec, error) {
+	var s Spec
+	if err := generateDefaultSpecWithPlatform(ctx, platform, c.ID, &s); err != nil {
 		return nil, err
 	}
 
-	return s, ApplyOpts(ctx, client, c, s, opts...)
+	return &s, ApplyOpts(ctx, client, c, &s, opts...)
+}
+
+func generateDefaultSpecWithPlatform(ctx context.Context, platform, id string, s *Spec) error {
+	plat, err := platforms.Parse(platform)
+	if err != nil {
+		return err
+	}
+
+	if plat.OS == "windows" {
+		err = populateDefaultWindowsSpec(ctx, s, id)
+	} else {
+		err = populateDefaultUnixSpec(ctx, s, id)
+		if err == nil && runtime.GOOS == "windows" {
+			// To run LCOW we have a Linux and Windows section. Add an empty one now.
+			s.Windows = &specs.Windows{}
+		}
+	}
+	return err
 }
 
 // ApplyOpts applys the options to the given spec, injecting data from the
@@ -50,7 +90,173 @@ func ApplyOpts(ctx context.Context, client Client, c *containers.Container, s *S
 	return nil
 }
 
-func createDefaultSpec(ctx context.Context, id string) (*Spec, error) {
-	var s Spec
-	return &s, populateDefaultSpec(ctx, &s, id)
+func defaultUnixCaps() []string {
+	return []string{
+		"CAP_CHOWN",
+		"CAP_DAC_OVERRIDE",
+		"CAP_FSETID",
+		"CAP_FOWNER",
+		"CAP_MKNOD",
+		"CAP_NET_RAW",
+		"CAP_SETGID",
+		"CAP_SETUID",
+		"CAP_SETFCAP",
+		"CAP_SETPCAP",
+		"CAP_NET_BIND_SERVICE",
+		"CAP_SYS_CHROOT",
+		"CAP_KILL",
+		"CAP_AUDIT_WRITE",
+	}
+}
+
+func defaultUnixNamespaces() []specs.LinuxNamespace {
+	return []specs.LinuxNamespace{
+		{
+			Type: specs.PIDNamespace,
+		},
+		{
+			Type: specs.IPCNamespace,
+		},
+		{
+			Type: specs.UTSNamespace,
+		},
+		{
+			Type: specs.MountNamespace,
+		},
+		{
+			Type: specs.NetworkNamespace,
+		},
+	}
+}
+
+func populateDefaultUnixSpec(ctx context.Context, s *Spec, id string) error {
+	ns, err := namespaces.NamespaceRequired(ctx)
+	if err != nil {
+		return err
+	}
+
+	*s = Spec{
+		Version: specs.Version,
+		Root: &specs.Root{
+			Path: defaultRootfsPath,
+		},
+		Process: &specs.Process{
+			Env:             defaultUnixEnv,
+			Cwd:             "/",
+			NoNewPrivileges: true,
+			User: specs.User{
+				UID: 0,
+				GID: 0,
+			},
+			Capabilities: &specs.LinuxCapabilities{
+				Bounding:    defaultUnixCaps(),
+				Permitted:   defaultUnixCaps(),
+				Inheritable: defaultUnixCaps(),
+				Effective:   defaultUnixCaps(),
+			},
+			Rlimits: []specs.POSIXRlimit{
+				{
+					Type: "RLIMIT_NOFILE",
+					Hard: uint64(1024),
+					Soft: uint64(1024),
+				},
+			},
+		},
+		Mounts: []specs.Mount{
+			{
+				Destination: "/proc",
+				Type:        "proc",
+				Source:      "proc",
+			},
+			{
+				Destination: "/dev",
+				Type:        "tmpfs",
+				Source:      "tmpfs",
+				Options:     []string{"nosuid", "strictatime", "mode=755", "size=65536k"},
+			},
+			{
+				Destination: "/dev/pts",
+				Type:        "devpts",
+				Source:      "devpts",
+				Options:     []string{"nosuid", "noexec", "newinstance", "ptmxmode=0666", "mode=0620", "gid=5"},
+			},
+			{
+				Destination: "/dev/shm",
+				Type:        "tmpfs",
+				Source:      "shm",
+				Options:     []string{"nosuid", "noexec", "nodev", "mode=1777", "size=65536k"},
+			},
+			{
+				Destination: "/dev/mqueue",
+				Type:        "mqueue",
+				Source:      "mqueue",
+				Options:     []string{"nosuid", "noexec", "nodev"},
+			},
+			{
+				Destination: "/sys",
+				Type:        "sysfs",
+				Source:      "sysfs",
+				Options:     []string{"nosuid", "noexec", "nodev", "ro"},
+			},
+			{
+				Destination: "/run",
+				Type:        "tmpfs",
+				Source:      "tmpfs",
+				Options:     []string{"nosuid", "strictatime", "mode=755", "size=65536k"},
+			},
+		},
+		Linux: &specs.Linux{
+			MaskedPaths: []string{
+				"/proc/acpi",
+				"/proc/kcore",
+				"/proc/keys",
+				"/proc/latency_stats",
+				"/proc/timer_list",
+				"/proc/timer_stats",
+				"/proc/sched_debug",
+				"/sys/firmware",
+				"/proc/scsi",
+			},
+			ReadonlyPaths: []string{
+				"/proc/asound",
+				"/proc/bus",
+				"/proc/fs",
+				"/proc/irq",
+				"/proc/sys",
+				"/proc/sysrq-trigger",
+			},
+			CgroupsPath: filepath.Join("/", ns, id),
+			Resources: &specs.LinuxResources{
+				Devices: []specs.LinuxDeviceCgroup{
+					{
+						Allow:  false,
+						Access: rwm,
+					},
+				},
+			},
+			Namespaces: defaultUnixNamespaces(),
+		},
+	}
+	return nil
+}
+
+func populateDefaultWindowsSpec(ctx context.Context, s *Spec, id string) error {
+	*s = Spec{
+		Version: specs.Version,
+		Root:    &specs.Root{},
+		Process: &specs.Process{
+			Cwd: `C:\`,
+			ConsoleSize: &specs.Box{
+				Width:  80,
+				Height: 20,
+			},
+		},
+		Windows: &specs.Windows{
+			IgnoreFlushesDuringBoot: true,
+			Network: &specs.WindowsNetwork{
+				AllowUnqualifiedDNSQuery: true,
+			},
+		},
+	}
+	return nil
 }

vendor/github.com/containerd/containerd/oci/spec_opts.go

@@ -19,12 +19,25 @@ package oci
 import (
 	"context"
 	"encoding/json"
+	"fmt"
 	"io/ioutil"
+	"os"
+	"path/filepath"
+	"strconv"
 	"strings"
 
 	"github.com/containerd/containerd/containers"
+	"github.com/containerd/containerd/content"
+	"github.com/containerd/containerd/images"
+	"github.com/containerd/containerd/mount"
+	"github.com/containerd/containerd/namespaces"
+	"github.com/containerd/containerd/platforms"
+	"github.com/containerd/continuity/fs"
+	"github.com/opencontainers/image-spec/specs-go/v1"
+	"github.com/opencontainers/runc/libcontainer/user"
 	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/pkg/errors"
+	"github.com/syndtr/gocapability/capability"
 )
 
 // SpecOpts sets spec specific information to a newly generated OCI spec
@@ -49,13 +62,45 @@ func setProcess(s *Spec) {
 	}
 }
 
+// setRoot sets Root to empty if unset
+func setRoot(s *Spec) {
+	if s.Root == nil {
+		s.Root = &specs.Root{}
+	}
+}
+
+// setLinux sets Linux to empty if unset
+func setLinux(s *Spec) {
+	if s.Linux == nil {
+		s.Linux = &specs.Linux{}
+	}
+}
+
+// setCapabilities sets Linux Capabilities to empty if unset
+func setCapabilities(s *Spec) {
+	setProcess(s)
+	if s.Process.Capabilities == nil {
+		s.Process.Capabilities = &specs.LinuxCapabilities{}
+	}
+}
+
 // WithDefaultSpec returns a SpecOpts that will populate the spec with default
 // values.
 //
 // Use as the first option to clear the spec, then apply options afterwards.
 func WithDefaultSpec() SpecOpts {
 	return func(ctx context.Context, _ Client, c *containers.Container, s *Spec) error {
-		return populateDefaultSpec(ctx, s, c.ID)
+		return generateDefaultSpecWithPlatform(ctx, platforms.DefaultString(), c.ID, s)
+	}
+}
+
+// WithDefaultSpecForPlatform returns a SpecOpts that will populate the spec
+// with default values for a given platform.
+//
+// Use as the first option to clear the spec, then apply options afterwards.
+func WithDefaultSpecForPlatform(platform string) SpecOpts {
+	return func(ctx context.Context, _ Client, c *containers.Container, s *Spec) error {
+		return generateDefaultSpecWithPlatform(ctx, platform, c.ID, s)
 	}
 }
 
@@ -81,32 +126,6 @@ func WithSpecFromFile(filename string) SpecOpts {
 	}
 }
 
-// WithProcessArgs replaces the args on the generated spec
-func WithProcessArgs(args ...string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setProcess(s)
-		s.Process.Args = args
-		return nil
-	}
-}
-
-// WithProcessCwd replaces the current working directory on the generated spec
-func WithProcessCwd(cwd string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setProcess(s)
-		s.Process.Cwd = cwd
-		return nil
-	}
-}
-
-// WithHostname sets the container's hostname
-func WithHostname(name string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		s.Hostname = name
-		return nil
-	}
-}
-
 // WithEnv appends environment variables
 func WithEnv(environmentVariables []string) SpecOpts {
 	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
@@ -118,14 +137,6 @@ func WithEnv(environmentVariables []string) SpecOpts {
 	}
 }
 
-// WithMounts appends mounts
-func WithMounts(mounts []specs.Mount) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		s.Mounts = append(s.Mounts, mounts...)
-		return nil
-	}
-}
-
 // replaceOrAppendEnvValues returns the defaults with the overrides either
 // replaced by env key or appended to the list
 func replaceOrAppendEnvValues(defaults, overrides []string) []string {
@@ -163,3 +174,826 @@ func replaceOrAppendEnvValues(defaults, overrides []string) []string {
 
 	return defaults
 }
+
+// WithProcessArgs replaces the args on the generated spec
+func WithProcessArgs(args ...string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setProcess(s)
+		s.Process.Args = args
+		return nil
+	}
+}
+
+// WithProcessCwd replaces the current working directory on the generated spec
+func WithProcessCwd(cwd string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setProcess(s)
+		s.Process.Cwd = cwd
+		return nil
+	}
+}
+
+// WithTTY sets the information on the spec as well as the environment variables for
+// using a TTY
+func WithTTY(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	setProcess(s)
+	s.Process.Terminal = true
+	if s.Linux != nil {
+		s.Process.Env = append(s.Process.Env, "TERM=xterm")
+	}
+
+	return nil
+}
+
+// WithTTYSize sets the information on the spec as well as the environment variables for
+// using a TTY
+func WithTTYSize(width, height int) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setProcess(s)
+		if s.Process.ConsoleSize == nil {
+			s.Process.ConsoleSize = &specs.Box{}
+		}
+		s.Process.ConsoleSize.Width = uint(width)
+		s.Process.ConsoleSize.Height = uint(height)
+		return nil
+	}
+}
+
+// WithHostname sets the container's hostname
+func WithHostname(name string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		s.Hostname = name
+		return nil
+	}
+}
+
+// WithMounts appends mounts
+func WithMounts(mounts []specs.Mount) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		s.Mounts = append(s.Mounts, mounts...)
+		return nil
+	}
+}
+
+// WithHostNamespace allows a task to run inside the host's linux namespace
+func WithHostNamespace(ns specs.LinuxNamespaceType) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setLinux(s)
+		for i, n := range s.Linux.Namespaces {
+			if n.Type == ns {
+				s.Linux.Namespaces = append(s.Linux.Namespaces[:i], s.Linux.Namespaces[i+1:]...)
+				return nil
+			}
+		}
+		return nil
+	}
+}
+
+// WithLinuxNamespace uses the passed in namespace for the spec. If a namespace of the same type already exists in the
+// spec, the existing namespace is replaced by the one provided.
+func WithLinuxNamespace(ns specs.LinuxNamespace) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setLinux(s)
+		for i, n := range s.Linux.Namespaces {
+			if n.Type == ns.Type {
+				before := s.Linux.Namespaces[:i]
+				after := s.Linux.Namespaces[i+1:]
+				s.Linux.Namespaces = append(before, ns)
+				s.Linux.Namespaces = append(s.Linux.Namespaces, after...)
+				return nil
+			}
+		}
+		s.Linux.Namespaces = append(s.Linux.Namespaces, ns)
+		return nil
+	}
+}
+
+// WithImageConfig configures the spec to from the configuration of an Image
+func WithImageConfig(image Image) SpecOpts {
+	return WithImageConfigArgs(image, nil)
+}
+
+// WithImageConfigArgs configures the spec to from the configuration of an Image with additional args that
+// replaces the CMD of the image
+func WithImageConfigArgs(image Image, args []string) SpecOpts {
+	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) error {
+		ic, err := image.Config(ctx)
+		if err != nil {
+			return err
+		}
+		var (
+			ociimage v1.Image
+			config   v1.ImageConfig
+		)
+		switch ic.MediaType {
+		case v1.MediaTypeImageConfig, images.MediaTypeDockerSchema2Config:
+			p, err := content.ReadBlob(ctx, image.ContentStore(), ic)
+			if err != nil {
+				return err
+			}
+
+			if err := json.Unmarshal(p, &ociimage); err != nil {
+				return err
+			}
+			config = ociimage.Config
+		default:
+			return fmt.Errorf("unknown image config media type %s", ic.MediaType)
+		}
+
+		setProcess(s)
+		if s.Linux != nil {
+			s.Process.Env = append(s.Process.Env, config.Env...)
+			cmd := config.Cmd
+			if len(args) > 0 {
+				cmd = args
+			}
+			s.Process.Args = append(config.Entrypoint, cmd...)
+
+			cwd := config.WorkingDir
+			if cwd == "" {
+				cwd = "/"
+			}
+			s.Process.Cwd = cwd
+			if config.User != "" {
+				return WithUser(config.User)(ctx, client, c, s)
+			}
+		} else if s.Windows != nil {
+			s.Process.Env = config.Env
+			s.Process.Args = append(config.Entrypoint, config.Cmd...)
+			s.Process.User = specs.User{
+				Username: config.User,
+			}
+		} else {
+			return errors.New("spec does not contain Linux or Windows section")
+		}
+		return nil
+	}
+}
+
+// WithRootFSPath specifies unmanaged rootfs path.
+func WithRootFSPath(path string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setRoot(s)
+		s.Root.Path = path
+		// Entrypoint is not set here (it's up to caller)
+		return nil
+	}
+}
+
+// WithRootFSReadonly sets specs.Root.Readonly to true
+func WithRootFSReadonly() SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setRoot(s)
+		s.Root.Readonly = true
+		return nil
+	}
+}
+
+// WithNoNewPrivileges sets no_new_privileges on the process for the container
+func WithNoNewPrivileges(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	setProcess(s)
+	s.Process.NoNewPrivileges = true
+	return nil
+}
+
+// WithHostHostsFile bind-mounts the host's /etc/hosts into the container as readonly
+func WithHostHostsFile(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	s.Mounts = append(s.Mounts, specs.Mount{
+		Destination: "/etc/hosts",
+		Type:        "bind",
+		Source:      "/etc/hosts",
+		Options:     []string{"rbind", "ro"},
+	})
+	return nil
+}
+
+// WithHostResolvconf bind-mounts the host's /etc/resolv.conf into the container as readonly
+func WithHostResolvconf(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	s.Mounts = append(s.Mounts, specs.Mount{
+		Destination: "/etc/resolv.conf",
+		Type:        "bind",
+		Source:      "/etc/resolv.conf",
+		Options:     []string{"rbind", "ro"},
+	})
+	return nil
+}
+
+// WithHostLocaltime bind-mounts the host's /etc/localtime into the container as readonly
+func WithHostLocaltime(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	s.Mounts = append(s.Mounts, specs.Mount{
+		Destination: "/etc/localtime",
+		Type:        "bind",
+		Source:      "/etc/localtime",
+		Options:     []string{"rbind", "ro"},
+	})
+	return nil
+}
+
+// WithUserNamespace sets the uid and gid mappings for the task
+// this can be called multiple times to add more mappings to the generated spec
+func WithUserNamespace(container, host, size uint32) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		var hasUserns bool
+		setLinux(s)
+		for _, ns := range s.Linux.Namespaces {
+			if ns.Type == specs.UserNamespace {
+				hasUserns = true
+				break
+			}
+		}
+		if !hasUserns {
+			s.Linux.Namespaces = append(s.Linux.Namespaces, specs.LinuxNamespace{
+				Type: specs.UserNamespace,
+			})
+		}
+		mapping := specs.LinuxIDMapping{
+			ContainerID: container,
+			HostID:      host,
+			Size:        size,
+		}
+		s.Linux.UIDMappings = append(s.Linux.UIDMappings, mapping)
+		s.Linux.GIDMappings = append(s.Linux.GIDMappings, mapping)
+		return nil
+	}
+}
+
+// WithCgroup sets the container's cgroup path
+func WithCgroup(path string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setLinux(s)
+		s.Linux.CgroupsPath = path
+		return nil
+	}
+}
+
+// WithNamespacedCgroup uses the namespace set on the context to create a
+// root directory for containers in the cgroup with the id as the subcgroup
+func WithNamespacedCgroup() SpecOpts {
+	return func(ctx context.Context, _ Client, c *containers.Container, s *Spec) error {
+		namespace, err := namespaces.NamespaceRequired(ctx)
+		if err != nil {
+			return err
+		}
+		setLinux(s)
+		s.Linux.CgroupsPath = filepath.Join("/", namespace, c.ID)
+		return nil
+	}
+}
+
+// WithUser sets the user to be used within the container.
+// It accepts a valid user string in OCI Image Spec v1.0.0:
+//   user, uid, user:group, uid:gid, uid:group, user:gid
+func WithUser(userstr string) SpecOpts {
+	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) error {
+		setProcess(s)
+		parts := strings.Split(userstr, ":")
+		switch len(parts) {
+		case 1:
+			v, err := strconv.Atoi(parts[0])
+			if err != nil {
+				// if we cannot parse as a uint they try to see if it is a username
+				return WithUsername(userstr)(ctx, client, c, s)
+			}
+			return WithUserID(uint32(v))(ctx, client, c, s)
+		case 2:
+			var (
+				username  string
+				groupname string
+			)
+			var uid, gid uint32
+			v, err := strconv.Atoi(parts[0])
+			if err != nil {
+				username = parts[0]
+			} else {
+				uid = uint32(v)
+			}
+			if v, err = strconv.Atoi(parts[1]); err != nil {
+				groupname = parts[1]
+			} else {
+				gid = uint32(v)
+			}
+			if username == "" && groupname == "" {
+				s.Process.User.UID, s.Process.User.GID = uid, gid
+				return nil
+			}
+			f := func(root string) error {
+				if username != "" {
+					user, err := getUserFromPath(root, func(u user.User) bool {
+						return u.Name == username
+					})
+					if err != nil {
+						return err
+					}
+					uid = uint32(user.Uid)
+				}
+				if groupname != "" {
+					gid, err = getGIDFromPath(root, func(g user.Group) bool {
+						return g.Name == groupname
+					})
+					if err != nil {
+						return err
+					}
+				}
+				s.Process.User.UID, s.Process.User.GID = uid, gid
+				return nil
+			}
+			if c.Snapshotter == "" && c.SnapshotKey == "" {
+				if !isRootfsAbs(s.Root.Path) {
+					return errors.New("rootfs absolute path is required")
+				}
+				return f(s.Root.Path)
+			}
+			if c.Snapshotter == "" {
+				return errors.New("no snapshotter set for container")
+			}
+			if c.SnapshotKey == "" {
+				return errors.New("rootfs snapshot not created for container")
+			}
+			snapshotter := client.SnapshotService(c.Snapshotter)
+			mounts, err := snapshotter.Mounts(ctx, c.SnapshotKey)
+			if err != nil {
+				return err
+			}
+			return mount.WithTempMount(ctx, mounts, f)
+		default:
+			return fmt.Errorf("invalid USER value %s", userstr)
+		}
+	}
+}
+
+// WithUIDGID allows the UID and GID for the Process to be set
+func WithUIDGID(uid, gid uint32) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setProcess(s)
+		s.Process.User.UID = uid
+		s.Process.User.GID = gid
+		return nil
+	}
+}
+
+// WithUserID sets the correct UID and GID for the container based
+// on the image's /etc/passwd contents. If /etc/passwd does not exist,
+// or uid is not found in /etc/passwd, it sets the requested uid,
+// additionally sets the gid to 0, and does not return an error.
+func WithUserID(uid uint32) SpecOpts {
+	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) (err error) {
+		setProcess(s)
+		if c.Snapshotter == "" && c.SnapshotKey == "" {
+			if !isRootfsAbs(s.Root.Path) {
+				return errors.Errorf("rootfs absolute path is required")
+			}
+			user, err := getUserFromPath(s.Root.Path, func(u user.User) bool {
+				return u.Uid == int(uid)
+			})
+			if err != nil {
+				if os.IsNotExist(err) || err == errNoUsersFound {
+					s.Process.User.UID, s.Process.User.GID = uid, 0
+					return nil
+				}
+				return err
+			}
+			s.Process.User.UID, s.Process.User.GID = uint32(user.Uid), uint32(user.Gid)
+			return nil
+
+		}
+		if c.Snapshotter == "" {
+			return errors.Errorf("no snapshotter set for container")
+		}
+		if c.SnapshotKey == "" {
+			return errors.Errorf("rootfs snapshot not created for container")
+		}
+		snapshotter := client.SnapshotService(c.Snapshotter)
+		mounts, err := snapshotter.Mounts(ctx, c.SnapshotKey)
+		if err != nil {
+			return err
+		}
+		return mount.WithTempMount(ctx, mounts, func(root string) error {
+			user, err := getUserFromPath(root, func(u user.User) bool {
+				return u.Uid == int(uid)
+			})
+			if err != nil {
+				if os.IsNotExist(err) || err == errNoUsersFound {
+					s.Process.User.UID, s.Process.User.GID = uid, 0
+					return nil
+				}
+				return err
+			}
+			s.Process.User.UID, s.Process.User.GID = uint32(user.Uid), uint32(user.Gid)
+			return nil
+		})
+	}
+}
+
+// WithUsername sets the correct UID and GID for the container
+// based on the the image's /etc/passwd contents. If /etc/passwd
+// does not exist, or the username is not found in /etc/passwd,
+// it returns error.
+func WithUsername(username string) SpecOpts {
+	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) (err error) {
+		setProcess(s)
+		if s.Linux != nil {
+			if c.Snapshotter == "" && c.SnapshotKey == "" {
+				if !isRootfsAbs(s.Root.Path) {
+					return errors.Errorf("rootfs absolute path is required")
+				}
+				user, err := getUserFromPath(s.Root.Path, func(u user.User) bool {
+					return u.Name == username
+				})
+				if err != nil {
+					return err
+				}
+				s.Process.User.UID, s.Process.User.GID = uint32(user.Uid), uint32(user.Gid)
+				return nil
+			}
+			if c.Snapshotter == "" {
+				return errors.Errorf("no snapshotter set for container")
+			}
+			if c.SnapshotKey == "" {
+				return errors.Errorf("rootfs snapshot not created for container")
+			}
+			snapshotter := client.SnapshotService(c.Snapshotter)
+			mounts, err := snapshotter.Mounts(ctx, c.SnapshotKey)
+			if err != nil {
+				return err
+			}
+			return mount.WithTempMount(ctx, mounts, func(root string) error {
+				user, err := getUserFromPath(root, func(u user.User) bool {
+					return u.Name == username
+				})
+				if err != nil {
+					return err
+				}
+				s.Process.User.UID, s.Process.User.GID = uint32(user.Uid), uint32(user.Gid)
+				return nil
+			})
+		} else if s.Windows != nil {
+			s.Process.User.Username = username
+		} else {
+			return errors.New("spec does not contain Linux or Windows section")
+		}
+		return nil
+	}
+}
+
+// WithAdditionalGIDs sets the OCI spec's additionalGids array to any additional groups listed
+// for a particular user in the /etc/groups file of the image's root filesystem
+// The passed in user can be either a uid or a username.
+func WithAdditionalGIDs(userstr string) SpecOpts {
+	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) (err error) {
+		setProcess(s)
+		setAdditionalGids := func(root string) error {
+			var username string
+			uid, err := strconv.Atoi(userstr)
+			if err == nil {
+				user, err := getUserFromPath(root, func(u user.User) bool {
+					return u.Uid == uid
+				})
+				if err != nil {
+					if os.IsNotExist(err) || err == errNoUsersFound {
+						return nil
+					}
+					return err
+				}
+				username = user.Name
+			} else {
+				username = userstr
+			}
+			gids, err := getSupplementalGroupsFromPath(root, func(g user.Group) bool {
+				// we only want supplemental groups
+				if g.Name == username {
+					return false
+				}
+				for _, entry := range g.List {
+					if entry == username {
+						return true
+					}
+				}
+				return false
+			})
+			if err != nil {
+				if os.IsNotExist(err) {
+					return nil
+				}
+				return err
+			}
+			s.Process.User.AdditionalGids = gids
+			return nil
+		}
+		if c.Snapshotter == "" && c.SnapshotKey == "" {
+			if !isRootfsAbs(s.Root.Path) {
+				return errors.Errorf("rootfs absolute path is required")
+			}
+			return setAdditionalGids(s.Root.Path)
+		}
+		if c.Snapshotter == "" {
+			return errors.Errorf("no snapshotter set for container")
+		}
+		if c.SnapshotKey == "" {
+			return errors.Errorf("rootfs snapshot not created for container")
+		}
+		snapshotter := client.SnapshotService(c.Snapshotter)
+		mounts, err := snapshotter.Mounts(ctx, c.SnapshotKey)
+		if err != nil {
+			return err
+		}
+		return mount.WithTempMount(ctx, mounts, setAdditionalGids)
+	}
+}
+
+// WithCapabilities sets Linux capabilities on the process
+func WithCapabilities(caps []string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setCapabilities(s)
+
+		s.Process.Capabilities.Bounding = caps
+		s.Process.Capabilities.Effective = caps
+		s.Process.Capabilities.Permitted = caps
+		s.Process.Capabilities.Inheritable = caps
+
+		return nil
+	}
+}
+
+// WithAllCapabilities sets all linux capabilities for the process
+var WithAllCapabilities = WithCapabilities(getAllCapabilities())
+
+func getAllCapabilities() []string {
+	last := capability.CAP_LAST_CAP
+	// hack for RHEL6 which has no /proc/sys/kernel/cap_last_cap
+	if last == capability.Cap(63) {
+		last = capability.CAP_BLOCK_SUSPEND
+	}
+	var caps []string
+	for _, cap := range capability.List() {
+		if cap > last {
+			continue
+		}
+		caps = append(caps, "CAP_"+strings.ToUpper(cap.String()))
+	}
+	return caps
+}
+
+// WithAmbientCapabilities set the Linux ambient capabilities for the process
+// Ambient capabilities should only be set for non-root users or the caller should
+// understand how these capabilities are used and set
+func WithAmbientCapabilities(caps []string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setCapabilities(s)
+
+		s.Process.Capabilities.Ambient = caps
+		return nil
+	}
+}
+
+var errNoUsersFound = errors.New("no users found")
+
+func getUserFromPath(root string, filter func(user.User) bool) (user.User, error) {
+	ppath, err := fs.RootPath(root, "/etc/passwd")
+	if err != nil {
+		return user.User{}, err
+	}
+	users, err := user.ParsePasswdFileFilter(ppath, filter)
+	if err != nil {
+		return user.User{}, err
+	}
+	if len(users) == 0 {
+		return user.User{}, errNoUsersFound
+	}
+	return users[0], nil
+}
+
+var errNoGroupsFound = errors.New("no groups found")
+
+func getGIDFromPath(root string, filter func(user.Group) bool) (gid uint32, err error) {
+	gpath, err := fs.RootPath(root, "/etc/group")
+	if err != nil {
+		return 0, err
+	}
+	groups, err := user.ParseGroupFileFilter(gpath, filter)
+	if err != nil {
+		return 0, err
+	}
+	if len(groups) == 0 {
+		return 0, errNoGroupsFound
+	}
+	g := groups[0]
+	return uint32(g.Gid), nil
+}
+
+func getSupplementalGroupsFromPath(root string, filter func(user.Group) bool) ([]uint32, error) {
+	gpath, err := fs.RootPath(root, "/etc/group")
+	if err != nil {
+		return []uint32{}, err
+	}
+	groups, err := user.ParseGroupFileFilter(gpath, filter)
+	if err != nil {
+		return []uint32{}, err
+	}
+	if len(groups) == 0 {
+		// if there are no additional groups; just return an empty set
+		return []uint32{}, nil
+	}
+	addlGids := []uint32{}
+	for _, grp := range groups {
+		addlGids = append(addlGids, uint32(grp.Gid))
+	}
+	return addlGids, nil
+}
+
+func isRootfsAbs(root string) bool {
+	return filepath.IsAbs(root)
+}
+
+// WithMaskedPaths sets the masked paths option
+func WithMaskedPaths(paths []string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setLinux(s)
+		s.Linux.MaskedPaths = paths
+		return nil
+	}
+}
+
+// WithReadonlyPaths sets the read only paths option
+func WithReadonlyPaths(paths []string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setLinux(s)
+		s.Linux.ReadonlyPaths = paths
+		return nil
+	}
+}
+
+// WithWriteableSysfs makes any sysfs mounts writeable
+func WithWriteableSysfs(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	for i, m := range s.Mounts {
+		if m.Type == "sysfs" {
+			var options []string
+			for _, o := range m.Options {
+				if o == "ro" {
+					o = "rw"
+				}
+				options = append(options, o)
+			}
+			s.Mounts[i].Options = options
+		}
+	}
+	return nil
+}
+
+// WithWriteableCgroupfs makes any cgroup mounts writeable
+func WithWriteableCgroupfs(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	for i, m := range s.Mounts {
+		if m.Type == "cgroup" {
+			var options []string
+			for _, o := range m.Options {
+				if o == "ro" {
+					o = "rw"
+				}
+				options = append(options, o)
+			}
+			s.Mounts[i].Options = options
+		}
+	}
+	return nil
+}
+
+// WithSelinuxLabel sets the process SELinux label
+func WithSelinuxLabel(label string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setProcess(s)
+		s.Process.SelinuxLabel = label
+		return nil
+	}
+}
+
+// WithApparmorProfile sets the Apparmor profile for the process
+func WithApparmorProfile(profile string) SpecOpts {
+	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+		setProcess(s)
+		s.Process.ApparmorProfile = profile
+		return nil
+	}
+}
+
+// WithSeccompUnconfined clears the seccomp profile
+func WithSeccompUnconfined(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	setLinux(s)
+	s.Linux.Seccomp = nil
+	return nil
+}
+
+// WithParentCgroupDevices uses the default cgroup setup to inherit the container's parent cgroup's
+// allowed and denied devices
+func WithParentCgroupDevices(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	setLinux(s)
+	if s.Linux.Resources == nil {
+		s.Linux.Resources = &specs.LinuxResources{}
+	}
+	s.Linux.Resources.Devices = nil
+	return nil
+}
+
+// WithDefaultUnixDevices adds the default devices for unix such as /dev/null, /dev/random to
+// the container's resource cgroup spec
+func WithDefaultUnixDevices(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
+	setLinux(s)
+	if s.Linux.Resources == nil {
+		s.Linux.Resources = &specs.LinuxResources{}
+	}
+	intptr := func(i int64) *int64 {
+		return &i
+	}
+	s.Linux.Resources.Devices = append(s.Linux.Resources.Devices, []specs.LinuxDeviceCgroup{
+		{
+			// "/dev/null",
+			Type:   "c",
+			Major:  intptr(1),
+			Minor:  intptr(3),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			// "/dev/random",
+			Type:   "c",
+			Major:  intptr(1),
+			Minor:  intptr(8),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			// "/dev/full",
+			Type:   "c",
+			Major:  intptr(1),
+			Minor:  intptr(7),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			// "/dev/tty",
+			Type:   "c",
+			Major:  intptr(5),
+			Minor:  intptr(0),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			// "/dev/zero",
+			Type:   "c",
+			Major:  intptr(1),
+			Minor:  intptr(5),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			// "/dev/urandom",
+			Type:   "c",
+			Major:  intptr(1),
+			Minor:  intptr(9),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			// "/dev/console",
+			Type:   "c",
+			Major:  intptr(5),
+			Minor:  intptr(1),
+			Access: rwm,
+			Allow:  true,
+		},
+		// /dev/pts/ - pts namespaces are "coming soon"
+		{
+			Type:   "c",
+			Major:  intptr(136),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			Type:   "c",
+			Major:  intptr(5),
+			Minor:  intptr(2),
+			Access: rwm,
+			Allow:  true,
+		},
+		{
+			// tuntap
+			Type:   "c",
+			Major:  intptr(10),
+			Minor:  intptr(200),
+			Access: rwm,
+			Allow:  true,
+		},
+	}...)
+	return nil
+}
+
+// WithPrivileged sets up options for a privileged container
+// TODO(justincormack) device handling
+var WithPrivileged = Compose(
+	WithAllCapabilities,
+	WithMaskedPaths(nil),
+	WithReadonlyPaths(nil),
+	WithWriteableSysfs,
+	WithWriteableCgroupfs,
+	WithSelinuxLabel(""),
+	WithApparmorProfile(""),
+	WithSeccompUnconfined,
+)

vendor/github.com/containerd/containerd/oci/spec_opts_unix.go

@@ -1,616 +0,0 @@
-// +build !windows
-
-/*
-   Copyright The containerd Authors.
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
-*/
-
-package oci
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"os"
-	"path/filepath"
-	"strconv"
-	"strings"
-
-	"github.com/containerd/containerd/containers"
-	"github.com/containerd/containerd/content"
-	"github.com/containerd/containerd/images"
-	"github.com/containerd/containerd/mount"
-	"github.com/containerd/containerd/namespaces"
-	"github.com/containerd/continuity/fs"
-	"github.com/opencontainers/image-spec/specs-go/v1"
-	"github.com/opencontainers/runc/libcontainer/user"
-	specs "github.com/opencontainers/runtime-spec/specs-go"
-	"github.com/pkg/errors"
-	"github.com/syndtr/gocapability/capability"
-)
-
-// WithTTY sets the information on the spec as well as the environment variables for
-// using a TTY
-func WithTTY(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	setProcess(s)
-	s.Process.Terminal = true
-	s.Process.Env = append(s.Process.Env, "TERM=xterm")
-	return nil
-}
-
-// setRoot sets Root to empty if unset
-func setRoot(s *Spec) {
-	if s.Root == nil {
-		s.Root = &specs.Root{}
-	}
-}
-
-// setLinux sets Linux to empty if unset
-func setLinux(s *Spec) {
-	if s.Linux == nil {
-		s.Linux = &specs.Linux{}
-	}
-}
-
-// setCapabilities sets Linux Capabilities to empty if unset
-func setCapabilities(s *Spec) {
-	setProcess(s)
-	if s.Process.Capabilities == nil {
-		s.Process.Capabilities = &specs.LinuxCapabilities{}
-	}
-}
-
-// WithHostNamespace allows a task to run inside the host's linux namespace
-func WithHostNamespace(ns specs.LinuxNamespaceType) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setLinux(s)
-		for i, n := range s.Linux.Namespaces {
-			if n.Type == ns {
-				s.Linux.Namespaces = append(s.Linux.Namespaces[:i], s.Linux.Namespaces[i+1:]...)
-				return nil
-			}
-		}
-		return nil
-	}
-}
-
-// WithLinuxNamespace uses the passed in namespace for the spec. If a namespace of the same type already exists in the
-// spec, the existing namespace is replaced by the one provided.
-func WithLinuxNamespace(ns specs.LinuxNamespace) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setLinux(s)
-		for i, n := range s.Linux.Namespaces {
-			if n.Type == ns.Type {
-				before := s.Linux.Namespaces[:i]
-				after := s.Linux.Namespaces[i+1:]
-				s.Linux.Namespaces = append(before, ns)
-				s.Linux.Namespaces = append(s.Linux.Namespaces, after...)
-				return nil
-			}
-		}
-		s.Linux.Namespaces = append(s.Linux.Namespaces, ns)
-		return nil
-	}
-}
-
-// WithImageConfig configures the spec to from the configuration of an Image
-func WithImageConfig(image Image) SpecOpts {
-	return WithImageConfigArgs(image, nil)
-}
-
-// WithImageConfigArgs configures the spec to from the configuration of an Image with additional args that
-// replaces the CMD of the image
-func WithImageConfigArgs(image Image, args []string) SpecOpts {
-	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) error {
-		ic, err := image.Config(ctx)
-		if err != nil {
-			return err
-		}
-		var (
-			ociimage v1.Image
-			config   v1.ImageConfig
-		)
-		switch ic.MediaType {
-		case v1.MediaTypeImageConfig, images.MediaTypeDockerSchema2Config:
-			p, err := content.ReadBlob(ctx, image.ContentStore(), ic)
-			if err != nil {
-				return err
-			}
-
-			if err := json.Unmarshal(p, &ociimage); err != nil {
-				return err
-			}
-			config = ociimage.Config
-		default:
-			return fmt.Errorf("unknown image config media type %s", ic.MediaType)
-		}
-
-		setProcess(s)
-		s.Process.Env = append(s.Process.Env, config.Env...)
-		cmd := config.Cmd
-		if len(args) > 0 {
-			cmd = args
-		}
-		s.Process.Args = append(config.Entrypoint, cmd...)
-		cwd := config.WorkingDir
-		if cwd == "" {
-			cwd = "/"
-		}
-		s.Process.Cwd = cwd
-		if config.User != "" {
-			return WithUser(config.User)(ctx, client, c, s)
-		}
-		return nil
-	}
-}
-
-// WithRootFSPath specifies unmanaged rootfs path.
-func WithRootFSPath(path string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setRoot(s)
-		s.Root.Path = path
-		// Entrypoint is not set here (it's up to caller)
-		return nil
-	}
-}
-
-// WithRootFSReadonly sets specs.Root.Readonly to true
-func WithRootFSReadonly() SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setRoot(s)
-		s.Root.Readonly = true
-		return nil
-	}
-}
-
-// WithNoNewPrivileges sets no_new_privileges on the process for the container
-func WithNoNewPrivileges(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	setProcess(s)
-	s.Process.NoNewPrivileges = true
-	return nil
-}
-
-// WithHostHostsFile bind-mounts the host's /etc/hosts into the container as readonly
-func WithHostHostsFile(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	s.Mounts = append(s.Mounts, specs.Mount{
-		Destination: "/etc/hosts",
-		Type:        "bind",
-		Source:      "/etc/hosts",
-		Options:     []string{"rbind", "ro"},
-	})
-	return nil
-}
-
-// WithHostResolvconf bind-mounts the host's /etc/resolv.conf into the container as readonly
-func WithHostResolvconf(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	s.Mounts = append(s.Mounts, specs.Mount{
-		Destination: "/etc/resolv.conf",
-		Type:        "bind",
-		Source:      "/etc/resolv.conf",
-		Options:     []string{"rbind", "ro"},
-	})
-	return nil
-}
-
-// WithHostLocaltime bind-mounts the host's /etc/localtime into the container as readonly
-func WithHostLocaltime(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	s.Mounts = append(s.Mounts, specs.Mount{
-		Destination: "/etc/localtime",
-		Type:        "bind",
-		Source:      "/etc/localtime",
-		Options:     []string{"rbind", "ro"},
-	})
-	return nil
-}
-
-// WithUserNamespace sets the uid and gid mappings for the task
-// this can be called multiple times to add more mappings to the generated spec
-func WithUserNamespace(container, host, size uint32) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		var hasUserns bool
-		setLinux(s)
-		for _, ns := range s.Linux.Namespaces {
-			if ns.Type == specs.UserNamespace {
-				hasUserns = true
-				break
-			}
-		}
-		if !hasUserns {
-			s.Linux.Namespaces = append(s.Linux.Namespaces, specs.LinuxNamespace{
-				Type: specs.UserNamespace,
-			})
-		}
-		mapping := specs.LinuxIDMapping{
-			ContainerID: container,
-			HostID:      host,
-			Size:        size,
-		}
-		s.Linux.UIDMappings = append(s.Linux.UIDMappings, mapping)
-		s.Linux.GIDMappings = append(s.Linux.GIDMappings, mapping)
-		return nil
-	}
-}
-
-// WithCgroup sets the container's cgroup path
-func WithCgroup(path string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setLinux(s)
-		s.Linux.CgroupsPath = path
-		return nil
-	}
-}
-
-// WithNamespacedCgroup uses the namespace set on the context to create a
-// root directory for containers in the cgroup with the id as the subcgroup
-func WithNamespacedCgroup() SpecOpts {
-	return func(ctx context.Context, _ Client, c *containers.Container, s *Spec) error {
-		namespace, err := namespaces.NamespaceRequired(ctx)
-		if err != nil {
-			return err
-		}
-		setLinux(s)
-		s.Linux.CgroupsPath = filepath.Join("/", namespace, c.ID)
-		return nil
-	}
-}
-
-// WithUser sets the user to be used within the container.
-// It accepts a valid user string in OCI Image Spec v1.0.0:
-//   user, uid, user:group, uid:gid, uid:group, user:gid
-func WithUser(userstr string) SpecOpts {
-	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) error {
-		setProcess(s)
-		parts := strings.Split(userstr, ":")
-		switch len(parts) {
-		case 1:
-			v, err := strconv.Atoi(parts[0])
-			if err != nil {
-				// if we cannot parse as a uint they try to see if it is a username
-				return WithUsername(userstr)(ctx, client, c, s)
-			}
-			return WithUserID(uint32(v))(ctx, client, c, s)
-		case 2:
-			var (
-				username  string
-				groupname string
-			)
-			var uid, gid uint32
-			v, err := strconv.Atoi(parts[0])
-			if err != nil {
-				username = parts[0]
-			} else {
-				uid = uint32(v)
-			}
-			if v, err = strconv.Atoi(parts[1]); err != nil {
-				groupname = parts[1]
-			} else {
-				gid = uint32(v)
-			}
-			if username == "" && groupname == "" {
-				s.Process.User.UID, s.Process.User.GID = uid, gid
-				return nil
-			}
-			f := func(root string) error {
-				if username != "" {
-					uid, _, err = getUIDGIDFromPath(root, func(u user.User) bool {
-						return u.Name == username
-					})
-					if err != nil {
-						return err
-					}
-				}
-				if groupname != "" {
-					gid, err = getGIDFromPath(root, func(g user.Group) bool {
-						return g.Name == groupname
-					})
-					if err != nil {
-						return err
-					}
-				}
-				s.Process.User.UID, s.Process.User.GID = uid, gid
-				return nil
-			}
-			if c.Snapshotter == "" && c.SnapshotKey == "" {
-				if !isRootfsAbs(s.Root.Path) {
-					return errors.New("rootfs absolute path is required")
-				}
-				return f(s.Root.Path)
-			}
-			if c.Snapshotter == "" {
-				return errors.New("no snapshotter set for container")
-			}
-			if c.SnapshotKey == "" {
-				return errors.New("rootfs snapshot not created for container")
-			}
-			snapshotter := client.SnapshotService(c.Snapshotter)
-			mounts, err := snapshotter.Mounts(ctx, c.SnapshotKey)
-			if err != nil {
-				return err
-			}
-			return mount.WithTempMount(ctx, mounts, f)
-		default:
-			return fmt.Errorf("invalid USER value %s", userstr)
-		}
-	}
-}
-
-// WithUIDGID allows the UID and GID for the Process to be set
-func WithUIDGID(uid, gid uint32) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setProcess(s)
-		s.Process.User.UID = uid
-		s.Process.User.GID = gid
-		return nil
-	}
-}
-
-// WithUserID sets the correct UID and GID for the container based
-// on the image's /etc/passwd contents. If /etc/passwd does not exist,
-// or uid is not found in /etc/passwd, it sets the requested uid,
-// additionally sets the gid to 0, and does not return an error.
-func WithUserID(uid uint32) SpecOpts {
-	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) (err error) {
-		setProcess(s)
-		if c.Snapshotter == "" && c.SnapshotKey == "" {
-			if !isRootfsAbs(s.Root.Path) {
-				return errors.Errorf("rootfs absolute path is required")
-			}
-			uuid, ugid, err := getUIDGIDFromPath(s.Root.Path, func(u user.User) bool {
-				return u.Uid == int(uid)
-			})
-			if err != nil {
-				if os.IsNotExist(err) || err == errNoUsersFound {
-					s.Process.User.UID, s.Process.User.GID = uid, 0
-					return nil
-				}
-				return err
-			}
-			s.Process.User.UID, s.Process.User.GID = uuid, ugid
-			return nil
-
-		}
-		if c.Snapshotter == "" {
-			return errors.Errorf("no snapshotter set for container")
-		}
-		if c.SnapshotKey == "" {
-			return errors.Errorf("rootfs snapshot not created for container")
-		}
-		snapshotter := client.SnapshotService(c.Snapshotter)
-		mounts, err := snapshotter.Mounts(ctx, c.SnapshotKey)
-		if err != nil {
-			return err
-		}
-		return mount.WithTempMount(ctx, mounts, func(root string) error {
-			uuid, ugid, err := getUIDGIDFromPath(root, func(u user.User) bool {
-				return u.Uid == int(uid)
-			})
-			if err != nil {
-				if os.IsNotExist(err) || err == errNoUsersFound {
-					s.Process.User.UID, s.Process.User.GID = uid, 0
-					return nil
-				}
-				return err
-			}
-			s.Process.User.UID, s.Process.User.GID = uuid, ugid
-			return nil
-		})
-	}
-}
-
-// WithUsername sets the correct UID and GID for the container
-// based on the the image's /etc/passwd contents. If /etc/passwd
-// does not exist, or the username is not found in /etc/passwd,
-// it returns error.
-func WithUsername(username string) SpecOpts {
-	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) (err error) {
-		setProcess(s)
-		if c.Snapshotter == "" && c.SnapshotKey == "" {
-			if !isRootfsAbs(s.Root.Path) {
-				return errors.Errorf("rootfs absolute path is required")
-			}
-			uid, gid, err := getUIDGIDFromPath(s.Root.Path, func(u user.User) bool {
-				return u.Name == username
-			})
-			if err != nil {
-				return err
-			}
-			s.Process.User.UID, s.Process.User.GID = uid, gid
-			return nil
-		}
-		if c.Snapshotter == "" {
-			return errors.Errorf("no snapshotter set for container")
-		}
-		if c.SnapshotKey == "" {
-			return errors.Errorf("rootfs snapshot not created for container")
-		}
-		snapshotter := client.SnapshotService(c.Snapshotter)
-		mounts, err := snapshotter.Mounts(ctx, c.SnapshotKey)
-		if err != nil {
-			return err
-		}
-		return mount.WithTempMount(ctx, mounts, func(root string) error {
-			uid, gid, err := getUIDGIDFromPath(root, func(u user.User) bool {
-				return u.Name == username
-			})
-			if err != nil {
-				return err
-			}
-			s.Process.User.UID, s.Process.User.GID = uid, gid
-			return nil
-		})
-	}
-}
-
-// WithCapabilities sets Linux capabilities on the process
-func WithCapabilities(caps []string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setCapabilities(s)
-
-		s.Process.Capabilities.Bounding = caps
-		s.Process.Capabilities.Effective = caps
-		s.Process.Capabilities.Permitted = caps
-		s.Process.Capabilities.Inheritable = caps
-
-		return nil
-	}
-}
-
-// WithAllCapabilities sets all linux capabilities for the process
-var WithAllCapabilities = WithCapabilities(getAllCapabilities())
-
-func getAllCapabilities() []string {
-	last := capability.CAP_LAST_CAP
-	// hack for RHEL6 which has no /proc/sys/kernel/cap_last_cap
-	if last == capability.Cap(63) {
-		last = capability.CAP_BLOCK_SUSPEND
-	}
-	var caps []string
-	for _, cap := range capability.List() {
-		if cap > last {
-			continue
-		}
-		caps = append(caps, "CAP_"+strings.ToUpper(cap.String()))
-	}
-	return caps
-}
-
-var errNoUsersFound = errors.New("no users found")
-
-func getUIDGIDFromPath(root string, filter func(user.User) bool) (uid, gid uint32, err error) {
-	ppath, err := fs.RootPath(root, "/etc/passwd")
-	if err != nil {
-		return 0, 0, err
-	}
-	users, err := user.ParsePasswdFileFilter(ppath, filter)
-	if err != nil {
-		return 0, 0, err
-	}
-	if len(users) == 0 {
-		return 0, 0, errNoUsersFound
-	}
-	u := users[0]
-	return uint32(u.Uid), uint32(u.Gid), nil
-}
-
-var errNoGroupsFound = errors.New("no groups found")
-
-func getGIDFromPath(root string, filter func(user.Group) bool) (gid uint32, err error) {
-	gpath, err := fs.RootPath(root, "/etc/group")
-	if err != nil {
-		return 0, err
-	}
-	groups, err := user.ParseGroupFileFilter(gpath, filter)
-	if err != nil {
-		return 0, err
-	}
-	if len(groups) == 0 {
-		return 0, errNoGroupsFound
-	}
-	g := groups[0]
-	return uint32(g.Gid), nil
-}
-
-func isRootfsAbs(root string) bool {
-	return filepath.IsAbs(root)
-}
-
-// WithMaskedPaths sets the masked paths option
-func WithMaskedPaths(paths []string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setLinux(s)
-		s.Linux.MaskedPaths = paths
-		return nil
-	}
-}
-
-// WithReadonlyPaths sets the read only paths option
-func WithReadonlyPaths(paths []string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setLinux(s)
-		s.Linux.ReadonlyPaths = paths
-		return nil
-	}
-}
-
-// WithWriteableSysfs makes any sysfs mounts writeable
-func WithWriteableSysfs(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	for i, m := range s.Mounts {
-		if m.Type == "sysfs" {
-			var options []string
-			for _, o := range m.Options {
-				if o == "ro" {
-					o = "rw"
-				}
-				options = append(options, o)
-			}
-			s.Mounts[i].Options = options
-		}
-	}
-	return nil
-}
-
-// WithWriteableCgroupfs makes any cgroup mounts writeable
-func WithWriteableCgroupfs(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	for i, m := range s.Mounts {
-		if m.Type == "cgroup" {
-			var options []string
-			for _, o := range m.Options {
-				if o == "ro" {
-					o = "rw"
-				}
-				options = append(options, o)
-			}
-			s.Mounts[i].Options = options
-		}
-	}
-	return nil
-}
-
-// WithSelinuxLabel sets the process SELinux label
-func WithSelinuxLabel(label string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setProcess(s)
-		s.Process.SelinuxLabel = label
-		return nil
-	}
-}
-
-// WithApparmorProfile sets the Apparmor profile for the process
-func WithApparmorProfile(profile string) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setProcess(s)
-		s.Process.ApparmorProfile = profile
-		return nil
-	}
-}
-
-// WithSeccompUnconfined clears the seccomp profile
-func WithSeccompUnconfined(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-	setLinux(s)
-	s.Linux.Seccomp = nil
-	return nil
-}
-
-// WithPrivileged sets up options for a privileged container
-// TODO(justincormack) device handling
-var WithPrivileged = Compose(
-	WithAllCapabilities,
-	WithMaskedPaths(nil),
-	WithReadonlyPaths(nil),
-	WithWriteableSysfs,
-	WithWriteableCgroupfs,
-	WithSelinuxLabel(""),
-	WithApparmorProfile(""),
-	WithSeccompUnconfined,
-)

vendor/github.com/containerd/containerd/oci/spec_opts_windows.go

@@ -1,89 +0,0 @@
-// +build windows
-
-/*
-   Copyright The containerd Authors.
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
-*/
-
-package oci
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-
-	"github.com/containerd/containerd/containers"
-	"github.com/containerd/containerd/content"
-	"github.com/containerd/containerd/images"
-	"github.com/opencontainers/image-spec/specs-go/v1"
-	specs "github.com/opencontainers/runtime-spec/specs-go"
-)
-
-// WithImageConfig configures the spec to from the configuration of an Image
-func WithImageConfig(image Image) SpecOpts {
-	return func(ctx context.Context, client Client, _ *containers.Container, s *Spec) error {
-		setProcess(s)
-		ic, err := image.Config(ctx)
-		if err != nil {
-			return err
-		}
-		var (
-			ociimage v1.Image
-			config   v1.ImageConfig
-		)
-		switch ic.MediaType {
-		case v1.MediaTypeImageConfig, images.MediaTypeDockerSchema2Config:
-			p, err := content.ReadBlob(ctx, image.ContentStore(), ic)
-			if err != nil {
-				return err
-			}
-			if err := json.Unmarshal(p, &ociimage); err != nil {
-				return err
-			}
-			config = ociimage.Config
-		default:
-			return fmt.Errorf("unknown image config media type %s", ic.MediaType)
-		}
-		s.Process.Env = config.Env
-		s.Process.Args = append(config.Entrypoint, config.Cmd...)
-		s.Process.User = specs.User{
-			Username: config.User,
-		}
-		return nil
-	}
-}
-
-// WithTTY sets the information on the spec as well as the environment variables for
-// using a TTY
-func WithTTY(width, height int) SpecOpts {
-	return func(_ context.Context, _ Client, _ *containers.Container, s *Spec) error {
-		setProcess(s)
-		s.Process.Terminal = true
-		if s.Process.ConsoleSize == nil {
-			s.Process.ConsoleSize = &specs.Box{}
-		}
-		s.Process.ConsoleSize.Width = uint(width)
-		s.Process.ConsoleSize.Height = uint(height)
-		return nil
-	}
-}
-
-// WithUsername sets the username on the process
-func WithUsername(username string) SpecOpts {
-	return func(ctx context.Context, client Client, c *containers.Container, s *Spec) error {
-		setProcess(s)
-		s.Process.User.Username = username
-		return nil
-	}
-}

vendor/github.com/containerd/containerd/oci/spec_unix.go

@@ -1,188 +0,0 @@
-// +build !windows
-
-/*
-   Copyright The containerd Authors.
-
-   Licensed under the Apache License, Version 2.0 (the "License");
-   you may not use this file except in compliance with the License.
-   You may obtain a copy of the License at
-
-       http://www.apache.org/licenses/LICENSE-2.0
-
-   Unless required by applicable law or agreed to in writing, software
-   distributed under the License is distributed on an "AS IS" BASIS,
-   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-   See the License for the specific language governing permissions and
-   limitations under the License.
-*/
-
-package oci
-
-import (
-	"context"
-	"path/filepath"
-
-	"github.com/containerd/containerd/namespaces"
-	specs "github.com/opencontainers/runtime-spec/specs-go"
-)
-
-const (
-	rwm               = "rwm"
-	defaultRootfsPath = "rootfs"
-)
-
-var (
-	defaultEnv = []string{
-		"PATH=/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin",
-	}
-)
-
-func defaultCaps() []string {
-	return []string{
-		"CAP_CHOWN",
-		"CAP_DAC_OVERRIDE",
-		"CAP_FSETID",
-		"CAP_FOWNER",
-		"CAP_MKNOD",
-		"CAP_NET_RAW",
-		"CAP_SETGID",
-		"CAP_SETUID",
-		"CAP_SETFCAP",
-		"CAP_SETPCAP",
-		"CAP_NET_BIND_SERVICE",
-		"CAP_SYS_CHROOT",
-		"CAP_KILL",
-		"CAP_AUDIT_WRITE",
-	}
-}
-
-func defaultNamespaces() []specs.LinuxNamespace {
-	return []specs.LinuxNamespace{
-		{
-			Type: specs.PIDNamespace,
-		},
-		{
-			Type: specs.IPCNamespace,
-		},
-		{
-			Type: specs.UTSNamespace,
-		},
-		{
-			Type: specs.MountNamespace,
-		},
-		{
-			Type: specs.NetworkNamespace,
-		},
-	}
-}
-
-func populateDefaultSpec(ctx context.Context, s *Spec, id string) error {
-	ns, err := namespaces.NamespaceRequired(ctx)
-	if err != nil {
-		return err
-	}
-
-	*s = Spec{
-		Version: specs.Version,
-		Root: &specs.Root{
-			Path: defaultRootfsPath,
-		},
-		Process: &specs.Process{
-			Env:             defaultEnv,
-			Cwd:             "/",
-			NoNewPrivileges: true,
-			User: specs.User{
-				UID: 0,
-				GID: 0,
-			},
-			Capabilities: &specs.LinuxCapabilities{
-				Bounding:    defaultCaps(),
-				Permitted:   defaultCaps(),
-				Inheritable: defaultCaps(),
-				Effective:   defaultCaps(),
-			},
-			Rlimits: []specs.POSIXRlimit{
-				{
-					Type: "RLIMIT_NOFILE",
-					Hard: uint64(1024),
-					Soft: uint64(1024),
-				},
-			},
-		},
-		Mounts: []specs.Mount{
-			{
-				Destination: "/proc",
-				Type:        "proc",
-				Source:      "proc",
-			},
-			{
-				Destination: "/dev",
-				Type:        "tmpfs",
-				Source:      "tmpfs",
-				Options:     []string{"nosuid", "strictatime", "mode=755", "size=65536k"},
-			},
-			{
-				Destination: "/dev/pts",
-				Type:        "devpts",
-				Source:      "devpts",
-				Options:     []string{"nosuid", "noexec", "newinstance", "ptmxmode=0666", "mode=0620", "gid=5"},
-			},
-			{
-				Destination: "/dev/shm",
-				Type:        "tmpfs",
-				Source:      "shm",
-				Options:     []string{"nosuid", "noexec", "nodev", "mode=1777", "size=65536k"},
-			},
-			{
-				Destination: "/dev/mqueue",
-				Type:        "mqueue",
-				Source:      "mqueue",
-				Options:     []string{"nosuid", "noexec", "nodev"},
-			},
-			{
-				Destination: "/sys",
-				Type:        "sysfs",
-				Source:      "sysfs",
-				Options:     []string{"nosuid", "noexec", "nodev", "ro"},
-			},
-			{
-				Destination: "/run",
-				Type:        "tmpfs",
-				Source:      "tmpfs",
-				Options:     []string{"nosuid", "strictatime", "mode=755", "size=65536k"},
-			},
-		},
-		Linux: &specs.Linux{
-			MaskedPaths: []string{
-				"/proc/acpi",
-				"/proc/kcore",
-				"/proc/keys",
-				"/proc/latency_stats",
-				"/proc/timer_list",
-				"/proc/timer_stats",
-				"/proc/sched_debug",
-				"/sys/firmware",
-				"/proc/scsi",
-			},
-			ReadonlyPaths: []string{
-				"/proc/asound",
-				"/proc/bus",
-				"/proc/fs",
-				"/proc/irq",
-				"/proc/sys",
-				"/proc/sysrq-trigger",
-			},
-			CgroupsPath: filepath.Join("/", ns, id),
-			Resources: &specs.LinuxResources{
-				Devices: []specs.LinuxDeviceCgroup{
-					{
-						Allow:  false,
-						Access: rwm,
-					},
-				},
-			},
-			Namespaces: defaultNamespaces(),
-		},
-	}
-	return nil
-}

vendor/github.com/containerd/containerd/platforms/compare.go

@@ -0,0 +1,192 @@
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package platforms
+
+import specs "github.com/opencontainers/image-spec/specs-go/v1"
+
+// MatchComparer is able to match and compare platforms to
+// filter and sort platforms.
+type MatchComparer interface {
+	Matcher
+
+	Less(specs.Platform, specs.Platform) bool
+}
+
+// Only returns a match comparer for a single platform
+// using default resolution logic for the platform.
+//
+// For ARMv7, will also match ARMv6 and ARMv5
+// For ARMv6, will also match ARMv5
+func Only(platform specs.Platform) MatchComparer {
+	platform = Normalize(platform)
+	if platform.Architecture == "arm" {
+		if platform.Variant == "v7" {
+			return orderedPlatformComparer{
+				matchers: []Matcher{
+					&matcher{
+						Platform: platform,
+					},
+					&matcher{
+						Platform: specs.Platform{
+							Architecture: platform.Architecture,
+							OS:           platform.OS,
+							OSVersion:    platform.OSVersion,
+							OSFeatures:   platform.OSFeatures,
+							Variant:      "v6",
+						},
+					},
+					&matcher{
+						Platform: specs.Platform{
+							Architecture: platform.Architecture,
+							OS:           platform.OS,
+							OSVersion:    platform.OSVersion,
+							OSFeatures:   platform.OSFeatures,
+							Variant:      "v5",
+						},
+					},
+				},
+			}
+		}
+		if platform.Variant == "v6" {
+			return orderedPlatformComparer{
+				matchers: []Matcher{
+					&matcher{
+						Platform: platform,
+					},
+					&matcher{
+						Platform: specs.Platform{
+							Architecture: platform.Architecture,
+							OS:           platform.OS,
+							OSVersion:    platform.OSVersion,
+							OSFeatures:   platform.OSFeatures,
+							Variant:      "v5",
+						},
+					},
+				},
+			}
+		}
+	}
+
+	return singlePlatformComparer{
+		Matcher: &matcher{
+			Platform: platform,
+		},
+	}
+}
+
+// Ordered returns a platform MatchComparer which matches any of the platforms
+// but orders them in order they are provided.
+func Ordered(platforms ...specs.Platform) MatchComparer {
+	matchers := make([]Matcher, len(platforms))
+	for i := range platforms {
+		matchers[i] = NewMatcher(platforms[i])
+	}
+	return orderedPlatformComparer{
+		matchers: matchers,
+	}
+}
+
+// Any returns a platform MatchComparer which matches any of the platforms
+// with no preference for ordering.
+func Any(platforms ...specs.Platform) MatchComparer {
+	matchers := make([]Matcher, len(platforms))
+	for i := range platforms {
+		matchers[i] = NewMatcher(platforms[i])
+	}
+	return anyPlatformComparer{
+		matchers: matchers,
+	}
+}
+
+// All is a platform MatchComparer which matches all platforms
+// with preference for ordering.
+var All MatchComparer = allPlatformComparer{}
+
+type singlePlatformComparer struct {
+	Matcher
+}
+
+func (c singlePlatformComparer) Less(p1, p2 specs.Platform) bool {
+	return c.Match(p1) && !c.Match(p2)
+}
+
+type orderedPlatformComparer struct {
+	matchers []Matcher
+}
+
+func (c orderedPlatformComparer) Match(platform specs.Platform) bool {
+	for _, m := range c.matchers {
+		if m.Match(platform) {
+			return true
+		}
+	}
+	return false
+}
+
+func (c orderedPlatformComparer) Less(p1 specs.Platform, p2 specs.Platform) bool {
+	for _, m := range c.matchers {
+		p1m := m.Match(p1)
+		p2m := m.Match(p2)
+		if p1m && !p2m {
+			return true
+		}
+		if p1m || p2m {
+			return false
+		}
+	}
+	return false
+}
+
+type anyPlatformComparer struct {
+	matchers []Matcher
+}
+
+func (c anyPlatformComparer) Match(platform specs.Platform) bool {
+	for _, m := range c.matchers {
+		if m.Match(platform) {
+			return true
+		}
+	}
+	return false
+}
+
+func (c anyPlatformComparer) Less(p1, p2 specs.Platform) bool {
+	var p1m, p2m bool
+	for _, m := range c.matchers {
+		if !p1m && m.Match(p1) {
+			p1m = true
+		}
+		if !p2m && m.Match(p2) {
+			p2m = true
+		}
+		if p1m && p2m {
+			return false
+		}
+	}
+	// If one matches, and the other does, sort match first
+	return p1m && !p2m
+}
+
+type allPlatformComparer struct{}
+
+func (allPlatformComparer) Match(specs.Platform) bool {
+	return true
+}
+
+func (allPlatformComparer) Less(specs.Platform, specs.Platform) bool {
+	return false
+}

vendor/github.com/containerd/containerd/platforms/defaults.go

@@ -22,8 +22,8 @@ import (
 	specs "github.com/opencontainers/image-spec/specs-go/v1"
 )
 
-// Default returns the default specifier for the platform.
-func Default() string {
+// DefaultString returns the default string specifier for the platform.
+func DefaultString() string {
 	return Format(DefaultSpec())
 }
 

vendor/github.com/containerd/containerd/platforms/defaults_unix.go

@@ -0,0 +1,24 @@
+// +build !windows
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package platforms
+
+// Default returns the default matcher for the platform.
+func Default() MatchComparer {
+	return Only(DefaultSpec())
+}

vendor/github.com/containerd/containerd/platforms/defaults_windows.go

@@ -1,3 +1,5 @@
+// +build windows
+
 /*
    Copyright The containerd Authors.
 
@@ -14,31 +16,16 @@
    limitations under the License.
 */
 
-package oci
+package platforms
 
 import (
-	"context"
-
-	specs "github.com/opencontainers/runtime-spec/specs-go"
+	specs "github.com/opencontainers/image-spec/specs-go/v1"
 )
 
-func populateDefaultSpec(ctx context.Context, s *Spec, id string) error {
-	*s = Spec{
-		Version: specs.Version,
-		Root:    &specs.Root{},
-		Process: &specs.Process{
-			Cwd: `C:\`,
-			ConsoleSize: &specs.Box{
-				Width:  80,
-				Height: 20,
-			},
-		},
-		Windows: &specs.Windows{
-			IgnoreFlushesDuringBoot: true,
-			Network: &specs.WindowsNetwork{
-				AllowUnqualifiedDNSQuery: true,
-			},
-		},
-	}
-	return nil
+// Default returns the default matcher for the platform.
+func Default() MatchComparer {
+	return Ordered(DefaultSpec(), specs.Platform{
+		OS:           "linux",
+		Architecture: "amd64",
+	})
 }

vendor/github.com/containerd/containerd/platforms/platforms.go

@@ -109,6 +109,7 @@ package platforms
 import (
 	"regexp"
 	"runtime"
+	"strconv"
 	"strings"
 
 	"github.com/containerd/containerd/errdefs"
@@ -230,6 +231,16 @@ func Parse(specifier string) (specs.Platform, error) {
 	return specs.Platform{}, errors.Wrapf(errdefs.ErrInvalidArgument, "%q: cannot parse platform specifier", specifier)
 }
 
+// MustParse is like Parses but panics if the specifier cannot be parsed.
+// Simplifies initialization of global variables.
+func MustParse(specifier string) specs.Platform {
+	p, err := Parse(specifier)
+	if err != nil {
+		panic("platform: Parse(" + strconv.Quote(specifier) + "): " + err.Error())
+	}
+	return p
+}
+
 // Format returns a string specifier from the provided platform specification.
 func Format(platform specs.Platform) string {
 	if platform.OS == "" {

vendor/github.com/containerd/containerd/remotes/docker/fetcher.go

@@ -117,7 +117,7 @@ func (r dockerFetcher) open(ctx context.Context, u, mediatype string, offset int
 			}
 		} else {
 			// TODO: Should any cases where use of content range
-			// without the proper header be considerd?
+			// without the proper header be considered?
 			// 206 responses?
 
 			// Discard up to offset

vendor/github.com/containerd/containerd/remotes/docker/httpreadseeker.go

@@ -134,7 +134,7 @@ func (hrs *httpReadSeeker) reader() (io.Reader, error) {
 		// There is an edge case here where offset == size of the content. If
 		// we seek, we will probably get an error for content that cannot be
 		// sought (?). In that case, we should err on committing the content,
-		// as the length is already satisified but we just return the empty
+		// as the length is already satisfied but we just return the empty
 		// reader instead.
 
 		hrs.rc = ioutil.NopCloser(bytes.NewReader([]byte{}))

vendor/github.com/containerd/containerd/remotes/docker/schema1/converter.go

@@ -272,8 +272,14 @@ func (c *Converter) fetchBlob(ctx context.Context, desc ocispec.Descriptor) erro
 			return err
 		}
 
-		// TODO: Check if blob -> diff id mapping already exists
-		// TODO: Check if blob empty label exists
+		reuse, err := c.reuseLabelBlobState(ctx, desc)
+		if err != nil {
+			return err
+		}
+
+		if reuse {
+			return nil
+		}
 
 		ra, err := c.contentStore.ReaderAt(ctx, desc)
 		if err != nil {
@@ -343,6 +349,17 @@ func (c *Converter) fetchBlob(ctx context.Context, desc ocispec.Descriptor) erro
 
 	state := calc.State()
 
+	cinfo := content.Info{
+		Digest: desc.Digest,
+		Labels: map[string]string{
+			"containerd.io/uncompressed": state.diffID.String(),
+		},
+	}
+
+	if _, err := c.contentStore.Update(ctx, cinfo, "labels.containerd.io/uncompressed"); err != nil {
+		return errors.Wrap(err, "failed to update uncompressed label")
+	}
+
 	c.mu.Lock()
 	c.blobMap[desc.Digest] = state
 	c.layerBlobs[state.diffID] = desc
@@ -351,6 +368,40 @@ func (c *Converter) fetchBlob(ctx context.Context, desc ocispec.Descriptor) erro
 	return nil
 }
 
+func (c *Converter) reuseLabelBlobState(ctx context.Context, desc ocispec.Descriptor) (bool, error) {
+	cinfo, err := c.contentStore.Info(ctx, desc.Digest)
+	if err != nil {
+		return false, errors.Wrap(err, "failed to get blob info")
+	}
+	desc.Size = cinfo.Size
+
+	diffID, ok := cinfo.Labels["containerd.io/uncompressed"]
+	if !ok {
+		return false, nil
+	}
+
+	bState := blobState{empty: false}
+
+	if bState.diffID, err = digest.Parse(diffID); err != nil {
+		log.G(ctx).WithField("id", desc.Digest).Warnf("failed to parse digest from label containerd.io/uncompressed: %v", diffID)
+		return false, nil
+	}
+
+	// NOTE: there is no need to read header to get compression method
+	// because there are only two kinds of methods.
+	if bState.diffID == desc.Digest {
+		desc.MediaType = images.MediaTypeDockerSchema2Layer
+	} else {
+		desc.MediaType = images.MediaTypeDockerSchema2LayerGzip
+	}
+
+	c.mu.Lock()
+	c.blobMap[desc.Digest] = bState
+	c.layerBlobs[bState.diffID] = desc
+	c.mu.Unlock()
+	return true, nil
+}
+
 func (c *Converter) schema1ManifestHistory() ([]ocispec.History, []digest.Digest, error) {
 	if c.pulledManifest == nil {
 		return nil, nil, errors.New("missing schema 1 manifest for conversion")

vendor/github.com/containerd/containerd/remotes/handlers.go

@@ -27,6 +27,7 @@ import (
 	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/images"
 	"github.com/containerd/containerd/log"
+	"github.com/containerd/containerd/platforms"
 	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
@@ -155,7 +156,7 @@ func push(ctx context.Context, provider content.Provider, pusher Pusher, desc oc
 //
 // Base handlers can be provided which will be called before any push specific
 // handlers.
-func PushContent(ctx context.Context, pusher Pusher, desc ocispec.Descriptor, provider content.Provider, platforms []string, baseHandlers ...images.Handler) error {
+func PushContent(ctx context.Context, pusher Pusher, desc ocispec.Descriptor, provider content.Provider, platform platforms.MatchComparer, baseHandlers ...images.Handler) error {
 	var m sync.Mutex
 	manifestStack := []ocispec.Descriptor{}
 
@@ -175,7 +176,7 @@ func PushContent(ctx context.Context, pusher Pusher, desc ocispec.Descriptor, pr
 	pushHandler := PushHandler(pusher, provider)
 
 	handlers := append(baseHandlers,
-		images.FilterPlatforms(images.ChildrenHandler(provider), platforms...),
+		images.FilterPlatforms(images.ChildrenHandler(provider), platform),
 		filterHandler,
 		pushHandler,
 	)

vendor/github.com/containerd/containerd/runtime/v1/linux/proc/exec.go

@@ -147,7 +147,7 @@ func (e *execProcess) start(ctx context.Context) (err error) {
 			return errors.Wrap(err, "creating new NULL IO")
 		}
 	} else {
-		if e.io, err = runc.NewPipeIO(e.parent.IoUID, e.parent.IoGID); err != nil {
+		if e.io, err = runc.NewPipeIO(e.parent.IoUID, e.parent.IoGID, withConditionalIO(e.stdio)); err != nil {
 			return errors.Wrap(err, "failed to create runc io pipes")
 		}
 	}
@@ -164,10 +164,7 @@ func (e *execProcess) start(ctx context.Context) (err error) {
 		return e.parent.runtimeError(err, "OCI runtime exec failed")
 	}
 	if e.stdio.Stdin != "" {
-		fifoCtx, cancel := context.WithTimeout(ctx, 15*time.Second)
-		defer cancel()
-
-		sc, err := fifo.OpenFifo(fifoCtx, e.stdio.Stdin, syscall.O_WRONLY|syscall.O_NONBLOCK, 0)
+		sc, err := fifo.OpenFifo(ctx, e.stdio.Stdin, syscall.O_WRONLY|syscall.O_NONBLOCK, 0)
 		if err != nil {
 			return errors.Wrapf(err, "failed to open stdin fifo %s", e.stdio.Stdin)
 		}
@@ -184,10 +181,7 @@ func (e *execProcess) start(ctx context.Context) (err error) {
 			return errors.Wrap(err, "failed to start console copy")
 		}
 	} else if !e.stdio.IsNull() {
-		fifoCtx, cancel := context.WithTimeout(ctx, 15*time.Second)
-		defer cancel()
-
-		if err := copyPipes(fifoCtx, e.io, e.stdio.Stdin, e.stdio.Stdout, e.stdio.Stderr, &e.wg, &copyWaitGroup); err != nil {
+		if err := copyPipes(ctx, e.io, e.stdio.Stdin, e.stdio.Stdout, e.stdio.Stderr, &e.wg, &copyWaitGroup); err != nil {
 			return errors.Wrap(err, "failed to start io pipe copy")
 		}
 	}

vendor/github.com/containerd/containerd/runtime/v1/linux/proc/exec_state.go

@@ -60,11 +60,11 @@ func (s *execCreatedState) Start(ctx context.Context) error {
 }
 
 func (s *execCreatedState) Delete(ctx context.Context) error {
-	s.p.mu.Lock()
-	defer s.p.mu.Unlock()
 	if err := s.p.delete(ctx); err != nil {
 		return err
 	}
+	s.p.mu.Lock()
+	defer s.p.mu.Unlock()
 	return s.transition("deleted")
 }
 
@@ -168,11 +168,11 @@ func (s *execStoppedState) Start(ctx context.Context) error {
 }
 
 func (s *execStoppedState) Delete(ctx context.Context) error {
-	s.p.mu.Lock()
-	defer s.p.mu.Unlock()
 	if err := s.p.delete(ctx); err != nil {
 		return err
 	}
+	s.p.mu.Lock()
+	defer s.p.mu.Unlock()
 	return s.transition("deleted")
 }
 

vendor/github.com/containerd/containerd/runtime/v1/linux/proc/init.go

@@ -123,7 +123,7 @@ func (p *Init) Create(ctx context.Context, r *CreateConfig) error {
 			return errors.Wrap(err, "creating new NULL IO")
 		}
 	} else {
-		if p.io, err = runc.NewPipeIO(p.IoUID, p.IoGID); err != nil {
+		if p.io, err = runc.NewPipeIO(p.IoUID, p.IoGID, withConditionalIO(p.stdio)); err != nil {
 			return errors.Wrap(err, "failed to create OCI runtime io pipes")
 		}
 	}
@@ -228,7 +228,7 @@ func (p *Init) Status(ctx context.Context) (string, error) {
 	defer p.mu.Unlock()
 	c, err := p.runtime.State(ctx, p.id)
 	if err != nil {
-		if os.IsNotExist(err) {
+		if strings.Contains(err.Error(), "does not exist") {
 			return "stopped", nil
 		}
 		return "", p.runtimeError(err, "OCI runtime state failed")
@@ -249,7 +249,6 @@ func (p *Init) setExited(status int) {
 }
 
 func (p *Init) delete(context context.Context) error {
-	p.KillAll(context)
 	p.wg.Wait()
 	err := p.runtime.Delete(context, p.id, nil)
 	// ignore errors if a runtime has already deleted the process
@@ -400,3 +399,11 @@ func (p *Init) runtimeError(rErr error, msg string) error {
 		return errors.Errorf("%s: %s", msg, rMsg)
 	}
 }
+
+func withConditionalIO(c proc.Stdio) runc.IOOpt {
+	return func(o *runc.IOOption) {
+		o.OpenStdin = c.Stdin != ""
+		o.OpenStdout = c.Stdout != ""
+		o.OpenStderr = c.Stderr != ""
+	}
+}

vendor/github.com/containerd/containerd/runtime/v1/linux/proc/io.go

@@ -109,10 +109,9 @@ func copyPipes(ctx context.Context, rio runc.IO, stdin, stdout, stderr string, w
 		i.dest(fw, fr)
 	}
 	if stdin == "" {
-		rio.Stdin().Close()
 		return nil
 	}
-	f, err := fifo.OpenFifo(ctx, stdin, syscall.O_RDONLY, 0)
+	f, err := fifo.OpenFifo(ctx, stdin, syscall.O_RDONLY|syscall.O_NONBLOCK, 0)
 	if err != nil {
 		return fmt.Errorf("containerd-shim: opening %s failed: %s", stdin, err)
 	}

vendor/github.com/containerd/containerd/runtime/v1/linux/runtime.go

@@ -204,7 +204,7 @@ func (r *Runtime) Create(ctx context.Context, id string, opts runtime.CreateOpts
 				log.G(ctx).WithError(err).WithFields(logrus.Fields{
 					"id":        id,
 					"namespace": namespace,
-				}).Warn("failed to clen up after killed shim")
+				}).Warn("failed to clean up after killed shim")
 			}
 		}
 		shimopt = ShimRemote(r.config, r.address, cgroup, exitHandler)
@@ -248,8 +248,7 @@ func (r *Runtime) Create(ctx context.Context, id string, opts runtime.CreateOpts
 	if err != nil {
 		return nil, errdefs.FromGRPC(err)
 	}
-	t, err := newTask(id, namespace, int(cr.Pid), s, r.events,
-		proc.NewRunc(ropts.RuntimeRoot, sopts.Bundle, namespace, rt, ropts.CriuPath, ropts.SystemdCgroup), r.tasks, bundle)
+	t, err := newTask(id, namespace, int(cr.Pid), s, r.events, r.tasks, bundle)
 	if err != nil {
 		return nil, err
 	}
@@ -341,15 +340,8 @@ func (r *Runtime) loadTasks(ctx context.Context, ns string) ([]*Task, error) {
 			}
 			continue
 		}
-		ropts, err := r.getRuncOptions(ctx, id)
-		if err != nil {
-			log.G(ctx).WithError(err).WithField("id", id).
-				Error("get runtime options")
-			continue
-		}
 
-		t, err := newTask(id, ns, pid, s, r.events,
-			proc.NewRunc(ropts.RuntimeRoot, bundle.path, ns, ropts.Runtime, ropts.CriuPath, ropts.SystemdCgroup), r.tasks, bundle)
+		t, err := newTask(id, ns, pid, s, r.events, r.tasks, bundle)
 		if err != nil {
 			log.G(ctx).WithError(err).Error("loading task type")
 			continue

vendor/github.com/containerd/containerd/runtime/v1/linux/task.go

@@ -31,8 +31,7 @@ import (
 	"github.com/containerd/containerd/log"
 	"github.com/containerd/containerd/runtime"
 	"github.com/containerd/containerd/runtime/v1/shim/client"
-	shim "github.com/containerd/containerd/runtime/v1/shim/v1"
-	runc "github.com/containerd/go-runc"
+	"github.com/containerd/containerd/runtime/v1/shim/v1"
 	"github.com/containerd/ttrpc"
 	"github.com/containerd/typeurl"
 	"github.com/gogo/protobuf/types"
@@ -52,7 +51,7 @@ type Task struct {
 	bundle    *bundle
 }
 
-func newTask(id, namespace string, pid int, shim *client.Client, events *exchange.Exchange, runtime *runc.Runc, list *runtime.TaskList, bundle *bundle) (*Task, error) {
+func newTask(id, namespace string, pid int, shim *client.Client, events *exchange.Exchange, list *runtime.TaskList, bundle *bundle) (*Task, error) {
 	var (
 		err error
 		cg  cgroups.Cgroup

vendor/github.com/containerd/containerd/runtime/v1/shim/service.go

@@ -20,7 +20,9 @@ package shim
 
 import (
 	"context"
+	"encoding/json"
 	"fmt"
+	"io/ioutil"
 	"os"
 	"path/filepath"
 	"sync"
@@ -41,6 +43,7 @@ import (
 	runc "github.com/containerd/go-runc"
 	"github.com/containerd/typeurl"
 	ptypes "github.com/gogo/protobuf/types"
+	specs "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
 	"google.golang.org/grpc/codes"
@@ -221,19 +224,21 @@ func (s *Service) Delete(ctx context.Context, r *ptypes.Empty) (*shimapi.DeleteR
 
 // DeleteProcess deletes an exec'd process
 func (s *Service) DeleteProcess(ctx context.Context, r *shimapi.DeleteProcessRequest) (*shimapi.DeleteResponse, error) {
-	s.mu.Lock()
-	defer s.mu.Unlock()
 	if r.ID == s.id {
 		return nil, status.Errorf(codes.InvalidArgument, "cannot delete init process with DeleteProcess")
 	}
+	s.mu.Lock()
 	p := s.processes[r.ID]
+	s.mu.Unlock()
 	if p == nil {
 		return nil, errors.Wrapf(errdefs.ErrNotFound, "process %s", r.ID)
 	}
 	if err := p.Delete(ctx); err != nil {
 		return nil, err
 	}
+	s.mu.Lock()
 	delete(s.processes, r.ID)
+	s.mu.Unlock()
 	return &shimapi.DeleteResponse{
 		ExitStatus: uint32(p.ExitStatus()),
 		ExitedAt:   p.ExitedAt(),
@@ -507,13 +512,22 @@ func (s *Service) processExits() {
 func (s *Service) checkProcesses(e runc.Exit) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
+
+	shouldKillAll, err := shouldKillAllOnExit(s.bundle)
+	if err != nil {
+		log.G(s.context).WithError(err).Error("failed to check shouldKillAll")
+	}
+
 	for _, p := range s.processes {
 		if p.Pid() == e.Pid {
-			if ip, ok := p.(*proc.Init); ok {
-				// Ensure all children are killed
-				if err := ip.KillAll(s.context); err != nil {
-					log.G(s.context).WithError(err).WithField("id", ip.ID()).
-						Error("failed to kill init's children")
+
+			if shouldKillAll {
+				if ip, ok := p.(*proc.Init); ok {
+					// Ensure all children are killed
+					if err := ip.KillAll(s.context); err != nil {
+						log.G(s.context).WithError(err).WithField("id", ip.ID()).
+							Error("failed to kill init's children")
+					}
 				}
 			}
 			p.SetExited(e.Status)
@@ -529,6 +543,25 @@ func (s *Service) checkProcesses(e runc.Exit) {
 	}
 }
 
+func shouldKillAllOnExit(bundlePath string) (bool, error) {
+	var bundleSpec specs.Spec
+	bundleConfigContents, err := ioutil.ReadFile(filepath.Join(bundlePath, "config.json"))
+	if err != nil {
+		return false, err
+	}
+	json.Unmarshal(bundleConfigContents, &bundleSpec)
+
+	if bundleSpec.Linux != nil {
+		for _, ns := range bundleSpec.Linux.Namespaces {
+			if ns.Type == specs.PIDNamespace {
+				return false, nil
+			}
+		}
+	}
+
+	return true, nil
+}
+
 func (s *Service) getContainerPids(ctx context.Context, id string) ([]uint32, error) {
 	s.mu.Lock()
 	defer s.mu.Unlock()

vendor/github.com/containerd/containerd/runtime/v2/binary.go

@@ -73,7 +73,7 @@ func (b *binary) Start(ctx context.Context) (_ *shim, err error) {
 		}
 	}()
 	// open the log pipe and block until the writer is ready
-	// this helps with syncronization of the shim
+	// this helps with synchronization of the shim
 	// copy the shim's logs to containerd's output
 	go func() {
 		defer f.Close()

vendor/github.com/containerd/containerd/runtime/v2/bundle.go

@@ -58,16 +58,7 @@ func NewBundle(ctx context.Context, root, state, id string, spec []byte) (b *Bun
 		Path:      filepath.Join(state, ns, id),
 		Namespace: ns,
 	}
-	paths := []string{b.Path, work}
-	// create base directories
-	for _, d := range paths {
-		if err := os.MkdirAll(filepath.Dir(d), 0711); err != nil {
-			return nil, err
-		}
-		if err := os.Mkdir(d, 0711); err != nil {
-			return nil, err
-		}
-	}
+	var paths []string
 	defer func() {
 		if err != nil {
 			for _, d := range paths {
@@ -75,6 +66,28 @@ func NewBundle(ctx context.Context, root, state, id string, spec []byte) (b *Bun
 			}
 		}
 	}()
+	// create state directory for the bundle
+	if err := os.MkdirAll(filepath.Dir(b.Path), 0711); err != nil {
+		return nil, err
+	}
+	if err := os.Mkdir(b.Path, 0711); err != nil {
+		return nil, err
+	}
+	paths = append(paths, b.Path)
+	// create working directory for the bundle
+	if err := os.MkdirAll(filepath.Dir(work), 0711); err != nil {
+		return nil, err
+	}
+	if err := os.Mkdir(work, 0711); err != nil {
+		if !os.IsExist(err) {
+			return nil, err
+		}
+		os.RemoveAll(work)
+		if err := os.Mkdir(work, 0711); err != nil {
+			return nil, err
+		}
+	}
+	paths = append(paths, work)
 	// create rootfs dir
 	if err := os.Mkdir(filepath.Join(b.Path, "rootfs"), 0711); err != nil {
 		return nil, err

vendor/github.com/containerd/containerd/runtime/v2/manager.go

@@ -30,10 +30,8 @@ import (
 	"github.com/containerd/containerd/metadata"
 	"github.com/containerd/containerd/mount"
 	"github.com/containerd/containerd/namespaces"
-	"github.com/containerd/containerd/platforms"
 	"github.com/containerd/containerd/plugin"
 	"github.com/containerd/containerd/runtime"
-	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 )
 
 func init() {
@@ -44,7 +42,7 @@ func init() {
 			plugin.MetadataPlugin,
 		},
 		InitFn: func(ic *plugin.InitContext) (interface{}, error) {
-			ic.Meta.Platforms = []ocispec.Platform{platforms.DefaultSpec()}
+			ic.Meta.Platforms = supportedPlatforms()
 			if err := os.MkdirAll(ic.Root, 0711); err != nil {
 				return nil, err
 			}

vendor/github.com/containerd/containerd/runtime/v2/manager_unix.go

@@ -0,0 +1,28 @@
+// +build !windows
+
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package v2
+
+import (
+	"github.com/containerd/containerd/platforms"
+	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
+)
+
+func supportedPlatforms() []ocispec.Platform {
+	return []ocispec.Platform{platforms.DefaultSpec()}
+}

vendor/github.com/containerd/containerd/runtime/v2/manager_windows.go

@@ -1,3 +1,5 @@
+// +build windows
+
 /*
    Copyright The containerd Authors.
 
@@ -14,18 +16,19 @@
    limitations under the License.
 */
 
-package containerd
+package v2
 
 import (
-	"context"
-
-	specs "github.com/opencontainers/runtime-spec/specs-go"
+	"github.com/containerd/containerd/platforms"
+	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 )
 
-// WithResources sets the provided resources on the spec for task updates
-func WithResources(resources *specs.WindowsResources) UpdateTaskOpts {
-	return func(ctx context.Context, client *Client, r *UpdateTaskInfo) error {
-		r.Resources = resources
-		return nil
+func supportedPlatforms() []ocispec.Platform {
+	return []ocispec.Platform{
+		platforms.DefaultSpec(),
+		{
+			OS:           "linux",
+			Architecture: "amd64",
+		},
 	}
 }

vendor/github.com/containerd/containerd/runtime/v2/shim.go

@@ -66,7 +66,7 @@ func loadShim(ctx context.Context, bundle *Bundle, events *exchange.Exchange, rt
 		}
 	}()
 	// open the log pipe and block until the writer is ready
-	// this helps with syncronization of the shim
+	// this helps with synchronization of the shim
 	// copy the shim's logs to containerd's output
 	go func() {
 		defer f.Close()

vendor/github.com/containerd/containerd/runtime/v2/shim/shim_windows.go

@@ -27,6 +27,7 @@ import (
 	"os"
 	"os/exec"
 	"sync"
+	"unsafe"
 
 	winio "github.com/Microsoft/go-winio"
 	"github.com/containerd/containerd/events"
@@ -35,6 +36,7 @@ import (
 	"github.com/containerd/typeurl"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/windows"
 )
 
 // setupSignals creates a new signal handler for all signals
@@ -51,8 +53,43 @@ func subreaper() error {
 	return nil
 }
 
+type fakeSignal struct {
+}
+
+func (fs *fakeSignal) String() string {
+	return ""
+}
+
+func (fs *fakeSignal) Signal() {
+}
+
 func setupDumpStacks(dump chan<- os.Signal) {
-	// TODO: JTERRY75: Make this based on events. signal.Notify(dump, syscall.SIGUSR1)
+	// Windows does not support signals like *nix systems. So instead of
+	// trapping on SIGUSR1 to dump stacks, we wait on a Win32 event to be
+	// signaled. ACL'd to builtin administrators and local system
+	event := "Global\\containerd-shim-runhcs-v1-" + fmt.Sprint(os.Getpid())
+	ev, _ := windows.UTF16PtrFromString(event)
+	sd, err := winio.SddlToSecurityDescriptor("D:P(A;;GA;;;BA)(A;;GA;;;SY)")
+	if err != nil {
+		logrus.Errorf("failed to get security descriptor for debug stackdump event %s: %s", event, err.Error())
+		return
+	}
+	var sa windows.SecurityAttributes
+	sa.Length = uint32(unsafe.Sizeof(sa))
+	sa.InheritHandle = 1
+	sa.SecurityDescriptor = uintptr(unsafe.Pointer(&sd[0]))
+	h, err := windows.CreateEvent(&sa, 0, 0, ev)
+	if h == 0 || err != nil {
+		logrus.Errorf("failed to create debug stackdump event %s: %s", event, err.Error())
+		return
+	}
+	go func() {
+		logrus.Debugf("Stackdump - waiting signal at %s", event)
+		for {
+			windows.WaitForSingleObject(h, windows.INFINITE)
+			dump <- new(fakeSignal)
+		}
+	}()
 }
 
 // serve serves the ttrpc API over a unix socket at the provided path

vendor/github.com/containerd/containerd/services/content/service.go

@@ -70,13 +70,13 @@ func init() {
 			if err != nil {
 				return nil, err
 			}
-			return newService(cs.(content.Store)), nil
+			return NewService(cs.(content.Store)), nil
 		},
 	})
 }
 
-// newService returns the content GRPC server
-func newService(cs content.Store) api.ContentServer {
+// NewService returns the content GRPC server
+func NewService(cs content.Store) api.ContentServer {
 	return &service{store: cs}
 }
 

vendor/github.com/containerd/containerd/services/diff/service_windows.go

@@ -19,5 +19,5 @@
 package diff
 
 var defaultDifferConfig = &config{
-	Order: []string{"windows"},
+	Order: []string{"windows", "windows-lcow"},
 }

vendor/github.com/containerd/containerd/snapshots/overlay/check.go

@@ -71,7 +71,7 @@ func supportsMultipleLowerDir(d string) error {
 }
 
 // Supported returns nil when the overlayfs is functional on the system with the root directory.
-// Suppported is not called during plugin initialization, but exposed for downstream projects which uses
+// Supported is not called during plugin initialization, but exposed for downstream projects which uses
 // this snapshotter as a library.
 func Supported(root string) error {
 	if err := os.MkdirAll(root, 0700); err != nil {

vendor/github.com/containerd/containerd/snapshots/overlay/overlay.go

@@ -168,7 +168,7 @@ func (o *snapshotter) Usage(ctx context.Context, key string) (snapshots.Usage, e
 	upperPath := o.upperPath(id)
 
 	if info.Kind == snapshots.KindActive {
-		du, err := fs.DiskUsage(upperPath)
+		du, err := fs.DiskUsage(ctx, upperPath)
 		if err != nil {
 			// TODO(stevvooe): Consider not reporting an error in this case.
 			return snapshots.Usage{}, err
@@ -225,7 +225,7 @@ func (o *snapshotter) Commit(ctx context.Context, name, key string, opts ...snap
 		return err
 	}
 
-	usage, err := fs.DiskUsage(o.upperPath(id))
+	usage, err := fs.DiskUsage(ctx, o.upperPath(id))
 	if err != nil {
 		return err
 	}

vendor/github.com/containerd/containerd/sys/mount_linux.go

@@ -0,0 +1,119 @@
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package sys
+
+import (
+	"runtime"
+	"syscall"
+	"unsafe"
+
+	"github.com/pkg/errors"
+	"golang.org/x/sys/unix"
+)
+
+// FMountat performs mount from the provided directory.
+func FMountat(dirfd uintptr, source, target, fstype string, flags uintptr, data string) error {
+	var (
+		sourceP, targetP, fstypeP, dataP *byte
+		pid                              uintptr
+		ws                               unix.WaitStatus
+		err                              error
+		errno                            syscall.Errno
+	)
+
+	sourceP, err = syscall.BytePtrFromString(source)
+	if err != nil {
+		return err
+	}
+
+	targetP, err = syscall.BytePtrFromString(target)
+	if err != nil {
+		return err
+	}
+
+	fstypeP, err = syscall.BytePtrFromString(fstype)
+	if err != nil {
+		return err
+	}
+
+	if data != "" {
+		dataP, err = syscall.BytePtrFromString(data)
+		if err != nil {
+			return err
+		}
+	}
+
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	pid, errno = forkAndMountat(dirfd,
+		uintptr(unsafe.Pointer(sourceP)),
+		uintptr(unsafe.Pointer(targetP)),
+		uintptr(unsafe.Pointer(fstypeP)),
+		flags,
+		uintptr(unsafe.Pointer(dataP)))
+
+	if errno != 0 {
+		return errors.Wrap(errno, "failed to fork thread")
+	}
+
+	_, err = unix.Wait4(int(pid), &ws, 0, nil)
+	for err == syscall.EINTR {
+		_, err = unix.Wait4(int(pid), &ws, 0, nil)
+	}
+
+	if err != nil {
+		return errors.Wrapf(err, "failed to find pid=%d process", pid)
+	}
+
+	errno = syscall.Errno(ws.ExitStatus())
+	if errno != 0 {
+		return errors.Wrap(errno, "failed to mount")
+	}
+	return nil
+}
+
+// forkAndMountat will fork thread, change working dir and mount.
+//
+// precondition: the runtime OS thread must be locked.
+func forkAndMountat(dirfd uintptr, source, target, fstype, flags, data uintptr) (pid uintptr, errno syscall.Errno) {
+	// block signal during clone
+	beforeFork()
+
+	// the cloned thread shares the open file descriptor, but the thread
+	// never be reused by runtime.
+	pid, _, errno = syscall.RawSyscall6(syscall.SYS_CLONE, uintptr(syscall.SIGCHLD)|syscall.CLONE_FILES, 0, 0, 0, 0, 0)
+	if errno != 0 || pid != 0 {
+		// restore all signals
+		afterFork()
+		return
+	}
+
+	// restore all signals
+	afterForkInChild()
+
+	// change working dir
+	_, _, errno = syscall.RawSyscall(syscall.SYS_FCHDIR, dirfd, 0, 0)
+	if errno != 0 {
+		goto childerr
+	}
+	_, _, errno = syscall.RawSyscall6(syscall.SYS_MOUNT, source, target, fstype, flags, data, 0)
+
+childerr:
+	syscall.RawSyscall(syscall.SYS_EXIT, uintptr(errno), 0, 0)
+	panic("unreachable")
+}

vendor/github.com/containerd/containerd/sys/socket_unix.go

@@ -42,7 +42,7 @@ func CreateUnixSocket(path string) (net.Listener, error) {
 	return net.Listen("unix", path)
 }
 
-// GetLocalListener returns a listerner out of a unix socket.
+// GetLocalListener returns a listener out of a unix socket.
 func GetLocalListener(path string, uid, gid int) (net.Listener, error) {
 	// Ensure parent directory is created
 	if err := mkdirAs(filepath.Dir(path), uid, gid); err != nil {

vendor/github.com/containerd/containerd/sys/subprocess_unsafe_linux.go

@@ -0,0 +1,30 @@
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/
+
+package sys
+
+import (
+	_ "unsafe" // required for go:linkname.
+)
+
+//go:linkname beforeFork syscall.runtime_BeforeFork
+func beforeFork()
+
+//go:linkname afterFork syscall.runtime_AfterFork
+func afterFork()
+
+//go:linkname afterForkInChild syscall.runtime_AfterForkInChild
+func afterForkInChild()

vendor/github.com/containerd/containerd/sys/subprocess_unsafe_linux.s

@@ -0,0 +1,15 @@
+/*
+   Copyright The containerd Authors.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+*/

vendor/github.com/containerd/containerd/task_opts.go

@@ -18,10 +18,18 @@ package containerd
 
 import (
 	"context"
+	"encoding/json"
+	"fmt"
 	"syscall"
 
+	"github.com/containerd/containerd/api/types"
+	"github.com/containerd/containerd/content"
 	"github.com/containerd/containerd/errdefs"
+	"github.com/containerd/containerd/images"
 	"github.com/containerd/containerd/mount"
+	imagespec "github.com/opencontainers/image-spec/specs-go/v1"
+	"github.com/opencontainers/runtime-spec/specs-go"
+	"github.com/pkg/errors"
 )
 
 // NewTaskOpts allows the caller to set options on a new task
@@ -35,6 +43,44 @@ func WithRootFS(mounts []mount.Mount) NewTaskOpts {
 	}
 }
 
+// WithTaskCheckpoint allows a task to be created with live runtime and memory data from a
+// previous checkpoint. Additional software such as CRIU may be required to
+// restore a task from a checkpoint
+func WithTaskCheckpoint(im Image) NewTaskOpts {
+	return func(ctx context.Context, c *Client, info *TaskInfo) error {
+		desc := im.Target()
+		id := desc.Digest
+		index, err := decodeIndex(ctx, c.ContentStore(), desc)
+		if err != nil {
+			return err
+		}
+		for _, m := range index.Manifests {
+			if m.MediaType == images.MediaTypeContainerd1Checkpoint {
+				info.Checkpoint = &types.Descriptor{
+					MediaType: m.MediaType,
+					Size_:     m.Size,
+					Digest:    m.Digest,
+				}
+				return nil
+			}
+		}
+		return fmt.Errorf("checkpoint not found in index %s", id)
+	}
+}
+
+func decodeIndex(ctx context.Context, store content.Provider, desc imagespec.Descriptor) (*imagespec.Index, error) {
+	var index imagespec.Index
+	p, err := content.ReadBlob(ctx, store, desc)
+	if err != nil {
+		return nil, err
+	}
+	if err := json.Unmarshal(p, &index); err != nil {
+		return nil, err
+	}
+
+	return &index, nil
+}
+
 // WithCheckpointName sets the image name for the checkpoint
 func WithCheckpointName(name string) CheckpointTaskOpts {
 	return func(r *CheckpointTaskInfo) error {
@@ -92,3 +138,19 @@ func WithKillExecID(execID string) KillOpts {
 		return nil
 	}
 }
+
+// WithResources sets the provided resources for task updates. Resources must be
+// either a *specs.LinuxResources or a *specs.WindowsResources
+func WithResources(resources interface{}) UpdateTaskOpts {
+	return func(ctx context.Context, client *Client, r *UpdateTaskInfo) error {
+		switch resources.(type) {
+		case *specs.LinuxResources:
+		case *specs.WindowsResources:
+		default:
+			return errors.New("WithResources requires a *specs.LinuxResources or *specs.WindowsResources")
+		}
+
+		r.Resources = resources
+		return nil
+	}
+}

vendor/github.com/containerd/containerd/task_opts_unix.go

@@ -1,3 +1,5 @@
+// +build !windows
+
 /*
    Copyright The containerd Authors.
 
@@ -18,20 +20,11 @@ package containerd
 
 import (
 	"context"
-	"errors"
 
 	"github.com/containerd/containerd/runtime/linux/runctypes"
-	"github.com/opencontainers/runtime-spec/specs-go"
+	"github.com/pkg/errors"
 )
 
-// WithResources sets the provided resources for task updates
-func WithResources(resources *specs.LinuxResources) UpdateTaskOpts {
-	return func(ctx context.Context, client *Client, r *UpdateTaskInfo) error {
-		r.Resources = resources
-		return nil
-	}
-}
-
 // WithNoNewKeyring causes tasks not to be created with a new keyring for secret storage.
 // There is an upper limit on the number of keyrings in a linux system
 func WithNoNewKeyring(ctx context.Context, c *Client, ti *TaskInfo) error {
@@ -46,3 +39,19 @@ func WithNoNewKeyring(ctx context.Context, c *Client, ti *TaskInfo) error {
 	opts.NoNewKeyring = true
 	return nil
 }
+
+// WithNoPivotRoot instructs the runtime not to you pivot_root
+func WithNoPivotRoot(_ context.Context, _ *Client, info *TaskInfo) error {
+	if info.Options == nil {
+		info.Options = &runctypes.CreateOptions{
+			NoPivotRoot: true,
+		}
+		return nil
+	}
+	opts, ok := info.Options.(*runctypes.CreateOptions)
+	if !ok {
+		return errors.New("invalid options type, expected runctypes.CreateOptions")
+	}
+	opts.NoPivotRoot = true
+	return nil
+}

vendor/github.com/containerd/containerd/vendor.conf

@@ -1,10 +1,10 @@
-github.com/containerd/go-runc edcf3de1f4971445c42d61f20d506b30612aa031
-github.com/containerd/console 4d8a41f4ce5b9bae77c41786ea2458330f43f081
+github.com/containerd/go-runc 5a6d9f37cfa36b15efba46dc7ea349fa9b7143c3
+github.com/containerd/console c12b1e7919c14469339a5d38f2f8ed9b64a9de23
 github.com/containerd/cgroups 5e610833b72089b37d0e615de9a92dfc043757c2
 github.com/containerd/typeurl a93fcdb778cd272c6e9b3028b2f42d813e785d40
 github.com/containerd/fifo 3d5202aec260678c48179c56f40e6f38a095738c
 github.com/containerd/btrfs 2e1aa0ddf94f91fa282b6ed87c23bf0d64911244
-github.com/containerd/continuity d3c23511c1bf5851696cba83143d9cbcd666869b
+github.com/containerd/continuity f44b615e492bdfb371aae2f76ec694d9da1db537
 github.com/coreos/go-systemd 48702e0da86bd25e76cfef347e2adeb434a0d0a6
 github.com/docker/go-metrics 4ea375f7759c82740c893fc030bc37088d2ec098
 github.com/docker/go-events 9461782956ad83b30282bf90e31fa6a70c255ba9
@@ -20,7 +20,7 @@ github.com/gogo/protobuf v1.0.0
 github.com/gogo/googleapis 08a7655d27152912db7aaf4f983275eaf8d128ef
 github.com/golang/protobuf v1.1.0
 github.com/opencontainers/runtime-spec d810dbc60d8c5aeeb3d054bd1132fab2121968ce # v1.0.1-43-gd810dbc
-github.com/opencontainers/runc 69663f0bd4b60df09991c08812a60108003fa340
+github.com/opencontainers/runc 20aff4f0488c6d4b8df4d85b4f63f1f704c11abd
 github.com/sirupsen/logrus v1.0.0
 github.com/urfave/cli 7bc6a0acffa589f415f88aca16cc1de5ffd66f9c
 golang.org/x/net b3756b4b77d7b13260a0a2ec658753cf48922eac
@@ -32,8 +32,8 @@ github.com/opencontainers/image-spec v1.0.1
 golang.org/x/sync 450f422ab23cf9881c94e2db30cac0eb1b7cf80c
 github.com/BurntSushi/toml a368813c5e648fee92e5f6c30e3944ff9d5e8895
 github.com/grpc-ecosystem/go-grpc-prometheus 6b7015e65d366bf3f19b2b2a000a831940f0f7e0
-github.com/Microsoft/go-winio v0.4.7
-github.com/Microsoft/hcsshim v0.6.11
+github.com/Microsoft/go-winio v0.4.10
+github.com/Microsoft/hcsshim 44c060121b68e8bdc40b411beba551f3b4ee9e55
 github.com/boltdb/bolt e9cf4fae01b5a8ff89d0ec6b32f0d9c9f79aefdd
 google.golang.org/genproto d80a6e20e776b0b17a324d0ba1ab50a39c8e8944
 golang.org/x/text 19e51611da83d6be54ddafce4a4af510cb3e9ea4
@@ -43,7 +43,7 @@ gotest.tools v2.1.0
 github.com/google/go-cmp v0.1.0
 
 # cri dependencies
-github.com/containerd/cri 661f3b0377db409fe0e5677115f02ce7b89fd17d https://github.com/dmcgowan/cri-containerd
+github.com/containerd/cri v1.11.1
 github.com/containerd/go-cni 5882530828ecf62032409b298a3e8b19e08b6534
 github.com/blang/semver v3.1.0
 github.com/containernetworking/cni v0.6.0
@@ -71,7 +71,7 @@ github.com/xeipuuv/gojsonschema 1d523034197ff1f222f6429836dd36a2457a1874
 golang.org/x/crypto 49796115aa4b964c318aad4f3084fdb41e9aa067
 golang.org/x/time f51c12702a4d776e4c1fa9b0fabab841babae631
 gopkg.in/inf.v0 3887ee99ecf07df5b447e9b00d9c0b2adaa9f3e4
-gopkg.in/yaml.v2 53feefa2559fb8dfa8d81baad31be332c97d6c77
+gopkg.in/yaml.v2 v2.2.1
 k8s.io/api 9e5ffd1f1320950b238cfce291b926411f0af722
 k8s.io/apimachinery ed135c5b96450fd24e5e981c708114fbbd950697
 k8s.io/apiserver a90e3a95c2e91b944bfca8225c4e0d12e42a9eb5
@@ -85,4 +85,4 @@ github.com/mistifyio/go-zfs 166add352731e515512690329794ee593f1aaff2
 github.com/pborman/uuid c65b2f87fee37d1c7854c9164a450713c28d50cd
 
 # aufs dependencies
-github.com/containerd/aufs a7fbd554da7a9eafbe5a460a421313a9fd18d988
+github.com/containerd/aufs ffa39970e26ad01d81f540b21e65f9c1841a5f92

vendor/github.com/containerd/containerd/version/version.go

@@ -21,7 +21,7 @@ var (
 	Package = "github.com/containerd/containerd"
 
 	// Version holds the complete version number. Filled in at linking time.
-	Version = "1.1.0+unknown"
+	Version = "1.2.0-beta.2+unknown"
 
 	// Revision is filled with the VCS (e.g. git) revision being used to build
 	// the program at linking time.

vendor/github.com/containerd/continuity/context.go

@@ -0,0 +1,657 @@
+package continuity
+
+import (
+	"bytes"
+	"fmt"
+	"io"
+	"log"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/containerd/continuity/devices"
+	driverpkg "github.com/containerd/continuity/driver"
+	"github.com/containerd/continuity/pathdriver"
+
+	"github.com/opencontainers/go-digest"
+)
+
+var (
+	// ErrNotFound represents the resource not found
+	ErrNotFound = fmt.Errorf("not found")
+	// ErrNotSupported represents the resource not supported
+	ErrNotSupported = fmt.Errorf("not supported")
+)
+
+// Context represents a file system context for accessing resources. The
+// responsibility of the context is to convert system specific resources to
+// generic Resource objects. Most of this is safe path manipulation, as well
+// as extraction of resource details.
+type Context interface {
+	Apply(Resource) error
+	Verify(Resource) error
+	Resource(string, os.FileInfo) (Resource, error)
+	Walk(filepath.WalkFunc) error
+}
+
+// SymlinkPath is intended to give the symlink target value
+// in a root context. Target and linkname are absolute paths
+// not under the given root.
+type SymlinkPath func(root, linkname, target string) (string, error)
+
+// ContextOptions represents options to create a new context.
+type ContextOptions struct {
+	Digester   Digester
+	Driver     driverpkg.Driver
+	PathDriver pathdriver.PathDriver
+	Provider   ContentProvider
+}
+
+// context represents a file system context for accessing resources.
+// Generally, all path qualified access and system considerations should land
+// here.
+type context struct {
+	driver     driverpkg.Driver
+	pathDriver pathdriver.PathDriver
+	root       string
+	digester   Digester
+	provider   ContentProvider
+}
+
+// NewContext returns a Context associated with root. The default driver will
+// be used, as returned by NewDriver.
+func NewContext(root string) (Context, error) {
+	return NewContextWithOptions(root, ContextOptions{})
+}
+
+// NewContextWithOptions returns a Context associate with the root.
+func NewContextWithOptions(root string, options ContextOptions) (Context, error) {
+	// normalize to absolute path
+	pathDriver := options.PathDriver
+	if pathDriver == nil {
+		pathDriver = pathdriver.LocalPathDriver
+	}
+
+	root = pathDriver.FromSlash(root)
+	root, err := pathDriver.Abs(pathDriver.Clean(root))
+	if err != nil {
+		return nil, err
+	}
+
+	driver := options.Driver
+	if driver == nil {
+		driver, err = driverpkg.NewSystemDriver()
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	digester := options.Digester
+	if digester == nil {
+		digester = simpleDigester{digest.Canonical}
+	}
+
+	// Check the root directory. Need to be a little careful here. We are
+	// allowing a link for now, but this may have odd behavior when
+	// canonicalizing paths. As long as all files are opened through the link
+	// path, this should be okay.
+	fi, err := driver.Stat(root)
+	if err != nil {
+		return nil, err
+	}
+
+	if !fi.IsDir() {
+		return nil, &os.PathError{Op: "NewContext", Path: root, Err: os.ErrInvalid}
+	}
+
+	return &context{
+		root:       root,
+		driver:     driver,
+		pathDriver: pathDriver,
+		digester:   digester,
+		provider:   options.Provider,
+	}, nil
+}
+
+// Resource returns the resource as path p, populating the entry with info
+// from fi. The path p should be the path of the resource in the context,
+// typically obtained through Walk or from the value of Resource.Path(). If fi
+// is nil, it will be resolved.
+func (c *context) Resource(p string, fi os.FileInfo) (Resource, error) {
+	fp, err := c.fullpath(p)
+	if err != nil {
+		return nil, err
+	}
+
+	if fi == nil {
+		fi, err = c.driver.Lstat(fp)
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	base, err := newBaseResource(p, fi)
+	if err != nil {
+		return nil, err
+	}
+
+	base.xattrs, err = c.resolveXAttrs(fp, fi, base)
+	if err == ErrNotSupported {
+		log.Printf("resolving xattrs on %s not supported", fp)
+	} else if err != nil {
+		return nil, err
+	}
+
+	// TODO(stevvooe): Handle windows alternate data streams.
+
+	if fi.Mode().IsRegular() {
+		dgst, err := c.digest(p)
+		if err != nil {
+			return nil, err
+		}
+
+		return newRegularFile(*base, base.paths, fi.Size(), dgst)
+	}
+
+	if fi.Mode().IsDir() {
+		return newDirectory(*base)
+	}
+
+	if fi.Mode()&os.ModeSymlink != 0 {
+		// We handle relative links vs absolute links by including a
+		// beginning slash for absolute links. Effectively, the bundle's
+		// root is treated as the absolute link anchor.
+		target, err := c.driver.Readlink(fp)
+		if err != nil {
+			return nil, err
+		}
+
+		return newSymLink(*base, target)
+	}
+
+	if fi.Mode()&os.ModeNamedPipe != 0 {
+		return newNamedPipe(*base, base.paths)
+	}
+
+	if fi.Mode()&os.ModeDevice != 0 {
+		deviceDriver, ok := c.driver.(driverpkg.DeviceInfoDriver)
+		if !ok {
+			log.Printf("device extraction not supported %s", fp)
+			return nil, ErrNotSupported
+		}
+
+		// character and block devices merely need to recover the
+		// major/minor device number.
+		major, minor, err := deviceDriver.DeviceInfo(fi)
+		if err != nil {
+			return nil, err
+		}
+
+		return newDevice(*base, base.paths, major, minor)
+	}
+
+	log.Printf("%q (%v) is not supported", fp, fi.Mode())
+	return nil, ErrNotFound
+}
+
+func (c *context) verifyMetadata(resource, target Resource) error {
+	if target.Mode() != resource.Mode() {
+		return fmt.Errorf("resource %q has incorrect mode: %v != %v", target.Path(), target.Mode(), resource.Mode())
+	}
+
+	if target.UID() != resource.UID() {
+		return fmt.Errorf("unexpected uid for %q: %v != %v", target.Path(), target.UID(), resource.GID())
+	}
+
+	if target.GID() != resource.GID() {
+		return fmt.Errorf("unexpected gid for %q: %v != %v", target.Path(), target.GID(), target.GID())
+	}
+
+	if xattrer, ok := resource.(XAttrer); ok {
+		txattrer, tok := target.(XAttrer)
+		if !tok {
+			return fmt.Errorf("resource %q has xattrs but target does not support them", resource.Path())
+		}
+
+		// For xattrs, only ensure that we have those defined in the resource
+		// and their values match. We can ignore other xattrs. In other words,
+		// we only verify that target has the subset defined by resource.
+		txattrs := txattrer.XAttrs()
+		for attr, value := range xattrer.XAttrs() {
+			tvalue, ok := txattrs[attr]
+			if !ok {
+				return fmt.Errorf("resource %q target missing xattr %q", resource.Path(), attr)
+			}
+
+			if !bytes.Equal(value, tvalue) {
+				return fmt.Errorf("xattr %q value differs for resource %q", attr, resource.Path())
+			}
+		}
+	}
+
+	switch r := resource.(type) {
+	case RegularFile:
+		// TODO(stevvooe): Another reason to use a record-based approach. We
+		// have to do another type switch to get this to work. This could be
+		// fixed with an Equal function, but let's study this a little more to
+		// be sure.
+		t, ok := target.(RegularFile)
+		if !ok {
+			return fmt.Errorf("resource %q target not a regular file", r.Path())
+		}
+
+		if t.Size() != r.Size() {
+			return fmt.Errorf("resource %q target has incorrect size: %v != %v", t.Path(), t.Size(), r.Size())
+		}
+	case Directory:
+		t, ok := target.(Directory)
+		if !ok {
+			return fmt.Errorf("resource %q target not a directory", t.Path())
+		}
+	case SymLink:
+		t, ok := target.(SymLink)
+		if !ok {
+			return fmt.Errorf("resource %q target not a symlink", t.Path())
+		}
+
+		if t.Target() != r.Target() {
+			return fmt.Errorf("resource %q target has mismatched target: %q != %q", t.Path(), t.Target(), r.Target())
+		}
+	case Device:
+		t, ok := target.(Device)
+		if !ok {
+			return fmt.Errorf("resource %q is not a device", t.Path())
+		}
+
+		if t.Major() != r.Major() || t.Minor() != r.Minor() {
+			return fmt.Errorf("resource %q has mismatched major/minor numbers: %d,%d != %d,%d", t.Path(), t.Major(), t.Minor(), r.Major(), r.Minor())
+		}
+	case NamedPipe:
+		t, ok := target.(NamedPipe)
+		if !ok {
+			return fmt.Errorf("resource %q is not a named pipe", t.Path())
+		}
+	default:
+		return fmt.Errorf("cannot verify resource: %v", resource)
+	}
+
+	return nil
+}
+
+// Verify the resource in the context. An error will be returned a discrepancy
+// is found.
+func (c *context) Verify(resource Resource) error {
+	fp, err := c.fullpath(resource.Path())
+	if err != nil {
+		return err
+	}
+
+	fi, err := c.driver.Lstat(fp)
+	if err != nil {
+		return err
+	}
+
+	target, err := c.Resource(resource.Path(), fi)
+	if err != nil {
+		return err
+	}
+
+	if target.Path() != resource.Path() {
+		return fmt.Errorf("resource paths do not match: %q != %q", target.Path(), resource.Path())
+	}
+
+	if err := c.verifyMetadata(resource, target); err != nil {
+		return err
+	}
+
+	if h, isHardlinkable := resource.(Hardlinkable); isHardlinkable {
+		hardlinkKey, err := newHardlinkKey(fi)
+		if err == errNotAHardLink {
+			if len(h.Paths()) > 1 {
+				return fmt.Errorf("%q is not a hardlink to %q", h.Paths()[1], resource.Path())
+			}
+		} else if err != nil {
+			return err
+		}
+
+		for _, path := range h.Paths()[1:] {
+			fpLink, err := c.fullpath(path)
+			if err != nil {
+				return err
+			}
+
+			fiLink, err := c.driver.Lstat(fpLink)
+			if err != nil {
+				return err
+			}
+
+			targetLink, err := c.Resource(path, fiLink)
+			if err != nil {
+				return err
+			}
+
+			hardlinkKeyLink, err := newHardlinkKey(fiLink)
+			if err != nil {
+				return err
+			}
+
+			if hardlinkKeyLink != hardlinkKey {
+				return fmt.Errorf("%q is not a hardlink to %q", path, resource.Path())
+			}
+
+			if err := c.verifyMetadata(resource, targetLink); err != nil {
+				return err
+			}
+		}
+	}
+
+	switch r := resource.(type) {
+	case RegularFile:
+		t, ok := target.(RegularFile)
+		if !ok {
+			return fmt.Errorf("resource %q target not a regular file", r.Path())
+		}
+
+		// TODO(stevvooe): This may need to get a little more sophisticated
+		// for digest comparison. We may want to actually calculate the
+		// provided digests, rather than the implementations having an
+		// overlap.
+		if !digestsMatch(t.Digests(), r.Digests()) {
+			return fmt.Errorf("digests for resource %q do not match: %v != %v", t.Path(), t.Digests(), r.Digests())
+		}
+	}
+
+	return nil
+}
+
+func (c *context) checkoutFile(fp string, rf RegularFile) error {
+	if c.provider == nil {
+		return fmt.Errorf("no file provider")
+	}
+	var (
+		r   io.ReadCloser
+		err error
+	)
+	for _, dgst := range rf.Digests() {
+		r, err = c.provider.Reader(dgst)
+		if err == nil {
+			break
+		}
+	}
+	if err != nil {
+		return fmt.Errorf("file content could not be provided: %v", err)
+	}
+	defer r.Close()
+
+	return atomicWriteFile(fp, r, rf.Size(), rf.Mode())
+}
+
+// Apply the resource to the contexts. An error will be returned if the
+// operation fails. Depending on the resource type, the resource may be
+// created. For resource that cannot be resolved, an error will be returned.
+func (c *context) Apply(resource Resource) error {
+	fp, err := c.fullpath(resource.Path())
+	if err != nil {
+		return err
+	}
+
+	if !strings.HasPrefix(fp, c.root) {
+		return fmt.Errorf("resource %v escapes root", resource)
+	}
+
+	var chmod = true
+	fi, err := c.driver.Lstat(fp)
+	if err != nil {
+		if !os.IsNotExist(err) {
+			return err
+		}
+	}
+
+	switch r := resource.(type) {
+	case RegularFile:
+		if fi == nil {
+			if err := c.checkoutFile(fp, r); err != nil {
+				return fmt.Errorf("error checking out file %q: %v", resource.Path(), err)
+			}
+			chmod = false
+		} else {
+			if !fi.Mode().IsRegular() {
+				return fmt.Errorf("file %q should be a regular file, but is not", resource.Path())
+			}
+			if fi.Size() != r.Size() {
+				if err := c.checkoutFile(fp, r); err != nil {
+					return fmt.Errorf("error checking out file %q: %v", resource.Path(), err)
+				}
+			} else {
+				for _, dgst := range r.Digests() {
+					f, err := os.Open(fp)
+					if err != nil {
+						return fmt.Errorf("failure opening file for read %q: %v", resource.Path(), err)
+					}
+					compared, err := dgst.Algorithm().FromReader(f)
+					if err == nil && dgst != compared {
+						if err := c.checkoutFile(fp, r); err != nil {
+							return fmt.Errorf("error checking out file %q: %v", resource.Path(), err)
+						}
+						break
+					}
+					if err1 := f.Close(); err == nil {
+						err = err1
+					}
+					if err != nil {
+						return fmt.Errorf("error checking digest for %q: %v", resource.Path(), err)
+					}
+				}
+			}
+		}
+	case Directory:
+		if fi == nil {
+			if err := c.driver.Mkdir(fp, resource.Mode()); err != nil {
+				return err
+			}
+		} else if !fi.Mode().IsDir() {
+			return fmt.Errorf("%q should be a directory, but is not", resource.Path())
+		}
+
+	case SymLink:
+		var target string // only possibly set if target resource is a symlink
+
+		if fi != nil {
+			if fi.Mode()&os.ModeSymlink != 0 {
+				target, err = c.driver.Readlink(fp)
+				if err != nil {
+					return err
+				}
+			}
+		}
+
+		if target != r.Target() {
+			if fi != nil {
+				if err := c.driver.Remove(fp); err != nil { // RemoveAll in case of directory?
+					return err
+				}
+			}
+
+			if err := c.driver.Symlink(r.Target(), fp); err != nil {
+				return err
+			}
+		}
+
+	case Device:
+		if fi == nil {
+			if err := c.driver.Mknod(fp, resource.Mode(), int(r.Major()), int(r.Minor())); err != nil {
+				return err
+			}
+		} else if (fi.Mode() & os.ModeDevice) == 0 {
+			return fmt.Errorf("%q should be a device, but is not", resource.Path())
+		} else {
+			major, minor, err := devices.DeviceInfo(fi)
+			if err != nil {
+				return err
+			}
+			if major != r.Major() || minor != r.Minor() {
+				if err := c.driver.Remove(fp); err != nil {
+					return err
+				}
+
+				if err := c.driver.Mknod(fp, resource.Mode(), int(r.Major()), int(r.Minor())); err != nil {
+					return err
+				}
+			}
+		}
+
+	case NamedPipe:
+		if fi == nil {
+			if err := c.driver.Mkfifo(fp, resource.Mode()); err != nil {
+				return err
+			}
+		} else if (fi.Mode() & os.ModeNamedPipe) == 0 {
+			return fmt.Errorf("%q should be a named pipe, but is not", resource.Path())
+		}
+	}
+
+	if h, isHardlinkable := resource.(Hardlinkable); isHardlinkable {
+		for _, path := range h.Paths() {
+			if path == resource.Path() {
+				continue
+			}
+
+			lp, err := c.fullpath(path)
+			if err != nil {
+				return err
+			}
+
+			if _, fi := c.driver.Lstat(lp); fi == nil {
+				c.driver.Remove(lp)
+			}
+			if err := c.driver.Link(fp, lp); err != nil {
+				return err
+			}
+		}
+	}
+
+	// Update filemode if file was not created
+	if chmod {
+		if err := c.driver.Lchmod(fp, resource.Mode()); err != nil {
+			return err
+		}
+	}
+
+	if err := c.driver.Lchown(fp, resource.UID(), resource.GID()); err != nil {
+		return err
+	}
+
+	if xattrer, ok := resource.(XAttrer); ok {
+		// For xattrs, only ensure that we have those defined in the resource
+		// and their values are set. We can ignore other xattrs. In other words,
+		// we only set xattres defined by resource but never remove.
+
+		if _, ok := resource.(SymLink); ok {
+			lxattrDriver, ok := c.driver.(driverpkg.LXAttrDriver)
+			if !ok {
+				return fmt.Errorf("unsupported symlink xattr for resource %q", resource.Path())
+			}
+			if err := lxattrDriver.LSetxattr(fp, xattrer.XAttrs()); err != nil {
+				return err
+			}
+		} else {
+			xattrDriver, ok := c.driver.(driverpkg.XAttrDriver)
+			if !ok {
+				return fmt.Errorf("unsupported xattr for resource %q", resource.Path())
+			}
+			if err := xattrDriver.Setxattr(fp, xattrer.XAttrs()); err != nil {
+				return err
+			}
+		}
+	}
+
+	return nil
+}
+
+// Walk provides a convenience function to call filepath.Walk correctly for
+// the context. Otherwise identical to filepath.Walk, the path argument is
+// corrected to be contained within the context.
+func (c *context) Walk(fn filepath.WalkFunc) error {
+	root := c.root
+	fi, err := c.driver.Lstat(c.root)
+	if err == nil && fi.Mode()&os.ModeSymlink != 0 {
+		root, err = c.driver.Readlink(c.root)
+		if err != nil {
+			return err
+		}
+	}
+	return c.pathDriver.Walk(root, func(p string, fi os.FileInfo, err error) error {
+		contained, err := c.containWithRoot(p, root)
+		return fn(contained, fi, err)
+	})
+}
+
+// fullpath returns the system path for the resource, joined with the context
+// root. The path p must be a part of the context.
+func (c *context) fullpath(p string) (string, error) {
+	p = c.pathDriver.Join(c.root, p)
+	if !strings.HasPrefix(p, c.root) {
+		return "", fmt.Errorf("invalid context path")
+	}
+
+	return p, nil
+}
+
+// contain cleans and santizes the filesystem path p to be an absolute path,
+// effectively relative to the context root.
+func (c *context) contain(p string) (string, error) {
+	return c.containWithRoot(p, c.root)
+}
+
+// containWithRoot cleans and santizes the filesystem path p to be an absolute path,
+// effectively relative to the passed root. Extra care should be used when calling this
+// instead of contain. This is needed for Walk, as if context root is a symlink,
+// it must be evaluated prior to the Walk
+func (c *context) containWithRoot(p string, root string) (string, error) {
+	sanitized, err := c.pathDriver.Rel(root, p)
+	if err != nil {
+		return "", err
+	}
+
+	// ZOMBIES(stevvooe): In certain cases, we may want to remap these to a
+	// "containment error", so the caller can decide what to do.
+	return c.pathDriver.Join("/", c.pathDriver.Clean(sanitized)), nil
+}
+
+// digest returns the digest of the file at path p, relative to the root.
+func (c *context) digest(p string) (digest.Digest, error) {
+	f, err := c.driver.Open(c.pathDriver.Join(c.root, p))
+	if err != nil {
+		return "", err
+	}
+	defer f.Close()
+
+	return c.digester.Digest(f)
+}
+
+// resolveXAttrs attempts to resolve the extended attributes for the resource
+// at the path fp, which is the full path to the resource. If the resource
+// cannot have xattrs, nil will be returned.
+func (c *context) resolveXAttrs(fp string, fi os.FileInfo, base *resource) (map[string][]byte, error) {
+	if fi.Mode().IsRegular() || fi.Mode().IsDir() {
+		xattrDriver, ok := c.driver.(driverpkg.XAttrDriver)
+		if !ok {
+			log.Println("xattr extraction not supported")
+			return nil, ErrNotSupported
+		}
+
+		return xattrDriver.Getxattr(fp)
+	}
+
+	if fi.Mode()&os.ModeSymlink != 0 {
+		lxattrDriver, ok := c.driver.(driverpkg.LXAttrDriver)
+		if !ok {
+			log.Println("xattr extraction for symlinks not supported")
+			return nil, ErrNotSupported
+		}
+
+		return lxattrDriver.LGetxattr(fp)
+	}
+
+	return nil, nil
+}

vendor/github.com/containerd/continuity/devices/devices.go

@@ -0,0 +1,5 @@
+package devices
+
+import "fmt"
+
+var ErrNotSupported = fmt.Errorf("not supported")

vendor/github.com/containerd/continuity/devices/devices_unix.go

@@ -0,0 +1,58 @@
+// +build linux darwin freebsd solaris
+
+package devices
+
+import (
+	"fmt"
+	"os"
+	"syscall"
+
+	"golang.org/x/sys/unix"
+)
+
+func DeviceInfo(fi os.FileInfo) (uint64, uint64, error) {
+	sys, ok := fi.Sys().(*syscall.Stat_t)
+	if !ok {
+		return 0, 0, fmt.Errorf("cannot extract device from os.FileInfo")
+	}
+
+	dev := uint64(sys.Rdev)
+	return uint64(unix.Major(dev)), uint64(unix.Minor(dev)), nil
+}
+
+// mknod provides a shortcut for syscall.Mknod
+func Mknod(p string, mode os.FileMode, maj, min int) error {
+	var (
+		m   = syscallMode(mode.Perm())
+		dev uint64
+	)
+
+	if mode&os.ModeDevice != 0 {
+		dev = unix.Mkdev(uint32(maj), uint32(min))
+
+		if mode&os.ModeCharDevice != 0 {
+			m |= unix.S_IFCHR
+		} else {
+			m |= unix.S_IFBLK
+		}
+	} else if mode&os.ModeNamedPipe != 0 {
+		m |= unix.S_IFIFO
+	}
+
+	return unix.Mknod(p, m, int(dev))
+}
+
+// syscallMode returns the syscall-specific mode bits from Go's portable mode bits.
+func syscallMode(i os.FileMode) (o uint32) {
+	o |= uint32(i.Perm())
+	if i&os.ModeSetuid != 0 {
+		o |= unix.S_ISUID
+	}
+	if i&os.ModeSetgid != 0 {
+		o |= unix.S_ISGID
+	}
+	if i&os.ModeSticky != 0 {
+		o |= unix.S_ISVTX
+	}
+	return
+}

vendor/github.com/containerd/continuity/devices/devices_windows.go

@@ -0,0 +1,11 @@
+package devices
+
+import (
+	"os"
+
+	"github.com/pkg/errors"
+)
+
+func DeviceInfo(fi os.FileInfo) (uint64, uint64, error) {
+	return 0, 0, errors.Wrap(ErrNotSupported, "cannot get device info on windows")
+}

vendor/github.com/containerd/continuity/digests.go

@@ -0,0 +1,88 @@
+package continuity
+
+import (
+	"fmt"
+	"io"
+	"sort"
+
+	"github.com/opencontainers/go-digest"
+)
+
+// Digester produces a digest for a given read stream
+type Digester interface {
+	Digest(io.Reader) (digest.Digest, error)
+}
+
+// ContentProvider produces a read stream for a given digest
+type ContentProvider interface {
+	Reader(digest.Digest) (io.ReadCloser, error)
+}
+
+type simpleDigester struct {
+	algorithm digest.Algorithm
+}
+
+func (sd simpleDigester) Digest(r io.Reader) (digest.Digest, error) {
+	digester := sd.algorithm.Digester()
+
+	if _, err := io.Copy(digester.Hash(), r); err != nil {
+		return "", err
+	}
+
+	return digester.Digest(), nil
+}
+
+// uniqifyDigests sorts and uniqifies the provided digest, ensuring that the
+// digests are not repeated and no two digests with the same algorithm have
+// different values. Because a stable sort is used, this has the effect of
+// "zipping" digest collections from multiple resources.
+func uniqifyDigests(digests ...digest.Digest) ([]digest.Digest, error) {
+	sort.Stable(digestSlice(digests)) // stable sort is important for the behavior here.
+	seen := map[digest.Digest]struct{}{}
+	algs := map[digest.Algorithm][]digest.Digest{} // detect different digests.
+
+	var out []digest.Digest
+	// uniqify the digests
+	for _, d := range digests {
+		if _, ok := seen[d]; ok {
+			continue
+		}
+
+		seen[d] = struct{}{}
+		algs[d.Algorithm()] = append(algs[d.Algorithm()], d)
+
+		if len(algs[d.Algorithm()]) > 1 {
+			return nil, fmt.Errorf("conflicting digests for %v found", d.Algorithm())
+		}
+
+		out = append(out, d)
+	}
+
+	return out, nil
+}
+
+// digestsMatch compares the two sets of digests to see if they match.
+func digestsMatch(as, bs []digest.Digest) bool {
+	all := append(as, bs...)
+
+	uniqified, err := uniqifyDigests(all...)
+	if err != nil {
+		// the only error uniqifyDigests returns is when the digests disagree.
+		return false
+	}
+
+	disjoint := len(as) + len(bs)
+	if len(uniqified) == disjoint {
+		// if these two sets have the same cardinality, we know both sides
+		// didn't share any digests.
+		return false
+	}
+
+	return true
+}
+
+type digestSlice []digest.Digest
+
+func (p digestSlice) Len() int           { return len(p) }
+func (p digestSlice) Less(i, j int) bool { return p[i] < p[j] }
+func (p digestSlice) Swap(i, j int)      { p[i], p[j] = p[j], p[i] }

vendor/github.com/containerd/continuity/driver/driver.go

@@ -0,0 +1,158 @@
+package driver
+
+import (
+	"fmt"
+	"io"
+	"os"
+)
+
+var ErrNotSupported = fmt.Errorf("not supported")
+
+// Driver provides all of the system-level functions in a common interface.
+// The context should call these with full paths and should never use the `os`
+// package or any other package to access resources on the filesystem. This
+// mechanism let's us carefully control access to the context and maintain
+// path and resource integrity. It also gives us an interface to reason about
+// direct resource access.
+//
+// Implementations don't need to do much other than meet the interface. For
+// example, it is not required to wrap os.FileInfo to return correct paths for
+// the call to Name().
+type Driver interface {
+	// Note that Open() returns a File interface instead of *os.File. This
+	// is because os.File is a struct, so if Open was to return *os.File,
+	// the only way to fulfill the interface would be to call os.Open()
+	Open(path string) (File, error)
+	OpenFile(path string, flag int, perm os.FileMode) (File, error)
+
+	Stat(path string) (os.FileInfo, error)
+	Lstat(path string) (os.FileInfo, error)
+	Readlink(p string) (string, error)
+	Mkdir(path string, mode os.FileMode) error
+	Remove(path string) error
+
+	Link(oldname, newname string) error
+	Lchmod(path string, mode os.FileMode) error
+	Lchown(path string, uid, gid int64) error
+	Symlink(oldname, newname string) error
+
+	MkdirAll(path string, perm os.FileMode) error
+	RemoveAll(path string) error
+
+	// TODO(aaronl): These methods might move outside the main Driver
+	// interface in the future as more platforms are added.
+	Mknod(path string, mode os.FileMode, major int, minor int) error
+	Mkfifo(path string, mode os.FileMode) error
+}
+
+// File is the interface for interacting with files returned by continuity's Open
+// This is needed since os.File is a struct, instead of an interface, so it can't
+// be used.
+type File interface {
+	io.ReadWriteCloser
+	io.Seeker
+	Readdir(n int) ([]os.FileInfo, error)
+}
+
+func NewSystemDriver() (Driver, error) {
+	// TODO(stevvooe): Consider having this take a "hint" path argument, which
+	// would be the context root. The hint could be used to resolve required
+	// filesystem support when assembling the driver to use.
+	return &driver{}, nil
+}
+
+// XAttrDriver should be implemented on operation systems and filesystems that
+// have xattr support for regular files and directories.
+type XAttrDriver interface {
+	// Getxattr returns all of the extended attributes for the file at path.
+	// Typically, this takes a syscall call to Listxattr and Getxattr.
+	Getxattr(path string) (map[string][]byte, error)
+
+	// Setxattr sets all of the extended attributes on file at path, following
+	// any symbolic links, if necessary. All attributes on the target are
+	// replaced by the values from attr. If the operation fails to set any
+	// attribute, those already applied will not be rolled back.
+	Setxattr(path string, attr map[string][]byte) error
+}
+
+// LXAttrDriver should be implemented by drivers on operating systems and
+// filesystems that support setting and getting extended attributes on
+// symbolic links. If this is not implemented, extended attributes will be
+// ignored on symbolic links.
+type LXAttrDriver interface {
+	// LGetxattr returns all of the extended attributes for the file at path
+	// and does not follow symlinks. Typically, this takes a syscall call to
+	// Llistxattr and Lgetxattr.
+	LGetxattr(path string) (map[string][]byte, error)
+
+	// LSetxattr sets all of the extended attributes on file at path, without
+	// following symbolic links. All attributes on the target are replaced by
+	// the values from attr. If the operation fails to set any attribute,
+	// those already applied will not be rolled back.
+	LSetxattr(path string, attr map[string][]byte) error
+}
+
+type DeviceInfoDriver interface {
+	DeviceInfo(fi os.FileInfo) (maj uint64, min uint64, err error)
+}
+
+// driver is a simple default implementation that sends calls out to the "os"
+// package. Extend the "driver" type in system-specific files to add support,
+// such as xattrs, which can add support at compile time.
+type driver struct{}
+
+var _ File = &os.File{}
+
+// LocalDriver is the exported Driver struct for convenience.
+var LocalDriver Driver = &driver{}
+
+func (d *driver) Open(p string) (File, error) {
+	return os.Open(p)
+}
+
+func (d *driver) OpenFile(path string, flag int, perm os.FileMode) (File, error) {
+	return os.OpenFile(path, flag, perm)
+}
+
+func (d *driver) Stat(p string) (os.FileInfo, error) {
+	return os.Stat(p)
+}
+
+func (d *driver) Lstat(p string) (os.FileInfo, error) {
+	return os.Lstat(p)
+}
+
+func (d *driver) Mkdir(p string, mode os.FileMode) error {
+	return os.Mkdir(p, mode)
+}
+
+// Remove is used to unlink files and remove directories.
+// This is following the golang os package api which
+// combines the operations into a higher level Remove
+// function. If explicit unlinking or directory removal
+// to mirror system call is required, they should be
+// split up at that time.
+func (d *driver) Remove(path string) error {
+	return os.Remove(path)
+}
+
+func (d *driver) Link(oldname, newname string) error {
+	return os.Link(oldname, newname)
+}
+
+func (d *driver) Lchown(name string, uid, gid int64) error {
+	// TODO: error out if uid excesses int bit width?
+	return os.Lchown(name, int(uid), int(gid))
+}
+
+func (d *driver) Symlink(oldname, newname string) error {
+	return os.Symlink(oldname, newname)
+}
+
+func (d *driver) MkdirAll(path string, perm os.FileMode) error {
+	return os.MkdirAll(path, perm)
+}
+
+func (d *driver) RemoveAll(path string) error {
+	return os.RemoveAll(path)
+}

vendor/github.com/containerd/continuity/driver/driver_unix.go

@@ -0,0 +1,114 @@
+// +build linux darwin freebsd solaris
+
+package driver
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"sort"
+
+	"github.com/containerd/continuity/devices"
+	"github.com/containerd/continuity/sysx"
+)
+
+func (d *driver) Mknod(path string, mode os.FileMode, major, minor int) error {
+	return devices.Mknod(path, mode, major, minor)
+}
+
+func (d *driver) Mkfifo(path string, mode os.FileMode) error {
+	if mode&os.ModeNamedPipe == 0 {
+		return errors.New("mode passed to Mkfifo does not have the named pipe bit set")
+	}
+	// mknod with a mode that has ModeNamedPipe set creates a fifo, not a
+	// device.
+	return devices.Mknod(path, mode, 0, 0)
+}
+
+// Getxattr returns all of the extended attributes for the file at path p.
+func (d *driver) Getxattr(p string) (map[string][]byte, error) {
+	xattrs, err := sysx.Listxattr(p)
+	if err != nil {
+		return nil, fmt.Errorf("listing %s xattrs: %v", p, err)
+	}
+
+	sort.Strings(xattrs)
+	m := make(map[string][]byte, len(xattrs))
+
+	for _, attr := range xattrs {
+		value, err := sysx.Getxattr(p, attr)
+		if err != nil {
+			return nil, fmt.Errorf("getting %q xattr on %s: %v", attr, p, err)
+		}
+
+		// NOTE(stevvooe): This append/copy tricky relies on unique
+		// xattrs. Break this out into an alloc/copy if xattrs are no
+		// longer unique.
+		m[attr] = append(m[attr], value...)
+	}
+
+	return m, nil
+}
+
+// Setxattr sets all of the extended attributes on file at path, following
+// any symbolic links, if necessary. All attributes on the target are
+// replaced by the values from attr. If the operation fails to set any
+// attribute, those already applied will not be rolled back.
+func (d *driver) Setxattr(path string, attrMap map[string][]byte) error {
+	for attr, value := range attrMap {
+		if err := sysx.Setxattr(path, attr, value, 0); err != nil {
+			return fmt.Errorf("error setting xattr %q on %s: %v", attr, path, err)
+		}
+	}
+
+	return nil
+}
+
+// LGetxattr returns all of the extended attributes for the file at path p
+// not following symbolic links.
+func (d *driver) LGetxattr(p string) (map[string][]byte, error) {
+	xattrs, err := sysx.LListxattr(p)
+	if err != nil {
+		return nil, fmt.Errorf("listing %s xattrs: %v", p, err)
+	}
+
+	sort.Strings(xattrs)
+	m := make(map[string][]byte, len(xattrs))
+
+	for _, attr := range xattrs {
+		value, err := sysx.LGetxattr(p, attr)
+		if err != nil {
+			return nil, fmt.Errorf("getting %q xattr on %s: %v", attr, p, err)
+		}
+
+		// NOTE(stevvooe): This append/copy tricky relies on unique
+		// xattrs. Break this out into an alloc/copy if xattrs are no
+		// longer unique.
+		m[attr] = append(m[attr], value...)
+	}
+
+	return m, nil
+}
+
+// LSetxattr sets all of the extended attributes on file at path, not
+// following any symbolic links. All attributes on the target are
+// replaced by the values from attr. If the operation fails to set any
+// attribute, those already applied will not be rolled back.
+func (d *driver) LSetxattr(path string, attrMap map[string][]byte) error {
+	for attr, value := range attrMap {
+		if err := sysx.LSetxattr(path, attr, value, 0); err != nil {
+			return fmt.Errorf("error setting xattr %q on %s: %v", attr, path, err)
+		}
+	}
+
+	return nil
+}
+
+func (d *driver) DeviceInfo(fi os.FileInfo) (maj uint64, min uint64, err error) {
+	return devices.DeviceInfo(fi)
+}
+
+// Readlink was forked on Windows to fix a Golang bug, use the "os" package here
+func (d *driver) Readlink(p string) (string, error) {
+	return os.Readlink(p)
+}

vendor/github.com/containerd/continuity/driver/driver_windows.go

@@ -0,0 +1,28 @@
+package driver
+
+import (
+	"os"
+
+	"github.com/containerd/continuity/sysx"
+	"github.com/pkg/errors"
+)
+
+func (d *driver) Mknod(path string, mode os.FileMode, major, minor int) error {
+	return errors.Wrap(ErrNotSupported, "cannot create device node on Windows")
+}
+
+func (d *driver) Mkfifo(path string, mode os.FileMode) error {
+	return errors.Wrap(ErrNotSupported, "cannot create fifo on Windows")
+}
+
+// Lchmod changes the mode of an file not following symlinks.
+func (d *driver) Lchmod(path string, mode os.FileMode) (err error) {
+	// TODO: Use Window's equivalent
+	return os.Chmod(path, mode)
+}
+
+// Readlink is forked in order to support Volume paths which are used
+// in container layers.
+func (d *driver) Readlink(p string) (string, error) {
+	return sysx.Readlink(p)
+}

vendor/github.com/containerd/continuity/driver/lchmod_linux.go

@@ -0,0 +1,19 @@
+package driver
+
+import (
+	"os"
+
+	"golang.org/x/sys/unix"
+)
+
+// Lchmod changes the mode of a file not following symlinks.
+func (d *driver) Lchmod(path string, mode os.FileMode) error {
+	// On Linux, file mode is not supported for symlinks,
+	// and fchmodat() does not support AT_SYMLINK_NOFOLLOW,
+	// so symlinks need to be skipped entirely.
+	if st, err := os.Stat(path); err == nil && st.Mode()&os.ModeSymlink != 0 {
+		return nil
+	}
+
+	return unix.Fchmodat(unix.AT_FDCWD, path, uint32(mode), 0)
+}

vendor/github.com/containerd/continuity/driver/lchmod_unix.go

@@ -0,0 +1,14 @@
+// +build darwin freebsd solaris
+
+package driver
+
+import (
+	"os"
+
+	"golang.org/x/sys/unix"
+)
+
+// Lchmod changes the mode of a file not following symlinks.
+func (d *driver) Lchmod(path string, mode os.FileMode) error {
+	return unix.Fchmodat(unix.AT_FDCWD, path, uint32(mode), unix.AT_SYMLINK_NOFOLLOW)
+}

vendor/github.com/containerd/continuity/driver/utils.go

@@ -0,0 +1,74 @@
+package driver
+
+import (
+	"io"
+	"io/ioutil"
+	"os"
+	"sort"
+)
+
+// ReadFile works the same as ioutil.ReadFile with the Driver abstraction
+func ReadFile(r Driver, filename string) ([]byte, error) {
+	f, err := r.Open(filename)
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	data, err := ioutil.ReadAll(f)
+	if err != nil {
+		return nil, err
+	}
+
+	return data, nil
+}
+
+// WriteFile works the same as ioutil.WriteFile with the Driver abstraction
+func WriteFile(r Driver, filename string, data []byte, perm os.FileMode) error {
+	f, err := r.OpenFile(filename, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, perm)
+	if err != nil {
+		return err
+	}
+	defer f.Close()
+
+	n, err := f.Write(data)
+	if err != nil {
+		return err
+	} else if n != len(data) {
+		return io.ErrShortWrite
+	}
+
+	return nil
+}
+
+// ReadDir works the same as ioutil.ReadDir with the Driver abstraction
+func ReadDir(r Driver, dirname string) ([]os.FileInfo, error) {
+	f, err := r.Open(dirname)
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	dirs, err := f.Readdir(-1)
+	if err != nil {
+		return nil, err
+	}
+
+	sort.Sort(fileInfos(dirs))
+	return dirs, nil
+}
+
+// Simple implementation of the sort.Interface for os.FileInfo
+type fileInfos []os.FileInfo
+
+func (fis fileInfos) Len() int {
+	return len(fis)
+}
+
+func (fis fileInfos) Less(i, j int) bool {
+	return fis[i].Name() < fis[j].Name()
+}
+
+func (fis fileInfos) Swap(i, j int) {
+	fis[i], fis[j] = fis[j], fis[i]
+}

vendor/github.com/containerd/continuity/fs/du.go

@@ -10,8 +10,8 @@ type Usage struct {
 
 // DiskUsage counts the number of inodes and disk usage for the resources under
 // path.
-func DiskUsage(roots ...string) (Usage, error) {
-	return diskUsage(roots...)
+func DiskUsage(ctx context.Context, roots ...string) (Usage, error) {
+	return diskUsage(ctx, roots...)
 }
 
 // DiffUsage counts the numbers of inodes and disk usage in the

vendor/github.com/containerd/continuity/fs/du_unix.go

@@ -24,7 +24,7 @@ func newInode(stat *syscall.Stat_t) inode {
 	}
 }
 
-func diskUsage(roots ...string) (Usage, error) {
+func diskUsage(ctx context.Context, roots ...string) (Usage, error) {
 
 	var (
 		size   int64
@@ -37,6 +37,12 @@ func diskUsage(roots ...string) (Usage, error) {
 				return err
 			}
 
+			select {
+			case <-ctx.Done():
+				return ctx.Err()
+			default:
+			}
+
 			inoKey := newInode(fi.Sys().(*syscall.Stat_t))
 			if _, ok := inodes[inoKey]; !ok {
 				inodes[inoKey] = struct{}{}

vendor/github.com/containerd/continuity/fs/du_windows.go

@@ -8,7 +8,7 @@ import (
 	"path/filepath"
 )
 
-func diskUsage(roots ...string) (Usage, error) {
+func diskUsage(ctx context.Context, roots ...string) (Usage, error) {
 	var (
 		size int64
 	)
@@ -21,6 +21,12 @@ func diskUsage(roots ...string) (Usage, error) {
 				return err
 			}
 
+			select {
+			case <-ctx.Done():
+				return ctx.Err()
+			default:
+			}
+
 			size += fi.Size()
 			return nil
 		}); err != nil {

vendor/github.com/containerd/continuity/groups_unix.go

@@ -0,0 +1,113 @@
+package continuity
+
+import (
+	"bufio"
+	"fmt"
+	"io"
+	"os"
+	"strconv"
+	"strings"
+)
+
+// TODO(stevvooe): This needs a lot of work before we can call it useful.
+
+type groupIndex struct {
+	byName map[string]*group
+	byGID  map[int]*group
+}
+
+func getGroupIndex() (*groupIndex, error) {
+	f, err := os.Open("/etc/group")
+	if err != nil {
+		return nil, err
+	}
+	defer f.Close()
+
+	groups, err := parseGroups(f)
+	if err != nil {
+		return nil, err
+	}
+
+	return newGroupIndex(groups), nil
+}
+
+func newGroupIndex(groups []group) *groupIndex {
+	gi := &groupIndex{
+		byName: make(map[string]*group),
+		byGID:  make(map[int]*group),
+	}
+
+	for i, group := range groups {
+		gi.byGID[group.gid] = &groups[i]
+		gi.byName[group.name] = &groups[i]
+	}
+
+	return gi
+}
+
+type group struct {
+	name    string
+	gid     int
+	members []string
+}
+
+func getGroupName(gid int) (string, error) {
+	f, err := os.Open("/etc/group")
+	if err != nil {
+		return "", err
+	}
+	defer f.Close()
+
+	groups, err := parseGroups(f)
+	if err != nil {
+		return "", err
+	}
+
+	for _, group := range groups {
+		if group.gid == gid {
+			return group.name, nil
+		}
+	}
+
+	return "", fmt.Errorf("no group for gid")
+}
+
+// parseGroups parses an /etc/group file for group names, ids and membership.
+// This is unix specific.
+func parseGroups(rd io.Reader) ([]group, error) {
+	var groups []group
+	scanner := bufio.NewScanner(rd)
+
+	for scanner.Scan() {
+		if strings.HasPrefix(scanner.Text(), "#") {
+			continue // skip comment
+		}
+
+		parts := strings.SplitN(scanner.Text(), ":", 4)
+
+		if len(parts) != 4 {
+			return nil, fmt.Errorf("bad entry: %q", scanner.Text())
+		}
+
+		name, _, sgid, smembers := parts[0], parts[1], parts[2], parts[3]
+
+		gid, err := strconv.Atoi(sgid)
+		if err != nil {
+			return nil, fmt.Errorf("bad gid: %q", gid)
+		}
+
+		members := strings.Split(smembers, ",")
+
+		groups = append(groups, group{
+			name:    name,
+			gid:     gid,
+			members: members,
+		})
+	}
+
+	if scanner.Err() != nil {
+		return nil, scanner.Err()
+	}
+
+	return groups, nil
+}

vendor/github.com/containerd/continuity/hardlinks.go

@@ -0,0 +1,57 @@
+package continuity
+
+import (
+	"fmt"
+	"os"
+)
+
+var (
+	errNotAHardLink = fmt.Errorf("invalid hardlink")
+)
+
+type hardlinkManager struct {
+	hardlinks map[hardlinkKey][]Resource
+}
+
+func newHardlinkManager() *hardlinkManager {
+	return &hardlinkManager{
+		hardlinks: map[hardlinkKey][]Resource{},
+	}
+}
+
+// Add attempts to add the resource to the hardlink manager. If the resource
+// cannot be considered as a hardlink candidate, errNotAHardLink is returned.
+func (hlm *hardlinkManager) Add(fi os.FileInfo, resource Resource) error {
+	if _, ok := resource.(Hardlinkable); !ok {
+		return errNotAHardLink
+	}
+
+	key, err := newHardlinkKey(fi)
+	if err != nil {
+		return err
+	}
+
+	hlm.hardlinks[key] = append(hlm.hardlinks[key], resource)
+
+	return nil
+}
+
+// Merge processes the current state of the hardlink manager and merges any
+// shared nodes into hardlinked resources.
+func (hlm *hardlinkManager) Merge() ([]Resource, error) {
+	var resources []Resource
+	for key, linked := range hlm.hardlinks {
+		if len(linked) < 1 {
+			return nil, fmt.Errorf("no hardlink entrys for dev, inode pair: %#v", key)
+		}
+
+		merged, err := Merge(linked...)
+		if err != nil {
+			return nil, fmt.Errorf("error merging hardlink: %v", err)
+		}
+
+		resources = append(resources, merged)
+	}
+
+	return resources, nil
+}

vendor/github.com/containerd/continuity/hardlinks_unix.go

@@ -0,0 +1,36 @@
+// +build linux darwin freebsd solaris
+
+package continuity
+
+import (
+	"fmt"
+	"os"
+	"syscall"
+)
+
+// hardlinkKey provides a tuple-key for managing hardlinks. This is system-
+// specific.
+type hardlinkKey struct {
+	dev   uint64
+	inode uint64
+}
+
+// newHardlinkKey returns a hardlink key for the provided file info. If the
+// resource does not represent a possible hardlink, errNotAHardLink will be
+// returned.
+func newHardlinkKey(fi os.FileInfo) (hardlinkKey, error) {
+	sys, ok := fi.Sys().(*syscall.Stat_t)
+	if !ok {
+		return hardlinkKey{}, fmt.Errorf("cannot resolve (*syscall.Stat_t) from os.FileInfo")
+	}
+
+	if sys.Nlink < 2 {
+		// NOTE(stevvooe): This is not always true for all filesystems. We
+		// should somehow detect this and provided a slow "polyfill" that
+		// leverages os.SameFile if we detect a filesystem where link counts
+		// is not really supported.
+		return hardlinkKey{}, errNotAHardLink
+	}
+
+	return hardlinkKey{dev: uint64(sys.Dev), inode: uint64(sys.Ino)}, nil
+}

vendor/github.com/containerd/continuity/hardlinks_windows.go

@@ -0,0 +1,12 @@
+package continuity
+
+import "os"
+
+type hardlinkKey struct{}
+
+func newHardlinkKey(fi os.FileInfo) (hardlinkKey, error) {
+	// NOTE(stevvooe): Obviously, this is not yet implemented. However, the
+	// makings of an implementation are available in src/os/types_windows.go. More
+	// investigation needs to be done to figure out exactly how to do this.
+	return hardlinkKey{}, errNotAHardLink
+}

vendor/github.com/containerd/continuity/ioutils.go

@@ -0,0 +1,47 @@
+package continuity
+
+import (
+	"bytes"
+	"io"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+)
+
+// AtomicWriteFile atomically writes data to a file by first writing to a
+// temp file and calling rename.
+func AtomicWriteFile(filename string, data []byte, perm os.FileMode) error {
+	buf := bytes.NewBuffer(data)
+	return atomicWriteFile(filename, buf, int64(len(data)), perm)
+}
+
+// atomicWriteFile writes data to a file by first writing to a temp
+// file and calling rename.
+func atomicWriteFile(filename string, r io.Reader, dataSize int64, perm os.FileMode) error {
+	f, err := ioutil.TempFile(filepath.Dir(filename), ".tmp-"+filepath.Base(filename))
+	if err != nil {
+		return err
+	}
+	err = os.Chmod(f.Name(), perm)
+	if err != nil {
+		f.Close()
+		return err
+	}
+	n, err := io.Copy(f, r)
+	if err == nil && n < dataSize {
+		f.Close()
+		return io.ErrShortWrite
+	}
+	if err != nil {
+		f.Close()
+		return err
+	}
+	if err := f.Sync(); err != nil {
+		f.Close()
+		return err
+	}
+	if err := f.Close(); err != nil {
+		return err
+	}
+	return os.Rename(f.Name(), filename)
+}

vendor/github.com/containerd/continuity/manifest.go

@@ -0,0 +1,144 @@
+package continuity
+
+import (
+	"fmt"
+	"io"
+	"log"
+	"os"
+	"sort"
+
+	pb "github.com/containerd/continuity/proto"
+	"github.com/golang/protobuf/proto"
+)
+
+// Manifest provides the contents of a manifest. Users of this struct should
+// not typically modify any fields directly.
+type Manifest struct {
+	// Resources specifies all the resources for a manifest in order by path.
+	Resources []Resource
+}
+
+func Unmarshal(p []byte) (*Manifest, error) {
+	var bm pb.Manifest
+
+	if err := proto.Unmarshal(p, &bm); err != nil {
+		return nil, err
+	}
+
+	var m Manifest
+	for _, b := range bm.Resource {
+		r, err := fromProto(b)
+		if err != nil {
+			return nil, err
+		}
+
+		m.Resources = append(m.Resources, r)
+	}
+
+	return &m, nil
+}
+
+func Marshal(m *Manifest) ([]byte, error) {
+	var bm pb.Manifest
+	for _, resource := range m.Resources {
+		bm.Resource = append(bm.Resource, toProto(resource))
+	}
+
+	return proto.Marshal(&bm)
+}
+
+func MarshalText(w io.Writer, m *Manifest) error {
+	var bm pb.Manifest
+	for _, resource := range m.Resources {
+		bm.Resource = append(bm.Resource, toProto(resource))
+	}
+
+	return proto.MarshalText(w, &bm)
+}
+
+// BuildManifest creates the manifest for the given context
+func BuildManifest(ctx Context) (*Manifest, error) {
+	resourcesByPath := map[string]Resource{}
+	hardlinks := newHardlinkManager()
+
+	if err := ctx.Walk(func(p string, fi os.FileInfo, err error) error {
+		if err != nil {
+			return fmt.Errorf("error walking %s: %v", p, err)
+		}
+
+		if p == string(os.PathSeparator) {
+			// skip root
+			return nil
+		}
+
+		resource, err := ctx.Resource(p, fi)
+		if err != nil {
+			if err == ErrNotFound {
+				return nil
+			}
+			log.Printf("error getting resource %q: %v", p, err)
+			return err
+		}
+
+		// add to the hardlink manager
+		if err := hardlinks.Add(fi, resource); err == nil {
+			// Resource has been accepted by hardlink manager so we don't add
+			// it to the resourcesByPath until we merge at the end.
+			return nil
+		} else if err != errNotAHardLink {
+			// handle any other case where we have a proper error.
+			return fmt.Errorf("adding hardlink %s: %v", p, err)
+		}
+
+		resourcesByPath[p] = resource
+
+		return nil
+	}); err != nil {
+		return nil, err
+	}
+
+	// merge and post-process the hardlinks.
+	hardlinked, err := hardlinks.Merge()
+	if err != nil {
+		return nil, err
+	}
+
+	for _, resource := range hardlinked {
+		resourcesByPath[resource.Path()] = resource
+	}
+
+	var resources []Resource
+	for _, resource := range resourcesByPath {
+		resources = append(resources, resource)
+	}
+
+	sort.Stable(ByPath(resources))
+
+	return &Manifest{
+		Resources: resources,
+	}, nil
+}
+
+// VerifyManifest verifies all the resources in a manifest
+// against files from the given context.
+func VerifyManifest(ctx Context, manifest *Manifest) error {
+	for _, resource := range manifest.Resources {
+		if err := ctx.Verify(resource); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// ApplyManifest applies on the resources in a manifest to
+// the given context.
+func ApplyManifest(ctx Context, manifest *Manifest) error {
+	for _, resource := range manifest.Resources {
+		if err := ctx.Apply(resource); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}

vendor/github.com/containerd/continuity/proto/gen.go

@@ -0,0 +1,3 @@
+package proto
+
+//go:generate protoc --go_out=. manifest.proto

vendor/github.com/containerd/continuity/proto/manifest.pb.go

@@ -0,0 +1,181 @@
+// Code generated by protoc-gen-go.
+// source: manifest.proto
+// DO NOT EDIT!
+
+/*
+Package proto is a generated protocol buffer package.
+
+It is generated from these files:
+	manifest.proto
+
+It has these top-level messages:
+	Manifest
+	Resource
+	XAttr
+	ADSEntry
+*/
+package proto
+
+import proto1 "github.com/golang/protobuf/proto"
+import fmt "fmt"
+import math "math"
+
+// Reference imports to suppress errors if they are not otherwise used.
+var _ = proto1.Marshal
+var _ = fmt.Errorf
+var _ = math.Inf
+
+// This is a compile-time assertion to ensure that this generated file
+// is compatible with the proto package it is being compiled against.
+// A compilation error at this line likely means your copy of the
+// proto package needs to be updated.
+const _ = proto1.ProtoPackageIsVersion2 // please upgrade the proto package
+
+// Manifest specifies the entries in a container bundle, keyed and sorted by
+// path.
+type Manifest struct {
+	Resource []*Resource `protobuf:"bytes,1,rep,name=resource" json:"resource,omitempty"`
+}
+
+func (m *Manifest) Reset()                    { *m = Manifest{} }
+func (m *Manifest) String() string            { return proto1.CompactTextString(m) }
+func (*Manifest) ProtoMessage()               {}
+func (*Manifest) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{0} }
+
+func (m *Manifest) GetResource() []*Resource {
+	if m != nil {
+		return m.Resource
+	}
+	return nil
+}
+
+type Resource struct {
+	// Path specifies the path from the bundle root. If more than one
+	// path is present, the entry may represent a hardlink, rather than using
+	// a link target. The path format is operating system specific.
+	Path []string `protobuf:"bytes,1,rep,name=path" json:"path,omitempty"`
+	// Uid specifies the user id for the resource.
+	Uid int64 `protobuf:"varint,2,opt,name=uid" json:"uid,omitempty"`
+	// Gid specifies the group id for the resource.
+	Gid int64 `protobuf:"varint,3,opt,name=gid" json:"gid,omitempty"`
+	// user and group are not currently used but their field numbers have been
+	// reserved for future use. As such, they are marked as deprecated.
+	User  string `protobuf:"bytes,4,opt,name=user" json:"user,omitempty"`
+	Group string `protobuf:"bytes,5,opt,name=group" json:"group,omitempty"`
+	// Mode defines the file mode and permissions. We've used the same
+	// bit-packing from Go's os package,
+	// http://golang.org/pkg/os/#FileMode, since they've done the work of
+	// creating a cross-platform layout.
+	Mode uint32 `protobuf:"varint,6,opt,name=mode" json:"mode,omitempty"`
+	// Size specifies the size in bytes of the resource. This is only valid
+	// for regular files.
+	Size uint64 `protobuf:"varint,7,opt,name=size" json:"size,omitempty"`
+	// Digest specifies the content digest of the target file. Only valid for
+	// regular files. The strings are formatted in OCI style, i.e. <alg>:<encoded>.
+	// For detailed information about the format, please refer to OCI Image Spec:
+	// https://github.com/opencontainers/image-spec/blob/master/descriptor.md#digests-and-verification
+	// The digests are sorted in lexical order and implementations may choose
+	// which algorithms they prefer.
+	Digest []string `protobuf:"bytes,8,rep,name=digest" json:"digest,omitempty"`
+	// Target defines the target of a hard or soft link. Absolute links start
+	// with a slash and specify the resource relative to the bundle root.
+	// Relative links do not start with a slash and are relative to the
+	// resource path.
+	Target string `protobuf:"bytes,9,opt,name=target" json:"target,omitempty"`
+	// Major specifies the major device number for character and block devices.
+	Major uint64 `protobuf:"varint,10,opt,name=major" json:"major,omitempty"`
+	// Minor specifies the minor device number for character and block devices.
+	Minor uint64 `protobuf:"varint,11,opt,name=minor" json:"minor,omitempty"`
+	// Xattr provides storage for extended attributes for the target resource.
+	Xattr []*XAttr `protobuf:"bytes,12,rep,name=xattr" json:"xattr,omitempty"`
+	// Ads stores one or more alternate data streams for the target resource.
+	Ads []*ADSEntry `protobuf:"bytes,13,rep,name=ads" json:"ads,omitempty"`
+}
+
+func (m *Resource) Reset()                    { *m = Resource{} }
+func (m *Resource) String() string            { return proto1.CompactTextString(m) }
+func (*Resource) ProtoMessage()               {}
+func (*Resource) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{1} }
+
+func (m *Resource) GetXattr() []*XAttr {
+	if m != nil {
+		return m.Xattr
+	}
+	return nil
+}
+
+func (m *Resource) GetAds() []*ADSEntry {
+	if m != nil {
+		return m.Ads
+	}
+	return nil
+}
+
+// XAttr encodes extended attributes for a resource.
+type XAttr struct {
+	// Name specifies the attribute name.
+	Name string `protobuf:"bytes,1,opt,name=name" json:"name,omitempty"`
+	// Data specifies the associated data for the attribute.
+	Data []byte `protobuf:"bytes,2,opt,name=data,proto3" json:"data,omitempty"`
+}
+
+func (m *XAttr) Reset()                    { *m = XAttr{} }
+func (m *XAttr) String() string            { return proto1.CompactTextString(m) }
+func (*XAttr) ProtoMessage()               {}
+func (*XAttr) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{2} }
+
+// ADSEntry encodes information for a Windows Alternate Data Stream.
+type ADSEntry struct {
+	// Name specifices the stream name.
+	Name string `protobuf:"bytes,1,opt,name=name" json:"name,omitempty"`
+	// Data specifies the stream data.
+	// See also the description about the digest below.
+	Data []byte `protobuf:"bytes,2,opt,name=data,proto3" json:"data,omitempty"`
+	// Digest is a CAS representation of the stream data.
+	//
+	// At least one of data or digest MUST be specified, and either one of them
+	// SHOULD be specified.
+	//
+	// How to access the actual data using the digest is implementation-specific,
+	// and implementations can choose not to implement digest.
+	// So, digest SHOULD be used only when the stream data is large.
+	Digest string `protobuf:"bytes,3,opt,name=digest" json:"digest,omitempty"`
+}
+
+func (m *ADSEntry) Reset()                    { *m = ADSEntry{} }
+func (m *ADSEntry) String() string            { return proto1.CompactTextString(m) }
+func (*ADSEntry) ProtoMessage()               {}
+func (*ADSEntry) Descriptor() ([]byte, []int) { return fileDescriptor0, []int{3} }
+
+func init() {
+	proto1.RegisterType((*Manifest)(nil), "proto.Manifest")
+	proto1.RegisterType((*Resource)(nil), "proto.Resource")
+	proto1.RegisterType((*XAttr)(nil), "proto.XAttr")
+	proto1.RegisterType((*ADSEntry)(nil), "proto.ADSEntry")
+}
+
+func init() { proto1.RegisterFile("manifest.proto", fileDescriptor0) }
+
+var fileDescriptor0 = []byte{
+	// 317 bytes of a gzipped FileDescriptorProto
+	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, 0x02, 0xff, 0x8c, 0x90, 0x4f, 0x4b, 0xf3, 0x40,
+	0x10, 0xc6, 0x49, 0x93, 0xf4, 0x4d, 0xa7, 0xed, 0xab, 0x2c, 0x52, 0xe6, 0x18, 0x73, 0x0a, 0x08,
+	0x15, 0xf4, 0xe0, 0xb9, 0xa2, 0x17, 0xc1, 0xcb, 0x7a, 0xf1, 0xba, 0xba, 0x6b, 0x5c, 0x21, 0xd9,
+	0xb0, 0xd9, 0x80, 0xfa, 0xe5, 0xfc, 0x6a, 0x32, 0xb3, 0x69, 0xd1, 0x9b, 0xa7, 0x3c, 0xcf, 0x6f,
+	0xfe, 0x64, 0xf6, 0x81, 0xff, 0xad, 0xea, 0xec, 0x8b, 0x19, 0xc2, 0xb6, 0xf7, 0x2e, 0x38, 0x91,
+	0xf3, 0xa7, 0xba, 0x82, 0xe2, 0x7e, 0x2a, 0x88, 0x33, 0x28, 0xbc, 0x19, 0xdc, 0xe8, 0x9f, 0x0d,
+	0x26, 0x65, 0x5a, 0x2f, 0x2f, 0x8e, 0x62, 0xf3, 0x56, 0x4e, 0x58, 0x1e, 0x1a, 0xaa, 0xaf, 0x19,
+	0x14, 0x7b, 0x2c, 0x04, 0x64, 0xbd, 0x0a, 0xaf, 0x3c, 0xb5, 0x90, 0xac, 0xc5, 0x31, 0xa4, 0xa3,
+	0xd5, 0x38, 0x2b, 0x93, 0x3a, 0x95, 0x24, 0x89, 0x34, 0x56, 0x63, 0x1a, 0x49, 0x63, 0xb5, 0xd8,
+	0x40, 0x36, 0x0e, 0xc6, 0x63, 0x56, 0x26, 0xf5, 0xe2, 0x7a, 0x86, 0x89, 0x64, 0x2f, 0x10, 0xf2,
+	0xc6, 0xbb, 0xb1, 0xc7, 0xfc, 0x50, 0x88, 0x80, 0xfe, 0xd4, 0x3a, 0x6d, 0x70, 0x5e, 0x26, 0xf5,
+	0x5a, 0xb2, 0x26, 0x36, 0xd8, 0x4f, 0x83, 0xff, 0xca, 0xa4, 0xce, 0x24, 0x6b, 0xb1, 0x81, 0xb9,
+	0xb6, 0x8d, 0x19, 0x02, 0x16, 0x7c, 0xd3, 0xe4, 0x88, 0x07, 0xe5, 0x1b, 0x13, 0x70, 0x41, 0xab,
+	0xe5, 0xe4, 0xc4, 0x09, 0xe4, 0xad, 0x7a, 0x73, 0x1e, 0x81, 0x97, 0x44, 0xc3, 0xd4, 0x76, 0xce,
+	0xe3, 0x72, 0xa2, 0x64, 0x44, 0x05, 0xf9, 0xbb, 0x0a, 0xc1, 0xe3, 0x8a, 0x43, 0x5a, 0x4d, 0x21,
+	0x3d, 0xee, 0x42, 0xf0, 0x32, 0x96, 0xc4, 0x29, 0xa4, 0x4a, 0x0f, 0xb8, 0xfe, 0x15, 0xe3, 0xee,
+	0xe6, 0xe1, 0xb6, 0x0b, 0xfe, 0x43, 0x52, 0xad, 0x3a, 0x87, 0x9c, 0x47, 0xe8, 0xfe, 0x4e, 0xb5,
+	0x94, 0x39, 0x5d, 0xc4, 0x9a, 0x98, 0x56, 0x41, 0x71, 0x7c, 0x2b, 0xc9, 0xba, 0xba, 0x83, 0x62,
+	0xbf, 0xe1, 0xaf, 0x33, 0x3f, 0x72, 0x48, 0xe3, 0x7b, 0xa3, 0x7b, 0x9a, 0xf3, 0x45, 0x97, 0xdf,
+	0x01, 0x00, 0x00, 0xff, 0xff, 0xef, 0x27, 0x99, 0xf7, 0x17, 0x02, 0x00, 0x00,
+}

vendor/github.com/containerd/continuity/proto/manifest.proto

@@ -0,0 +1,97 @@
+syntax = "proto3";
+
+package proto;
+
+// Manifest specifies the entries in a container bundle, keyed and sorted by
+// path.
+message Manifest {
+    repeated Resource resource = 1;
+}
+
+message Resource {
+    // Path specifies the path from the bundle root. If more than one
+    // path is present, the entry may represent a hardlink, rather than using
+    // a link target. The path format is operating system specific.
+    repeated string path = 1;
+
+    // NOTE(stevvooe): Need to define clear precedence for user/group/uid/gid precedence.
+
+    // Uid specifies the user id for the resource.
+    int64 uid = 2;
+
+    // Gid specifies the group id for the resource.
+    int64 gid = 3;
+
+    // user and group are not currently used but their field numbers have been
+    // reserved for future use. As such, they are marked as deprecated.
+    string user = 4 [deprecated=true]; // "deprecated" stands for "reserved" here
+    string group = 5 [deprecated=true]; // "deprecated" stands for "reserved" here
+
+    // Mode defines the file mode and permissions. We've used the same
+    // bit-packing from Go's os package,
+    // http://golang.org/pkg/os/#FileMode, since they've done the work of
+    // creating a cross-platform layout.
+    uint32 mode = 6;
+
+    // NOTE(stevvooe): Beyond here, we start defining type specific fields.
+
+    // Size specifies the size in bytes of the resource. This is only valid
+    // for regular files.
+    uint64 size = 7;
+
+    // Digest specifies the content digest of the target file. Only valid for
+    // regular files. The strings are formatted in OCI style, i.e. <alg>:<encoded>.
+    // For detailed information about the format, please refer to OCI Image Spec:
+    // https://github.com/opencontainers/image-spec/blob/master/descriptor.md#digests-and-verification
+    // The digests are sorted in lexical order and implementations may choose
+    // which algorithms they prefer.
+    repeated string digest = 8;
+
+    // Target defines the target of a hard or soft link. Absolute links start
+    // with a slash and specify the resource relative to the bundle root.
+    // Relative links do not start with a slash and are relative to the
+    // resource path.
+    string target = 9;
+
+    // Major specifies the major device number for character and block devices.
+    uint64 major = 10;
+
+    // Minor specifies the minor device number for character and block devices.
+    uint64 minor = 11;
+
+    // Xattr provides storage for extended attributes for the target resource.
+    repeated XAttr xattr = 12;
+
+    // Ads stores one or more alternate data streams for the target resource.
+    repeated ADSEntry ads = 13;
+
+}
+
+// XAttr encodes extended attributes for a resource.
+message XAttr {
+    // Name specifies the attribute name.
+    string name = 1;
+
+    // Data specifies the associated data for the attribute.
+    bytes data = 2;
+}
+
+// ADSEntry encodes information for a Windows Alternate Data Stream.
+message ADSEntry {
+    // Name specifices the stream name.
+    string name = 1;
+
+    // Data specifies the stream data.
+    // See also the description about the digest below.
+    bytes data = 2;
+
+    // Digest is a CAS representation of the stream data.
+    //
+    // At least one of data or digest MUST be specified, and either one of them
+    // SHOULD be specified.
+    //
+    // How to access the actual data using the digest is implementation-specific,
+    // and implementations can choose not to implement digest.
+    // So, digest SHOULD be used only when the stream data is large.
+    string digest = 3;
+}

vendor/github.com/containerd/continuity/resource.go

@@ -0,0 +1,574 @@
+package continuity
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"reflect"
+	"sort"
+
+	pb "github.com/containerd/continuity/proto"
+	"github.com/opencontainers/go-digest"
+)
+
+// TODO(stevvooe): A record based model, somewhat sketched out at the bottom
+// of this file, will be more flexible. Another possibly is to tie the package
+// interface directly to the protobuf type. This will have efficiency
+// advantages at the cost coupling the nasty codegen types to the exported
+// interface.
+
+type Resource interface {
+	// Path provides the primary resource path relative to the bundle root. In
+	// cases where resources have more than one path, such as with hard links,
+	// this will return the primary path, which is often just the first entry.
+	Path() string
+
+	// Mode returns the
+	Mode() os.FileMode
+
+	UID() int64
+	GID() int64
+}
+
+// ByPath provides the canonical sort order for a set of resources. Use with
+// sort.Stable for deterministic sorting.
+type ByPath []Resource
+
+func (bp ByPath) Len() int           { return len(bp) }
+func (bp ByPath) Swap(i, j int)      { bp[i], bp[j] = bp[j], bp[i] }
+func (bp ByPath) Less(i, j int) bool { return bp[i].Path() < bp[j].Path() }
+
+type XAttrer interface {
+	XAttrs() map[string][]byte
+}
+
+// Hardlinkable is an interface that a resource type satisfies if it can be a
+// hardlink target.
+type Hardlinkable interface {
+	// Paths returns all paths of the resource, including the primary path
+	// returned by Resource.Path. If len(Paths()) > 1, the resource is a hard
+	// link.
+	Paths() []string
+}
+
+type RegularFile interface {
+	Resource
+	XAttrer
+	Hardlinkable
+
+	Size() int64
+	Digests() []digest.Digest
+}
+
+// Merge two or more Resources into new file. Typically, this should be
+// used to merge regular files as hardlinks. If the files are not identical,
+// other than Paths and Digests, the merge will fail and an error will be
+// returned.
+func Merge(fs ...Resource) (Resource, error) {
+	if len(fs) < 1 {
+		return nil, fmt.Errorf("please provide a resource to merge")
+	}
+
+	if len(fs) == 1 {
+		return fs[0], nil
+	}
+
+	var paths []string
+	var digests []digest.Digest
+	bypath := map[string][]Resource{}
+
+	// The attributes are all compared against the first to make sure they
+	// agree before adding to the above collections. If any of these don't
+	// correctly validate, the merge fails.
+	prototype := fs[0]
+	xattrs := make(map[string][]byte)
+
+	// initialize xattrs for use below. All files must have same xattrs.
+	if prototypeXAttrer, ok := prototype.(XAttrer); ok {
+		for attr, value := range prototypeXAttrer.XAttrs() {
+			xattrs[attr] = value
+		}
+	}
+
+	for _, f := range fs {
+		h, isHardlinkable := f.(Hardlinkable)
+		if !isHardlinkable {
+			return nil, errNotAHardLink
+		}
+
+		if f.Mode() != prototype.Mode() {
+			return nil, fmt.Errorf("modes do not match: %v != %v", f.Mode(), prototype.Mode())
+		}
+
+		if f.UID() != prototype.UID() {
+			return nil, fmt.Errorf("uid does not match: %v != %v", f.UID(), prototype.UID())
+		}
+
+		if f.GID() != prototype.GID() {
+			return nil, fmt.Errorf("gid does not match: %v != %v", f.GID(), prototype.GID())
+		}
+
+		if xattrer, ok := f.(XAttrer); ok {
+			fxattrs := xattrer.XAttrs()
+			if !reflect.DeepEqual(fxattrs, xattrs) {
+				return nil, fmt.Errorf("resource %q xattrs do not match: %v != %v", f, fxattrs, xattrs)
+			}
+		}
+
+		for _, p := range h.Paths() {
+			pfs, ok := bypath[p]
+			if !ok {
+				// ensure paths are unique by only appending on a new path.
+				paths = append(paths, p)
+			}
+
+			bypath[p] = append(pfs, f)
+		}
+
+		if regFile, isRegFile := f.(RegularFile); isRegFile {
+			prototypeRegFile, prototypeIsRegFile := prototype.(RegularFile)
+			if !prototypeIsRegFile {
+				return nil, errors.New("prototype is not a regular file")
+			}
+
+			if regFile.Size() != prototypeRegFile.Size() {
+				return nil, fmt.Errorf("size does not match: %v != %v", regFile.Size(), prototypeRegFile.Size())
+			}
+
+			digests = append(digests, regFile.Digests()...)
+		} else if device, isDevice := f.(Device); isDevice {
+			prototypeDevice, prototypeIsDevice := prototype.(Device)
+			if !prototypeIsDevice {
+				return nil, errors.New("prototype is not a device")
+			}
+
+			if device.Major() != prototypeDevice.Major() {
+				return nil, fmt.Errorf("major number does not match: %v != %v", device.Major(), prototypeDevice.Major())
+			}
+			if device.Minor() != prototypeDevice.Minor() {
+				return nil, fmt.Errorf("minor number does not match: %v != %v", device.Minor(), prototypeDevice.Minor())
+			}
+		} else if _, isNamedPipe := f.(NamedPipe); isNamedPipe {
+			_, prototypeIsNamedPipe := prototype.(NamedPipe)
+			if !prototypeIsNamedPipe {
+				return nil, errors.New("prototype is not a named pipe")
+			}
+		} else {
+			return nil, errNotAHardLink
+		}
+	}
+
+	sort.Stable(sort.StringSlice(paths))
+
+	// Choose a "canonical" file. Really, it is just the first file to sort
+	// against. We also effectively select the very first digest as the
+	// "canonical" one for this file.
+	first := bypath[paths[0]][0]
+
+	resource := resource{
+		paths:  paths,
+		mode:   first.Mode(),
+		uid:    first.UID(),
+		gid:    first.GID(),
+		xattrs: xattrs,
+	}
+
+	switch typedF := first.(type) {
+	case RegularFile:
+		var err error
+		digests, err = uniqifyDigests(digests...)
+		if err != nil {
+			return nil, err
+		}
+
+		return &regularFile{
+			resource: resource,
+			size:     typedF.Size(),
+			digests:  digests,
+		}, nil
+	case Device:
+		return &device{
+			resource: resource,
+			major:    typedF.Major(),
+			minor:    typedF.Minor(),
+		}, nil
+
+	case NamedPipe:
+		return &namedPipe{
+			resource: resource,
+		}, nil
+
+	default:
+		return nil, errNotAHardLink
+	}
+}
+
+type Directory interface {
+	Resource
+	XAttrer
+
+	// Directory is a no-op method to identify directory objects by interface.
+	Directory()
+}
+
+type SymLink interface {
+	Resource
+
+	// Target returns the target of the symlink contained in the .
+	Target() string
+}
+
+type NamedPipe interface {
+	Resource
+	Hardlinkable
+	XAttrer
+
+	// Pipe is a no-op method to allow consistent resolution of NamedPipe
+	// interface.
+	Pipe()
+}
+
+type Device interface {
+	Resource
+	Hardlinkable
+	XAttrer
+
+	Major() uint64
+	Minor() uint64
+}
+
+type resource struct {
+	paths    []string
+	mode     os.FileMode
+	uid, gid int64
+	xattrs   map[string][]byte
+}
+
+var _ Resource = &resource{}
+
+func (r *resource) Path() string {
+	if len(r.paths) < 1 {
+		return ""
+	}
+
+	return r.paths[0]
+}
+
+func (r *resource) Mode() os.FileMode {
+	return r.mode
+}
+
+func (r *resource) UID() int64 {
+	return r.uid
+}
+
+func (r *resource) GID() int64 {
+	return r.gid
+}
+
+type regularFile struct {
+	resource
+	size    int64
+	digests []digest.Digest
+}
+
+var _ RegularFile = &regularFile{}
+
+// newRegularFile returns the RegularFile, using the populated base resource
+// and one or more digests of the content.
+func newRegularFile(base resource, paths []string, size int64, dgsts ...digest.Digest) (RegularFile, error) {
+	if !base.Mode().IsRegular() {
+		return nil, fmt.Errorf("not a regular file")
+	}
+
+	base.paths = make([]string, len(paths))
+	copy(base.paths, paths)
+
+	// make our own copy of digests
+	ds := make([]digest.Digest, len(dgsts))
+	copy(ds, dgsts)
+
+	return &regularFile{
+		resource: base,
+		size:     size,
+		digests:  ds,
+	}, nil
+}
+
+func (rf *regularFile) Paths() []string {
+	paths := make([]string, len(rf.paths))
+	copy(paths, rf.paths)
+	return paths
+}
+
+func (rf *regularFile) Size() int64 {
+	return rf.size
+}
+
+func (rf *regularFile) Digests() []digest.Digest {
+	digests := make([]digest.Digest, len(rf.digests))
+	copy(digests, rf.digests)
+	return digests
+}
+
+func (rf *regularFile) XAttrs() map[string][]byte {
+	xattrs := make(map[string][]byte, len(rf.xattrs))
+
+	for attr, value := range rf.xattrs {
+		xattrs[attr] = append(xattrs[attr], value...)
+	}
+
+	return xattrs
+}
+
+type directory struct {
+	resource
+}
+
+var _ Directory = &directory{}
+
+func newDirectory(base resource) (Directory, error) {
+	if !base.Mode().IsDir() {
+		return nil, fmt.Errorf("not a directory")
+	}
+
+	return &directory{
+		resource: base,
+	}, nil
+}
+
+func (d *directory) Directory() {}
+
+func (d *directory) XAttrs() map[string][]byte {
+	xattrs := make(map[string][]byte, len(d.xattrs))
+
+	for attr, value := range d.xattrs {
+		xattrs[attr] = append(xattrs[attr], value...)
+	}
+
+	return xattrs
+}
+
+type symLink struct {
+	resource
+	target string
+}
+
+var _ SymLink = &symLink{}
+
+func newSymLink(base resource, target string) (SymLink, error) {
+	if base.Mode()&os.ModeSymlink == 0 {
+		return nil, fmt.Errorf("not a symlink")
+	}
+
+	return &symLink{
+		resource: base,
+		target:   target,
+	}, nil
+}
+
+func (l *symLink) Target() string {
+	return l.target
+}
+
+type namedPipe struct {
+	resource
+}
+
+var _ NamedPipe = &namedPipe{}
+
+func newNamedPipe(base resource, paths []string) (NamedPipe, error) {
+	if base.Mode()&os.ModeNamedPipe == 0 {
+		return nil, fmt.Errorf("not a namedpipe")
+	}
+
+	base.paths = make([]string, len(paths))
+	copy(base.paths, paths)
+
+	return &namedPipe{
+		resource: base,
+	}, nil
+}
+
+func (np *namedPipe) Pipe() {}
+
+func (np *namedPipe) Paths() []string {
+	paths := make([]string, len(np.paths))
+	copy(paths, np.paths)
+	return paths
+}
+
+func (np *namedPipe) XAttrs() map[string][]byte {
+	xattrs := make(map[string][]byte, len(np.xattrs))
+
+	for attr, value := range np.xattrs {
+		xattrs[attr] = append(xattrs[attr], value...)
+	}
+
+	return xattrs
+}
+
+type device struct {
+	resource
+	major, minor uint64
+}
+
+var _ Device = &device{}
+
+func newDevice(base resource, paths []string, major, minor uint64) (Device, error) {
+	if base.Mode()&os.ModeDevice == 0 {
+		return nil, fmt.Errorf("not a device")
+	}
+
+	base.paths = make([]string, len(paths))
+	copy(base.paths, paths)
+
+	return &device{
+		resource: base,
+		major:    major,
+		minor:    minor,
+	}, nil
+}
+
+func (d *device) Paths() []string {
+	paths := make([]string, len(d.paths))
+	copy(paths, d.paths)
+	return paths
+}
+
+func (d *device) XAttrs() map[string][]byte {
+	xattrs := make(map[string][]byte, len(d.xattrs))
+
+	for attr, value := range d.xattrs {
+		xattrs[attr] = append(xattrs[attr], value...)
+	}
+
+	return xattrs
+}
+
+func (d device) Major() uint64 {
+	return d.major
+}
+
+func (d device) Minor() uint64 {
+	return d.minor
+}
+
+// toProto converts a resource to a protobuf record. We'd like to push this
+// the individual types but we want to keep this all together during
+// prototyping.
+func toProto(resource Resource) *pb.Resource {
+	b := &pb.Resource{
+		Path: []string{resource.Path()},
+		Mode: uint32(resource.Mode()),
+		Uid:  resource.UID(),
+		Gid:  resource.GID(),
+	}
+
+	if xattrer, ok := resource.(XAttrer); ok {
+		// Sorts the XAttrs by name for consistent ordering.
+		keys := []string{}
+		xattrs := xattrer.XAttrs()
+		for k := range xattrs {
+			keys = append(keys, k)
+		}
+		sort.Strings(keys)
+
+		for _, k := range keys {
+			b.Xattr = append(b.Xattr, &pb.XAttr{Name: k, Data: xattrs[k]})
+		}
+	}
+
+	switch r := resource.(type) {
+	case RegularFile:
+		b.Path = r.Paths()
+		b.Size = uint64(r.Size())
+
+		for _, dgst := range r.Digests() {
+			b.Digest = append(b.Digest, dgst.String())
+		}
+	case SymLink:
+		b.Target = r.Target()
+	case Device:
+		b.Major, b.Minor = r.Major(), r.Minor()
+		b.Path = r.Paths()
+	case NamedPipe:
+		b.Path = r.Paths()
+	}
+
+	// enforce a few stability guarantees that may not be provided by the
+	// resource implementation.
+	sort.Strings(b.Path)
+
+	return b
+}
+
+// fromProto converts from a protobuf Resource to a Resource interface.
+func fromProto(b *pb.Resource) (Resource, error) {
+	base := &resource{
+		paths: b.Path,
+		mode:  os.FileMode(b.Mode),
+		uid:   b.Uid,
+		gid:   b.Gid,
+	}
+
+	base.xattrs = make(map[string][]byte, len(b.Xattr))
+
+	for _, attr := range b.Xattr {
+		base.xattrs[attr.Name] = attr.Data
+	}
+
+	switch {
+	case base.Mode().IsRegular():
+		dgsts := make([]digest.Digest, len(b.Digest))
+		for i, dgst := range b.Digest {
+			// TODO(stevvooe): Should we be validating at this point?
+			dgsts[i] = digest.Digest(dgst)
+		}
+
+		return newRegularFile(*base, b.Path, int64(b.Size), dgsts...)
+	case base.Mode().IsDir():
+		return newDirectory(*base)
+	case base.Mode()&os.ModeSymlink != 0:
+		return newSymLink(*base, b.Target)
+	case base.Mode()&os.ModeNamedPipe != 0:
+		return newNamedPipe(*base, b.Path)
+	case base.Mode()&os.ModeDevice != 0:
+		return newDevice(*base, b.Path, b.Major, b.Minor)
+	}
+
+	return nil, fmt.Errorf("unknown resource record (%#v): %s", b, base.Mode())
+}
+
+// NOTE(stevvooe): An alternative model that supports inline declaration.
+// Convenient for unit testing where inline declarations may be desirable but
+// creates an awkward API for the standard use case.
+
+// type ResourceKind int
+
+// const (
+// 	ResourceRegularFile = iota + 1
+// 	ResourceDirectory
+// 	ResourceSymLink
+// 	Resource
+// )
+
+// type Resource struct {
+// 	Kind         ResourceKind
+// 	Paths        []string
+// 	Mode         os.FileMode
+// 	UID          string
+// 	GID          string
+// 	Size         int64
+// 	Digests      []digest.Digest
+// 	Target       string
+// 	Major, Minor int
+// 	XAttrs       map[string][]byte
+// }
+
+// type RegularFile struct {
+// 	Paths   []string
+//  Size 	int64
+// 	Digests []digest.Digest
+// 	Perm    os.FileMode // os.ModePerm + sticky, setuid, setgid
+// }

vendor/github.com/containerd/continuity/resource_unix.go

@@ -0,0 +1,37 @@
+// +build linux darwin freebsd solaris
+
+package continuity
+
+import (
+	"fmt"
+	"os"
+	"syscall"
+)
+
+// newBaseResource returns a *resource, populated with data from p and fi,
+// where p will be populated directly.
+func newBaseResource(p string, fi os.FileInfo) (*resource, error) {
+	// TODO(stevvooe): This need to be resolved for the container's root,
+	// where here we are really getting the host OS's value. We need to allow
+	// this be passed in and fixed up to make these uid/gid mappings portable.
+	// Either this can be part of the driver or we can achieve it through some
+	// other mechanism.
+	sys, ok := fi.Sys().(*syscall.Stat_t)
+	if !ok {
+		// TODO(stevvooe): This may not be a hard error for all platforms. We
+		// may want to move this to the driver.
+		return nil, fmt.Errorf("unable to resolve syscall.Stat_t from (os.FileInfo).Sys(): %#v", fi)
+	}
+
+	return &resource{
+		paths: []string{p},
+		mode:  fi.Mode(),
+
+		uid: int64(sys.Uid),
+		gid: int64(sys.Gid),
+
+		// NOTE(stevvooe): Population of shared xattrs field is deferred to
+		// the resource types that populate it. Since they are a property of
+		// the context, they must set there.
+	}, nil
+}