Runtime v2

Signed-off-by: Michael Crosby <crosbymichael@gmail.com>

runtime/v2/runc/options/doc.go

@@ -0,0 +1,17 @@
+/*
+   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 options

runtime/v2/runc/options/oci.proto

@@ -0,0 +1,39 @@
+syntax = "proto3";
+
+package containerd.runc.v1;
+
+import weak "gogoproto/gogo.proto";
+
+option go_package = "github.com/containerd/containerd/runtime/v2/runc/options;options";
+
+message Options {
+	bool no_pivot_root = 1;
+	bool open_tcp = 2;
+	bool external_unix_sockets = 3;
+	bool terminal = 4;
+	bool file_locks = 5;
+	repeated string empty_namespaces = 6;
+	string cgroups_mode = 7;
+	bool no_new_keyring = 8;
+	string shim_cgroup = 9;
+	uint32 io_uid = 10;
+	uint32 io_gid = 11;
+	string binary_name = 12;
+	string root = 13;
+	string criu_path = 14;
+	bool systemd_cgroup = 15;
+}
+
+message CheckpointOptions {
+	bool exit = 1;
+	bool open_tcp = 2;
+	bool external_unix_sockets = 3;
+	bool terminal = 4;
+	bool file_locks = 5;
+	repeated string empty_namespaces = 6;
+	string cgroups_mode = 7;
+}
+
+message ProcessDetails {
+	string exec_id = 1;
+}

runtime/v2/runc/service.go

@@ -0,0 +1,665 @@
+// +build linux
+
+/*
+   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 runc
+
+import (
+	"context"
+	"io/ioutil"
+	"os"
+	"path/filepath"
+	"sync"
+	"time"
+
+	"github.com/containerd/cgroups"
+	"github.com/containerd/console"
+	eventstypes "github.com/containerd/containerd/api/events"
+	"github.com/containerd/containerd/api/types/task"
+	"github.com/containerd/containerd/errdefs"
+	"github.com/containerd/containerd/events"
+	"github.com/containerd/containerd/mount"
+	"github.com/containerd/containerd/namespaces"
+	"github.com/containerd/containerd/runtime"
+	rproc "github.com/containerd/containerd/runtime/proc"
+	"github.com/containerd/containerd/runtime/v1/linux/proc"
+	"github.com/containerd/containerd/runtime/v2/runc/options"
+	"github.com/containerd/containerd/runtime/v2/shim"
+	taskAPI "github.com/containerd/containerd/runtime/v2/task"
+	runcC "github.com/containerd/go-runc"
+	"github.com/containerd/typeurl"
+	ptypes "github.com/gogo/protobuf/types"
+	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
+)
+
+var (
+	empty   = &ptypes.Empty{}
+	bufPool = sync.Pool{
+		New: func() interface{} {
+			buffer := make([]byte, 32<<10)
+			return &buffer
+		},
+	}
+)
+
+var _ = (taskAPI.TaskService)(&service{})
+
+// New returns a new shim service that can be used via GRPC
+func New(ctx context.Context, id string, publisher events.Publisher) (shim.Shim, error) {
+	s := &service{
+		id:        id,
+		context:   ctx,
+		processes: make(map[string]rproc.Process),
+		events:    make(chan interface{}, 128),
+		ec:        shim.Default.Subscribe(),
+	}
+	go s.processExits()
+	runcC.Monitor = shim.Default
+	if err := s.initPlatform(); err != nil {
+		return nil, errors.Wrap(err, "failed to initialized platform behavior")
+	}
+	go s.forward(publisher)
+	return s, nil
+}
+
+// service is the shim implementation of a remote shim over GRPC
+type service struct {
+	mu sync.Mutex
+
+	context   context.Context
+	processes map[string]rproc.Process
+	events    chan interface{}
+	platform  rproc.Platform
+	ec        chan runcC.Exit
+
+	id string
+	// Filled by Create()
+	bundle string
+	cg     cgroups.Cgroup
+}
+
+func (s *service) Cleanup(ctx context.Context) (*taskAPI.DeleteResponse, error) {
+	path, err := os.Getwd()
+	if err != nil {
+		return nil, err
+	}
+	ns, err := namespaces.NamespaceRequired(ctx)
+	if err != nil {
+		return nil, err
+	}
+	runtime, _ := s.readRuntime(path)
+	if runtime != "" {
+		r := proc.NewRunc(proc.RuncRoot, path, ns, runtime, "", false)
+		if err := r.Delete(ctx, s.id, &runcC.DeleteOpts{
+			Force: true,
+		}); err != nil {
+			logrus.WithError(err).Warn("runc delete")
+		}
+	}
+	if err := mount.UnmountAll(filepath.Join(path, "rootfs"), 0); err != nil {
+		logrus.WithError(err).Warn("failed to cleanup rootfs mount")
+	}
+	return &taskAPI.DeleteResponse{
+		ExitedAt:   time.Now(),
+		ExitStatus: 128 + uint32(unix.SIGKILL),
+	}, nil
+}
+
+func (s *service) readRuntime(path string) (string, error) {
+	data, err := ioutil.ReadFile(filepath.Join(path, "runtime"))
+	if err != nil {
+		return "", err
+	}
+	return string(data), nil
+}
+
+func (s *service) writeRuntime(path, runtime string) error {
+	return ioutil.WriteFile(filepath.Join(path, "runtime"), []byte(runtime), 0600)
+}
+
+// Create a new initial process and container with the underlying OCI runtime
+func (s *service) Create(ctx context.Context, r *taskAPI.CreateTaskRequest) (_ *taskAPI.CreateTaskResponse, err error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	ns, err := namespaces.NamespaceRequired(ctx)
+	if err != nil {
+		return nil, errors.Wrap(err, "create namespace")
+	}
+
+	var opts options.Options
+	if r.Options != nil {
+		v, err := typeurl.UnmarshalAny(r.Options)
+		if err != nil {
+			return nil, err
+		}
+		opts = *v.(*options.Options)
+	}
+
+	var mounts []proc.Mount
+	for _, m := range r.Rootfs {
+		mounts = append(mounts, proc.Mount{
+			Type:    m.Type,
+			Source:  m.Source,
+			Target:  m.Target,
+			Options: m.Options,
+		})
+	}
+	config := &proc.CreateConfig{
+		ID:               r.ID,
+		Bundle:           r.Bundle,
+		Runtime:          opts.BinaryName,
+		Rootfs:           mounts,
+		Terminal:         r.Terminal,
+		Stdin:            r.Stdin,
+		Stdout:           r.Stdout,
+		Stderr:           r.Stderr,
+		Checkpoint:       r.Checkpoint,
+		ParentCheckpoint: r.ParentCheckpoint,
+		Options:          r.Options,
+	}
+	if err := s.writeRuntime(r.Bundle, opts.BinaryName); err != nil {
+		return nil, err
+	}
+	rootfs := filepath.Join(r.Bundle, "rootfs")
+	defer func() {
+		if err != nil {
+			if err2 := mount.UnmountAll(rootfs, 0); err2 != nil {
+				logrus.WithError(err2).Warn("failed to cleanup rootfs mount")
+			}
+		}
+	}()
+	for _, rm := range mounts {
+		m := &mount.Mount{
+			Type:    rm.Type,
+			Source:  rm.Source,
+			Options: rm.Options,
+		}
+		if err := m.Mount(rootfs); err != nil {
+			return nil, errors.Wrapf(err, "failed to mount rootfs component %v", m)
+		}
+	}
+	process, err := newInit(
+		ctx,
+		r.Bundle,
+		filepath.Join(r.Bundle, "work"),
+		ns,
+		s.platform,
+		config,
+		&opts,
+	)
+	if err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	if err := process.Create(ctx, config); err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	// save the main task id and bundle to the shim for additional requests
+	s.id = r.ID
+	s.bundle = r.Bundle
+	pid := process.Pid()
+	if pid > 0 {
+		cg, err := cgroups.Load(cgroups.V1, cgroups.PidPath(pid))
+		if err != nil {
+			logrus.WithError(err).Errorf("loading cgroup for %d", pid)
+		}
+		s.cg = cg
+	}
+	s.processes[r.ID] = process
+	return &taskAPI.CreateTaskResponse{
+		Pid: uint32(pid),
+	}, nil
+
+}
+
+// Start a process
+func (s *service) Start(ctx context.Context, r *taskAPI.StartRequest) (*taskAPI.StartResponse, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[r.ID]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrNotFound, "process %s", r.ID)
+	}
+	if err := p.Start(ctx); err != nil {
+		return nil, err
+	}
+	if s.cg == nil && p.Pid() > 0 {
+		cg, err := cgroups.Load(cgroups.V1, cgroups.PidPath(p.Pid()))
+		if err != nil {
+			logrus.WithError(err).Errorf("loading cgroup for %d", p.Pid())
+		}
+		s.cg = cg
+	}
+	return &taskAPI.StartResponse{
+		ID:  p.ID(),
+		Pid: uint32(p.Pid()),
+	}, nil
+}
+
+// Delete the initial process and container
+func (s *service) Delete(ctx context.Context, r *taskAPI.DeleteRequest) (*taskAPI.DeleteResponse, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	p := s.processes[r.ID]
+	if p == nil {
+		return nil, errors.Wrapf(errdefs.ErrNotFound, "process %s", r.ID)
+	}
+	if err := p.Delete(ctx); err != nil {
+		return nil, err
+	}
+	delete(s.processes, r.ID)
+	if r.ID == s.id && s.platform != nil {
+		s.platform.Close()
+	}
+	return &taskAPI.DeleteResponse{
+		ExitStatus: uint32(p.ExitStatus()),
+		ExitedAt:   p.ExitedAt(),
+		Pid:        uint32(p.Pid()),
+	}, nil
+}
+
+// Exec an additional process inside the container
+func (s *service) Exec(ctx context.Context, r *taskAPI.ExecProcessRequest) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if p := s.processes[r.ID]; p != nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrAlreadyExists, "id %s", r.ID)
+	}
+
+	p := s.processes[s.id]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+	}
+
+	process, err := p.(*proc.Init).Exec(ctx, s.bundle, &proc.ExecConfig{
+		ID:       r.ID,
+		Terminal: r.Terminal,
+		Stdin:    r.Stdin,
+		Stdout:   r.Stdout,
+		Stderr:   r.Stderr,
+		Spec:     r.Spec,
+	})
+	if err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	s.processes[r.ID] = process
+	return empty, nil
+}
+
+// ResizePty of a process
+func (s *service) ResizePty(ctx context.Context, r *taskAPI.ResizePtyRequest) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if r.ID == "" {
+		return nil, errdefs.ToGRPCf(errdefs.ErrInvalidArgument, "id not provided")
+	}
+	ws := console.WinSize{
+		Width:  uint16(r.Width),
+		Height: uint16(r.Height),
+	}
+	p := s.processes[r.ID]
+	if p == nil {
+		return nil, errors.Errorf("process does not exist %s", r.ID)
+	}
+	if err := p.Resize(ws); err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	return empty, nil
+}
+
+// State returns runtime state information for a process
+func (s *service) State(ctx context.Context, r *taskAPI.StateRequest) (*taskAPI.StateResponse, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[r.ID]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrNotFound, "process id %s", r.ID)
+	}
+	st, err := p.Status(ctx)
+	if err != nil {
+		return nil, err
+	}
+	status := task.StatusUnknown
+	switch st {
+	case "created":
+		status = task.StatusCreated
+	case "running":
+		status = task.StatusRunning
+	case "stopped":
+		status = task.StatusStopped
+	case "paused":
+		status = task.StatusPaused
+	case "pausing":
+		status = task.StatusPausing
+	}
+	sio := p.Stdio()
+	return &taskAPI.StateResponse{
+		ID:         p.ID(),
+		Bundle:     s.bundle,
+		Pid:        uint32(p.Pid()),
+		Status:     status,
+		Stdin:      sio.Stdin,
+		Stdout:     sio.Stdout,
+		Stderr:     sio.Stderr,
+		Terminal:   sio.Terminal,
+		ExitStatus: uint32(p.ExitStatus()),
+		ExitedAt:   p.ExitedAt(),
+	}, nil
+}
+
+// Pause the container
+func (s *service) Pause(ctx context.Context, r *ptypes.Empty) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[s.id]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+	}
+	if err := p.(*proc.Init).Pause(ctx); err != nil {
+		return nil, err
+	}
+	return empty, nil
+}
+
+// Resume the container
+func (s *service) Resume(ctx context.Context, r *ptypes.Empty) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[s.id]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+	}
+	if err := p.(*proc.Init).Resume(ctx); err != nil {
+		return nil, err
+	}
+	return empty, nil
+}
+
+// Kill a process with the provided signal
+func (s *service) Kill(ctx context.Context, r *taskAPI.KillRequest) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if r.ID == "" {
+		p := s.processes[s.id]
+		if p == nil {
+			return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+		}
+		if err := p.Kill(ctx, r.Signal, r.All); err != nil {
+			return nil, errdefs.ToGRPC(err)
+		}
+		return empty, nil
+	}
+
+	p := s.processes[r.ID]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrNotFound, "process id %s not found", r.ID)
+	}
+	if err := p.Kill(ctx, r.Signal, r.All); err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	return empty, nil
+}
+
+// Pids returns all pids inside the container
+func (s *service) Pids(ctx context.Context, r *taskAPI.PidsRequest) (*taskAPI.PidsResponse, error) {
+	pids, err := s.getContainerPids(ctx, r.ID)
+	if err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	var processes []*task.ProcessInfo
+	for _, pid := range pids {
+		pInfo := task.ProcessInfo{
+			Pid: pid,
+		}
+		for _, p := range s.processes {
+			if p.Pid() == int(pid) {
+				d := &options.ProcessDetails{
+					ExecID: p.ID(),
+				}
+				a, err := typeurl.MarshalAny(d)
+				if err != nil {
+					return nil, errors.Wrapf(err, "failed to marshal process %d info", pid)
+				}
+				pInfo.Info = a
+				break
+			}
+		}
+		processes = append(processes, &pInfo)
+	}
+	return &taskAPI.PidsResponse{
+		Processes: processes,
+	}, nil
+}
+
+// CloseIO of a process
+func (s *service) CloseIO(ctx context.Context, r *taskAPI.CloseIORequest) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[r.ID]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrNotFound, "process does not exist %s", r.ID)
+	}
+	if stdin := p.Stdin(); stdin != nil {
+		if err := stdin.Close(); err != nil {
+			return nil, errors.Wrap(err, "close stdin")
+		}
+	}
+	return empty, nil
+}
+
+// Checkpoint the container
+func (s *service) Checkpoint(ctx context.Context, r *taskAPI.CheckpointTaskRequest) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[s.id]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+	}
+	var opts options.CheckpointOptions
+	if r.Options != nil {
+		v, err := typeurl.UnmarshalAny(r.Options)
+		if err != nil {
+			return nil, err
+		}
+		opts = *v.(*options.CheckpointOptions)
+	}
+	if err := p.(*proc.Init).Checkpoint(ctx, &proc.CheckpointConfig{
+		Path:                     r.Path,
+		Exit:                     opts.Exit,
+		AllowOpenTCP:             opts.OpenTcp,
+		AllowExternalUnixSockets: opts.ExternalUnixSockets,
+		AllowTerminal:            opts.Terminal,
+		FileLocks:                opts.FileLocks,
+		EmptyNamespaces:          opts.EmptyNamespaces,
+	}); err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	return empty, nil
+}
+
+// Connect returns shim information such as the shim's pid
+func (s *service) Connect(ctx context.Context, r *taskAPI.ConnectRequest) (*taskAPI.ConnectResponse, error) {
+	return &taskAPI.ConnectResponse{
+		ShimPid: uint32(os.Getpid()),
+		TaskPid: uint32(s.processes[s.id].Pid()),
+	}, nil
+}
+
+func (s *service) Shutdown(ctx context.Context, r *taskAPI.ShutdownRequest) (*ptypes.Empty, error) {
+	os.Exit(0)
+	return empty, nil
+}
+
+func (s *service) Stats(ctx context.Context, r *taskAPI.StatsRequest) (*taskAPI.StatsResponse, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	if s.cg == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrNotFound, "cgroup does not exist")
+	}
+	stats, err := s.cg.Stat(cgroups.IgnoreNotExist)
+	if err != nil {
+		return nil, err
+	}
+	data, err := typeurl.MarshalAny(stats)
+	if err != nil {
+		return nil, err
+	}
+	return &taskAPI.StatsResponse{
+		Stats: data,
+	}, nil
+}
+
+// Update a running container
+func (s *service) Update(ctx context.Context, r *taskAPI.UpdateTaskRequest) (*ptypes.Empty, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[s.id]
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+	}
+	if err := p.(*proc.Init).Update(ctx, r.Resources); err != nil {
+		return nil, errdefs.ToGRPC(err)
+	}
+	return empty, nil
+}
+
+// Wait for a process to exit
+func (s *service) Wait(ctx context.Context, r *taskAPI.WaitRequest) (*taskAPI.WaitResponse, error) {
+	s.mu.Lock()
+	p := s.processes[r.ID]
+	s.mu.Unlock()
+	if p == nil {
+		return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+	}
+	p.Wait()
+
+	return &taskAPI.WaitResponse{
+		ExitStatus: uint32(p.ExitStatus()),
+		ExitedAt:   p.ExitedAt(),
+	}, nil
+}
+
+func (s *service) processExits() {
+	for e := range s.ec {
+		s.checkProcesses(e)
+	}
+}
+
+func (s *service) checkProcesses(e runcC.Exit) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	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 {
+					logrus.WithError(err).WithField("id", ip.ID()).
+						Error("failed to kill init's children")
+				}
+			}
+			p.SetExited(e.Status)
+			s.events <- &eventstypes.TaskExit{
+				ContainerID: s.id,
+				ID:          p.ID(),
+				Pid:         uint32(e.Pid),
+				ExitStatus:  uint32(e.Status),
+				ExitedAt:    p.ExitedAt(),
+			}
+			return
+		}
+	}
+}
+
+func (s *service) getContainerPids(ctx context.Context, id string) ([]uint32, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	p := s.processes[s.id]
+	if p == nil {
+		return nil, errors.Wrapf(errdefs.ErrFailedPrecondition, "container must be created")
+	}
+
+	ps, err := p.(*proc.Init).Runtime().Ps(ctx, id)
+	if err != nil {
+		return nil, err
+	}
+	pids := make([]uint32, 0, len(ps))
+	for _, pid := range ps {
+		pids = append(pids, uint32(pid))
+	}
+	return pids, nil
+}
+
+func (s *service) forward(publisher events.Publisher) {
+	for e := range s.events {
+		if err := publisher.Publish(s.context, getTopic(s.context, e), e); err != nil {
+			logrus.WithError(err).Error("post event")
+		}
+	}
+}
+
+func getTopic(ctx context.Context, e interface{}) string {
+	switch e.(type) {
+	case *eventstypes.TaskCreate:
+		return runtime.TaskCreateEventTopic
+	case *eventstypes.TaskStart:
+		return runtime.TaskStartEventTopic
+	case *eventstypes.TaskOOM:
+		return runtime.TaskOOMEventTopic
+	case *eventstypes.TaskExit:
+		return runtime.TaskExitEventTopic
+	case *eventstypes.TaskDelete:
+		return runtime.TaskDeleteEventTopic
+	case *eventstypes.TaskExecAdded:
+		return runtime.TaskExecAddedEventTopic
+	case *eventstypes.TaskExecStarted:
+		return runtime.TaskExecStartedEventTopic
+	case *eventstypes.TaskPaused:
+		return runtime.TaskPausedEventTopic
+	case *eventstypes.TaskResumed:
+		return runtime.TaskResumedEventTopic
+	case *eventstypes.TaskCheckpointed:
+		return runtime.TaskCheckpointedEventTopic
+	default:
+		logrus.Warnf("no topic for type %#v", e)
+	}
+	return runtime.TaskUnknownTopic
+}
+
+func newInit(ctx context.Context, path, workDir, namespace string, platform rproc.Platform, r *proc.CreateConfig, options *options.Options) (*proc.Init, error) {
+	rootfs := filepath.Join(path, "rootfs")
+	runtime := proc.NewRunc(options.Root, path, namespace, options.BinaryName, options.CriuPath, options.SystemdCgroup)
+	p := proc.New(r.ID, runtime, rproc.Stdio{
+		Stdin:    r.Stdin,
+		Stdout:   r.Stdout,
+		Stderr:   r.Stderr,
+		Terminal: r.Terminal,
+	})
+	p.Bundle = r.Bundle
+	p.Platform = platform
+	p.Rootfs = rootfs
+	p.WorkDir = workDir
+	p.IoUID = int(options.IoUid)
+	p.IoGID = int(options.IoGid)
+	p.NoPivotRoot = options.NoPivotRoot
+	p.NoNewKeyring = options.NoNewKeyring
+	return p, nil
+}

runtime/v2/runc/service_linux.go

@@ -0,0 +1,111 @@
+/*
+   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 runc
+
+import (
+	"context"
+	"io"
+	"sync"
+	"syscall"
+
+	"github.com/containerd/console"
+	"github.com/containerd/fifo"
+	"github.com/pkg/errors"
+)
+
+type linuxPlatform struct {
+	epoller *console.Epoller
+}
+
+func (p *linuxPlatform) CopyConsole(ctx context.Context, console console.Console, stdin, stdout, stderr string, wg, cwg *sync.WaitGroup) (console.Console, error) {
+	if p.epoller == nil {
+		return nil, errors.New("uninitialized epoller")
+	}
+
+	epollConsole, err := p.epoller.Add(console)
+	if err != nil {
+		return nil, err
+	}
+
+	if stdin != "" {
+		in, err := fifo.OpenFifo(ctx, stdin, syscall.O_RDONLY, 0)
+		if err != nil {
+			return nil, err
+		}
+		cwg.Add(1)
+		go func() {
+			cwg.Done()
+			p := bufPool.Get().(*[]byte)
+			defer bufPool.Put(p)
+			io.CopyBuffer(epollConsole, in, *p)
+		}()
+	}
+
+	outw, err := fifo.OpenFifo(ctx, stdout, syscall.O_WRONLY, 0)
+	if err != nil {
+		return nil, err
+	}
+	outr, err := fifo.OpenFifo(ctx, stdout, syscall.O_RDONLY, 0)
+	if err != nil {
+		return nil, err
+	}
+	wg.Add(1)
+	cwg.Add(1)
+	go func() {
+		cwg.Done()
+		p := bufPool.Get().(*[]byte)
+		defer bufPool.Put(p)
+		io.CopyBuffer(outw, epollConsole, *p)
+		epollConsole.Close()
+		outr.Close()
+		outw.Close()
+		wg.Done()
+	}()
+	return epollConsole, nil
+}
+
+func (p *linuxPlatform) ShutdownConsole(ctx context.Context, cons console.Console) error {
+	if p.epoller == nil {
+		return errors.New("uninitialized epoller")
+	}
+	epollConsole, ok := cons.(*console.EpollConsole)
+	if !ok {
+		return errors.Errorf("expected EpollConsole, got %#v", cons)
+	}
+	return epollConsole.Shutdown(p.epoller.CloseConsole)
+}
+
+func (p *linuxPlatform) Close() error {
+	return p.epoller.Close()
+}
+
+// initialize a single epoll fd to manage our consoles. `initPlatform` should
+// only be called once.
+func (s *service) initPlatform() error {
+	if s.platform != nil {
+		return nil
+	}
+	epoller, err := console.NewEpoller()
+	if err != nil {
+		return errors.Wrap(err, "failed to initialize epoller")
+	}
+	s.platform = &linuxPlatform{
+		epoller: epoller,
+	}
+	go epoller.Wait()
+	return nil
+}