summaryrefslogtreecommitdiffhomepage
path: root/pkg/shim/v1
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/shim/v1')
-rw-r--r--pkg/shim/v1/proc/deleted_state.go49
-rw-r--r--pkg/shim/v1/proc/exec.go282
-rw-r--r--pkg/shim/v1/proc/exec_state.go154
-rw-r--r--pkg/shim/v1/proc/init.go462
-rw-r--r--pkg/shim/v1/proc/init_state.go182
-rw-r--r--pkg/shim/v1/proc/io.go167
-rw-r--r--pkg/shim/v1/proc/process.go37
-rw-r--r--pkg/shim/v1/proc/types.go70
-rw-r--r--pkg/shim/v1/proc/utils.go92
-rw-r--r--pkg/shim/v1/shim/platform.go110
-rw-r--r--pkg/shim/v1/shim/service.go579
-rw-r--r--pkg/shim/v1/utils/utils.go57
-rw-r--r--pkg/shim/v1/utils/volumes.go156
-rw-r--r--pkg/shim/v1/utils/volumes_test.go309
14 files changed, 2706 insertions, 0 deletions
diff --git a/pkg/shim/v1/proc/deleted_state.go b/pkg/shim/v1/proc/deleted_state.go
new file mode 100644
index 000000000..0196c96dd
--- /dev/null
+++ b/pkg/shim/v1/proc/deleted_state.go
@@ -0,0 +1,49 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "context"
+
+ "github.com/containerd/console"
+ "github.com/containerd/containerd/errdefs"
+ "github.com/containerd/containerd/runtime/proc"
+ "github.com/pkg/errors"
+)
+
+type deletedState struct{}
+
+func (*deletedState) Resize(ws console.WinSize) error {
+ return errors.Errorf("cannot resize a deleted process")
+}
+
+func (*deletedState) Start(ctx context.Context) error {
+ return errors.Errorf("cannot start a deleted process")
+}
+
+func (*deletedState) Delete(ctx context.Context) error {
+ return errors.Wrap(errdefs.ErrNotFound, "cannot delete a deleted process")
+}
+
+func (*deletedState) Kill(ctx context.Context, sig uint32, all bool) error {
+ return errors.Wrap(errdefs.ErrNotFound, "cannot kill a deleted process")
+}
+
+func (*deletedState) SetExited(status int) {}
+
+func (*deletedState) Exec(ctx context.Context, path string, r *ExecConfig) (proc.Process, error) {
+ return nil, errors.Errorf("cannot exec in a deleted state")
+}
diff --git a/pkg/shim/v1/proc/exec.go b/pkg/shim/v1/proc/exec.go
new file mode 100644
index 000000000..17199960e
--- /dev/null
+++ b/pkg/shim/v1/proc/exec.go
@@ -0,0 +1,282 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "context"
+ "fmt"
+ "io"
+ "os"
+ "path/filepath"
+ "sync"
+ "syscall"
+ "time"
+
+ "github.com/containerd/console"
+ "github.com/containerd/containerd/errdefs"
+ "github.com/containerd/containerd/runtime/proc"
+ "github.com/containerd/fifo"
+ runc "github.com/containerd/go-runc"
+ specs "github.com/opencontainers/runtime-spec/specs-go"
+ "github.com/pkg/errors"
+ "golang.org/x/sys/unix"
+
+ runsc "github.com/google/gvisor-containerd-shim/pkg/go-runsc"
+)
+
+type execProcess struct {
+ wg sync.WaitGroup
+
+ execState execState
+
+ mu sync.Mutex
+ id string
+ console console.Console
+ io runc.IO
+ status int
+ exited time.Time
+ pid int
+ internalPid int
+ closers []io.Closer
+ stdin io.Closer
+ stdio proc.Stdio
+ path string
+ spec specs.Process
+
+ parent *Init
+ waitBlock chan struct{}
+}
+
+func (e *execProcess) Wait() {
+ <-e.waitBlock
+}
+
+func (e *execProcess) ID() string {
+ return e.id
+}
+
+func (e *execProcess) Pid() int {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+ return e.pid
+}
+
+func (e *execProcess) ExitStatus() int {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+ return e.status
+}
+
+func (e *execProcess) ExitedAt() time.Time {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+ return e.exited
+}
+
+func (e *execProcess) SetExited(status int) {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+
+ e.execState.SetExited(status)
+}
+
+func (e *execProcess) setExited(status int) {
+ e.status = status
+ e.exited = time.Now()
+ e.parent.Platform.ShutdownConsole(context.Background(), e.console)
+ close(e.waitBlock)
+}
+
+func (e *execProcess) Delete(ctx context.Context) error {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+
+ return e.execState.Delete(ctx)
+}
+
+func (e *execProcess) delete(ctx context.Context) error {
+ e.wg.Wait()
+ if e.io != nil {
+ for _, c := range e.closers {
+ c.Close()
+ }
+ e.io.Close()
+ }
+ pidfile := filepath.Join(e.path, fmt.Sprintf("%s.pid", e.id))
+ // silently ignore error
+ os.Remove(pidfile)
+ internalPidfile := filepath.Join(e.path, fmt.Sprintf("%s-internal.pid", e.id))
+ // silently ignore error
+ os.Remove(internalPidfile)
+ return nil
+}
+
+func (e *execProcess) Resize(ws console.WinSize) error {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+
+ return e.execState.Resize(ws)
+}
+
+func (e *execProcess) resize(ws console.WinSize) error {
+ if e.console == nil {
+ return nil
+ }
+ return e.console.Resize(ws)
+}
+
+func (e *execProcess) Kill(ctx context.Context, sig uint32, _ bool) error {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+
+ return e.execState.Kill(ctx, sig, false)
+}
+
+func (e *execProcess) kill(ctx context.Context, sig uint32, _ bool) error {
+ internalPid := e.internalPid
+ if internalPid != 0 {
+ if err := e.parent.runtime.Kill(ctx, e.parent.id, int(sig), &runsc.KillOpts{
+ Pid: internalPid,
+ }); err != nil {
+ // If this returns error, consider the process has already stopped.
+ // TODO: Fix after signal handling is fixed.
+ return errors.Wrapf(errdefs.ErrNotFound, err.Error())
+ }
+ }
+ return nil
+}
+
+func (e *execProcess) Stdin() io.Closer {
+ return e.stdin
+}
+
+func (e *execProcess) Stdio() proc.Stdio {
+ return e.stdio
+}
+
+func (e *execProcess) Start(ctx context.Context) error {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+
+ return e.execState.Start(ctx)
+}
+
+func (e *execProcess) start(ctx context.Context) (err error) {
+ var (
+ socket *runc.Socket
+ pidfile = filepath.Join(e.path, fmt.Sprintf("%s.pid", e.id))
+ internalPidfile = filepath.Join(e.path, fmt.Sprintf("%s-internal.pid", e.id))
+ )
+ if e.stdio.Terminal {
+ if socket, err = runc.NewTempConsoleSocket(); err != nil {
+ return errors.Wrap(err, "failed to create runc console socket")
+ }
+ defer socket.Close()
+ } else if e.stdio.IsNull() {
+ if e.io, err = runc.NewNullIO(); err != nil {
+ return errors.Wrap(err, "creating new NULL IO")
+ }
+ } else {
+ 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")
+ }
+ }
+ opts := &runsc.ExecOpts{
+ PidFile: pidfile,
+ InternalPidFile: internalPidfile,
+ IO: e.io,
+ Detach: true,
+ }
+ if socket != nil {
+ opts.ConsoleSocket = socket
+ }
+ eventCh := e.parent.Monitor.Subscribe()
+ defer func() {
+ // Unsubscribe if an error is returned.
+ if err != nil {
+ e.parent.Monitor.Unsubscribe(eventCh)
+ }
+ }()
+ if err := e.parent.runtime.Exec(ctx, e.parent.id, e.spec, opts); err != nil {
+ close(e.waitBlock)
+ return e.parent.runtimeError(err, "OCI runtime exec failed")
+ }
+ if e.stdio.Stdin != "" {
+ sc, err := fifo.OpenFifo(context.Background(), 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)
+ }
+ e.closers = append(e.closers, sc)
+ e.stdin = sc
+ }
+ var copyWaitGroup sync.WaitGroup
+ ctx, cancel := context.WithTimeout(ctx, 30*time.Second)
+ defer cancel()
+ if socket != nil {
+ console, err := socket.ReceiveMaster()
+ if err != nil {
+ return errors.Wrap(err, "failed to retrieve console master")
+ }
+ if e.console, err = e.parent.Platform.CopyConsole(ctx, console, e.stdio.Stdin, e.stdio.Stdout, e.stdio.Stderr, &e.wg, &copyWaitGroup); err != nil {
+ return errors.Wrap(err, "failed to start console copy")
+ }
+ } else if !e.stdio.IsNull() {
+ 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")
+ }
+ }
+ copyWaitGroup.Wait()
+ pid, err := runc.ReadPidFile(opts.PidFile)
+ if err != nil {
+ return errors.Wrap(err, "failed to retrieve OCI runtime exec pid")
+ }
+ e.pid = pid
+ internalPid, err := runc.ReadPidFile(opts.InternalPidFile)
+ if err != nil {
+ return errors.Wrap(err, "failed to retrieve OCI runtime exec internal pid")
+ }
+ e.internalPid = internalPid
+ go func() {
+ defer e.parent.Monitor.Unsubscribe(eventCh)
+ for event := range eventCh {
+ if event.Pid == e.pid {
+ ExitCh <- Exit{
+ Timestamp: event.Timestamp,
+ ID: e.id,
+ Status: event.Status,
+ }
+ break
+ }
+ }
+ }()
+ return nil
+}
+
+func (e *execProcess) Status(ctx context.Context) (string, error) {
+ e.mu.Lock()
+ defer e.mu.Unlock()
+ // if we don't have a pid then the exec process has just been created
+ if e.pid == 0 {
+ return "created", nil
+ }
+ // if we have a pid and it can be signaled, the process is running
+ // TODO(random-liu): Use `runsc kill --pid`.
+ if err := unix.Kill(e.pid, 0); err == nil {
+ return "running", nil
+ }
+ // else if we have a pid but it can nolonger be signaled, it has stopped
+ return "stopped", nil
+}
diff --git a/pkg/shim/v1/proc/exec_state.go b/pkg/shim/v1/proc/exec_state.go
new file mode 100644
index 000000000..5416cb601
--- /dev/null
+++ b/pkg/shim/v1/proc/exec_state.go
@@ -0,0 +1,154 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "context"
+
+ "github.com/containerd/console"
+ "github.com/pkg/errors"
+)
+
+type execState interface {
+ Resize(console.WinSize) error
+ Start(context.Context) error
+ Delete(context.Context) error
+ Kill(context.Context, uint32, bool) error
+ SetExited(int)
+}
+
+type execCreatedState struct {
+ p *execProcess
+}
+
+func (s *execCreatedState) transition(name string) error {
+ switch name {
+ case "running":
+ s.p.execState = &execRunningState{p: s.p}
+ case "stopped":
+ s.p.execState = &execStoppedState{p: s.p}
+ case "deleted":
+ s.p.execState = &deletedState{}
+ default:
+ return errors.Errorf("invalid state transition %q to %q", stateName(s), name)
+ }
+ return nil
+}
+
+func (s *execCreatedState) Resize(ws console.WinSize) error {
+ return s.p.resize(ws)
+}
+
+func (s *execCreatedState) Start(ctx context.Context) error {
+ if err := s.p.start(ctx); err != nil {
+ return err
+ }
+ return s.transition("running")
+}
+
+func (s *execCreatedState) Delete(ctx context.Context) error {
+ if err := s.p.delete(ctx); err != nil {
+ return err
+ }
+ return s.transition("deleted")
+}
+
+func (s *execCreatedState) Kill(ctx context.Context, sig uint32, all bool) error {
+ return s.p.kill(ctx, sig, all)
+}
+
+func (s *execCreatedState) SetExited(status int) {
+ s.p.setExited(status)
+
+ if err := s.transition("stopped"); err != nil {
+ panic(err)
+ }
+}
+
+type execRunningState struct {
+ p *execProcess
+}
+
+func (s *execRunningState) transition(name string) error {
+ switch name {
+ case "stopped":
+ s.p.execState = &execStoppedState{p: s.p}
+ default:
+ return errors.Errorf("invalid state transition %q to %q", stateName(s), name)
+ }
+ return nil
+}
+
+func (s *execRunningState) Resize(ws console.WinSize) error {
+ return s.p.resize(ws)
+}
+
+func (s *execRunningState) Start(ctx context.Context) error {
+ return errors.Errorf("cannot start a running process")
+}
+
+func (s *execRunningState) Delete(ctx context.Context) error {
+ return errors.Errorf("cannot delete a running process")
+}
+
+func (s *execRunningState) Kill(ctx context.Context, sig uint32, all bool) error {
+ return s.p.kill(ctx, sig, all)
+}
+
+func (s *execRunningState) SetExited(status int) {
+ s.p.setExited(status)
+
+ if err := s.transition("stopped"); err != nil {
+ panic(err)
+ }
+}
+
+type execStoppedState struct {
+ p *execProcess
+}
+
+func (s *execStoppedState) transition(name string) error {
+ switch name {
+ case "deleted":
+ s.p.execState = &deletedState{}
+ default:
+ return errors.Errorf("invalid state transition %q to %q", stateName(s), name)
+ }
+ return nil
+}
+
+func (s *execStoppedState) Resize(ws console.WinSize) error {
+ return errors.Errorf("cannot resize a stopped container")
+}
+
+func (s *execStoppedState) Start(ctx context.Context) error {
+ return errors.Errorf("cannot start a stopped process")
+}
+
+func (s *execStoppedState) Delete(ctx context.Context) error {
+ if err := s.p.delete(ctx); err != nil {
+ return err
+ }
+ return s.transition("deleted")
+}
+
+func (s *execStoppedState) Kill(ctx context.Context, sig uint32, all bool) error {
+ return s.p.kill(ctx, sig, all)
+}
+
+func (s *execStoppedState) SetExited(status int) {
+ // no op
+}
diff --git a/pkg/shim/v1/proc/init.go b/pkg/shim/v1/proc/init.go
new file mode 100644
index 000000000..fadfc9146
--- /dev/null
+++ b/pkg/shim/v1/proc/init.go
@@ -0,0 +1,462 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "context"
+ "encoding/json"
+ "io"
+ "path/filepath"
+ "strings"
+ "sync"
+ "syscall"
+ "time"
+
+ "github.com/containerd/console"
+ "github.com/containerd/containerd/errdefs"
+ "github.com/containerd/containerd/log"
+ "github.com/containerd/containerd/mount"
+ "github.com/containerd/containerd/runtime/proc"
+ "github.com/containerd/fifo"
+ runc "github.com/containerd/go-runc"
+ specs "github.com/opencontainers/runtime-spec/specs-go"
+ "github.com/pkg/errors"
+
+ runsc "github.com/google/gvisor-containerd-shim/pkg/go-runsc"
+)
+
+// InitPidFile name of the file that contains the init pid
+const InitPidFile = "init.pid"
+
+// Init represents an initial process for a container
+type Init struct {
+ wg sync.WaitGroup
+ initState initState
+
+ // mu is used to ensure that `Start()` and `Exited()` calls return in
+ // the right order when invoked in separate go routines.
+ // This is the case within the shim implementation as it makes use of
+ // the reaper interface.
+ mu sync.Mutex
+
+ waitBlock chan struct{}
+
+ WorkDir string
+
+ id string
+ Bundle string
+ console console.Console
+ Platform proc.Platform
+ io runc.IO
+ runtime *runsc.Runsc
+ status int
+ exited time.Time
+ pid int
+ closers []io.Closer
+ stdin io.Closer
+ stdio proc.Stdio
+ Rootfs string
+ IoUID int
+ IoGID int
+ Sandbox bool
+ UserLog string
+ Monitor ProcessMonitor
+}
+
+// NewRunsc returns a new runsc instance for a process
+func NewRunsc(root, path, namespace, runtime string, config map[string]string) *runsc.Runsc {
+ if root == "" {
+ root = RunscRoot
+ }
+ return &runsc.Runsc{
+ Command: runtime,
+ PdeathSignal: syscall.SIGKILL,
+ Log: filepath.Join(path, "log.json"),
+ LogFormat: runc.JSON,
+ Root: filepath.Join(root, namespace),
+ Config: config,
+ }
+}
+
+// New returns a new init process
+func New(id string, runtime *runsc.Runsc, stdio proc.Stdio) *Init {
+ p := &Init{
+ id: id,
+ runtime: runtime,
+ stdio: stdio,
+ status: 0,
+ waitBlock: make(chan struct{}),
+ }
+ p.initState = &createdState{p: p}
+ return p
+}
+
+// Create the process with the provided config
+func (p *Init) Create(ctx context.Context, r *CreateConfig) (err error) {
+ var socket *runc.Socket
+ if r.Terminal {
+ if socket, err = runc.NewTempConsoleSocket(); err != nil {
+ return errors.Wrap(err, "failed to create OCI runtime console socket")
+ }
+ defer socket.Close()
+ } else if hasNoIO(r) {
+ if p.io, err = runc.NewNullIO(); err != nil {
+ return errors.Wrap(err, "creating new NULL IO")
+ }
+ } else {
+ 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")
+ }
+ }
+ pidFile := filepath.Join(p.Bundle, InitPidFile)
+ opts := &runsc.CreateOpts{
+ PidFile: pidFile,
+ }
+ if socket != nil {
+ opts.ConsoleSocket = socket
+ }
+ if p.Sandbox {
+ opts.IO = p.io
+ // UserLog is only useful for sandbox.
+ opts.UserLog = p.UserLog
+ }
+ if err := p.runtime.Create(ctx, r.ID, r.Bundle, opts); err != nil {
+ return p.runtimeError(err, "OCI runtime create failed")
+ }
+ if r.Stdin != "" {
+ sc, err := fifo.OpenFifo(context.Background(), r.Stdin, syscall.O_WRONLY|syscall.O_NONBLOCK, 0)
+ if err != nil {
+ return errors.Wrapf(err, "failed to open stdin fifo %s", r.Stdin)
+ }
+ p.stdin = sc
+ p.closers = append(p.closers, sc)
+ }
+ var copyWaitGroup sync.WaitGroup
+ ctx, cancel := context.WithTimeout(ctx, 30*time.Second)
+ defer cancel()
+ if socket != nil {
+ console, err := socket.ReceiveMaster()
+ if err != nil {
+ return errors.Wrap(err, "failed to retrieve console master")
+ }
+ console, err = p.Platform.CopyConsole(ctx, console, r.Stdin, r.Stdout, r.Stderr, &p.wg, &copyWaitGroup)
+ if err != nil {
+ return errors.Wrap(err, "failed to start console copy")
+ }
+ p.console = console
+ } else if !hasNoIO(r) {
+ if err := copyPipes(ctx, p.io, r.Stdin, r.Stdout, r.Stderr, &p.wg, &copyWaitGroup); err != nil {
+ return errors.Wrap(err, "failed to start io pipe copy")
+ }
+ }
+
+ copyWaitGroup.Wait()
+ pid, err := runc.ReadPidFile(pidFile)
+ if err != nil {
+ return errors.Wrap(err, "failed to retrieve OCI runtime container pid")
+ }
+ p.pid = pid
+ return nil
+}
+
+// Wait for the process to exit
+func (p *Init) Wait() {
+ <-p.waitBlock
+}
+
+// ID of the process
+func (p *Init) ID() string {
+ return p.id
+}
+
+// Pid of the process
+func (p *Init) Pid() int {
+ return p.pid
+}
+
+// ExitStatus of the process
+func (p *Init) ExitStatus() int {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+ return p.status
+}
+
+// ExitedAt at time when the process exited
+func (p *Init) ExitedAt() time.Time {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+ return p.exited
+}
+
+// Status of the process
+func (p *Init) Status(ctx context.Context) (string, error) {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+ c, err := p.runtime.State(ctx, p.id)
+ if err != nil {
+ if strings.Contains(err.Error(), "does not exist") {
+ return "stopped", nil
+ }
+ return "", p.runtimeError(err, "OCI runtime state failed")
+ }
+ return p.convertStatus(c.Status), nil
+}
+
+// Start the init process
+func (p *Init) Start(ctx context.Context) error {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+
+ return p.initState.Start(ctx)
+}
+
+func (p *Init) start(ctx context.Context) error {
+ var cio runc.IO
+ if !p.Sandbox {
+ cio = p.io
+ }
+ if err := p.runtime.Start(ctx, p.id, cio); err != nil {
+ return p.runtimeError(err, "OCI runtime start failed")
+ }
+ go func() {
+ status, err := p.runtime.Wait(context.Background(), p.id)
+ if err != nil {
+ log.G(ctx).WithError(err).Errorf("Failed to wait for container %q", p.id)
+ // TODO(random-liu): Handle runsc kill error.
+ if err := p.killAll(ctx); err != nil {
+ log.G(ctx).WithError(err).Errorf("Failed to kill container %q", p.id)
+ }
+ status = internalErrorCode
+ }
+ ExitCh <- Exit{
+ Timestamp: time.Now(),
+ ID: p.id,
+ Status: status,
+ }
+ }()
+ return nil
+}
+
+// SetExited of the init process with the next status
+func (p *Init) SetExited(status int) {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+
+ p.initState.SetExited(status)
+}
+
+func (p *Init) setExited(status int) {
+ p.exited = time.Now()
+ p.status = status
+ p.Platform.ShutdownConsole(context.Background(), p.console)
+ close(p.waitBlock)
+}
+
+// Delete the init process
+func (p *Init) Delete(ctx context.Context) error {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+
+ return p.initState.Delete(ctx)
+}
+
+func (p *Init) delete(ctx context.Context) error {
+ p.killAll(ctx)
+ p.wg.Wait()
+ err := p.runtime.Delete(ctx, p.id, nil)
+ // ignore errors if a runtime has already deleted the process
+ // but we still hold metadata and pipes
+ //
+ // this is common during a checkpoint, runc will delete the container state
+ // after a checkpoint and the container will no longer exist within runc
+ if err != nil {
+ if strings.Contains(err.Error(), "does not exist") {
+ err = nil
+ } else {
+ err = p.runtimeError(err, "failed to delete task")
+ }
+ }
+ if p.io != nil {
+ for _, c := range p.closers {
+ c.Close()
+ }
+ p.io.Close()
+ }
+ if err2 := mount.UnmountAll(p.Rootfs, 0); err2 != nil {
+ log.G(ctx).WithError(err2).Warn("failed to cleanup rootfs mount")
+ if err == nil {
+ err = errors.Wrap(err2, "failed rootfs umount")
+ }
+ }
+ return err
+}
+
+// Resize the init processes console
+func (p *Init) Resize(ws console.WinSize) error {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+
+ if p.console == nil {
+ return nil
+ }
+ return p.console.Resize(ws)
+}
+
+func (p *Init) resize(ws console.WinSize) error {
+ if p.console == nil {
+ return nil
+ }
+ return p.console.Resize(ws)
+}
+
+// Kill the init process
+func (p *Init) Kill(ctx context.Context, signal uint32, all bool) error {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+
+ return p.initState.Kill(ctx, signal, all)
+}
+
+func (p *Init) kill(context context.Context, signal uint32, all bool) error {
+ var (
+ killErr error
+ backoff = 100 * time.Millisecond
+ )
+ timeout := 1 * time.Second
+ for start := time.Now(); time.Now().Sub(start) < timeout; {
+ c, err := p.runtime.State(context, p.id)
+ if err != nil {
+ if strings.Contains(err.Error(), "does not exist") {
+ return errors.Wrapf(errdefs.ErrNotFound, "no such process")
+ }
+ return p.runtimeError(err, "OCI runtime state failed")
+ }
+ // For runsc, signal only works when container is running state.
+ // If the container is not in running state, directly return
+ // "no such process"
+ if p.convertStatus(c.Status) == "stopped" {
+ return errors.Wrapf(errdefs.ErrNotFound, "no such process")
+ }
+ killErr = p.runtime.Kill(context, p.id, int(signal), &runsc.KillOpts{
+ All: all,
+ })
+ if killErr == nil {
+ return nil
+ }
+ time.Sleep(backoff)
+ backoff *= 2
+ }
+ return p.runtimeError(killErr, "kill timeout")
+}
+
+// KillAll processes belonging to the init process
+func (p *Init) KillAll(context context.Context) error {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+ return p.killAll(context)
+}
+
+func (p *Init) killAll(context context.Context) error {
+ p.runtime.Kill(context, p.id, int(syscall.SIGKILL), &runsc.KillOpts{
+ All: true,
+ })
+ // Ignore error handling for `runsc kill --all` for now.
+ // * If it doesn't return error, it is good;
+ // * If it returns error, consider the container has already stopped.
+ // TODO: Fix `runsc kill --all` error handling.
+ return nil
+}
+
+// Stdin of the process
+func (p *Init) Stdin() io.Closer {
+ return p.stdin
+}
+
+// Runtime returns the OCI runtime configured for the init process
+func (p *Init) Runtime() *runsc.Runsc {
+ return p.runtime
+}
+
+// Exec returns a new child process
+func (p *Init) Exec(ctx context.Context, path string, r *ExecConfig) (proc.Process, error) {
+ p.mu.Lock()
+ defer p.mu.Unlock()
+
+ return p.initState.Exec(ctx, path, r)
+}
+
+// exec returns a new exec'd process
+func (p *Init) exec(ctx context.Context, path string, r *ExecConfig) (proc.Process, error) {
+ // process exec request
+ var spec specs.Process
+ if err := json.Unmarshal(r.Spec.Value, &spec); err != nil {
+ return nil, err
+ }
+ spec.Terminal = r.Terminal
+
+ e := &execProcess{
+ id: r.ID,
+ path: path,
+ parent: p,
+ spec: spec,
+ stdio: proc.Stdio{
+ Stdin: r.Stdin,
+ Stdout: r.Stdout,
+ Stderr: r.Stderr,
+ Terminal: r.Terminal,
+ },
+ waitBlock: make(chan struct{}),
+ }
+ e.execState = &execCreatedState{p: e}
+ return e, nil
+}
+
+// Stdio of the process
+func (p *Init) Stdio() proc.Stdio {
+ return p.stdio
+}
+
+func (p *Init) runtimeError(rErr error, msg string) error {
+ if rErr == nil {
+ return nil
+ }
+
+ rMsg, err := getLastRuntimeError(p.runtime)
+ switch {
+ case err != nil:
+ return errors.Wrapf(rErr, "%s: %s (%s)", msg, "unable to retrieve OCI runtime error", err.Error())
+ case rMsg == "":
+ return errors.Wrap(rErr, msg)
+ default:
+ return errors.Errorf("%s: %s", msg, rMsg)
+ }
+}
+
+func (p *Init) convertStatus(status string) string {
+ if status == "created" && !p.Sandbox && p.status == internalErrorCode {
+ // Treat start failure state for non-root container as stopped.
+ return "stopped"
+ }
+ return status
+}
+
+func withConditionalIO(c proc.Stdio) runc.IOOpt {
+ return func(o *runc.IOOption) {
+ o.OpenStdin = c.Stdin != ""
+ o.OpenStdout = c.Stdout != ""
+ o.OpenStderr = c.Stderr != ""
+ }
+}
diff --git a/pkg/shim/v1/proc/init_state.go b/pkg/shim/v1/proc/init_state.go
new file mode 100644
index 000000000..868646b6c
--- /dev/null
+++ b/pkg/shim/v1/proc/init_state.go
@@ -0,0 +1,182 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "context"
+
+ "github.com/containerd/console"
+ "github.com/containerd/containerd/errdefs"
+ "github.com/containerd/containerd/runtime/proc"
+ "github.com/pkg/errors"
+)
+
+type initState interface {
+ Resize(console.WinSize) error
+ Start(context.Context) error
+ Delete(context.Context) error
+ Exec(context.Context, string, *ExecConfig) (proc.Process, error)
+ Kill(context.Context, uint32, bool) error
+ SetExited(int)
+}
+
+type createdState struct {
+ p *Init
+}
+
+func (s *createdState) transition(name string) error {
+ switch name {
+ case "running":
+ s.p.initState = &runningState{p: s.p}
+ case "stopped":
+ s.p.initState = &stoppedState{p: s.p}
+ case "deleted":
+ s.p.initState = &deletedState{}
+ default:
+ return errors.Errorf("invalid state transition %q to %q", stateName(s), name)
+ }
+ return nil
+}
+
+func (s *createdState) Resize(ws console.WinSize) error {
+ return s.p.resize(ws)
+}
+
+func (s *createdState) Start(ctx context.Context) error {
+ if err := s.p.start(ctx); err != nil {
+ // Containerd doesn't allow deleting container in created state.
+ // However, for gvisor, a non-root container in created state can
+ // only go to running state. If the container can't be started,
+ // it can only stay in created state, and never be deleted.
+ // To work around that, we treat non-root container in start failure
+ // state as stopped.
+ if !s.p.Sandbox {
+ s.p.io.Close()
+ s.p.setExited(internalErrorCode)
+ if err := s.transition("stopped"); err != nil {
+ panic(err)
+ }
+ }
+ return err
+ }
+ return s.transition("running")
+}
+
+func (s *createdState) Delete(ctx context.Context) error {
+ if err := s.p.delete(ctx); err != nil {
+ return err
+ }
+ return s.transition("deleted")
+}
+
+func (s *createdState) Kill(ctx context.Context, sig uint32, all bool) error {
+ return s.p.kill(ctx, sig, all)
+}
+
+func (s *createdState) SetExited(status int) {
+ s.p.setExited(status)
+
+ if err := s.transition("stopped"); err != nil {
+ panic(err)
+ }
+}
+
+func (s *createdState) Exec(ctx context.Context, path string, r *ExecConfig) (proc.Process, error) {
+ return s.p.exec(ctx, path, r)
+}
+
+type runningState struct {
+ p *Init
+}
+
+func (s *runningState) transition(name string) error {
+ switch name {
+ case "stopped":
+ s.p.initState = &stoppedState{p: s.p}
+ default:
+ return errors.Errorf("invalid state transition %q to %q", stateName(s), name)
+ }
+ return nil
+}
+
+func (s *runningState) Resize(ws console.WinSize) error {
+ return s.p.resize(ws)
+}
+
+func (s *runningState) Start(ctx context.Context) error {
+ return errors.Errorf("cannot start a running process")
+}
+
+func (s *runningState) Delete(ctx context.Context) error {
+ return errors.Errorf("cannot delete a running process")
+}
+
+func (s *runningState) Kill(ctx context.Context, sig uint32, all bool) error {
+ return s.p.kill(ctx, sig, all)
+}
+
+func (s *runningState) SetExited(status int) {
+ s.p.setExited(status)
+
+ if err := s.transition("stopped"); err != nil {
+ panic(err)
+ }
+}
+
+func (s *runningState) Exec(ctx context.Context, path string, r *ExecConfig) (proc.Process, error) {
+ return s.p.exec(ctx, path, r)
+}
+
+type stoppedState struct {
+ p *Init
+}
+
+func (s *stoppedState) transition(name string) error {
+ switch name {
+ case "deleted":
+ s.p.initState = &deletedState{}
+ default:
+ return errors.Errorf("invalid state transition %q to %q", stateName(s), name)
+ }
+ return nil
+}
+
+func (s *stoppedState) Resize(ws console.WinSize) error {
+ return errors.Errorf("cannot resize a stopped container")
+}
+
+func (s *stoppedState) Start(ctx context.Context) error {
+ return errors.Errorf("cannot start a stopped process")
+}
+
+func (s *stoppedState) Delete(ctx context.Context) error {
+ if err := s.p.delete(ctx); err != nil {
+ return err
+ }
+ return s.transition("deleted")
+}
+
+func (s *stoppedState) Kill(ctx context.Context, sig uint32, all bool) error {
+ return errdefs.ToGRPCf(errdefs.ErrNotFound, "process %s not found", s.p.id)
+}
+
+func (s *stoppedState) SetExited(status int) {
+ // no op
+}
+
+func (s *stoppedState) Exec(ctx context.Context, path string, r *ExecConfig) (proc.Process, error) {
+ return nil, errors.Errorf("cannot exec in a stopped state")
+}
diff --git a/pkg/shim/v1/proc/io.go b/pkg/shim/v1/proc/io.go
new file mode 100644
index 000000000..2677b4e54
--- /dev/null
+++ b/pkg/shim/v1/proc/io.go
@@ -0,0 +1,167 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "context"
+ "fmt"
+ "io"
+ "os"
+ "sync"
+ "sync/atomic"
+ "syscall"
+
+ "github.com/containerd/containerd/log"
+ "github.com/containerd/fifo"
+ runc "github.com/containerd/go-runc"
+)
+
+// TODO(random-liu): This file can be a util.
+
+var bufPool = sync.Pool{
+ New: func() interface{} {
+ buffer := make([]byte, 32<<10)
+ return &buffer
+ },
+}
+
+func copyPipes(ctx context.Context, rio runc.IO, stdin, stdout, stderr string, wg, cwg *sync.WaitGroup) error {
+ var sameFile *countingWriteCloser
+ for _, i := range []struct {
+ name string
+ dest func(wc io.WriteCloser, rc io.Closer)
+ }{
+ {
+ name: stdout,
+ dest: func(wc io.WriteCloser, rc io.Closer) {
+ wg.Add(1)
+ cwg.Add(1)
+ go func() {
+ cwg.Done()
+ p := bufPool.Get().(*[]byte)
+ defer bufPool.Put(p)
+ if _, err := io.CopyBuffer(wc, rio.Stdout(), *p); err != nil {
+ log.G(ctx).Warn("error copying stdout")
+ }
+ wg.Done()
+ wc.Close()
+ if rc != nil {
+ rc.Close()
+ }
+ }()
+ },
+ }, {
+ name: stderr,
+ dest: func(wc io.WriteCloser, rc io.Closer) {
+ wg.Add(1)
+ cwg.Add(1)
+ go func() {
+ cwg.Done()
+ p := bufPool.Get().(*[]byte)
+ defer bufPool.Put(p)
+ if _, err := io.CopyBuffer(wc, rio.Stderr(), *p); err != nil {
+ log.G(ctx).Warn("error copying stderr")
+ }
+ wg.Done()
+ wc.Close()
+ if rc != nil {
+ rc.Close()
+ }
+ }()
+ },
+ },
+ } {
+ ok, err := isFifo(i.name)
+ if err != nil {
+ return err
+ }
+ var (
+ fw io.WriteCloser
+ fr io.Closer
+ )
+ if ok {
+ if fw, err = fifo.OpenFifo(ctx, i.name, syscall.O_WRONLY, 0); err != nil {
+ return fmt.Errorf("gvisor-containerd-shim: opening %s failed: %s", i.name, err)
+ }
+ if fr, err = fifo.OpenFifo(ctx, i.name, syscall.O_RDONLY, 0); err != nil {
+ return fmt.Errorf("gvisor-containerd-shim: opening %s failed: %s", i.name, err)
+ }
+ } else {
+ if sameFile != nil {
+ sameFile.count++
+ i.dest(sameFile, nil)
+ continue
+ }
+ if fw, err = os.OpenFile(i.name, syscall.O_WRONLY|syscall.O_APPEND, 0); err != nil {
+ return fmt.Errorf("gvisor-containerd-shim: opening %s failed: %s", i.name, err)
+ }
+ if stdout == stderr {
+ sameFile = &countingWriteCloser{
+ WriteCloser: fw,
+ count: 1,
+ }
+ }
+ }
+ i.dest(fw, fr)
+ }
+ if stdin == "" {
+ return nil
+ }
+ f, err := fifo.OpenFifo(context.Background(), stdin, syscall.O_RDONLY|syscall.O_NONBLOCK, 0)
+ if err != nil {
+ return fmt.Errorf("gvisor-containerd-shim: opening %s failed: %s", stdin, err)
+ }
+ cwg.Add(1)
+ go func() {
+ cwg.Done()
+ p := bufPool.Get().(*[]byte)
+ defer bufPool.Put(p)
+
+ io.CopyBuffer(rio.Stdin(), f, *p)
+ rio.Stdin().Close()
+ f.Close()
+ }()
+ return nil
+}
+
+// countingWriteCloser masks io.Closer() until close has been invoked a certain number of times.
+type countingWriteCloser struct {
+ io.WriteCloser
+ count int64
+}
+
+func (c *countingWriteCloser) Close() error {
+ if atomic.AddInt64(&c.count, -1) > 0 {
+ return nil
+ }
+ return c.WriteCloser.Close()
+}
+
+// isFifo checks if a file is a fifo
+// if the file does not exist then it returns false
+func isFifo(path string) (bool, error) {
+ stat, err := os.Stat(path)
+ if err != nil {
+ if os.IsNotExist(err) {
+ return false, nil
+ }
+ return false, err
+ }
+ if stat.Mode()&os.ModeNamedPipe == os.ModeNamedPipe {
+ return true, nil
+ }
+ return false, nil
+}
diff --git a/pkg/shim/v1/proc/process.go b/pkg/shim/v1/proc/process.go
new file mode 100644
index 000000000..1bfa99f4c
--- /dev/null
+++ b/pkg/shim/v1/proc/process.go
@@ -0,0 +1,37 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "github.com/pkg/errors"
+)
+
+// RunscRoot is the path to the root runsc state directory
+const RunscRoot = "/run/containerd/runsc"
+
+func stateName(v interface{}) string {
+ switch v.(type) {
+ case *runningState, *execRunningState:
+ return "running"
+ case *createdState, *execCreatedState:
+ return "created"
+ case *deletedState:
+ return "deleted"
+ case *stoppedState:
+ return "stopped"
+ }
+ panic(errors.Errorf("invalid state %v", v))
+}
diff --git a/pkg/shim/v1/proc/types.go b/pkg/shim/v1/proc/types.go
new file mode 100644
index 000000000..dcd43bcca
--- /dev/null
+++ b/pkg/shim/v1/proc/types.go
@@ -0,0 +1,70 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "time"
+
+ google_protobuf "github.com/gogo/protobuf/types"
+
+ runc "github.com/containerd/go-runc"
+)
+
+// Mount holds filesystem mount configuration
+type Mount struct {
+ Type string
+ Source string
+ Target string
+ Options []string
+}
+
+// CreateConfig hold task creation configuration
+type CreateConfig struct {
+ ID string
+ Bundle string
+ Runtime string
+ Rootfs []Mount
+ Terminal bool
+ Stdin string
+ Stdout string
+ Stderr string
+ Options *google_protobuf.Any
+}
+
+// ExecConfig holds exec creation configuration
+type ExecConfig struct {
+ ID string
+ Terminal bool
+ Stdin string
+ Stdout string
+ Stderr string
+ Spec *google_protobuf.Any
+}
+
+// Exit is the type of exit events
+type Exit struct {
+ Timestamp time.Time
+ ID string
+ Status int
+}
+
+// ProcessMonitor monitors process exit changes
+type ProcessMonitor interface {
+ // Subscribe to process exit changes
+ Subscribe() chan runc.Exit
+ // Unsubscribe to process exit changes
+ Unsubscribe(c chan runc.Exit)
+}
diff --git a/pkg/shim/v1/proc/utils.go b/pkg/shim/v1/proc/utils.go
new file mode 100644
index 000000000..84bd702ab
--- /dev/null
+++ b/pkg/shim/v1/proc/utils.go
@@ -0,0 +1,92 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2018 The gVisor 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
+//
+// https://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 proc
+
+import (
+ "encoding/json"
+ "io"
+ "os"
+ "strings"
+ "time"
+
+ runsc "github.com/google/gvisor-containerd-shim/pkg/go-runsc"
+)
+
+const (
+ internalErrorCode = 128
+ bufferSize = 32
+)
+
+// ExitCh is the exit events channel for containers and exec processes
+// inside the sandbox.
+var ExitCh = make(chan Exit, bufferSize)
+
+// TODO(random-liu): This can be a utility.
+
+// TODO(mlaventure): move to runc package?
+func getLastRuntimeError(r *runsc.Runsc) (string, error) {
+ if r.Log == "" {
+ return "", nil
+ }
+
+ f, err := os.OpenFile(r.Log, os.O_RDONLY, 0400)
+ if err != nil {
+ return "", err
+ }
+
+ var (
+ errMsg string
+ log struct {
+ Level string
+ Msg string
+ Time time.Time
+ }
+ )
+
+ dec := json.NewDecoder(f)
+ for err = nil; err == nil; {
+ if err = dec.Decode(&log); err != nil && err != io.EOF {
+ return "", err
+ }
+ if log.Level == "error" {
+ errMsg = strings.TrimSpace(log.Msg)
+ }
+ }
+
+ return errMsg, nil
+}
+
+func copyFile(to, from string) error {
+ ff, err := os.Open(from)
+ if err != nil {
+ return err
+ }
+ defer ff.Close()
+ tt, err := os.Create(to)
+ if err != nil {
+ return err
+ }
+ defer tt.Close()
+
+ p := bufPool.Get().(*[]byte)
+ defer bufPool.Put(p)
+ _, err = io.CopyBuffer(tt, ff, *p)
+ return err
+}
+
+func hasNoIO(r *CreateConfig) bool {
+ return r.Stdin == "" && r.Stdout == "" && r.Stderr == ""
+}
diff --git a/pkg/shim/v1/shim/platform.go b/pkg/shim/v1/shim/platform.go
new file mode 100644
index 000000000..86252c3f5
--- /dev/null
+++ b/pkg/shim/v1/shim/platform.go
@@ -0,0 +1,110 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2019 The gVisor 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
+//
+// https://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 shim
+
+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
+}
diff --git a/pkg/shim/v1/shim/service.go b/pkg/shim/v1/shim/service.go
new file mode 100644
index 000000000..27e915c14
--- /dev/null
+++ b/pkg/shim/v1/shim/service.go
@@ -0,0 +1,579 @@
+// Copyright 2018 The containerd Authors.
+// Copyright 2019 The gVisor 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
+//
+// https://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 shim
+
+import (
+ "context"
+ "fmt"
+ "os"
+ "path/filepath"
+ "sync"
+
+ "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/log"
+ "github.com/containerd/containerd/mount"
+ "github.com/containerd/containerd/namespaces"
+ "github.com/containerd/containerd/runtime"
+ "github.com/containerd/containerd/runtime/linux/runctypes"
+ rproc "github.com/containerd/containerd/runtime/proc"
+ "github.com/containerd/containerd/runtime/v1/shim"
+ shimapi "github.com/containerd/containerd/runtime/v1/shim/v1"
+ "github.com/containerd/typeurl"
+ ptypes "github.com/gogo/protobuf/types"
+ "github.com/pkg/errors"
+ "github.com/sirupsen/logrus"
+ "google.golang.org/grpc/codes"
+ "google.golang.org/grpc/status"
+
+ runsc "github.com/google/gvisor-containerd-shim/pkg/go-runsc"
+ "github.com/google/gvisor-containerd-shim/pkg/v1/proc"
+ "github.com/google/gvisor-containerd-shim/pkg/v1/utils"
+)
+
+var (
+ empty = &ptypes.Empty{}
+ bufPool = sync.Pool{
+ New: func() interface{} {
+ buffer := make([]byte, 32<<10)
+ return &buffer
+ },
+ }
+)
+
+// Config contains shim specific configuration
+type Config struct {
+ Path string
+ Namespace string
+ WorkDir string
+ RuntimeRoot string
+ RunscConfig map[string]string
+}
+
+// NewService returns a new shim service that can be used via GRPC
+func NewService(config Config, publisher events.Publisher) (*Service, error) {
+ if config.Namespace == "" {
+ return nil, fmt.Errorf("shim namespace cannot be empty")
+ }
+ ctx := namespaces.WithNamespace(context.Background(), config.Namespace)
+ ctx = log.WithLogger(ctx, logrus.WithFields(logrus.Fields{
+ "namespace": config.Namespace,
+ "path": config.Path,
+ "pid": os.Getpid(),
+ }))
+ s := &Service{
+ config: config,
+ context: ctx,
+ processes: make(map[string]rproc.Process),
+ events: make(chan interface{}, 128),
+ ec: proc.ExitCh,
+ }
+ go s.processExits()
+ 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
+
+ config Config
+ context context.Context
+ processes map[string]rproc.Process
+ events chan interface{}
+ platform rproc.Platform
+ ec chan proc.Exit
+
+ // Filled by Create()
+ id string
+ bundle string
+}
+
+// Create a new initial process and container with the underlying OCI runtime
+func (s *Service) Create(ctx context.Context, r *shimapi.CreateTaskRequest) (_ *shimapi.CreateTaskResponse, err error) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ 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,
+ })
+ }
+
+ rootfs := filepath.Join(r.Bundle, "rootfs")
+ if err := os.Mkdir(rootfs, 0711); err != nil && !os.IsExist(err) {
+ return nil, err
+ }
+
+ config := &proc.CreateConfig{
+ ID: r.ID,
+ Bundle: r.Bundle,
+ Runtime: r.Runtime,
+ Rootfs: mounts,
+ Terminal: r.Terminal,
+ Stdin: r.Stdin,
+ Stdout: r.Stdout,
+ Stderr: r.Stderr,
+ Options: r.Options,
+ }
+ defer func() {
+ if err != nil {
+ if err2 := mount.UnmountAll(rootfs, 0); err2 != nil {
+ log.G(ctx).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,
+ s.config.Path,
+ s.config.WorkDir,
+ s.config.RuntimeRoot,
+ s.config.Namespace,
+ s.config.RunscConfig,
+ s.platform,
+ config,
+ )
+ 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()
+ s.processes[r.ID] = process
+ return &shimapi.CreateTaskResponse{
+ Pid: uint32(pid),
+ }, nil
+}
+
+// Start a process
+func (s *Service) Start(ctx context.Context, r *shimapi.StartRequest) (*shimapi.StartResponse, error) {
+ p, err := s.getExecProcess(r.ID)
+ if err != nil {
+ return nil, err
+ }
+ if err := p.Start(ctx); err != nil {
+ return nil, err
+ }
+ return &shimapi.StartResponse{
+ ID: p.ID(),
+ Pid: uint32(p.Pid()),
+ }, nil
+}
+
+// Delete the initial process and container
+func (s *Service) Delete(ctx context.Context, r *ptypes.Empty) (*shimapi.DeleteResponse, error) {
+ p, err := s.getInitProcess()
+ if err != nil {
+ return nil, err
+ }
+ if err := p.Delete(ctx); err != nil {
+ return nil, err
+ }
+ s.mu.Lock()
+ delete(s.processes, s.id)
+ s.mu.Unlock()
+ s.platform.Close()
+ return &shimapi.DeleteResponse{
+ ExitStatus: uint32(p.ExitStatus()),
+ ExitedAt: p.ExitedAt(),
+ Pid: uint32(p.Pid()),
+ }, nil
+}
+
+// DeleteProcess deletes an exec'd process
+func (s *Service) DeleteProcess(ctx context.Context, r *shimapi.DeleteProcessRequest) (*shimapi.DeleteResponse, error) {
+ if r.ID == s.id {
+ return nil, status.Errorf(codes.InvalidArgument, "cannot delete init process with DeleteProcess")
+ }
+ p, err := s.getExecProcess(r.ID)
+ if err != nil {
+ return nil, err
+ }
+ 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(),
+ Pid: uint32(p.Pid()),
+ }, nil
+}
+
+// Exec an additional process inside the container
+func (s *Service) Exec(ctx context.Context, r *shimapi.ExecProcessRequest) (*ptypes.Empty, error) {
+ s.mu.Lock()
+
+ if p := s.processes[r.ID]; p != nil {
+ s.mu.Unlock()
+ return nil, errdefs.ToGRPCf(errdefs.ErrAlreadyExists, "id %s", r.ID)
+ }
+
+ p := s.processes[s.id]
+ s.mu.Unlock()
+ if p == nil {
+ return nil, errdefs.ToGRPCf(errdefs.ErrFailedPrecondition, "container must be created")
+ }
+
+ process, err := p.(*proc.Init).Exec(ctx, s.config.Path, &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.mu.Lock()
+ s.processes[r.ID] = process
+ s.mu.Unlock()
+ return empty, nil
+}
+
+// ResizePty of a process
+func (s *Service) ResizePty(ctx context.Context, r *shimapi.ResizePtyRequest) (*ptypes.Empty, error) {
+ if r.ID == "" {
+ return nil, errdefs.ToGRPCf(errdefs.ErrInvalidArgument, "id not provided")
+ }
+ ws := console.WinSize{
+ Width: uint16(r.Width),
+ Height: uint16(r.Height),
+ }
+ p, err := s.getExecProcess(r.ID)
+ if err != nil {
+ return nil, err
+ }
+ 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 *shimapi.StateRequest) (*shimapi.StateResponse, error) {
+ p, err := s.getExecProcess(r.ID)
+ if err != nil {
+ return nil, err
+ }
+ 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
+ }
+ sio := p.Stdio()
+ return &shimapi.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) {
+ return empty, errdefs.ToGRPC(errdefs.ErrNotImplemented)
+}
+
+// Resume the container
+func (s *Service) Resume(ctx context.Context, r *ptypes.Empty) (*ptypes.Empty, error) {
+ return empty, errdefs.ToGRPC(errdefs.ErrNotImplemented)
+}
+
+// Kill a process with the provided signal
+func (s *Service) Kill(ctx context.Context, r *shimapi.KillRequest) (*ptypes.Empty, error) {
+ if r.ID == "" {
+ p, err := s.getInitProcess()
+ if err != nil {
+ return nil, err
+ }
+ if err := p.Kill(ctx, r.Signal, r.All); err != nil {
+ return nil, errdefs.ToGRPC(err)
+ }
+ return empty, nil
+ }
+
+ p, err := s.getExecProcess(r.ID)
+ if err != nil {
+ return nil, err
+ }
+ if err := p.Kill(ctx, r.Signal, r.All); err != nil {
+ return nil, errdefs.ToGRPC(err)
+ }
+ return empty, nil
+}
+
+// ListPids returns all pids inside the container
+func (s *Service) ListPids(ctx context.Context, r *shimapi.ListPidsRequest) (*shimapi.ListPidsResponse, 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 := &runctypes.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 &shimapi.ListPidsResponse{
+ Processes: processes,
+ }, nil
+}
+
+// CloseIO of a process
+func (s *Service) CloseIO(ctx context.Context, r *shimapi.CloseIORequest) (*ptypes.Empty, error) {
+ p, err := s.getExecProcess(r.ID)
+ if err != nil {
+ return nil, err
+ }
+ 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 *shimapi.CheckpointTaskRequest) (*ptypes.Empty, error) {
+ return empty, errdefs.ToGRPC(errdefs.ErrNotImplemented)
+}
+
+// ShimInfo returns shim information such as the shim's pid
+func (s *Service) ShimInfo(ctx context.Context, r *ptypes.Empty) (*shimapi.ShimInfoResponse, error) {
+ return &shimapi.ShimInfoResponse{
+ ShimPid: uint32(os.Getpid()),
+ }, nil
+}
+
+// Update a running container
+func (s *Service) Update(ctx context.Context, r *shimapi.UpdateTaskRequest) (*ptypes.Empty, error) {
+ return empty, errdefs.ToGRPC(errdefs.ErrNotImplemented)
+}
+
+// Wait for a process to exit
+func (s *Service) Wait(ctx context.Context, r *shimapi.WaitRequest) (*shimapi.WaitResponse, error) {
+ p, err := s.getExecProcess(r.ID)
+ if err != nil {
+ return nil, err
+ }
+ p.Wait()
+
+ return &shimapi.WaitResponse{
+ ExitStatus: uint32(p.ExitStatus()),
+ ExitedAt: p.ExitedAt(),
+ }, nil
+}
+
+func (s *Service) processExits() {
+ for e := range s.ec {
+ s.checkProcesses(e)
+ }
+}
+
+func (s *Service) allProcesses() []rproc.Process {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+
+ res := make([]rproc.Process, 0, len(s.processes))
+ for _, p := range s.processes {
+ res = append(res, p)
+ }
+ return res
+}
+
+func (s *Service) checkProcesses(e proc.Exit) {
+ for _, p := range s.allProcesses() {
+ if p.ID() == e.ID {
+ 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)
+ s.events <- &eventstypes.TaskExit{
+ ContainerID: s.id,
+ ID: p.ID(),
+ Pid: uint32(p.Pid()),
+ ExitStatus: uint32(e.Status),
+ ExitedAt: p.ExitedAt(),
+ }
+ return
+ }
+ }
+}
+
+func (s *Service) getContainerPids(ctx context.Context, id string) ([]uint32, error) {
+ p, err := s.getInitProcess()
+ if err != nil {
+ return nil, err
+ }
+
+ 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 {
+ log.G(s.context).WithError(err).Error("post event")
+ }
+ }
+}
+
+// getInitProcess returns initial process
+func (s *Service) getInitProcess() (rproc.Process, 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")
+ }
+ return p, nil
+}
+
+// getExecProcess returns exec process
+func (s *Service) getExecProcess(id string) (rproc.Process, error) {
+ s.mu.Lock()
+ defer s.mu.Unlock()
+ p := s.processes[id]
+ if p == nil {
+ return nil, errdefs.ToGRPCf(errdefs.ErrNotFound, "process %s does not exist", id)
+ }
+ return p, nil
+}
+
+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
+ default:
+ logrus.Warnf("no topic for type %#v", e)
+ }
+ return runtime.TaskUnknownTopic
+}
+
+func newInit(ctx context.Context, path, workDir, runtimeRoot, namespace string, config map[string]string, platform rproc.Platform, r *proc.CreateConfig) (*proc.Init, error) {
+ var options runctypes.CreateOptions
+ if r.Options != nil {
+ v, err := typeurl.UnmarshalAny(r.Options)
+ if err != nil {
+ return nil, err
+ }
+ options = *v.(*runctypes.CreateOptions)
+ }
+
+ spec, err := utils.ReadSpec(r.Bundle)
+ if err != nil {
+ return nil, errors.Wrap(err, "read oci spec")
+ }
+ if err := utils.UpdateVolumeAnnotations(r.Bundle, spec); err != nil {
+ return nil, errors.Wrap(err, "update volume annotations")
+ }
+
+ runsc.FormatLogPath(r.ID, config)
+ rootfs := filepath.Join(path, "rootfs")
+ runtime := proc.NewRunsc(runtimeRoot, path, namespace, r.Runtime, config)
+ 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.Sandbox = utils.IsSandbox(spec)
+ p.UserLog = utils.UserLogPath(spec)
+ p.Monitor = shim.Default
+ return p, nil
+}
diff --git a/pkg/shim/v1/utils/utils.go b/pkg/shim/v1/utils/utils.go
new file mode 100644
index 000000000..7a400af1c
--- /dev/null
+++ b/pkg/shim/v1/utils/utils.go
@@ -0,0 +1,57 @@
+// Copyright 2018 The gVisor 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
+//
+// https://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 utils
+
+import (
+ "encoding/json"
+ "io/ioutil"
+ "os"
+ "path/filepath"
+
+ "github.com/containerd/cri/pkg/annotations"
+ specs "github.com/opencontainers/runtime-spec/specs-go"
+)
+
+// ReadSpec reads OCI spec from the bundle directory.
+func ReadSpec(bundle string) (*specs.Spec, error) {
+ f, err := os.Open(filepath.Join(bundle, "config.json"))
+ if err != nil {
+ return nil, err
+ }
+ b, err := ioutil.ReadAll(f)
+ if err != nil {
+ return nil, err
+ }
+ var spec specs.Spec
+ if err := json.Unmarshal(b, &spec); err != nil {
+ return nil, err
+ }
+ return &spec, nil
+}
+
+// IsSandbox checks whether a container is a sandbox container.
+func IsSandbox(spec *specs.Spec) bool {
+ t, ok := spec.Annotations[annotations.ContainerType]
+ return !ok || t == annotations.ContainerTypeSandbox
+}
+
+// UserLogPath gets user log path from OCI annotation.
+func UserLogPath(spec *specs.Spec) string {
+ sandboxLogDir := spec.Annotations[annotations.SandboxLogDir]
+ if sandboxLogDir == "" {
+ return ""
+ }
+ return filepath.Join(sandboxLogDir, "gvisor.log")
+}
diff --git a/pkg/shim/v1/utils/volumes.go b/pkg/shim/v1/utils/volumes.go
new file mode 100644
index 000000000..7323e7245
--- /dev/null
+++ b/pkg/shim/v1/utils/volumes.go
@@ -0,0 +1,156 @@
+// Copyright 2018 The gVisor 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
+//
+// https://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 utils
+
+import (
+ "encoding/json"
+ "fmt"
+ "io/ioutil"
+ "path/filepath"
+ "strings"
+
+ "github.com/containerd/cri/pkg/annotations"
+ specs "github.com/opencontainers/runtime-spec/specs-go"
+ "github.com/pkg/errors"
+ "github.com/sirupsen/logrus"
+)
+
+const volumeKeyPrefix = "dev.gvisor.spec.mount."
+
+var kubeletPodsDir = "/var/lib/kubelet/pods"
+
+// volumeName gets volume name from volume annotation key, example:
+// dev.gvisor.spec.mount.NAME.share
+func volumeName(k string) string {
+ return strings.SplitN(strings.TrimPrefix(k, volumeKeyPrefix), ".", 2)[0]
+}
+
+// volumeFieldName gets volume field name from volume annotation key, example:
+// `type` is the field of dev.gvisor.spec.mount.NAME.type
+func volumeFieldName(k string) string {
+ parts := strings.Split(strings.TrimPrefix(k, volumeKeyPrefix), ".")
+ return parts[len(parts)-1]
+}
+
+// podUID gets pod UID from the pod log path.
+func podUID(s *specs.Spec) (string, error) {
+ sandboxLogDir := s.Annotations[annotations.SandboxLogDir]
+ if sandboxLogDir == "" {
+ return "", errors.New("no sandbox log path annotation")
+ }
+ fields := strings.Split(filepath.Base(sandboxLogDir), "_")
+ switch len(fields) {
+ case 1: // This is the old CRI logging path
+ return fields[0], nil
+ case 3: // This is the new CRI logging path
+ return fields[2], nil
+ }
+ return "", errors.Errorf("unexpected sandbox log path %q", sandboxLogDir)
+}
+
+// isVolumeKey checks whether an annotation key is for volume.
+func isVolumeKey(k string) bool {
+ return strings.HasPrefix(k, volumeKeyPrefix)
+}
+
+// volumeSourceKey constructs the annotation key for volume source.
+func volumeSourceKey(volume string) string {
+ return volumeKeyPrefix + volume + ".source"
+}
+
+// volumePath searches the volume path in the kubelet pod directory.
+func volumePath(volume, uid string) (string, error) {
+ // TODO: Support subpath when gvisor supports pod volume bind mount.
+ volumeSearchPath := fmt.Sprintf("%s/%s/volumes/*/%s", kubeletPodsDir, uid, volume)
+ dirs, err := filepath.Glob(volumeSearchPath)
+ if err != nil {
+ return "", err
+ }
+ if len(dirs) != 1 {
+ return "", errors.Errorf("unexpected matched volume list %v", dirs)
+ }
+ return dirs[0], nil
+}
+
+// isVolumePath checks whether a string is the volume path.
+func isVolumePath(volume, path string) (bool, error) {
+ // TODO: Support subpath when gvisor supports pod volume bind mount.
+ volumeSearchPath := fmt.Sprintf("%s/*/volumes/*/%s", kubeletPodsDir, volume)
+ return filepath.Match(volumeSearchPath, path)
+}
+
+// UpdateVolumeAnnotations add necessary OCI annotations for gvisor
+// volume optimization.
+func UpdateVolumeAnnotations(bundle string, s *specs.Spec) error {
+ var (
+ uid string
+ err error
+ )
+ if IsSandbox(s) {
+ uid, err = podUID(s)
+ if err != nil {
+ // Skip if we can't get pod UID, because this doesn't work
+ // for containerd 1.1.
+ logrus.WithError(err).Error("Can't get pod uid")
+ return nil
+ }
+ }
+ var updated bool
+ for k, v := range s.Annotations {
+ if !isVolumeKey(k) {
+ continue
+ }
+ if volumeFieldName(k) != "type" {
+ continue
+ }
+ volume := volumeName(k)
+ if uid != "" {
+ // This is a sandbox
+ path, err := volumePath(volume, uid)
+ if err != nil {
+ return errors.Wrapf(err, "get volume path for %q", volume)
+ }
+ s.Annotations[volumeSourceKey(volume)] = path
+ updated = true
+ } else {
+ // This is a container
+ for i := range s.Mounts {
+ // An error is returned for sandbox if source annotation
+ // is not successfully applied, so it is guaranteed that
+ // the source annotation for sandbox has already been
+ // successfully applied at this point.
+ // The volume name is unique inside a pod, so matching without
+ // podUID is fine here.
+ // TODO: Pass podUID down to shim for containers to do
+ // more accurate matching.
+ if yes, _ := isVolumePath(volume, s.Mounts[i].Source); yes {
+ // gVisor requires the container mount type to match
+ // sandbox mount type.
+ s.Mounts[i].Type = v
+ updated = true
+ }
+ }
+ }
+ }
+ if !updated {
+ return nil
+ }
+ // Update bundle
+ b, err := json.Marshal(s)
+ if err != nil {
+ return err
+ }
+ return ioutil.WriteFile(filepath.Join(bundle, "config.json"), b, 0666)
+}
diff --git a/pkg/shim/v1/utils/volumes_test.go b/pkg/shim/v1/utils/volumes_test.go
new file mode 100644
index 000000000..4b2639545
--- /dev/null
+++ b/pkg/shim/v1/utils/volumes_test.go
@@ -0,0 +1,309 @@
+// Copyright 2019 The gVisor 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
+//
+// https://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 utils
+
+import (
+ "encoding/json"
+ "fmt"
+ "io/ioutil"
+ "os"
+ "path/filepath"
+ "reflect"
+ "testing"
+
+ "github.com/containerd/cri/pkg/annotations"
+ specs "github.com/opencontainers/runtime-spec/specs-go"
+)
+
+func TestUpdateVolumeAnnotations(t *testing.T) {
+ dir, err := ioutil.TempDir("", "test-update-volume-annotations")
+ if err != nil {
+ t.Fatalf("create tempdir: %v", err)
+ }
+ defer os.RemoveAll(dir)
+ kubeletPodsDir = dir
+
+ const (
+ testPodUID = "testuid"
+ testVolumeName = "testvolume"
+ testLogDirPath = "/var/log/pods/testns_testname_" + testPodUID
+ testLegacyLogDirPath = "/var/log/pods/" + testPodUID
+ )
+ testVolumePath := fmt.Sprintf("%s/%s/volumes/kubernetes.io~empty-dir/%s", dir, testPodUID, testVolumeName)
+
+ if err := os.MkdirAll(testVolumePath, 0755); err != nil {
+ t.Fatalf("Create test volume: %v", err)
+ }
+
+ for _, test := range []struct {
+ desc string
+ spec *specs.Spec
+ expected *specs.Spec
+ expectErr bool
+ expectUpdate bool
+ }{
+ {
+ desc: "volume annotations for sandbox",
+ spec: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.SandboxLogDir: testLogDirPath,
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ expected: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.SandboxLogDir: testLogDirPath,
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ "dev.gvisor.spec.mount." + testVolumeName + ".source": testVolumePath,
+ },
+ },
+ expectUpdate: true,
+ },
+ {
+ desc: "volume annotations for sandbox with legacy log path",
+ spec: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.SandboxLogDir: testLegacyLogDirPath,
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ expected: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.SandboxLogDir: testLegacyLogDirPath,
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ "dev.gvisor.spec.mount." + testVolumeName + ".source": testVolumePath,
+ },
+ },
+ expectUpdate: true,
+ },
+ {
+ desc: "tmpfs: volume annotations for container",
+ spec: &specs.Spec{
+ Mounts: []specs.Mount{
+ {
+ Destination: "/test",
+ Type: "bind",
+ Source: testVolumePath,
+ Options: []string{"ro"},
+ },
+ {
+ Destination: "/random",
+ Type: "bind",
+ Source: "/random",
+ Options: []string{"ro"},
+ },
+ },
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeContainer,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ expected: &specs.Spec{
+ Mounts: []specs.Mount{
+ {
+ Destination: "/test",
+ Type: "tmpfs",
+ Source: testVolumePath,
+ Options: []string{"ro"},
+ },
+ {
+ Destination: "/random",
+ Type: "bind",
+ Source: "/random",
+ Options: []string{"ro"},
+ },
+ },
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeContainer,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ expectUpdate: true,
+ },
+ {
+ desc: "bind: volume annotations for container",
+ spec: &specs.Spec{
+ Mounts: []specs.Mount{
+ {
+ Destination: "/test",
+ Type: "bind",
+ Source: testVolumePath,
+ Options: []string{"ro"},
+ },
+ },
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeContainer,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "container",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "bind",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ expected: &specs.Spec{
+ Mounts: []specs.Mount{
+ {
+ Destination: "/test",
+ Type: "bind",
+ Source: testVolumePath,
+ Options: []string{"ro"},
+ },
+ },
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeContainer,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "container",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "bind",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ expectUpdate: true,
+ },
+ {
+ desc: "should not return error without pod log directory",
+ spec: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ expected: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ "dev.gvisor.spec.mount." + testVolumeName + ".share": "pod",
+ "dev.gvisor.spec.mount." + testVolumeName + ".type": "tmpfs",
+ "dev.gvisor.spec.mount." + testVolumeName + ".options": "ro",
+ },
+ },
+ },
+ {
+ desc: "should return error if volume path does not exist",
+ spec: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.SandboxLogDir: testLogDirPath,
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ "dev.gvisor.spec.mount.notexist.share": "pod",
+ "dev.gvisor.spec.mount.notexist.type": "tmpfs",
+ "dev.gvisor.spec.mount.notexist.options": "ro",
+ },
+ },
+ expectErr: true,
+ },
+ {
+ desc: "no volume annotations for sandbox",
+ spec: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.SandboxLogDir: testLogDirPath,
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ },
+ },
+ expected: &specs.Spec{
+ Annotations: map[string]string{
+ annotations.SandboxLogDir: testLogDirPath,
+ annotations.ContainerType: annotations.ContainerTypeSandbox,
+ },
+ },
+ },
+ {
+ desc: "no volume annotations for container",
+ spec: &specs.Spec{
+ Mounts: []specs.Mount{
+ {
+ Destination: "/test",
+ Type: "bind",
+ Source: "/test",
+ Options: []string{"ro"},
+ },
+ {
+ Destination: "/random",
+ Type: "bind",
+ Source: "/random",
+ Options: []string{"ro"},
+ },
+ },
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeContainer,
+ },
+ },
+ expected: &specs.Spec{
+ Mounts: []specs.Mount{
+ {
+ Destination: "/test",
+ Type: "bind",
+ Source: "/test",
+ Options: []string{"ro"},
+ },
+ {
+ Destination: "/random",
+ Type: "bind",
+ Source: "/random",
+ Options: []string{"ro"},
+ },
+ },
+ Annotations: map[string]string{
+ annotations.ContainerType: annotations.ContainerTypeContainer,
+ },
+ },
+ },
+ } {
+ t.Run(test.desc, func(t *testing.T) {
+ bundle, err := ioutil.TempDir(dir, "test-bundle")
+ if err != nil {
+ t.Fatalf("Create test bundle: %v", err)
+ }
+ err = UpdateVolumeAnnotations(bundle, test.spec)
+ if test.expectErr {
+ if err == nil {
+ t.Fatal("Expected error, but got nil")
+ }
+ return
+ }
+ if err != nil {
+ t.Fatalf("Unexpected error: %v", err)
+ }
+ if !reflect.DeepEqual(test.expected, test.spec) {
+ t.Fatalf("Expected %+v, got %+v", test.expected, test.spec)
+ }
+ if test.expectUpdate {
+ b, err := ioutil.ReadFile(filepath.Join(bundle, "config.json"))
+ if err != nil {
+ t.Fatalf("Read spec from bundle: %v", err)
+ }
+ var spec specs.Spec
+ if err := json.Unmarshal(b, &spec); err != nil {
+ t.Fatalf("Unmarshal spec: %v", err)
+ }
+ if !reflect.DeepEqual(test.expected, &spec) {
+ t.Fatalf("Expected %+v, got %+v", test.expected, &spec)
+ }
+ }
+ })
+ }
+}
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-02 10:35:52 +0000