Merge pull request #317 from mlaventure/handle-exec-clingy-children

Handle exec clingy children

containerd-shim/main.go

@@ -106,7 +106,7 @@ func start(log *os.File) error {
 		case s := <-signals:
 			switch s {
 			case syscall.SIGCHLD:
-				exits, _ := osutils.Reap()
+				exits, _ := osutils.Reap(false)
 				for _, e := range exits {
 					// check to see if runtime is one of the processes that has exited
 					if e.Pid == p.pid() {
@@ -117,6 +117,9 @@ func start(log *os.File) error {
 			}
 			// runtime has exited so the shim can also exit
 			if exitShim {
+				// Wait for all the childs this process may have created
+				// (only needed for exec, but it won't hurt when done on init)
+				osutils.Reap(true)
 				// Let containerd take care of calling the runtime delete
 				f.Close()
 				p.Wait()

containerd/main.go

@@ -23,7 +23,6 @@ import (
 	"github.com/docker/containerd/api/grpc/server"
 	"github.com/docker/containerd/api/grpc/types"
 	"github.com/docker/containerd/api/http/pprof"
-	"github.com/docker/containerd/osutils"
 	"github.com/docker/containerd/supervisor"
 	"github.com/docker/docker/pkg/listeners"
 	"github.com/rcrowley/go-metrics"
@@ -160,7 +159,6 @@ func main() {
 func daemon(context *cli.Context) error {
 	s := make(chan os.Signal, 2048)
 	signal.Notify(s, syscall.SIGTERM, syscall.SIGINT)
-	osutils.SetSubreaper(1)
 	sv, err := supervisor.New(
 		context.String("state-dir"),
 		context.String("runtime"),

osutils/reaper.go

@@ -12,13 +12,17 @@ type Exit struct {
 
 // Reap reaps all child processes for the calling process and returns their
 // exit information
-func Reap() (exits []Exit, err error) {
+func Reap(wait bool) (exits []Exit, err error) {
 	var (
 		ws  syscall.WaitStatus
 		rus syscall.Rusage
 	)
+	flag := syscall.WNOHANG
+	if wait {
+		flag = 0
+	}
 	for {
-		pid, err := syscall.Wait4(-1, &ws, syscall.WNOHANG, &rus)
+		pid, err := syscall.Wait4(-1, &ws, flag, &rus)
 		if err != nil {
 			if err == syscall.ECHILD {
 				return exits, nil

runtime/container.go

@@ -14,6 +14,7 @@ import (
 	"github.com/Sirupsen/logrus"
 	"github.com/docker/containerd/specs"
 	ocs "github.com/opencontainers/runtime-spec/specs-go"
+	"golang.org/x/sys/unix"
 )
 
 // Container defines the operations allowed on a container
@@ -480,13 +481,34 @@ func (c *container) createCmd(pid string, cmd *exec.Cmd, p *process) error {
 		}
 		return err
 	}
+	// We need the pid file to have been written to run
+	defer func() {
 		go func() {
 			err := p.cmd.Wait()
 			if err == nil {
 				p.cmdSuccess = true
 			}
+
+			if same, err := p.isSameProcess(); same && p.pid > 0 {
+				// The process changed its PR_SET_PDEATHSIG, so force
+				// kill it
+				logrus.Infof("containerd: %s:%s (pid %v) has become an orphan, killing it", p.container.id, p.id, p.pid)
+				err = unix.Kill(p.pid, syscall.SIGKILL)
+				if err != nil && err != syscall.ESRCH {
+					logrus.Errorf("containerd: unable to SIGKILL %s:%s (pid %v): %v", p.container.id, p.id, p.pid, err)
+				} else {
+					for {
+						err = unix.Kill(p.pid, 0)
+						if err != nil {
+							break
+						}
+						time.Sleep(5 * time.Millisecond)
+					}
+				}
+			}
 			close(p.cmdDoneCh)
 		}()
+	}()
 	if err := c.waitForCreate(p, cmd); err != nil {
 		return err
 	}

runtime/process.go

@@ -228,24 +228,31 @@ func (p *process) Resize(w, h int) error {
 	return err
 }
 
+func (p *process) updateExitStatusFile(status int) (int, error) {
+	p.stateLock.Lock()
+	p.state = Stopped
+	p.stateLock.Unlock()
+	err := ioutil.WriteFile(filepath.Join(p.root, ExitStatusFile), []byte(fmt.Sprintf("%d", status)), 0644)
+	return status, err
+}
+
 func (p *process) handleSigkilledShim(rst int, rerr error) (int, error) {
 	if p.cmd == nil || p.cmd.Process == nil {
 		e := unix.Kill(p.pid, 0)
 		if e == syscall.ESRCH {
-			return rst, rerr
+			logrus.Warnf("containerd: %s:%s (pid %d) does not exist", p.container.id, p.id, p.pid)
+			// The process died while containerd was down (probably of
+			// SIGKILL, but no way to be sure)
+			return p.updateExitStatusFile(255)
 		}
 
 		// If it's not the same process, just mark it stopped and set
 		// the status to 255
 		if same, err := p.isSameProcess(); !same {
 			logrus.Warnf("containerd: %s:%s (pid %d) is not the same process anymore (%v)", p.container.id, p.id, p.pid, err)
-			p.stateLock.Lock()
-			p.state = Stopped
-			p.stateLock.Unlock()
 			// Create the file so we get the exit event generated once monitor kicks in
-			// without going to this all process again
+			// without having to go through all this process again
-			rerr = ioutil.WriteFile(filepath.Join(p.root, ExitStatusFile), []byte("255"), 0644)
+			return p.updateExitStatusFile(255)
-			return 255, nil
 		}
 
 		ppid, err := readProcStatField(p.pid, 4)
@@ -255,19 +262,21 @@ func (p *process) handleSigkilledShim(rst int, rerr error) (int, error) {
 		if ppid == "1" {
 			logrus.Warnf("containerd: %s:%s shim died, killing associated process", p.container.id, p.id)
 			unix.Kill(p.pid, syscall.SIGKILL)
+			if err != nil && err != syscall.ESRCH {
+				return 255, fmt.Errorf("containerd: unable to SIGKILL %s:%s (pid %v): %v", p.container.id, p.id, p.pid, err)
+			}
+
 			// wait for the process to die
 			for {
 				e := unix.Kill(p.pid, 0)
 				if e == syscall.ESRCH {
 					break
 				}
-				time.Sleep(10 * time.Millisecond)
+				time.Sleep(5 * time.Millisecond)
 			}
-
-			rst = 128 + int(syscall.SIGKILL)
 			// Create the file so we get the exit event generated once monitor kicks in
-			// without going to this all process again
+			// without having to go through all this process again
-			rerr = ioutil.WriteFile(filepath.Join(p.root, ExitStatusFile), []byte(fmt.Sprintf("%d", rst)), 0644)
+			return p.updateExitStatusFile(128 + int(syscall.SIGKILL))
 		}
 
 		return rst, rerr
@@ -286,29 +295,8 @@ func (p *process) handleSigkilledShim(rst int, rerr error) (int, error) {
 	if shimStatus.Signaled() && shimStatus.Signal() == syscall.SIGKILL {
 		logrus.Debugf("containerd: ExitStatus(container: %s, process: %s): shim was SIGKILL'ed reaping its child with pid %d", p.container.id, p.id, p.pid)
 
-		var (
-			status unix.WaitStatus
-			rusage unix.Rusage
-			wpid   int
-		)
-
-		// Some processes change their PR_SET_PDEATHSIG, so force kill them
-		unix.Kill(p.pid, syscall.SIGKILL)
-
-		for wpid == 0 {
-			wpid, e = unix.Wait4(p.pid, &status, unix.WNOHANG, &rusage)
-			if e != nil {
-				logrus.Debugf("containerd: ExitStatus(container: %s, process: %s): Wait4(%d): %v", p.container.id, p.id, p.pid, rerr)
-				return rst, rerr
-			}
-		}
-
-		if wpid == p.pid {
 		rerr = nil
 		rst = 128 + int(shimStatus.Signal())
-		} else {
-			logrus.Errorf("containerd: ExitStatus(container: %s, process: %s): unexpected returned pid from wait4 %v (expected %v)", p.container.id, p.id, wpid, p.pid)
-		}
 
 		p.stateLock.Lock()
 		p.state = Stopped

supervisor/exit.go

@@ -73,6 +73,11 @@ func (s *Supervisor) execExit(t *ExecExitTask) error {
 	if err := container.RemoveProcess(t.PID); err != nil {
 		logrus.WithField("error", err).Error("containerd: find container for pid")
 	}
+	// If the exec spawned children which are still using its IO
+	// waiting here will block until they die or close their IO
+	// descriptors.
+	// Hence, we use a go routine to avoid block all other operations
+	go func() {
 		t.Process.Wait()
 		s.notifySubscribers(Event{
 			Timestamp: time.Now(),
@@ -81,5 +86,6 @@ func (s *Supervisor) execExit(t *ExecExitTask) error {
 			PID:       t.PID,
 			Status:    t.Status,
 		})
+	}()
 	return nil
 }