utils: Add waitLocalProcess function

Refactored some of the hypervisors to remove the duplicated code used to
trigger a shutdown.

Also added some unit tests.

Signed-off-by: James O. D. Hunt <james.o.hunt@intel.com>

src/runtime/virtcontainers/utils/utils.go

@@ -12,7 +12,10 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"syscall"
+	"time"
 
+	"github.com/sirupsen/logrus"
 	"github.com/vishvananda/netlink"
 
 	pbTypes "github.com/kata-containers/kata-containers/src/runtime/virtcontainers/pkg/agent/protocols"
@@ -312,3 +315,63 @@ func ConvertNetlinkFamily(netlinkFamily int32) pbTypes.IPFamily {
 		return pbTypes.IPFamily_v4
 	}
 }
+
+// WaitLocalProcess waits for the specified process for up to timeoutSecs seconds.
+//
+// Notes:
+//
+// - If the initial signal is zero, the specified process is assumed to be
+//   attempting to stop itself.
+// - If the initial signal is not zero, it will be sent to the process before
+//   checking if it is running.
+// - If the process has not ended after the timeout value, it will be forcibly killed.
+func WaitLocalProcess(pid int, timeoutSecs uint, initialSignal syscall.Signal, logger *logrus.Entry) error {
+	var err error
+
+	// Don't support process groups
+	if pid <= 0 {
+		return errors.New("can only wait for a single process")
+	}
+
+	if initialSignal != syscall.Signal(0) {
+		if err = syscall.Kill(pid, initialSignal); err != nil {
+			if err == syscall.ESRCH {
+				return nil
+			}
+
+			return fmt.Errorf("Failed to send initial signal %v to process %v: %v", initialSignal, pid, err)
+		}
+	}
+
+	pidRunning := true
+
+	// Wait for the VM process to terminate
+	startTime := time.Now()
+
+	for {
+		if err = syscall.Kill(pid, syscall.Signal(0)); err != nil {
+			pidRunning = false
+			break
+		}
+
+		if time.Since(startTime).Seconds() >= float64(timeoutSecs) {
+			pidRunning = true
+
+			logger.Warnf("process %v still running after waiting %ds", pid, timeoutSecs)
+
+			break
+		}
+
+		// Brief pause to avoid a busy loop
+		time.Sleep(time.Duration(50) * time.Millisecond)
+	}
+
+	if pidRunning {
+		// Force process to die
+		if err = syscall.Kill(pid, syscall.SIGKILL); err != nil {
+			return fmt.Errorf("Failed to stop process %v: %s", pid, err)
+		}
+	}
+
+	return nil
+}

src/runtime/virtcontainers/utils/utils_test.go

@@ -9,14 +9,19 @@ import (
 	"fmt"
 	"io/ioutil"
 	"os"
+	"os/exec"
 	"path/filepath"
 	"reflect"
 	"strings"
+	"syscall"
 	"testing"
 
+	"github.com/sirupsen/logrus"
 	"github.com/stretchr/testify/assert"
 )
 
+const waitLocalProcessTimeoutSecs = 3
+
 func TestFileCopySuccessful(t *testing.T) {
 	assert := assert.New(t)
 	fileContent := "testContent"
@@ -375,3 +380,99 @@ func TestToBytes(t *testing.T) {
 	expected := uint64(1073741824)
 	assert.Equal(expected, result)
 }
+
+func TestWaitLocalProcessInvalidSignal(t *testing.T) {
+	assert := assert.New(t)
+
+	const invalidSignal = syscall.Signal(999)
+
+	cmd := exec.Command("sleep", "999")
+	err := cmd.Start()
+	assert.NoError(err)
+
+	pid := cmd.Process.Pid
+
+	logger := logrus.WithField("foo", "bar")
+
+	err = WaitLocalProcess(pid, waitLocalProcessTimeoutSecs, invalidSignal, logger)
+	assert.Error(err)
+
+	err = syscall.Kill(pid, syscall.SIGTERM)
+	assert.NoError(err)
+
+	err = cmd.Wait()
+
+	// This will error because we killed the process without the knowledge
+	// of exec.Command.
+	assert.Error(err)
+}
+
+func TestWaitLocalProcessInvalidPid(t *testing.T) {
+	assert := assert.New(t)
+
+	invalidPids := []int{-999, -173, -3, -2, -1, 0}
+
+	logger := logrus.WithField("foo", "bar")
+
+	for i, pid := range invalidPids {
+		msg := fmt.Sprintf("test[%d]: %v", i, pid)
+
+		err := WaitLocalProcess(pid, waitLocalProcessTimeoutSecs, syscall.Signal(0), logger)
+		assert.Error(err, msg)
+	}
+}
+
+func TestWaitLocalProcessBrief(t *testing.T) {
+	assert := assert.New(t)
+
+	cmd := exec.Command("true")
+	err := cmd.Start()
+	assert.NoError(err)
+
+	pid := cmd.Process.Pid
+
+	logger := logrus.WithField("foo", "bar")
+
+	err = WaitLocalProcess(pid, waitLocalProcessTimeoutSecs, syscall.SIGKILL, logger)
+	assert.NoError(err)
+
+	_ = cmd.Wait()
+}
+
+func TestWaitLocalProcessLongRunningPreKill(t *testing.T) {
+	assert := assert.New(t)
+
+	cmd := exec.Command("sleep", "999")
+	err := cmd.Start()
+	assert.NoError(err)
+
+	pid := cmd.Process.Pid
+
+	logger := logrus.WithField("foo", "bar")
+
+	err = WaitLocalProcess(pid, waitLocalProcessTimeoutSecs, syscall.SIGKILL, logger)
+	assert.NoError(err)
+
+	_ = cmd.Wait()
+}
+
+func TestWaitLocalProcessLongRunning(t *testing.T) {
+	assert := assert.New(t)
+
+	cmd := exec.Command("sleep", "999")
+	err := cmd.Start()
+	assert.NoError(err)
+
+	pid := cmd.Process.Pid
+
+	logger := logrus.WithField("foo", "bar")
+
+	// Don't wait for long as the process isn't actually trying to stop,
+	// so it will have to timeout and then be killed.
+	const timeoutSecs = 1
+
+	err = WaitLocalProcess(pid, timeoutSecs, syscall.Signal(0), logger)
+	assert.NoError(err)
+
+	_ = cmd.Wait()
+}