qemu: check guest status with qmp query-status

When guest panics or stops with unexpected internal
error, qemu process might still be running but we can
find out such situation with qmp. Then monitor can still
report such failures to watchers.

Fixes: #1963
Signed-off-by: Peng Tao <[email protected]>

virtcontainers/acrn.go

@@ -7,7 +7,6 @@ package virtcontainers
 
 import (
 	"context"
-	"errors"
 	"fmt"
 	"os"
 	"os/exec"
@@ -16,13 +15,15 @@ import (
 	"syscall"
 	"time"
 
+	opentracing "github.com/opentracing/opentracing-go"
+	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
+
 	"github.com/kata-containers/runtime/virtcontainers/device/config"
 	persistapi "github.com/kata-containers/runtime/virtcontainers/persist/api"
 	"github.com/kata-containers/runtime/virtcontainers/store"
 	"github.com/kata-containers/runtime/virtcontainers/types"
 	"github.com/kata-containers/runtime/virtcontainers/utils"
-	opentracing "github.com/opentracing/opentracing-go"
-	"github.com/sirupsen/logrus"
 )
 
 // AcrnState keeps Acrn's state
@@ -648,3 +649,11 @@ func (a *acrn) load(s persistapi.HypervisorState) {
 	a.info.PID = s.Pid
 	a.state.UUID = s.UUID
 }
+
+func (a *acrn) check() error {
+	if err := syscall.Kill(a.pid(), syscall.Signal(0)); err != nil {
+		return errors.Wrapf(err, "failed to ping acrn process")
+	}
+
+	return nil
+}

virtcontainers/fc.go

@@ -1006,3 +1006,11 @@ func (fc *firecracker) save() (s persistapi.HypervisorState) {
 func (fc *firecracker) load(s persistapi.HypervisorState) {
 	fc.info.PID = s.Pid
 }
+
+func (fc *firecracker) check() error {
+	if err := syscall.Kill(fc.pid(), syscall.Signal(0)); err != nil {
+		return errors.Wrapf(err, "failed to ping fc process")
+	}
+
+	return nil
+}

virtcontainers/hypervisor.go

@@ -671,6 +671,7 @@ type hypervisor interface {
 	pid() int
 	fromGrpc(ctx context.Context, hypervisorConfig *HypervisorConfig, store *store.VCStore, j []byte) error
 	toGrpc() ([]byte, error)
+	check() error
 
 	save() persistapi.HypervisorState
 	load(persistapi.HypervisorState)

virtcontainers/mock_hypervisor.go

@@ -121,3 +121,7 @@ func (m *mockHypervisor) save() (s persistapi.HypervisorState) {
 }
 
 func (m *mockHypervisor) load(s persistapi.HypervisorState) {}
+
+func (m *mockHypervisor) check() error {
+	return nil
+}

virtcontainers/mock_hypervisor_test.go

@@ -82,3 +82,9 @@ func TestMockHypervisorDisconnect(t *testing.T) {
 
 	m.disconnect()
 }
+
+func TestMockHypervisorCheck(t *testing.T) {
+	var m *mockHypervisor
+
+	assert.NoError(t, m.check())
+}

virtcontainers/monitor.go

@@ -7,7 +7,6 @@ package virtcontainers
 
 import (
 	"sync"
-	"syscall"
 	"time"
 
 	"github.com/pkg/errors"
@@ -127,7 +126,7 @@ func (m *monitor) watchAgent() {
 }
 
 func (m *monitor) watchHypervisor() error {
-	if err := syscall.Kill(m.sandbox.hypervisor.pid(), syscall.Signal(0)); err != nil {
+	if err := m.sandbox.hypervisor.check(); err != nil {
 		m.notify(errors.Wrapf(err, "failed to ping hypervisor process"))
 		return err
 	}

virtcontainers/qemu.go

@@ -10,7 +10,6 @@ import (
 	"context"
 	"encoding/hex"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"io/ioutil"
 	"math"
@@ -24,17 +23,17 @@ import (
 	"unsafe"
 
 	govmmQemu "github.com/intel/govmm/qemu"
-	"github.com/kata-containers/runtime/virtcontainers/pkg/uuid"
 	"github.com/opentracing/opentracing-go"
+	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
 
 	"github.com/kata-containers/runtime/virtcontainers/device/config"
 	persistapi "github.com/kata-containers/runtime/virtcontainers/persist/api"
+	"github.com/kata-containers/runtime/virtcontainers/pkg/uuid"
 	"github.com/kata-containers/runtime/virtcontainers/store"
 	"github.com/kata-containers/runtime/virtcontainers/types"
 	"github.com/kata-containers/runtime/virtcontainers/utils"
-
-	"golang.org/x/sys/unix"
 )
 
 // romFile is the file name of the ROM that can be used for virtio-pci devices.
@@ -2015,3 +2014,21 @@ func (q *qemu) load(s persistapi.HypervisorState) {
 		})
 	}
 }
+
+func (q *qemu) check() error {
+	err := q.qmpSetup()
+	if err != nil {
+		return err
+	}
+
+	status, err := q.qmpMonitorCh.qmp.ExecuteQueryStatus(q.qmpMonitorCh.ctx)
+	if err != nil {
+		return err
+	}
+
+	if status.Status == "internal-error" || status.Status == "guest-panicked" {
+		return errors.Errorf("guest failure: %s", status.Status)
+	}
+
+	return nil
+}