func waitInitReady(ctx *VmContext) {
conn, err := UnixSocketConnect(ctx.HyperSockName)
if err != nil {
glog.Error("Cannot connect to hyper socket ", err.Error())
ctx.Hub <- &InitFailedEvent{
Reason: "Cannot connect to hyper socket " + err.Error(),
}
return
}
glog.Info("Wating for init messages...")
msg, err := readVmMessage(conn.(*net.UnixConn))
if err != nil {
glog.Error("read init message failed... ", err.Error())
ctx.Hub <- &InitFailedEvent{
Reason: "read init message failed... " + err.Error(),
}
conn.Close()
} else if msg.code == INIT_READY {
glog.Info("Get init ready message")
ctx.Hub <- &InitConnectedEvent{conn: conn.(*net.UnixConn)}
go waitCmdToInit(ctx, conn.(*net.UnixConn))
} else {
glog.Warningf("Get init message %d", msg.code)
ctx.Hub <- &InitFailedEvent{
Reason: fmt.Sprintf("Get init message %d", msg.code),
}
conn.Close()
}
}
func InitDriver() *XenDriver {
xd := &XenDriver{}
if err := xd.Initialize(); err == nil {
glog.Info("Xen Driver Loaded.")
globalDriver = xd
return globalDriver
} else {
glog.Info("Xen Driver Load failed: ", err.Error())
return nil
}
}
func qmpCommander(handler chan QmpInteraction, conn *net.UnixConn, session *QmpSession, feedback chan QmpInteraction) {
glog.V(1).Info("Begin process command session")
for _, cmd := range session.commands {
msg, err := json.Marshal(*cmd)
if err != nil {
handler <- qmpFail("cannot marshal command", session.callback)
return
}
success := false
var qe *QmpError = nil
for repeat := 0; !success && repeat < 3; repeat++ {
if len(cmd.Scm) > 0 {
glog.V(1).Infof("send cmd with scm (%d bytes) (%d) %s", len(cmd.Scm), repeat+1, string(msg))
f, _ := conn.File()
fd := f.Fd()
syscall.Sendmsg(int(fd), msg, cmd.Scm, nil, 0)
} else {
glog.V(1).Infof("sending command (%d) %s", repeat+1, string(msg))
conn.Write(msg)
}
res, ok := <-feedback
if !ok {
glog.Info("QMP command result chan closed")
return
}
switch res.MessageType() {
case QMP_RESULT:
success = true
break
//success
case QMP_ERROR:
glog.Warning("got one qmp error")
qe = res.(*QmpError)
time.Sleep(1000 * time.Millisecond)
case QMP_INTERNAL_ERROR:
glog.Info("QMP quit... commander quit... ")
return
}
}
if !success {
handler <- qe.Finish(session.callback)
return
}
}
handler <- session.Finish()
return
}
func waitConsoleOutput(ctx *VmContext) {
conn, err := unixSocketConnect(ctx.consoleSockName)
if err != nil {
glog.Error("failed to connected to ", ctx.consoleSockName, " ", err.Error())
return
}
glog.V(1).Info("connected to ", ctx.consoleSockName)
tc, err := telnet.NewConn(conn)
if err != nil {
glog.Error("fail to init telnet connection to ", ctx.consoleSockName, ": ", err.Error())
return
}
glog.V(1).Infof("connected %s as telnet mode.", ctx.consoleSockName)
cout := make(chan string, 128)
go ttyLiner(tc, cout)
for {
line, ok := <-cout
if ok {
glog.V(1).Info("[console] ", line)
} else {
glog.Info("console output end")
break
}
}
}
func stateCleaning(ctx *VmContext, ev QemuEvent) {
if processed := commonStateHandler(ctx, ev, false); processed {
} else if processed, success := deviceRemoveHandler(ctx, ev); processed {
if !success {
glog.Warning("fail to unplug devices for stop")
ctx.poweroffVM(true, "fail to unplug devices")
ctx.Become(stateDestroying, "DESTROYING")
} else if ctx.deviceReady() {
// ctx.reset()
// ctx.unsetTimeout()
// ctx.reportPodStopped()
// glog.V(1).Info("device ready, could run pod.")
// ctx.Become(stateInit, "INIT")
ctx.vm <- &DecodedMessage{
code: INIT_READY,
message: []byte{},
}
glog.V(1).Info("device ready, could run pod.")
}
} else if processed := initFailureHandler(ctx, ev); processed {
ctx.poweroffVM(true, "fail to unplug devices")
ctx.Become(stateDestroying, "DESTROYING")
} else {
switch ev.Event() {
case COMMAND_RELEASE:
glog.Info("vm cleaning to idle, got release, quit")
ctx.reportVmShutdown()
ctx.Become(stateDestroying, "DESTROYING")
case EVENT_QEMU_TIMEOUT:
glog.Warning("Qemu did not exit in time, try to stop it")
ctx.poweroffVM(true, "pod stopp/unplug timeout")
ctx.Become(stateDestroying, "DESTROYING")
case COMMAND_ACK:
ack := ev.(*CommandAck)
glog.V(1).Infof("[cleaning] Got reply to %d: '%s'", ack.reply, string(ack.msg))
if ack.reply == INIT_READY {
ctx.reset()
ctx.unsetTimeout()
ctx.reportPodStopped()
glog.Info("init has been acknowledged, could run pod.")
ctx.Become(stateInit, "INIT")
}
default:
glog.V(1).Info("got event message while cleaning")
}
}
}
// InitDeviceContext will init device info in context
func (ctx *VmContext) InitDeviceContext(spec *pod.UserPod, wg *sync.WaitGroup,
cInfo []*ContainerInfo, vInfo []*VolumeInfo) {
ctx.lock.Lock()
defer ctx.lock.Unlock()
for i := 0; i < InterfaceCount; i++ {
ctx.progress.adding.networks[i] = true
}
if cInfo == nil {
cInfo = []*ContainerInfo{}
}
if vInfo == nil {
vInfo = []*VolumeInfo{}
}
ctx.initVolumeMap(spec)
if glog.V(3) {
for i, c := range cInfo {
glog.Infof("#%d Container Info:", i)
b, err := json.MarshalIndent(c, "...|", " ")
if err == nil {
glog.Info("\n", string(b))
}
}
}
containers := make([]VmContainer, len(spec.Containers))
for i, container := range spec.Containers {
ctx.initContainerInfo(i, &containers[i], &container)
ctx.setContainerInfo(i, &containers[i], cInfo[i])
if spec.Tty {
containers[i].Tty = ctx.attachId
ctx.attachId++
ctx.ptys.ttys[containers[i].Tty] = newAttachments(i, true)
}
}
ctx.vmSpec = &VmPod{
Hostname: spec.Name,
Containers: containers,
Interfaces: nil,
Routes: nil,
ShareDir: ShareDirTag,
}
for _, vol := range vInfo {
ctx.setVolumeInfo(vol)
}
ctx.userSpec = spec
ctx.wg = wg
}
func stateDestroying(ctx *VmContext, ev QemuEvent) {
if processed, _ := deviceRemoveHandler(ctx, ev); processed {
if closed := ctx.tryClose(); closed {
glog.Info("resources reclaimed, quit...")
}
} else {
switch ev.Event() {
case EVENT_QMP_EVENT:
if ev.(*QmpEvent).Type == QMP_EVENT_SHUTDOWN {
glog.Info("Got QMP shutdown event")
ctx.unsetTimeout()
if closed := ctx.onQemuExit(false); closed {
glog.Info("VM Context closed.")
}
}
case EVENT_QEMU_KILL:
glog.Info("Got Qemu force killed message")
ctx.unsetTimeout()
if closed := ctx.onQemuExit(true); closed {
glog.Info("VM Context closed.")
}
case ERROR_INTERRUPTED:
glog.V(1).Info("Connection interrupted while destroying")
case COMMAND_RELEASE:
glog.Info("vm destroying, got release")
ctx.reportVmShutdown()
case EVENT_QEMU_TIMEOUT:
glog.Info("Device removing timeout")
ctx.Close()
default:
glog.Warning("got event during vm cleaning up")
}
}
}
func DriversProbe() hypervisor.HypervisorDriver {
xd := xen.InitDriver()
if xd != nil {
glog.Info("Xen Driver Loaded.")
return xd
}
qd := &qemu.QemuDriver{}
if err := qd.Initialize(); err == nil {
glog.Info("Qemu Driver Loaded")
return qd
} else {
glog.Info("Qemu Driver Load failed: ", err.Error())
}
glog.Error("No driver available")
return nil
}
func (qc *QemuContext) Kill(ctx *hypervisor.VmContext) {
defer func() {
err := recover()
if glog.V(1) && err != nil {
glog.Info("kill qemu, but channel has already been closed")
}
}()
qc.wdt <- "kill"
}
func stateTerminating(ctx *VmContext, ev QemuEvent) {
switch ev.Event() {
case EVENT_QMP_EVENT:
if ev.(*QmpEvent).Type == QMP_EVENT_SHUTDOWN {
glog.Info("Got QMP shutdown event while terminating, go to cleaning up")
ctx.unsetTimeout()
if closed := ctx.onQemuExit(true); !closed {
ctx.Become(stateDestroying, "DESTROYING")
}
}
case EVENT_QEMU_KILL:
glog.Info("Got Qemu force killed message, go to cleaning up")
ctx.unsetTimeout()
if closed := ctx.onQemuExit(true); !closed {
ctx.Become(stateDestroying, "DESTROYING")
}
case COMMAND_RELEASE:
glog.Info("vm terminating, got release")
ctx.reportVmShutdown()
case COMMAND_ACK:
ack := ev.(*CommandAck)
glog.V(1).Infof("[Terminating] Got reply to %d: '%s'", ack.reply, string(ack.msg))
if ack.reply == INIT_DESTROYPOD {
glog.Info("POD destroyed ", string(ack.msg))
ctx.poweroffVM(false, "")
}
case ERROR_CMD_FAIL:
ack := ev.(*CommandError)
if ack.context.code == INIT_DESTROYPOD {
glog.Warning("Destroy pod failed")
ctx.poweroffVM(true, "Destroy pod failed")
}
case EVENT_QEMU_TIMEOUT:
glog.Warning("Qemu did not exit in time, try to stop it")
ctx.poweroffVM(true, "vm terminating timeout")
case ERROR_INTERRUPTED:
glog.V(1).Info("Connection interrupted while terminating")
default:
glog.V(1).Info("got event during terminating")
}
}
func (daemon *Daemon) CmdPodRun(job *engine.Job) error {
// we can only support 1024 Pods
if daemon.GetRunningPodNum() >= 1024 {
return fmt.Errorf("Pod full, the maximum Pod is 1024!")
}
podArgs := job.Args[0]
vmId := fmt.Sprintf("vm-%s", pod.RandStr(10, "alpha"))
podId := fmt.Sprintf("pod-%s", pod.RandStr(10, "alpha"))
glog.Info(podArgs)
code, cause, err := daemon.StartPod(podId, vmId, podArgs)
if err != nil {
daemon.KillVm(vmId)
glog.Error(err.Error())
return err
}
if err := daemon.WritePodAndContainers(podId); err != nil {
glog.V(1).Info("Found an error while saveing the Containers info")
return err
}
data, err := daemon.GetPodByName(podId)
if err != nil {
return err
}
userPod, err := pod.ProcessPodBytes(data)
if err != nil {
return err
}
vm := &Vm{
Id: vmId,
Pod: daemon.podList[podId],
Status: types.S_VM_ASSOCIATED,
Cpu: userPod.Resource.Vcpu,
Mem: userPod.Resource.Memory,
}
daemon.podList[podId].Vm = vmId
daemon.AddVm(vm)
// Prepare the qemu status to client
v := &engine.Env{}
v.Set("ID", podId)
v.SetInt("Code", code)
v.Set("Cause", cause)
if _, err := v.WriteTo(job.Stdout); err != nil {
return err
}
return nil
}
func (pts *pseudoTtys) ptyConnect(ctx *VmContext, container int, session uint64, tty *TtyIO) {
pts.lock.Lock()
if ta, ok := pts.ttys[session]; ok {
ta.attach(tty)
} else {
pts.ttys[session] = newAttachmentsWithTty(container, false, tty)
}
pts.lock.Unlock()
if tty.Stdin != nil {
go func() {
buf := make([]byte, 32)
defer pts.Detach(ctx, session, tty)
defer func() { recover() }()
for {
nr, err := tty.Stdin.Read(buf)
if err != nil {
glog.Info("a stdin closed, ", err.Error())
return
} else if nr == 1 && buf[0] == ExitChar {
glog.Info("got stdin detach char, exit term")
return
}
glog.V(3).Infof("trying to input char: %d and %d chars", buf[0], nr)
mbuf := make([]byte, nr)
copy(mbuf, buf[:nr])
pts.channel <- &ttyMessage{
session: session,
message: mbuf[:nr],
}
}
}()
}
return
}
// launchQemu run qemu and wait it's quit, includes
func launchQemu(ctx *VmContext) {
qemu, err := exec.LookPath("qemu-system-x86_64")
if err != nil {
ctx.hub <- &QemuExitEvent{message: "can not find qemu executable"}
return
}
args := ctx.QemuArguments()
if glog.V(1) {
glog.Info("cmdline arguments: ", strings.Join(args, " "))
}
go waitConsoleOutput(ctx)
pipe := make([]int, 2)
err = syscall.Pipe(pipe)
if err != nil {
glog.Error("fail to create pipe")
ctx.hub <- &QemuExitEvent{message: "fail to create pipe"}
return
}
err = daemon(qemu, append([]string{"qemu-system-x86_64"}, args...), pipe[1])
if err != nil {
//fail to daemonize
glog.Error("try to start qemu failed")
ctx.hub <- &QemuExitEvent{message: "try to start qemu failed"}
return
}
buf := make([]byte, 4)
nr, err := syscall.Read(pipe[0], buf)
if err != nil || nr != 4 {
glog.Error("try to start qemu failed")
ctx.hub <- &QemuExitEvent{message: "try to start qemu failed"}
return
}
syscall.Close(pipe[1])
syscall.Close(pipe[0])
pid := binary.BigEndian.Uint32(buf[:nr])
glog.V(1).Infof("starting daemon with pid: %d", pid)
err = ctx.watchPid(int(pid))
if err != nil {
glog.Error("watch qemu process failed")
ctx.hub <- &QemuExitEvent{message: "watch qemu process failed"}
return
}
}
// launchQemu run qemu and wait it's quit, includes
func launchQemu(qc *QemuContext, ctx *hypervisor.VmContext) {
qemu := qc.driver.executable
if qemu == "" {
ctx.Hub <- &hypervisor.VmStartFailEvent{Message: "can not find qemu executable"}
return
}
args := qc.arguments(ctx)
if glog.V(1) {
glog.Info("cmdline arguments: ", strings.Join(args, " "))
}
pipe := make([]int, 2)
err := syscall.Pipe(pipe)
if err != nil {
glog.Error("fail to create pipe")
ctx.Hub <- &hypervisor.VmStartFailEvent{Message: "fail to create pipe"}
return
}
err = daemon(qemu, append([]string{"qemu-system-x86_64"}, args...), pipe[1])
if err != nil {
//fail to daemonize
glog.Error("try to start qemu failed")
ctx.Hub <- &hypervisor.VmStartFailEvent{Message: "try to start qemu failed"}
return
}
buf := make([]byte, 4)
nr, err := syscall.Read(pipe[0], buf)
if err != nil || nr != 4 {
glog.Error("try to start qemu failed")
ctx.Hub <- &hypervisor.VmStartFailEvent{Message: "try to start qemu failed"}
return
}
syscall.Close(pipe[1])
syscall.Close(pipe[0])
pid := binary.BigEndian.Uint32(buf[:nr])
glog.V(1).Infof("starting daemon with pid: %d", pid)
err = ctx.DCtx.(*QemuContext).watchPid(int(pid), ctx.Hub)
if err != nil {
glog.Error("watch qemu process failed")
ctx.Hub <- &hypervisor.VmStartFailEvent{Message: "watch qemu process failed"}
return
}
}
// state machine
func commonStateHandler(ctx *VmContext, ev QemuEvent, hasPod bool) bool {
processed := true
switch ev.Event() {
case EVENT_QMP_EVENT:
if ev.(*QmpEvent).Type == QMP_EVENT_SHUTDOWN {
glog.Info("Got QMP shutdown event, go to cleaning up")
ctx.unsetTimeout()
if closed := ctx.onQemuExit(hasPod); !closed {
ctx.Become(stateDestroying, "DESTROYING")
}
} else {
processed = false
}
case ERROR_INTERRUPTED:
glog.Info("Connection interrupted, quit...")
ctx.exitVM(true, "connection to VM broken", hasPod)
case COMMAND_SHUTDOWN:
glog.Info("got shutdown command, shutting down")
ctx.exitVM(false, "", hasPod)
default:
processed = false
}
return processed
}
func statePodStopping(ctx *VmContext, ev QemuEvent) {
if processed := commonStateHandler(ctx, ev, true); processed {
} else {
switch ev.Event() {
case COMMAND_RELEASE:
glog.Info("pod stopping, got release, quit.")
ctx.unsetTimeout()
ctx.shutdownVM(false, "got release, quit")
ctx.Become(stateTerminating, "TERMINATING")
ctx.reportVmShutdown()
case COMMAND_ACK:
ack := ev.(*CommandAck)
glog.V(1).Infof("[Stopping] got init ack to %d", ack.reply)
if ack.reply == INIT_STOPPOD {
glog.Info("POD stopped ", string(ack.msg))
ctx.detatchDevice()
ctx.Become(stateCleaning, "CLEANING")
}
case ERROR_CMD_FAIL:
ack := ev.(*CommandError)
if ack.context.code == INIT_STOPPOD {
ctx.unsetTimeout()
ctx.shutdownVM(true, "Stop pod failed as init report")
ctx.Become(stateTerminating, "TERMINATING")
glog.Error("Stop pod failed as init report")
}
case EVENT_QEMU_TIMEOUT:
reason := "stopping POD timeout"
ctx.shutdownVM(true, reason)
ctx.Become(stateTerminating, "TERMINATING")
glog.Error(reason)
default:
glog.Warning("got unexpected event during pod stopping")
}
}
}
func stateInit(ctx *VmContext, ev QemuEvent) {
if processed := commonStateHandler(ctx, ev, false); processed {
//processed by common
} else if processed := initFailureHandler(ctx, ev); processed {
ctx.shutdownVM(true, "Fail during init environment")
ctx.Become(stateDestroying, "DESTROYING")
} else {
switch ev.Event() {
case EVENT_QEMU_EXIT:
glog.Error("Qemu did not start up properly, go to cleaning up")
ctx.reportVmFault("Qemu did not start up properly, go to cleaning up")
ctx.Close()
case EVENT_INIT_CONNECTED:
glog.Info("begin to wait vm commands")
ctx.reportVmRun()
case COMMAND_RELEASE:
glog.Info("no pod on vm, got release, quit.")
ctx.shutdownVM(false, "")
ctx.Become(stateDestroying, "DESTRYING")
ctx.reportVmShutdown()
case COMMAND_EXEC:
ctx.execCmd(ev.(*ExecCommand))
case COMMAND_WINDOWSIZE:
cmd := ev.(*WindowSizeCommand)
ctx.setWindowSize(cmd.ClientTag, cmd.Size)
case COMMAND_RUN_POD, COMMAND_REPLACE_POD:
glog.Info("got spec, prepare devices")
if ok := ctx.prepareDevice(ev.(*RunPodCommand)); ok {
ctx.setTimeout(60)
ctx.Become(stateStarting, "STARTING")
}
default:
glog.Warning("got event during pod initiating")
}
}
}
func stateRunning(ctx *VmContext, ev QemuEvent) {
if processed := commonStateHandler(ctx, ev, true); processed {
} else if processed := initFailureHandler(ctx, ev); processed {
ctx.shutdownVM(true, "Fail during reconnect to a running pod")
ctx.Become(stateTerminating, "TERMINATING")
} else {
switch ev.Event() {
case COMMAND_STOP_POD:
ctx.stopPod()
ctx.Become(statePodStopping, "STOPPING")
case COMMAND_RELEASE:
glog.Info("pod is running, got release command, let qemu fly")
ctx.Become(nil, "NONE")
ctx.reportSuccess("", nil)
case COMMAND_EXEC:
ctx.execCmd(ev.(*ExecCommand))
case COMMAND_ATTACH:
ctx.attachCmd(ev.(*AttachCommand))
case COMMAND_WINDOWSIZE:
cmd := ev.(*WindowSizeCommand)
if ctx.userSpec.Tty {
ctx.setWindowSize(cmd.ClientTag, cmd.Size)
}
case EVENT_POD_FINISH:
result := ev.(*PodFinished)
ctx.reportPodFinished(result)
ctx.shutdownVM(false, "")
ctx.Become(stateTerminating, "TERMINATING")
case COMMAND_ACK:
ack := ev.(*CommandAck)
glog.V(1).Infof("[running] got init ack to %d", ack.reply)
case ERROR_CMD_FAIL:
ack := ev.(*CommandError)
if ack.context.code == INIT_EXECCMD {
cmd := ExecCommand{}
json.Unmarshal(ack.context.message, &cmd)
ctx.ptys.Close(ctx, cmd.Sequence)
glog.V(0).Infof("Exec command %s on session %d failed", cmd.Command[0], cmd.Sequence)
}
default:
glog.Warning("got unexpected event during pod running")
}
}
}
func (ctx *VmContext) prepareDevice(cmd *RunPodCommand) bool {
if len(cmd.Spec.Containers) != len(cmd.Containers) {
ctx.reportBadRequest("Spec and Container Info mismatch")
return false
}
ctx.InitDeviceContext(cmd.Spec, cmd.Containers, cmd.Volumes)
if glog.V(2) {
res, _ := json.MarshalIndent(*ctx.vmSpec, " ", " ")
glog.Info("initial vm spec: ", string(res))
}
ctx.allocateNetworks()
ctx.addBlockDevices()
return true
}
func qmpReceiver(ch chan QmpInteraction, decoder *json.Decoder) {
glog.V(0).Info("Begin receive QMP message")
for {
rsp := &QmpResponse{}
if err := decoder.Decode(rsp); err == io.EOF {
glog.Info("QMP exit as got EOF")
ch <- &QmpInternalError{cause: err.Error()}
return
} else if err != nil {
glog.Error("QMP receive and decode error: ", err.Error())
ch <- &QmpInternalError{cause: err.Error()}
return
}
msg := rsp.msg
ch <- msg
if msg.MessageType() == QMP_EVENT && msg.(*QmpEvent).Type == QMP_EVENT_SHUTDOWN {
glog.V(0).Info("Shutdown, quit QMP receiver")
return
}
}
}
func waitCmdToInit(ctx *VmContext, init *net.UnixConn) {
looping := true
cmds := []*DecodedMessage{}
var pingTimer *time.Timer = nil
var pongTimer *time.Timer = nil
go waitInitAck(ctx, init)
for looping {
cmd, ok := <-ctx.vm
if !ok {
glog.Info("vm channel closed, quit")
break
}
if cmd.code == INIT_ACK || cmd.code == INIT_ERROR {
if len(cmds) > 0 {
if cmds[0].code == INIT_DESTROYPOD {
glog.Info("got response of shutdown command, last round of command to init")
looping = false
}
if cmd.code == INIT_ACK {
if cmds[0].code != INIT_PING {
ctx.Hub <- &CommandAck{
reply: cmds[0].code,
msg: cmd.message,
}
}
} else {
ctx.Hub <- &CommandError{
context: cmds[0],
msg: cmd.message,
}
}
cmds = cmds[1:]
if pongTimer != nil {
glog.V(1).Info("ack got, clear pong timer")
pongTimer.Stop()
pongTimer = nil
}
if pingTimer == nil {
pingTimer = time.AfterFunc(30*time.Second, func() {
defer func() { recover() }()
glog.V(1).Info("Send ping message to init")
ctx.vm <- &DecodedMessage{
code: INIT_PING,
message: []byte{},
}
pingTimer = nil
})
} else {
pingTimer.Reset(30 * time.Second)
}
} else {
glog.Error("got ack but no command in queue")
}
} else if cmd.code == INIT_FINISHPOD {
num := len(cmd.message) / 4
results := make([]uint32, num)
for i := 0; i < num; i++ {
results[i] = binary.BigEndian.Uint32(cmd.message[i*4 : i*4+4])
}
for _, c := range cmds {
if c.code == INIT_DESTROYPOD {
glog.Info("got pod finish message after having send destroy message")
looping = false
ctx.Hub <- &CommandAck{
reply: c.code,
}
break
}
}
glog.V(1).Infof("Pod finished, returned %d values", num)
ctx.Hub <- &PodFinished{
result: results,
}
} else {
if glog.V(1) {
glog.Infof("send command %d to init, payload: '%s'.", cmd.code, string(cmd.message))
}
init.Write(newVmMessage(cmd))
cmds = append(cmds, cmd)
if pongTimer == nil {
glog.V(1).Info("message sent, set pong timer")
pongTimer = time.AfterFunc(30*time.Second, func() {
ctx.Hub <- &Interrupted{Reason: "init not reply ping mesg"}
})
}
}
}
if pingTimer != nil {
pingTimer.Stop()
}
if pongTimer != nil {
pongTimer.Stop()
}
}
func stateStarting(ctx *VmContext, ev QemuEvent) {
if processed := commonStateHandler(ctx, ev, true); processed {
//processed by common
} else if processed := deviceInitHandler(ctx, ev); processed {
if ctx.deviceReady() {
glog.V(1).Info("device ready, could run pod.")
ctx.startPod()
}
} else if processed := initFailureHandler(ctx, ev); processed {
ctx.shutdownVM(true, "Fail during init pod running environment")
ctx.Become(stateTerminating, "TERMINATING")
} else {
switch ev.Event() {
case EVENT_QEMU_EXIT:
glog.Info("Qemu did not start up properly, go to cleaning up")
if closed := ctx.onQemuExit(true); !closed {
ctx.Become(stateDestroying, "DESTROYING")
}
case EVENT_INIT_CONNECTED:
glog.Info("begin to wait vm commands")
ctx.reportVmRun()
case COMMAND_RELEASE:
glog.Info("pod starting, got release, please wait")
ctx.reportBusy("")
case COMMAND_ATTACH:
ctx.attachCmd(ev.(*AttachCommand))
case COMMAND_WINDOWSIZE:
cmd := ev.(*WindowSizeCommand)
if ctx.userSpec.Tty {
ctx.setWindowSize(cmd.ClientTag, cmd.Size)
}
case COMMAND_ACK:
ack := ev.(*CommandAck)
glog.V(1).Infof("[starting] got init ack to %d", ack.reply)
if ack.reply == INIT_STARTPOD {
ctx.unsetTimeout()
var pinfo []byte = []byte{}
persist, err := ctx.dump()
if err == nil {
buf, err := persist.serialize()
if err == nil {
pinfo = buf
}
}
ctx.reportSuccess("Start POD success", pinfo)
ctx.Become(stateRunning, "RUNNING")
glog.Info("pod start success ", string(ack.msg))
}
case ERROR_CMD_FAIL:
ack := ev.(*CommandError)
if ack.context.code == INIT_STARTPOD {
reason := "Start POD failed"
ctx.shutdownVM(true, reason)
ctx.Become(stateTerminating, "TERMINATING")
glog.Error(reason)
}
case EVENT_QEMU_TIMEOUT:
reason := "Start POD timeout"
ctx.shutdownVM(true, reason)
ctx.Become(stateTerminating, "TERMINATING")
glog.Error(reason)
default:
glog.Warning("got event during pod initiating")
}
}
}
func daemon(cmd string, argv []string, pipe int) error {
// create a subprocess
pid, err := fork(false)
if err != nil {
return err
} else if pid > 0 {
go func() {
wp, err := syscall.Wait4(int(pid), nil, 0, nil)
if err == nil {
glog.V(3).Infof("collect child %d", wp)
} else {
glog.Errorf("error during wait %d: %s", pid, err.Error())
}
}()
// return the parent
return nil
}
// exit the created one, create the daemon
_, err = fork(true)
if err != nil {
glog.Error("second fork failed: ", err.Error())
os.Exit(-1)
}
cur := os.Getpid()
glog.V(1).Infof("qemu daemon pid %d.", cur)
//Change the file mode mask
_ = syscall.Umask(0)
// create a new SID for the child process
s_ret, err := syscall.Setsid()
if err != nil {
glog.Info("Error: syscall.Setsid errno: ", err.Error())
os.Exit(-1)
}
if s_ret < 0 {
glog.Errorf("setsid return negative value: %d", s_ret)
os.Exit(-1)
}
os.Chdir("/")
f, e := os.OpenFile("/dev/null", os.O_RDWR, 0)
if e == nil {
fd := f.Fd()
syscall.Dup2(int(fd), int(os.Stdin.Fd()))
syscall.Dup2(int(fd), int(os.Stdout.Fd()))
syscall.Dup2(int(fd), int(os.Stderr.Fd()))
}
buf := make([]byte, 4)
binary.BigEndian.PutUint32(buf, uint32(cur))
syscall.Write(pipe, buf)
syscall.Close(pipe)
fds := listFd()
for _, fd := range fds {
if f, err := strconv.Atoi(fd); err == nil && f > 2 {
glog.V(1).Infof("close fd %d", f)
syscall.Close(f)
}
}
err = syscall.Exec(cmd, argv, []string{})
if err != nil {
glog.Error("fail to exec qemu process")
os.Exit(-1)
}
return nil
}
func qmpInitializer(ctx *VmContext) {
conn, err := unixSocketConnect(ctx.qmpSockName)
if err != nil {
glog.Error("failed to connected to ", ctx.qmpSockName, " ", err.Error())
ctx.qmp <- qmpFail(err.Error(), nil)
return
}
glog.V(1).Info("connected to ", ctx.qmpSockName)
var msg map[string]interface{}
decoder := json.NewDecoder(conn)
defer func() {
if err != nil {
conn.Close()
}
}()
glog.Info("begin qmp init...")
err = decoder.Decode(&msg)
if err != nil {
glog.Error("get qmp welcome failed: ", err.Error())
ctx.qmp <- qmpFail(err.Error(), nil)
return
}
glog.Info("got qmp welcome, now sending command qmp_capabilities")
cmd, err := json.Marshal(QmpCommand{Execute: "qmp_capabilities"})
if err != nil {
glog.Error("qmp_capabilities marshal failed ", err.Error())
ctx.qmp <- qmpFail(err.Error(), nil)
return
}
_, err = conn.Write(cmd)
if err != nil {
glog.Error("qmp_capabilities send failed ", err.Error())
ctx.qmp <- qmpFail(err.Error(), nil)
return
}
glog.Info("waiting for response")
rsp := &QmpResponse{}
err = decoder.Decode(rsp)
if err != nil {
glog.Error("response receive failed ", err.Error())
ctx.qmp <- qmpFail(err.Error(), nil)
return
}
glog.Info("got for response")
if rsp.msg.MessageType() == QMP_RESULT {
glog.Info("QMP connection initialized")
ctx.qmp <- &QmpInit{
conn: conn.(*net.UnixConn),
decoder: decoder,
}
return
}
ctx.qmp <- qmpFail("handshake failed", nil)
}
func qmpHandler(ctx *VmContext) {
go qmpInitializer(ctx)
timer := time.AfterFunc(10*time.Second, func() {
glog.Warning("Initializer Timeout.")
ctx.qmp <- &QmpTimeout{}
})
type msgHandler func(QmpInteraction)
var handler msgHandler = nil
var conn *net.UnixConn = nil
buf := []*QmpSession{}
res := make(chan QmpInteraction, 128)
loop := func(msg QmpInteraction) {
switch msg.MessageType() {
case QMP_SESSION:
glog.Info("got new session")
buf = append(buf, msg.(*QmpSession))
if len(buf) == 1 {
go qmpCommander(ctx.qmp, conn, msg.(*QmpSession), res)
}
case QMP_FINISH:
glog.Infof("session finished, buffer size %d", len(buf))
r := msg.(*QmpFinish)
if r.success {
glog.V(1).Info("success ")
if r.callback != nil {
ctx.hub <- r.callback
}
} else {
reason := "unknown"
if c, ok := r.reason["error"]; ok {
reason = c.(string)
}
glog.Error("QMP command failed ", reason)
ctx.hub <- &DeviceFailed{
session: r.callback,
}
}
buf = buf[1:]
if len(buf) > 0 {
go qmpCommander(ctx.qmp, conn, buf[0], res)
}
case QMP_RESULT, QMP_ERROR:
res <- msg
case QMP_EVENT:
ev := msg.(*QmpEvent)
ctx.hub <- ev
if ev.Type == QMP_EVENT_SHUTDOWN {
glog.Info("got QMP shutdown event, quit...")
handler = nil
}
case QMP_INTERNAL_ERROR:
res <- msg
handler = nil
glog.Info("QMP handler quit as received ", msg.(*QmpInternalError).cause)
ctx.hub <- &Interrupted{reason: msg.(*QmpInternalError).cause}
case QMP_QUIT:
handler = nil
}
}
initializing := func(msg QmpInteraction) {
switch msg.MessageType() {
case QMP_INIT:
timer.Stop()
init := msg.(*QmpInit)
conn = init.conn
handler = loop
glog.Info("QMP initialzed, go into main QMP loop")
//routine for get message
go qmpReceiver(ctx.qmp, init.decoder)
if len(buf) > 0 {
go qmpCommander(ctx.qmp, conn, buf[0], res)
}
case QMP_FINISH:
finish := msg.(*QmpFinish)
if !finish.success {
timer.Stop()
ctx.hub <- &InitFailedEvent{
reason: finish.reason["error"].(string),
}
handler = nil
glog.Error("QMP initialize failed")
}
case QMP_TIMEOUT:
ctx.hub <- &InitFailedEvent{
reason: "QMP Init timeout",
}
handler = nil
glog.Error("QMP initialize timeout")
case QMP_SESSION:
glog.Info("got new session during initializing")
buf = append(buf, msg.(*QmpSession))
}
}
handler = initializing
for handler != nil {
msg := <-ctx.qmp
handler(msg)
}
}
func (daemon *Daemon) CmdPodStart(job *engine.Job) error {
// we can only support 1024 Pods
if daemon.GetRunningPodNum() >= 1024 {
return fmt.Errorf("Pod full, the maximum Pod is 1024!")
}
podId := job.Args[0]
vmId := job.Args[1]
glog.Info("pod:%s, vm:%s", podId, vmId)
// Do the status check for the given pod
if pod, ok := daemon.podList[podId]; ok {
if pod.Status == types.S_POD_RUNNING {
return fmt.Errorf("The pod(%s) is running, can not start it", podId)
} else {
if pod.Type == "kubernetes" && pod.Status != types.S_POD_CREATED {
return fmt.Errorf("The pod(%s) is finished with kubernetes type, can not start it again", podId)
}
}
} else {
return fmt.Errorf("The pod(%s) can not be found, please create it first", podId)
}
data, err := daemon.GetPodByName(podId)
if err != nil {
return err
}
userPod, err := pod.ProcessPodBytes(data)
if err != nil {
return err
}
if vmId == "" {
vmId = fmt.Sprintf("vm-%s", pod.RandStr(10, "alpha"))
} else {
if _, ok := daemon.vmList[vmId]; !ok {
return fmt.Errorf("The VM %s doesn't exist", vmId)
}
if userPod.Resource.Vcpu != daemon.vmList[vmId].Cpu {
return fmt.Errorf("The new pod's cpu setting is different the current VM's cpu")
}
if userPod.Resource.Memory != daemon.vmList[vmId].Mem {
return fmt.Errorf("The new pod's memory setting is different the current VM's memory")
}
}
code, cause, err := daemon.StartPod(podId, vmId, "")
if err != nil {
daemon.KillVm(vmId)
glog.Error(err.Error())
return err
}
vm := &Vm{
Id: vmId,
Pod: daemon.podList[podId],
Status: types.S_VM_ASSOCIATED,
Cpu: userPod.Resource.Vcpu,
Mem: userPod.Resource.Memory,
}
daemon.podList[podId].Vm = vmId
daemon.AddVm(vm)
// Prepare the qemu status to client
v := &engine.Env{}
v.Set("ID", vmId)
v.SetInt("Code", code)
v.Set("Cause", cause)
if _, err := v.WriteTo(job.Stdout); err != nil {
return err
}
return nil
}