1
0
Fork 0
mirror of https://github.com/moby/moby.git synced 2022-11-09 12:21:53 -05:00

Merge pull request #25849 from darrenstahlmsft/LibcontainerdRaces

Lock all calls to hcsshim to prevent close races
This commit is contained in:
John Howard 2016-09-20 10:34:54 -07:00 committed by GitHub
commit 4348878242
7 changed files with 95 additions and 47 deletions

View file

@ -121,13 +121,10 @@ func (daemon *Daemon) Kill(container *container.Container) error {
return nil
}
if container.IsRunning() {
container.WaitStop(2 * time.Second)
if container.IsRunning() {
if _, err2 := container.WaitStop(2 * time.Second); err2 != nil {
return err
}
}
}
// 2. Wait for the process to die, in last resort, try to kill the process directly
if err := killProcessDirectly(container); err != nil {

View file

@ -162,7 +162,9 @@ func (daemon *Daemon) AttachStreams(id string, iop libcontainerd.IOPipe) error {
if iop.Stdin != nil {
go func() {
io.Copy(iop.Stdin, stdin)
iop.Stdin.Close()
if err := iop.Stdin.Close(); err != nil {
logrus.Error(err)
}
}()
}
} else {
@ -172,7 +174,9 @@ func (daemon *Daemon) AttachStreams(id string, iop libcontainerd.IOPipe) error {
if (c != nil && !c.Config.Tty) || (ec != nil && !ec.Tty && runtime.GOOS == "windows") {
// tty is enabled, so dont close containerd's iopipe stdin.
if iop.Stdin != nil {
iop.Stdin.Close()
if err := iop.Stdin.Close(); err != nil {
logrus.Error(err)
}
}
}
}

View file

@ -46,11 +46,23 @@ func (daemon *Daemon) containerStop(container *container.Container, seconds int)
stopSignal := container.StopSignal()
// 1. Send a stop signal
if err := daemon.killPossiblyDeadProcess(container, stopSignal); err != nil {
logrus.Infof("Failed to send signal %d to the process, force killing", stopSignal)
// While normally we might "return err" here we're not going to
// because if we can't stop the container by this point then
// its probably because its already stopped. Meaning, between
// the time of the IsRunning() call above and now it stopped.
// Also, since the err return will be environment specific we can't
// look for any particular (common) error that would indicate
// that the process is already dead vs something else going wrong.
// So, instead we'll give it up to 2 more seconds to complete and if
// by that time the container is still running, then the error
// we got is probably valid and so we force kill it.
if _, err := container.WaitStop(2 * time.Second); err != nil {
logrus.Infof("Container failed to stop after sending signal %d to the process, force killing", stopSignal)
if err := daemon.killPossiblyDeadProcess(container, 9); err != nil {
return err
}
}
}
// 2. Wait for the process to exit on its own
if _, err := container.WaitStop(time.Duration(seconds) * time.Second); err != nil {

View file

@ -29,9 +29,9 @@ func (clnt *client) appendContainer(cont *container) {
clnt.containers[cont.containerID] = cont
clnt.mapMutex.Unlock()
}
func (clnt *client) deleteContainer(friendlyName string) {
func (clnt *client) deleteContainer(containerID string) {
clnt.mapMutex.Lock()
delete(clnt.containers, friendlyName)
delete(clnt.containers, containerID)
clnt.mapMutex.Unlock()
}

View file

@ -38,6 +38,8 @@ const defaultOwner = "docker"
// Create is the entrypoint to create a container from a spec, and if successfully
// created, start it too.
func (clnt *client) Create(containerID string, checkpoint string, checkpointDir string, spec Spec, options ...CreateOption) error {
clnt.lock(containerID)
defer clnt.unlock(containerID)
logrus.Debugln("libcontainerd: client.Create() with spec", spec)
configuration := &hcsshim.ContainerConfig{
@ -221,6 +223,13 @@ func (clnt *client) AddProcess(ctx context.Context, containerID, processFriendly
return err
}
pid := newProcess.Pid()
openedProcess, err := container.hcsContainer.OpenProcess(pid)
if err != nil {
logrus.Errorf("AddProcess %s OpenProcess() failed %s", containerID, err)
return err
}
stdin, stdout, stderr, err = newProcess.Stdio()
if err != nil {
logrus.Errorf("libcontainerd: %s getting std pipes failed %s", containerID, err)
@ -238,8 +247,6 @@ func (clnt *client) AddProcess(ctx context.Context, containerID, processFriendly
iopipe.Stderr = openReaderFromPipe(stderr)
}
pid := newProcess.Pid()
proc := &process{
processCommon: processCommon{
containerID: containerID,
@ -248,7 +255,7 @@ func (clnt *client) AddProcess(ctx context.Context, containerID, processFriendly
systemPid: uint32(pid),
},
commandLine: createProcessParms.CommandLine,
hcsProcess: newProcess,
hcsProcess: openedProcess,
}
// Add the process to the container's list of processes
@ -280,7 +287,7 @@ func (clnt *client) Signal(containerID string, sig int) error {
err error
)
// Get the container as we need it to find the pid of the process.
// Get the container as we need it to get the container handle.
clnt.lock(containerID)
defer clnt.unlock(containerID)
if cont, err = clnt.getContainer(containerID); err != nil {

View file

@ -35,6 +35,8 @@ func (ctr *container) newProcess(friendlyName string) *process {
}
}
// start starts a created container.
// Caller needs to lock container ID before calling this method.
func (ctr *container) start() error {
var err error
isServicing := false
@ -78,7 +80,7 @@ func (ctr *container) start() error {
createProcessParms.CommandLine = strings.Join(ctr.ociSpec.Process.Args, " ")
// Start the command running in the container.
hcsProcess, err := ctr.hcsContainer.CreateProcess(createProcessParms)
newProcess, err := ctr.hcsContainer.CreateProcess(createProcessParms)
if err != nil {
logrus.Errorf("libcontainerd: CreateProcess() failed %s", err)
if err := ctr.terminate(); err != nil {
@ -90,10 +92,21 @@ func (ctr *container) start() error {
}
ctr.startedAt = time.Now()
pid := newProcess.Pid()
openedProcess, err := ctr.hcsContainer.OpenProcess(pid)
if err != nil {
logrus.Errorf("OpenProcess() failed %s", err)
if err := ctr.terminate(); err != nil {
logrus.Errorf("Failed to cleanup after a failed OpenProcess. %s", err)
} else {
logrus.Debugln("Cleaned up after failed OpenProcess by calling Terminate")
}
return err
}
// Save the hcs Process and PID
ctr.process.friendlyName = InitFriendlyName
pid := hcsProcess.Pid()
ctr.process.hcsProcess = hcsProcess
ctr.process.hcsProcess = openedProcess
// If this is a servicing container, wait on the process synchronously here and
// if it succeeds, wait for it cleanly shutdown and merge into the parent container.
@ -110,7 +123,7 @@ func (ctr *container) start() error {
var stdout, stderr io.ReadCloser
var stdin io.WriteCloser
stdin, stdout, stderr, err = hcsProcess.Stdio()
stdin, stdout, stderr, err = newProcess.Stdio()
if err != nil {
logrus.Errorf("libcontainerd: failed to get stdio pipes: %s", err)
if err := ctr.terminate(); err != nil {
@ -121,7 +134,7 @@ func (ctr *container) start() error {
iopipe := &IOPipe{Terminal: ctr.ociSpec.Process.Terminal}
iopipe.Stdin = createStdInCloser(stdin, hcsProcess)
iopipe.Stdin = createStdInCloser(stdin, newProcess)
// Convert io.ReadClosers to io.Readers
if stdout != nil {
@ -151,6 +164,7 @@ func (ctr *container) start() error {
State: StateStart,
Pid: ctr.systemPid, // Not sure this is needed? Double-check monitor.go in daemon BUGBUG @jhowardmsft
}}
logrus.Debugf("libcontainerd: start() completed OK, %+v", si)
return ctr.client.backend.StateChanged(ctr.containerID, si)
}
@ -182,10 +196,6 @@ func (ctr *container) waitProcessExitCode(process *process) int {
// has exited to avoid a container being dropped on the floor.
}
if err := process.hcsProcess.Close(); err != nil {
logrus.Errorf("libcontainerd: hcsProcess.Close(): %v", err)
}
return exitCode
}
@ -197,6 +207,8 @@ func (ctr *container) waitExit(process *process, isFirstProcessToStart bool) err
logrus.Debugln("libcontainerd: waitExit() on pid", process.systemPid)
exitCode := ctr.waitProcessExitCode(process)
// Lock the container while shutting down
ctr.client.lock(ctr.containerID)
// Assume the container has exited
si := StateInfo{
@ -212,6 +224,7 @@ func (ctr *container) waitExit(process *process, isFirstProcessToStart bool) err
// But it could have been an exec'd process which exited
if !isFirstProcessToStart {
si.State = StateExitProcess
ctr.cleanProcess(process.friendlyName)
} else {
updatePending, err := ctr.hcsContainer.HasPendingUpdates()
if err != nil {
@ -237,6 +250,7 @@ func (ctr *container) waitExit(process *process, isFirstProcessToStart bool) err
} else if restart {
si.State = StateRestart
ctr.restarting = true
ctr.client.deleteContainer(ctr.containerID)
waitRestart = wait
}
}
@ -244,10 +258,17 @@ func (ctr *container) waitExit(process *process, isFirstProcessToStart bool) err
// Remove process from list if we have exited
// We need to do so here in case the Message Handler decides to restart it.
if si.State == StateExit {
ctr.client.deleteContainer(ctr.friendlyName)
ctr.client.deleteContainer(ctr.containerID)
}
}
if err := process.hcsProcess.Close(); err != nil {
logrus.Errorf("libcontainerd: hcsProcess.Close(): %v", err)
}
// Unlock here before we call back into the daemon to update state
ctr.client.unlock(ctr.containerID)
// Call into the backend to notify it of the state change.
logrus.Debugf("libcontainerd: waitExit() calling backend.StateChanged %+v", si)
if err := ctr.client.backend.StateChanged(ctr.containerID, si); err != nil {
@ -257,7 +278,6 @@ func (ctr *container) waitExit(process *process, isFirstProcessToStart bool) err
go func() {
err := <-waitRestart
ctr.restarting = false
ctr.client.deleteContainer(ctr.friendlyName)
if err == nil {
if err = ctr.client.Create(ctr.containerID, "", "", ctr.ociSpec, ctr.options...); err != nil {
logrus.Errorf("libcontainerd: error restarting %v", err)
@ -276,6 +296,14 @@ func (ctr *container) waitExit(process *process, isFirstProcessToStart bool) err
return nil
}
// cleanProcess removes process from the map.
// Caller needs to lock container ID before calling this method.
func (ctr *container) cleanProcess(id string) {
delete(ctr.processes, id)
}
// shutdown shuts down the container in HCS
// Caller needs to lock container ID before calling this method.
func (ctr *container) shutdown() error {
const shutdownTimeout = time.Minute * 5
err := ctr.hcsContainer.Shutdown()
@ -297,6 +325,8 @@ func (ctr *container) shutdown() error {
return nil
}
// terminate terminates the container in HCS
// Caller needs to lock container ID before calling this method.
func (ctr *container) terminate() error {
const terminateTimeout = time.Minute * 5
err := ctr.hcsContainer.Terminate()

View file

@ -4,6 +4,7 @@ import (
"io"
"github.com/Microsoft/hcsshim"
"github.com/docker/docker/pkg/ioutils"
)
// process keeps the state for both main container process and exec process.
@ -29,26 +30,23 @@ func openReaderFromPipe(p io.ReadCloser) io.Reader {
return r
}
type stdInCloser struct {
io.WriteCloser
hcsshim.Process
}
func createStdInCloser(pipe io.WriteCloser, process hcsshim.Process) *stdInCloser {
return &stdInCloser{
WriteCloser: pipe,
Process: process,
}
}
func (stdin *stdInCloser) Close() error {
if err := stdin.WriteCloser.Close(); err != nil {
func createStdInCloser(pipe io.WriteCloser, process hcsshim.Process) io.WriteCloser {
return ioutils.NewWriteCloserWrapper(pipe, func() error {
if err := pipe.Close(); err != nil {
return err
}
return stdin.Process.CloseStdin()
}
// We do not need to lock container ID here, even though
// we are calling into hcsshim. This is safe, because the
// only place that closes this process handle is this method.
err := process.CloseStdin()
if err != nil && !hcsshim.IsNotExist(err) {
// This error will occur if the compute system is currently shutting down
if perr, ok := err.(*hcsshim.ProcessError); ok && perr.Err != hcsshim.ErrVmcomputeOperationInvalidState {
return err
}
}
func (stdin *stdInCloser) Write(p []byte) (n int, err error) {
return stdin.WriteCloser.Write(p)
return process.Close()
})
}