moby/daemon/start.go
Djordje Lukic 32d58144fd c8d: Use reference counting while mounting a snapshot
Some snapshotters (like overlayfs or zfs) can't mount the same
directories twice. For example if the same directroy is used as an upper
directory in two mounts the kernel will output this warning:

    overlayfs: upperdir is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.

And indeed accessing the files from both mounts will result in an "No
such file or directory" error.

This change introduces reference counts for the mounts, if a directory
is already mounted the mount interface will only increment the mount
counter and return the mount target effectively making sure that the
filesystem doesn't end up in an undefined behavior.

Signed-off-by: Djordje Lukic <djordje.lukic@docker.com>
2023-06-07 15:50:01 +02:00

260 lines
8.8 KiB
Go

package daemon // import "github.com/docker/docker/daemon"
import (
"context"
"runtime"
"time"
"github.com/docker/docker/api/types"
containertypes "github.com/docker/docker/api/types/container"
"github.com/docker/docker/container"
"github.com/docker/docker/errdefs"
"github.com/docker/docker/libcontainerd"
"github.com/pkg/errors"
"github.com/sirupsen/logrus"
)
// ContainerStart starts a container.
func (daemon *Daemon) ContainerStart(ctx context.Context, name string, hostConfig *containertypes.HostConfig, checkpoint string, checkpointDir string) error {
daemonCfg := daemon.config()
if checkpoint != "" && !daemonCfg.Experimental {
return errdefs.InvalidParameter(errors.New("checkpoint is only supported in experimental mode"))
}
ctr, err := daemon.GetContainer(name)
if err != nil {
return err
}
validateState := func() error {
ctr.Lock()
defer ctr.Unlock()
if ctr.Paused {
return errdefs.Conflict(errors.New("cannot start a paused container, try unpause instead"))
}
if ctr.Running {
return containerNotModifiedError{running: true}
}
if ctr.RemovalInProgress || ctr.Dead {
return errdefs.Conflict(errors.New("container is marked for removal and cannot be started"))
}
return nil
}
if err := validateState(); err != nil {
return err
}
// Windows does not have the backwards compatibility issue here.
if runtime.GOOS != "windows" {
// This is kept for backward compatibility - hostconfig should be passed when
// creating a container, not during start.
if hostConfig != nil {
logrus.Warn("DEPRECATED: Setting host configuration options when the container starts is deprecated and has been removed in Docker 1.12")
oldNetworkMode := ctr.HostConfig.NetworkMode
if err := daemon.setSecurityOptions(&daemonCfg.Config, ctr, hostConfig); err != nil {
return errdefs.InvalidParameter(err)
}
if err := daemon.mergeAndVerifyLogConfig(&hostConfig.LogConfig); err != nil {
return errdefs.InvalidParameter(err)
}
if err := daemon.setHostConfig(ctr, hostConfig); err != nil {
return errdefs.InvalidParameter(err)
}
newNetworkMode := ctr.HostConfig.NetworkMode
if string(oldNetworkMode) != string(newNetworkMode) {
// if user has change the network mode on starting, clean up the
// old networks. It is a deprecated feature and has been removed in Docker 1.12
ctr.NetworkSettings.Networks = nil
}
if err := ctr.CheckpointTo(daemon.containersReplica); err != nil {
return errdefs.System(err)
}
ctr.InitDNSHostConfig()
}
} else {
if hostConfig != nil {
return errdefs.InvalidParameter(errors.New("Supplying a hostconfig on start is not supported. It should be supplied on create"))
}
}
// check if hostConfig is in line with the current system settings.
// It may happen cgroups are umounted or the like.
if _, err = daemon.verifyContainerSettings(daemonCfg, ctr.HostConfig, nil, false); err != nil {
return errdefs.InvalidParameter(err)
}
// Adapt for old containers in case we have updates in this function and
// old containers never have chance to call the new function in create stage.
if hostConfig != nil {
if err := daemon.adaptContainerSettings(&daemonCfg.Config, ctr.HostConfig, false); err != nil {
return errdefs.InvalidParameter(err)
}
}
return daemon.containerStart(ctx, daemonCfg, ctr, checkpoint, checkpointDir, true)
}
// containerStart prepares the container to run by setting up everything the
// container needs, such as storage and networking, as well as links
// between containers. The container is left waiting for a signal to
// begin running.
func (daemon *Daemon) containerStart(ctx context.Context, daemonCfg *configStore, container *container.Container, checkpoint string, checkpointDir string, resetRestartManager bool) (retErr error) {
start := time.Now()
container.Lock()
defer container.Unlock()
if resetRestartManager && container.Running { // skip this check if already in restarting step and resetRestartManager==false
return nil
}
if container.RemovalInProgress || container.Dead {
return errdefs.Conflict(errors.New("container is marked for removal and cannot be started"))
}
if checkpointDir != "" {
// TODO(mlaventure): how would we support that?
return errdefs.Forbidden(errors.New("custom checkpointdir is not supported"))
}
// if we encounter an error during start we need to ensure that any other
// setup has been cleaned up properly
defer func() {
if retErr != nil {
container.SetError(retErr)
// if no one else has set it, make sure we don't leave it at zero
if container.ExitCode() == 0 {
container.SetExitCode(exitUnknown)
}
if err := container.CheckpointTo(daemon.containersReplica); err != nil {
logrus.Errorf("%s: failed saving state on start failure: %v", container.ID, err)
}
container.Reset(false)
daemon.Cleanup(container)
// if containers AutoRemove flag is set, remove it after clean up
if container.HostConfig.AutoRemove {
container.Unlock()
if err := daemon.containerRm(&daemonCfg.Config, container.ID, &types.ContainerRmConfig{ForceRemove: true, RemoveVolume: true}); err != nil {
logrus.Errorf("can't remove container %s: %v", container.ID, err)
}
container.Lock()
}
}
}()
if err := daemon.conditionalMountOnStart(container); err != nil {
return err
}
if err := daemon.initializeNetworking(&daemonCfg.Config, container); err != nil {
return err
}
spec, err := daemon.createSpec(ctx, daemonCfg, container)
if err != nil {
return errdefs.System(err)
}
if resetRestartManager {
container.ResetRestartManager(true)
container.HasBeenManuallyStopped = false
}
if err := daemon.saveAppArmorConfig(container); err != nil {
return err
}
if checkpoint != "" {
checkpointDir, err = getCheckpointDir(checkpointDir, checkpoint, container.Name, container.ID, container.CheckpointDir(), false)
if err != nil {
return err
}
}
shim, createOptions, err := daemon.getLibcontainerdCreateOptions(daemonCfg, container)
if err != nil {
return err
}
ctr, err := libcontainerd.ReplaceContainer(ctx, daemon.containerd, container.ID, spec, shim, createOptions)
if err != nil {
return setExitCodeFromError(container.SetExitCode, err)
}
// TODO(mlaventure): we need to specify checkpoint options here
tsk, err := ctr.Start(context.TODO(), // Passing ctx to ctr.Start caused integration tests to be stuck in the cleanup phase
checkpointDir, container.StreamConfig.Stdin() != nil || container.Config.Tty,
container.InitializeStdio)
if err != nil {
if err := ctr.Delete(context.Background()); err != nil {
logrus.WithError(err).WithField("container", container.ID).
Error("failed to delete failed start container")
}
return setExitCodeFromError(container.SetExitCode, err)
}
container.HasBeenManuallyRestarted = false
container.SetRunning(ctr, tsk, true)
container.HasBeenStartedBefore = true
daemon.setStateCounter(container)
daemon.initHealthMonitor(container)
if err := container.CheckpointTo(daemon.containersReplica); err != nil {
logrus.WithError(err).WithField("container", container.ID).
Errorf("failed to store container")
}
daemon.LogContainerEvent(container, "start")
containerActions.WithValues("start").UpdateSince(start)
return nil
}
// Cleanup releases any network resources allocated to the container along with any rules
// around how containers are linked together. It also unmounts the container's root filesystem.
func (daemon *Daemon) Cleanup(container *container.Container) {
// Microsoft HCS containers get in a bad state if host resources are
// released while the container still exists.
if ctr, ok := container.C8dContainer(); ok {
if err := ctr.Delete(context.Background()); err != nil {
logrus.Errorf("%s cleanup: failed to delete container from containerd: %v", container.ID, err)
}
}
daemon.releaseNetwork(container)
if err := container.UnmountIpcMount(); err != nil {
logrus.Warnf("%s cleanup: failed to unmount IPC: %s", container.ID, err)
}
if err := daemon.conditionalUnmountOnCleanup(container); err != nil {
// FIXME: remove once reference counting for graphdrivers has been refactored
// Ensure that all the mounts are gone
if mountid, err := daemon.imageService.GetLayerMountID(container.ID); err == nil {
daemon.cleanupMountsByID(mountid)
}
}
if err := container.UnmountSecrets(); err != nil {
logrus.Warnf("%s cleanup: failed to unmount secrets: %s", container.ID, err)
}
if err := recursiveUnmount(container.Root); err != nil {
logrus.WithError(err).WithField("container", container.ID).Warn("Error while cleaning up container resource mounts.")
}
for _, eConfig := range container.ExecCommands.Commands() {
daemon.unregisterExecCommand(container, eConfig)
}
if container.BaseFS != "" {
if err := container.UnmountVolumes(daemon.LogVolumeEvent); err != nil {
logrus.Warnf("%s cleanup: Failed to umount volumes: %v", container.ID, err)
}
}
container.CancelAttachContext()
}