rancherOS/cmd/power/power.go

package power

import (
	"errors"
	"fmt"
	"os"
	"path/filepath"
	"strconv"
	"strings"
	"syscall"
	"time"

	"golang.org/x/net/context"

	"github.com/docker/engine-api/types"
	"github.com/docker/engine-api/types/container"
	"github.com/docker/engine-api/types/filters"
	"github.com/rancher/os/cmd/control/install"
	"github.com/rancher/os/config"
	"github.com/rancher/os/log"

	"github.com/rancher/os/docker"
	"github.com/rancher/os/util"
)

// You can't shutdown the system from a process in console because we want to stop the console container.
// If you do that you kill yourself.  So we spawn a separate container to do power operations
// This can up because on shutdown we want ssh to gracefully die, terminating ssh connections and not just hanging tcp session
//
// Be careful of container name. only [a-zA-Z0-9][a-zA-Z0-9_.-] are allowed
func runDocker(name string) error {
	if os.ExpandEnv("${IN_DOCKER}") == "true" {
		return nil
	}

	client, err := docker.NewSystemClient()
	if err != nil {
		return err
	}

	cmd := os.Args
	log.Debugf("runDocker cmd: %s", cmd)

	if name == "" {
		name = filepath.Base(os.Args[0])
	}

	containerName := strings.TrimPrefix(strings.Join(strings.Split(name, "/"), "-"), "-")

	existing, err := client.ContainerInspect(context.Background(), containerName)
	if err == nil && existing.ID != "" {
		// remove the old version of reboot
		err := client.ContainerRemove(context.Background(), types.ContainerRemoveOptions{
			ContainerID: existing.ID,
		})

		if err != nil {
			return err
		}
	}

	currentContainerID, err := util.GetCurrentContainerID()
	if err != nil {
		return err
	}

	currentContainer, err := client.ContainerInspect(context.Background(), currentContainerID)
	if err != nil {
		return err
	}

	powerContainer, err := client.ContainerCreate(context.Background(),
		&container.Config{
			Image: currentContainer.Config.Image,
			Cmd:   cmd,
			Env: []string{
				"IN_DOCKER=true",
			},
		},
		&container.HostConfig{
			PidMode:     "host",
			NetworkMode: "none",
			VolumesFrom: []string{
				currentContainer.ID,
			},
			Privileged: true,
		}, nil, containerName)
	if err != nil {
		return err
	}

	err = client.ContainerStart(context.Background(), powerContainer.ID)
	if err != nil {
		return err
	}

	reader, err := client.ContainerLogs(context.Background(), types.ContainerLogsOptions{
		ContainerID: powerContainer.ID,
		ShowStderr:  true,
		ShowStdout:  true,
		Follow:      true,
	})
	if err != nil {
		log.Fatal(err)
	}

	for {
		p := make([]byte, 4096)
		n, err := reader.Read(p)
		if err != nil {
			log.Error(err)
			if n == 0 {
				reader.Close()
				break
			}
		}
		if n > 0 {
			fmt.Print(string(p))
		}
	}

	if err != nil {
		log.Fatal(err)
	}

	os.Exit(0)
	return nil
}

func reboot(name string, force bool, code uint) {
	if os.Geteuid() != 0 {
		log.Fatalf("%s: Need to be root", os.Args[0])
	}

	// Add shutdown timeout
	cfg := config.LoadConfig()
	timeoutValue := cfg.Rancher.ShutdownTimeout
	if timeoutValue == 0 {
		timeoutValue = 60
	}
	if timeoutValue < 5 {
		timeoutValue = 5
	}
	log.Infof("Setting %s timeout to %d (rancher.shutdown_timeout set to %d)", os.Args[0], timeoutValue, cfg.Rancher.ShutdownTimeout)

	go func() {
		timeout := time.After(time.Duration(timeoutValue) * time.Second)
		tick := time.Tick(100 * time.Millisecond)
		// Keep trying until we're timed out or got a result or got an error
		for {
			select {
			// Got a timeout! fail with a timeout error
			case <-timeout:
				log.Errorf("Container shutdown taking too long, forcing %s.", os.Args[0])
				syscall.Sync()
				syscall.Reboot(int(code))
			case <-tick:
				fmt.Printf(".")
			}
		}
	}()

	// reboot -f should work even when system-docker is having problems
	if !force {
		if kexecFlag || previouskexecFlag || kexecAppendFlag != "" {
			// pass through the cmdline args
			name = ""
		}
		if err := runDocker(name); err != nil {
			log.Fatal(err)
		}
	}

	if kexecFlag || previouskexecFlag || kexecAppendFlag != "" {
		// need to mount boot dir, or `system-docker run -v /:/host -w /host/boot` ?
		baseName := "/mnt/new_img"
		_, _, err := install.MountDevice(baseName, "", "", false)
		if err != nil {
			log.Errorf("ERROR: can't Kexec: %s", err)
			return
		}
		defer util.Unmount(baseName)
		Kexec(previouskexecFlag, filepath.Join(baseName, install.BootDir), kexecAppendFlag)
		return
	}

	if !force {
		err := shutDownContainers()
		if err != nil {
			log.Error(err)
		}
	}

	syscall.Sync()
	err := syscall.Reboot(int(code))
	if err != nil {
		log.Fatal(err)
	}
}

func shutDownContainers() error {
	var err error
	shutDown := true
	timeout := 2
	for i, arg := range os.Args {
		if arg == "-f" || arg == "--f" || arg == "--force" {
			shutDown = false
		}
		if arg == "-t" || arg == "--t" || arg == "--timeout" {
			if len(os.Args) > i+1 {
				t, err := strconv.Atoi(os.Args[i+1])
				if err != nil {
					return err
				}
				timeout = t
			} else {
				log.Error("please specify a timeout")
			}
		}
	}
	if !shutDown {
		return nil
	}
	client, err := docker.NewSystemClient()

	if err != nil {
		return err
	}

	filter := filters.NewArgs()
	filter.Add("status", "running")

	opts := types.ContainerListOptions{
		All:    true,
		Filter: filter,
	}

	containers, err := client.ContainerList(context.Background(), opts)
	if err != nil {
		return err
	}

	currentContainerID, err := util.GetCurrentContainerID()
	if err != nil {
		return err
	}

	var stopErrorStrings []string
	consoleContainerIdx := -1

	for idx, container := range containers {
		if container.ID == currentContainerID {
			continue
		}
		if container.Names[0] == "/console" {
			consoleContainerIdx = idx
			continue
		}

		log.Infof("Stopping %s : %s", container.Names[0], container.ID[:12])
		stopErr := client.ContainerStop(context.Background(), container.ID, timeout)
		if stopErr != nil {
			log.Errorf("------- Error Stopping %s : %s", container.Names[0], stopErr.Error())
			stopErrorStrings = append(stopErrorStrings, " ["+container.ID+"] "+stopErr.Error())
		}
	}

	// lets see what containers are still running and only wait on those
	containers, err = client.ContainerList(context.Background(), opts)
	if err != nil {
		return err
	}

	var waitErrorStrings []string

	for idx, container := range containers {
		if container.ID == currentContainerID {
			continue
		}
		if container.Names[0] == "/console" {
			consoleContainerIdx = idx
			continue
		}
		log.Infof("Waiting %s : %s", container.Names[0], container.ID[:12])
		_, waitErr := client.ContainerWait(context.Background(), container.ID)
		if waitErr != nil {
			log.Errorf("------- Error Waiting %s : %s", container.Names[0], waitErr.Error())
			waitErrorStrings = append(waitErrorStrings, " ["+container.ID+"] "+waitErr.Error())
		}
	}

	// and now stop the console
	if consoleContainerIdx != -1 {
		container := containers[consoleContainerIdx]
		log.Infof("Console Stopping %v : %s", container.Names, container.ID[:12])
		stopErr := client.ContainerStop(context.Background(), container.ID, timeout)
		if stopErr != nil {
			log.Errorf("------- Error Stopping %v : %s", container.Names, stopErr.Error())
			stopErrorStrings = append(stopErrorStrings, " ["+container.ID+"] "+stopErr.Error())
		}

		log.Infof("Console Waiting %v : %s", container.Names, container.ID[:12])
		_, waitErr := client.ContainerWait(context.Background(), container.ID)
		if waitErr != nil {
			log.Errorf("------- Error Waiting %v : %s", container.Names, waitErr.Error())
			waitErrorStrings = append(waitErrorStrings, " ["+container.ID+"] "+waitErr.Error())
		}
	}

	if len(waitErrorStrings) != 0 || len(stopErrorStrings) != 0 {
		return errors.New("error while stopping \n1. STOP Errors [" + strings.Join(stopErrorStrings, ",") + "] \n2. WAIT Errors [" + strings.Join(waitErrorStrings, ",") + "]")
	}

	return nil
}