Recovery console

cmd/control/cli.go

@@ -101,6 +101,13 @@ func Main() {
 			SkipFlagParsing: true,
 			Action:          preloadImagesAction,
 		},
+		{
+			Name:            "recovery-init",
+			Hidden:          true,
+			HideHelp:        true,
+			SkipFlagParsing: true,
+			Action:          recoveryInitAction,
+		},
 		{
 			Name:            "switch-console",
 			Hidden:          true,

cmd/control/console_init.go

@@ -82,7 +82,7 @@ func consoleInitFunc() error {
 		log.Error(err)
 	}
 
-	if err := writeRespawn(); err != nil {
+	if err := writeRespawn("rancher", true); err != nil {
 		log.Error(err)
 	}
 
@@ -155,7 +155,7 @@ func consoleInitFunc() error {
 	return syscall.Exec(respawnBinPath, []string{"respawn", "-f", "/etc/respawn.conf"}, os.Environ())
 }
 
-func generateRespawnConf(cmdline string) string {
+func generateRespawnConf(cmdline, user string, sshd bool) string {
 	var respawnConf bytes.Buffer
 
 	for i := 1; i < 7; i++ {
@@ -163,7 +163,7 @@ func generateRespawnConf(cmdline string) string {
 
 		respawnConf.WriteString(gettyCmd)
 		if strings.Contains(cmdline, fmt.Sprintf("rancher.autologin=%s", tty)) {
-			respawnConf.WriteString(" --autologin rancher")
+			respawnConf.WriteString(fmt.Sprintf(" --autologin %s", user))
 		}
 		respawnConf.WriteString(fmt.Sprintf(" --noclear %s linux\n", tty))
 	}
@@ -175,23 +175,25 @@ func generateRespawnConf(cmdline string) string {
 
 		respawnConf.WriteString(gettyCmd)
 		if strings.Contains(cmdline, fmt.Sprintf("rancher.autologin=%s", tty)) {
-			respawnConf.WriteString(" --autologin rancher")
+			respawnConf.WriteString(fmt.Sprintf(" --autologin %s", user))
 		}
 		respawnConf.WriteString(fmt.Sprintf(" %s\n", tty))
 	}
 
+	if sshd {
 		respawnConf.WriteString("/usr/sbin/sshd -D")
+	}
 
 	return respawnConf.String()
 }
 
-func writeRespawn() error {
+func writeRespawn(user string, sshd bool) error {
 	cmdline, err := ioutil.ReadFile("/proc/cmdline")
 	if err != nil {
 		return err
 	}
 
-	respawn := generateRespawnConf(string(cmdline))
+	respawn := generateRespawnConf(string(cmdline), user, sshd)
 
 	files, err := ioutil.ReadDir("/etc/respawn.conf.d")
 	if err == nil {

cmd/control/recovery_init.go

@@ -0,0 +1,25 @@
+package control
+
+import (
+	"os"
+	"os/exec"
+	"syscall"
+
+	log "github.com/Sirupsen/logrus"
+	"github.com/codegangsta/cli"
+)
+
+func recoveryInitAction(c *cli.Context) error {
+	if err := writeRespawn("root", false); err != nil {
+		log.Error(err)
+	}
+
+	os.Setenv("TERM", "linux")
+
+	respawnBinPath, err := exec.LookPath("respawn")
+	if err != nil {
+		return err
+	}
+
+	return syscall.Exec(respawnBinPath, []string{"respawn", "-f", "/etc/respawn.conf"}, os.Environ())
+}

config/schema.go

@@ -37,6 +37,7 @@ var schema = `{
         "no_sharedroot": {"type": "boolean"},
         "log": {"type": "boolean"},
         "force_console_rebuild": {"type": "boolean"},
+        "recovery": {"type": "boolean"},
         "disable": {"$ref": "#/definitions/list_of_strings"},
         "services_include": {"type": "object"},
         "modules": {"$ref": "#/definitions/list_of_strings"},

config/types.go

@@ -113,6 +113,7 @@ type RancherConfig struct {
 	NoSharedRoot        bool                                      `yaml:"no_sharedroot,omitempty"`
 	Log                 bool                                      `yaml:"log,omitempty"`
 	ForceConsoleRebuild bool                                      `yaml:"force_console_rebuild,omitempty"`
+	Recovery            bool                                      `yaml:"recovery,omitempty"`
 	Disable             []string                                  `yaml:"disable,omitempty"`
 	ServicesInclude     map[string]bool                           `yaml:"services_include,omitempty"`
 	Modules             []string                                  `yaml:"modules,omitempty"`

init/init.go

@@ -263,6 +263,12 @@ func RunInit() error {
 			return cfg, nil
 		}},
 		config.CfgFuncData{"load modules", loadModules},
+		config.CfgFuncData("recovery console", func(cfg *config.CloudConfig) (*config.CloudConfig, error) {
+			if cfg.Rancher.Recovery {
+				recovery(nil)
+			}
+			return cfg, nil
+		},
 		config.CfgFuncData{"b2d env", func(cfg *config.CloudConfig) (*config.CloudConfig, error) {
 			if dev := util.ResolveDevice("LABEL=B2D_STATE"); dev != "" {
 				boot2DockerEnvironment = true
@@ -411,7 +417,7 @@ func RunInit() error {
 
 	cfg, err := config.ChainCfgFuncs(nil, initFuncs)
 	if err != nil {
-		return err
+		recovery(err)
 	}
 
 	launchConfig, args := getLaunchConfig(cfg, &cfg.Rancher.SystemDocker)
@@ -422,6 +428,7 @@ func RunInit() error {
 	_, err = dfs.LaunchDocker(launchConfig, config.SystemDockerBin, args...)
 	if err != nil {
 		log.Errorf("Error Launching System Docker: %s", err)
+		recovery(err)
 		return err
 	}
 	// Code never gets here - rancher.system_docker.exec=true

init/recovery.go

@@ -0,0 +1,90 @@
+package init
+
+import (
+	log "github.com/Sirupsen/logrus"
+	composeConfig "github.com/docker/libcompose/config"
+	"github.com/docker/libcompose/yaml"
+	"github.com/rancher/os/compose"
+	"github.com/rancher/os/config"
+)
+
+var (
+	recoveryDockerService = composeConfig.ServiceConfigV1{
+		Image: config.OsBase,
+		Command: yaml.Command{
+			"ros",
+			"recovery-init",
+		},
+		Labels: map[string]string{
+			config.DetachLabel: "false",
+			config.ScopeLabel:  "system",
+		},
+		LogDriver:  "json-file",
+		Net:        "host",
+		Uts:        "host",
+		Pid:        "host",
+		Ipc:        "host",
+		Privileged: true,
+		Volumes: []string{
+			"/dev:/host/dev",
+			"/etc/ssl/certs/ca-certificates.crt:/etc/ssl/certs/ca-certificates.crt.rancher",
+			"/lib/modules:/lib/modules",
+			"/lib/firmware:/lib/firmware",
+			"/usr/bin/ros:/usr/bin/ros:ro",
+			"/usr/bin/ros:/usr/bin/cloud-init-save",
+			"/usr/bin/ros:/usr/bin/respawn:ro",
+			"/usr/share/ros:/usr/share/ros:ro",
+			"/var/lib/rancher:/var/lib/rancher",
+			"/var/lib/rancher/conf:/var/lib/rancher/conf",
+		},
+	}
+)
+
+func recoveryServices(cfg *config.CloudConfig) (*config.CloudConfig, error) {
+	_, err := compose.RunServiceSet("recovery", cfg, map[string]*composeConfig.ServiceConfigV1{
+		"recovery": &recoveryDockerService,
+	})
+	return nil, err
+}
+
+func recovery(initFailure error) {
+	if initFailure != nil {
+		log.Errorf("RancherOS has failed to boot: %v", initFailure)
+	}
+	log.Info("Launching recovery console")
+
+	var recoveryConfig config.CloudConfig
+	recoveryConfig.Rancher.Defaults = config.Defaults{
+		Network: config.NetworkConfig{
+			DNS: config.DNSConfig{
+				Nameservers: []string{
+					"8.8.8.8",
+					"8.8.4.4",
+				},
+			},
+		},
+	}
+	recoveryConfig.Rancher.BootstrapDocker = config.DockerConfig{
+		EngineOpts: config.EngineOpts{
+			Bridge:        "none",
+			StorageDriver: "overlay",
+			Restart:       &[]bool{false}[0],
+			Graph:         "/var/lib/recovery-docker",
+			Group:         "root",
+			Host:          []string{"unix:///var/run/system-docker.sock"},
+			UserlandProxy: &[]bool{false}[0],
+		},
+	}
+
+	_, err := startDocker(&recoveryConfig)
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	_, err = config.ChainCfgFuncs(&recoveryConfig,
+		loadImages,
+		recoveryServices)
+	if err != nil {
+		log.Fatal(err)
+	}
+}

scripts/schema.json

@@ -35,6 +35,7 @@
         "no_sharedroot": {"type": "boolean"},
         "log": {"type": "boolean"},
         "force_console_rebuild": {"type": "boolean"},
+        "recovery": {"type": "boolean"},
         "disable": {"$ref": "#/definitions/list_of_strings"},
         "services_include": {"type": "object"},
         "modules": {"$ref": "#/definitions/list_of_strings"},