// Finalize expects that the setns calls have been setup and that is has joined an
// existing namespace
func FinalizeSetns(container *libcontainer.Config, args []string) error {
// clear the current processes env and replace it with the environment defined on the container
if err := LoadContainerEnvironment(container); err != nil {
return err
}
if err := FinalizeNamespace(container); err != nil {
return err
}
if err := apparmor.ApplyProfile(container.AppArmorProfile); err != nil {
return fmt.Errorf("set apparmor profile %s: %s", container.AppArmorProfile, err)
}
if container.ProcessLabel != "" {
if err := label.SetProcessLabel(container.ProcessLabel); err != nil {
return err
}
}
if err := system.Execv(args[0], args[0:], os.Environ()); err != nil {
return err
}
panic("unreachable")
}
// ExecIn uses an existing pid and joins the pid's namespaces with the new command.
func ExecIn(container *libcontainer.Config, state *libcontainer.State, args []string) error {
// Enter the namespace and then finish setup
args, err := getNsEnterCommand(strconv.Itoa(state.InitPid), container, "", args)
if err != nil {
return err
}
finalArgs := append([]string{os.Args[0]}, args...)
if err := system.Execv(finalArgs[0], finalArgs[0:], os.Environ()); err != nil {
return err
}
panic("unreachable")
}
// ExecIn uses an existing pid and joins the pid's namespaces with the new command.
func ExecIn(container *libcontainer.Config, state *libcontainer.State, args []string) error {
// TODO(vmarmol): If this gets too long, send it over a pipe to the child.
// Marshall the container into JSON since it won't be available in the namespace.
containerJson, err := json.Marshal(container)
if err != nil {
return err
}
// Enter the namespace and then finish setup
finalArgs := []string{os.Args[0], "nsenter", "--nspid", strconv.Itoa(state.InitPid), "--containerjson", string(containerJson), "--"}
finalArgs = append(finalArgs, args...)
if err := system.Execv(finalArgs[0], finalArgs[0:], os.Environ()); err != nil {
return err
}
panic("unreachable")
}
func (l *linuxSetnsInit) Init() error {
if err := setupRlimits(l.config.Config); err != nil {
return err
}
if err := finalizeNamespace(l.config); err != nil {
return err
}
if err := apparmor.ApplyProfile(l.config.Config.AppArmorProfile); err != nil {
return err
}
if l.config.Config.ProcessLabel != "" {
if err := label.SetProcessLabel(l.config.Config.ProcessLabel); err != nil {
return err
}
}
return system.Execv(l.config.Args[0], l.config.Args[0:], os.Environ())
}
// Run a command in a container after entering the namespace.
func NsEnter(container *libcontainer.Config, args []string) error {
// clear the current processes env and replace it with the environment
// defined on the container
if err := LoadContainerEnvironment(container); err != nil {
return err
}
if err := FinalizeNamespace(container); err != nil {
return err
}
if container.ProcessLabel != "" {
if err := label.SetProcessLabel(container.ProcessLabel); err != nil {
return err
}
}
if err := system.Execv(args[0], args[0:], container.Env); err != nil {
return err
}
panic("unreachable")
}
func (l *linuxStandardInit) Init() error {
// join any namespaces via a path to the namespace fd if provided
if err := joinExistingNamespaces(l.config.Config.Namespaces); err != nil {
return err
}
var console *linuxConsole
if l.config.Console != "" {
console = newConsoleFromPath(l.config.Console)
if err := console.dupStdio(); err != nil {
return err
}
}
if _, err := syscall.Setsid(); err != nil {
return err
}
if console != nil {
if err := system.Setctty(); err != nil {
return err
}
}
if err := setupNetwork(l.config); err != nil {
return err
}
if err := setupRoute(l.config.Config); err != nil {
return err
}
if err := setupRlimits(l.config.Config); err != nil {
return err
}
label.Init()
// InitializeMountNamespace() can be executed only for a new mount namespace
if l.config.Config.Namespaces.Contains(configs.NEWNS) {
if err := setupRootfs(l.config.Config, console); err != nil {
return err
}
}
if hostname := l.config.Config.Hostname; hostname != "" {
if err := syscall.Sethostname([]byte(hostname)); err != nil {
return err
}
}
if err := apparmor.ApplyProfile(l.config.Config.AppArmorProfile); err != nil {
return err
}
if err := label.SetProcessLabel(l.config.Config.ProcessLabel); err != nil {
return err
}
for _, path := range l.config.Config.ReadonlyPaths {
if err := remountReadonly(path); err != nil {
return err
}
}
for _, path := range l.config.Config.MaskPaths {
if err := maskFile(path); err != nil {
return err
}
}
pdeath, err := system.GetParentDeathSignal()
if err != nil {
return err
}
if err := finalizeNamespace(l.config); err != nil {
return err
}
// finalizeNamespace can change user/group which clears the parent death
// signal, so we restore it here.
if err := pdeath.Restore(); err != nil {
return err
}
// Signal self if parent is already dead. Does nothing if running in a new
// PID namespace, as Getppid will always return 0.
if syscall.Getppid() == 1 {
return syscall.Kill(syscall.Getpid(), syscall.SIGKILL)
}
return system.Execv(l.config.Args[0], l.config.Args[0:], os.Environ())
}
// TODO(vishh): This is part of the libcontainer API and it does much more than just namespaces related work.
// Move this to libcontainer package.
// Init is the init process that first runs inside a new namespace to setup mounts, users, networking,
// and other options required for the new container.
// The caller of Init function has to ensure that the go runtime is locked to an OS thread
// (using runtime.LockOSThread) else system calls like setns called within Init may not work as intended.
func Init(container *libcontainer.Config, uncleanRootfs, consolePath string, syncPipe *syncpipe.SyncPipe, args []string) (err error) {
defer func() {
if err != nil {
syncPipe.ReportChildError(err)
}
}()
rootfs, err := utils.ResolveRootfs(uncleanRootfs)
if err != nil {
return err
}
// clear the current processes env and replace it with the environment
// defined on the container
if err := LoadContainerEnvironment(container); err != nil {
return err
}
// We always read this as it is a way to sync with the parent as well
var networkState *network.NetworkState
if err := syncPipe.ReadFromParent(&networkState); err != nil {
return err
}
if consolePath != "" {
if err := console.OpenAndDup(consolePath); err != nil {
return err
}
}
if _, err := syscall.Setsid(); err != nil {
return fmt.Errorf("setsid %s", err)
}
if consolePath != "" {
if err := system.Setctty(); err != nil {
return fmt.Errorf("setctty %s", err)
}
}
if err := ipc.Initialize(container.IpcNsPath); err != nil {
return fmt.Errorf("setup IPC %s", err)
}
if err := setupNetwork(container, networkState); err != nil {
return fmt.Errorf("setup networking %s", err)
}
if err := setupRoute(container); err != nil {
return fmt.Errorf("setup route %s", err)
}
label.Init()
if err := mount.InitializeMountNamespace(rootfs,
consolePath,
container.RestrictSys,
(*mount.MountConfig)(container.MountConfig)); err != nil {
return fmt.Errorf("setup mount namespace %s", err)
}
if container.Hostname != "" {
if err := syscall.Sethostname([]byte(container.Hostname)); err != nil {
return fmt.Errorf("sethostname %s", err)
}
}
if err := apparmor.ApplyProfile(container.AppArmorProfile); err != nil {
return fmt.Errorf("set apparmor profile %s: %s", container.AppArmorProfile, err)
}
if err := label.SetProcessLabel(container.ProcessLabel); err != nil {
return fmt.Errorf("set process label %s", err)
}
// TODO: (crosbymichael) make this configurable at the Config level
if container.RestrictSys {
if err := restrict.Restrict("proc/sys", "proc/sysrq-trigger", "proc/irq", "proc/bus"); err != nil {
return err
}
}
pdeathSignal, err := system.GetParentDeathSignal()
if err != nil {
return fmt.Errorf("get parent death signal %s", err)
}
if err := FinalizeNamespace(container); err != nil {
return fmt.Errorf("finalize namespace %s", err)
}
// FinalizeNamespace can change user/group which clears the parent death
// signal, so we restore it here.
if err := RestoreParentDeathSignal(pdeathSignal); err != nil {
return fmt.Errorf("restore parent death signal %s", err)
}
return system.Execv(args[0], args[0:], os.Environ())
}
// TODO(vishh): This is part of the libcontainer API and it does much more than just namespaces related work.
// Move this to libcontainer package.
// Init is the init process that first runs inside a new namespace to setup mounts, users, networking,
// and other options required for the new container.
// The caller of Init function has to ensure that the go runtime is locked to an OS thread
// (using runtime.LockOSThread) else system calls like setns called within Init may not work as intended.
func Init(container *libcontainer.Config, uncleanRootfs, consolePath string, pipe *os.File, args []string) (err error) {
defer func() {
// if we have an error during the initialization of the container's init then send it back to the
// parent process in the form of an initError.
if err != nil {
// ensure that any data sent from the parent is consumed so it doesn't
// receive ECONNRESET when the child writes to the pipe.
ioutil.ReadAll(pipe)
if err := json.NewEncoder(pipe).Encode(initError{
Message: err.Error(),
}); err != nil {
panic(err)
}
}
// ensure that this pipe is always closed
pipe.Close()
}()
rootfs, err := utils.ResolveRootfs(uncleanRootfs)
if err != nil {
return err
}
// clear the current processes env and replace it with the environment
// defined on the container
if err := LoadContainerEnvironment(container); err != nil {
return err
}
// We always read this as it is a way to sync with the parent as well
var networkState *network.NetworkState
if err := json.NewDecoder(pipe).Decode(&networkState); err != nil {
return err
}
// join any namespaces via a path to the namespace fd if provided
if err := joinExistingNamespaces(container.Namespaces); err != nil {
return err
}
if consolePath != "" {
if err := console.OpenAndDup(consolePath); err != nil {
return err
}
}
if _, err := syscall.Setsid(); err != nil {
return fmt.Errorf("setsid %s", err)
}
if consolePath != "" {
if err := system.Setctty(); err != nil {
return fmt.Errorf("setctty %s", err)
}
}
if err := setupNetwork(container, networkState); err != nil {
return fmt.Errorf("setup networking %s", err)
}
if err := setupRoute(container); err != nil {
return fmt.Errorf("setup route %s", err)
}
if err := setupRlimits(container); err != nil {
return fmt.Errorf("setup rlimits %s", err)
}
label.Init()
if err := mount.InitializeMountNamespace(rootfs,
consolePath,
container.RestrictSys,
(*mount.MountConfig)(container.MountConfig)); err != nil {
return fmt.Errorf("setup mount namespace %s", err)
}
if container.Hostname != "" {
if err := syscall.Sethostname([]byte(container.Hostname)); err != nil {
return fmt.Errorf("unable to sethostname %q: %s", container.Hostname, err)
}
}
if err := apparmor.ApplyProfile(container.AppArmorProfile); err != nil {
return fmt.Errorf("set apparmor profile %s: %s", container.AppArmorProfile, err)
}
if err := label.SetProcessLabel(container.ProcessLabel); err != nil {
return fmt.Errorf("set process label %s", err)
}
// TODO: (crosbymichael) make this configurable at the Config level
if container.RestrictSys {
if err := restrict.Restrict("proc/sys", "proc/sysrq-trigger", "proc/irq", "proc/bus"); err != nil {
return err
}
}
pdeathSignal, err := system.GetParentDeathSignal()
if err != nil {
return fmt.Errorf("get parent death signal %s", err)
}
if err := FinalizeNamespace(container); err != nil {
return fmt.Errorf("finalize namespace %s", err)
}
// FinalizeNamespace can change user/group which clears the parent death
// signal, so we restore it here.
if err := RestoreParentDeathSignal(pdeathSignal); err != nil {
return fmt.Errorf("restore parent death signal %s", err)
}
return system.Execv(args[0], args[0:], os.Environ())
}
func (l *linuxStandardInit) Init() error {
// join any namespaces via a path to the namespace fd if provided
if err := joinExistingNamespaces(l.config.Config.Namespaces); err != nil {
return err
}
var console *linuxConsole
if l.config.Console != "" {
console = newConsoleFromPath(l.config.Console)
if err := console.dupStdio(); err != nil {
return err
}
}
if _, err := syscall.Setsid(); err != nil {
return err
}
if console != nil {
if err := system.Setctty(); err != nil {
return err
}
}
if err := setupNetwork(l.config); err != nil {
return err
}
if err := setupRoute(l.config.Config); err != nil {
return err
}
if err := setupRlimits(l.config.Config); err != nil {
return err
}
label.Init()
// InitializeMountNamespace() can be executed only for a new mount namespace
if l.config.Config.Namespaces.Contains(configs.NEWNS) {
if err := setupRootfs(l.config.Config, console); err != nil {
return err
}
}
if hostname := l.config.Config.Hostname; hostname != "" {
if err := syscall.Sethostname([]byte(hostname)); err != nil {
return err
}
}
if err := apparmor.ApplyProfile(l.config.Config.AppArmorProfile); err != nil {
return err
}
if err := label.SetProcessLabel(l.config.Config.ProcessLabel); err != nil {
return err
}
for _, path := range l.config.Config.ReadonlyPaths {
if err := remountReadonly(path); err != nil {
return err
}
}
for _, path := range l.config.Config.MaskPaths {
if err := maskFile(path); err != nil {
return err
}
}
pdeath, err := system.GetParentDeathSignal()
if err != nil {
return err
}
if err := finalizeNamespace(l.config); err != nil {
return err
}
// finalizeNamespace can change user/group which clears the parent death
// signal, so we restore it here.
if err := pdeath.Restore(); err != nil {
return err
}
// compare the parent from the inital start of the init process and make sure that it did not change.
// if the parent changes that means it died and we were reparened to something else so we should
// just kill ourself and not cause problems for someone else.
if syscall.Getppid() != l.parentPid {
return syscall.Kill(syscall.Getpid(), syscall.SIGKILL)
}
return system.Execv(l.config.Args[0], l.config.Args[0:], os.Environ())
}
func initDefault(container *libcontainer.Config, uncleanRootfs, consolePath string, pipe *os.File, args []string) (err error) {
rootfs, err := utils.ResolveRootfs(uncleanRootfs)
if err != nil {
return err
}
// clear the current processes env and replace it with the environment
// defined on the container
if err := LoadContainerEnvironment(container); err != nil {
return err
}
// We always read this as it is a way to sync with the parent as well
var networkState *network.NetworkState
if err := json.NewDecoder(pipe).Decode(&networkState); err != nil {
return err
}
// join any namespaces via a path to the namespace fd if provided
if err := joinExistingNamespaces(container.Namespaces); err != nil {
return err
}
if consolePath != "" {
if err := console.OpenAndDup(consolePath); err != nil {
return err
}
}
if _, err := syscall.Setsid(); err != nil {
return fmt.Errorf("setsid %s", err)
}
if consolePath != "" {
if err := system.Setctty(); err != nil {
return fmt.Errorf("setctty %s", err)
}
}
cloneFlags := GetNamespaceFlags(container.Namespaces)
if (cloneFlags & syscall.CLONE_NEWNET) == 0 {
if len(container.Networks) != 0 || len(container.Routes) != 0 {
return fmt.Errorf("unable to apply network parameters without network namespace")
}
} else {
if err := setupNetwork(container, networkState); err != nil {
return fmt.Errorf("setup networking %s", err)
}
if err := setupRoute(container); err != nil {
return fmt.Errorf("setup route %s", err)
}
}
if err := setupRlimits(container); err != nil {
return fmt.Errorf("setup rlimits %s", err)
}
label.Init()
// InitializeMountNamespace() can be executed only for a new mount namespace
if (cloneFlags & syscall.CLONE_NEWNS) == 0 {
if container.MountConfig != nil {
return fmt.Errorf("mount config is set without mount namespace")
}
} else if err := mount.InitializeMountNamespace(rootfs,
consolePath,
container.RestrictSys,
0, // Default Root Uid
0, // Default Root Gid
(*mount.MountConfig)(container.MountConfig)); err != nil {
return fmt.Errorf("setup mount namespace %s", err)
}
if container.Hostname != "" {
if (cloneFlags & syscall.CLONE_NEWUTS) == 0 {
return fmt.Errorf("unable to set the hostname without UTS namespace")
}
if err := syscall.Sethostname([]byte(container.Hostname)); err != nil {
return fmt.Errorf("unable to sethostname %q: %s", container.Hostname, err)
}
}
if err := apparmor.ApplyProfile(container.AppArmorProfile); err != nil {
return fmt.Errorf("set apparmor profile %s: %s", container.AppArmorProfile, err)
}
if err := label.SetProcessLabel(container.ProcessLabel); err != nil {
return fmt.Errorf("set process label %s", err)
}
// TODO: (crosbymichael) make this configurable at the Config level
if container.RestrictSys {
if (cloneFlags & syscall.CLONE_NEWNS) == 0 {
return fmt.Errorf("unable to restrict access to kernel files without mount namespace")
}
if err := restrict.Restrict("proc/sys", "proc/sysrq-trigger", "proc/irq", "proc/bus"); err != nil {
return err
}
}
pdeathSignal, err := system.GetParentDeathSignal()
if err != nil {
return fmt.Errorf("get parent death signal %s", err)
}
if err := FinalizeNamespace(container); err != nil {
return fmt.Errorf("finalize namespace %s", err)
}
// FinalizeNamespace can change user/group which clears the parent death
// signal, so we restore it here.
if err := RestoreParentDeathSignal(pdeathSignal); err != nil {
return fmt.Errorf("restore parent death signal %s", err)
}
return system.Execv(args[0], args[0:], os.Environ())
}
func initUserNs(container *libcontainer.Config, uncleanRootfs, consolePath string, pipe *os.File, args []string) (err error) {
// clear the current processes env and replace it with the environment
// defined on the container
if err := LoadContainerEnvironment(container); err != nil {
return err
}
// We always read this as it is a way to sync with the parent as well
var networkState *network.NetworkState
if err := json.NewDecoder(pipe).Decode(&networkState); err != nil {
return err
}
// join any namespaces via a path to the namespace fd if provided
if err := joinExistingNamespaces(container.Namespaces); err != nil {
return err
}
if consolePath != "" {
if err := console.OpenAndDup("/dev/console"); err != nil {
return err
}
}
if _, err := syscall.Setsid(); err != nil {
return fmt.Errorf("setsid %s", err)
}
if consolePath != "" {
if err := system.Setctty(); err != nil {
return fmt.Errorf("setctty %s", err)
}
}
if container.WorkingDir == "" {
container.WorkingDir = "/"
}
if err := setupRlimits(container); err != nil {
return fmt.Errorf("setup rlimits %s", err)
}
cloneFlags := GetNamespaceFlags(container.Namespaces)
if container.Hostname != "" {
if (cloneFlags & syscall.CLONE_NEWUTS) == 0 {
return fmt.Errorf("unable to set the hostname without UTS namespace")
}
if err := syscall.Sethostname([]byte(container.Hostname)); err != nil {
return fmt.Errorf("unable to sethostname %q: %s", container.Hostname, err)
}
}
if err := apparmor.ApplyProfile(container.AppArmorProfile); err != nil {
return fmt.Errorf("set apparmor profile %s: %s", container.AppArmorProfile, err)
}
if err := label.SetProcessLabel(container.ProcessLabel); err != nil {
return fmt.Errorf("set process label %s", err)
}
if container.RestrictSys {
if (cloneFlags & syscall.CLONE_NEWNS) == 0 {
return fmt.Errorf("unable to restrict access to kernel files without mount namespace")
}
if err := restrict.Restrict("proc/sys", "proc/sysrq-trigger", "proc/irq", "proc/bus"); err != nil {
return err
}
}
pdeathSignal, err := system.GetParentDeathSignal()
if err != nil {
return fmt.Errorf("get parent death signal %s", err)
}
if err := FinalizeNamespace(container); err != nil {
return fmt.Errorf("finalize namespace %s", err)
}
// FinalizeNamespace can change user/group which clears the parent death
// signal, so we restore it here.
if err := RestoreParentDeathSignal(pdeathSignal); err != nil {
return fmt.Errorf("restore parent death signal %s", err)
}
return system.Execv(args[0], args[0:], os.Environ())
}