// InitializeMountNamespace sets up the devices, mount points, and filesystems for use inside a
// new mount namespace.
func InitializeMountNamespace(rootfs, console string, sysReadonly bool, mountConfig *MountConfig) error {
var (
err error
flag = syscall.MS_PRIVATE
)
if mountConfig.NoPivotRoot {
flag = syscall.MS_SLAVE
}
if err := syscall.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil {
return fmt.Errorf("mounting / with flags %X %s", (flag | syscall.MS_REC), err)
}
if err := syscall.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil {
return fmt.Errorf("mouting %s as bind %s", rootfs, err)
}
if err := mountSystem(rootfs, sysReadonly, mountConfig); err != nil {
return fmt.Errorf("mount system %s", err)
}
if err := setupBindmounts(rootfs, mountConfig); err != nil {
return fmt.Errorf("bind mounts %s", err)
}
if err := nodes.CreateDeviceNodes(rootfs, mountConfig.DeviceNodes); err != nil {
return fmt.Errorf("create device nodes %s", err)
}
if err := SetupPtmx(rootfs, console, mountConfig.MountLabel); err != nil {
return err
}
// stdin, stdout and stderr could be pointing to /dev/null from parent namespace.
// Re-open them inside this namespace.
if err := reOpenDevNull(rootfs); err != nil {
return fmt.Errorf("Failed to reopen /dev/null %s", err)
}
if err := setupDevSymlinks(rootfs); err != nil {
return fmt.Errorf("dev symlinks %s", err)
}
if err := syscall.Chdir(rootfs); err != nil {
return fmt.Errorf("chdir into %s %s", rootfs, err)
}
if mountConfig.NoPivotRoot {
err = MsMoveRoot(rootfs)
} else {
err = PivotRoot(rootfs)
}
if err != nil {
return err
}
if mountConfig.ReadonlyFs {
if err := SetReadonly(); err != nil {
return fmt.Errorf("set readonly %s", err)
}
}
syscall.Umask(0022)
return nil
}
// InitializeMountNamespace sets up the devices, mount points, and filesystems for use inside a
// new mount namespace.
func InitializeMountNamespace(rootfs, console string, mountConfig *MountConfig) error {
var (
err error
flag = syscall.MS_PRIVATE
)
if mountConfig.NoPivotRoot {
flag = syscall.MS_SLAVE
}
if err := system.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil {
return fmt.Errorf("mounting / with flags %X %s", (flag | syscall.MS_REC), err)
}
if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil {
return fmt.Errorf("mouting %s as bind %s", rootfs, err)
}
if err := mountSystem(rootfs, mountConfig); err != nil {
return fmt.Errorf("mount system %s", err)
}
if err := setupBindmounts(rootfs, mountConfig); err != nil {
return fmt.Errorf("bind mounts %s", err)
}
if err := nodes.CreateDeviceNodes(rootfs, mountConfig.DeviceNodes); err != nil {
return fmt.Errorf("create device nodes %s", err)
}
if err := SetupPtmx(rootfs, console, mountConfig.MountLabel); err != nil {
return err
}
if err := setupDevSymlinks(rootfs); err != nil {
return fmt.Errorf("dev symlinks %s", err)
}
if err := system.Chdir(rootfs); err != nil {
return fmt.Errorf("chdir into %s %s", rootfs, err)
}
if mountConfig.NoPivotRoot {
err = MsMoveRoot(rootfs)
} else {
err = PivotRoot(rootfs)
}
if err != nil {
return err
}
if mountConfig.ReadonlyFs {
if err := SetReadonly(); err != nil {
return fmt.Errorf("set readonly %s", err)
}
}
system.Umask(0022)
return nil
}
func (d *driver) Run(c *execdriver.Command, pipes *execdriver.Pipes, startCallback execdriver.StartCallback) (int, error) {
var (
term execdriver.Terminal
err error
)
if c.ProcessConfig.Tty {
term, err = NewTtyConsole(&c.ProcessConfig, pipes)
} else {
term, err = execdriver.NewStdConsole(&c.ProcessConfig, pipes)
}
c.ProcessConfig.Terminal = term
c.Mounts = append(c.Mounts, execdriver.Mount{
Source: d.initPath,
Destination: c.InitPath,
Writable: false,
Private: true,
})
if err := d.generateEnvConfig(c); err != nil {
return -1, err
}
configPath, err := d.generateLXCConfig(c)
if err != nil {
return -1, err
}
params := []string{
"lxc-start",
"-n", c.ID,
"-f", configPath,
"--",
c.InitPath,
}
if c.Network.Interface != nil {
params = append(params,
"-g", c.Network.Interface.Gateway,
"-i", fmt.Sprintf("%s/%d", c.Network.Interface.IPAddress, c.Network.Interface.IPPrefixLen),
)
}
params = append(params,
"-mtu", strconv.Itoa(c.Network.Mtu),
)
if c.ProcessConfig.User != "" {
params = append(params, "-u", c.ProcessConfig.User)
}
if c.ProcessConfig.Privileged {
if d.apparmor {
params[0] = path.Join(d.root, "lxc-start-unconfined")
}
params = append(params, "-privileged")
}
if c.WorkingDir != "" {
params = append(params, "-w", c.WorkingDir)
}
if len(c.CapAdd) > 0 {
params = append(params, fmt.Sprintf("-cap-add=%s", strings.Join(c.CapAdd, ":")))
}
if len(c.CapDrop) > 0 {
params = append(params, fmt.Sprintf("-cap-drop=%s", strings.Join(c.CapDrop, ":")))
}
params = append(params, "--", c.ProcessConfig.Entrypoint)
params = append(params, c.ProcessConfig.Arguments...)
if d.sharedRoot {
// lxc-start really needs / to be non-shared, or all kinds of stuff break
// when lxc-start unmount things and those unmounts propagate to the main
// mount namespace.
// What we really want is to clone into a new namespace and then
// mount / MS_REC|MS_SLAVE, but since we can't really clone or fork
// without exec in go we have to do this horrible shell hack...
shellString :=
"mount --make-rslave /; exec " +
utils.ShellQuoteArguments(params)
params = []string{
"unshare", "-m", "--", "/bin/sh", "-c", shellString,
}
}
var (
name = params[0]
arg = params[1:]
)
aname, err := exec.LookPath(name)
if err != nil {
aname = name
}
c.ProcessConfig.Path = aname
c.ProcessConfig.Args = append([]string{name}, arg...)
if err := nodes.CreateDeviceNodes(c.Rootfs, c.AutoCreatedDevices); err != nil {
return -1, err
}
if err := c.ProcessConfig.Start(); err != nil {
return -1, err
}
var (
waitErr error
waitLock = make(chan struct{})
)
go func() {
if err := c.ProcessConfig.Wait(); err != nil {
if _, ok := err.(*exec.ExitError); !ok { // Do not propagate the error if it's simply a status code != 0
waitErr = err
}
}
close(waitLock)
}()
// Poll lxc for RUNNING status
pid, err := d.waitForStart(c, waitLock)
if err != nil {
if c.ProcessConfig.Process != nil {
c.ProcessConfig.Process.Kill()
c.ProcessConfig.Wait()
}
return -1, err
}
c.ContainerPid = pid
if startCallback != nil {
startCallback(&c.ProcessConfig, pid)
}
<-waitLock
return getExitCode(c), waitErr
}
func (d *driver) Run(c *execdriver.Command, pipes *execdriver.Pipes, startCallback execdriver.StartCallback) (execdriver.ExitStatus, error) {
var (
term execdriver.Terminal
err error
dataPath = d.containerDir(c.ID)
)
if c.ProcessConfig.Tty {
term, err = NewTtyConsole(&c.ProcessConfig, pipes)
} else {
term, err = execdriver.NewStdConsole(&c.ProcessConfig, pipes)
}
c.ProcessConfig.Terminal = term
container, err := d.createContainer(c)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
d.Lock()
d.activeContainers[c.ID] = &activeContainer{
container: container,
cmd: &c.ProcessConfig.Cmd,
}
d.Unlock()
c.Mounts = append(c.Mounts, execdriver.Mount{
Source: d.initPath,
Destination: c.InitPath,
Writable: false,
Private: true,
})
if err := d.generateEnvConfig(c); err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
configPath, err := d.generateLXCConfig(c)
if err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
params := []string{
"lxc-start",
"-n", c.ID,
"-f", configPath,
}
if c.Network.ContainerID != "" {
params = append(params,
"--share-net", c.Network.ContainerID,
)
}
if c.Ipc != nil {
if c.Ipc.ContainerID != "" {
params = append(params,
"--share-ipc", c.Ipc.ContainerID,
)
} else if c.Ipc.HostIpc {
params = append(params,
"--share-ipc", "1",
)
}
}
params = append(params,
"--",
c.InitPath,
)
if c.Network.Interface != nil {
params = append(params,
"-g", c.Network.Interface.Gateway,
"-i", fmt.Sprintf("%s/%d", c.Network.Interface.IPAddress, c.Network.Interface.IPPrefixLen),
)
}
params = append(params,
"-mtu", strconv.Itoa(c.Network.Mtu),
)
if c.ProcessConfig.User != "" {
params = append(params, "-u", c.ProcessConfig.User)
}
if c.ProcessConfig.Privileged {
if d.apparmor {
params[0] = path.Join(d.root, "lxc-start-unconfined")
}
params = append(params, "-privileged")
}
if c.WorkingDir != "" {
params = append(params, "-w", c.WorkingDir)
}
params = append(params, "--", c.ProcessConfig.Entrypoint)
params = append(params, c.ProcessConfig.Arguments...)
if d.sharedRoot {
// lxc-start really needs / to be non-shared, or all kinds of stuff break
// when lxc-start unmount things and those unmounts propagate to the main
// mount namespace.
// What we really want is to clone into a new namespace and then
// mount / MS_REC|MS_SLAVE, but since we can't really clone or fork
// without exec in go we have to do this horrible shell hack...
shellString :=
"mount --make-rslave /; exec " +
utils.ShellQuoteArguments(params)
params = []string{
"unshare", "-m", "--", "/bin/sh", "-c", shellString,
}
}
log.Debugf("lxc params %s", params)
var (
name = params[0]
arg = params[1:]
)
aname, err := exec.LookPath(name)
if err != nil {
aname = name
}
c.ProcessConfig.Path = aname
c.ProcessConfig.Args = append([]string{name}, arg...)
if err := nodes.CreateDeviceNodes(c.Rootfs, c.AutoCreatedDevices); err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
if err := c.ProcessConfig.Start(); err != nil {
return execdriver.ExitStatus{ExitCode: -1}, err
}
var (
waitErr error
waitLock = make(chan struct{})
)
go func() {
if err := c.ProcessConfig.Wait(); err != nil {
if _, ok := err.(*exec.ExitError); !ok { // Do not propagate the error if it's simply a status code != 0
waitErr = err
}
}
close(waitLock)
}()
terminate := func(terr error) (execdriver.ExitStatus, error) {
if c.ProcessConfig.Process != nil {
c.ProcessConfig.Process.Kill()
c.ProcessConfig.Wait()
}
return execdriver.ExitStatus{ExitCode: -1}, terr
}
// Poll lxc for RUNNING status
pid, err := d.waitForStart(c, waitLock)
if err != nil {
return terminate(err)
}
cgroupPaths, err := cgroupPaths(c.ID)
if err != nil {
return terminate(err)
}
state := &libcontainer.State{
InitPid: pid,
CgroupPaths: cgroupPaths,
}
if err := libcontainer.SaveState(dataPath, state); err != nil {
return terminate(err)
}
c.ContainerPid = pid
if startCallback != nil {
log.Debugf("Invoking startCallback")
startCallback(&c.ProcessConfig, pid)
}
oomKill := false
oomKillNotification, err := libcontainer.NotifyOnOOM(state)
if err == nil {
_, oomKill = <-oomKillNotification
log.Debugf("oomKill error %s waitErr %s", oomKill, waitErr)
} else {
log.Warnf("WARNING: Your kernel does not support OOM notifications: %s", err)
}
<-waitLock
// check oom error
exitCode := getExitCode(c)
if oomKill {
exitCode = 137
}
return execdriver.ExitStatus{ExitCode: exitCode, OOMKilled: oomKill}, waitErr
}