lxc: Work around lxc-start need for private mounts

lxc-start requires / to be mounted private, otherwise the changes
it does inside the container (both mounts and unmounts) will propagate
out to the host.

We work around this by starting up lxc-start in its own namespace where
we set / to rshared.

Unfortunately go can't really execute any code between clone and exec,
so we can't do this in a nice way. Instead we have a horrible hack that
use the unshare command, the shell and the mount command...

container.go

@@ -863,7 +863,13 @@ func (container *Container) Start() (err error) {
 		return err
 	}
 
+	var lxcStart string = "lxc-start"
+	if container.hostConfig.Privileged && container.runtime.capabilities.AppArmor {
+		lxcStart = path.Join(container.runtime.config.Root, "lxc-start-unconfined")
+	}
+
 	params := []string{
+		lxcStart,
 		"-n", container.ID,
 		"-f", container.lxcConfigPath(),
 		"--",
@@ -956,11 +962,24 @@ func (container *Container) Start() (err error) {
 	params = append(params, "--", container.Path)
 	params = append(params, container.Args...)
 
-	var lxcStart string = "lxc-start"
-	if container.hostConfig.Privileged && container.runtime.capabilities.AppArmor {
-		lxcStart = path.Join(container.runtime.config.Root, "lxc-start-unconfined")
+	if RootIsShared() {
+		// 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,
+		}
 	}
-	container.cmd = exec.Command(lxcStart, params...)
+
+	container.cmd = exec.Command(params[0], params[1:]...)
+
 	// Setup logging of stdout and stderr to disk
 	if err := container.runtime.LogToDisk(container.stdout, container.logPath("json"), "stdout"); err != nil {
 		return err

utils.go

@@ -4,6 +4,7 @@ import (
 	"fmt"
 	"github.com/dotcloud/docker/namesgenerator"
 	"github.com/dotcloud/docker/utils"
+	"io/ioutil"
 	"strconv"
 	"strings"
 )
@@ -301,6 +302,20 @@ func parseLink(rawLink string) (map[string]string, error) {
 	return utils.PartParser("name:alias", rawLink)
 }
 
+func RootIsShared() bool {
+	if data, err := ioutil.ReadFile("/proc/self/mountinfo"); err == nil {
+		for _, line := range strings.Split(string(data), "\n") {
+			cols := strings.Split(line, " ")
+			if len(cols) >= 6 && cols[4] == "/" {
+				return strings.HasPrefix(cols[6], "shared")
+			}
+		}
+	}
+
+	// No idea, probably safe to assume so
+	return true
+}
+
 type checker struct {
 	runtime *Runtime
 }