Merge pull request #304 from mrunalp/feature/user_namespaces

Adds user namespace support to libcontainer

config.go

@@ -124,6 +124,11 @@ type Config struct {
 	// AdditionalGroups specifies the gids that should be added to supplementary groups
 	// in addition to those that the user belongs to.
 	AdditionalGroups []int `json:"additional_groups,omitempty"`
+	// UidMappings is an array of User ID mappings for User Namespaces
+	UidMappings []IDMap `json:"uid_mappings,omitempty"`
+
+	// GidMappings is an array of Group ID mappings for User Namespaces
+	GidMappings []IDMap `json:"gid_mappings,omitempty"`
 }
 
 // Routes can be specified to create entries in the route table as the container is started
@@ -152,3 +157,10 @@ type Rlimit struct {
 	Hard uint64 `json:"hard,omitempty"`
 	Soft uint64 `json:"soft,omitempty"`
 }
+
+// IDMap represents UID/GID Mappings for User Namespaces.
+type IDMap struct {
+	ContainerID int `json:"container_id,omitempty"`
+	HostID      int `json:"host_id,omitempty"`
+	Size        int `json:"size,omitempty"`
+}

console/console.go

@@ -13,15 +13,15 @@ import (
 )
 
 // Setup initializes the proper /dev/console inside the rootfs path
-func Setup(rootfs, consolePath, mountLabel string) error {
+func Setup(rootfs, consolePath, mountLabel string, hostRootUid, hostRootGid int) error {
 	oldMask := syscall.Umask(0000)
 	defer syscall.Umask(oldMask)
 
 	if err := os.Chmod(consolePath, 0600); err != nil {
 		return err
 	}
 
-	if err := os.Chown(consolePath, 0, 0); err != nil {
+	if err := os.Chown(consolePath, hostRootUid, hostRootGid); err != nil {
 		return err
 	}
 

integration/execin_test.go

@@ -115,6 +115,7 @@ func TestExecInRlimit(t *testing.T) {
 func startLongRunningContainer(config *libcontainer.Config) (*exec.Cmd, string, chan error) {
 	containerErr := make(chan error, 1)
 	containerCmd := &exec.Cmd{}
+	setupContainerCmd := &exec.Cmd{}
 	var statePath string
 
 	createCmd := func(container *libcontainer.Config, console, dataPath, init string,
@@ -124,14 +125,20 @@ func startLongRunningContainer(config *libcontainer.Config) (*exec.Cmd, string,
 		return containerCmd
 	}
 
+	setupCmd := func(container *libcontainer.Config, console, dataPath, init string) *exec.Cmd {
+		setupContainerCmd = namespaces.DefaultSetupCommand(container, console, dataPath, init)
+		statePath = dataPath
+		return setupContainerCmd
+	}
+
 	var containerStart sync.WaitGroup
 	containerStart.Add(1)
 	go func() {
 		buffers := newStdBuffers()
 		_, err := namespaces.Exec(config,
 			buffers.Stdin, buffers.Stdout, buffers.Stderr,
 			"", config.RootFs, []string{"sleep", "10"},
-			createCmd, containerStart.Done)
+			createCmd, setupCmd, containerStart.Done)
 		containerErr <- err
 	}()
 	containerStart.Wait()

integration/utils_test.go

@@ -90,6 +90,6 @@ func runContainer(config *libcontainer.Config, console string, args ...string) (
 
 	buffers = newStdBuffers()
 	exitCode, err = namespaces.Exec(config, buffers.Stdin, buffers.Stdout, buffers.Stderr,
-		console, config.RootFs, args, namespaces.DefaultCreateCommand, nil)
+		console, config.RootFs, args, namespaces.DefaultCreateCommand, namespaces.DefaultSetupCommand, nil)
 	return
 }

mount/init.go

@@ -25,7 +25,7 @@ type mount struct {
 
 // 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 {
+func InitializeMountNamespace(rootfs, console string, sysReadonly bool, hostRootUid, hostRootGid int, mountConfig *MountConfig) error {
 	var (
 		err  error
 		flag = syscall.MS_PRIVATE
@@ -58,14 +58,17 @@ func InitializeMountNamespace(rootfs, console string, sysReadonly bool, mountCon
 		return fmt.Errorf("create device nodes %s", err)
 	}
 
-	if err := SetupPtmx(rootfs, console, mountConfig.MountLabel); err != nil {
+	if err := SetupPtmx(rootfs, console, mountConfig.MountLabel, hostRootUid, hostRootGid); 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)
+	// FIXME: Need to fix this for user namespaces.
+	if hostRootUid == 0 {
+		if err := reOpenDevNull(rootfs); err != nil {
+			return fmt.Errorf("Failed to reopen /dev/null %s", err)
+		}
 	}
 
 	if err := setupDevSymlinks(rootfs); err != nil {

mount/ptmx.go

@@ -10,7 +10,7 @@ import (
 	"github.com/docker/libcontainer/console"
 )
 
-func SetupPtmx(rootfs, consolePath, mountLabel string) error {
+func SetupPtmx(rootfs, consolePath, mountLabel string, hostRootUid, hostRootGid int) error {
 	ptmx := filepath.Join(rootfs, "dev/ptmx")
 	if err := os.Remove(ptmx); err != nil && !os.IsNotExist(err) {
 		return err
@@ -21,7 +21,7 @@ func SetupPtmx(rootfs, consolePath, mountLabel string) error {
 	}
 
 	if consolePath != "" {
-		if err := console.Setup(rootfs, consolePath, mountLabel); err != nil {
+		if err := console.Setup(rootfs, consolePath, mountLabel, hostRootUid, hostRootGid); err != nil {
 			return err
 		}
 	}

namespaces/create.go

@@ -8,3 +8,4 @@ import (
 )
 
 type CreateCommand func(container *libcontainer.Config, console, dataPath, init string, childPipe *os.File, args []string) *exec.Cmd
+type SetupCommand func(container *libcontainer.Config, console, dataPath, init string) *exec.Cmd

namespaces/exec.go

@@ -4,6 +4,7 @@ package namespaces
 
 import (
 	"encoding/json"
+	"fmt"
 	"io"
 	"os"
 	"os/exec"
@@ -25,7 +26,7 @@ const (
 // Move this to libcontainer package.
 // Exec performs setup outside of a namespace so that a container can be
 // executed.  Exec is a high level function for working with container namespaces.
-func Exec(container *libcontainer.Config, stdin io.Reader, stdout, stderr io.Writer, console, dataPath string, args []string, createCommand CreateCommand, startCallback func()) (int, error) {
+func Exec(container *libcontainer.Config, stdin io.Reader, stdout, stderr io.Writer, console, dataPath string, args []string, createCommand CreateCommand, setupCommand SetupCommand, startCallback func()) (int, error) {
 	var err error
 
 	// create a pipe so that we can syncronize with the namespaced process and
@@ -74,14 +75,6 @@ func Exec(container *libcontainer.Config, stdin io.Reader, stdout, stderr io.Wri
 	if err := InitializeNetworking(container, command.Process.Pid, &networkState); err != nil {
 		return terminate(err)
 	}
-	// send the state to the container's init process then shutdown writes for the parent
-	if err := json.NewEncoder(parent).Encode(networkState); err != nil {
-		return terminate(err)
-	}
-	// shutdown writes for the parent side of the pipe
-	if err := syscall.Shutdown(int(parent.Fd()), syscall.SHUT_WR); err != nil {
-		return terminate(err)
-	}
 
 	state := &libcontainer.State{
 		InitPid:       command.Process.Pid,
@@ -95,6 +88,26 @@ func Exec(container *libcontainer.Config, stdin io.Reader, stdout, stderr io.Wri
 	}
 	defer libcontainer.DeleteState(dataPath)
 
+	// Start the setup process to setup the init process
+	if container.Namespaces.Contains(libcontainer.NEWUSER) {
+		setupCmd := setupCommand(container, console, dataPath, os.Args[0])
+		output, err := setupCmd.CombinedOutput()
+		if err != nil || setupCmd.ProcessState.Sys().(syscall.WaitStatus).ExitStatus() != 0 {
+			command.Process.Kill()
+			command.Wait()
+			return -1, fmt.Errorf("setup failed: %s %s", err, output)
+		}
+	}
+
+	// send the state to the container's init process then shutdown writes for the parent
+	if err := json.NewEncoder(parent).Encode(networkState); err != nil {
+		return terminate(err)
+	}
+	// shutdown writes for the parent side of the pipe
+	if err := syscall.Shutdown(int(parent.Fd()), syscall.SHUT_WR); err != nil {
+		return terminate(err)
+	}
+
 	// wait for the child process to fully complete and receive an error message
 	// if one was encoutered
 	var ierr *initError
@@ -157,6 +170,75 @@ func killAllPids(container *libcontainer.Config) error {
 	return err
 }
 
+// Utility function that gets a host ID for a container ID from user namespace map
+// if that ID is present in the map.
+func hostIDFromMapping(containerID int, uMap []libcontainer.IDMap) (int, bool) {
+	for _, m := range uMap {
+		if (containerID >= m.ContainerID) && (containerID <= (m.ContainerID + m.Size - 1)) {
+			hostID := m.HostID + (containerID - m.ContainerID)
+			return hostID, true
+		}
+	}
+	return -1, false
+}
+
+// Gets the root uid for the process on host which could be non-zero
+// when user namespaces are enabled.
+func GetHostRootGid(container *libcontainer.Config) (int, error) {
+	if container.Namespaces.Contains(libcontainer.NEWUSER) {
+		if container.GidMappings == nil {
+			return -1, fmt.Errorf("User namespaces enabled, but no gid mappings found.")
+		}
+		hostRootGid, found := hostIDFromMapping(0, container.GidMappings)
+		if !found {
+			return -1, fmt.Errorf("User namespaces enabled, but no root user mapping found.")
+		}
+		return hostRootGid, nil
+	}
+
+	// Return default root uid 0
+	return 0, nil
+}
+
+// Gets the root uid for the process on host which could be non-zero
+// when user namespaces are enabled.
+func GetHostRootUid(container *libcontainer.Config) (int, error) {
+	if container.Namespaces.Contains(libcontainer.NEWUSER) {
+		if container.UidMappings == nil {
+			return -1, fmt.Errorf("User namespaces enabled, but no user mappings found.")
+		}
+		hostRootUid, found := hostIDFromMapping(0, container.UidMappings)
+		if !found {
+			return -1, fmt.Errorf("User namespaces enabled, but no root user mapping found.")
+		}
+		return hostRootUid, nil
+	}
+
+	// Return default root uid 0
+	return 0, nil
+}
+
+// Converts IDMap to SysProcIDMap array and adds it to SysProcAttr.
+func AddUidGidMappings(sys *syscall.SysProcAttr, container *libcontainer.Config) {
+	if container.UidMappings != nil {
+		sys.UidMappings = make([]syscall.SysProcIDMap, len(container.UidMappings))
+		for i, um := range container.UidMappings {
+			sys.UidMappings[i].ContainerID = um.ContainerID
+			sys.UidMappings[i].HostID = um.HostID
+			sys.UidMappings[i].Size = um.Size
+		}
+	}
+
+	if container.GidMappings != nil {
+		sys.GidMappings = make([]syscall.SysProcIDMap, len(container.GidMappings))
+		for i, gm := range container.GidMappings {
+			sys.GidMappings[i].ContainerID = gm.ContainerID
+			sys.GidMappings[i].HostID = gm.HostID
+			sys.GidMappings[i].Size = gm.Size
+		}
+	}
+}
+
 // DefaultCreateCommand will return an exec.Cmd with the Cloneflags set to the proper namespaces
 // defined on the container's configuration and use the current binary as the init with the
 // args provided
@@ -187,6 +269,46 @@ func DefaultCreateCommand(container *libcontainer.Config, console, dataPath, ini
 	command.SysProcAttr.Pdeathsig = syscall.SIGKILL
 	command.ExtraFiles = []*os.File{pipe}
 
+	if container.Namespaces.Contains(libcontainer.NEWUSER) {
+		AddUidGidMappings(command.SysProcAttr, container)
+
+		// Default to root user when user namespaces are enabled.
+		if command.SysProcAttr.Credential == nil {
+			command.SysProcAttr.Credential = &syscall.Credential{}
+		}
+	}
+
+	return command
+}
+
+// DefaultSetupCommand will return an exec.Cmd that joins the init process to set it up.
+//
+// console: the /dev/console to setup inside the container
+// init: the program executed inside the namespaces
+// root: the path to the container json file and information
+// args: the arguments to pass to the container to run as the user's program
+func DefaultSetupCommand(container *libcontainer.Config, console, dataPath, init string) *exec.Cmd {
+	// get our binary name from arg0 so we can always reexec ourself
+	env := []string{
+		"console=" + console,
+		"data_path=" + dataPath,
+	}
+
+	if dataPath == "" {
+		dataPath, _ = os.Getwd()
+	}
+
+	if container.RootFs == "" {
+		container.RootFs, _ = os.Getwd()
+	}
+	args := []string{dataPath, container.RootFs, console}
+
+	command := exec.Command(init, append([]string{"exec", "--func", "setup", "--"}, args...)...)
+
+	// make sure the process is executed inside the context of the rootfs
+	command.Dir = container.RootFs
+	command.Env = append(os.Environ(), env...)
+
 	return command
 }
 

namespaces/execin.go

@@ -16,7 +16,9 @@ import (
 	"github.com/docker/libcontainer/apparmor"
 	"github.com/docker/libcontainer/cgroups"
 	"github.com/docker/libcontainer/label"
+	"github.com/docker/libcontainer/mount"
 	"github.com/docker/libcontainer/system"
+	"github.com/docker/libcontainer/utils"
 )
 
 // ExecIn reexec's the initPath with the argv 0 rewrite to "nsenter" so that it is able to run the
@@ -127,6 +129,61 @@ func FinalizeSetns(container *libcontainer.Config, args []string) error {
 	panic("unreachable")
 }
 
+// SetupContainer is run to setup mounts and networking related operations
+// for a user namespace enabled process as a user namespace root doesn't
+// have permissions to perform these operations.
+// The setup process joins all the namespaces of user namespace enabled init
+// except the user namespace, so it run as root in the root user namespace
+// to perform these operations.
+func SetupContainer(container *libcontainer.Config, dataPath, uncleanRootfs, consolePath string) 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
+	}
+
+	state, err := libcontainer.GetState(dataPath)
+	if err != nil && !os.IsNotExist(err) {
+		return fmt.Errorf("unable to read state: %s", err)
+	}
+
+	if err := setupNetwork(container, &state.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()
+
+	hostRootUid, err := GetHostRootUid(container)
+	if err != nil {
+		return fmt.Errorf("failed to get hostRootUid %s", err)
+	}
+
+	hostRootGid, err := GetHostRootGid(container)
+	if err != nil {
+		return fmt.Errorf("failed to get hostRootGid %s", err)
+	}
+
+	if err := mount.InitializeMountNamespace(rootfs,
+		consolePath,
+		container.RestrictSys,
+		hostRootUid,
+		hostRootGid,
+		(*mount.MountConfig)(container.MountConfig)); err != nil {
+		return fmt.Errorf("setup mount namespace %s", err)
+	}
+
+	return nil
+}
+
 func EnterCgroups(state *libcontainer.State, pid int) error {
 	return cgroups.EnterPid(state.CgroupPaths, pid)
 }