podman, userNS: configure an intermediate mount namespace

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>

Closes: #690
Approved by: mheon

6 files changed, 168 insertions, 20 deletions

diff --git a/libpod/container.go b/libpod/container.go
index 5fa27af58..e35a620ed 100644
--- a/libpod/container.go
+++ b/libpod/container.go
@@ -115,11 +115,18 @@ type containerState struct {
 	ConfigPath string `json:"configPath,omitempty"`
 	// RunDir is a per-boot directory for container content
 	RunDir string `json:"runDir,omitempty"`
+	// DestinationRunDir is where the files in RunDir will be accessible for the container.
+	// It is different than RunDir when using userNS
+	DestinationRunDir string `json:"destinationRunDir,omitempty"`
 	// Mounted indicates whether the container's storage has been mounted
 	// for use
 	Mounted bool `json:"mounted,omitempty"`
-	// MountPoint contains the path to the container's mounted storage
+	// Mountpoint contains the path to the container's mounted storage as given
+	// by containers/storage.  It can be different than RealMountpoint when
+	// usernamespaces are used
 	Mountpoint string `json:"mountPoint,omitempty"`
+	// RealMountpoint contains the path to the container's mounted storage
+	RealMountpoint string `json:"realMountPoint,omitempty"`
 	// StartedTime is the time the container was started
 	StartedTime time.Time `json:"startedTime,omitempty"`
 	// FinishedTime is the time the container finished executing
@@ -152,6 +159,10 @@ type containerState struct {
 	// This maps the path the file will be mounted to in the container to
 	// the path of the file on disk outside the container
 	BindMounts map[string]string `json:"bindMounts,omitempty"`
+
+	// UserNSRoot is the directory used as root for the container when using
+	// user namespaces.
+	UserNSRoot string `json:"userNSRoot,omitempty"`
 }
 
 // ExecSession contains information on an active exec session
diff --git a/libpod/container_internal.go b/libpod/container_internal.go
index 18b56e23c..5417c8a4f 100644
--- a/libpod/container_internal.go
+++ b/libpod/container_internal.go
@@ -16,6 +16,7 @@ import (
 	"github.com/containers/storage"
 	"github.com/containers/storage/pkg/archive"
 	"github.com/containers/storage/pkg/chrootarchive"
+	"github.com/containers/storage/pkg/idtools"
 	"github.com/docker/docker/pkg/mount"
 	"github.com/docker/docker/pkg/stringid"
 	spec "github.com/opencontainers/runtime-spec/specs-go"
@@ -196,8 +197,33 @@ func (c *Container) setupStorage(ctx context.Context) error {
 		return errors.Wrapf(err, "error creating container storage")
 	}
 
+	if len(c.config.IDMappings.UIDMap) != 0 || len(c.config.IDMappings.GIDMap) != 0 {
+		info, err := os.Stat(c.runtime.config.TmpDir)
+		if err != nil {
+			return errors.Wrapf(err, "cannot stat `%s`", c.runtime.config.TmpDir)
+		}
+		if err := os.Chmod(c.runtime.config.TmpDir, info.Mode()|0111); err != nil {
+			return errors.Wrapf(err, "cannot chmod `%s`", c.runtime.config.TmpDir)
+		}
+		root := filepath.Join(c.runtime.config.TmpDir, "containers-root", c.ID())
+		if err := os.MkdirAll(root, 0755); err != nil {
+			return errors.Wrapf(err, "error creating userNS tmpdir for container %s", c.ID())
+		}
+		if err := os.Chown(root, c.RootUID(), c.RootGID()); err != nil {
+			return err
+		}
+		c.state.UserNSRoot, err = filepath.EvalSymlinks(root)
+		if err != nil {
+			return errors.Wrapf(err, "failed to eval symlinks for %s", root)
+		}
+	}
+
 	c.config.StaticDir = containerInfo.Dir
 	c.state.RunDir = containerInfo.RunDir
+	c.state.DestinationRunDir = c.state.RunDir
+	if c.state.UserNSRoot != "" {
+		c.state.DestinationRunDir = filepath.Join(c.state.UserNSRoot, "rundir")
+	}
 
 	// Set the default Entrypoint and Command
 	c.config.Entrypoint = containerInfo.Config.Config.Entrypoint
@@ -230,6 +256,12 @@ func (c *Container) teardownStorage() error {
 		return errors.Wrapf(err, "failed to cleanup container %s storage", c.ID())
 	}
 
+	if c.state.UserNSRoot != "" {
+		if err := os.RemoveAll(c.state.UserNSRoot); err != nil {
+			return errors.Wrapf(err, "error removing userns root %q", c.state.UserNSRoot)
+		}
+	}
+
 	if err := c.runtime.storageService.DeleteContainer(c.ID()); err != nil {
 		// If the container has already been removed, warn but do not
 		// error - we wanted it gone, it is already gone.
@@ -261,9 +293,35 @@ func (c *Container) refresh() error {
 	if err != nil {
 		return errors.Wrapf(err, "error retrieving temporary directory for container %s", c.ID())
 	}
+
+	if len(c.config.IDMappings.UIDMap) != 0 || len(c.config.IDMappings.GIDMap) != 0 {
+		info, err := os.Stat(c.runtime.config.TmpDir)
+		if err != nil {
+			return errors.Wrapf(err, "cannot stat `%s`", c.runtime.config.TmpDir)
+		}
+		if err := os.Chmod(c.runtime.config.TmpDir, info.Mode()|0111); err != nil {
+			return errors.Wrapf(err, "cannot chmod `%s`", c.runtime.config.TmpDir)
+		}
+		root := filepath.Join(c.runtime.config.TmpDir, "containers-root", c.ID())
+		if err := os.MkdirAll(root, 0755); err != nil {
+			return errors.Wrapf(err, "error creating userNS tmpdir for container %s", c.ID())
+		}
+		if err := os.Chown(root, c.RootUID(), c.RootGID()); err != nil {
+			return err
+		}
+		c.state.UserNSRoot, err = filepath.EvalSymlinks(root)
+		if err != nil {
+			return errors.Wrapf(err, "failed to eval symlinks for %s", root)
+		}
+	}
+
 	c.state.RunDir = dir
+	c.state.DestinationRunDir = c.state.RunDir
+	if c.state.UserNSRoot != "" {
+		c.state.DestinationRunDir = filepath.Join(c.state.UserNSRoot, "rundir")
+	}
 
-	if err := c.runtime.state.SaveContainer(c); err != nil {
+	if err := c.save(); err != nil {
 		return errors.Wrapf(err, "error refreshing state for container %s", c.ID())
 	}
 
@@ -600,6 +658,10 @@ func (c *Container) mountStorage() (err error) {
 		return errors.Wrapf(err, "unable to determine if %q is mounted", c.config.ShmDir)
 	}
 
+	if err := os.Chown(c.config.ShmDir, c.RootUID(), c.RootGID()); err != nil {
+		return err
+	}
+
 	if !mounted {
 		shmOptions := fmt.Sprintf("mode=1777,size=%d", c.config.ShmSize)
 		if err := unix.Mount("shm", c.config.ShmDir, "tmpfs", unix.MS_NOEXEC|unix.MS_NOSUID|unix.MS_NODEV,
@@ -607,7 +669,7 @@ func (c *Container) mountStorage() (err error) {
 			return errors.Wrapf(err, "failed to mount shm tmpfs %q", c.config.ShmDir)
 		}
 		if err := os.Chown(c.config.ShmDir, c.RootUID(), c.RootGID()); err != nil {
-			return err
+			return errors.Wrapf(err, "failed to chown %s", c.config.ShmDir)
 		}
 	}
 
@@ -617,6 +679,11 @@ func (c *Container) mountStorage() (err error) {
 	}
 	c.state.Mounted = true
 	c.state.Mountpoint = mountPoint
+	if c.state.UserNSRoot == "" {
+		c.state.RealMountpoint = c.state.Mountpoint
+	} else {
+		c.state.RealMountpoint = filepath.Join(c.state.UserNSRoot, "mountpoint")
+	}
 
 	logrus.Debugf("Created root filesystem for container %s at %s", c.ID(), c.state.Mountpoint)
 
@@ -716,6 +783,10 @@ func (c *Container) cleanup() error {
 
 // Make standard bind mounts to include in the container
 func (c *Container) makeBindMounts() error {
+	if err := os.Chown(c.state.RunDir, c.RootUID(), c.RootGID()); err != nil {
+		return errors.Wrapf(err, "error chown %s", c.state.RunDir)
+	}
+
 	if c.state.BindMounts == nil {
 		c.state.BindMounts = make(map[string]string)
 	}
@@ -724,11 +795,8 @@ func (c *Container) makeBindMounts() error {
 	c.state.BindMounts["/dev/shm"] = c.config.ShmDir
 
 	// Make /etc/resolv.conf
-	if path, ok := c.state.BindMounts["/etc/resolv.conf"]; ok {
+	if _, ok := c.state.BindMounts["/etc/resolv.conf"]; ok {
 		// If it already exists, delete so we can recreate
-		if err := os.Remove(path); err != nil {
-			return errors.Wrapf(err, "error removing resolv.conf for container %s", c.ID())
-		}
 		delete(c.state.BindMounts, "/etc/resolv.conf")
 	}
 	newResolv, err := c.generateResolvConf()
@@ -738,11 +806,8 @@ func (c *Container) makeBindMounts() error {
 	c.state.BindMounts["/etc/resolv.conf"] = newResolv
 
 	// Make /etc/hosts
-	if path, ok := c.state.BindMounts["/etc/hosts"]; ok {
+	if _, ok := c.state.BindMounts["/etc/hosts"]; ok {
 		// If it already exists, delete so we can recreate
-		if err := os.Remove(path); err != nil {
-			return errors.Wrapf(err, "error removing hosts file for container %s", c.ID())
-		}
 		delete(c.state.BindMounts, "/etc/hosts")
 	}
 	newHosts, err := c.generateHosts()
@@ -773,7 +838,7 @@ func (c *Container) makeBindMounts() error {
 	}
 
 	// Add Secret Mounts
-	secretMounts := secrets.SecretMountsWithUIDGID(c.config.MountLabel, c.state.RunDir, c.runtime.config.DefaultMountsFile, c.RootUID(), c.RootGID())
+	secretMounts := secrets.SecretMountsWithUIDGID(c.config.MountLabel, c.state.RunDir, c.runtime.config.DefaultMountsFile, c.state.DestinationRunDir, c.RootUID(), c.RootGID())
 	for _, mount := range secretMounts {
 		if _, ok := c.state.BindMounts[mount.Destination]; !ok {
 			c.state.BindMounts[mount.Destination] = mount.Source
@@ -786,6 +851,11 @@ func (c *Container) makeBindMounts() error {
 // writeStringToRundir copies the provided file to the runtimedir
 func (c *Container) writeStringToRundir(destFile, output string) (string, error) {
 	destFileName := filepath.Join(c.state.RunDir, destFile)
+
+	if err := os.Remove(destFileName); err != nil && !os.IsNotExist(err) {
+		return "", errors.Wrapf(err, "error removing %s for container %s", destFile, c.ID())
+	}
+
 	f, err := os.Create(destFileName)
 	if err != nil {
 		return "", errors.Wrapf(err, "unable to create %s", destFileName)
@@ -802,7 +872,8 @@ func (c *Container) writeStringToRundir(destFile, output string) (string, error)
 	if err := label.Relabel(destFileName, c.config.MountLabel, false); err != nil {
 		return "", err
 	}
-	return destFileName, nil
+
+	return filepath.Join(c.state.DestinationRunDir, destFile), nil
 }
 
 type resolvConf struct {
@@ -1035,7 +1106,11 @@ func (c *Container) generateSpec(ctx context.Context) (*spec.Spec, error) {
 		}
 	}
 
-	g.SetRootPath(c.state.Mountpoint)
+	if err := idtools.MkdirAllAs(c.state.RealMountpoint, 0700, c.RootUID(), c.RootGID()); err != nil {
+		return nil, err
+	}
+
+	g.SetRootPath(c.state.RealMountpoint)
 	g.AddAnnotation(crioAnnotations.Created, c.config.CreatedTime.Format(time.RFC3339Nano))
 	g.AddAnnotation("org.opencontainers.image.stopSignal", fmt.Sprintf("%d", c.config.StopSignal))
 
diff --git a/libpod/oci.go b/libpod/oci.go
index b59525de7..9c842f2c9 100644
--- a/libpod/oci.go
+++ b/libpod/oci.go
@@ -8,11 +8,14 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"runtime"
 	"strconv"
+	"sync"
 	"syscall"
 	"time"
 
 	"github.com/containerd/cgroups"
+	"github.com/containers/storage/pkg/idtools"
 	"github.com/coreos/go-systemd/activation"
 	spec "github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/pkg/errors"
@@ -186,6 +189,53 @@ func waitPidsStop(pids []int, timeout time.Duration) error {
 // TODO terminal support for container
 // Presently just ignoring conmon opts related to it
 func (r *OCIRuntime) createContainer(ctr *Container, cgroupParent string) (err error) {
+	if ctr.state.UserNSRoot == "" {
+		// no need of an intermediate mount ns
+		return r.createOCIContainer(ctr, cgroupParent)
+	}
+	var wg sync.WaitGroup
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		runtime.LockOSThread()
+
+		fd, err := os.Open(fmt.Sprintf("/proc/%d/task/%d/ns/mnt", os.Getpid(), unix.Gettid()))
+		if err != nil {
+			return
+		}
+		defer fd.Close()
+
+		// create a new mountns on the current thread
+		if err = unix.Unshare(unix.CLONE_NEWNS); err != nil {
+			return
+		}
+		defer unix.Setns(int(fd.Fd()), unix.CLONE_NEWNS)
+
+		// don't spread our mounts around
+		err = unix.Mount("/", "/", "none", unix.MS_REC|unix.MS_SLAVE, "")
+		if err != nil {
+			return
+		}
+		err = unix.Mount(ctr.state.Mountpoint, ctr.state.RealMountpoint, "none", unix.MS_BIND, "")
+		if err != nil {
+			return
+		}
+		if err := idtools.MkdirAllAs(ctr.state.DestinationRunDir, 0700, ctr.RootUID(), ctr.RootGID()); err != nil {
+			return
+		}
+
+		err = unix.Mount(ctr.state.RunDir, ctr.state.DestinationRunDir, "none", unix.MS_BIND, "")
+		if err != nil {
+			return
+		}
+		err = r.createOCIContainer(ctr, cgroupParent)
+	}()
+	wg.Wait()
+
+	return err
+}
+
+func (r *OCIRuntime) createOCIContainer(ctr *Container, cgroupParent string) (err error) {
 	var stderrBuf bytes.Buffer
 
 	parentPipe, childPipe, err := newPipe()
diff --git a/libpod/runtime.go b/libpod/runtime.go
index 9de70da1b..dde81a59d 100644
--- a/libpod/runtime.go
+++ b/libpod/runtime.go
@@ -340,6 +340,14 @@ func makeRuntime(runtime *Runtime) error {
 		SignaturePolicyPath: runtime.config.SignaturePolicyPath,
 	}
 
+	// Create the tmpDir
+	if err := os.MkdirAll(runtime.config.TmpDir, 0751); err != nil {
+		// The directory is allowed to exist
+		if !os.IsExist(err) {
+			return errors.Wrapf(err, "error creating tmpdir %s", runtime.config.TmpDir)
+		}
+	}
+
 	// Make an OCI runtime to perform container operations
 	ociRuntime, err := newOCIRuntime("runc", runtime.ociRuntimePath,
 		runtime.conmonPath, runtime.config.ConmonEnvVars,
diff --git a/libpod/runtime_ctr.go b/libpod/runtime_ctr.go
index 3f27df5b9..4708e0c8f 100644
--- a/libpod/runtime_ctr.go
+++ b/libpod/runtime_ctr.go
@@ -76,7 +76,11 @@ func (r *Runtime) NewContainer(ctx context.Context, rSpec *spec.Spec, options ..
 		ctr.config.LogPath = filepath.Join(ctr.config.StaticDir, "ctr.log")
 	}
 	if ctr.config.ShmDir == "" {
-		ctr.config.ShmDir = filepath.Join(ctr.bundlePath(), "shm")
+		if ctr.state.UserNSRoot == "" {
+			ctr.config.ShmDir = filepath.Join(ctr.bundlePath(), "shm")
+		} else {
+			ctr.config.ShmDir = filepath.Join(ctr.state.UserNSRoot, "shm")
+		}
 		if err := os.MkdirAll(ctr.config.ShmDir, 0700); err != nil {
 			if !os.IsExist(err) {
 				return nil, errors.Wrapf(err, "unable to create shm %q dir", ctr.config.ShmDir)
diff --git a/pkg/secrets/secrets.go b/pkg/secrets/secrets.go
index 29ccd4592..0e48c7a12 100644
--- a/pkg/secrets/secrets.go
+++ b/pkg/secrets/secrets.go
@@ -128,11 +128,11 @@ func getMountsMap(path string) (string, string, error) {
 
 // SecretMounts copies, adds, and mounts the secrets to the container root filesystem
 func SecretMounts(mountLabel, containerWorkingDir, mountFile string) []rspec.Mount {
-	return SecretMountsWithUIDGID(mountLabel, containerWorkingDir, mountFile, 0, 0)
+	return SecretMountsWithUIDGID(mountLabel, containerWorkingDir, mountFile, containerWorkingDir, 0, 0)
 }
 
 // SecretMountsWithUIDGID specifies the uid/gid of the owner
-func SecretMountsWithUIDGID(mountLabel, containerWorkingDir, mountFile string, uid, gid int) []rspec.Mount {
+func SecretMountsWithUIDGID(mountLabel, containerWorkingDir, mountFile, mountPrefix string, uid, gid int) []rspec.Mount {
 	var (
 		secretMounts []rspec.Mount
 		mountFiles   []string
@@ -146,7 +146,7 @@ func SecretMountsWithUIDGID(mountLabel, containerWorkingDir, mountFile string, u
 		mountFiles = append(mountFiles, mountFile)
 	}
 	for _, file := range mountFiles {
-		mounts, err := addSecretsFromMountsFile(file, mountLabel, containerWorkingDir, uid, gid)
+		mounts, err := addSecretsFromMountsFile(file, mountLabel, containerWorkingDir, mountPrefix, uid, gid)
 		if err != nil {
 			logrus.Warnf("error mounting secrets, skipping: %v", err)
 		}
@@ -175,7 +175,7 @@ func rchown(chowndir string, uid, gid int) error {
 
 // addSecretsFromMountsFile copies the contents of host directory to container directory
 // and returns a list of mounts
-func addSecretsFromMountsFile(filePath, mountLabel, containerWorkingDir string, uid, gid int) ([]rspec.Mount, error) {
+func addSecretsFromMountsFile(filePath, mountLabel, containerWorkingDir, mountPrefix string, uid, gid int) ([]rspec.Mount, error) {
 	var mounts []rspec.Mount
 	defaultMountsPaths := getMounts(filePath)
 	for _, path := range defaultMountsPaths {
@@ -226,7 +226,7 @@ func addSecretsFromMountsFile(filePath, mountLabel, containerWorkingDir string,
 		}
 
 		m := rspec.Mount{
-			Source:      ctrDirOnHost,
+			Source:      filepath.Join(mountPrefix, ctrDir),
 			Destination: ctrDir,
 			Type:        "bind",
 			Options:     []string{"bind"},