Support running nested SELinux container separation

Currently Podman prevents SELinux container separation,
when running within a container. This PR adds a new
--security-opt label=nested

When setting this option, Podman unmasks and mountsi
/sys/fs/selinux into the containers making /sys/fs/selinux
fully exposed. Secondly Podman sets the attribute
run.oci.mount_context_type=rootcontext

This attribute tells crun to mount volumes with rootcontext=MOUNTLABEL
as opposed to context=MOUNTLABEL.

With these two settings Podman inside the container is allowed to set
its own SELinux labels on tmpfs file systems mounted into its parents
container, while still being confined by SELinux. Thus you can have
nested SELinux labeling inside of a container.

Signed-off-by: Daniel J Walsh <dwalsh@redhat.com>

cmd/podman/containers/create.go

@@ -257,7 +257,7 @@ func CreateInit(c *cobra.Command, vals entities.ContainerCreateOptions, isInfra
 					if registry.IsRemote() {
 						return vals, errors.New("the '--group-add keep-groups' option is not supported in remote mode")
 					}
-					vals.Annotation = append(vals.Annotation, "run.oci.keep_original_groups=1")
+					vals.Annotation = append(vals.Annotation, fmt.Sprintf("%s=1", define.RunOCIKeepOriginalGroups))
 				} else {
 					groups = append(groups, g)
 				}

docs/source/markdown/options/security-opt.md

@@ -18,6 +18,8 @@ Security Options
 
 Note: Labeling can be disabled for all <<|pods/>>containers by setting label=false in the **containers.conf** (`/etc/containers/containers.conf` or `$HOME/.config/containers/containers.conf`) file.
 
+- **label=nested**: Allows SELinux modifications within the container. Containers are allowed to modify SELinux labels on files and processes, as long as SELinux policy allows. Without **nested**, containers view SELinux as disabled, even when it is enabled on the host. Containers are prevented from setting any labels.
+
 - **mask**=_/path/1:/path/2_: The paths to mask separated by a colon. A masked path cannot be accessed inside the container<<s within the pod|>>.
 
 - **no-new-privileges**: Disable container processes from gaining additional privileges.

libpod/container_config.go

@@ -219,6 +219,8 @@ type ContainerSecurityConfig struct {
 	// Libpod - mostly used in rootless containers where the user running
 	// Libpod wants to retain their UID inside the container.
 	AddCurrentUserPasswdEntry bool `json:"addCurrentUserPasswdEntry,omitempty"`
+	// LabelNested, allow labeling separation from within a container
+	LabelNested bool `json:"label_nested"`
 }
 
 // ContainerNameSpaceConfig is an embedded sub-config providing

libpod/container_internal_linux.go

@@ -32,8 +32,13 @@ var (
 )
 
 func (c *Container) mountSHM(shmOptions string) error {
+	contextType := "context"
+	if c.config.LabelNested {
+		contextType = "rootcontext"
+	}
+
 	if err := unix.Mount("shm", c.config.ShmDir, "tmpfs", unix.MS_NOEXEC|unix.MS_NOSUID|unix.MS_NODEV,
-		label.FormatMountLabel(shmOptions, c.config.MountLabel)); err != nil {
+		label.FormatMountLabelByType(shmOptions, c.config.MountLabel, contextType)); err != nil {
 		return fmt.Errorf("failed to mount shm tmpfs %q: %w", c.config.ShmDir, err)
 	}
 	return nil

libpod/define/annotations.go

@@ -1,6 +1,12 @@
 package define
 
 const (
+	// RunOCIMountContextType tells the OCI runtime which context mount
+	// type to use. context, rootcontext, fscontext, defcontext
+	RunOCIMountContextType = "run.oci.mount_context_type"
+	// RunOCIKeepOriginalGroups tells the OCI runtime to leak the users
+	// current groups into the container
+	RunOCIKeepOriginalGroups = "run.oci.keep_original_groups"
 	// InspectAnnotationCIDFile is used by Inspect to determine if a
 	// container ID file was created for the container.
 	// If an annotation with this key is found in the OCI spec, it will be
@@ -58,7 +64,6 @@ const (
 	// If an annotation with this key is found in the OCI spec, it will be
 	// used in the output of Inspect().
 	InspectAnnotationApparmor = "io.podman.annotations.apparmor"
-
 	// InspectResponseTrue is a boolean True response for an inspect
 	// annotation.
 	InspectResponseTrue = "TRUE"

libpod/options.go

@@ -2341,3 +2341,16 @@ func WithMountAllDevices() CtrCreateOption {
 		return nil
 	}
 }
+
+// WithLabelNested sets the LabelNested flag allowing label separation within container
+func WithLabelNested(nested bool) CtrCreateOption {
+	return func(ctr *Container) error {
+		if ctr.valid {
+			return define.ErrCtrFinalized
+		}
+
+		ctr.config.LabelNested = nested
+
+		return nil
+	}
+}

libpod/runtime_volume_common.go

@@ -120,15 +120,13 @@ func (r *Runtime) newVolume(ctx context.Context, noCreatePluginVolume bool, opti
 		volume.config.StorageImageID = image.ID()
 
 		// Create a backing container in c/storage.
-		storageConfig := storage.ContainerOptions{
-			LabelOpts: []string{"filetype:container_file_t:s0"},
-		}
+		storageConfig := storage.ContainerOptions{}
 		if len(volume.config.MountLabel) > 0 {
 			context, err := selinux.NewContext(volume.config.MountLabel)
 			if err != nil {
 				return nil, fmt.Errorf("failed to get SELinux context from %s: %w", volume.config.MountLabel, err)
 			}
-			storageConfig.LabelOpts = []string{fmt.Sprintf("filetype:%s:s0", context["type"])}
+			storageConfig.LabelOpts = []string{fmt.Sprintf("filetype:%s", context["type"])}
 		}
 		if _, err := r.storageService.CreateContainerStorage(ctx, r.imageContext, imgString, image.ID(), volume.config.StorageName, volume.config.StorageID, storageConfig); err != nil {
 			return nil, fmt.Errorf("creating backing storage for image driver: %w", err)

pkg/specgen/generate/container_create.go

@@ -482,6 +482,9 @@ func createContainerOptions(rt *libpod.Runtime, s *specgen.SpecGenerator, pod *l
 			options = append(options, libpod.WithLogDriver(s.LogConfiguration.Driver))
 		}
 	}
+	if s.ContainerSecurityConfig.LabelNested {
+		options = append(options, libpod.WithLabelNested(s.ContainerSecurityConfig.LabelNested))
+	}
 	// Security options
 	if len(s.SelinuxOpts) > 0 {
 		options = append(options, libpod.WithSecLabels(s.SelinuxOpts))

pkg/specgen/specgen.go

@@ -396,6 +396,10 @@ type ContainerSecurityConfig struct {
 	// mount temporary file systems
 	ReadWriteTmpfs bool `json:"read_write_tmpfs,omitempty"`
 
+	// LabelNested indicates whether or not the container is allowed to
+	// run fully nested containers including labelling
+	LabelNested bool `json:"label_nested,omitempty"`
+
 	// Umask is the umask the init process of the container will be run with.
 	Umask string `json:"umask,omitempty"`
 	// ProcOpts are the options used for the proc mount.

pkg/specgenutil/specgen.go

@@ -620,9 +620,6 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions
 	}
 
 	for _, opt := range c.SecurityOpt {
-		if opt == "no-new-privileges" {
-			s.ContainerSecurityConfig.NoNewPrivileges = true
-		} else {
 		// Docker deprecated the ":" syntax but still supports it,
 		// so we need to as well
 		var con []string
@@ -631,7 +628,8 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions
 		} else {
 			con = strings.SplitN(opt, ":", 2)
 		}
-			if len(con) != 2 {
+		if len(con) != 2 &&
+			con[0] != "no-new-privileges" {
 			return fmt.Errorf("invalid --security-opt 1: %q", opt)
 		}
 		switch con[0] {
@@ -639,6 +637,10 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions
 			s.ContainerSecurityConfig.ApparmorProfile = con[1]
 			s.Annotations[define.InspectAnnotationApparmor] = con[1]
 		case "label":
+			if con[1] == "nested" {
+				s.ContainerSecurityConfig.LabelNested = true
+				continue
+			}
 			// TODO selinux opts and label opts are the same thing
 			s.ContainerSecurityConfig.SelinuxOpts = append(s.ContainerSecurityConfig.SelinuxOpts, con[1])
 			s.Annotations[define.InspectAnnotationLabel] = strings.Join(s.ContainerSecurityConfig.SelinuxOpts, ",label=")
@@ -659,16 +661,18 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions
 		case "unmask":
 			s.ContainerSecurityConfig.Unmask = append(s.ContainerSecurityConfig.Unmask, con[1:]...)
 		case "no-new-privileges":
-				noNewPrivileges, err := strconv.ParseBool(con[1])
+			noNewPrivileges := true
+			if len(con) == 2 {
+				noNewPrivileges, err = strconv.ParseBool(con[1])
 				if err != nil {
 					return fmt.Errorf("invalid --security-opt 2: %q", opt)
 				}
+			}
 			s.ContainerSecurityConfig.NoNewPrivileges = noNewPrivileges
 		default:
 			return fmt.Errorf("invalid --security-opt 2: %q", opt)
 		}
 	}
-	}
 
 	if len(s.SeccompPolicy) == 0 || len(c.SeccompPolicy) != 0 {
 		s.SeccompPolicy = c.SeccompPolicy
@@ -690,6 +694,17 @@ func FillOutSpecGen(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions
 	if len(s.Volumes) == 0 || len(c.Volume) != 0 {
 		s.Volumes = volumes
 	}
+
+	if s.ContainerSecurityConfig.LabelNested {
+		// Need to unmask the SELinux file system
+		s.Unmask = append(s.Unmask, "/sys/fs/selinux", "/proc")
+		s.Mounts = append(s.Mounts, specs.Mount{
+			Source:      "/sys/fs/selinux",
+			Destination: "/sys/fs/selinux",
+			Type:        define.TypeBind,
+		})
+		s.Annotations[define.RunOCIMountContextType] = "rootcontext"
+	}
 	// TODO make sure these work in clone
 	if len(s.OverlayVolumes) == 0 {
 		s.OverlayVolumes = overlayVolumes

test/e2e/containers_conf_test.go

@@ -7,6 +7,7 @@ import (
 	"path/filepath"
 	"strings"
 
+	"github.com/containers/podman/v4/libpod/define"
 	. "github.com/containers/podman/v4/test/utils"
 	. "github.com/onsi/ginkgo"
 	. "github.com/onsi/gomega"
@@ -111,7 +112,7 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		result := podmanTest.Podman([]string{"top", "test1", "capeff"})
 		result.WaitWithDefaultTimeout()
 		Expect(result).Should(Exit(0))
-		Expect(result.Out.Contents()).To(
+		Expect(result.OutputToString()).To(
 			And(
 				ContainSubstring("FOWNER"),
 				ContainSubstring("SETFCAP"),
@@ -128,7 +129,7 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		result := podmanTest.Podman([]string{"container", "top", "test1", "capeff"})
 		result.WaitWithDefaultTimeout()
 		Expect(result).Should(Exit(0))
-		Expect(result.Out.Contents()).ToNot(
+		Expect(result.OutputToString()).ToNot(
 			And(
 				ContainSubstring("SETUID"),
 				ContainSubstring("FOWNER"),
@@ -266,7 +267,7 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
 		Expect(session.OutputToStringArray()).To(ContainElement(HavePrefix("search")))
-		Expect(session.Out.Contents()).To(
+		Expect(session.OutputToString()).To(
 			And(
 				ContainSubstring("foobar.com"),
 				ContainSubstring("1.2.3.4"),
@@ -322,7 +323,7 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
 		Expect(session.OutputToStringArray()).To(ContainElement(HavePrefix("search")))
-		Expect(session.Out.Contents()).To(
+		Expect(session.OutputToString()).To(
 			And(
 				ContainSubstring("foobar.com"),
 				ContainSubstring("1.2.3.4"),
@@ -333,26 +334,26 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		session = podmanTest.Podman([]string{"run", "--rm", ALPINE, "cat", "/proc/sys/net/ipv4/ping_group_range"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
-		Expect(session.Out.Contents()).To(ContainSubstring("1000"))
+		Expect(session.OutputToString()).To(ContainSubstring("1000"))
 
 		// shm-size
 		session = podmanTest.Podman([]string{"run", ALPINE, "grep", "shm", "/proc/self/mounts"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
-		Expect(session.Out.Contents()).To(ContainSubstring("size=200k"))
+		Expect(session.OutputToString()).To(ContainSubstring("size=200k"))
 
 		// ulimits
 		session = podmanTest.Podman([]string{"run", "--rm", fedoraMinimal, "ulimit", "-n"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
-		Expect(session.Out.Contents()).To(ContainSubstring("500"))
+		Expect(session.OutputToString()).To(ContainSubstring("500"))
 
 		// Configuration that comes from remote client
 		// Timezone
 		session = podmanTest.Podman([]string{"run", ALPINE, "date", "+'%H %Z'"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
-		Expect(session.Out.Contents()).To(
+		Expect(session.OutputToString()).To(
 			Or(
 				ContainSubstring("EST"),
 				ContainSubstring("EDT"),
@@ -373,14 +374,14 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 
 		inspect := podmanTest.Podman([]string{"inspect", "--format", "{{ .Config.Annotations }}", "test"})
 		inspect.WaitWithDefaultTimeout()
-		Expect(inspect.Out.Contents()).To(ContainSubstring("run.oci.keep_original_groups:1"))
+		Expect(inspect.OutputToString()).To(ContainSubstring(fmt.Sprintf("%s:1", define.RunOCIKeepOriginalGroups)))
 	})
 
 	It("--add-host and no-hosts=true fails", func() {
 		session := podmanTest.Podman([]string{"run", "-dt", "--add-host", "test1:127.0.0.1", ALPINE, "top"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).To(ExitWithError())
-		Expect(session.Err.Contents()).To(ContainSubstring("--no-hosts and --add-host cannot be set together"))
+		Expect(session.ErrorToString()).To(ContainSubstring("--no-hosts and --add-host cannot be set together"))
 
 		session = podmanTest.Podman([]string{"run", "-dt", "--add-host", "test1:127.0.0.1", "--no-hosts=false", ALPINE, "top"})
 		session.WaitWithDefaultTimeout()
@@ -391,12 +392,12 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		session := podmanTest.Podman([]string{"run", "--rm", "--name", "test", ALPINE, "cat", "/etc/hosts"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
-		Expect(session.Out.Contents()).ToNot(ContainSubstring("test"))
+		Expect(session.OutputToString()).ToNot(ContainSubstring("test"))
 
 		session = podmanTest.Podman([]string{"run", "--rm", "--name", "test", "--no-hosts=false", ALPINE, "cat", "/etc/hosts"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
-		Expect(session.Out.Contents()).To(ContainSubstring("test"))
+		Expect(session.OutputToString()).To(ContainSubstring("test"))
 	})
 
 	It("seccomp profile path", func() {
@@ -462,7 +463,7 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		session = podmanTest.Podman([]string{"info", "--format", "{{.Store.ImageCopyTmpDir}}"})
 		session.WaitWithDefaultTimeout()
 		Expect(session).Should(Exit(0))
-		Expect(session.Out.Contents()).To(ContainSubstring(storagePath))
+		Expect(session.OutputToString()).To(ContainSubstring(storagePath))
 
 		containersConf = []byte("[engine]\nimage_copy_tmp_dir=\"storage1\"")
 		err = os.WriteFile(configPath, containersConf, os.ModePerm)
@@ -472,7 +473,7 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 			session = podmanTest.Podman([]string{"info", "--format", "{{.Store.ImageCopyTmpDir}}"})
 			session.WaitWithDefaultTimeout()
 			Expect(session).Should(Exit(125))
-			Expect(session.Err.Contents()).To(ContainSubstring("invalid image_copy_tmp_dir value \"storage1\" (relative paths are not accepted)"))
+			Expect(session.ErrorToString()).To(ContainSubstring("invalid image_copy_tmp_dir value \"storage1\" (relative paths are not accepted)"))
 
 			os.Setenv("TMPDIR", "/hoge")
 			session = podmanTest.Podman([]string{"info", "--format", "{{.Store.ImageCopyTmpDir}}"})
@@ -490,7 +491,7 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		result := podmanTest.Podman([]string{"system", "service", "--help"})
 		result.WaitWithDefaultTimeout()
 		Expect(result).Should(Exit(0))
-		Expect(result.Out.Contents()).To(ContainSubstring("(default 1234)"))
+		Expect(result.OutputToString()).To(ContainSubstring("(default 1234)"))
 	})
 
 	It("bad infra_image name", func() {
@@ -512,17 +513,17 @@ var _ = Describe("Verify podman containers.conf usage", func() {
 		result := podmanTest.Podman([]string{"pod", "create", "--infra-image", infra2})
 		result.WaitWithDefaultTimeout()
 		Expect(result).Should(Exit(125))
-		Expect(result.Err.Contents()).To(ContainSubstring(error2String))
+		Expect(result.ErrorToString()).To(ContainSubstring(error2String))
 
 		result = podmanTest.Podman([]string{"pod", "create"})
 		result.WaitWithDefaultTimeout()
 		Expect(result).Should(Exit(125))
-		Expect(result.Err.Contents()).To(ContainSubstring(errorString))
+		Expect(result.ErrorToString()).To(ContainSubstring(errorString))
 
 		result = podmanTest.Podman([]string{"create", "--pod", "new:pod1", ALPINE})
 		result.WaitWithDefaultTimeout()
 		Expect(result).Should(Exit(125))
-		Expect(result.Err.Contents()).To(ContainSubstring(errorString))
+		Expect(result.ErrorToString()).To(ContainSubstring(errorString))
 	})
 
 	It("set .engine.remote=true", func() {

test/system/410-selinux.bats

@@ -277,4 +277,20 @@ function check_label() {
     is "$output" "${RELABEL} $tmpdir" "Shared Relabel Correctly"
 }
 
+@test "podman selinux nested" {
+    skip_if_no_selinux
+
+    ROOTCONTEXT='rw,rootcontext="system_u:object_r:container_file_t:s0:c1,c2"'
+    SELINUXMNT="selinuxfs.*(rw,nosuid,noexec,relatime)"
+
+    SELINUXMNT="tmpfs.*selinux.*\(ro"
+    run_podman run --rm --security-opt label=level:s0:c1,c2 $IMAGE mount
+    assert "$output" !~ "${ROOTCONTEXT}" "Don't use rootcontext"
+    assert "$output" =~ "${SELINUXMNT}" "Mount SELinux file system readwrite"
+
+    run_podman run --rm --security-opt label=nested --security-opt label=level:s0:c1,c2 $IMAGE mount
+    assert "$output" =~ "${ROOTCONTEXT}" "Uses rootcontext"
+    assert "$output" =~ "${SELINUXMNT}" "Mount SELinux file system readwrite"
+}
+
 # vim: filetype=sh