Merge pull request containerd#3533 from austinvazquez/support-systempaths-unconfined

Add container run --security-opt systempaths=unconfined

Author: AkihiroSuda

cmd/nerdctl/container/container_run.go

@@ -183,6 +183,7 @@ func setCreateFlags(cmd *cobra.Command) {
 			"seccomp=", "seccomp=" + defaults.SeccompProfileName, "seccomp=unconfined",
 			"apparmor=", "apparmor=" + defaults.AppArmorProfileName, "apparmor=unconfined",
 			"no-new-privileges",
+			"systempaths=unconfined",
 			"privileged-without-host-devices"}, cobra.ShellCompDirectiveNoFileComp
 	})
 	// cap-add and cap-drop are defined as StringSlice, not StringArray, to allow specifying "--cap-add=CAP_SYS_ADMIN,CAP_NET_ADMIN" (compatible with Podman)

cmd/nerdctl/container/container_run_security_linux_test.go

@@ -193,6 +193,61 @@ func TestRunSeccompCapSysPtrace(t *testing.T) {
 	// Docker/Moby 's seccomp profile allows ptrace(2) by default, but containerd does not (yet): https://github.com/containerd/containerd/issues/6802
 }
 
+func TestRunSystemPathsUnconfined(t *testing.T) {
+	base := testutil.NewBase(t)
+
+	const findmnt = "`apk add -q findmnt && findmnt -R /proc && findmnt -R /sys`"
+	result := base.Cmd("run", "--rm", testutil.AlpineImage, "sh", "-euxc", findmnt).Run()
+	defaultContainerOutput := result.Combined()
+
+	var confined []string
+
+	for _, path := range []string{
+		"/proc/kcore",
+		"/proc/keys",
+		"/proc/latency_stats",
+		"/proc/sched_debug",
+		"/proc/scsi",
+		"/proc/timer_list",
+		"/proc/timer_stats",
+		"/sys/firmware",
+		"/sys/fs/selinux",
+	} {
+		// Not each distribution will support every masked path here.
+		if strings.Contains(defaultContainerOutput, path) {
+			confined = append(confined, path)
+		}
+	}
+
+	assert.Check(t, len(confined) != 0, "Default container has no confined paths to validate")
+
+	result = base.Cmd("run", "--rm", "--security-opt", "systempaths=unconfined", testutil.AlpineImage, "sh", "-euxc", findmnt).Run()
+	unconfinedContainerOutput := result.Combined()
+
+	for _, path := range confined {
+		assert.Assert(t, !strings.Contains(unconfinedContainerOutput, path), fmt.Sprintf("%s should not be masked when unconfined", path))
+	}
+
+	for _, path := range []string{
+		"/proc/acpi",
+		"/proc/bus",
+		"/proc/fs",
+		"/proc/irq",
+		"/proc/sysrq-trigger",
+		"/proc/sys",
+	} {
+		findmntPath := fmt.Sprintf("`apk add -q findmnt && findmnt %s`", path)
+
+		result := base.Cmd("run", "--rm", testutil.AlpineImage, "sh", "-euxc", findmntPath).Run()
+
+		// Not each distribution will support every read-only path here.
+		if strings.Contains(result.Combined(), path) {
+			result = base.Cmd("run", "--rm", "--security-opt", "systempaths=unconfined", testutil.AlpineImage, "sh", "-euxc", findmntPath).Run()
+			assert.Assert(t, !strings.Contains(result.Combined(), "ro,"), fmt.Sprintf("%s should not be read-only when unconfined", path))
+		}
+	}
+}
+
 func TestRunPrivileged(t *testing.T) {
 	// docker does not support --privileged-without-host-devices
 	testutil.DockerIncompatible(t)

docs/command-reference.md

@@ -230,6 +230,7 @@ Security flags:
 - :whale: `--security-opt seccomp=<PROFILE_JSON_FILE>`: specify custom seccomp profile
 - :whale: `--security-opt apparmor=<PROFILE>`: specify custom AppArmor profile
 - :whale: `--security-opt no-new-privileges`: disallow privilege escalation, e.g., setuid and file capabilities
+- :whale: `--security-opt systempaths=unconfined`: Turn off confinement for system paths (masked paths, read-only paths) for the container
 - :nerd_face: `--security-opt privileged-without-host-devices`: Don't pass host devices to privileged containers
 - :whale: `--cap-add=<CAP>`: Add Linux capabilities
 - :whale: `--cap-drop=<CAP>`: Drop Linux capabilities

pkg/cmd/container/run_security_linux.go

@@ -45,10 +45,14 @@ var privilegedWithoutDevicesOpts = []oci.SpecOpts{
 	oci.WithNewPrivileges,
 }
 
+const (
+	systemPathsUnconfined = "unconfined"
+)
+
 func generateSecurityOpts(privileged bool, securityOptsMap map[string]string) ([]oci.SpecOpts, error) {
 	for k := range securityOptsMap {
 		switch k {
-		case "seccomp", "apparmor", "no-new-privileges", "privileged-without-host-devices":
+		case "seccomp", "apparmor", "no-new-privileges", "systempaths", "privileged-without-host-devices":
 		default:
 			log.L.Warnf("unknown security-opt: %q", k)
 		}
@@ -99,6 +103,13 @@ func generateSecurityOpts(privileged bool, securityOptsMap map[string]string) ([
 		opts = append(opts, oci.WithNewPrivileges)
 	}
 
+	if value, ok := securityOptsMap["systempaths"]; ok && value == systemPathsUnconfined {
+		opts = append(opts, oci.WithMaskedPaths(nil))
+		opts = append(opts, oci.WithReadonlyPaths(nil))
+	} else if ok && value != systemPathsUnconfined {
+		return nil, errors.New(`invalid security-opt "systempaths=unconfined"`)
+	}
+
 	privilegedWithoutHostDevices, err := maputil.MapBoolValueAsOpt(securityOptsMap, "privileged-without-host-devices")
 	if err != nil {
 		return nil, err