linux: Support setting execution domain via linux personality

carry opencontainers#3126

Co-authored-by: Aditya R <arajan@redhat.com>
Signed-off-by: Zheao.Li <me@manjusaka.me>

docs/spec-conformance.md

@@ -7,7 +7,6 @@ The following features are not implemented yet:
 
 Spec version | Feature                                  | PR
 -------------|------------------------------------------|----------------------------------------------------------
-v1.0.2       | `.linux.personality`                     | [#3126](https://github.com/opencontainers/runc/pull/3126)
 v1.1.0       | `SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV` | [#3862](https://github.com/opencontainers/runc/pull/3862)
 v1.1.0       | rsvd hugetlb cgroup                      | TODO ([#3859](https://github.com/opencontainers/runc/issues/3859))
 v1.1.0       | `.process.ioPriority`                    | [#3783](https://github.com/opencontainers/runc/pull/3783)

libcontainer/configs/config.go

@@ -219,6 +219,9 @@ type Config struct {
 
 	// Scheduler represents the scheduling attributes for a process.
 	Scheduler *Scheduler `json:"scheduler,omitempty"`
+
+	// Personality contains configuration for the Linux personality syscall.
+	Personality *LinuxPersonality `json:"personality,omitempty"`
 }
 
 // Scheduler is based on the Linux sched_setattr(2) syscall.

libcontainer/configs/config_linux.go

@@ -9,6 +9,19 @@ var (
 	errNoGroupMap = errors.New("User namespaces enabled, but no group mapping found.")
 )
 
+// Please check https://man7.org/linux/man-pages/man2/personality.2.html for const details.
+// https://raw.githubusercontent.com/torvalds/linux/master/include/uapi/linux/personality.h
+const (
+	PerLinux   = 0x0000
+	PerLinux32 = 0x0008
+)
+
+type LinuxPersonality struct {
+	// Domain for the personality
+	// can only contain values "LINUX" and "LINUX32"
+	Domain int `json:"domain"`
+}
+
 // HostUID gets the translated uid for the process on host which could be
 // different when user namespaces are enabled.
 func (c Config) HostUID(containerId int) (int, error) {

libcontainer/init_linux.go

@@ -654,6 +654,10 @@ func setupScheduler(config *configs.Config) error {
 	return nil
 }
 
+func setupPersonality(config *configs.Config) error {
+	return system.SetLinuxPersonality(config.Personality.Domain)
+}
+
 // signalAllProcesses freezes then iterates over all the processes inside the
 // manager's cgroups sending the signal s to them.
 func signalAllProcesses(m cgroups.Manager, s unix.Signal) error {

libcontainer/setns_init_linux.go

@@ -93,6 +93,11 @@ func (l *linuxSetnsInit) Init() error {
 	if err := apparmor.ApplyProfile(l.config.AppArmorProfile); err != nil {
 		return err
 	}
+	if l.config.Config.Personality != nil {
+		if err := setupPersonality(l.config.Config); err != nil {
+			return err
+		}
+	}
 	// Check for the arg early to make sure it exists.
 	name, err := exec.LookPath(l.config.Args[0])
 	if err != nil {

libcontainer/specconv/spec_linux.go

@@ -434,6 +434,18 @@ func CreateLibcontainerConfig(opts *CreateOpts) (*configs.Config, error) {
 				MemBwSchema:   spec.Linux.IntelRdt.MemBwSchema,
 			}
 		}
+		if spec.Linux.Personality != nil {
+			if len(spec.Linux.Personality.Flags) > 0 {
+				logrus.Warnf("ignoring unsupported personality flags: %+v because personality flag has not supported at this time", spec.Linux.Personality.Flags)
+			}
+			domain, err := getLinuxPersonalityFromStr(string(spec.Linux.Personality.Domain))
+			if err != nil {
+				return nil, err
+			}
+			config.Personality = &configs.LinuxPersonality{
+				Domain: domain,
+			}
+		}
 	}
 
 	// Set the host UID that should own the container's cgroup.
@@ -571,6 +583,16 @@ func checkPropertyName(s string) error {
 	return nil
 }
 
+// getLinuxPersonalityFromStr converts the string domain received from spec to equivalent integer.
+func getLinuxPersonalityFromStr(domain string) (int, error) {
+	if domain == string(specs.PerLinux32) {
+		return configs.PerLinux32, nil
+	} else if domain == string(specs.PerLinux) {
+		return configs.PerLinux, nil
+	}
+	return -1, fmt.Errorf("invalid personality domain %s", domain)
+}
+
 // Some systemd properties are documented as having "Sec" suffix
 // (e.g. TimeoutStopSec) but are expected to have "USec" suffix
 // here, so let's provide conversion to improve compatibility.

libcontainer/standard_init_linux.go

@@ -233,6 +233,14 @@ func (l *linuxStandardInit) Init() error {
 			return err
 		}
 	}
+
+	// Set personality if specified.
+	if l.config.Config.Personality != nil {
+		if err := setupPersonality(l.config.Config); err != nil {
+			return err
+		}
+	}
+
 	// Close the pipe to signal that we have completed our init.
 	logrus.Debugf("init: closing the pipe to signal completion")
 	_ = l.pipe.Close()

libcontainer/system/linux.go

@@ -214,3 +214,13 @@ func Copy(dst io.Writer, src io.Reader) (copied int64, err error) {
 fallback:
 	return io.Copy(dst, src)
 }
+
+// SetLinuxPersonality sets the Linux execution personality. For more information see the personality syscall documentation.
+// checkout getLinuxPersonalityFromStr() from libcontainer/specconv/spec_linux.go for type conversion.
+func SetLinuxPersonality(personality int) error {
+	_, _, errno := syscall.Syscall(syscall.SYS_PERSONALITY, uintptr(personality), 0, 0)
+	if errno != 0 {
+		return &os.SyscallError{Syscall: "set_personality", Err: errno}
+	}
+	return nil
+}

tests/integration/personality.bats

@@ -0,0 +1,64 @@
+#!/usr/bin/env bats
+
+load helpers
+
+function setup() {
+	requires arch_x86_64
+	setup_busybox
+}
+
+function teardown() {
+	teardown_bundle
+}
+
+@test "runc run personality for i686" {
+	update_config '
+      .process.args = ["/bin/sh", "-c", "uname -a"]
+			| .linux.personality = {
+                "domain": "LINUX32",
+                "flags": []
+			}'
+
+	runc run test_busybox
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"i686"* ]]
+}
+
+@test "runc run personality with exec for i686" {
+	update_config '
+      .linux.personality = {
+                "domain": "LINUX32",
+      }'
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_busybox
+	[ "$status" -eq 0 ]
+	runc exec test_busybox /bin/sh -c "uname -a"
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"i686"* ]]
+}
+
+@test "runc run personality for x86_64" {
+	update_config '
+      .process.args = ["/bin/sh", "-c", "uname -a"]
+			| .linux.personality = {
+                "domain": "LINUX",
+                "flags": []
+			}'
+
+	runc run test_busybox
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"x86_64"* ]]
+}
+
+@test "runc run personality with exec for x86_64" {
+	update_config '
+      .linux.personality = {
+                "domain": "LINUX",
+      }'
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_busybox
+	[ "$status" -eq 0 ]
+	runc exec test_busybox /bin/sh -c "uname -a"
+	[ "$status" -eq 0 ]
+	[[ "$output" == *"x86_64"* ]]
+}