carry opencontainers#3126: linux: Support setting execution domain vi…

…a linux personality

Signed-off-by: Zheao.Li <[email protected]>

libcontainer/configs/config.go

@@ -223,6 +223,9 @@ 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 (
+	PER_LINUX   = 0x0000
+	PER_LINUX32 = 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/specconv/spec_linux.go

@@ -436,6 +436,15 @@ func CreateLibcontainerConfig(opts *CreateOpts) (*configs.Config, error) {
 				MemBwSchema:   spec.Linux.IntelRdt.MemBwSchema,
 			}
 		}
+		if spec.Linux.Personality != nil {
+			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.
@@ -573,6 +582,17 @@ func checkPropertyName(s string) error {
 	return nil
 }
 
+// getLinuxPersonalityFromStr converts the string domain received from spec to equivalent integer.
+func getLinuxPersonalityFromStr(domain string) (int, error) {
+	// defaults to PER_LINUX
+	if domain == "LINUX32" {
+		return configs.PER_LINUX32, nil
+	} else if domain == "LINUX" {
+		return configs.PER_LINUX, 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/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(persona int) error {
+	_, _, errno := syscall.Syscall(syscall.SYS_PERSONALITY, uintptr(persona), 0, 0)
+	if errno != 0 {
+		return &os.SyscallError{Syscall: "set_personality", Err: errno}
+	}
+	return nil
+}