Vendor in files for runc/libcontainer

vndr was previously removing .go files in runc/libcontainer since they weren't used anywhere
in cri-o, but kpod stats will use them

Signed-off-by: Ryan Cole <rcyoalne@gmail.com>

vendor/github.com/opencontainers/runc/libcontainer/configs/validate/rootless.go

@@ -0,0 +1,117 @@
+package validate
+
+import (
+	"fmt"
+	"os"
+	"reflect"
+	"strings"
+
+	"github.com/opencontainers/runc/libcontainer/configs"
+)
+
+var (
+	geteuid = os.Geteuid
+	getegid = os.Getegid
+)
+
+func (v *ConfigValidator) rootless(config *configs.Config) error {
+	if err := rootlessMappings(config); err != nil {
+		return err
+	}
+	if err := rootlessMount(config); err != nil {
+		return err
+	}
+	// Currently, cgroups cannot effectively be used in rootless containers.
+	// The new cgroup namespace doesn't really help us either because it doesn't
+	// have nice interactions with the user namespace (we're working with upstream
+	// to fix this).
+	if err := rootlessCgroup(config); err != nil {
+		return err
+	}
+
+	// XXX: We currently can't verify the user config at all, because
+	//      configs.Config doesn't store the user-related configs. So this
+	//      has to be verified by setupUser() in init_linux.go.
+
+	return nil
+}
+
+func rootlessMappings(config *configs.Config) error {
+	rootuid, err := config.HostRootUID()
+	if err != nil {
+		return fmt.Errorf("failed to get root uid from uidMappings: %v", err)
+	}
+	if euid := geteuid(); euid != 0 {
+		if !config.Namespaces.Contains(configs.NEWUSER) {
+			return fmt.Errorf("rootless containers require user namespaces")
+		}
+		if rootuid != euid {
+			return fmt.Errorf("rootless containers cannot map container root to a different host user")
+		}
+	}
+
+	rootgid, err := config.HostRootGID()
+	if err != nil {
+		return fmt.Errorf("failed to get root gid from gidMappings: %v", err)
+	}
+
+	// Similar to the above test, we need to make sure that we aren't trying to
+	// map to a group ID that we don't have the right to be.
+	if rootgid != getegid() {
+		return fmt.Errorf("rootless containers cannot map container root to a different host group")
+	}
+
+	// We can only map one user and group inside a container (our own).
+	if len(config.UidMappings) != 1 || config.UidMappings[0].Size != 1 {
+		return fmt.Errorf("rootless containers cannot map more than one user")
+	}
+	if len(config.GidMappings) != 1 || config.GidMappings[0].Size != 1 {
+		return fmt.Errorf("rootless containers cannot map more than one group")
+	}
+
+	return nil
+}
+
+// cgroup verifies that the user isn't trying to set any cgroup limits or paths.
+func rootlessCgroup(config *configs.Config) error {
+	// Nothing set at all.
+	if config.Cgroups == nil || config.Cgroups.Resources == nil {
+		return nil
+	}
+
+	// Used for comparing to the zero value.
+	left := reflect.ValueOf(*config.Cgroups.Resources)
+	right := reflect.Zero(left.Type())
+
+	// This is all we need to do, since specconv won't add cgroup options in
+	// rootless mode.
+	if !reflect.DeepEqual(left.Interface(), right.Interface()) {
+		return fmt.Errorf("cannot specify resource limits in rootless container")
+	}
+
+	return nil
+}
+
+// mount verifies that the user isn't trying to set up any mounts they don't have
+// the rights to do. In addition, it makes sure that no mount has a `uid=` or
+// `gid=` option that doesn't resolve to root.
+func rootlessMount(config *configs.Config) error {
+	// XXX: We could whitelist allowed devices at this point, but I'm not
+	//      convinced that's a good idea. The kernel is the best arbiter of
+	//      access control.
+
+	for _, mount := range config.Mounts {
+		// Check that the options list doesn't contain any uid= or gid= entries
+		// that don't resolve to root.
+		for _, opt := range strings.Split(mount.Data, ",") {
+			if strings.HasPrefix(opt, "uid=") && opt != "uid=0" {
+				return fmt.Errorf("cannot specify uid= mount options in rootless containers where argument isn't 0")
+			}
+			if strings.HasPrefix(opt, "gid=") && opt != "gid=0" {
+				return fmt.Errorf("cannot specify gid= mount options in rootless containers where argument isn't 0")
+			}
+		}
+	}
+
+	return nil
+}

vendor/github.com/opencontainers/runc/libcontainer/configs/validate/validator.go

@@ -0,0 +1,195 @@
+package validate
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"github.com/opencontainers/runc/libcontainer/configs"
+	selinux "github.com/opencontainers/selinux/go-selinux"
+)
+
+type Validator interface {
+	Validate(*configs.Config) error
+}
+
+func New() Validator {
+	return &ConfigValidator{}
+}
+
+type ConfigValidator struct {
+}
+
+func (v *ConfigValidator) Validate(config *configs.Config) error {
+	if err := v.rootfs(config); err != nil {
+		return err
+	}
+	if err := v.network(config); err != nil {
+		return err
+	}
+	if err := v.hostname(config); err != nil {
+		return err
+	}
+	if err := v.security(config); err != nil {
+		return err
+	}
+	if err := v.usernamespace(config); err != nil {
+		return err
+	}
+	if err := v.sysctl(config); err != nil {
+		return err
+	}
+	if config.Rootless {
+		if err := v.rootless(config); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// rootfs validates if the rootfs is an absolute path and is not a symlink
+// to the container's root filesystem.
+func (v *ConfigValidator) rootfs(config *configs.Config) error {
+	if _, err := os.Stat(config.Rootfs); err != nil {
+		if os.IsNotExist(err) {
+			return fmt.Errorf("rootfs (%s) does not exist", config.Rootfs)
+		}
+		return err
+	}
+	cleaned, err := filepath.Abs(config.Rootfs)
+	if err != nil {
+		return err
+	}
+	if cleaned, err = filepath.EvalSymlinks(cleaned); err != nil {
+		return err
+	}
+	if filepath.Clean(config.Rootfs) != cleaned {
+		return fmt.Errorf("%s is not an absolute path or is a symlink", config.Rootfs)
+	}
+	return nil
+}
+
+func (v *ConfigValidator) network(config *configs.Config) error {
+	if !config.Namespaces.Contains(configs.NEWNET) {
+		if len(config.Networks) > 0 || len(config.Routes) > 0 {
+			return fmt.Errorf("unable to apply network settings without a private NET namespace")
+		}
+	}
+	return nil
+}
+
+func (v *ConfigValidator) hostname(config *configs.Config) error {
+	if config.Hostname != "" && !config.Namespaces.Contains(configs.NEWUTS) {
+		return fmt.Errorf("unable to set hostname without a private UTS namespace")
+	}
+	return nil
+}
+
+func (v *ConfigValidator) security(config *configs.Config) error {
+	// restrict sys without mount namespace
+	if (len(config.MaskPaths) > 0 || len(config.ReadonlyPaths) > 0) &&
+		!config.Namespaces.Contains(configs.NEWNS) {
+		return fmt.Errorf("unable to restrict sys entries without a private MNT namespace")
+	}
+	if config.ProcessLabel != "" && !selinux.GetEnabled() {
+		return fmt.Errorf("selinux label is specified in config, but selinux is disabled or not supported")
+	}
+
+	return nil
+}
+
+func (v *ConfigValidator) usernamespace(config *configs.Config) error {
+	if config.Namespaces.Contains(configs.NEWUSER) {
+		if _, err := os.Stat("/proc/self/ns/user"); os.IsNotExist(err) {
+			return fmt.Errorf("USER namespaces aren't enabled in the kernel")
+		}
+	} else {
+		if config.UidMappings != nil || config.GidMappings != nil {
+			return fmt.Errorf("User namespace mappings specified, but USER namespace isn't enabled in the config")
+		}
+	}
+	return nil
+}
+
+// sysctl validates that the specified sysctl keys are valid or not.
+// /proc/sys isn't completely namespaced and depending on which namespaces
+// are specified, a subset of sysctls are permitted.
+func (v *ConfigValidator) sysctl(config *configs.Config) error {
+	validSysctlMap := map[string]bool{
+		"kernel.msgmax":          true,
+		"kernel.msgmnb":          true,
+		"kernel.msgmni":          true,
+		"kernel.sem":             true,
+		"kernel.shmall":          true,
+		"kernel.shmmax":          true,
+		"kernel.shmmni":          true,
+		"kernel.shm_rmid_forced": true,
+	}
+
+	for s := range config.Sysctl {
+		if validSysctlMap[s] || strings.HasPrefix(s, "fs.mqueue.") {
+			if config.Namespaces.Contains(configs.NEWIPC) {
+				continue
+			} else {
+				return fmt.Errorf("sysctl %q is not allowed in the hosts ipc namespace", s)
+			}
+		}
+		if strings.HasPrefix(s, "net.") {
+			if config.Namespaces.Contains(configs.NEWNET) {
+				if path := config.Namespaces.PathOf(configs.NEWNET); path != "" {
+					if err := checkHostNs(s, path); err != nil {
+						return err
+					}
+				}
+				continue
+			} else {
+				return fmt.Errorf("sysctl %q is not allowed in the hosts network namespace", s)
+			}
+		}
+		return fmt.Errorf("sysctl %q is not in a separate kernel namespace", s)
+	}
+
+	return nil
+}
+
+func isSymbolicLink(path string) (bool, error) {
+	fi, err := os.Lstat(path)
+	if err != nil {
+		return false, err
+	}
+
+	return fi.Mode()&os.ModeSymlink == os.ModeSymlink, nil
+}
+
+// checkHostNs checks whether network sysctl is used in host namespace.
+func checkHostNs(sysctlConfig string, path string) error {
+	var currentProcessNetns = "/proc/self/ns/net"
+	// readlink on the current processes network namespace
+	destOfCurrentProcess, err := os.Readlink(currentProcessNetns)
+	if err != nil {
+		return fmt.Errorf("read soft link %q error", currentProcessNetns)
+	}
+
+	// First check if the provided path is a symbolic link
+	symLink, err := isSymbolicLink(path)
+	if err != nil {
+		return fmt.Errorf("could not check that %q is a symlink: %v", path, err)
+	}
+
+	if symLink == false {
+		// The provided namespace is not a symbolic link,
+		// it is not the host namespace.
+		return nil
+	}
+
+	// readlink on the path provided in the struct
+	destOfContainer, err := os.Readlink(path)
+	if err != nil {
+		return fmt.Errorf("read soft link %q error", path)
+	}
+	if destOfContainer == destOfCurrentProcess {
+		return fmt.Errorf("sysctl %q is not allowed in the hosts network namespace", sysctlConfig)
+	}
+	return nil
+}