cri: implement RuntimeConfig rpc

The rpc only reports one field, i.e. the cgroup driver, to kubelet.
Containerd determines the effective cgroup driver by looking at all
runtime handlers, starting from the default runtime handler (the rest in
alphabetical order), and returning the cgroup driver setting of the
first runtime handler that supports one. If no runtime handler supports
cgroup driver (i.e. has a config option for it) containerd falls back to
auto-detection, returning systemd if systemd is running and cgroupfs
otherwise.

This patch implements the CRI server side of Kubernetes KEP-4033:
https://github.com/kubernetes/enhancements/tree/master/keps/sig-node/4033-group-driver-detection-over-cri

Signed-off-by: Markus Lehtonen <markus.lehtonen@intel.com>

vendor/github.com/opencontainers/runc/libcontainer/utils/cmsg.go

@@ -0,0 +1,96 @@
+package utils
+
+/*
+ * Copyright 2016, 2017 SUSE LLC
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+import (
+	"fmt"
+	"os"
+
+	"golang.org/x/sys/unix"
+)
+
+// MaxSendfdLen is the maximum length of the name of a file descriptor being
+// sent using SendFd. The name of the file handle returned by RecvFd will never
+// be larger than this value.
+const MaxNameLen = 4096
+
+// oobSpace is the size of the oob slice required to store a single FD. Note
+// that unix.UnixRights appears to make the assumption that fd is always int32,
+// so sizeof(fd) = 4.
+var oobSpace = unix.CmsgSpace(4)
+
+// RecvFd waits for a file descriptor to be sent over the given AF_UNIX
+// socket. The file name of the remote file descriptor will be recreated
+// locally (it is sent as non-auxiliary data in the same payload).
+func RecvFd(socket *os.File) (*os.File, error) {
+	// For some reason, unix.Recvmsg uses the length rather than the capacity
+	// when passing the msg_controllen and other attributes to recvmsg.  So we
+	// have to actually set the length.
+	name := make([]byte, MaxNameLen)
+	oob := make([]byte, oobSpace)
+
+	sockfd := socket.Fd()
+	n, oobn, _, _, err := unix.Recvmsg(int(sockfd), name, oob, 0)
+	if err != nil {
+		return nil, err
+	}
+
+	if n >= MaxNameLen || oobn != oobSpace {
+		return nil, fmt.Errorf("recvfd: incorrect number of bytes read (n=%d oobn=%d)", n, oobn)
+	}
+
+	// Truncate.
+	name = name[:n]
+	oob = oob[:oobn]
+
+	scms, err := unix.ParseSocketControlMessage(oob)
+	if err != nil {
+		return nil, err
+	}
+	if len(scms) != 1 {
+		return nil, fmt.Errorf("recvfd: number of SCMs is not 1: %d", len(scms))
+	}
+	scm := scms[0]
+
+	fds, err := unix.ParseUnixRights(&scm)
+	if err != nil {
+		return nil, err
+	}
+	if len(fds) != 1 {
+		return nil, fmt.Errorf("recvfd: number of fds is not 1: %d", len(fds))
+	}
+	fd := uintptr(fds[0])
+
+	return os.NewFile(fd, string(name)), nil
+}
+
+// SendFd sends a file descriptor over the given AF_UNIX socket. In
+// addition, the file.Name() of the given file will also be sent as
+// non-auxiliary data in the same payload (allowing to send contextual
+// information for a file descriptor).
+func SendFd(socket *os.File, name string, fd uintptr) error {
+	if len(name) >= MaxNameLen {
+		return fmt.Errorf("sendfd: filename too long: %s", name)
+	}
+	return SendFds(socket, []byte(name), int(fd))
+}
+
+// SendFds sends a list of files descriptor and msg over the given AF_UNIX socket.
+func SendFds(socket *os.File, msg []byte, fds ...int) error {
+	oob := unix.UnixRights(fds...)
+	return unix.Sendmsg(int(socket.Fd()), msg, oob, nil, 0)
+}

vendor/github.com/opencontainers/runc/libcontainer/utils/utils.go

@@ -0,0 +1,167 @@
+package utils
+
+import (
+	"encoding/binary"
+	"encoding/json"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"unsafe"
+
+	securejoin "github.com/cyphar/filepath-securejoin"
+	"golang.org/x/sys/unix"
+)
+
+const (
+	exitSignalOffset = 128
+)
+
+// NativeEndian is the native byte order of the host system.
+var NativeEndian binary.ByteOrder
+
+func init() {
+	// Copied from <golang.org/x/net/internal/socket/sys.go>.
+	i := uint32(1)
+	b := (*[4]byte)(unsafe.Pointer(&i))
+	if b[0] == 1 {
+		NativeEndian = binary.LittleEndian
+	} else {
+		NativeEndian = binary.BigEndian
+	}
+}
+
+// ExitStatus returns the correct exit status for a process based on if it
+// was signaled or exited cleanly
+func ExitStatus(status unix.WaitStatus) int {
+	if status.Signaled() {
+		return exitSignalOffset + int(status.Signal())
+	}
+	return status.ExitStatus()
+}
+
+// WriteJSON writes the provided struct v to w using standard json marshaling
+func WriteJSON(w io.Writer, v interface{}) error {
+	data, err := json.Marshal(v)
+	if err != nil {
+		return err
+	}
+	_, err = w.Write(data)
+	return err
+}
+
+// CleanPath makes a path safe for use with filepath.Join. This is done by not
+// only cleaning the path, but also (if the path is relative) adding a leading
+// '/' and cleaning it (then removing the leading '/'). This ensures that a
+// path resulting from prepending another path will always resolve to lexically
+// be a subdirectory of the prefixed path. This is all done lexically, so paths
+// that include symlinks won't be safe as a result of using CleanPath.
+func CleanPath(path string) string {
+	// Deal with empty strings nicely.
+	if path == "" {
+		return ""
+	}
+
+	// Ensure that all paths are cleaned (especially problematic ones like
+	// "/../../../../../" which can cause lots of issues).
+	path = filepath.Clean(path)
+
+	// If the path isn't absolute, we need to do more processing to fix paths
+	// such as "../../../../<etc>/some/path". We also shouldn't convert absolute
+	// paths to relative ones.
+	if !filepath.IsAbs(path) {
+		path = filepath.Clean(string(os.PathSeparator) + path)
+		// This can't fail, as (by definition) all paths are relative to root.
+		path, _ = filepath.Rel(string(os.PathSeparator), path)
+	}
+
+	// Clean the path again for good measure.
+	return filepath.Clean(path)
+}
+
+// stripRoot returns the passed path, stripping the root path if it was
+// (lexicially) inside it. Note that both passed paths will always be treated
+// as absolute, and the returned path will also always be absolute. In
+// addition, the paths are cleaned before stripping the root.
+func stripRoot(root, path string) string {
+	// Make the paths clean and absolute.
+	root, path = CleanPath("/"+root), CleanPath("/"+path)
+	switch {
+	case path == root:
+		path = "/"
+	case root == "/":
+		// do nothing
+	case strings.HasPrefix(path, root+"/"):
+		path = strings.TrimPrefix(path, root+"/")
+	}
+	return CleanPath("/" + path)
+}
+
+// WithProcfd runs the passed closure with a procfd path (/proc/self/fd/...)
+// corresponding to the unsafePath resolved within the root. Before passing the
+// fd, this path is verified to have been inside the root -- so operating on it
+// through the passed fdpath should be safe. Do not access this path through
+// the original path strings, and do not attempt to use the pathname outside of
+// the passed closure (the file handle will be freed once the closure returns).
+func WithProcfd(root, unsafePath string, fn func(procfd string) error) error {
+	// Remove the root then forcefully resolve inside the root.
+	unsafePath = stripRoot(root, unsafePath)
+	path, err := securejoin.SecureJoin(root, unsafePath)
+	if err != nil {
+		return fmt.Errorf("resolving path inside rootfs failed: %w", err)
+	}
+
+	// Open the target path.
+	fh, err := os.OpenFile(path, unix.O_PATH|unix.O_CLOEXEC, 0)
+	if err != nil {
+		return fmt.Errorf("open o_path procfd: %w", err)
+	}
+	defer fh.Close()
+
+	// Double-check the path is the one we expected.
+	procfd := "/proc/self/fd/" + strconv.Itoa(int(fh.Fd()))
+	if realpath, err := os.Readlink(procfd); err != nil {
+		return fmt.Errorf("procfd verification failed: %w", err)
+	} else if realpath != path {
+		return fmt.Errorf("possibly malicious path detected -- refusing to operate on %s", realpath)
+	}
+
+	// Run the closure.
+	return fn(procfd)
+}
+
+// SearchLabels searches a list of key-value pairs for the provided key and
+// returns the corresponding value. The pairs must be separated with '='.
+func SearchLabels(labels []string, query string) string {
+	for _, l := range labels {
+		parts := strings.SplitN(l, "=", 2)
+		if len(parts) < 2 {
+			continue
+		}
+		if parts[0] == query {
+			return parts[1]
+		}
+	}
+	return ""
+}
+
+// Annotations returns the bundle path and user defined annotations from the
+// libcontainer state.  We need to remove the bundle because that is a label
+// added by libcontainer.
+func Annotations(labels []string) (bundle string, userAnnotations map[string]string) {
+	userAnnotations = make(map[string]string)
+	for _, l := range labels {
+		parts := strings.SplitN(l, "=", 2)
+		if len(parts) < 2 {
+			continue
+		}
+		if parts[0] == "bundle" {
+			bundle = parts[1]
+		} else {
+			userAnnotations[parts[0]] = parts[1]
+		}
+	}
+	return
+}

vendor/github.com/opencontainers/runc/libcontainer/utils/utils_unix.go

@@ -0,0 +1,69 @@
+//go:build !windows
+// +build !windows
+
+package utils
+
+import (
+	"fmt"
+	"os"
+	"strconv"
+
+	"golang.org/x/sys/unix"
+)
+
+// EnsureProcHandle returns whether or not the given file handle is on procfs.
+func EnsureProcHandle(fh *os.File) error {
+	var buf unix.Statfs_t
+	if err := unix.Fstatfs(int(fh.Fd()), &buf); err != nil {
+		return fmt.Errorf("ensure %s is on procfs: %w", fh.Name(), err)
+	}
+	if buf.Type != unix.PROC_SUPER_MAGIC {
+		return fmt.Errorf("%s is not on procfs", fh.Name())
+	}
+	return nil
+}
+
+// CloseExecFrom applies O_CLOEXEC to all file descriptors currently open for
+// the process (except for those below the given fd value).
+func CloseExecFrom(minFd int) error {
+	fdDir, err := os.Open("/proc/self/fd")
+	if err != nil {
+		return err
+	}
+	defer fdDir.Close()
+
+	if err := EnsureProcHandle(fdDir); err != nil {
+		return err
+	}
+
+	fdList, err := fdDir.Readdirnames(-1)
+	if err != nil {
+		return err
+	}
+	for _, fdStr := range fdList {
+		fd, err := strconv.Atoi(fdStr)
+		// Ignore non-numeric file names.
+		if err != nil {
+			continue
+		}
+		// Ignore descriptors lower than our specified minimum.
+		if fd < minFd {
+			continue
+		}
+		// Intentionally ignore errors from unix.CloseOnExec -- the cases where
+		// this might fail are basically file descriptors that have already
+		// been closed (including and especially the one that was created when
+		// os.ReadDir did the "opendir" syscall).
+		unix.CloseOnExec(fd)
+	}
+	return nil
+}
+
+// NewSockPair returns a new unix socket pair
+func NewSockPair(name string) (parent *os.File, child *os.File, err error) {
+	fds, err := unix.Socketpair(unix.AF_LOCAL, unix.SOCK_STREAM|unix.SOCK_CLOEXEC, 0)
+	if err != nil {
+		return nil, nil, err
+	}
+	return os.NewFile(uintptr(fds[1]), name+"-p"), os.NewFile(uintptr(fds[0]), name+"-c"), nil
+}