refactor: move some c code to go

Move all the stage-0 c code and some of the stage-2
c code to go code, because they are not related to
namespaces, and could be implemented by golang.

Signed-off-by: lifubang <[email protected]>

Author: lifubang

libcontainer/configs/config.go

@@ -27,6 +27,11 @@ type IDMap struct {
 	Size        int64 `json:"size"`
 }
 
+// ToString is to serialize the IDMap to a string.
+func (i IDMap) ToString() string {
+	return fmt.Sprintf("%d %d %d", i.ContainerID, i.HostID, i.Size)
+}
+
 // Seccomp represents syscall restrictions
 // By default, only the native architecture of the kernel is allowed to be used
 // for syscalls. Additional architectures can be added by specifying them in

libcontainer/container_linux.go

@@ -536,6 +536,10 @@ func (c *Container) newParentProcess(p *Process) (parentProcess, error) {
 	cmd.Env = append(cmd.Env,
 		"_LIBCONTAINER_INITPIPE="+strconv.Itoa(stdioFdCount+len(cmd.ExtraFiles)-1),
 	)
+	cmd.ExtraFiles = append(cmd.ExtraFiles, comm.stage1SockChild)
+	cmd.Env = append(cmd.Env,
+		"_LIBCONTAINER_STAGE1PIPE="+strconv.Itoa(stdioFdCount+len(cmd.ExtraFiles)-1),
+	)
 	cmd.ExtraFiles = append(cmd.ExtraFiles, comm.syncSockChild.File())
 	cmd.Env = append(cmd.Env,
 		"_LIBCONTAINER_SYNCPIPE="+strconv.Itoa(stdioFdCount+len(cmd.ExtraFiles)-1),
@@ -1022,17 +1026,6 @@ func (c *Container) orderNamespacePaths(namespaces map[configs.NamespaceType]str
 	return paths, nil
 }
 
-func encodeIDMapping(idMap []configs.IDMap) ([]byte, error) {
-	data := bytes.NewBuffer(nil)
-	for _, im := range idMap {
-		line := fmt.Sprintf("%d %d %d\n", im.ContainerID, im.HostID, im.Size)
-		if _, err := data.WriteString(line); err != nil {
-			return nil, err
-		}
-	}
-	return data.Bytes(), nil
-}
-
 // netlinkError is an error wrapper type for use by custom netlink message
 // types. Panics with errors are wrapped in netlinkError so that the recover
 // in bootstrapData can distinguish intentional panics.
@@ -1079,59 +1072,6 @@ func (c *Container) bootstrapData(cloneFlags uintptr, nsMaps map[configs.Namespa
 		})
 	}
 
-	// write namespace paths only when we are not joining an existing user ns
-	_, joinExistingUser := nsMaps[configs.NEWUSER]
-	if !joinExistingUser {
-		// write uid mappings
-		if len(c.config.UIDMappings) > 0 {
-			if c.config.RootlessEUID {
-				// We resolve the paths for new{u,g}idmap from
-				// the context of runc to avoid doing a path
-				// lookup in the nsexec context.
-				if path, err := exec.LookPath("newuidmap"); err == nil {
-					r.AddData(&Bytemsg{
-						Type:  UidmapPathAttr,
-						Value: []byte(path),
-					})
-				}
-			}
-			b, err := encodeIDMapping(c.config.UIDMappings)
-			if err != nil {
-				return nil, err
-			}
-			r.AddData(&Bytemsg{
-				Type:  UidmapAttr,
-				Value: b,
-			})
-		}
-
-		// write gid mappings
-		if len(c.config.GIDMappings) > 0 {
-			b, err := encodeIDMapping(c.config.GIDMappings)
-			if err != nil {
-				return nil, err
-			}
-			r.AddData(&Bytemsg{
-				Type:  GidmapAttr,
-				Value: b,
-			})
-			if c.config.RootlessEUID {
-				if path, err := exec.LookPath("newgidmap"); err == nil {
-					r.AddData(&Bytemsg{
-						Type:  GidmapPathAttr,
-						Value: []byte(path),
-					})
-				}
-			}
-			if requiresRootOrMappingTool(c.config) {
-				r.AddData(&Boolmsg{
-					Type:  SetgroupAttr,
-					Value: true,
-				})
-			}
-		}
-	}
-
 	if c.config.OomScoreAdj != nil {
 		// write oom_score_adj
 		r.AddData(&Bytemsg{
@@ -1140,12 +1080,6 @@ func (c *Container) bootstrapData(cloneFlags uintptr, nsMaps map[configs.Namespa
 		})
 	}
 
-	// write rootless
-	r.AddData(&Boolmsg{
-		Type:  RootlessEUIDAttr,
-		Value: c.config.RootlessEUID,
-	})
-
 	// write boottime and monotonic time ns offsets.
 	if c.config.TimeOffsets != nil {
 		var offsetSpec bytes.Buffer
@@ -1186,9 +1120,9 @@ func ignoreTerminateErrors(err error) error {
 	return err
 }
 
-func requiresRootOrMappingTool(c *configs.Config) bool {
+func requiresRootOrMappingTool(gidMappings []configs.IDMap) bool {
 	gidMap := []configs.IDMap{
 		{ContainerID: 0, HostID: int64(os.Getegid()), Size: 1},
 	}
-	return !reflect.DeepEqual(c.GIDMappings, gidMap)
+	return !reflect.DeepEqual(gidMappings, gidMap)
 }

libcontainer/container_setup_linux.go

@@ -0,0 +1,147 @@
+package libcontainer
+
+import (
+	"encoding/binary"
+	"fmt"
+	"io"
+	"os"
+
+	"github.com/opencontainers/runc/libcontainer/configs"
+	"github.com/opencontainers/runc/libcontainer/system"
+	"github.com/sirupsen/logrus"
+	"github.com/vishvananda/netlink/nl"
+	"golang.org/x/sys/execabs"
+	"golang.org/x/sys/unix"
+)
+
+// NsExecSyncMsg is used for communication between the parent and child during
+// container setup.
+type NsExecSyncMsg uint32
+
+const (
+	SyncUsermapPls NsExecSyncMsg = iota + 0x40
+	SyncUsermapAck
+	SyncRecvPidPls
+	SyncRecvPidAck
+	SyncTimeOffsetsPls
+	SyncTimeOffsetsAck
+)
+
+const bufSize = 4
+
+// setupNsExec is used to help nsexec to setup the container and wait the container's pid.
+func (s *containerProcess) setupNsExec(syncSock *os.File) error {
+	logrus.Debugf("waiting nsexec to report the container's pid")
+	err := ParseNsExecSync(syncSock, func(msg NsExecSyncMsg) error {
+		switch msg {
+		case SyncUsermapPls:
+			logrus.Debugf("nsexec has requested userns mappings")
+			if err := s.setupUsermap(); err != nil {
+				return err
+			}
+			return AckNsExecSync(syncSock, SyncUsermapAck)
+		case SyncTimeOffsetsPls:
+			logrus.Debugf("nsexec has requested to configure timens offsets")
+			if err := system.UpdateTimeNsOffsets(s.cmd.Process.Pid, s.container.config.TimeOffsets); err != nil {
+				return err
+			}
+			return AckNsExecSync(syncSock, SyncTimeOffsetsAck)
+		case SyncRecvPidPls:
+			logrus.Debugf("nsexec has reported pid")
+			var pid uint32
+			if err := binary.Read(syncSock, nl.NativeEndian(), &pid); err != nil {
+				return err
+			}
+			s.childPid = int(pid)
+			return AckNsExecSync(syncSock, SyncRecvPidAck)
+		default:
+			return fmt.Errorf("unexpected message %d", msg)
+		}
+	})
+
+	return err
+}
+
+// ParseNsExecSync runs the given callback function on each message received
+// from the child. It will return once the child sends SYNC_RECVPID_PLS.
+func ParseNsExecSync(r io.Reader, fn func(NsExecSyncMsg) error) error {
+	var (
+		msg NsExecSyncMsg
+		buf [bufSize]byte
+	)
+
+	native := nl.NativeEndian()
+
+	for {
+		if _, err := io.ReadAtLeast(r, buf[:], bufSize); err != nil {
+			return err
+		}
+		msg = NsExecSyncMsg(native.Uint32(buf[:]))
+		if err := fn(msg); err != nil {
+			return err
+		}
+		if msg == SyncRecvPidPls {
+			break
+		}
+	}
+	return nil
+}
+
+// AckNsExecSync is used to send a message to the child.
+func AckNsExecSync(f *os.File, msg NsExecSyncMsg) error {
+	var buf [bufSize]byte
+	native := nl.NativeEndian()
+	native.PutUint32(buf[:], uint32(msg))
+	if _, err := unix.Write(int(f.Fd()), buf[:]); err != nil {
+		logrus.Debugf("failed to write message to nsexec: %v", err)
+		return err
+	}
+	return nil
+}
+
+// setupUsermap is used to set up the user mappings.
+func (s *containerProcess) setupUsermap() error {
+	var uidMapPath, gidMapPath string
+
+	// Enable setgroups(2) if we've been asked to. But we also have to explicitly
+	// disable setgroups(2) if we're creating a rootless container for single-entry
+	// mapping. (this is required since Linux 3.19).
+	// For rootless multi-entry mapping, we should use newuidmap/newgidmap
+	// to do mapping user namespace.
+	if s.config.Config.RootlessEUID && !requiresRootOrMappingTool(s.config.Config.GIDMappings) {
+		_ = system.UpdateSetgroups(s.cmd.Process.Pid, system.SetgroupsDeny)
+	}
+
+	nsMaps := make(map[configs.NamespaceType]string)
+	for _, ns := range s.container.config.Namespaces {
+		if ns.Path != "" {
+			nsMaps[ns.Type] = ns.Path
+		}
+	}
+	_, joinExistingUser := nsMaps[configs.NEWUSER]
+	if !joinExistingUser {
+		// write uid mappings
+		if len(s.container.config.UIDMappings) > 0 {
+			if s.container.config.RootlessEUID {
+				if path, err := execabs.LookPath("newuidmap"); err == nil {
+					uidMapPath = path
+				}
+			}
+		}
+
+		// write gid mappings
+		if len(s.container.config.GIDMappings) > 0 {
+			if s.container.config.RootlessEUID {
+				if path, err := execabs.LookPath("newgidmap"); err == nil {
+					gidMapPath = path
+				}
+			}
+		}
+	}
+
+	/* Set up mappings. */
+	if err := system.UpdateUidmap(uidMapPath, s.cmd.Process.Pid, s.container.config.UIDMappings); err != nil {
+		return err
+	}
+	return system.UpdateGidmap(gidMapPath, s.cmd.Process.Pid, s.container.config.GIDMappings)
+}

libcontainer/init_linux.go

@@ -12,6 +12,7 @@ import (
 	"runtime/debug"
 	"strconv"
 	"syscall"
+	"unsafe"
 
 	"github.com/containerd/console"
 	"github.com/opencontainers/runtime-spec/specs-go"
@@ -33,11 +34,6 @@ const (
 	initStandard initType = "standard"
 )
 
-type pid struct {
-	Pid           int `json:"stage2_pid"`
-	PidFirstChild int `json:"stage1_pid"`
-}
-
 // network is an internal struct used to setup container networks.
 type network struct {
 	configs.Network
@@ -174,6 +170,11 @@ func startInitialization() (retErr error) {
 
 	logrus.SetOutput(logPipe)
 	logrus.SetFormatter(new(logrus.JSONFormatter))
+
+	/* For debugging. */
+	procName := append([]byte("runc:[2:INIT]"), 0)
+	_ = unix.Prctl(unix.PR_SET_NAME, uintptr(unsafe.Pointer(&procName[0])), 0, 0, 0)
+
 	logrus.Debug("child process in init()")
 
 	// Only init processes have FIFOFD.
@@ -227,6 +228,24 @@ func startInitialization() (retErr error) {
 		return err
 	}
 
+	if _, err := unix.Setsid(); err != nil {
+		return os.NewSyscallError("setsid", err)
+	}
+
+	if err := unix.Setuid(0); err != nil {
+		return os.NewSyscallError("setuid", err)
+	}
+
+	if err := unix.Setgid(0); err != nil {
+		return os.NewSyscallError("setgid", err)
+	}
+
+	if !config.Config.RootlessEUID && requiresRootOrMappingTool(config.Config.GIDMappings) {
+		if err := unix.Setgroups([]int{0}); err != nil {
+			return os.NewSyscallError("setgroups", err)
+		}
+	}
+
 	// If init succeeds, it will not return, hence none of the defers will be called.
 	return containerInit(it, &config, syncPipe, consoleSocket, pidfdSocket, fifoFile, logPipe)
 }

libcontainer/message_linux.go

@@ -11,17 +11,11 @@ import (
 // list of known message types we want to send to bootstrap program
 // The number is randomly chosen to not conflict with known netlink types
 const (
-	InitMsg          uint16 = 62000
-	CloneFlagsAttr   uint16 = 27281
-	NsPathsAttr      uint16 = 27282
-	UidmapAttr       uint16 = 27283
-	GidmapAttr       uint16 = 27284
-	SetgroupAttr     uint16 = 27285
-	OomScoreAdjAttr  uint16 = 27286
-	RootlessEUIDAttr uint16 = 27287
-	UidmapPathAttr   uint16 = 27288
-	GidmapPathAttr   uint16 = 27289
-	TimeOffsetsAttr  uint16 = 27290
+	InitMsg         uint16 = 62000
+	CloneFlagsAttr  uint16 = 27281
+	NsPathsAttr     uint16 = 27282
+	OomScoreAdjAttr uint16 = 27286
+	TimeOffsetsAttr uint16 = 27290
 )
 
 type Int32msg struct {

libcontainer/nsenter/log.c

@@ -58,7 +58,7 @@ void write_log(int level, const char *format, ...)
 		if (stage == NULL)
 			goto out;
 	} else {
-		ret = asprintf(&stage, "nsexec-%d", current_stage);
+		ret = asprintf(&stage, "nsexec-%d", current_stage + 1);
 		if (ret < 0) {
 			stage = NULL;
 			goto out;