merge branch 'pr-1125'

LGTMs: @hqhq @mrunalp
Closes #1125

libcontainer/configs/config.go

@@ -85,11 +85,6 @@ type Config struct {
 	// that the parent process dies.
 	ParentDeathSignal int `json:"parent_death_signal"`
 
-	// PivotDir allows a custom directory inside the container's root filesystem to be used as pivot, when NoPivotRoot is not set.
-	// When a custom PivotDir not set, a temporary dir inside the root filesystem will be used. The pivot dir needs to be writeable.
-	// This is required when using read only root filesystems. In these cases, a read/writeable path can be (bind) mounted somewhere inside the root filesystem to act as pivot.
-	PivotDir string `json:"pivot_dir"`
-
 	// Path to a directory containing the container's root filesystem.
 	Rootfs string `json:"rootfs"`
 

libcontainer/rootfs_linux.go

@@ -84,7 +84,7 @@ func setupRootfs(config *configs.Config, console *linuxConsole, pipe io.ReadWrit
 	if config.NoPivotRoot {
 		err = msMoveRoot(config.Rootfs)
 	} else {
-		err = pivotRoot(config.Rootfs, config.PivotDir)
+		err = pivotRoot(config.Rootfs)
 	}
 	if err != nil {
 		return newSystemErrorWithCause(err, "jailing process inside rootfs")
@@ -590,48 +590,66 @@ func setupPtmx(config *configs.Config, console *linuxConsole) error {
 	return nil
 }
 
-func pivotRoot(rootfs, pivotBaseDir string) (err error) {
+// pivotRoot will call pivot_root such that rootfs becomes the new root
-	if pivotBaseDir == "" {
+// filesystem, and everything else is cleaned up.
-		pivotBaseDir = "/"
+func pivotRoot(rootfs string) error {
-	}
+	// While the documentation may claim otherwise, pivot_root(".", ".") is
-	tmpDir := filepath.Join(rootfs, pivotBaseDir)
+	// actually valid. What this results in is / being the new root but
-	if err := os.MkdirAll(tmpDir, 0755); err != nil {
+	// /proc/self/cwd being the old root. Since we can play around with the cwd
-		return fmt.Errorf("can't create tmp dir %s, error %v", tmpDir, err)
+	// with pivot_root this allows us to pivot without creating directories in
-	}
+	// the rootfs. Shout-outs to the LXC developers for giving us this idea.
-	pivotDir, err := ioutil.TempDir(tmpDir, ".pivot_root")
+
+	oldroot, err := syscall.Open("/", syscall.O_DIRECTORY|syscall.O_RDONLY, 0)
 	if err != nil {
-		return fmt.Errorf("can't create pivot_root dir %s, error %v", pivotDir, err)
-	}
-	defer func() {
-		errVal := os.Remove(pivotDir)
-		if err == nil {
-			err = errVal
-		}
-	}()
-	if err := syscall.PivotRoot(rootfs, pivotDir); err != nil {
-		// Make the parent mount private
-		if err := rootfsParentMountPrivate(rootfs); err != nil {
 		return err
 	}
+	defer syscall.Close(oldroot)
+
+	newroot, err := syscall.Open(rootfs, syscall.O_DIRECTORY|syscall.O_RDONLY, 0)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(newroot)
+
+	// Change to the new root so that the pivot_root actually acts on it.
+	if err := syscall.Fchdir(newroot); err != nil {
+		return err
+	}
+
+	if err := syscall.PivotRoot(".", "."); err != nil {
+		// Make the parent mount private
+		if err := rootfsParentMountPrivate("."); err != nil {
+			return err
+		}
+
 		// Try again
-		if err := syscall.PivotRoot(rootfs, pivotDir); err != nil {
+		if err := syscall.PivotRoot(".", "."); err != nil {
 			return fmt.Errorf("pivot_root %s", err)
 		}
 	}
-	if err := syscall.Chdir("/"); err != nil {
-		return fmt.Errorf("chdir / %s", err)
-	}
-	// path to pivot dir now changed, update
-	pivotDir = filepath.Join(pivotBaseDir, filepath.Base(pivotDir))
 
-	// Make pivotDir rprivate to make sure any of the unmounts don't
+	// Currently our "." is oldroot (according to the current kernel code).
-	// propagate to parent.
+	// However, purely for safety, we will fchdir(oldroot) since there isn't
-	if err := syscall.Mount("", pivotDir, "", syscall.MS_PRIVATE|syscall.MS_REC, ""); err != nil {
+	// really any guarantee from the kernel what /proc/self/cwd will be after a
+	// pivot_root(2).
+
+	if err := syscall.Fchdir(oldroot); err != nil {
 		return err
 	}
 
-	if err := syscall.Unmount(pivotDir, syscall.MNT_DETACH); err != nil {
+	// Make oldroot rprivate to make sure our unmounts don't propogate to the
-		return fmt.Errorf("unmount pivot_root dir %s", err)
+	// host (and thus bork the machine).
+	if err := syscall.Mount("", ".", "", syscall.MS_PRIVATE|syscall.MS_REC, ""); err != nil {
+		return err
+	}
+	// Preform the unmount. MNT_DETACH allows us to unmount /proc/self/cwd.
+	if err := syscall.Unmount(".", syscall.MNT_DETACH); err != nil {
+		return err
+	}
+
+	// Switch back to our shiny new root.
+	if err := syscall.Chdir("/"); err != nil {
+		return fmt.Errorf("chdir / %s", err)
 	}
 	return nil
 }