[1.1 backport] Prohibit /proc and /sys to be symlinks

libcontainer/rootfs_linux.go

@@ -398,32 +398,43 @@ func doTmpfsCopyUp(m *configs.Mount, rootfs, mountLabel string) (Err error) {
 
 func mountToRootfs(m *configs.Mount, c *mountConfig) error {
 	rootfs := c.root
-	mountLabel := c.label
-	mountFd := c.fd
-	dest, err := securejoin.SecureJoin(rootfs, m.Destination)
-	if err != nil {
-		return err
-	}
 
+	// procfs and sysfs are special because we need to ensure they are actually
+	// mounted on a specific path in a container without any funny business.
 	switch m.Device {
 	case "proc", "sysfs":
 		// If the destination already exists and is not a directory, we bail
-		// out This is to avoid mounting through a symlink or similar -- which
+		// out. This is to avoid mounting through a symlink or similar -- which
 		// has been a "fun" attack scenario in the past.
 		// TODO: This won't be necessary once we switch to libpathrs and we can
 		//       stop all of these symlink-exchange attacks.
+		dest := filepath.Clean(m.Destination)
+		if !strings.HasPrefix(dest, rootfs) {
+			// Do not use securejoin as it resolves symlinks.
+			dest = filepath.Join(rootfs, dest)
+		}
 		if fi, err := os.Lstat(dest); err != nil {
 			if !os.IsNotExist(err) {
 				return err
 			}
-		} else if fi.Mode()&os.ModeDir == 0 {
+		} else if !fi.IsDir() {
 			return fmt.Errorf("filesystem %q must be mounted on ordinary directory", m.Device)
 		}
 		if err := os.MkdirAll(dest, 0o755); err != nil {
 			return err
 		}
-		// Selinux kernels do not support labeling of /proc or /sys
+		// Selinux kernels do not support labeling of /proc or /sys.
 		return mountPropagate(m, rootfs, "", nil)
+	}
+
+	mountLabel := c.label
+	mountFd := c.fd
+	dest, err := securejoin.SecureJoin(rootfs, m.Destination)
+	if err != nil {
+		return err
+	}
+
+	switch m.Device {
 	case "mqueue":
 		if err := os.MkdirAll(dest, 0o755); err != nil {
 			return err

tests/integration/mask.bats

@@ -56,3 +56,22 @@ function teardown() {
 	[ "$status" -eq 1 ]
 	[[ "${output}" == *"Operation not permitted"* ]]
 }
+
+@test "mask paths [prohibit symlink /proc]" {
+	ln -s /symlink rootfs/proc
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_busybox
+	[ "$status" -eq 1 ]
+	[[ "${output}" == *"must be mounted on ordinary directory"* ]]
+}
+
+@test "mask paths [prohibit symlink /sys]" {
+	# In rootless containers, /sys is a bind mount not a real sysfs.
+	requires root
+
+	ln -s /symlink rootfs/sys
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_busybox
+	[ "$status" -eq 1 ]
+	# On cgroup v1, this may fail before checking if /sys is a symlink,
+	# so we merely check that it fails, and do not check the exact error
+	# message like for /proc above.
+}