Support mount option "rro" (recursive read-only) using mount_setattr

The new mount option "rro" makes the mount point recursively read-only,
by calling `mount_setattr(2)` with `MOUNT_ATTR_RDONLY` and `AT_RECURSIVE`.
https://man7.org/linux/man-pages/man2/mount_setattr.2.html

Requires kernel >= 5.12.

The "rro" option string conforms to the proposal in util-linux/util-linux Issue 1501.

Fix issue 2823

Signed-off-by: Akihiro Suda <[email protected]>

libcontainer/configs/mount.go

@@ -6,6 +6,10 @@ const (
 	// EXT_COPYUP is a directive to copy up the contents of a directory when
 	// a tmpfs is mounted over it.
 	EXT_COPYUP = 1 << iota //nolint:golint // ignore "don't use ALL_CAPS" warning
+
+	// EXT_RRO enables recursive read-only (RRO) mount, with MOUNT_ATTR_RDONLY , AT_RECURSIVE .
+	// Requires kernel >= 5.12.
+	EXT_RRO //nolint:golint
 )
 
 type Mount struct {

libcontainer/rootfs_linux.go

@@ -21,6 +21,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/devices"
 	"github.com/opencontainers/runc/libcontainer/userns"
 	"github.com/opencontainers/runc/libcontainer/utils"
+	"github.com/opencontainers/runc/libcontainer/utils/syscallutil"
 	"github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/opencontainers/selinux/go-selinux/label"
 	"github.com/sirupsen/logrus"
@@ -474,7 +475,11 @@ func mountToRootfs(m *configs.Mount, c *mountConfig) error {
 				return err
 			}
 		}
-
+		if m.Extensions&configs.EXT_RRO == configs.EXT_RRO {
+			if err := setRRO(m, rootfs); err != nil {
+				return err
+			}
+		}
 		if m.Relabel != "" {
 			if err := label.Validate(m.Relabel); err != nil {
 				return err
@@ -1113,3 +1118,13 @@ func mountPropagate(m *configs.Mount, rootfs string, mountLabel string, mountFd
 	}
 	return nil
 }
+
+// setRRO sets mount attribute MOUNT_ATTR_RDONLY, with AT_RECURSIVE
+func setRRO(m *configs.Mount, rootfs string) error {
+	return utils.WithProcfd(rootfs, m.Destination, func(procfd string) error {
+		attr := syscallutil.MountAttr{
+			AttrSet: syscallutil.MOUNT_ATTR_RDONLY,
+		}
+		return syscallutil.MountSetattr(-1, procfd, syscallutil.AT_RECURSIVE, &attr)
+	})
+}

libcontainer/specconv/spec_linux.go

@@ -821,6 +821,9 @@ func parseMountOptions(options []string) (int, []int, string, int) {
 		flag  int
 	}{
 		"tmpcopyup": {false, configs.EXT_COPYUP},
+		// rro = recursive read-only.
+		// "rro" option string is being proposed to `mount(8)` too: https://github.com/util-linux/util-linux/issues/1501
+		"rro": {false, configs.EXT_RRO},
 	}
 	for _, o := range options {
 		// If the option does not exist in the flags table or the flag

libcontainer/utils/syscallutil/syscallutil.go

@@ -0,0 +1,56 @@
+// Package syscallutil provdes addenda to golang.org/x/sys/unix
+package syscallutil
+
+import (
+	"unsafe"
+
+	"golang.org/x/sys/unix"
+)
+
+// nolint
+const (
+	AT_EMPTY_PATH          = unix.AT_EMPTY_PATH
+	AT_RECURSIVE           = 0x8000 // https://github.com/torvalds/linux/blob/v5.12/include/uapi/linux/fcntl.h#L112
+	AT_SYMLINK_NOFOLLOW    = unix.AT_SYMLINK_NOFOLLOW
+	AT_NO_AUTOMOUNT        = unix.AT_NO_AUTOMOUNT
+	MOUNT_ATTR_RDONLY      = 0x00000001 // https://github.com/torvalds/linux/blob/v5.12/include/uapi/linux/mount.h#L113
+	MOUNT_ATTR_NOSUID      = 0x00000002
+	MOUNT_ATTR_NODEV       = 0x00000004
+	MOUNT_ATTR_NOEXEC      = 0x00000008
+	MOUNT_ATTR__ATIME      = 0x00000070
+	MOUNT_ATTR_RELATIME    = 0x00000000
+	MOUNT_ATTR_NOATIME     = 0x00000010
+	MOUNT_ATTR_STRICTATIME = 0x00000020
+	MOUNT_ATTR_NODIRATIME  = 0x00000080
+	MOUNT_ATTR_IDMAP       = 0x00100000
+	MOUNT_ATTR_SIZE_VER0   = 32 // https://github.com/torvalds/linux/blob/v5.12/include/uapi/linux/mount.h#L135
+)
+
+// MountAttr corresponds to struct mount_attr, version 0, appeared in kernel 5.12.
+// https://github.com/torvalds/linux/blob/v5.12/include/uapi/linux/mount.h#L124-L132
+type MountAttr struct {
+	AttrSet     uint64 // __u64 attr_set
+	AttrClr     uint64 // __u64 attr_clr
+	Propagation uint64 // __u64 propagation
+	UsernsFd    uint64 // __u64 userns_fd
+}
+
+// MountSetattr is a wrapper for mount_setattr(2).
+//
+// int syscall(SYS_mount_setattr, int dirfd, const char *pathname, unsigned int flags, struct mount_attr *attr, size_t size);
+//
+// Requires kernel >= 5.12.
+// https://man7.org/linux/man-pages/man2/mount_setattr.2.html
+func MountSetattr(dirfd int, pathname string, flags uint, attr *MountAttr) error {
+	pathnamePtr, err := unix.BytePtrFromString(pathname)
+	if err != nil {
+		return err
+	}
+	_, _, errno := unix.Syscall6(unix.SYS_MOUNT_SETATTR,
+		uintptr(dirfd), uintptr(unsafe.Pointer(pathnamePtr)), uintptr(flags),
+		uintptr(unsafe.Pointer(attr)), unsafe.Sizeof(*attr), 0)
+	if errno != 0 {
+		return errno
+	}
+	return nil
+}

tests/integration/mounts_recursive.bats

@@ -0,0 +1,78 @@
+#!/usr/bin/env bats
+
+load helpers
+
+TESTVOLUME="${BATS_RUN_TMPDIR}/mounts_recursive"
+
+function setup_volume() {
+	# requires root (in the current user namespace) to mount tmpfs outside runc
+	requires root
+
+	mkdir -p "${TESTVOLUME}"
+	mount -t tmpfs none "${TESTVOLUME}"
+	echo "foo" >"${TESTVOLUME}/foo"
+
+	mkdir "${TESTVOLUME}/subvol"
+	mount -t tmpfs none "${TESTVOLUME}/subvol"
+	echo "bar" >"${TESTVOLUME}/subvol/bar"
+}
+
+function teardown_volume() {
+	umount -R "${TESTVOLUME}"
+}
+
+function setup() {
+	setup_volume
+	setup_busybox
+}
+
+function teardown() {
+	teardown_volume
+	teardown_bundle
+}
+
+@test "runc run [rbind,ro mount is read-only but not recursively]" {
+	update_config ".mounts += [{source: \"${TESTVOLUME}\" , destination: \"/mnt\", options: [\"rbind\",\"ro\"]}]"
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_rbind_ro
+	[ "$status" -eq 0 ]
+
+	runc exec test_rbind_ro touch /mnt/foo
+	[ "$status" -eq 1 ]
+	[[ "${output}" == *"Read-only file system"* ]]
+
+	runc exec test_rbind_ro touch /mnt/subvol/bar
+	[ "$status" -eq 0 ]
+}
+
+@test "runc run [rbind,rro mount is recursively read-only]" {
+	requires_kernel 5.12
+	update_config ".mounts += [{source: \"${TESTVOLUME}\" , destination: \"/mnt\", options: [\"rbind\",\"rro\"]}]"
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_rbind_rro
+	[ "$status" -eq 0 ]
+
+	runc exec test_rbind_rro touch /mnt/foo
+	[ "$status" -eq 1 ]
+	[[ "${output}" == *"Read-only file system"* ]]
+
+	runc exec test_rbind_rro touch /mnt/subvol/bar
+	[ "$status" -eq 1 ]
+	[[ "${output}" == *"Read-only file system"* ]]
+}
+
+@test "runc run [rbind,ro,rro mount is recursively read-only too]" {
+	requires_kernel 5.12
+	update_config ".mounts += [{source: \"${TESTVOLUME}\" , destination: \"/mnt\", options: [\"rbind\",\"ro\",\"rro\"]}]"
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_rbind_ro_rro
+	[ "$status" -eq 0 ]
+
+	runc exec test_rbind_ro_rro touch /mnt/foo
+	[ "$status" -eq 1 ]
+	[[ "${output}" == *"Read-only file system"* ]]
+
+	runc exec test_rbind_ro_rro touch /mnt/subvol/bar
+	[ "$status" -eq 1 ]
+	[[ "${output}" == *"Read-only file system"* ]]
+}