rootless: fix /sys/fs/cgroup mounts

It was found that rootless runc makes `/sys/fs/cgroup` writable in following conditons:

1. when runc is executed inside the user namespace, and the config.json does not specify the cgroup namespace to be unshared
   (e.g.., `(docker|podman|nerdctl) run --cgroupns=host`, with Rootless Docker/Podman/nerdctl)
2. or, when runc is executed outside the user namespace, and `/sys` is mounted with `rbind, ro`
   (e.g., `runc spec --rootless`; this condition is very rare)

A container may gain the write access to user-owned cgroup hierarchy `/sys/fs/cgroup/user.slice/...` on the host.
Other users's cgroup hierarchies are not affected.

To fix the issue, this commit does:
1. Remount `/sys/fs/cgroup` to apply `MS_RDONLY` when it is being bind-mounted
2. Mask `/sys/fs/cgroup` when the bind source is unavailable

Fix CVE-2023-25809 (GHSA-m8cg-xc2p-r3fc)

Co-authored-by: Kir Kolyshkin <kolyshkin@gmail.com>
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
(cherry picked from commit df4eae4)
Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>

libcontainer/rootfs_linux.go

@@ -329,26 +329,41 @@ func mountCgroupV2(m *configs.Mount, c *mountConfig) error {
 	if err := os.MkdirAll(dest, 0o755); err != nil {
 		return err
 	}
-	return utils.WithProcfd(c.root, m.Destination, func(procfd string) error {
-		if err := mount(m.Source, m.Destination, procfd, "cgroup2", uintptr(m.Flags), m.Data); err != nil {
-			// when we are in UserNS but CgroupNS is not unshared, we cannot mount cgroup2 (#2158)
-			if errors.Is(err, unix.EPERM) || errors.Is(err, unix.EBUSY) {
-				src := fs2.UnifiedMountpoint
-				if c.cgroupns && c.cgroup2Path != "" {
-					// Emulate cgroupns by bind-mounting
-					// the container cgroup path rather than
-					// the whole /sys/fs/cgroup.
-					src = c.cgroup2Path
-				}
-				err = mount(src, m.Destination, procfd, "", uintptr(m.Flags)|unix.MS_BIND, "")
-				if c.rootlessCgroups && errors.Is(err, unix.ENOENT) {
-					err = nil
-				}
-			}
-			return err
-		}
-		return nil
+	err = utils.WithProcfd(c.root, m.Destination, func(procfd string) error {
+		return mount(m.Source, m.Destination, procfd, "cgroup2", uintptr(m.Flags), m.Data)
 	})
+	if err == nil || !(errors.Is(err, unix.EPERM) || errors.Is(err, unix.EBUSY)) {
+		return err
+	}
+
+	// When we are in UserNS but CgroupNS is not unshared, we cannot mount
+	// cgroup2 (#2158), so fall back to bind mount.
+	bindM := &configs.Mount{
+		Device:           "bind",
+		Source:           fs2.UnifiedMountpoint,
+		Destination:      m.Destination,
+		Flags:            unix.MS_BIND | m.Flags,
+		PropagationFlags: m.PropagationFlags,
+	}
+	if c.cgroupns && c.cgroup2Path != "" {
+		// Emulate cgroupns by bind-mounting the container cgroup path
+		// rather than the whole /sys/fs/cgroup.
+		bindM.Source = c.cgroup2Path
+	}
+	// mountToRootfs() handles remounting for MS_RDONLY.
+	// No need to set c.fd here, because mountToRootfs() calls utils.WithProcfd() by itself in mountPropagate().
+	err = mountToRootfs(bindM, c)
+	if c.rootlessCgroups && errors.Is(err, unix.ENOENT) {
+		// ENOENT (for `src = c.cgroup2Path`) happens when rootless runc is being executed
+		// outside the userns+mountns.
+		//
+		// Mask `/sys/fs/cgroup` to ensure it is read-only, even when `/sys` is mounted
+		// with `rbind,ro` (`runc spec --rootless` produces `rbind,ro` for `/sys`).
+		err = utils.WithProcfd(c.root, m.Destination, func(procfd string) error {
+			return maskPath(procfd, c.label)
+		})
+	}
+	return err
 }
 
 func doTmpfsCopyUp(m *configs.Mount, rootfs, mountLabel string) (Err error) {

tests/integration/mounts.bats

@@ -63,3 +63,20 @@ function teardown() {
 	runc run test_busybox
 	[ "$status" -eq 0 ]
 }
+
+# https://github.com/opencontainers/runc/security/advisories/GHSA-m8cg-xc2p-r3fc
+@test "runc run [ro /sys/fs/cgroup mount]" {
+	# With cgroup namespace
+	update_config '.process.args |= ["sh", "-euc", "for f in `grep /sys/fs/cgroup /proc/mounts | awk \"{print \\\\$2}\"| uniq`; do grep -w $f /proc/mounts | tail -n1; done"]'
+	runc run test_busybox
+	[ "$status" -eq 0 ]
+	[ "${#lines[@]}" -ne 0 ]
+	for line in "${lines[@]}"; do [[ "${line}" == *'ro,'* ]]; done
+
+	# Without cgroup namespace
+	update_config '.linux.namespaces -= [{"type": "cgroup"}]'
+	runc run test_busybox
+	[ "$status" -eq 0 ]
+	[ "${#lines[@]}" -ne 0 ]
+	for line in "${lines[@]}"; do [[ "${line}" == *'ro,'* ]]; done
+}