File CVE-2021-30465.patch of Package runc.21021

Overview Repositories Revisions Requests Users Attributes Meta

File CVE-2021-30465.patch of Package runc.21021

From f55e108b7284971c7051c56d516ae7f5633fec33 Mon Sep 17 00:00:00 2001
From: Kir Kolyshkin <kolyshkin@gmail.com>
Date: Tue, 23 Feb 2021 17:58:07 -0800
Subject: [PATCH] rootfs: add mount destination validation

Because the target of a mount is inside a container (which may be a
volume that is shared with another container), there exists a race
condition where the target of the mount may change to a path containing
a symlink after we have sanitised the path -- resulting in us
inadvertently mounting the path outside of the container.

This is not immediately useful because we are in a mount namespace with
MS_SLAVE mount propagation applied to "/", so we cannot mount on top of
host paths in the host namespace. However, if any subsequent mountpoints
in the configuration use a subdirectory of that host path as a source,
those subsequent mounts will use an attacker-controlled source path
(resolved within the host rootfs) -- allowing the bind-mounting of "/"
into the container.

While arguably configuration issues like this are not entirely within
runc's threat model, within the context of Kubernetes (and possibly
other container managers that provide semi-arbitrary container creation
privileges to untrusted users) this is a legitimate issue. Since we
cannot block mounting from the host into the container, we need to block
the first stage of this attack (mounting onto a path outside the
container).

The long-term plan to solve this would be to migrate to libpathrs, but
as a stop-gap we implement libpathrs-like path verification through
readlink(/proc/self/fd/$n) and then do mount operations through the
procfd once it's been verified to be inside the container. The target
could move after we've checked it, but if it is inside the container
then we can assume that it is safe for the same reason that libpathrs
operations would be safe.

A slight wrinkle is the "copyup" functionality we provide for tmpfs,
which is the only case where we want to do a mount on the host
filesystem. To facilitate this, I split out the copy-up functionality
entirely so that the logic isn't interspersed with the regular tmpfs
logic. In addition, all dependencies on m.Destination being overwritten
have been removed since that pattern was just begging to be a source of
more mount-target bugs (we do still have to modify m.Destination for
tmpfs-copyup but we only do it temporarily).

--8<-----------------------------------------------------------------------

This is a backport of the fix for CVE-2021-30465 to the v1.0.0-rc93
release. However, because the patch does not apply cleanly it was
necessary to backport the following commits (from [1]):

* deb8a8dd7767 ("libct/newInitConfig: nit")
 * 1e476578b6cd ("libct/rootfs: introduce and use mountConfig")
 * 3826db196d59 ("libct/rootfs/mountCgroupV2: minor refactor")
 * ff692f289b60 ("Fix cgroup2 mount for rootless case")

And the patch itself was modified to remove hardenings for code which
didn't exist in v1.0.0-rc93 (in particular, the mount changes in [2]).
All of these changes have been squashed into one patch.

[1]: https://github.com/opencontainers/runc/pull/2818
[2]: https://github.com/opencontainers/runc/pull/2798

--8<-----------------------------------------------------------------------

Fixes: CVE-2021-30465
SUSE-Bugs: bsc#1185405
Reported-by: Etienne Champetier <champetier.etienne@gmail.com>
Co-authored-by: Noah Meyerhans <nmeyerha@amazon.com>
Reviewed-by: Samuel Karp <skarp@amazon.com>
Reviewed-by: Kir Kolyshkin <kolyshkin@gmail.com> (@kolyshkin)
Reviewed-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
---
 libcontainer/container_linux.go  |  11 +-
 libcontainer/init_linux.go       |   1 +
 libcontainer/rootfs_linux.go     | 291 +++++++++++++++++--------------
 libcontainer/specconv/example.go |  18 +-
 libcontainer/utils/utils.go      |  54 ++++++
 libcontainer/utils/utils_test.go |  35 ++++
 6 files changed, 263 insertions(+), 147 deletions(-)

diff --git a/libcontainer/container_linux.go b/libcontainer/container_linux.go
index 3dca29e4c3f2..70b388b1252e 100644
--- a/libcontainer/container_linux.go
+++ b/libcontainer/container_linux.go
@@ -594,6 +594,9 @@ func (c *linuxContainer) newInitConfig(process *Process) *initConfig {
 		AppArmorProfile:  c.config.AppArmorProfile,
 		ProcessLabel:     c.config.ProcessLabel,
 		Rlimits:          c.config.Rlimits,
+		CreateConsole:    process.ConsoleSocket != nil,
+		ConsoleWidth:     process.ConsoleWidth,
+		ConsoleHeight:    process.ConsoleHeight,
 	}
 	if process.NoNewPrivileges != nil {
 		cfg.NoNewPrivileges = *process.NoNewPrivileges
@@ -607,9 +610,10 @@ func (c *linuxContainer) newInitConfig(process *Process) *initConfig {
 	if len(process.Rlimits) > 0 {
 		cfg.Rlimits = process.Rlimits
 	}
-	cfg.CreateConsole = process.ConsoleSocket != nil
-	cfg.ConsoleWidth = process.ConsoleWidth
-	cfg.ConsoleHeight = process.ConsoleHeight
+	if cgroups.IsCgroup2UnifiedMode() {
+		cfg.Cgroup2Path = c.cgroupManager.Path("")
+	}
+
 	return cfg
 }
 
@@ -1198,7 +1202,6 @@ func (c *linuxContainer) makeCriuRestoreMountpoints(m *configs.Mount) error {
 		if err := checkProcMount(c.config.Rootfs, dest, ""); err != nil {
 			return err
 		}
-		m.Destination = dest
 		if err := os.MkdirAll(dest, 0755); err != nil {
 			return err
 		}
diff --git a/libcontainer/init_linux.go b/libcontainer/init_linux.go
index c57af0eebb8b..681797099f46 100644
--- a/libcontainer/init_linux.go
+++ b/libcontainer/init_linux.go
@@ -70,6 +70,7 @@ type initConfig struct {
 	RootlessEUID     bool                  `json:"rootless_euid,omitempty"`
 	RootlessCgroups  bool                  `json:"rootless_cgroups,omitempty"`
 	SpecState        *specs.State          `json:"spec_state,omitempty"`
+	Cgroup2Path      string                `json:"cgroup2_path,omitempty"`
 }
 
 type initer interface {
diff --git a/libcontainer/rootfs_linux.go b/libcontainer/rootfs_linux.go
index 411496ab7c6d..96be669c365e 100644
--- a/libcontainer/rootfs_linux.go
+++ b/libcontainer/rootfs_linux.go
@@ -17,6 +17,7 @@ import (
 	"github.com/moby/sys/mountinfo"
 	"github.com/mrunalp/fileutils"
 	"github.com/opencontainers/runc/libcontainer/cgroups"
+	"github.com/opencontainers/runc/libcontainer/cgroups/fs2"
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"github.com/opencontainers/runc/libcontainer/devices"
 	"github.com/opencontainers/runc/libcontainer/system"
@@ -24,11 +25,20 @@ import (
 	libcontainerUtils "github.com/opencontainers/runc/libcontainer/utils"
 	"github.com/opencontainers/runtime-spec/specs-go"
 	"github.com/opencontainers/selinux/go-selinux/label"
+	"github.com/sirupsen/logrus"
 	"golang.org/x/sys/unix"
 )
 
 const defaultMountFlags = unix.MS_NOEXEC | unix.MS_NOSUID | unix.MS_NODEV
 
+type mountConfig struct {
+	root            string
+	label           string
+	cgroup2Path     string
+	rootlessCgroups bool
+	cgroupns        bool
+}
+
 // needsSetupDev returns true if /dev needs to be set up.
 func needsSetupDev(config *configs.Config) bool {
 	for _, m := range config.Mounts {
@@ -48,7 +58,13 @@ func prepareRootfs(pipe io.ReadWriter, iConfig *initConfig) (err error) {
 		return newSystemErrorWithCause(err, "preparing rootfs")
 	}
 
-	hasCgroupns := config.Namespaces.Contains(configs.NEWCGROUP)
+	mountConfig := &mountConfig{
+		root:            config.Rootfs,
+		label:           config.MountLabel,
+		cgroup2Path:     iConfig.Cgroup2Path,
+		rootlessCgroups: iConfig.RootlessCgroups,
+		cgroupns:        config.Namespaces.Contains(configs.NEWCGROUP),
+	}
 	setupDev := needsSetupDev(config)
 	for _, m := range config.Mounts {
 		for _, precmd := range m.PremountCmds {
@@ -56,7 +72,7 @@ func prepareRootfs(pipe io.ReadWriter, iConfig *initConfig) (err error) {
 				return newSystemErrorWithCause(err, "running premount command")
 			}
 		}
-		if err := mountToRootfs(m, config.Rootfs, config.MountLabel, hasCgroupns); err != nil {
+		if err := mountToRootfs(m, mountConfig); err != nil {
 			return newSystemErrorWithCausef(err, "mounting %q to rootfs at %q", m.Source, m.Destination)
 		}
 
@@ -213,8 +229,6 @@ func prepareBindMount(m *configs.Mount, rootfs string) error {
 	if err := checkProcMount(rootfs, dest, m.Source); err != nil {
 		return err
 	}
-	// update the mount with the correct dest after symlinks are resolved.
-	m.Destination = dest
 	if err := createIfNotExists(dest, stat.IsDir()); err != nil {
 		return err
 	}
@@ -222,7 +236,7 @@ func prepareBindMount(m *configs.Mount, rootfs string) error {
 	return nil
 }
 
-func mountCgroupV1(m *configs.Mount, rootfs, mountLabel string, enableCgroupns bool) error {
+func mountCgroupV1(m *configs.Mount, c *mountConfig) error {
 	binds, err := getCgroupMounts(m)
 	if err != nil {
 		return err
@@ -242,31 +256,34 @@ func mountCgroupV1(m *configs.Mount, rootfs, mountLabel string, enableCgroupns b
 		Data:             "mode=755",
 		PropagationFlags: m.PropagationFlags,
 	}
-	if err := mountToRootfs(tmpfs, rootfs, mountLabel, enableCgroupns); err != nil {
+	if err := mountToRootfs(tmpfs, c); err != nil {
 		return err
 	}
 	for _, b := range binds {
-		if enableCgroupns {
-			subsystemPath := filepath.Join(rootfs, b.Destination)
+		if c.cgroupns {
+			subsystemPath := filepath.Join(c.root, b.Destination)
 			if err := os.MkdirAll(subsystemPath, 0755); err != nil {
 				return err
 			}
-			flags := defaultMountFlags
-			if m.Flags&unix.MS_RDONLY != 0 {
-				flags = flags | unix.MS_RDONLY
-			}
-			cgroupmount := &configs.Mount{
-				Source:      "cgroup",
-				Device:      "cgroup", // this is actually fstype
-				Destination: subsystemPath,
-				Flags:       flags,
-				Data:        filepath.Base(subsystemPath),
-			}
-			if err := mountNewCgroup(cgroupmount); err != nil {
+			if err := utils.WithProcfd(c.root, b.Destination, func(procfd string) error {
+				flags := defaultMountFlags
+				if m.Flags&unix.MS_RDONLY != 0 {
+					flags = flags | unix.MS_RDONLY
+				}
+				var (
+					source = "cgroup"
+					data   = filepath.Base(subsystemPath)
+				)
+				if data == "systemd" {
+					data = cgroups.CgroupNamePrefix + data
+					source = "systemd"
+				}
+				return unix.Mount(source, procfd, "cgroup", uintptr(flags), data)
+			}); err != nil {
 				return err
 			}
 		} else {
-			if err := mountToRootfs(b, rootfs, mountLabel, enableCgroupns); err != nil {
+			if err := mountToRootfs(b, c); err != nil {
 				return err
 			}
 		}
@@ -276,7 +293,7 @@ func mountCgroupV1(m *configs.Mount, rootfs, mountLabel string, enableCgroupns b
 			// symlink(2) is very dumb, it will just shove the path into
 			// the link and doesn't do any checks or relative path
 			// conversion. Also, don't error out if the cgroup already exists.
-			if err := os.Symlink(mc, filepath.Join(rootfs, m.Destination, ss)); err != nil && !os.IsExist(err) {
+			if err := os.Symlink(mc, filepath.Join(c.root, m.Destination, ss)); err != nil && !os.IsExist(err) {
 				return err
 			}
 		}
@@ -284,30 +301,87 @@ func mountCgroupV1(m *configs.Mount, rootfs, mountLabel string, enableCgroupns b
 	return nil
 }
 
-func mountCgroupV2(m *configs.Mount, rootfs, mountLabel string, enableCgroupns bool) error {
-	cgroupPath, err := securejoin.SecureJoin(rootfs, m.Destination)
+func mountCgroupV2(m *configs.Mount, c *mountConfig) error {
+	dest, err := securejoin.SecureJoin(c.root, m.Destination)
 	if err != nil {
 		return err
 	}
-	if err := os.MkdirAll(cgroupPath, 0755); err != nil {
+	if err := os.MkdirAll(dest, 0755); err != nil {
 		return err
 	}
-	if err := unix.Mount(m.Source, cgroupPath, "cgroup2", uintptr(m.Flags), m.Data); err != nil {
-		// when we are in UserNS but CgroupNS is not unshared, we cannot mount cgroup2 (#2158)
-		if err == unix.EPERM || err == unix.EBUSY {
-			return unix.Mount("/sys/fs/cgroup", cgroupPath, "", uintptr(m.Flags)|unix.MS_BIND, "")
+	return utils.WithProcfd(c.root, m.Destination, func(procfd string) error {
+		if err := unix.Mount(m.Source, 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 err == unix.EPERM || 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 = unix.Mount(src, procfd, "", uintptr(m.Flags)|unix.MS_BIND, "")
+				if err == unix.ENOENT && c.rootlessCgroups {
+					err = nil
+				}
+			}
+			return err
 		}
+		return nil
+	})
+}
+
+func doTmpfsCopyUp(m *configs.Mount, rootfs, mountLabel string) (Err error) {
+	// Set up a scratch dir for the tmpfs on the host.
+	tmpdir, err := prepareTmp("/tmp")
+	if err != nil {
+		return newSystemErrorWithCause(err, "tmpcopyup: failed to setup tmpdir")
+	}
+	defer cleanupTmp(tmpdir)
+	tmpDir, err := ioutil.TempDir(tmpdir, "runctmpdir")
+	if err != nil {
+		return newSystemErrorWithCause(err, "tmpcopyup: failed to create tmpdir")
+	}
+	defer os.RemoveAll(tmpDir)
+
+	// Configure the *host* tmpdir as if it's the container mount. We change
+	// m.Destination since we are going to mount *on the host*.
+	oldDest := m.Destination
+	m.Destination = tmpDir
+	err = mountPropagate(m, "/", mountLabel)
+	m.Destination = oldDest
+	if err != nil {
 		return err
 	}
-	return nil
+	defer func() {
+		if Err != nil {
+			if err := unix.Unmount(tmpDir, unix.MNT_DETACH); err != nil {
+				logrus.Warnf("tmpcopyup: failed to unmount tmpdir on error: %v", err)
+			}
+		}
+	}()
+
+	return utils.WithProcfd(rootfs, m.Destination, func(procfd string) (Err error) {
+		// Copy the container data to the host tmpdir. We append "/" to force
+		// CopyDirectory to resolve the symlink rather than trying to copy the
+		// symlink itself.
+		if err := fileutils.CopyDirectory(procfd+"/", tmpDir); err != nil {
+			return fmt.Errorf("tmpcopyup: failed to copy %s to %s (%s): %w", m.Destination, procfd, tmpDir, err)
+		}
+		// Now move the mount into the container.
+		if err := unix.Mount(tmpDir, procfd, "", unix.MS_MOVE, ""); err != nil {
+			return fmt.Errorf("tmpcopyup: failed to move mount %s to %s (%s): %w", tmpDir, procfd, m.Destination, err)
+		}
+		return nil
+	})
 }
 
-func mountToRootfs(m *configs.Mount, rootfs, mountLabel string, enableCgroupns bool) error {
-	var (
-		dest = m.Destination
-	)
-	if !strings.HasPrefix(dest, rootfs) {
-		dest = filepath.Join(rootfs, dest)
+func mountToRootfs(m *configs.Mount, c *mountConfig) error {
+	rootfs := c.root
+	mountLabel := c.label
+	dest, err := securejoin.SecureJoin(rootfs, m.Destination)
+	if err != nil {
+		return err
 	}
 
 	switch m.Device {
@@ -338,53 +412,21 @@ func mountToRootfs(m *configs.Mount, rootfs, mountLabel string, enableCgroupns b
 		}
 		return label.SetFileLabel(dest, mountLabel)
 	case "tmpfs":
-		copyUp := m.Extensions&configs.EXT_COPYUP == configs.EXT_COPYUP
-		tmpDir := ""
-		// dest might be an absolute symlink, so it needs
-		// to be resolved under rootfs.
-		dest, err := securejoin.SecureJoin(rootfs, m.Destination)
-		if err != nil {
-			return err
-		}
-		m.Destination = dest
 		stat, err := os.Stat(dest)
 		if err != nil {
 			if err := os.MkdirAll(dest, 0755); err != nil {
 				return err
 			}
 		}
-		if copyUp {
-			tmpdir, err := prepareTmp("/tmp")
-			if err != nil {
-				return newSystemErrorWithCause(err, "tmpcopyup: failed to setup tmpdir")
-			}
-			defer cleanupTmp(tmpdir)
-			tmpDir, err = ioutil.TempDir(tmpdir, "runctmpdir")
-			if err != nil {
-				return newSystemErrorWithCause(err, "tmpcopyup: failed to create tmpdir")
-			}
-			defer os.RemoveAll(tmpDir)
-			m.Destination = tmpDir
+
+		if m.Extensions&configs.EXT_COPYUP == configs.EXT_COPYUP {
+			err = doTmpfsCopyUp(m, rootfs, mountLabel)
+		} else {
+			err = mountPropagate(m, rootfs, mountLabel)
 		}
-		if err := mountPropagate(m, rootfs, mountLabel); err != nil {
+		if err != nil {
 			return err
 		}
-		if copyUp {
-			if err := fileutils.CopyDirectory(dest, tmpDir); err != nil {
-				errMsg := fmt.Errorf("tmpcopyup: failed to copy %s to %s: %v", dest, tmpDir, err)
-				if err1 := unix.Unmount(tmpDir, unix.MNT_DETACH); err1 != nil {
-					return newSystemErrorWithCausef(err1, "tmpcopyup: %v: failed to unmount", errMsg)
-				}
-				return errMsg
-			}
-			if err := unix.Mount(tmpDir, dest, "", unix.MS_MOVE, ""); err != nil {
-				errMsg := fmt.Errorf("tmpcopyup: failed to move mount %s to %s: %v", tmpDir, dest, err)
-				if err1 := unix.Unmount(tmpDir, unix.MNT_DETACH); err1 != nil {
-					return newSystemErrorWithCausef(err1, "tmpcopyup: %v: failed to unmount", errMsg)
-				}
-				return errMsg
-			}
-		}
 		if stat != nil {
 			if err = os.Chmod(dest, stat.Mode()); err != nil {
 				return err
@@ -424,23 +466,13 @@ func mountToRootfs(m *configs.Mount, rootfs, mountLabel string, enableCgroupns b
 		}
 	case "cgroup":
 		if cgroups.IsCgroup2UnifiedMode() {
-			return mountCgroupV2(m, rootfs, mountLabel, enableCgroupns)
+			return mountCgroupV2(m, c)
 		}
-		return mountCgroupV1(m, rootfs, mountLabel, enableCgroupns)
+		return mountCgroupV1(m, c)
 	default:
-		// ensure that the destination of the mount is resolved of symlinks at mount time because
-		// any previous mounts can invalidate the next mount's destination.
-		// this can happen when a user specifies mounts within other mounts to cause breakouts or other
-		// evil stuff to try to escape the container's rootfs.
-		var err error
-		if dest, err = securejoin.SecureJoin(rootfs, m.Destination); err != nil {
-			return err
-		}
 		if err := checkProcMount(rootfs, dest, m.Source); err != nil {
 			return err
 		}
-		// update the mount with the correct dest after symlinks are resolved.
-		m.Destination = dest
 		if err := os.MkdirAll(dest, 0755); err != nil {
 			return err
 		}
@@ -623,7 +655,7 @@ func createDevices(config *configs.Config) error {
 	return nil
 }
 
-func bindMountDeviceNode(dest string, node *devices.Device) error {
+func bindMountDeviceNode(rootfs, dest string, node *devices.Device) error {
 	f, err := os.Create(dest)
 	if err != nil && !os.IsExist(err) {
 		return err
@@ -631,7 +663,9 @@ func bindMountDeviceNode(dest string, node *devices.Device) error {
 	if f != nil {
 		f.Close()
 	}
-	return unix.Mount(node.Path, dest, "bind", unix.MS_BIND, "")
+	return utils.WithProcfd(rootfs, dest, func(procfd string) error {
+		return unix.Mount(node.Path, procfd, "bind", unix.MS_BIND, "")
+	})
 }
 
 // Creates the device node in the rootfs of the container.
@@ -640,18 +674,21 @@ func createDeviceNode(rootfs string, node *devices.Device, bind bool) error {
 		// The node only exists for cgroup reasons, ignore it here.
 		return nil
 	}
-	dest := filepath.Join(rootfs, node.Path)
+	dest, err := securejoin.SecureJoin(rootfs, node.Path)
+	if err != nil {
+		return err
+	}
 	if err := os.MkdirAll(filepath.Dir(dest), 0755); err != nil {
 		return err
 	}
 	if bind {
-		return bindMountDeviceNode(dest, node)
+		return bindMountDeviceNode(rootfs, dest, node)
 	}
 	if err := mknodDevice(dest, node); err != nil {
 		if os.IsExist(err) {
 			return nil
 		} else if os.IsPermission(err) {
-			return bindMountDeviceNode(dest, node)
+			return bindMountDeviceNode(rootfs, dest, node)
 		}
 		return err
 	}
@@ -987,61 +1024,47 @@ func writeSystemProperty(key, value string) error {
 }
 
 func remount(m *configs.Mount, rootfs string) error {
-	var (
-		dest = m.Destination
-	)
-	if !strings.HasPrefix(dest, rootfs) {
-		dest = filepath.Join(rootfs, dest)
-	}
-	return unix.Mount(m.Source, dest, m.Device, uintptr(m.Flags|unix.MS_REMOUNT), "")
+	return utils.WithProcfd(rootfs, m.Destination, func(procfd string) error {
+		return unix.Mount(m.Source, procfd, m.Device, uintptr(m.Flags|unix.MS_REMOUNT), "")
+	})
 }
 
 // Do the mount operation followed by additional mounts required to take care
-// of propagation flags.
+// of propagation flags. This will always be scoped inside the container rootfs.
 func mountPropagate(m *configs.Mount, rootfs string, mountLabel string) error {
 	var (
-		dest  = m.Destination
 		data  = label.FormatMountLabel(m.Data, mountLabel)
 		flags = m.Flags
 	)
-	if libcontainerUtils.CleanPath(dest) == "/dev" {
+	// Delay mounting the filesystem read-only if we need to do further
+	// operations on it. We need to set up files in "/dev" and tmpfs mounts may
+	// need to be chmod-ed after mounting. The mount will be remounted ro later
+	// in finalizeRootfs() if necessary.
+	if libcontainerUtils.CleanPath(m.Destination) == "/dev" || m.Device == "tmpfs" {
 		flags &= ^unix.MS_RDONLY
 	}
 
-	// Mount it rw to allow chmod operation. A remount will be performed
-	// later to make it ro if set.
-	if m.Device == "tmpfs" {
-		flags &= ^unix.MS_RDONLY
-	}
-
-	copyUp := m.Extensions&configs.EXT_COPYUP == configs.EXT_COPYUP
-	if !(copyUp || strings.HasPrefix(dest, rootfs)) {
-		dest = filepath.Join(rootfs, dest)
-	}
-
-	if err := unix.Mount(m.Source, dest, m.Device, uintptr(flags), data); err != nil {
-		return err
-	}
-
-	for _, pflag := range m.PropagationFlags {
-		if err := unix.Mount("", dest, "", uintptr(pflag), ""); err != nil {
-			return err
+	// Because the destination is inside a container path which might be
+	// mutating underneath us, we verify that we are actually going to mount
+	// inside the container with WithProcfd() -- mounting through a procfd
+	// mounts on the target.
+	if err := utils.WithProcfd(rootfs, m.Destination, func(procfd string) error {
+		return unix.Mount(m.Source, procfd, m.Device, uintptr(flags), data)
+	}); err != nil {
+		return fmt.Errorf("mount through procfd: %w", err)
+	}
+	// We have to apply mount propagation flags in a separate WithProcfd() call
+	// because the previous call invalidates the passed procfd -- the mount
+	// target needs to be re-opened.
+	if err := utils.WithProcfd(rootfs, m.Destination, func(procfd string) error {
+		for _, pflag := range m.PropagationFlags {
+			if err := unix.Mount("", procfd, "", uintptr(pflag), ""); err != nil {
+				return err
+			}
 		}
-	}
-	return nil
-}
-
-func mountNewCgroup(m *configs.Mount) error {
-	var (
-		data   = m.Data
-		source = m.Source
-	)
-	if data == "systemd" {
-		data = cgroups.CgroupNamePrefix + data
-		source = "systemd"
-	}
-	if err := unix.Mount(source, m.Destination, m.Device, uintptr(m.Flags), data); err != nil {
-		return err
+		return nil
+	}); err != nil {
+		return fmt.Errorf("change mount propagation through procfd: %w", err)
 	}
 	return nil
 }
diff --git a/libcontainer/specconv/example.go b/libcontainer/specconv/example.go
index 8a201bc78dd9..56bab3bfbfa5 100644
--- a/libcontainer/specconv/example.go
+++ b/libcontainer/specconv/example.go
@@ -2,6 +2,7 @@ package specconv
 
 import (
 	"os"
+	"path/filepath"
 	"strings"
 
 	"github.com/opencontainers/runc/libcontainer/cgroups"
@@ -200,8 +201,14 @@ func ToRootless(spec *specs.Spec) {
 	// Fix up mounts.
 	var mounts []specs.Mount
 	for _, mount := range spec.Mounts {
-		// Ignore all mounts that are under /sys.
-		if strings.HasPrefix(mount.Destination, "/sys") {
+		// Replace the /sys mount with an rbind.
+		if filepath.Clean(mount.Destination) == "/sys" {
+			mounts = append(mounts, specs.Mount{
+				Source:      "/sys",
+				Destination: "/sys",
+				Type:        "none",
+				Options:     []string{"rbind", "nosuid", "noexec", "nodev", "ro"},
+			})
 			continue
 		}
 
@@ -216,13 +223,6 @@ func ToRootless(spec *specs.Spec) {
 		mount.Options = options
 		mounts = append(mounts, mount)
 	}
-	// Add the sysfs mount as an rbind.
-	mounts = append(mounts, specs.Mount{
-		Source:      "/sys",
-		Destination: "/sys",
-		Type:        "none",
-		Options:     []string{"rbind", "nosuid", "noexec", "nodev", "ro"},
-	})
 	spec.Mounts = mounts
 
 	// Remove cgroup settings.
diff --git a/libcontainer/utils/utils.go b/libcontainer/utils/utils.go
index 1b72b7a1c1ba..cd78f23e1bd0 100644
--- a/libcontainer/utils/utils.go
+++ b/libcontainer/utils/utils.go
@@ -3,12 +3,15 @@ package utils
 import (
 	"encoding/binary"
 	"encoding/json"
+	"fmt"
 	"io"
 	"os"
 	"path/filepath"
+	"strconv"
 	"strings"
 	"unsafe"
 
+	"github.com/cyphar/filepath-securejoin"
 	"golang.org/x/sys/unix"
 )
 
@@ -88,6 +91,57 @@ func CleanPath(path string) string {
 	return filepath.Clean(path)
 }
 
+// stripRoot returns the passed path, stripping the root path if it was
+// (lexicially) inside it. Note that both passed paths will always be treated
+// as absolute, and the returned path will also always be absolute. In
+// addition, the paths are cleaned before stripping the root.
+func stripRoot(root, path string) string {
+	// Make the paths clean and absolute.
+	root, path = CleanPath("/"+root), CleanPath("/"+path)
+	switch {
+	case path == root:
+		path = "/"
+	case root == "/":
+		// do nothing
+	case strings.HasPrefix(path, root+"/"):
+		path = strings.TrimPrefix(path, root+"/")
+	}
+	return CleanPath("/" + path)
+}
+
+// WithProcfd runs the passed closure with a procfd path (/proc/self/fd/...)
+// corresponding to the unsafePath resolved within the root. Before passing the
+// fd, this path is verified to have been inside the root -- so operating on it
+// through the passed fdpath should be safe. Do not access this path through
+// the original path strings, and do not attempt to use the pathname outside of
+// the passed closure (the file handle will be freed once the closure returns).
+func WithProcfd(root, unsafePath string, fn func(procfd string) error) error {
+	// Remove the root then forcefully resolve inside the root.
+	unsafePath = stripRoot(root, unsafePath)
+	path, err := securejoin.SecureJoin(root, unsafePath)
+	if err != nil {
+		return fmt.Errorf("resolving path inside rootfs failed: %v", err)
+	}
+
+	// Open the target path.
+	fh, err := os.OpenFile(path, unix.O_PATH|unix.O_CLOEXEC, 0)
+	if err != nil {
+		return fmt.Errorf("open o_path procfd: %w", err)
+	}
+	defer fh.Close()
+
+	// Double-check the path is the one we expected.
+	procfd := "/proc/self/fd/" + strconv.Itoa(int(fh.Fd()))
+	if realpath, err := os.Readlink(procfd); err != nil {
+		return fmt.Errorf("procfd verification failed: %w", err)
+	} else if realpath != path {
+		return fmt.Errorf("possibly malicious path detected -- refusing to operate on %s", realpath)
+	}
+
+	// Run the closure.
+	return fn(procfd)
+}
+
 // SearchLabels searches a list of key-value pairs for the provided key and
 // returns the corresponding value. The pairs must be separated with '='.
 func SearchLabels(labels []string, query string) string {
diff --git a/libcontainer/utils/utils_test.go b/libcontainer/utils/utils_test.go
index 7f38ed169a6b..d33662238d36 100644
--- a/libcontainer/utils/utils_test.go
+++ b/libcontainer/utils/utils_test.go
@@ -143,3 +143,38 @@ func TestCleanPath(t *testing.T) {
 		t.Errorf("expected to receive '/foo' and received %s", path)
 	}
 }
+
+func TestStripRoot(t *testing.T) {
+	for _, test := range []struct {
+		root, path, out string
+	}{
+		// Works with multiple components.
+		{"/a/b", "/a/b/c", "/c"},
+		{"/hello/world", "/hello/world/the/quick-brown/fox", "/the/quick-brown/fox"},
+		// '/' must be a no-op.
+		{"/", "/a/b/c", "/a/b/c"},
+		// Must be the correct order.
+		{"/a/b", "/a/c/b", "/a/c/b"},
+		// Must be at start.
+		{"/abc/def", "/foo/abc/def/bar", "/foo/abc/def/bar"},
+		// Must be a lexical parent.
+		{"/foo/bar", "/foo/barSAMECOMPONENT", "/foo/barSAMECOMPONENT"},
+		// Must only strip the root once.
+		{"/foo/bar", "/foo/bar/foo/bar/baz", "/foo/bar/baz"},
+		// Deal with .. in a fairly sane way.
+		{"/foo/bar", "/foo/bar/../baz", "/foo/baz"},
+		{"/foo/bar", "../../../../../../foo/bar/baz", "/baz"},
+		{"/foo/bar", "/../../../../../../foo/bar/baz", "/baz"},
+		{"/foo/bar/../baz", "/foo/baz/bar", "/bar"},
+		{"/foo/bar/../baz", "/foo/baz/../bar/../baz/./foo", "/foo"},
+		// All paths are made absolute before stripping.
+		{"foo/bar", "/foo/bar/baz/bee", "/baz/bee"},
+		{"/foo/bar", "foo/bar/baz/beef", "/baz/beef"},
+		{"foo/bar", "foo/bar/baz/beets", "/baz/beets"},
+	} {
+		got := stripRoot(test.root, test.path)
+		if got != test.out {
+			t.Errorf("stripRoot(%q, %q) -- got %q, expected %q", test.root, test.path, got, test.out)
+		}
+	}
+}
-- 
2.31.1

Places

File CVE-2021-30465.patch of Package runc.21021

Places