Merge pull request #123060 from dims/update-to-runc-v1.1.12

Update to runc 1.1.12

go.mod

@@ -49,7 +49,7 @@ require (
 	github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822
 	github.com/onsi/ginkgo/v2 v2.15.0
 	github.com/onsi/gomega v1.31.0
-	github.com/opencontainers/runc v1.1.11
+	github.com/opencontainers/runc v1.1.12
 	github.com/opencontainers/selinux v1.11.0
 	github.com/pkg/errors v0.9.1
 	github.com/pmezard/go-difflib v1.0.0

go.sum

@@ -628,8 +628,8 @@ github.com/opencontainers/go-digest v1.0.0 h1:apOUWs51W5PlhuyGyz9FCeeBIOUDA/6nW8
 github.com/opencontainers/go-digest v1.0.0/go.mod h1:0JzlMkj0TRzQZfJkVvzbP0HBR3IKzErnv2BNG4W4MAM=
 github.com/opencontainers/image-spec v1.0.2 h1:9yCKha/T5XdGtO0q9Q9a6T5NUCsTn/DrBg0D7ufOcFM=
 github.com/opencontainers/image-spec v1.0.2/go.mod h1:BtxoFyWECRxE4U/7sNtV5W15zMzWCbyJoFRP3s7yZA0=
-github.com/opencontainers/runc v1.1.11 h1:9LjxyVlE0BPMRP2wuQDRlHV4941Jp9rc3F0+YKimopA=
-github.com/opencontainers/runc v1.1.11/go.mod h1:S+lQwSfncpBha7XTy/5lBwWgm5+y5Ma/O44Ekby9FK8=
+github.com/opencontainers/runc v1.1.12 h1:BOIssBaW1La0/qbNZHXOOa71dZfZEQOzW7dqQf3phss=
+github.com/opencontainers/runc v1.1.12/go.mod h1:S+lQwSfncpBha7XTy/5lBwWgm5+y5Ma/O44Ekby9FK8=
 github.com/opencontainers/runtime-spec v1.0.2/go.mod h1:jwyrGlmzljRJv/Fgzds9SsS/C5hL+LL3ko9hs6T5lQ0=
 github.com/opencontainers/runtime-spec v1.0.3-0.20200929063507-e6143ca7d51d/go.mod h1:jwyrGlmzljRJv/Fgzds9SsS/C5hL+LL3ko9hs6T5lQ0=
 github.com/opencontainers/runtime-spec v1.0.3-0.20220909204839-494a5a6aca78 h1:R5M2qXZiK/mWPMT4VldCOiSL9HIAMuxQZWdG0CSM5+4=

vendor/github.com/opencontainers/runc/libcontainer/cgroups/file.go

@@ -77,16 +77,16 @@ var (
 	// TestMode is set to true by unit tests that need "fake" cgroupfs.
 	TestMode bool
 
-	cgroupFd     int = -1
-	prepOnce     sync.Once
-	prepErr      error
-	resolveFlags uint64
+	cgroupRootHandle *os.File
+	prepOnce         sync.Once
+	prepErr          error
+	resolveFlags     uint64
 )
 
 func prepareOpenat2() error {
 	prepOnce.Do(func() {
 		fd, err := unix.Openat2(-1, cgroupfsDir, &unix.OpenHow{
-			Flags: unix.O_DIRECTORY | unix.O_PATH,
+			Flags: unix.O_DIRECTORY | unix.O_PATH | unix.O_CLOEXEC,
 		})
 		if err != nil {
 			prepErr = &os.PathError{Op: "openat2", Path: cgroupfsDir, Err: err}
@@ -97,15 +97,16 @@ func prepareOpenat2() error {
 			}
 			return
 		}
+		file := os.NewFile(uintptr(fd), cgroupfsDir)
+
 		var st unix.Statfs_t
-		if err = unix.Fstatfs(fd, &st); err != nil {
+		if err := unix.Fstatfs(int(file.Fd()), &st); err != nil {
 			prepErr = &os.PathError{Op: "statfs", Path: cgroupfsDir, Err: err}
 			logrus.Warnf("falling back to securejoin: %s", prepErr)
 			return
 		}
 
-		cgroupFd = fd
-
+		cgroupRootHandle = file
 		resolveFlags = unix.RESOLVE_BENEATH | unix.RESOLVE_NO_MAGICLINKS
 		if st.Type == unix.CGROUP2_SUPER_MAGIC {
 			// cgroupv2 has a single mountpoint and no "cpu,cpuacct" symlinks
@@ -132,7 +133,7 @@ func openFile(dir, file string, flags int) (*os.File, error) {
 		return openFallback(path, flags, mode)
 	}
 
-	fd, err := unix.Openat2(cgroupFd, relPath,
+	fd, err := unix.Openat2(int(cgroupRootHandle.Fd()), relPath,
 		&unix.OpenHow{
 			Resolve: resolveFlags,
 			Flags:   uint64(flags) | unix.O_CLOEXEC,
@@ -140,20 +141,20 @@ func openFile(dir, file string, flags int) (*os.File, error) {
 		})
 	if err != nil {
 		err = &os.PathError{Op: "openat2", Path: path, Err: err}
-		// Check if cgroupFd is still opened to cgroupfsDir
+		// Check if cgroupRootHandle is still opened to cgroupfsDir
 		// (happens when this package is incorrectly used
 		// across the chroot/pivot_root/mntns boundary, or
 		// when /sys/fs/cgroup is remounted).
 		//
 		// TODO: if such usage will ever be common, amend this
-		// to reopen cgroupFd and retry openat2.
-		fdStr := strconv.Itoa(cgroupFd)
+		// to reopen cgroupRootHandle and retry openat2.
+		fdStr := strconv.Itoa(int(cgroupRootHandle.Fd()))
 		fdDest, _ := os.Readlink("/proc/self/fd/" + fdStr)
 		if fdDest != cgroupfsDir {
-			// Wrap the error so it is clear that cgroupFd
+			// Wrap the error so it is clear that cgroupRootHandle
 			// is opened to an unexpected/wrong directory.
-			err = fmt.Errorf("cgroupFd %s unexpectedly opened to %s != %s: %w",
-				fdStr, fdDest, cgroupfsDir, err)
+			err = fmt.Errorf("cgroupRootHandle %d unexpectedly opened to %s != %s: %w",
+				cgroupRootHandle.Fd(), fdDest, cgroupfsDir, err)
 		}
 		return nil, err
 	}

vendor/github.com/opencontainers/runc/libcontainer/cgroups/fs/paths.go

@@ -83,6 +83,7 @@ func tryDefaultCgroupRoot() string {
 	if err != nil {
 		return ""
 	}
+	defer dir.Close()
 	names, err := dir.Readdirnames(1)
 	if err != nil {
 		return ""

vendor/github.com/opencontainers/runc/libcontainer/container_linux.go

@@ -353,6 +353,15 @@ func (c *linuxContainer) start(process *Process) (retErr error) {
 		}()
 	}
 
+	// Before starting "runc init", mark all non-stdio open files as O_CLOEXEC
+	// to make sure we don't leak any files into "runc init". Any files to be
+	// passed to "runc init" through ExtraFiles will get dup2'd by the Go
+	// runtime and thus their O_CLOEXEC flag will be cleared. This is some
+	// additional protection against attacks like CVE-2024-21626, by making
+	// sure we never leak files to "runc init" we didn't intend to.
+	if err := utils.CloseExecFrom(3); err != nil {
+		return fmt.Errorf("unable to mark non-stdio fds as cloexec: %w", err)
+	}
 	if err := parent.start(); err != nil {
 		return fmt.Errorf("unable to start container process: %w", err)
 	}

vendor/github.com/opencontainers/runc/libcontainer/init_linux.go

@@ -8,6 +8,7 @@ import (
 	"io"
 	"net"
 	"os"
+	"path/filepath"
 	"strings"
 	"unsafe"
 
@@ -135,6 +136,32 @@ func populateProcessEnvironment(env []string) error {
 	return nil
 }
 
+// verifyCwd ensures that the current directory is actually inside the mount
+// namespace root of the current process.
+func verifyCwd() error {
+	// getcwd(2) on Linux detects if cwd is outside of the rootfs of the
+	// current mount namespace root, and in that case prefixes "(unreachable)"
+	// to the returned string. glibc's getcwd(3) and Go's Getwd() both detect
+	// when this happens and return ENOENT rather than returning a non-absolute
+	// path. In both cases we can therefore easily detect if we have an invalid
+	// cwd by checking the return value of getcwd(3). See getcwd(3) for more
+	// details, and CVE-2024-21626 for the security issue that motivated this
+	// check.
+	//
+	// We have to use unix.Getwd() here because os.Getwd() has a workaround for
+	// $PWD which involves doing stat(.), which can fail if the current
+	// directory is inaccessible to the container process.
+	if wd, err := unix.Getwd(); errors.Is(err, unix.ENOENT) {
+		return errors.New("current working directory is outside of container mount namespace root -- possible container breakout detected")
+	} else if err != nil {
+		return fmt.Errorf("failed to verify if current working directory is safe: %w", err)
+	} else if !filepath.IsAbs(wd) {
+		// We shouldn't ever hit this, but check just in case.
+		return fmt.Errorf("current working directory is not absolute -- possible container breakout detected: cwd is %q", wd)
+	}
+	return nil
+}
+
 // finalizeNamespace drops the caps, sets the correct user
 // and working dir, and closes any leaked file descriptors
 // before executing the command inside the namespace
@@ -193,6 +220,10 @@ func finalizeNamespace(config *initConfig) error {
 			return fmt.Errorf("chdir to cwd (%q) set in config.json failed: %w", config.Cwd, err)
 		}
 	}
+	// Make sure our final working directory is inside the container.
+	if err := verifyCwd(); err != nil {
+		return err
+	}
 	if err := system.ClearKeepCaps(); err != nil {
 		return fmt.Errorf("unable to clear keep caps: %w", err)
 	}

vendor/github.com/opencontainers/runc/libcontainer/setns_init_linux.go

@@ -4,6 +4,7 @@ import (
 	"errors"
 	"fmt"
 	"os"
+	"os/exec"
 	"strconv"
 
 	"github.com/opencontainers/selinux/go-selinux"
@@ -14,6 +15,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/keys"
 	"github.com/opencontainers/runc/libcontainer/seccomp"
 	"github.com/opencontainers/runc/libcontainer/system"
+	"github.com/opencontainers/runc/libcontainer/utils"
 )
 
 // linuxSetnsInit performs the container's initialization for running a new process
@@ -82,6 +84,21 @@ func (l *linuxSetnsInit) Init() error {
 	if err := apparmor.ApplyProfile(l.config.AppArmorProfile); err != nil {
 		return err
 	}
+
+	// Check for the arg before waiting to make sure it exists and it is
+	// returned as a create time error.
+	name, err := exec.LookPath(l.config.Args[0])
+	if err != nil {
+		return err
+	}
+	// exec.LookPath in Go < 1.20 might return no error for an executable
+	// residing on a file system mounted with noexec flag, so perform this
+	// extra check now while we can still return a proper error.
+	// TODO: remove this once go < 1.20 is not supported.
+	if err := eaccess(name); err != nil {
+		return &os.PathError{Op: "eaccess", Path: name, Err: err}
+	}
+
 	// Set seccomp as close to execve as possible, so as few syscalls take
 	// place afterward (reducing the amount of syscalls that users need to
 	// enable in their seccomp profiles).
@@ -101,5 +118,23 @@ func (l *linuxSetnsInit) Init() error {
 		return &os.PathError{Op: "close log pipe", Path: "fd " + strconv.Itoa(l.logFd), Err: err}
 	}
 
-	return system.Execv(l.config.Args[0], l.config.Args[0:], os.Environ())
+	// Close all file descriptors we are not passing to the container. This is
+	// necessary because the execve target could use internal runc fds as the
+	// execve path, potentially giving access to binary files from the host
+	// (which can then be opened by container processes, leading to container
+	// escapes). Note that because this operation will close any open file
+	// descriptors that are referenced by (*os.File) handles from underneath
+	// the Go runtime, we must not do any file operations after this point
+	// (otherwise the (*os.File) finaliser could close the wrong file). See
+	// CVE-2024-21626 for more information as to why this protection is
+	// necessary.
+	//
+	// This is not needed for runc-dmz, because the extra execve(2) step means
+	// that all O_CLOEXEC file descriptors have already been closed and thus
+	// the second execve(2) from runc-dmz cannot access internal file
+	// descriptors from runc.
+	if err := utils.UnsafeCloseFrom(l.config.PassedFilesCount + 3); err != nil {
+		return err
+	}
+	return system.Exec(name, l.config.Args[0:], os.Environ())
 }

vendor/github.com/opencontainers/runc/libcontainer/standard_init_linux.go

@@ -17,6 +17,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/keys"
 	"github.com/opencontainers/runc/libcontainer/seccomp"
 	"github.com/opencontainers/runc/libcontainer/system"
+	"github.com/opencontainers/runc/libcontainer/utils"
 )
 
 type linuxStandardInit struct {
@@ -258,5 +259,23 @@ func (l *linuxStandardInit) Init() error {
 		return err
 	}
 
+	// Close all file descriptors we are not passing to the container. This is
+	// necessary because the execve target could use internal runc fds as the
+	// execve path, potentially giving access to binary files from the host
+	// (which can then be opened by container processes, leading to container
+	// escapes). Note that because this operation will close any open file
+	// descriptors that are referenced by (*os.File) handles from underneath
+	// the Go runtime, we must not do any file operations after this point
+	// (otherwise the (*os.File) finaliser could close the wrong file). See
+	// CVE-2024-21626 for more information as to why this protection is
+	// necessary.
+	//
+	// This is not needed for runc-dmz, because the extra execve(2) step means
+	// that all O_CLOEXEC file descriptors have already been closed and thus
+	// the second execve(2) from runc-dmz cannot access internal file
+	// descriptors from runc.
+	if err := utils.UnsafeCloseFrom(l.config.PassedFilesCount + 3); err != nil {
+		return err
+	}
 	return system.Exec(name, l.config.Args[0:], os.Environ())
 }

vendor/github.com/opencontainers/runc/libcontainer/utils/utils_unix.go

@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"os"
 	"strconv"
+	_ "unsafe" // for go:linkname
 
 	"golang.org/x/sys/unix"
 )
@@ -23,9 +24,11 @@ func EnsureProcHandle(fh *os.File) error {
 	return nil
 }
 
-// CloseExecFrom applies O_CLOEXEC to all file descriptors currently open for
-// the process (except for those below the given fd value).
-func CloseExecFrom(minFd int) error {
+type fdFunc func(fd int)
+
+// fdRangeFrom calls the passed fdFunc for each file descriptor that is open in
+// the current process.
+func fdRangeFrom(minFd int, fn fdFunc) error {
 	fdDir, err := os.Open("/proc/self/fd")
 	if err != nil {
 		return err
@@ -50,15 +53,60 @@ func CloseExecFrom(minFd int) error {
 		if fd < minFd {
 			continue
 		}
-		// Intentionally ignore errors from unix.CloseOnExec -- the cases where
-		// this might fail are basically file descriptors that have already
-		// been closed (including and especially the one that was created when
-		// os.ReadDir did the "opendir" syscall).
-		unix.CloseOnExec(fd)
+		// Ignore the file descriptor we used for readdir, as it will be closed
+		// when we return.
+		if uintptr(fd) == fdDir.Fd() {
+			continue
+		}
+		// Run the closure.
+		fn(fd)
 	}
 	return nil
 }
 
+// CloseExecFrom sets the O_CLOEXEC flag on all file descriptors greater or
+// equal to minFd in the current process.
+func CloseExecFrom(minFd int) error {
+	return fdRangeFrom(minFd, unix.CloseOnExec)
+}
+
+//go:linkname runtime_IsPollDescriptor internal/poll.IsPollDescriptor
+
+// In order to make sure we do not close the internal epoll descriptors the Go
+// runtime uses, we need to ensure that we skip descriptors that match
+// "internal/poll".IsPollDescriptor. Yes, this is a Go runtime internal thing,
+// unfortunately there's no other way to be sure we're only keeping the file
+// descriptors the Go runtime needs. Hopefully nothing blows up doing this...
+func runtime_IsPollDescriptor(fd uintptr) bool //nolint:revive
+
+// UnsafeCloseFrom closes all file descriptors greater or equal to minFd in the
+// current process, except for those critical to Go's runtime (such as the
+// netpoll management descriptors).
+//
+// NOTE: That this function is incredibly dangerous to use in most Go code, as
+// closing file descriptors from underneath *os.File handles can lead to very
+// bad behaviour (the closed file descriptor can be re-used and then any
+// *os.File operations would apply to the wrong file). This function is only
+// intended to be called from the last stage of runc init.
+func UnsafeCloseFrom(minFd int) error {
+	// We must not close some file descriptors.
+	return fdRangeFrom(minFd, func(fd int) {
+		if runtime_IsPollDescriptor(uintptr(fd)) {
+			// These are the Go runtimes internal netpoll file descriptors.
+			// These file descriptors are operated on deep in the Go scheduler,
+			// and closing those files from underneath Go can result in panics.
+			// There is no issue with keeping them because they are not
+			// executable and are not useful to an attacker anyway. Also we
+			// don't have any choice.
+			return
+		}
+		// There's nothing we can do about errors from close(2), and the
+		// only likely error to be seen is EBADF which indicates the fd was
+		// already closed (in which case, we got what we wanted).
+		_ = unix.Close(fd)
+	})
+}
+
 // NewSockPair returns a new unix socket pair
 func NewSockPair(name string) (parent *os.File, child *os.File, err error) {
 	fds, err := unix.Socketpair(unix.AF_LOCAL, unix.SOCK_STREAM|unix.SOCK_CLOEXEC, 0)

vendor/modules.txt

@@ -547,7 +547,7 @@ github.com/onsi/gomega/types
 # github.com/opencontainers/go-digest v1.0.0
 ## explicit; go 1.13
 github.com/opencontainers/go-digest
-# github.com/opencontainers/runc v1.1.11
+# github.com/opencontainers/runc v1.1.12
 ## explicit; go 1.17
 github.com/opencontainers/runc/libcontainer
 github.com/opencontainers/runc/libcontainer/apparmor