// applyLayer is the entry-point for docker-applylayer on re-exec. This is not
// used on Windows as it does not support chroot, hence no point sandboxing
// through chroot and rexec.
func applyLayer() {
var (
tmpDir string
err error
options *archive.TarOptions
)
runtime.LockOSThread()
flag.Parse()
inUserns := rsystem.RunningInUserNS()
if err := chroot(flag.Arg(0)); err != nil {
fatal(err)
}
// We need to be able to set any perms
oldmask, err := system.Umask(0)
defer system.Umask(oldmask)
if err != nil {
fatal(err)
}
if err := json.Unmarshal([]byte(os.Getenv("OPT")), &options); err != nil {
fatal(err)
}
if inUserns {
options.InUserNS = true
}
if tmpDir, err = ioutil.TempDir("/", "temp-docker-extract"); err != nil {
fatal(err)
}
os.Setenv("TMPDIR", tmpDir)
size, err := archive.UnpackLayer("/", os.Stdin, options)
os.RemoveAll(tmpDir)
if err != nil {
fatal(err)
}
encoder := json.NewEncoder(os.Stdout)
if err := encoder.Encode(applyLayerResponse{size}); err != nil {
fatal(fmt.Errorf("unable to encode layerSize JSON: %s", err))
}
if _, err := flush(os.Stdin); err != nil {
fatal(err)
}
os.Exit(0)
}
// Create the device nodes in the container.
func createDevices(config *configs.Config) error {
useBindMount := system.RunningInUserNS() || config.Namespaces.Contains(configs.NEWUSER)
oldMask := syscall.Umask(0000)
for _, node := range config.Devices {
// containers running in a user namespace are not allowed to mknod
// devices so we can just bind mount it from the host.
if err := createDeviceNode(config.Rootfs, node, useBindMount); err != nil {
syscall.Umask(oldMask)
return err
}
}
syscall.Umask(oldMask)
return nil
}
// applyLayerHandler parses a diff in the standard layer format from `layer`, and
// applies it to the directory `dest`. Returns the size in bytes of the
// contents of the layer.
func applyLayerHandler(dest string, layer io.Reader, options *archive.TarOptions, decompress bool) (size int64, err error) {
dest = filepath.Clean(dest)
if decompress {
decompressed, err := archive.DecompressStream(layer)
if err != nil {
return 0, err
}
defer decompressed.Close()
layer = decompressed
}
if options == nil {
options = &archive.TarOptions{}
if rsystem.RunningInUserNS() {
options.InUserNS = true
}
}
if options.ExcludePatterns == nil {
options.ExcludePatterns = []string{}
}
data, err := json.Marshal(options)
if err != nil {
return 0, fmt.Errorf("ApplyLayer json encode: %v", err)
}
cmd := reexec.Command("docker-applyLayer", dest)
cmd.Stdin = layer
cmd.Env = append(cmd.Env, fmt.Sprintf("OPT=%s", data))
outBuf, errBuf := new(bytes.Buffer), new(bytes.Buffer)
cmd.Stdout, cmd.Stderr = outBuf, errBuf
if err = cmd.Run(); err != nil {
return 0, fmt.Errorf("ApplyLayer %s stdout: %s stderr: %s", err, outBuf, errBuf)
}
// Stdout should be a valid JSON struct representing an applyLayerResponse.
response := applyLayerResponse{}
decoder := json.NewDecoder(outBuf)
if err = decoder.Decode(&response); err != nil {
return 0, fmt.Errorf("unable to decode ApplyLayer JSON response: %s", err)
}
return response.LayerSize, nil
}
func (s *DevicesGroup) Set(path string, cgroup *configs.Cgroup) error {
if system.RunningInUserNS() {
return nil
}
devices := cgroup.Resources.Devices
if len(devices) > 0 {
for _, dev := range devices {
file := "devices.deny"
if dev.Allow {
file = "devices.allow"
}
if err := writeFile(path, file, dev.CgroupString()); err != nil {
return err
}
}
return nil
}
if cgroup.Resources.AllowAllDevices != nil {
if *cgroup.Resources.AllowAllDevices == false {
if err := writeFile(path, "devices.deny", "a"); err != nil {
return err
}
for _, dev := range cgroup.Resources.AllowedDevices {
if err := writeFile(path, "devices.allow", dev.CgroupString()); err != nil {
return err
}
}
return nil
}
if err := writeFile(path, "devices.allow", "a"); err != nil {
return err
}
}
for _, dev := range cgroup.Resources.DeniedDevices {
if err := writeFile(path, "devices.deny", dev.CgroupString()); err != nil {
return err
}
}
return nil
}
func setupOOMScoreAdj(score int) error {
f, err := os.OpenFile("/proc/self/oom_score_adj", os.O_WRONLY, 0)
if err != nil {
return err
}
defer f.Close()
stringScore := strconv.Itoa(score)
_, err = f.WriteString(stringScore)
if os.IsPermission(err) {
// Setting oom_score_adj does not work in an
// unprivileged container. Ignore the error, but log
// it if we appear not to be in that situation.
if !rsystem.RunningInUserNS() {
logrus.Debugf("Permission denied writing %q to /proc/self/oom_score_adj", stringScore)
}
return nil
}
return err
}
func setOOMScore(pid, score int) error {
oomScoreAdjPath := fmt.Sprintf("/proc/%d/oom_score_adj", pid)
f, err := os.OpenFile(oomScoreAdjPath, os.O_WRONLY, 0)
if err != nil {
return err
}
stringScore := strconv.Itoa(score)
_, err = f.WriteString(stringScore)
f.Close()
if os.IsPermission(err) {
// Setting oom_score_adj does not work in an
// unprivileged container. Ignore the error, but log
// it if we appear not to be in that situation.
if !system.RunningInUserNS() {
logrus.Debugf("Permission denied writing %q to %s", stringScore, oomScoreAdjPath)
}
return nil
}
return err
}
// handleTarTypeBlockCharFifo is an OS-specific helper function used by
// createTarFile to handle the following types of header: Block; Char; Fifo
func handleTarTypeBlockCharFifo(hdr *tar.Header, path string) error {
if rsystem.RunningInUserNS() {
// cannot create a device if running in user namespace
return nil
}
mode := uint32(hdr.Mode & 07777)
switch hdr.Typeflag {
case tar.TypeBlock:
mode |= syscall.S_IFBLK
case tar.TypeChar:
mode |= syscall.S_IFCHR
case tar.TypeFifo:
mode |= syscall.S_IFIFO
}
if err := system.Mknod(path, mode, int(system.Mkdev(hdr.Devmajor, hdr.Devminor))); err != nil {
return err
}
return nil
}
// Return a nil error if the kernel supports aufs
// We cannot modprobe because inside dind modprobe fails
// to run
func supportsAufs() error {
// We can try to modprobe aufs first before looking at
// proc/filesystems for when aufs is supported
exec.Command("modprobe", "aufs").Run()
if rsystem.RunningInUserNS() {
return ErrAufsNested
}
f, err := os.Open("/proc/filesystems")
if err != nil {
return err
}
defer f.Close()
s := bufio.NewScanner(f)
for s.Scan() {
if strings.Contains(s.Text(), "aufs") {
return nil
}
}
return ErrAufsNotSupported
}
func copyDir(srcDir, dstDir string, flags copyFlags) error {
err := filepath.Walk(srcDir, func(srcPath string, f os.FileInfo, err error) error {
if err != nil {
return err
}
// Rebase path
relPath, err := filepath.Rel(srcDir, srcPath)
if err != nil {
return err
}
dstPath := filepath.Join(dstDir, relPath)
if err != nil {
return err
}
stat, ok := f.Sys().(*syscall.Stat_t)
if !ok {
return fmt.Errorf("Unable to get raw syscall.Stat_t data for %s", srcPath)
}
isHardlink := false
switch f.Mode() & os.ModeType {
case 0: // Regular file
if flags©Hardlink != 0 {
isHardlink = true
if err := os.Link(srcPath, dstPath); err != nil {
return err
}
} else {
if err := copyRegular(srcPath, dstPath, f.Mode()); err != nil {
return err
}
}
case os.ModeDir:
if err := os.Mkdir(dstPath, f.Mode()); err != nil && !os.IsExist(err) {
return err
}
case os.ModeSymlink:
link, err := os.Readlink(srcPath)
if err != nil {
return err
}
if err := os.Symlink(link, dstPath); err != nil {
return err
}
case os.ModeNamedPipe:
fallthrough
case os.ModeSocket:
if rsystem.RunningInUserNS() {
// cannot create a device if running in user namespace
return nil
}
if err := syscall.Mkfifo(dstPath, stat.Mode); err != nil {
return err
}
case os.ModeDevice:
if err := syscall.Mknod(dstPath, stat.Mode, int(stat.Rdev)); err != nil {
return err
}
default:
return fmt.Errorf("Unknown file type for %s\n", srcPath)
}
// Everything below is copying metadata from src to dst. All this metadata
// already shares an inode for hardlinks.
if isHardlink {
return nil
}
if err := os.Lchown(dstPath, int(stat.Uid), int(stat.Gid)); err != nil {
return err
}
if err := copyXattr(srcPath, dstPath, "security.capability"); err != nil {
return err
}
// We need to copy this attribute if it appears in an overlay upper layer, as
// this function is used to copy those. It is set by overlay if a directory
// is removed and then re-created and should not inherit anything from the
// same dir in the lower dir.
if err := copyXattr(srcPath, dstPath, "trusted.overlay.opaque"); err != nil {
return err
}
isSymlink := f.Mode()&os.ModeSymlink != 0
// There is no LChmod, so ignore mode for symlink. Also, this
// must happen after chown, as that can modify the file mode
if !isSymlink {
if err := os.Chmod(dstPath, f.Mode()); err != nil {
return err
}
}
// system.Chtimes doesn't support a NOFOLLOW flag atm
if !isSymlink {
aTime := time.Unix(int64(stat.Atim.Sec), int64(stat.Atim.Nsec))
mTime := time.Unix(int64(stat.Mtim.Sec), int64(stat.Mtim.Nsec))
if err := system.Chtimes(dstPath, aTime, mTime); err != nil {
return err
}
} else {
ts := []syscall.Timespec{stat.Atim, stat.Mtim}
if err := system.LUtimesNano(dstPath, ts); err != nil {
return err
}
}
return nil
})
return err
}
// chroot on linux uses pivot_root instead of chroot
// pivot_root takes a new root and an old root.
// Old root must be a sub-dir of new root, it is where the current rootfs will reside after the call to pivot_root.
// New root is where the new rootfs is set to.
// Old root is removed after the call to pivot_root so it is no longer available under the new root.
// This is similar to how libcontainer sets up a container's rootfs
func chroot(path string) (err error) {
// if the engine is running in a user namespace we need to use actual chroot
if rsystem.RunningInUserNS() {
return realChroot(path)
}
if err := syscall.Unshare(syscall.CLONE_NEWNS); err != nil {
return fmt.Errorf("Error creating mount namespace before pivot: %v", err)
}
// make everything in new ns private
if err := mount.MakeRPrivate("/"); err != nil {
return err
}
// ensure path is a mountpoint
if err := mount.MakePrivate(path); err != nil {
return err
}
// setup oldRoot for pivot_root
pivotDir, err := ioutil.TempDir(path, ".pivot_root")
if err != nil {
return fmt.Errorf("Error setting up pivot dir: %v", err)
}
var mounted bool
defer func() {
if mounted {
// make sure pivotDir is not mounted before we try to remove it
if errCleanup := syscall.Unmount(pivotDir, syscall.MNT_DETACH); errCleanup != nil {
if err == nil {
err = errCleanup
}
return
}
}
errCleanup := os.Remove(pivotDir)
// pivotDir doesn't exist if pivot_root failed and chroot+chdir was successful
// because we already cleaned it up on failed pivot_root
if errCleanup != nil && !os.IsNotExist(errCleanup) {
errCleanup = fmt.Errorf("Error cleaning up after pivot: %v", errCleanup)
if err == nil {
err = errCleanup
}
}
}()
if err := syscall.PivotRoot(path, pivotDir); err != nil {
// If pivot fails, fall back to the normal chroot after cleaning up temp dir
if err := os.Remove(pivotDir); err != nil {
return fmt.Errorf("Error cleaning up after failed pivot: %v", err)
}
return realChroot(path)
}
mounted = true
// This is the new path for where the old root (prior to the pivot) has been moved to
// This dir contains the rootfs of the caller, which we need to remove so it is not visible during extraction
pivotDir = filepath.Join("/", filepath.Base(pivotDir))
if err := syscall.Chdir("/"); err != nil {
return fmt.Errorf("Error changing to new root: %v", err)
}
// Make the pivotDir (where the old root lives) private so it can be unmounted without propagating to the host
if err := syscall.Mount("", pivotDir, "", syscall.MS_PRIVATE|syscall.MS_REC, ""); err != nil {
return fmt.Errorf("Error making old root private after pivot: %v", err)
}
// Now unmount the old root so it's no longer visible from the new root
if err := syscall.Unmount(pivotDir, syscall.MNT_DETACH); err != nil {
return fmt.Errorf("Error while unmounting old root after pivot: %v", err)
}
mounted = false
return nil
}