func setupBindmounts(rootfs string, bindMounts libcontainer.Mounts) error { for _, m := range bindMounts.OfType("bind") { var ( flags = syscall.MS_BIND | syscall.MS_REC dest = filepath.Join(rootfs, m.Destination) ) if !m.Writable { flags = flags | syscall.MS_RDONLY } if err := system.Mount(m.Source, dest, "bind", uintptr(flags), ""); err != nil { return fmt.Errorf("mounting %s into %s %s", m.Source, dest, err) } if !m.Writable { if err := system.Mount(m.Source, dest, "bind", uintptr(flags|syscall.MS_REMOUNT), ""); err != nil { return fmt.Errorf("remounting %s into %s %s", m.Source, dest, err) } } if m.Private { if err := system.Mount("", dest, "none", uintptr(syscall.MS_PRIVATE), ""); err != nil { return fmt.Errorf("mounting %s private %s", dest, err) } } } return nil }
// setupNewMountNamespace is used to initialize a new mount namespace for an new // container in the rootfs that is specified. // // There is no need to unmount the new mounts because as soon as the mount namespace // is no longer in use, the mounts will be removed automatically func setupNewMountNamespace(rootfs, console string, readonly bool) error { // mount as slave so that the new mounts do not propagate to the host if err := system.Mount("", "/", "", syscall.MS_SLAVE|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mounting / as slave %s", err) } if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mouting %s as bind %s", rootfs, err) } if readonly { if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mounting %s as readonly %s", rootfs, err) } } if err := mountSystem(rootfs); err != nil { return fmt.Errorf("mount system %s", err) } if err := copyDevNodes(rootfs); err != nil { return fmt.Errorf("copy dev nodes %s", err) } // In non-privileged mode, this fails. Discard the error. setupLoopbackDevices(rootfs) if err := setupDev(rootfs); err != nil { return err } if console != "" { if err := setupPtmx(rootfs, console); err != nil { return err } } if err := system.Chdir(rootfs); err != nil { return fmt.Errorf("chdir into %s %s", rootfs, err) } pivotDir, err := ioutil.TempDir(rootfs, ".pivot_root") if err != nil { return fmt.Errorf("can't create pivot_root dir %s", pivotDir, err) } if err := system.Pivotroot(rootfs, pivotDir); err != nil { return fmt.Errorf("pivot_root %s", err) } if err := system.Chdir("/"); err != nil { return fmt.Errorf("chdir / %s", err) } // path to pivot dir now changed, update pivotDir = filepath.Join("/", filepath.Base(pivotDir)) if err := system.Unmount(pivotDir, syscall.MNT_DETACH); err != nil { return fmt.Errorf("unmount pivot_root dir %s", err) } if err := os.Remove(pivotDir); err != nil { return fmt.Errorf("remove pivot_root dir %s", err) } system.Umask(0022) return nil }
// InitializeMountNamespace setups up the devices, mount points, and filesystems for use inside a // new mount namepsace func InitializeMountNamespace(rootfs, console string, container *libcontainer.Container) error { var ( err error flag = syscall.MS_PRIVATE ) if container.NoPivotRoot { flag = syscall.MS_SLAVE } if err := system.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil { return fmt.Errorf("mounting / with flags %X %s", (flag | syscall.MS_REC), err) } if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mouting %s as bind %s", rootfs, err) } if err := mountSystem(rootfs, container); err != nil { return fmt.Errorf("mount system %s", err) } if err := setupBindmounts(rootfs, container.Mounts); err != nil { return fmt.Errorf("bind mounts %s", err) } if err := nodes.CopyN(rootfs, nodes.DefaultNodes, true); err != nil { return fmt.Errorf("copy dev nodes %s", err) } if err := nodes.CopyN(rootfs, nodes.AdditionalNodes, false); err != nil { return fmt.Errorf("copy additional dev nodes %s", err) } if err := SetupPtmx(rootfs, console, container.Context["mount_label"]); err != nil { return err } if err := setupDevSymlinks(rootfs); err != nil { return fmt.Errorf("dev symlinks %s", err) } if err := system.Chdir(rootfs); err != nil { return fmt.Errorf("chdir into %s %s", rootfs, err) } if container.NoPivotRoot { err = MsMoveRoot(rootfs) } else { err = PivotRoot(rootfs) } if err != nil { return err } if container.ReadonlyFs { if err := SetReadonly(); err != nil { return fmt.Errorf("set readonly %s", err) } } system.Umask(0022) return nil }
// InitializeMountNamespace setups up the devices, mount points, and filesystems for use inside a // new mount namepsace func InitializeMountNamespace(rootfs, console string, container *libcontainer.Container) error { var ( err error flag = syscall.MS_PRIVATE ) if container.NoPivotRoot { flag = syscall.MS_SLAVE } if err := system.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil { return fmt.Errorf("mounting / as slave %s", err) } if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mouting %s as bind %s", rootfs, err) } if err := mountSystem(rootfs, container); err != nil { return fmt.Errorf("mount system %s", err) } if err := setupBindmounts(rootfs, container.Mounts); err != nil { return fmt.Errorf("bind mounts %s", err) } if err := nodes.CopyN(rootfs, nodes.DefaultNodes); err != nil { return fmt.Errorf("copy dev nodes %s", err) } if restrictionPath := container.Context["restriction_path"]; restrictionPath != "" { if err := restrict.Restrict(rootfs, restrictionPath); err != nil { return fmt.Errorf("restrict %s", err) } } if err := SetupPtmx(rootfs, console, container.Context["mount_label"]); err != nil { return err } if err := system.Chdir(rootfs); err != nil { return fmt.Errorf("chdir into %s %s", rootfs, err) } if container.NoPivotRoot { err = MsMoveRoot(rootfs) } else { err = PivotRoot(rootfs) } if err != nil { return err } if container.ReadonlyFs { if err := SetReadonly(); err != nil { return fmt.Errorf("set readonly %s", err) } } system.Umask(0022) return nil }
// This has to be called while the container still has CAP_SYS_ADMIN (to be able to perform mounts). // However, afterwards, CAP_SYS_ADMIN should be dropped (otherwise the user will be able to revert those changes). func Restrict(mounts ...string) error { // remount proc and sys as readonly for _, dest := range mounts { if err := system.Mount("", dest, "", syscall.MS_REMOUNT|syscall.MS_RDONLY, ""); err != nil { return fmt.Errorf("unable to remount %s readonly: %s", dest, err) } } if err := system.Mount("/dev/null", "/proc/kcore", "", syscall.MS_BIND, ""); err != nil { return fmt.Errorf("unable to bind-mount /dev/null over /proc/kcore") } return nil }
// InitializeMountNamespace sets up the devices, mount points, and filesystems for use inside a // new mount namespace. func InitializeMountNamespace(rootfs, console string, mountConfig *MountConfig) error { var ( err error flag = syscall.MS_PRIVATE ) if mountConfig.NoPivotRoot { flag = syscall.MS_SLAVE } if err := system.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil { return fmt.Errorf("mounting / with flags %X %s", (flag | syscall.MS_REC), err) } if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mouting %s as bind %s", rootfs, err) } if err := mountSystem(rootfs, mountConfig); err != nil { return fmt.Errorf("mount system %s", err) } if err := setupBindmounts(rootfs, mountConfig); err != nil { return fmt.Errorf("bind mounts %s", err) } if err := nodes.CreateDeviceNodes(rootfs, mountConfig.DeviceNodes); err != nil { return fmt.Errorf("create device nodes %s", err) } if err := SetupPtmx(rootfs, console, mountConfig.MountLabel); err != nil { return err } if err := setupDevSymlinks(rootfs); err != nil { return fmt.Errorf("dev symlinks %s", err) } if err := system.Chdir(rootfs); err != nil { return fmt.Errorf("chdir into %s %s", rootfs, err) } if mountConfig.NoPivotRoot { err = MsMoveRoot(rootfs) } else { err = PivotRoot(rootfs) } if err != nil { return err } if mountConfig.ReadonlyFs { if err := SetReadonly(); err != nil { return fmt.Errorf("set readonly %s", err) } } system.Umask(0022) return nil }
// setupNewMountNamespace is used to initialize a new mount namespace for an new // container in the rootfs that is specified. // // There is no need to unmount the new mounts because as soon as the mount namespace // is no longer in use, the mounts will be removed automatically func setupNewMountNamespace(rootfs, console string, readonly, noPivotRoot bool) error { flag := syscall.MS_PRIVATE if noPivotRoot { flag = syscall.MS_SLAVE } if err := system.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil { return fmt.Errorf("mounting / as slave %s", err) } if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mouting %s as bind %s", rootfs, err) } if readonly { if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mounting %s as readonly %s", rootfs, err) } } if err := mountSystem(rootfs); err != nil { return fmt.Errorf("mount system %s", err) } if err := copyDevNodes(rootfs); err != nil { return fmt.Errorf("copy dev nodes %s", err) } // In non-privileged mode, this fails. Discard the error. setupLoopbackDevices(rootfs) if err := setupDev(rootfs); err != nil { return err } if console != "" { if err := setupPtmx(rootfs, console); err != nil { return err } } if err := system.Chdir(rootfs); err != nil { return fmt.Errorf("chdir into %s %s", rootfs, err) } if noPivotRoot { if err := rootMsMove(rootfs); err != nil { return err } } else { if err := rootPivot(rootfs); err != nil { return err } } system.Umask(0022) return nil }
// setupConsole ensures that the container has a proper /dev/console setup func setupConsole(rootfs, console string) error { oldMask := system.Umask(0000) defer system.Umask(oldMask) stat, err := os.Stat(console) if err != nil { return fmt.Errorf("stat console %s %s", console, err) } var ( st = stat.Sys().(*syscall.Stat_t) dest = filepath.Join(rootfs, "dev/console") ) if err := os.Remove(dest); err != nil && !os.IsNotExist(err) { return fmt.Errorf("remove %s %s", dest, err) } if err := os.Chmod(console, 0600); err != nil { return err } if err := os.Chown(console, 0, 0); err != nil { return err } if err := system.Mknod(dest, (st.Mode&^07777)|0600, int(st.Rdev)); err != nil { return fmt.Errorf("mknod %s %s", dest, err) } if err := system.Mount(console, dest, "bind", syscall.MS_BIND, ""); err != nil { return fmt.Errorf("bind %s to %s %s", console, dest, err) } return nil }
// Setup initializes the proper /dev/console inside the rootfs path func Setup(rootfs, consolePath, mountLabel string) error { oldMask := system.Umask(0000) defer system.Umask(oldMask) if err := os.Chmod(consolePath, 0600); err != nil { return err } if err := os.Chown(consolePath, 0, 0); err != nil { return err } if err := label.SetFileLabel(consolePath, mountLabel); err != nil { return fmt.Errorf("set file label %s %s", consolePath, err) } dest := filepath.Join(rootfs, "dev/console") f, err := os.Create(dest) if err != nil && !os.IsExist(err) { return fmt.Errorf("create %s %s", dest, err) } if f != nil { f.Close() } if err := system.Mount(consolePath, dest, "bind", syscall.MS_BIND, ""); err != nil { return fmt.Errorf("bind %s to %s %s", consolePath, dest, err) } return nil }
func mountReadonly(path string) error { if err := system.Mount("", path, "", syscall.MS_REMOUNT|syscall.MS_RDONLY, ""); err != nil { if err == syscall.EINVAL { // Probably not a mountpoint, use bind-mount if err := system.Mount(path, path, "", syscall.MS_BIND, ""); err != nil { return err } if err := system.Mount(path, path, "", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC|defaultMountFlags, ""); err != nil { return err } } else { return err } } return nil }
// setupNewMountNamespace is used to initialize a new mount namespace for an new // container in the rootfs that is specified. // // There is no need to unmount the new mounts because as soon as the mount namespace // is no longer in use, the mounts will be removed automatically func setupNewMountNamespace(rootfs, console string, readonly bool) error { // mount as slave so that the new mounts do not propagate to the host if err := system.Mount("", "/", "", syscall.MS_SLAVE|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mounting / as slave %s", err) } if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mouting %s as bind %s", rootfs, err) } if readonly { if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mounting %s as readonly %s", rootfs, err) } } if err := mountSystem(rootfs); err != nil { return fmt.Errorf("mount system %s", err) } if err := copyDevNodes(rootfs); err != nil { return fmt.Errorf("copy dev nodes %s", err) } // In non-privileged mode, this fails. Discard the error. setupLoopbackDevices(rootfs) if err := setupDev(rootfs); err != nil { return err } if console != "" { if err := setupPtmx(rootfs, console); err != nil { return err } } if err := system.Chdir(rootfs); err != nil { return fmt.Errorf("chdir into %s %s", rootfs, err) } if err := system.Mount(rootfs, "/", "", syscall.MS_MOVE, ""); err != nil { return fmt.Errorf("mount move %s into / %s", rootfs, err) } if err := system.Chroot("."); err != nil { return fmt.Errorf("chroot . %s", err) } if err := system.Chdir("/"); err != nil { return fmt.Errorf("chdir / %s", err) } system.Umask(0022) return nil }
func setupBindmounts(rootfs string, mountConfig *MountConfig) error { bindMounts := mountConfig.Mounts for _, m := range bindMounts.OfType("bind") { var ( flags = syscall.MS_BIND | syscall.MS_REC dest = filepath.Join(rootfs, m.Destination) ) if !m.Writable { flags = flags | syscall.MS_RDONLY } stat, err := os.Stat(m.Source) if err != nil { return err } dest, err = symlink.FollowSymlinkInScope(dest, rootfs) if err != nil { return err } if err := createIfNotExists(dest, stat.IsDir()); err != nil { return fmt.Errorf("Creating new bind-mount target, %s", err) } if err := system.Mount(m.Source, dest, "bind", uintptr(flags), ""); err != nil { return fmt.Errorf("mounting %s into %s %s", m.Source, dest, err) } if !m.Writable { if err := system.Mount(m.Source, dest, "bind", uintptr(flags|syscall.MS_REMOUNT), ""); err != nil { return fmt.Errorf("remounting %s into %s %s", m.Source, dest, err) } } if m.Relabel != "" { if err := label.Relabel(m.Source, mountConfig.MountLabel, m.Relabel); err != nil { return fmt.Errorf("relabeling %s to %s %s", m.Source, mountConfig.MountLabel, err) } } if m.Private { if err := system.Mount("", dest, "none", uintptr(syscall.MS_PRIVATE), ""); err != nil { return fmt.Errorf("mounting %s private %s", dest, err) } } } return nil }
func MsMoveRoot(rootfs string) error { if err := system.Mount(rootfs, "/", "", syscall.MS_MOVE, ""); err != nil { return fmt.Errorf("mount move %s into / %s", rootfs, err) } if err := system.Chroot("."); err != nil { return fmt.Errorf("chroot . %s", err) } return system.Chdir("/") }
// remountProc is used to detach and remount the proc filesystem // commonly needed with running a new process inside an existing container func remountProc() error { if err := system.Unmount("/proc", syscall.MNT_DETACH); err != nil { return err } if err := system.Mount("proc", "/proc", "proc", uintptr(defaultMountFlags), ""); err != nil { return err } return nil }
// mountSystem sets up linux specific system mounts like sys, proc, shm, and devpts // inside the mount namespace func mountSystem(rootfs string, mountConfig *MountConfig) error { for _, m := range newSystemMounts(rootfs, mountConfig.MountLabel, mountConfig.Mounts) { if err := os.MkdirAll(m.path, 0755); err != nil && !os.IsExist(err) { return fmt.Errorf("mkdirall %s %s", m.path, err) } if err := system.Mount(m.source, m.path, m.device, uintptr(m.flags), m.data); err != nil { return fmt.Errorf("mounting %s into %s %s", m.source, m.path, err) } } return nil }
// mountSystem sets up linux specific system mounts like sys, proc, shm, and devpts // inside the mount namespace func mountSystem(rootfs string, container *libcontainer.Container) error { for _, m := range newSystemMounts(rootfs, container.Context["mount_label"], container.Mounts) { if err := os.MkdirAll(m.path, 0755); err != nil && !os.IsExist(err) { return fmt.Errorf("mkdirall %s %s", m.path, err) } if err := system.Mount(m.source, m.path, m.device, uintptr(m.flags), m.data); err != nil { return fmt.Errorf("mounting %s into %s %s", m.source, m.path, err) } } return nil }
// This has to be called while the container still has CAP_SYS_ADMIN (to be able to perform mounts). // However, afterwards, CAP_SYS_ADMIN should be dropped (otherwise the user will be able to revert those changes). func Restrict(mounts ...string) error { // remount proc and sys as readonly for _, dest := range mounts { if err := mountReadonly(dest); err != nil { return fmt.Errorf("unable to remount %s readonly: %s", dest, err) } } if err := system.Mount("/dev/null", "/proc/kcore", "", syscall.MS_BIND, ""); err != nil && !os.IsNotExist(err) { return fmt.Errorf("unable to bind-mount /dev/null over /proc/kcore: %s", err) } return nil }
func remountSys() error { if err := system.Unmount("/sys", syscall.MNT_DETACH); err != nil { if err != syscall.EINVAL { return err } } else { if err := system.Mount("sysfs", "/sys", "sysfs", uintptr(defaultMountFlags), ""); err != nil { return err } } return nil }
func mountReadonly(path string) error { for i := 0; i < 5; i++ { if err := system.Mount("", path, "", syscall.MS_REMOUNT|syscall.MS_RDONLY, ""); err != nil { switch err { case syscall.EINVAL: // Probably not a mountpoint, use bind-mount if err := system.Mount(path, path, "", syscall.MS_BIND, ""); err != nil { return err } return system.Mount(path, path, "", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC|defaultMountFlags, "") case syscall.EBUSY: time.Sleep(100 * time.Millisecond) continue default: return err } } return nil } return fmt.Errorf("unable to mount %s as readonly max retries reached", path) }
// Restrict locks down access to many areas of proc // by using the asumption that the user does not have mount caps to // revert the changes made here func Restrict(rootfs, empty string) error { for dest, source := range restrictions { dest = filepath.Join(rootfs, dest) // we don't have a "/dev/null" for dirs so have the requester pass a dir // for us to bind mount switch source { case "": source = empty default: source = filepath.Join(rootfs, source) } if err := system.Mount(source, dest, "bind", flags, ""); err != nil { if os.IsNotExist(err) { continue } return fmt.Errorf("unable to mount %s over %s %s", source, dest, err) } if err := system.Mount("", dest, "bind", flags|syscall.MS_REMOUNT, ""); err != nil { return fmt.Errorf("unable to mount %s over %s %s", source, dest, err) } } return nil }
// mountSystem sets up linux specific system mounts like sys, proc, shm, and devpts // inside the mount namespace func mountSystem(rootfs string) error { for _, m := range []struct { source string path string device string flags int data string }{ {source: "proc", path: filepath.Join(rootfs, "proc"), device: "proc", flags: defaultMountFlags}, {source: "sysfs", path: filepath.Join(rootfs, "sys"), device: "sysfs", flags: defaultMountFlags}, {source: "shm", path: filepath.Join(rootfs, "dev", "shm"), device: "tmpfs", flags: defaultMountFlags, data: "mode=1777,size=65536k"}, {source: "devpts", path: filepath.Join(rootfs, "dev", "pts"), device: "devpts", flags: syscall.MS_NOSUID | syscall.MS_NOEXEC, data: "newinstance,ptmxmode=0666,mode=620,gid=5"}, } { if err := os.MkdirAll(m.path, 0755); err != nil && !os.IsExist(err) { return fmt.Errorf("mkdirall %s %s", m.path, err) } if err := system.Mount(m.source, m.path, m.device, uintptr(m.flags), m.data); err != nil { return fmt.Errorf("mounting %s into %s %s", m.source, m.path, err) } } return nil }
func SetReadonly() error { return system.Mount("/", "/", "bind", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC, "") }
// setupNewMountNamespace is used to initialize a new mount namespace for an new // container in the rootfs that is specified. // // There is no need to unmount the new mounts because as soon as the mount namespace // is no longer in use, the mounts will be removed automatically func setupNewMountNamespace(rootfs string, bindMounts []libcontainer.Mount, console string, readonly, noPivotRoot bool, mountLabel string) error { flag := syscall.MS_PRIVATE if noPivotRoot { flag = syscall.MS_SLAVE } if err := system.Mount("", "/", "", uintptr(flag|syscall.MS_REC), ""); err != nil { return fmt.Errorf("mounting / as slave %s", err) } if err := system.Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mouting %s as bind %s", rootfs, err) } if err := mountSystem(rootfs, mountLabel); err != nil { return fmt.Errorf("mount system %s", err) } for _, m := range bindMounts { flags := syscall.MS_BIND | syscall.MS_REC if !m.Writable { flags = flags | syscall.MS_RDONLY } dest := filepath.Join(rootfs, m.Destination) if err := system.Mount(m.Source, dest, "bind", uintptr(flags), ""); err != nil { return fmt.Errorf("mounting %s into %s %s", m.Source, dest, err) } if m.Private { if err := system.Mount("", dest, "none", uintptr(syscall.MS_PRIVATE), ""); err != nil { return fmt.Errorf("mounting %s private %s", dest, err) } } } if err := copyDevNodes(rootfs); err != nil { return fmt.Errorf("copy dev nodes %s", err) } // In non-privileged mode, this fails. Discard the error. setupLoopbackDevices(rootfs) if err := setupPtmx(rootfs, console, mountLabel); err != nil { return err } if err := system.Chdir(rootfs); err != nil { return fmt.Errorf("chdir into %s %s", rootfs, err) } if noPivotRoot { if err := rootMsMove(rootfs); err != nil { return err } } else { if err := rootPivot(rootfs); err != nil { return err } } if readonly { if err := system.Mount("/", "/", "bind", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC, ""); err != nil { return fmt.Errorf("mounting %s as readonly %s", rootfs, err) } } system.Umask(0022) return nil }