Improve general quality of libcontainer

Docker-DCO-1.1-Signed-off-by: Guillaume J. Charmes <guillaume.charmes@docker.com> (github: creack)
This commit is contained in:
Guillaume J. Charmes 2014-02-18 23:13:36 -08:00 committed by Michael Crosby
parent 1142945769
commit 93d41e53ae
12 changed files with 159 additions and 238 deletions

View file

@ -10,6 +10,9 @@ import (
"github.com/dotcloud/docker/pkg/libcontainer/network"
"github.com/dotcloud/docker/pkg/libcontainer/utils"
"os"
exec_ "os/exec"
"path"
"path/filepath"
)
var (
@ -52,6 +55,18 @@ func exec(container *libcontainer.Container) error {
container.NetNsFd = netFile.Fd()
}
self, err := exec_.LookPath(os.Args[0])
if err != nil {
return err
}
if output, err := exec_.Command("cp", self, path.Join(container.RootFs, ".nsinit")).CombinedOutput(); err != nil {
return fmt.Errorf("Error exec cp: %s, (%s)", err, output)
} else {
println(self, container.RootFs)
fmt.Printf("-----> %s\n", output)
}
println("----")
pid, err := namespaces.ExecContainer(container)
if err != nil {
return fmt.Errorf("error exec container %s", err)
@ -77,25 +92,25 @@ func exec(container *libcontainer.Container) error {
}
func execIn(container *libcontainer.Container) error {
f, err := os.Open("/root/nsroot/test")
if err != nil {
return err
}
container.NetNsFd = f.Fd()
pid, err := namespaces.ExecIn(container, &libcontainer.Command{
Env: container.Command.Env,
Args: []string{
newCommand,
},
})
if err != nil {
return fmt.Errorf("error exexin container %s", err)
}
exitcode, err := utils.WaitOnPid(pid)
if err != nil {
return fmt.Errorf("error waiting on child %s", err)
}
os.Exit(exitcode)
// f, err := os.Open("/root/nsroot/test")
// if err != nil {
// return err
// }
// container.NetNsFd = f.Fd()
// pid, err := namespaces.ExecIn(container, &libcontainer.Command{
// Env: container.Command.Env,
// Args: []string{
// newCommand,
// },
// })
// if err != nil {
// return fmt.Errorf("error exexin container %s", err)
// }
// exitcode, err := utils.WaitOnPid(pid)
// if err != nil {
// return fmt.Errorf("error waiting on child %s", err)
// }
// os.Exit(exitcode)
return nil
}
@ -143,11 +158,13 @@ func printErr(err error) {
}
func main() {
var (
err error
cliCmd = flag.Arg(0)
config = "/root/development/gocode/src/github.com/dotcloud/docker/pkg/libcontainer/container.json" //flag.Arg(1)
)
cliCmd := flag.Arg(0)
config, err := filepath.Abs(flag.Arg(1))
if err != nil {
printErr(err)
}
println("cli:", cliCmd, "config:", config)
f, err := os.Open(config)
if err != nil {
printErr(err)

View file

@ -12,7 +12,7 @@
"TERM=xterm"
]
},
"rootfs": "/root/main/mycontainer",
"rootfs": "/var/lib/docker/containers/ee76122136d691d63e09d24168a91ddb2ef9fdcf210b4de5c50aa76354892f4b/root",
"namespaces": [
"NEWIPC",
"NEWNS",

View file

@ -6,8 +6,8 @@ package namespaces
import (
"errors"
"fmt"
"github.com/dotcloud/docker/pkg/libcontainer"
"github.com/dotcloud/docker/pkg/system"
"io"
"log"
"os"
@ -44,12 +44,15 @@ func ExecContainer(container *libcontainer.Container) (pid int, err error) {
// we need CLONE_VFORK so we can wait on the child
flag := uintptr(getNamespaceFlags(container.Namespaces) | CLONE_VFORK)
command := exec.Command(nsinit, "-master", strconv.Itoa(int(master.Fd())), "-console", console, "init")
command := exec.Command(nsinit, "-master", strconv.Itoa(int(master.Fd())), "-console", console, "init", "container.json")
// command.Stdin = os.Stdin
// command.Stdout = os.Stdout
// command.Stderr = os.Stderr
command.SysProcAttr = &syscall.SysProcAttr{}
command.SysProcAttr.Cloneflags = flag
command.ExtraFiles = []*os.File{master}
// command.SysProcAttr.Setctty = true
//command.ExtraFiles = []*os.File{master}
println("vvvvvvvvv")
if err := command.Start(); err != nil {
return -1, err
}
@ -68,104 +71,18 @@ func ExecContainer(container *libcontainer.Container) (pid int, err error) {
return pid, nil
}
// ExecIn will spawn a new command inside an existing container's namespaces. The existing container's
// pid and namespace configuration is needed along with the specific capabilities that should
// be dropped once inside the namespace.
func ExecIn(container *libcontainer.Container, cmd *libcontainer.Command) (int, error) {
return -1, fmt.Errorf("not implemented")
/*
if container.NsPid <= 0 {
return -1, libcontainer.ErrInvalidPid
}
fds, err := getNsFds(container)
if err != nil {
return -1, err
}
if container.NetNsFd > 0 {
fds = append(fds, container.NetNsFd)
}
pid, err := fork()
if err != nil {
for _, fd := range fds {
syscall.Close(int(fd))
}
return -1, err
}
if pid == 0 {
for _, fd := range fds {
if fd > 0 {
if err := JoinExistingNamespace(fd, ""); err != nil {
for _, fd := range fds {
syscall.Close(int(fd))
}
writeError("join existing namespace for %d %s", fd, err)
}
}
syscall.Close(int(fd))
}
if container.Namespaces.Contains(libcontainer.CLONE_NEWNS) &&
container.Namespaces.Contains(libcontainer.CLONE_NEWPID) {
// important:
//
// we need to fork and unshare so that re can remount proc and sys within
// the namespace so the CLONE_NEWPID namespace will take effect
// if we don't fork we would end up unmounting proc and sys for the entire
// namespace
child, err := fork()
if err != nil {
writeError("fork child %s", err)
}
if child == 0 {
if err := unshare(CLONE_NEWNS); err != nil {
writeError("unshare newns %s", err)
}
if err := remountProc(); err != nil {
writeError("remount proc %s", err)
}
if err := remountSys(); err != nil {
writeError("remount sys %s", err)
}
if err := capabilities.DropCapabilities(container); err != nil {
writeError("drop caps %s", err)
}
if err := exec(cmd.Args[0], cmd.Args[0:], cmd.Env); err != nil {
writeError("exec %s", err)
}
panic("unreachable")
}
exit, err := utils.WaitOnPid(child)
if err != nil {
writeError("wait on child %s", err)
}
os.Exit(exit)
}
if err := exec(cmd.Args[0], cmd.Args[0:], cmd.Env); err != nil {
writeError("exec %s", err)
}
panic("unreachable")
}
return pid, err
*/
}
func createMasterAndConsole() (*os.File, string, error) {
master, err := os.OpenFile("/dev/ptmx", syscall.O_RDWR|syscall.O_NOCTTY|syscall.O_CLOEXEC, 0)
if err != nil {
return nil, "", err
}
console, err := Ptsname(master)
console, err := system.Ptsname(master)
if err != nil {
return nil, "", err
}
if err := Unlockpt(master); err != nil {
if err := system.Unlockpt(master); err != nil {
return nil, "", err
}
return master, console, nil

View file

@ -2,6 +2,7 @@ package namespaces
import (
"fmt"
"github.com/dotcloud/docker/pkg/system"
"log"
"os"
"path/filepath"
@ -14,16 +15,16 @@ var (
)
func SetupNewMountNamespace(rootfs, console string, readonly bool) error {
if err := Mount("", "/", "", syscall.MS_SLAVE|syscall.MS_REC, ""); err != nil {
if err := system.Mount("", "/", "", syscall.MS_SLAVE|syscall.MS_REC, ""); err != nil {
return fmt.Errorf("mounting / as slave %s", err)
}
if err := Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REC, ""); err != nil {
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 := Mount(rootfs, rootfs, "bind", syscall.MS_BIND|syscall.MS_REMOUNT|syscall.MS_RDONLY|syscall.MS_REC, ""); err != nil {
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)
}
}
@ -52,29 +53,30 @@ func SetupNewMountNamespace(rootfs, console string, readonly bool) error {
return err
}
if err := Chdir(rootfs); err != nil {
if err := system.Chdir(rootfs); err != nil {
return fmt.Errorf("chdir into %s %s", rootfs, err)
}
if err := Mount(rootfs, "/", "", syscall.MS_MOVE, ""); err != nil {
if err := system.Mount(rootfs, "/", "", syscall.MS_MOVE, ""); err != nil {
return fmt.Errorf("mount move %s into / %s", rootfs, err)
}
if err := Chroot("."); err != nil {
if err := system.Chroot("."); err != nil {
return fmt.Errorf("chroot . %s", err)
}
if err := Chdir("/"); err != nil {
if err := system.Chdir("/"); err != nil {
return fmt.Errorf("chdir / %s", err)
}
Umask(0022)
system.Umask(0022)
return nil
}
func copyDevNodes(rootfs string) error {
Umask(0000)
oldMask := system.Umask(0000)
defer system.Umask(oldMask)
for _, node := range []string{
"null",
@ -95,7 +97,7 @@ func copyDevNodes(rootfs string) error {
)
log.Printf("copy %s to %s %d\n", node, dest, st.Rdev)
if err := Mknod(dest, st.Mode, int(st.Rdev)); err != nil && !os.IsExist(err) {
if err := system.Mknod(dest, st.Mode, int(st.Rdev)); err != nil && !os.IsExist(err) {
return fmt.Errorf("copy %s %s", node, err)
}
}
@ -125,7 +127,8 @@ func setupDev(rootfs string) error {
}
func setupConsole(rootfs, console string) error {
Umask(0000)
oldMask := system.Umask(0000)
defer system.Umask(oldMask)
stat, err := os.Stat(console)
if err != nil {
@ -145,11 +148,11 @@ func setupConsole(rootfs, console string) error {
return err
}
if err := Mknod(dest, (st.Mode&^07777)|0600, int(st.Rdev)); err != nil {
if err := system.Mknod(dest, (st.Mode&^07777)|0600, int(st.Rdev)); err != nil {
return fmt.Errorf("mknod %s %s", dest, err)
}
if err := Mount(console, dest, "bind", syscall.MS_BIND, ""); err != nil {
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
@ -158,7 +161,7 @@ func setupConsole(rootfs, console string) error {
// mountSystem sets up linux specific system mounts like sys, proc, shm, and devpts
// inside the mount namespace
func mountSystem(rootfs string) error {
mounts := []struct {
for _, m := range []struct {
source string
path string
device string
@ -171,12 +174,11 @@ func mountSystem(rootfs string) error {
{source: "shm", path: filepath.Join(rootfs, "dev", "shm"), device: "tmpfs", flags: defaults, data: "mode=1777"},
{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"},
{source: "tmpfs", path: filepath.Join(rootfs, "run"), device: "tmpfs", flags: syscall.MS_NOSUID | syscall.MS_NODEV | syscall.MS_STRICTATIME, data: "mode=755"},
}
for _, m := range mounts {
} {
if err := os.MkdirAll(m.path, 0755); err != nil && !os.IsExist(err) {
return fmt.Errorf("mkdirall %s %s", m.path, err)
}
if err := Mount(m.source, m.path, m.device, uintptr(m.flags), m.data); err != nil {
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)
}
}
@ -184,22 +186,22 @@ func mountSystem(rootfs string) error {
}
func remountProc() error {
if err := Unmount("/proc", syscall.MNT_DETACH); err != nil {
if err := system.Unmount("/proc", syscall.MNT_DETACH); err != nil {
return err
}
if err := Mount("proc", "/proc", "proc", uintptr(defaults), ""); err != nil {
if err := system.Mount("proc", "/proc", "proc", uintptr(defaults), ""); err != nil {
return err
}
return nil
}
func remountSys() error {
if err := Unmount("/sys", syscall.MNT_DETACH); err != nil {
if err := system.Unmount("/sys", syscall.MNT_DETACH); err != nil {
if err != syscall.EINVAL {
return err
}
} else {
if err := Mount("sysfs", "/sys", "sysfs", uintptr(defaults), ""); err != nil {
if err := system.Mount("sysfs", "/sys", "sysfs", uintptr(defaults), ""); err != nil {
return err
}
}

View file

@ -1,32 +0,0 @@
/*
TODO
pivot root
cgroups
more mount stuff that I probably am forgetting
apparmor
*/
package namespaces
import (
"github.com/dotcloud/docker/pkg/libcontainer"
)
// JoinExistingNamespace uses the fd of an existing linux namespace and
// has the current process join that namespace or the spacespace specified by ns
func JoinExistingNamespace(fd uintptr, ns libcontainer.Namespace) error {
flag := namespaceMap[ns]
if err := Setns(fd, uintptr(flag)); err != nil {
return err
}
return nil
}
// getNamespaceFlags parses the container's Namespaces options to set the correct
// flags on clone, unshare, and setns
func getNamespaceFlags(namespaces libcontainer.Namespaces) (flag int) {
for _, ns := range namespaces {
flag |= namespaceMap[ns]
}
return
}

View file

@ -33,3 +33,12 @@ var namespaceFileMap = map[libcontainer.Namespace]string{
libcontainer.CLONE_NEWPID: "pid",
libcontainer.CLONE_NEWNET: "net",
}
// getNamespaceFlags parses the container's Namespaces options to set the correct
// flags on clone, unshare, and setns
func getNamespaceFlags(namespaces libcontainer.Namespaces) (flag int) {
for _, ns := range namespaces {
flag |= namespaceMap[ns]
}
return
}

View file

@ -5,6 +5,7 @@ import (
"github.com/dotcloud/docker/pkg/libcontainer"
"github.com/dotcloud/docker/pkg/libcontainer/capabilities"
"github.com/dotcloud/docker/pkg/libcontainer/namespaces"
"github.com/dotcloud/docker/pkg/system"
"log"
"os"
"path/filepath"
@ -14,10 +15,12 @@ import (
// InitNamespace should be run inside an existing namespace to setup
// common mounts, drop capabilities, and setup network interfaces
func InitNamespace(container *libcontainer.Container) error {
println("|||||||||||||")
if err := setLogFile(container); err != nil {
return err
}
println(container.LogFile)
log.Printf("--------->")
rootfs, err := resolveRootfs(container)
if err != nil {
return err
@ -26,7 +29,7 @@ func InitNamespace(container *libcontainer.Container) error {
// any errors encoutered inside the namespace we should write
// out to a log or a pipe to our parent and exit(1)
// because writing to stderr will not work after we close
if err := closeMasterAndStd(container.Master); err != nil {
if err := closeMasterAndStd(os.NewFile(container.Master, "/dev/ptmx")); err != nil {
log.Fatalf("close master and std %s", err)
return err
}
@ -49,15 +52,15 @@ func InitNamespace(container *libcontainer.Container) error {
}
*/
if _, err := namespaces.Setsid(); err != nil {
if _, err := system.Setsid(); err != nil {
log.Fatalf("setsid %s", err)
return err
}
if err := namespaces.Setctty(); err != nil {
if err := system.Setctty(); err != nil {
log.Fatalf("setctty %s", err)
return err
}
if err := namespaces.ParentDeathSignal(); err != nil {
if err := system.ParentDeathSignal(); err != nil {
log.Fatalf("parent deth signal %s", err)
return err
}
@ -65,7 +68,7 @@ func InitNamespace(container *libcontainer.Container) error {
log.Fatalf("setup mount namespace %s", err)
return err
}
if err := namespaces.Sethostname(container.ID); err != nil {
if err := system.Sethostname(container.ID); err != nil {
log.Fatalf("sethostname %s", err)
return err
}
@ -78,12 +81,12 @@ func InitNamespace(container *libcontainer.Container) error {
return err
}
if container.WorkingDir != "" {
if err := namespaces.Chdir(container.WorkingDir); err != nil {
if err := system.Chdir(container.WorkingDir); err != nil {
log.Fatalf("chdir to %s %s", container.WorkingDir, err)
return err
}
}
if err := namespaces.Exec(container.Command.Args[0], container.Command.Args[0:], container.Command.Env); err != nil {
if err := system.Exec(container.Command.Args[0], container.Command.Args[0:], container.Command.Env); err != nil {
log.Fatalf("exec %s", err)
return err
}
@ -98,24 +101,23 @@ func resolveRootfs(container *libcontainer.Container) (string, error) {
return filepath.EvalSymlinks(rootfs)
}
func closeMasterAndStd(master uintptr) error {
namespaces.Closefd(master)
namespaces.Closefd(0)
namespaces.Closefd(1)
namespaces.Closefd(2)
func closeMasterAndStd(master *os.File) error {
master.Close()
os.Stdin.Close()
os.Stdout.Close()
os.Stderr.Close()
return nil
}
func setupUser(container *libcontainer.Container) error {
// TODO: honor user passed on container
if err := namespaces.Setgroups(nil); err != nil {
if err := system.Setgroups(nil); err != nil {
return err
}
if err := namespaces.Setresgid(0, 0, 0); err != nil {
if err := system.Setresgid(0, 0, 0); err != nil {
return err
}
if err := namespaces.Setresuid(0, 0, 0); err != nil {
if err := system.Setresuid(0, 0, 0); err != nil {
return err
}
return nil
@ -126,15 +128,16 @@ func dupSlave(slave *os.File) error {
if slave.Fd() != 0 {
return fmt.Errorf("slave fd not 0 %d", slave.Fd())
}
if err := namespaces.Dup2(slave.Fd(), 1); err != nil {
if err := system.Dup2(slave.Fd(), 1); err != nil {
return err
}
if err := namespaces.Dup2(slave.Fd(), 2); err != nil {
if err := system.Dup2(slave.Fd(), 2); err != nil {
return err
}
return nil
}
// openTerminal is a clone of os.OpenFile without the O_CLOEXEC addition.
func openTerminal(name string, flag int) (*os.File, error) {
r, e := syscall.Open(name, flag, 0)
if e != nil {

View file

@ -72,17 +72,3 @@ func setupEnvironment(container *libcontainer.Container) {
addEnvIfNotSet(container, "USER", "root")
addEnvIfNotSet(container, "LOGNAME", "root")
}
func getMasterAndConsole(container *libcontainer.Container) (string, *os.File, error) {
master, err := Openpmtx()
if err != nil {
return "", nil, err
}
console, err := Ptsname(master)
if err != nil {
master.Close()
return "", nil, err
}
return console, master, nil
}

View file

@ -1,15 +1,7 @@
package namespaces
package system
import (
"fmt"
"os"
"syscall"
"unsafe"
)
const (
TIOCGPTN = 0x80045430
TIOCSPTLCK = 0x40045431
)
func Chroot(dir string) error {
@ -60,14 +52,6 @@ func Clone(flags uintptr) (int, error) {
return int(pid), nil
}
func Setns(fd uintptr, flags uintptr) error {
_, _, err := syscall.RawSyscall(SYS_SETNS, fd, flags, 0)
if err != 0 {
return err
}
return nil
}
func UsetCloseOnExec(fd uintptr) error {
if _, _, err := syscall.Syscall(syscall.SYS_FCNTL, fd, syscall.F_SETFD, 0); err != 0 {
return err
@ -95,11 +79,6 @@ func Setsid() (int, error) {
return syscall.Setsid()
}
func Unlockpt(f *os.File) error {
var u int
return Ioctl(f.Fd(), TIOCSPTLCK, uintptr(unsafe.Pointer(&u)))
}
func Ioctl(fd uintptr, flag, data uintptr) error {
if _, _, err := syscall.Syscall(syscall.SYS_IOCTL, fd, flag, data); err != 0 {
return err
@ -107,18 +86,6 @@ func Ioctl(fd uintptr, flag, data uintptr) error {
return nil
}
func Ptsname(f *os.File) (string, error) {
var n int
if err := Ioctl(f.Fd(), TIOCGPTN, uintptr(unsafe.Pointer(&n))); err != nil {
return "", err
}
return fmt.Sprintf("/dev/pts/%d", n), nil
}
func Openpmtx() (*os.File, error) {
return os.OpenFile("/dev/ptmx", syscall.O_RDONLY|syscall.O_NOCTTY|syscall.O_CLOEXEC, 0)
}
func Closefd(fd uintptr) error {
return syscall.Close(int(fd))
}
@ -132,7 +99,7 @@ func Mknod(path string, mode uint32, dev int) error {
}
func ParentDeathSignal() error {
if _, _, err := syscall.RawSyscall6(syscall.SYS_PRCTL, syscall.PR_SET_PDEATHSIG, uintptr(syscall.SIGKILL), 0, 0, 0, 0); err != 0 {
if _, _, err := syscall.RawSyscall(syscall.SYS_PRCTL, syscall.PR_SET_PDEATHSIG, uintptr(syscall.SIGKILL), 0); err != 0 {
return err
}
return nil

31
pkg/system/pty_linux.go Normal file
View file

@ -0,0 +1,31 @@
package system
import (
"fmt"
"os"
"syscall"
"unsafe"
)
// Unlockpt unlocks the slave pseudoterminal device corresponding to the master pseudoterminal referred to by f.
// Unlockpt should be called before opening the slave side of a pseudoterminal.
func Unlockpt(f *os.File) error {
var u int
return Ioctl(f.Fd(), syscall.TIOCSPTLCK, uintptr(unsafe.Pointer(&u)))
}
// Ptsname retrieves the name of the first available pts for the given master.
func Ptsname(f *os.File) (string, error) {
var n int
if err := Ioctl(f.Fd(), syscall.TIOCGPTN, uintptr(unsafe.Pointer(&n))); err != nil {
return "", err
}
return fmt.Sprintf("/dev/pts/%d", n), nil
}
// OpenPtmx opens /dev/ptmx, i.e. the PTY master.
func OpenPtmx() (*os.File, error) {
// O_NOCTTY and O_CLOEXEC are not present in os package so we use the syscall's one for all.
return os.OpenFile("/dev/ptmx", syscall.O_RDONLY|syscall.O_NOCTTY|syscall.O_CLOEXEC, 0)
}

13
pkg/system/setns_linux.go Normal file
View file

@ -0,0 +1,13 @@
package system
import (
"syscall"
)
func Setns(fd uintptr, flags uintptr) error {
_, _, err := syscall.RawSyscall(SYS_SETNS, fd, flags, 0)
if err != 0 {
return err
}
return nil
}

View file

@ -0,0 +1,8 @@
// +build linux,amd64
package system
// Via http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=7b21fddd087678a70ad64afc0f632e0f1071b092
const (
SYS_SETNS = 308
)