update libcontainer,gocapability,cadvisor

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

@@ -37,7 +37,7 @@ type Manager interface {
 	// restore the object later.
 	GetPaths() map[string]string
 
-	// Set the cgroup as configured.
+	// Sets the cgroup as configured.
 	Set(container *configs.Config) error
 }
 

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

@@ -104,6 +104,8 @@ func (m *Manager) Apply(pid int) (err error) {
 	if m.Cgroups == nil {
 		return nil
 	}
+	m.mu.Lock()
+	defer m.mu.Unlock()
 
 	var c = m.Cgroups
 
@@ -128,8 +130,6 @@ func (m *Manager) Apply(pid int) (err error) {
 		return cgroups.EnterPid(m.Paths, pid)
 	}
 
-	m.mu.Lock()
-	defer m.mu.Unlock()
 	paths := make(map[string]string)
 	for _, sys := range subsystems {
 		if err := sys.Apply(d); err != nil {
@@ -195,18 +195,10 @@ func (m *Manager) Set(container *configs.Config) error {
 	if m.Cgroups.Paths != nil {
 		return nil
 	}
-	for _, sys := range subsystems {
-		// Generate fake cgroup data.
-		d, err := getCgroupData(container.Cgroups, -1)
-		if err != nil {
-			return err
-		}
-		// Get the path, but don't error out if the cgroup wasn't found.
-		path, err := d.path(sys.Name())
-		if err != nil && !cgroups.IsNotFound(err) {
-			return err
-		}
 
+	paths := m.GetPaths()
+	for _, sys := range subsystems {
+		path := paths[sys.Name()]
 		if err := sys.Set(path, container.Cgroups); err != nil {
 			return err
 		}
@@ -223,14 +215,8 @@ func (m *Manager) Set(container *configs.Config) error {
 // Freeze toggles the container's freezer cgroup depending on the state
 // provided
 func (m *Manager) Freeze(state configs.FreezerState) error {
-	d, err := getCgroupData(m.Cgroups, 0)
-	if err != nil {
-		return err
-	}
-	dir, err := d.path("freezer")
-	if err != nil {
-		return err
-	}
+	paths := m.GetPaths()
+	dir := paths["freezer"]
 	prevState := m.Cgroups.Resources.Freezer
 	m.Cgroups.Resources.Freezer = state
 	freezer, err := subsystems.Get("freezer")
@@ -246,28 +232,13 @@ func (m *Manager) Freeze(state configs.FreezerState) error {
 }
 
 func (m *Manager) GetPids() ([]int, error) {
-	dir, err := getCgroupPath(m.Cgroups)
-	if err != nil {
-		return nil, err
-	}
-	return cgroups.GetPids(dir)
+	paths := m.GetPaths()
+	return cgroups.GetPids(paths["devices"])
 }
 
 func (m *Manager) GetAllPids() ([]int, error) {
-	dir, err := getCgroupPath(m.Cgroups)
-	if err != nil {
-		return nil, err
-	}
-	return cgroups.GetAllPids(dir)
-}
-
-func getCgroupPath(c *configs.Cgroup) (string, error) {
-	d, err := getCgroupData(c, 0)
-	if err != nil {
-		return "", err
-	}
-
-	return d.path("devices")
+	paths := m.GetPaths()
+	return cgroups.GetAllPids(paths["devices"])
 }
 
 func getCgroupData(c *configs.Cgroup, pid int) (*cgroupData, error) {

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

@@ -22,10 +22,48 @@ func (s *CpuGroup) Name() string {
 func (s *CpuGroup) Apply(d *cgroupData) error {
 	// We always want to join the cpu group, to allow fair cpu scheduling
 	// on a container basis
-	_, err := d.join("cpu")
+	path, err := d.path("cpu")
 	if err != nil && !cgroups.IsNotFound(err) {
 		return err
 	}
+	return s.ApplyDir(path, d.config, d.pid)
+}
+
+func (s *CpuGroup) ApplyDir(path string, cgroup *configs.Cgroup, pid int) error {
+	// This might happen if we have no cpu cgroup mounted.
+	// Just do nothing and don't fail.
+	if path == "" {
+		return nil
+	}
+	if err := os.MkdirAll(path, 0755); err != nil {
+		return err
+	}
+	// We should set the real-Time group scheduling settings before moving
+	// in the process because if the process is already in SCHED_RR mode
+	// and no RT bandwidth is set, adding it will fail.
+	if err := s.SetRtSched(path, cgroup); err != nil {
+		return err
+	}
+	// because we are not using d.join we need to place the pid into the procs file
+	// unlike the other subsystems
+	if err := cgroups.WriteCgroupProc(path, pid); err != nil {
+		return err
+	}
+
+	return nil
+}
+
+func (s *CpuGroup) SetRtSched(path string, cgroup *configs.Cgroup) error {
+	if cgroup.Resources.CpuRtPeriod != 0 {
+		if err := writeFile(path, "cpu.rt_period_us", strconv.FormatInt(cgroup.Resources.CpuRtPeriod, 10)); err != nil {
+			return err
+		}
+	}
+	if cgroup.Resources.CpuRtRuntime != 0 {
+		if err := writeFile(path, "cpu.rt_runtime_us", strconv.FormatInt(cgroup.Resources.CpuRtRuntime, 10)); err != nil {
+			return err
+		}
+	}
 	return nil
 }
 
@@ -45,15 +83,8 @@ func (s *CpuGroup) Set(path string, cgroup *configs.Cgroup) error {
 			return err
 		}
 	}
-	if cgroup.Resources.CpuRtPeriod != 0 {
-		if err := writeFile(path, "cpu.rt_period_us", strconv.FormatInt(cgroup.Resources.CpuRtPeriod, 10)); err != nil {
-			return err
-		}
-	}
-	if cgroup.Resources.CpuRtRuntime != 0 {
-		if err := writeFile(path, "cpu.rt_runtime_us", strconv.FormatInt(cgroup.Resources.CpuRtRuntime, 10)); err != nil {
-			return err
-		}
+	if err := s.SetRtSched(path, cgroup); err != nil {
+		return err
 	}
 
 	return nil

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

@@ -16,9 +16,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/configs"
 )
 
-const (
-	cgroupKernelMemoryLimit = "memory.kmem.limit_in_bytes"
-)
+const cgroupKernelMemoryLimit = "memory.kmem.limit_in_bytes"
 
 type MemoryGroup struct {
 }
@@ -38,8 +36,10 @@ func (s *MemoryGroup) Apply(d *cgroupData) (err error) {
 				return err
 			}
 		}
-		if err := EnableKernelMemoryAccounting(path); err != nil {
-			return err
+		if d.config.KernelMemory != 0 {
+			if err := EnableKernelMemoryAccounting(path); err != nil {
+				return err
+			}
 		}
 	}
 	defer func() {
@@ -62,13 +62,10 @@ func EnableKernelMemoryAccounting(path string) error {
 	// We have to limit the kernel memory here as it won't be accounted at all
 	// until a limit is set on the cgroup and limit cannot be set once the
 	// cgroup has children, or if there are already tasks in the cgroup.
-	kernelMemoryLimit := int64(1)
-	if err := setKernelMemory(path, kernelMemoryLimit); err != nil {
-		return err
-	}
-	kernelMemoryLimit = int64(-1)
-	if err := setKernelMemory(path, kernelMemoryLimit); err != nil {
-		return err
+	for _, i := range []int64{1, -1} {
+		if err := setKernelMemory(path, i); err != nil {
+			return err
+		}
 	}
 	return nil
 }

vendor/github.com/opencontainers/runc/libcontainer/cgroups/systemd/apply_systemd.go

@@ -8,7 +8,6 @@ import (
 	"io/ioutil"
 	"os"
 	"path/filepath"
-	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -67,12 +66,14 @@ var subsystems = subsystemSet{
 
 const (
 	testScopeWait = 4
+	testSliceWait = 4
 )
 
 var (
 	connLock                        sync.Mutex
 	theConn                         *systemdDbus.Conn
 	hasStartTransientUnit           bool
+	hasStartTransientSliceUnit      bool
 	hasTransientDefaultDependencies bool
 	hasDelegate                     bool
 )
@@ -159,8 +160,36 @@ func UseSystemd() bool {
 			}
 		}
 
+		// Assume we have the ability to start a transient unit as a slice
+		// This was broken until systemd v229, but has been back-ported on RHEL environments >= 219
+		// For details, see: https://bugzilla.redhat.com/show_bug.cgi?id=1370299
+		hasStartTransientSliceUnit = true
+
+		// To ensure simple clean-up, we create a slice off the root with no hierarchy
+		slice := fmt.Sprintf("libcontainer_%d_systemd_test_default.slice", os.Getpid())
+		if _, err := theConn.StartTransientUnit(slice, "replace", nil, nil); err != nil {
+			if _, ok := err.(dbus.Error); ok {
+				hasStartTransientSliceUnit = false
+			}
+		}
+
+		for i := 0; i <= testSliceWait; i++ {
+			if _, err := theConn.StopUnit(slice, "replace", nil); err != nil {
+				if dbusError, ok := err.(dbus.Error); ok {
+					if strings.Contains(dbusError.Name, "org.freedesktop.systemd1.NoSuchUnit") {
+						hasStartTransientSliceUnit = false
+						break
+					}
+				}
+			} else {
+				break
+			}
+			time.Sleep(time.Millisecond)
+		}
+
 		// Not critical because of the stop unit logic above.
 		theConn.StopUnit(scope, "replace", nil)
+		theConn.StopUnit(slice, "replace", nil)
 	}
 	return hasStartTransientUnit
 }
@@ -194,11 +223,24 @@ func (m *Manager) Apply(pid int) error {
 		slice = c.Parent
 	}
 
-	properties = append(properties,
-		systemdDbus.PropSlice(slice),
-		systemdDbus.PropDescription("docker container "+c.Name),
-		newProp("PIDs", []uint32{uint32(pid)}),
-	)
+	properties = append(properties, systemdDbus.PropDescription("libcontainer container "+c.Name))
+
+	// if we create a slice, the parent is defined via a Wants=
+	if strings.HasSuffix(unitName, ".slice") {
+		// This was broken until systemd v229, but has been back-ported on RHEL environments >= 219
+		if !hasStartTransientSliceUnit {
+			return fmt.Errorf("systemd version does not support ability to start a slice as transient unit")
+		}
+		properties = append(properties, systemdDbus.PropWants(slice))
+	} else {
+		// otherwise, we use Slice=
+		properties = append(properties, systemdDbus.PropSlice(slice))
+	}
+
+	// only add pid if its valid, -1 is used w/ general slice creation.
+	if pid != -1 {
+		properties = append(properties, newProp("PIDs", []uint32{uint32(pid)}))
+	}
 
 	if hasDelegate {
 		// This is only supported on systemd versions 218 and above.
@@ -302,10 +344,9 @@ func join(c *configs.Cgroup, subsystem string, pid int) (string, error) {
 	if err := os.MkdirAll(path, 0755); err != nil {
 		return "", err
 	}
-	if err := writeFile(path, "cgroup.procs", strconv.Itoa(pid)); err != nil {
+	if err := cgroups.WriteCgroupProc(path, pid); err != nil {
 		return "", err
 	}
-
 	return path, nil
 }
 
@@ -350,7 +391,7 @@ func joinCgroups(c *configs.Cgroup, pid int) error {
 // systemd represents slice heirarchy using `-`, so we need to follow suit when
 // generating the path of slice. Essentially, test-a-b.slice becomes
 // test.slice/test-a.slice/test-a-b.slice.
-func expandSlice(slice string) (string, error) {
+func ExpandSlice(slice string) (string, error) {
 	suffix := ".slice"
 	// Name has to end with ".slice", but can't be just ".slice".
 	if len(slice) < len(suffix) || !strings.HasSuffix(slice, suffix) {
@@ -364,6 +405,10 @@ func expandSlice(slice string) (string, error) {
 
 	var path, prefix string
 	sliceName := strings.TrimSuffix(slice, suffix)
+	// if input was -.slice, we should just return root now
+	if sliceName == "-" {
+		return "/", nil
+	}
 	for _, component := range strings.Split(sliceName, "-") {
 		// test--a.slice isn't permitted, nor is -test.slice.
 		if component == "" {
@@ -396,7 +441,7 @@ func getSubsystemPath(c *configs.Cgroup, subsystem string) (string, error) {
 		slice = c.Parent
 	}
 
-	slice, err = expandSlice(slice)
+	slice, err = ExpandSlice(slice)
 	if err != nil {
 		return "", err
 	}
@@ -483,7 +528,11 @@ func (m *Manager) Set(container *configs.Config) error {
 }
 
 func getUnitName(c *configs.Cgroup) string {
-	return fmt.Sprintf("%s-%s.scope", c.ScopePrefix, c.Name)
+	// by default, we create a scope unless the user explicitly asks for a slice.
+	if !strings.HasSuffix(c.Name, ".slice") {
+		return fmt.Sprintf("%s-%s.scope", c.ScopePrefix, c.Name)
+	}
+	return c.Name
 }
 
 func setKernelMemory(c *configs.Cgroup) error {

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

@@ -139,7 +139,7 @@ func (m Mount) GetThisCgroupDir(cgroups map[string]string) (string, error) {
 	return getControllerPath(m.Subsystems[0], cgroups)
 }
 
-func getCgroupMountsHelper(ss map[string]bool, mi io.Reader) ([]Mount, error) {
+func getCgroupMountsHelper(ss map[string]bool, mi io.Reader, all bool) ([]Mount, error) {
 	res := make([]Mount, 0, len(ss))
 	scanner := bufio.NewScanner(mi)
 	numFound := 0
@@ -166,7 +166,9 @@ func getCgroupMountsHelper(ss map[string]bool, mi io.Reader) ([]Mount, error) {
 			} else {
 				m.Subsystems = append(m.Subsystems, opt)
 			}
-			numFound++
+			if !all {
+				numFound++
+			}
 		}
 		res = append(res, m)
 	}
@@ -176,23 +178,25 @@ func getCgroupMountsHelper(ss map[string]bool, mi io.Reader) ([]Mount, error) {
 	return res, nil
 }
 
-func GetCgroupMounts() ([]Mount, error) {
+// GetCgroupMounts returns the mounts for the cgroup subsystems.
+// all indicates whether to return just the first instance or all the mounts.
+func GetCgroupMounts(all bool) ([]Mount, error) {
 	f, err := os.Open("/proc/self/mountinfo")
 	if err != nil {
 		return nil, err
 	}
 	defer f.Close()
 
-	all, err := ParseCgroupFile("/proc/self/cgroup")
+	allSubsystems, err := ParseCgroupFile("/proc/self/cgroup")
 	if err != nil {
 		return nil, err
 	}
 
 	allMap := make(map[string]bool)
-	for s := range all {
+	for s := range allSubsystems {
 		allMap[s] = true
 	}
-	return getCgroupMountsHelper(allMap, f)
+	return getCgroupMountsHelper(allMap, f, all)
 }
 
 // GetAllSubsystems returns all the cgroup subsystems supported by the kernel