func (s *MemoryGroup) Apply(d *cgroupData) (err error) {
path, err := d.path("memory")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
if memoryAssigned(d.config) {
if path != "" {
if err := os.MkdirAll(path, 0755); err != nil {
return err
}
}
if err := s.Set(path, d.config); err != nil {
return err
}
}
defer func() {
if err != nil {
os.RemoveAll(path)
}
}()
// We need to join memory cgroup after set memory limits, because
// kmem.limit_in_bytes can only be set when the cgroup is empty.
_, err = d.join("memory")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
return nil
}
func (s *PerfEventGroup) Apply(d *cgroupData) error {
// we just want to join this group even though we don't set anything
if _, err := d.join("perf_event"); err != nil && !cgroups.IsNotFound(err) {
return err
}
return nil
}
func (s *CpusetGroup) Apply(d *cgroupData) error {
dir, err := d.path("cpuset")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
return s.ApplyDir(dir, d.config, d.pid)
}
func joinCpu(c *configs.Cgroup, pid int) error {
path, err := getSubsystemPath(c, "cpu")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
if c.CpuQuota != 0 {
if err = writeFile(path, "cpu.cfs_quota_us", strconv.FormatInt(c.CpuQuota, 10)); err != nil {
return err
}
}
if c.CpuPeriod != 0 {
if err = writeFile(path, "cpu.cfs_period_us", strconv.FormatInt(c.CpuPeriod, 10)); err != nil {
return err
}
}
if c.CpuRtPeriod != 0 {
if err = writeFile(path, "cpu.rt_period_us", strconv.FormatInt(c.CpuRtPeriod, 10)); err != nil {
return err
}
}
if c.CpuRtRuntime != 0 {
if err = writeFile(path, "cpu.rt_runtime_us", strconv.FormatInt(c.CpuRtRuntime, 10)); err != nil {
return err
}
}
return nil
}
// systemd does not atm set up the cpuset controller, so we must manually
// join it. Additionally that is a very finicky controller where each
// level must have a full setup as the default for a new directory is "no cpus"
func joinCpuset(c *configs.Cgroup, pid int) error {
path, err := getSubsystemPath(c, "cpuset")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
s := &fs.CpusetGroup{}
return s.ApplyDir(path, c, pid)
}
func joinPerfEvent(c *configs.Cgroup, pid int) error {
path, err := join(c, "perf_event", pid)
if err != nil && !cgroups.IsNotFound(err) {
return err
}
perfEvent, err := subsystems.Get("perf_event")
if err != nil {
return err
}
return perfEvent.Set(path, c)
}
func joinHugetlb(c *configs.Cgroup, pid int) error {
path, err := join(c, "hugetlb", pid)
if err != nil && !cgroups.IsNotFound(err) {
return err
}
hugetlb, err := subsystems.Get("hugetlb")
if err != nil {
return err
}
return hugetlb.Set(path, c)
}
func joinNetCls(c *configs.Cgroup, pid int) error {
path, err := join(c, "net_cls", pid)
if err != nil && !cgroups.IsNotFound(err) {
return err
}
netcls, err := subsystems.Get("net_cls")
if err != nil {
return err
}
return netcls.Set(path, c)
}
func joinFreezer(c *configs.Cgroup, pid int) error {
path, err := join(c, "freezer", pid)
if err != nil && !cgroups.IsNotFound(err) {
return err
}
freezer, err := subsystems.Get("freezer")
if err != nil {
return err
}
return freezer.Set(path, c)
}
func (s *BlkioGroup) Apply(d *cgroupData) error {
dir, err := d.join("blkio")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
if err := s.Set(dir, d.config); err != nil {
return err
}
return nil
}
func (s *CpuGroup) Apply(d *cgroupData) error {
// We always want to join the cpu group, to allow fair cpu scheduling
// on a container basis
dir, err := d.join("cpu")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
if err := s.Set(dir, d.config); err != nil {
return err
}
return nil
}
func (m *Manager) Apply(pid int) (err error) {
if m.Cgroups == nil {
return nil
}
var c = m.Cgroups
d, err := getCgroupData(m.Cgroups, pid)
if err != nil {
return err
}
paths := make(map[string]string)
defer func() {
if err != nil {
cgroups.RemovePaths(paths)
}
}()
for _, sys := range subsystems {
if err := sys.Apply(d); err != nil {
return err
}
// TODO: Apply should, ideally, be reentrant or be broken up into a separate
// create and join phase so that the cgroup hierarchy for a container can be
// created then join consists of writing the process pids to cgroup.procs
p, err := d.path(sys.Name())
if err != nil {
if cgroups.IsNotFound(err) {
continue
}
return err
}
paths[sys.Name()] = p
}
m.Paths = paths
if paths["cpu"] != "" {
if err := CheckCpushares(paths["cpu"], c.CpuShares); err != nil {
return err
}
}
return nil
}
func setKernelMemory(c *configs.Cgroup) error {
path, err := getSubsystemPath(c, "memory")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
if err := os.MkdirAll(path, 0755); err != nil {
return err
}
if c.KernelMemory > 0 {
err = writeFile(path, "memory.kmem.limit_in_bytes", strconv.FormatInt(c.KernelMemory, 10))
if err != nil {
return err
}
}
return nil
}
func joinMemory(c *configs.Cgroup, pid int) error {
path, err := getSubsystemPath(c, "memory")
if err != nil && !cgroups.IsNotFound(err) {
return err
}
// -1 disables memoryswap
if c.MemorySwap > 0 {
err = writeFile(path, "memory.memsw.limit_in_bytes", strconv.FormatInt(c.MemorySwap, 10))
if err != nil {
return err
}
}
if c.MemoryReservation > 0 {
err = writeFile(path, "memory.soft_limit_in_bytes", strconv.FormatInt(c.MemoryReservation, 10))
if err != nil {
return err
}
}
if c.OomKillDisable {
if err := writeFile(path, "memory.oom_control", "1"); err != nil {
return err
}
}
if c.MemorySwappiness >= 0 && c.MemorySwappiness <= 100 {
err = writeFile(path, "memory.swappiness", strconv.FormatInt(c.MemorySwappiness, 10))
if err != nil {
return err
}
} else if c.MemorySwappiness == -1 {
return nil
} else {
return fmt.Errorf("invalid value:%d. valid memory swappiness range is 0-100", c.MemorySwappiness)
}
return nil
}
func (m *Manager) Apply(pid int) error {
var (
c = m.Cgroups
unitName = getUnitName(c)
slice = "system.slice"
properties []systemdDbus.Property
)
if c.Slice != "" {
slice = c.Slice
}
properties = append(properties,
systemdDbus.PropSlice(slice),
systemdDbus.PropDescription("docker container "+c.Name),
newProp("PIDs", []uint32{uint32(pid)}),
)
// Always enable accounting, this gets us the same behaviour as the fs implementation,
// plus the kernel has some problems with joining the memory cgroup at a later time.
properties = append(properties,
newProp("MemoryAccounting", true),
newProp("CPUAccounting", true),
newProp("BlockIOAccounting", true))
if hasTransientDefaultDependencies {
properties = append(properties,
newProp("DefaultDependencies", false))
}
if c.Memory != 0 {
properties = append(properties,
newProp("MemoryLimit", uint64(c.Memory)))
}
if c.CpuShares != 0 {
properties = append(properties,
newProp("CPUShares", uint64(c.CpuShares)))
}
if c.BlkioWeight != 0 {
properties = append(properties,
newProp("BlockIOWeight", uint64(c.BlkioWeight)))
}
// We need to set kernel memory before processes join cgroup because
// kmem.limit_in_bytes can only be set when the cgroup is empty.
// And swap memory limit needs to be set after memory limit, only
// memory limit is handled by systemd, so it's kind of ugly here.
if c.KernelMemory > 0 {
if err := setKernelMemory(c); err != nil {
return err
}
}
if _, err := theConn.StartTransientUnit(unitName, "replace", properties, nil); err != nil {
return err
}
if err := joinDevices(c, pid); err != nil {
return err
}
// TODO: CpuQuota and CpuPeriod not available in systemd
// we need to manually join the cpu.cfs_quota_us and cpu.cfs_period_us
if err := joinCpu(c, pid); err != nil {
return err
}
// TODO: MemoryReservation and MemorySwap not available in systemd
if err := joinMemory(c, pid); err != nil {
return err
}
// we need to manually join the freezer, net_cls, net_prio and cpuset cgroup in systemd
// because it does not currently support it via the dbus api.
if err := joinFreezer(c, pid); err != nil {
return err
}
if err := joinNetPrio(c, pid); err != nil {
return err
}
if err := joinNetCls(c, pid); err != nil {
return err
}
if err := joinCpuset(c, pid); err != nil {
return err
}
if err := joinHugetlb(c, pid); err != nil {
return err
}
if err := joinPerfEvent(c, pid); err != nil {
return err
}
// FIXME: Systemd does have `BlockIODeviceWeight` property, but we got problem
// using that (at least on systemd 208, see https://github.com/opencontainers/runc/libcontainer/pull/354),
// so use fs work around for now.
if err := joinBlkio(c, pid); err != nil {
return err
}
paths := make(map[string]string)
for _, s := range subsystems {
subsystemPath, err := getSubsystemPath(m.Cgroups, s.Name())
if err != nil {
// Don't fail if a cgroup hierarchy was not found, just skip this subsystem
if cgroups.IsNotFound(err) {
continue
}
return err
}
paths[s.Name()] = subsystemPath
}
m.Paths = paths
if paths["cpu"] != "" {
if err := fs.CheckCpushares(paths["cpu"], c.CpuShares); err != nil {
return err
}
}
return nil
}