Support process.scheduler

Spec: https://github.com/opencontainers/runtime-spec/pull/1188
Fix: https://github.com/opencontainers/runc/issues/3895

Co-authored-by: lifubang <lifubang@acmcoder.com>
Signed-off-by: utam0k <k0ma@utam0k.jp>
Signed-off-by: lifubang <lifubang@acmcoder.com>

docs/spec-conformance.md

@@ -12,7 +12,6 @@ v1.0.0       | `SCMP_ARCH_PARISC64`                     | Unplanned, due to lack
 v1.0.2       | `.linux.personality`                     | [#3126](https://github.com/opencontainers/runc/pull/3126)
 v1.1.0       | `SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV` | [#3862](https://github.com/opencontainers/runc/pull/3862)
 v1.1.0       | rsvd hugetlb cgroup                      | TODO ([#3859](https://github.com/opencontainers/runc/issues/3859))
-v1.1.0       | `.process.scheduler`                     | TODO ([#3895](https://github.com/opencontainers/runc/issues/3895))
 v1.1.0       | `.process.ioPriority`                    | [#3783](https://github.com/opencontainers/runc/pull/3783)
 
 

libcontainer/configs/config.go

@@ -8,6 +8,7 @@ import (
 	"time"
 
 	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
 
 	"github.com/opencontainers/runc/libcontainer/devices"
 	"github.com/opencontainers/runtime-spec/specs-go"
@@ -219,6 +220,68 @@ type Config struct {
 
 	// TimeOffsets specifies the offset for supporting time namespaces.
 	TimeOffsets map[string]specs.LinuxTimeOffset `json:"time_offsets,omitempty"`
+
+	// Scheduler represents the scheduling attributes for a process.
+	Scheduler *Scheduler `json:"scheduler,omitempty"`
+}
+
+// Scheduler is based on the Linux sched_setattr(2) syscall.
+type Scheduler = specs.Scheduler
+
+// ToSchedAttr is to convert *configs.Scheduler to *unix.SchedAttr.
+func ToSchedAttr(scheduler *Scheduler) (*unix.SchedAttr, error) {
+	var policy uint32
+	switch scheduler.Policy {
+	case specs.SchedOther:
+		policy = 0
+	case specs.SchedFIFO:
+		policy = 1
+	case specs.SchedRR:
+		policy = 2
+	case specs.SchedBatch:
+		policy = 3
+	case specs.SchedISO:
+		policy = 4
+	case specs.SchedIdle:
+		policy = 5
+	case specs.SchedDeadline:
+		policy = 6
+	default:
+		return nil, fmt.Errorf("invalid scheduler policy: %s", scheduler.Policy)
+	}
+
+	var flags uint64
+	for _, flag := range scheduler.Flags {
+		switch flag {
+		case specs.SchedFlagResetOnFork:
+			flags |= 0x01
+		case specs.SchedFlagReclaim:
+			flags |= 0x02
+		case specs.SchedFlagDLOverrun:
+			flags |= 0x04
+		case specs.SchedFlagKeepPolicy:
+			flags |= 0x08
+		case specs.SchedFlagKeepParams:
+			flags |= 0x10
+		case specs.SchedFlagUtilClampMin:
+			flags |= 0x20
+		case specs.SchedFlagUtilClampMax:
+			flags |= 0x40
+		default:
+			return nil, fmt.Errorf("invalid scheduler flag: %s", flag)
+		}
+	}
+
+	return &unix.SchedAttr{
+		Size:     unix.SizeofSchedAttr,
+		Policy:   policy,
+		Flags:    flags,
+		Nice:     scheduler.Nice,
+		Priority: uint32(scheduler.Priority),
+		Runtime:  scheduler.Runtime,
+		Deadline: scheduler.Deadline,
+		Period:   scheduler.Period,
+	}, nil
 }
 
 type (

libcontainer/configs/validate/validator.go

@@ -11,6 +11,7 @@ import (
 	"github.com/opencontainers/runc/libcontainer/cgroups"
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"github.com/opencontainers/runc/libcontainer/intelrdt"
+	"github.com/opencontainers/runtime-spec/specs-go"
 	selinux "github.com/opencontainers/selinux/go-selinux"
 	"github.com/sirupsen/logrus"
 	"golang.org/x/sys/unix"
@@ -30,6 +31,7 @@ func Validate(config *configs.Config) error {
 		intelrdtCheck,
 		rootlessEUIDCheck,
 		mountsStrict,
+		scheduler,
 	}
 	for _, c := range checks {
 		if err := c(config); err != nil {
@@ -353,3 +355,24 @@ func isHostNetNS(path string) (bool, error) {
 
 	return (st1.Dev == st2.Dev) && (st1.Ino == st2.Ino), nil
 }
+
+// scheduler is to validate scheduler configs according to https://man7.org/linux/man-pages/man2/sched_setattr.2.html
+func scheduler(config *configs.Config) error {
+	s := config.Scheduler
+	if s == nil {
+		return nil
+	}
+	if s.Policy == "" {
+		return errors.New("scheduler policy is required")
+	}
+	if s.Nice < -20 || s.Nice > 19 {
+		return fmt.Errorf("invalid scheduler.nice: %d", s.Nice)
+	}
+	if s.Priority != 0 && (s.Policy != specs.SchedFIFO && s.Policy != specs.SchedRR) {
+		return errors.New("scheduler.priority can only be specified for SchedFIFO or SchedRR policy")
+	}
+	if s.Policy != specs.SchedDeadline && (s.Runtime != 0 || s.Deadline != 0 || s.Period != 0) {
+		return errors.New("scheduler runtime/deadline/period can only be specified for SchedDeadline policy")
+	}
+	return nil
+}

libcontainer/configs/validate/validator_test.go

@@ -616,3 +616,53 @@ func TestValidateIDMapMounts(t *testing.T) {
 		})
 	}
 }
+
+func TestValidateScheduler(t *testing.T) {
+	testCases := []struct {
+		isErr     bool
+		policy    string
+		niceValue int32
+		priority  int32
+		runtime   uint64
+		deadline  uint64
+		period    uint64
+	}{
+		{isErr: true, niceValue: 0},
+		{isErr: false, policy: "SCHED_OTHER", niceValue: 19},
+		{isErr: false, policy: "SCHED_OTHER", niceValue: -20},
+		{isErr: true, policy: "SCHED_OTHER", niceValue: 20},
+		{isErr: true, policy: "SCHED_OTHER", niceValue: -21},
+		{isErr: true, policy: "SCHED_OTHER", priority: 100},
+		{isErr: false, policy: "SCHED_FIFO", priority: 100},
+		{isErr: true, policy: "SCHED_FIFO", runtime: 20},
+		{isErr: true, policy: "SCHED_BATCH", deadline: 30},
+		{isErr: true, policy: "SCHED_IDLE", period: 40},
+		{isErr: true, policy: "SCHED_DEADLINE", priority: 100},
+		{isErr: false, policy: "SCHED_DEADLINE", runtime: 200},
+		{isErr: false, policy: "SCHED_DEADLINE", deadline: 300},
+		{isErr: false, policy: "SCHED_DEADLINE", period: 400},
+	}
+
+	for _, tc := range testCases {
+		scheduler := configs.Scheduler{
+			Policy:   specs.LinuxSchedulerPolicy(tc.policy),
+			Nice:     tc.niceValue,
+			Priority: tc.priority,
+			Runtime:  tc.runtime,
+			Deadline: tc.deadline,
+			Period:   tc.period,
+		}
+		config := &configs.Config{
+			Rootfs:    "/var",
+			Scheduler: &scheduler,
+		}
+
+		err := Validate(config)
+		if tc.isErr && err == nil {
+			t.Errorf("scheduler: %d, expected error, got nil", tc.niceValue)
+		}
+		if !tc.isErr && err != nil {
+			t.Errorf("scheduler: %d, expected nil, got error %v", tc.niceValue, err)
+		}
+	}
+}

libcontainer/init_linux.go

@@ -640,6 +640,20 @@ func setupRlimits(limits []configs.Rlimit, pid int) error {
 	return nil
 }
 
+func setupScheduler(config *configs.Config) error {
+	attr, err := configs.ToSchedAttr(config.Scheduler)
+	if err != nil {
+		return err
+	}
+	if err := unix.SchedSetAttr(0, attr, 0); err != nil {
+		if errors.Is(err, unix.EPERM) && config.Cgroups.CpusetCpus != "" {
+			return errors.New("process scheduler can't be used together with AllowedCPUs")
+		}
+		return fmt.Errorf("error setting scheduler: %w", err)
+	}
+	return nil
+}
+
 // signalAllProcesses freezes then iterates over all the processes inside the
 // manager's cgroups sending the signal s to them.
 func signalAllProcesses(m cgroups.Manager, s unix.Signal) error {

libcontainer/process.go

@@ -95,6 +95,8 @@ type Process struct {
 	//
 	// For cgroup v2, the only key allowed is "".
 	SubCgroupPaths map[string]string
+
+	Scheduler *configs.Scheduler
 }
 
 // Wait waits for the process to exit.

libcontainer/setns_init_linux.go

@@ -65,6 +65,12 @@ func (l *linuxSetnsInit) Init() error {
 		unix.Umask(int(*l.config.Config.Umask))
 	}
 
+	if l.config.Config.Scheduler != nil {
+		if err := setupScheduler(l.config.Config); err != nil {
+			return err
+		}
+	}
+
 	if err := selinux.SetExecLabel(l.config.ProcessLabel); err != nil {
 		return err
 	}

libcontainer/specconv/spec_linux.go

@@ -494,6 +494,10 @@ func CreateLibcontainerConfig(opts *CreateOpts) (*configs.Config, error) {
 				Ambient:     spec.Process.Capabilities.Ambient,
 			}
 		}
+		if spec.Process.Scheduler != nil {
+			s := *spec.Process.Scheduler
+			config.Scheduler = &s
+		}
 	}
 	createHooks(spec, config)
 	config.Version = specs.Version

libcontainer/standard_init_linux.go

@@ -159,6 +159,13 @@ func (l *linuxStandardInit) Init() error {
 			return &os.SyscallError{Syscall: "prctl(SET_NO_NEW_PRIVS)", Err: err}
 		}
 	}
+
+	if l.config.Config.Scheduler != nil {
+		if err := setupScheduler(l.config.Config); err != nil {
+			return err
+		}
+	}
+
 	// Tell our parent that we're ready to Execv. This must be done before the
 	// Seccomp rules have been applied, because we need to be able to read and
 	// write to a socket.

tests/integration/scheduler.bats

@@ -0,0 +1,34 @@
+#!/usr/bin/env bats
+
+load helpers
+
+function setup() {
+	requires root
+	setup_debian
+}
+
+function teardown() {
+	teardown_bundle
+}
+
+@test "scheduler is applied" {
+	update_config ' .process.scheduler = {"policy": "SCHED_DEADLINE", "nice": 19, "priority": 0, "runtime": 42000, "deadline": 1000000, "period": 1000000, }'
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_scheduler
+	[ "$status" -eq 0 ]
+
+	runc exec test_scheduler chrt -p 1
+	[ "$status" -eq 0 ]
+
+	[[ "${lines[0]}" == *"scheduling policy: SCHED_DEADLINE" ]]
+	[[ "${lines[1]}" == *"priority: 0" ]]
+	[[ "${lines[2]}" == *"runtime/deadline/period parameters: 42000/1000000/1000000" ]]
+}
+
+@test "scheduler vs cpus" {
+	update_config ' .linux.resources.cpu.cpus = "0"
+		| .process.scheduler = {"policy": "SCHED_DEADLINE", "nice": 19, "runtime": 42000, "deadline": 1000000, "period": 1000000, }'
+
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_scheduler
+	[ "$status" -eq 1 ]
+}

utils_linux.go

@@ -61,6 +61,11 @@ func newProcess(p specs.Process) (*libcontainer.Process, error) {
 		lp.ConsoleHeight = uint16(p.ConsoleSize.Height)
 	}
 
+	if p.Scheduler != nil {
+		s := *p.Scheduler
+		lp.Scheduler = &s
+	}
+
 	if p.Capabilities != nil {
 		lp.Capabilities = &configs.Capabilities{}
 		lp.Capabilities.Bounding = p.Capabilities.Bounding