Support process.scheduler

Spec: opencontainers/runtime-spec#1188 Fix: opencontainers/runc#3895 Co-authored-by: lifubang <[email protected]> Signed-off-by: utam0k <[email protected]> Signed-off-by: lifubang <[email protected]>
dims · Oct 4, 2023 · b494c40 · b494c40
1 parent 9ab8b91
commit b494c40
Show file tree

Hide file tree

Showing 3 changed files with 136 additions and 0 deletions.
diff --git a/libcontainer/configs/config.go b/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 (

diff --git a/libcontainer/configs/validate/validator.go b/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
+}
diff --git a/libcontainer/configs/validate/validator_test.go b/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)
+		}
+	}
+}