Merge pull request #4459 from kolyshkin/prio-nits

Fixups to scheduler/priority settings

libcontainer/configs/config.go

@@ -286,12 +286,6 @@ func ToSchedAttr(scheduler *Scheduler) (*unix.SchedAttr, error) {
 	}, nil
 }
 
-var IOPrioClassMapping = map[specs.IOPriorityClass]int{
-	specs.IOPRIO_CLASS_RT:   1,
-	specs.IOPRIO_CLASS_BE:   2,
-	specs.IOPRIO_CLASS_IDLE: 3,
-}
-
 type IOPriority = specs.LinuxIOPriority
 
 type (

libcontainer/configs/validate/validator.go

@@ -406,5 +406,13 @@ func ioPriority(config *configs.Config) error {
 	if priority < 0 || priority > 7 {
 		return fmt.Errorf("invalid ioPriority.Priority: %d", priority)
 	}
+
+	switch class := config.IOPriority.Class; class {
+	case specs.IOPRIO_CLASS_RT, specs.IOPRIO_CLASS_BE, specs.IOPRIO_CLASS_IDLE:
+		// Valid class, do nothing.
+	default:
+		return fmt.Errorf("invalid ioPriority.Class: %q", class)
+	}
+
 	return nil
 }

libcontainer/configs/validate/validator_test.go

@@ -847,15 +847,21 @@ func TestValidateIOPriority(t *testing.T) {
 	testCases := []struct {
 		isErr    bool
 		priority int
+		class    specs.IOPriorityClass
 	}{
-		{isErr: false, priority: 0},
-		{isErr: false, priority: 7},
-		{isErr: true, priority: -1},
+		{isErr: false, priority: 0, class: specs.IOPRIO_CLASS_IDLE},
+		{isErr: false, priority: 7, class: specs.IOPRIO_CLASS_RT},
+		{isErr: false, priority: 3, class: specs.IOPRIO_CLASS_BE},
+		// Invalid priority.
+		{isErr: true, priority: -1, class: specs.IOPRIO_CLASS_BE},
+		// Invalid class.
+		{isErr: true, priority: 3, class: specs.IOPriorityClass("IOPRIO_CLASS_WOW")},
 	}
 
 	for _, tc := range testCases {
 		ioPriroty := configs.IOPriority{
 			Priority: tc.priority,
+			Class:    tc.class,
 		}
 		config := &configs.Config{
 			Rootfs:     "/var",

libcontainer/init_linux.go

@@ -662,6 +662,9 @@ func setupRlimits(limits []configs.Rlimit, pid int) error {
 }
 
 func setupScheduler(config *configs.Config) error {
+	if config.Scheduler == nil {
+		return nil
+	}
 	attr, err := configs.ToSchedAttr(config.Scheduler)
 	if err != nil {
 		return err
@@ -675,6 +678,35 @@ func setupScheduler(config *configs.Config) error {
 	return nil
 }
 
+func setupIOPriority(config *configs.Config) error {
+	const ioprioWhoPgrp = 1
+
+	ioprio := config.IOPriority
+	if ioprio == nil {
+		return nil
+	}
+	class := 0
+	switch ioprio.Class {
+	case specs.IOPRIO_CLASS_RT:
+		class = 1
+	case specs.IOPRIO_CLASS_BE:
+		class = 2
+	case specs.IOPRIO_CLASS_IDLE:
+		class = 3
+	default:
+		return fmt.Errorf("invalid io priority class: %s", ioprio.Class)
+	}
+
+	// Combine class and priority into a single value
+	// https://github.com/torvalds/linux/blob/v5.18/include/uapi/linux/ioprio.h#L5-L17
+	iop := (class << 13) | ioprio.Priority
+	_, _, errno := unix.RawSyscall(unix.SYS_IOPRIO_SET, ioprioWhoPgrp, 0, uintptr(iop))
+	if errno != 0 {
+		return fmt.Errorf("failed to set io priority: %w", errno)
+	}
+	return nil
+}
+
 func setupPersonality(config *configs.Config) error {
 	return system.SetLinuxPersonality(config.Personality.Domain)
 }

libcontainer/process_linux.go

@@ -166,12 +166,6 @@ type setnsProcess struct {
 func (p *setnsProcess) start() (retErr error) {
 	defer p.comm.closeParent()
 
-	if p.process.IOPriority != nil {
-		if err := setIOPriority(p.process.IOPriority); err != nil {
-			return err
-		}
-	}
-
 	// get the "before" value of oom kill count
 	oom, _ := p.manager.OOMKillCount()
 	err := p.cmd.Start()
@@ -905,21 +899,3 @@ func (p *Process) InitializeIO(rootuid, rootgid int) (i *IO, err error) {
 	}
 	return i, nil
 }
-
-func setIOPriority(ioprio *configs.IOPriority) error {
-	const ioprioWhoPgrp = 1
-
-	class, ok := configs.IOPrioClassMapping[ioprio.Class]
-	if !ok {
-		return fmt.Errorf("invalid io priority class: %s", ioprio.Class)
-	}
-
-	// Combine class and priority into a single value
-	// https://github.com/torvalds/linux/blob/v5.18/include/uapi/linux/ioprio.h#L5-L17
-	iop := (class << 13) | ioprio.Priority
-	_, _, errno := unix.RawSyscall(unix.SYS_IOPRIO_SET, ioprioWhoPgrp, 0, uintptr(iop))
-	if errno != 0 {
-		return fmt.Errorf("failed to set io priority: %w", errno)
-	}
-	return nil
-}

libcontainer/setns_init_linux.go

@@ -71,12 +71,13 @@ 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 := setupScheduler(l.config.Config); err != nil {
+		return err
 	}
 
+	if err := setupIOPriority(l.config.Config); err != nil {
+		return err
+	}
 	// Tell our parent that we're ready to exec. This must be done before the
 	// Seccomp rules have been applied, because we need to be able to read and
 	// write to a socket.

libcontainer/standard_init_linux.go

@@ -155,15 +155,12 @@ func (l *linuxStandardInit) Init() error {
 		}
 	}
 
-	if l.config.Config.Scheduler != nil {
-		if err := setupScheduler(l.config.Config); err != nil {
-			return err
-		}
+	if err := setupScheduler(l.config.Config); err != nil {
+		return err
 	}
-	if l.config.Config.IOPriority != nil {
-		if err := setIOPriority(l.config.Config.IOPriority); err != nil {
-			return err
-		}
+
+	if err := setupIOPriority(l.config.Config); err != nil {
+		return err
 	}
 
 	// Tell our parent that we're ready to exec. This must be done before the

utils_linux.go

@@ -55,6 +55,8 @@ func newProcess(p specs.Process) (*libcontainer.Process, error) {
 		Label:           p.SelinuxLabel,
 		NoNewPrivileges: &p.NoNewPrivileges,
 		AppArmorProfile: p.ApparmorProfile,
+		Scheduler:       p.Scheduler,
+		IOPriority:      p.IOPriority,
 	}
 
 	if p.ConsoleSize != nil {
@@ -62,16 +64,6 @@ 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.IOPriority != nil {
-		ioPriority := *p.IOPriority
-		lp.IOPriority = &ioPriority
-	}
-
 	if p.Capabilities != nil {
 		lp.Capabilities = &configs.Capabilities{}
 		lp.Capabilities.Bounding = p.Capabilities.Bounding