Move tcpip.Clock impl to Timekeeper

...and pass it explicitly.

This reverts commit b63e61828d.

PiperOrigin-RevId: 380039167

pkg/sentry/fsimpl/testutil/kernel.go

@@ -80,12 +80,8 @@ func Boot() (*kernel.Kernel, error) {
 	}
 
 	// Create timekeeper.
-	tk, err := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange())
-	if err != nil {
-		return nil, fmt.Errorf("creating timekeeper: %v", err)
-	}
+	tk := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange())
 	tk.SetClocks(time.NewCalibratedClocks())
-	k.SetTimekeeper(tk)
 
 	creds := auth.NewRootCredentials(auth.NewRootUserNamespace())
 
@@ -94,6 +90,7 @@ func Boot() (*kernel.Kernel, error) {
 	if err = k.Init(kernel.InitKernelArgs{
 		ApplicationCores:            uint(runtime.GOMAXPROCS(-1)),
 		FeatureSet:                  cpuid.HostFeatureSet(),
+		Timekeeper:                  tk,
 		RootUserNamespace:           creds.UserNamespace,
 		Vdso:                        vdso,
 		RootUTSNamespace:            kernel.NewUTSNamespace("hostname", "domain", creds.UserNamespace),

pkg/sentry/kernel/kernel.go

@@ -143,12 +143,6 @@ type Kernel struct {
 	// to CreateProcess, and is protected by extMu.
 	globalInit *ThreadGroup
 
-	// realtimeClock is a ktime.Clock based on timekeeper's Realtime.
-	realtimeClock *timekeeperClock
-
-	// monotonicClock is a ktime.Clock based on timekeeper's Monotonic.
-	monotonicClock *timekeeperClock
-
 	// syslog is the kernel log.
 	syslog syslog
 
@@ -306,6 +300,9 @@ type InitKernelArgs struct {
 	// FeatureSet is the emulated CPU feature set.
 	FeatureSet *cpuid.FeatureSet
 
+	// Timekeeper manages time for all tasks in the system.
+	Timekeeper *Timekeeper
+
 	// RootUserNamespace is the root user namespace.
 	RootUserNamespace *auth.UserNamespace
 
@@ -345,24 +342,18 @@ type InitKernelArgs struct {
 	PIDNamespace *PIDNamespace
 }
 
-// SetTimekeeper sets Kernel.timekeeper. SetTimekeeper must be called before
-// Init.
-func (k *Kernel) SetTimekeeper(tk *Timekeeper) {
-	k.timekeeper = tk
-}
-
 // Init initialize the Kernel with no tasks.
 //
 // Callers must manually set Kernel.Platform and call Kernel.SetMemoryFile
-// and Kernel.SetTimekeeper before calling Init.
+// before calling Init.
 func (k *Kernel) Init(args InitKernelArgs) error {
 	if args.FeatureSet == nil {
 		return fmt.Errorf("args.FeatureSet is nil")
 	}
-	if k.timekeeper == nil {
-		return fmt.Errorf("timekeeper is nil")
+	if args.Timekeeper == nil {
+		return fmt.Errorf("args.Timekeeper is nil")
 	}
-	if k.timekeeper.clocks == nil {
+	if args.Timekeeper.clocks == nil {
 		return fmt.Errorf("must call Timekeeper.SetClocks() before Kernel.Init()")
 	}
 	if args.RootUserNamespace == nil {
@@ -373,6 +364,7 @@ func (k *Kernel) Init(args InitKernelArgs) error {
 	}
 
 	k.featureSet = args.FeatureSet
+	k.timekeeper = args.Timekeeper
 	k.tasks = newTaskSet(args.PIDNamespace)
 	k.rootUserNamespace = args.RootUserNamespace
 	k.rootUTSNamespace = args.RootUTSNamespace
@@ -397,8 +389,6 @@ func (k *Kernel) Init(args InitKernelArgs) error {
 	}
 	k.extraAuxv = args.ExtraAuxv
 	k.vdso = args.Vdso
-	k.realtimeClock = &timekeeperClock{tk: k.timekeeper, c: sentrytime.Realtime}
-	k.monotonicClock = &timekeeperClock{tk: k.timekeeper, c: sentrytime.Monotonic}
 	k.futexes = futex.NewManager()
 	k.netlinkPorts = port.New()
 	k.ptraceExceptions = make(map[*Task]*Task)
@@ -531,6 +521,8 @@ func (k *Kernel) SaveTo(ctx context.Context, w wire.Writer) error {
 	}
 	log.Infof("CPUID save took [%s].", time.Since(cpuidStart))
 
+	// Save the timekeeper's state.
+
 	// Save the kernel state.
 	kernelStart := time.Now()
 	stats, err := state.Save(ctx, w, k)
@@ -675,7 +667,7 @@ func (k *Kernel) invalidateUnsavableMappings(ctx context.Context) error {
 }
 
 // LoadFrom returns a new Kernel loaded from args.
-func (k *Kernel) LoadFrom(ctx context.Context, r wire.Reader, net inet.Stack, clocks sentrytime.Clocks, vfsOpts *vfs.CompleteRestoreOptions) error {
+func (k *Kernel) LoadFrom(ctx context.Context, r wire.Reader, timeReady chan struct{}, net inet.Stack, clocks sentrytime.Clocks, vfsOpts *vfs.CompleteRestoreOptions) error {
 	loadStart := time.Now()
 
 	initAppCores := k.applicationCores
@@ -722,6 +714,11 @@ func (k *Kernel) LoadFrom(ctx context.Context, r wire.Reader, net inet.Stack, cl
 	log.Infof("Overall load took [%s]", time.Since(loadStart))
 
 	k.Timekeeper().SetClocks(clocks)
+
+	if timeReady != nil {
+		close(timeReady)
+	}
+
 	if net != nil {
 		net.Resume()
 	}
@@ -1103,7 +1100,7 @@ func (k *Kernel) Start() error {
 	}
 
 	k.started = true
-	k.cpuClockTicker = ktime.NewTimer(k.monotonicClock, newKernelCPUClockTicker(k))
+	k.cpuClockTicker = ktime.NewTimer(k.timekeeper.monotonicClock, newKernelCPUClockTicker(k))
 	k.cpuClockTicker.Swap(ktime.Setting{
 		Enabled: true,
 		Period:  linux.ClockTick,
@@ -1258,7 +1255,7 @@ func (k *Kernel) incRunningTasks() {
 		// These cause very different value of cpuClock. But again, since
 		// nothing was running while the ticker was disabled, those differences
 		// don't matter.
-		setting, exp := k.cpuClockTickerSetting.At(k.monotonicClock.Now())
+		setting, exp := k.cpuClockTickerSetting.At(k.timekeeper.monotonicClock.Now())
 		if exp > 0 {
 			atomic.AddUint64(&k.cpuClock, exp)
 		}
@@ -1468,12 +1465,12 @@ func (k *Kernel) ApplicationCores() uint {
 
 // RealtimeClock returns the application CLOCK_REALTIME clock.
 func (k *Kernel) RealtimeClock() ktime.Clock {
-	return k.realtimeClock
+	return k.timekeeper.realtimeClock
 }
 
 // MonotonicClock returns the application CLOCK_MONOTONIC clock.
 func (k *Kernel) MonotonicClock() ktime.Clock {
-	return k.monotonicClock
+	return k.timekeeper.monotonicClock
 }
 
 // CPUClockNow returns the current value of k.cpuClock.
@@ -1553,32 +1550,6 @@ func (k *Kernel) SetSaveError(err error) {
 	}
 }
 
-var _ tcpip.Clock = (*Kernel)(nil)
-
-// Now implements tcpip.Clock.NowNanoseconds.
-func (k *Kernel) Now() time.Time {
-	nsec, err := k.timekeeper.GetTime(sentrytime.Realtime)
-	if err != nil {
-		panic("timekeeper.GetTime(sentrytime.Realtime): " + err.Error())
-	}
-	return time.Unix(0, nsec)
-}
-
-// NowMonotonic implements tcpip.Clock.NowMonotonic.
-func (k *Kernel) NowMonotonic() tcpip.MonotonicTime {
-	nsec, err := k.timekeeper.GetTime(sentrytime.Monotonic)
-	if err != nil {
-		panic("timekeeper.GetTime(sentrytime.Monotonic): " + err.Error())
-	}
-	var mt tcpip.MonotonicTime
-	return mt.Add(time.Duration(nsec) * time.Nanosecond)
-}
-
-// AfterFunc implements tcpip.Clock.AfterFunc.
-func (k *Kernel) AfterFunc(d time.Duration, f func()) tcpip.Timer {
-	return ktime.TcpipAfterFunc(k.realtimeClock, d, f)
-}
-
 // SetMemoryFile sets Kernel.mf. SetMemoryFile must be called before Init or
 // LoadFrom.
 func (k *Kernel) SetMemoryFile(mf *pgalloc.MemoryFile) {

pkg/sentry/kernel/thread_group.go

@@ -278,7 +278,7 @@ func (k *Kernel) NewThreadGroup(mntns *fs.MountNamespace, pidns *PIDNamespace, s
 		limits:            limits,
 		mounts:            mntns,
 	}
-	tg.itimerRealTimer = ktime.NewTimer(k.monotonicClock, &itimerRealListener{tg: tg})
+	tg.itimerRealTimer = ktime.NewTimer(k.timekeeper.monotonicClock, &itimerRealListener{tg: tg})
 	tg.timers = make(map[linux.TimerID]*IntervalTimer)
 	tg.oldRSeqCritical.Store(&OldRSeqCriticalRegion{})
 	return tg

pkg/sentry/kernel/timekeeper.go

@@ -25,6 +25,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/pgalloc"
 	sentrytime "gvisor.dev/gvisor/pkg/sentry/time"
 	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/tcpip"
 )
 
 // Timekeeper manages all of the kernel clocks.
@@ -39,6 +40,12 @@ type Timekeeper struct {
 	// It is set only once, by SetClocks.
 	clocks sentrytime.Clocks `state:"nosave"`
 
+	// realtimeClock is a ktime.Clock based on timekeeper's Realtime.
+	realtimeClock *timekeeperClock
+
+	// monotonicClock is a ktime.Clock based on timekeeper's Monotonic.
+	monotonicClock *timekeeperClock
+
 	// bootTime is the realtime when the system "booted". i.e., when
 	// SetClocks was called in the initial (not restored) run.
 	bootTime ktime.Time
@@ -90,10 +97,13 @@ type Timekeeper struct {
 // NewTimekeeper does not take ownership of paramPage.
 //
 // SetClocks must be called on the returned Timekeeper before it is usable.
-func NewTimekeeper(mfp pgalloc.MemoryFileProvider, paramPage memmap.FileRange) (*Timekeeper, error) {
-	return &Timekeeper{
+func NewTimekeeper(mfp pgalloc.MemoryFileProvider, paramPage memmap.FileRange) *Timekeeper {
+	t := Timekeeper{
 		params: NewVDSOParamPage(mfp, paramPage),
-	}, nil
+	}
+	t.realtimeClock = &timekeeperClock{tk: &t, c: sentrytime.Realtime}
+	t.monotonicClock = &timekeeperClock{tk: &t, c: sentrytime.Monotonic}
+	return &t
 }
 
 // SetClocks the backing clock source.
@@ -167,6 +177,32 @@ func (t *Timekeeper) SetClocks(c sentrytime.Clocks) {
 	}
 }
 
+var _ tcpip.Clock = (*Timekeeper)(nil)
+
+// Now implements tcpip.Clock.
+func (t *Timekeeper) Now() time.Time {
+	nsec, err := t.GetTime(sentrytime.Realtime)
+	if err != nil {
+		panic("timekeeper.GetTime(sentrytime.Realtime): " + err.Error())
+	}
+	return time.Unix(0, nsec)
+}
+
+// NowMonotonic implements tcpip.Clock.
+func (t *Timekeeper) NowMonotonic() tcpip.MonotonicTime {
+	nsec, err := t.GetTime(sentrytime.Monotonic)
+	if err != nil {
+		panic("timekeeper.GetTime(sentrytime.Monotonic): " + err.Error())
+	}
+	var mt tcpip.MonotonicTime
+	return mt.Add(time.Duration(nsec) * time.Nanosecond)
+}
+
+// AfterFunc implements tcpip.Clock.
+func (t *Timekeeper) AfterFunc(d time.Duration, f func()) tcpip.Timer {
+	return ktime.TcpipAfterFunc(t.realtimeClock, d, f)
+}
+
 // startUpdater starts an update goroutine that keeps the clocks updated.
 //
 // mu must be held.

pkg/sentry/state/state.go

@@ -110,7 +110,7 @@ type LoadOpts struct {
 }
 
 // Load loads the given kernel, setting the provided platform and stack.
-func (opts LoadOpts) Load(ctx context.Context, k *kernel.Kernel, n inet.Stack, clocks time.Clocks, vfsOpts *vfs.CompleteRestoreOptions) error {
+func (opts LoadOpts) Load(ctx context.Context, k *kernel.Kernel, timeReady chan struct{}, n inet.Stack, clocks time.Clocks, vfsOpts *vfs.CompleteRestoreOptions) error {
 	// Open the file.
 	r, m, err := statefile.NewReader(opts.Source, opts.Key)
 	if err != nil {
@@ -120,5 +120,5 @@ func (opts LoadOpts) Load(ctx context.Context, k *kernel.Kernel, n inet.Stack, c
 	previousMetadata = m
 
 	// Restore the Kernel object graph.
-	return k.LoadFrom(ctx, r, n, clocks, vfsOpts)
+	return k.LoadFrom(ctx, r, timeReady, n, clocks, vfsOpts)
 }

runsc/boot/controller.go

@@ -439,7 +439,7 @@ func (cm *containerManager) Restore(o *RestoreOpts, _ *struct{}) error {
 
 	// Load the state.
 	loadOpts := state.LoadOpts{Source: specFile}
-	if err := loadOpts.Load(ctx, k, networkStack, time.NewCalibratedClocks(), &vfs.CompleteRestoreOptions{}); err != nil {
+	if err := loadOpts.Load(ctx, k, nil, networkStack, time.NewCalibratedClocks(), &vfs.CompleteRestoreOptions{}); err != nil {
 		return err
 	}
 

runsc/boot/loader.go

@@ -278,19 +278,15 @@ func New(args Args) (*Loader, error) {
 	}
 
 	// Create timekeeper.
-	tk, err := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange())
-	if err != nil {
-		return nil, fmt.Errorf("creating timekeeper: %w", err)
-	}
+	tk := kernel.NewTimekeeper(k, vdso.ParamPage.FileRange())
 	tk.SetClocks(time.NewCalibratedClocks())
-	k.SetTimekeeper(tk)
 
 	if err := enableStrace(args.Conf); err != nil {
 		return nil, fmt.Errorf("enabling strace: %w", err)
 	}
 
 	// Create root network namespace/stack.
-	netns, err := newRootNetworkNamespace(args.Conf, k, k)
+	netns, err := newRootNetworkNamespace(args.Conf, tk, k)
 	if err != nil {
 		return nil, fmt.Errorf("creating network: %w", err)
 	}
@@ -332,6 +328,7 @@ func New(args Args) (*Loader, error) {
 	// to createVFS in order to mount (among other things) procfs.
 	if err = k.Init(kernel.InitKernelArgs{
 		FeatureSet:                  cpuid.HostFeatureSet(),
+		Timekeeper:                  tk,
 		RootUserNamespace:           creds.UserNamespace,
 		RootNetworkNamespace:        netns,
 		ApplicationCores:            uint(args.NumCPU),