kernel: hold TaskSet mutex across exitPtrace() and exitChildren()

This just saves needlessly bouncing the mutex. Compare Linux
kernel/exit.c:exit_notify() => write_lock_irq(&tasklist_lock);
forget_original_parent() => exit_ptrace().

PiperOrigin-RevId: 681181467

pkg/sentry/kernel/ptrace.go

@@ -567,10 +567,10 @@ func (t *Task) ptraceDetach(target *Task, sig linux.Signal) error {
 	return nil
 }
 
-// exitPtrace is called in the exit path to detach all of t's tracees.
-func (t *Task) exitPtrace() {
-	t.tg.pidns.owner.mu.Lock()
-	defer t.tg.pidns.owner.mu.Unlock()
+// exitPtraceLocked is called in the exit path to detach all of t's tracees.
+//
+// Preconditions: The TaskSet mutex must be locked for writing.
+func (t *Task) exitPtraceLocked() {
 	for target := range t.ptraceTracees {
 		if target.ptraceOpts.ExitKill {
 			target.tg.signalHandlers.mu.Lock()

pkg/sentry/kernel/task_exit.go

@@ -315,14 +315,15 @@ func (*runExitMain) execute(t *Task) taskRunState {
 		t.tg.Release(t)
 	}
 
+	t.tg.pidns.owner.mu.Lock()
 	// Detach tracees.
-	t.exitPtrace()
-
+	t.exitPtraceLocked()
 	// Reparent the task's children.
-	t.exitChildren()
+	t.exitChildrenLocked()
+	t.tg.pidns.owner.mu.Unlock()
 
-	// Don't tail-call runExitNotify, as exitChildren may have initiated a stop
-	// to wait for a PID namespace to die.
+	// Don't tail-call runExitNotify, as exitChildrenLocked may have initiated
+	// a stop to wait for a PID namespace to die.
 	return (*runExitNotify)(nil)
 }
 
@@ -360,9 +361,8 @@ func (t *Task) exitThreadGroup() bool {
 	return last
 }
 
-func (t *Task) exitChildren() {
-	t.tg.pidns.owner.mu.Lock()
-	defer t.tg.pidns.owner.mu.Unlock()
+// Preconditions: The TaskSet mutex must be locked for writing.
+func (t *Task) exitChildrenLocked() {
 	newParent := t.findReparentTargetLocked()
 	if newParent == nil {
 		// "If the init process of a PID namespace terminates, the kernel