X-Git-Url: http://git.tuebingen.mpg.de/?a=blobdiff_plain;f=sched.c;h=fe57e03eae01a1d4360ae41cdf61f893a9888128;hb=deb3326320166531abc35ce805dee4df6b49e936;hp=8b68667bf453c481df6950ec727be3b134037f99;hpb=bfdd36f2879b8dee1f6d3330bc49fbb5644d3caf;p=paraslash.git

diff --git a/sched.c b/sched.c
index 8b68667b..fe57e03e 100644
--- a/sched.c
+++ b/sched.c
@@ -21,25 +21,6 @@
 static struct timeval now_struct;
 struct timeval *now = &now_struct;
 
-/*
- * Remove a task from the scheduler.
- *
- * \param t The task to remove.
- *
- * If the pre_select pointer of \a t is not \p NULL, it is removed from
- * the pre_select list of the scheduler. Same goes for \a post_select.
- */
-static void unregister_task(struct task *t)
-{
-	assert(t->error < 0);
-	PARA_INFO_LOG("unregistering %s (%s)\n", t->status,
-		para_strerror(-t->error));
-	if (t->pre_select)
-		list_del(&t->pre_select_node);
-	if (t->post_select)
-		list_del(&t->post_select_node);
-}
-
 static inline bool timeout_is_zero(struct sched *s)
 {
 	struct timeval *tv = &s->select_timeout;
@@ -50,7 +31,9 @@ static void sched_preselect(struct sched *s)
 {
 	struct task *t, *tmp;
 
-	list_for_each_entry_safe(t, tmp, &s->pre_select_list, pre_select_node) {
+	list_for_each_entry_safe(t, tmp, &s->task_list, node) {
+		if (t->error < 0)
+			continue;
 		if (t->notification != 0)
 			sched_min_delay(s);
 		if (t->pre_select)
@@ -58,6 +41,14 @@ static void sched_preselect(struct sched *s)
 	}
 }
 
+static void unlink_and_free_task(struct task *t)
+{
+	PARA_INFO_LOG("freeing task %s\n", t->status);
+	list_del(&t->node);
+	if (t->owned_by_sched)
+		free(t);
+}
+
 //#define SCHED_DEBUG 1
 static inline void call_post_select(struct sched *s, struct task *t)
 {
@@ -78,23 +69,30 @@ static inline void call_post_select(struct sched *s, struct task *t)
 #endif
 }
 
-static void sched_post_select(struct sched *s)
+static unsigned sched_post_select(struct sched *s)
 {
 	struct task *t, *tmp;
+	unsigned num_running_tasks = 0;
 
-	list_for_each_entry_safe(t, tmp, &s->post_select_list, post_select_node) {
-		if (t->error >= 0)
-			call_post_select(s, t);
-//		PARA_INFO_LOG("%s: %d\n", t->status, t->ret);
-		t->notification = 0;
-		if (t->error >= 0)
+	list_for_each_entry_safe(t, tmp, &s->task_list, node) {
+		if (t->error < 0) {
+			if (t->dead) /* task has been reaped */
+				unlink_and_free_task(t);
 			continue;
-		unregister_task(t);
+		}
+		call_post_select(s, t);
+		t->notification = 0;
+		if (t->error < 0) {
+			if (!t->owned_by_sched)
+				list_del(&t->node);
+		} else
+			num_running_tasks++;
 	}
+	return num_running_tasks;
 }
 
 /**
- * The core function for all paraslash programs.
+ * The core function of all paraslash programs.
  *
  * \param s Pointer to the scheduler struct.
  *
@@ -103,14 +101,15 @@ static void sched_post_select(struct sched *s)
  * the fd sets of \a s.  Next, it calls para_select() and makes the result available
  * to the registered tasks by calling their post_select hook.
  *
- * \return Zero if no more tasks are left in either of the two lists, negative
- * if para_select returned an error.
+ * \return Zero if no more tasks are left in the task list, negative if the
+ * select function returned an error.
  *
- * \sa task, now.
+ * \sa \ref task, \ref now.
  */
 int schedule(struct sched *s)
 {
 	int ret;
+	unsigned num_running_tasks;
 
 	if (!s->select_function)
 		s->select_function = para_select;
@@ -136,41 +135,124 @@ again:
 		FD_ZERO(&s->wfds);
 	}
 	clock_get_realtime(now);
-	sched_post_select(s);
-	if (list_empty(&s->pre_select_list) && list_empty(&s->post_select_list))
+	num_running_tasks = sched_post_select(s);
+	if (num_running_tasks == 0)
 		return 0;
 	goto again;
 }
 
 /**
- * Add a task to the scheduler.
+ * Obtain the error status of a task and deallocate its resources.
  *
- * \param t The task to add.
- * \param s The scheduler instance to add the task to.
+ * \param tptr Identifies the task to reap.
  *
- * If the pre_select pointer of \a t is not \p NULL, it is added to
- * the pre_select list of the scheduler. Same goes for post_select.
+ * This function is similar to wait(2) in that it returns information about a
+ * terminated task and allows to release the resources associated with the
+ * task. Until this function is called, the terminated task remains in a zombie
+ * state.
  *
- * \sa task::pre_select, task::post_select
+ * \return If \a tptr is \p NULL, or \a *tptr is \p NULL, the function does
+ * nothing and returns zero. Otherwise, it is checked whether the task
+ * identified by \a tptr is still running. If it is, the function returns zero
+ * and again, no action is taken. Otherwise the (negative) error code of the
+ * terminated task is returned and \a *tptr is set to \p NULL. The task will
+ * then be removed removed from the scheduler task list.
+ *
+ * \sa \ref sched_shutdown(), wait(2).
  */
-void register_task(struct sched *s, struct task *t)
+int task_reap(struct task **tptr)
 {
-	PARA_INFO_LOG("registering %s (%p)\n", t->status, t);
-	t->notification = 0;
-	if (!s->pre_select_list.next)
-		INIT_LIST_HEAD(&s->pre_select_list);
-	if (!s->post_select_list.next)
-		INIT_LIST_HEAD(&s->post_select_list);
-	if (t->pre_select) {
-		PARA_DEBUG_LOG("pre_select: %p\n", &t->pre_select);
-		list_add_tail(&t->pre_select_node, &s->pre_select_list);
-	}
-	if (t->post_select) {
-		PARA_DEBUG_LOG("post_select: %p\n", &t->post_select);
-		list_add_tail(&t->post_select_node, &s->post_select_list);
+	struct task *t;
+
+	if (!tptr)
+		return 0;
+	t = *tptr;
+	if (!t)
+		return 0;
+	if (!t->owned_by_sched)
+		return 0;
+	if (t->error >= 0)
+		return 0;
+	if (t->dead) /* will be freed in sched_post_select() */
+		return 0;
+	/*
+	 * With list_for_each_entry_safe() it is only safe to remove the
+	 * _current_ list item. Since we are being called from the loop in
+	 * schedule() via some task's ->post_select() function, freeing the
+	 * given task here would result in use-after-free bugs in schedule().
+	 * So we only set t->dead which tells schedule() to free the task in
+	 * the next iteration of its loop.
+	 */
+	t->dead = true;
+	*tptr = NULL;
+	return t->error;
+}
+
+/**
+ * Deallocate all resources of all tasks of a scheduler instance.
+ *
+ * \param s The scheduler instance.
+ *
+ * This should only be called after \ref schedule() has returned.
+ */
+void sched_shutdown(struct sched *s)
+{
+	struct task *t, *tmp;
+
+	list_for_each_entry_safe(t, tmp, &s->task_list, node) {
+		if (t->error >= 0)
+			/* The task list should contain only terminated tasks. */
+			PARA_WARNING_LOG("shutting down running task %s\n",
+				t->status);
+		unlink_and_free_task(t);
 	}
 }
 
+/**
+ * Add a task to the scheduler task list.
+ *
+ * \param info Task information supplied by the caller.
+ * \param s The scheduler instance.
+ *
+ * \return A pointer to a newly allocated task structure. It will be
+ * freed by sched_shutdown().
+ */
+struct task *task_register(struct task_info *info, struct sched *s)
+{
+	struct task *t = para_malloc(sizeof(*t));
+
+	assert(info->post_select);
+
+	if (!s->task_list.next)
+		INIT_LIST_HEAD(&s->task_list);
+
+	snprintf(t->status, sizeof(t->status) - 1, "%s", info->name);
+	t->status[sizeof(t->status) - 1] = '\0';
+	t->notification = 0;
+	t->error = 0;
+	t->dead = false;
+	t->pre_select = info->pre_select;
+	t->post_select = info->post_select;
+	t->context = info->context;
+	t->owned_by_sched = true;
+	list_add_tail(&t->node, &s->task_list);
+	return t;
+}
+
+/**
+ * Obtain the context pointer of a task.
+ *
+ * \param t Return this task's context pointer.
+ *
+ * \return A pointer to the memory location specified previously as \a
+ * task_info->context when the task was registered with \ref task_register().
+ */
+void *task_context(struct task *t)
+{
+	assert(t->owned_by_sched);
+	return t->context;
+}
+
 /**
  * Get the list of all registered tasks.
  *
@@ -186,21 +268,14 @@ char *get_task_list(struct sched *s)
 	struct task *t, *tmp;
 	char *msg = NULL;
 
-	list_for_each_entry_safe(t, tmp, &s->pre_select_list, pre_select_node) {
+	list_for_each_entry_safe(t, tmp, &s->task_list, node) {
 		char *tmp_msg;
-		tmp_msg = make_message("%s%p\tpre\t%s\n", msg? msg : "", t, t->status);
+		tmp_msg = make_message("%s%p\t%s\t%s\n", msg? msg : "", t,
+			t->error < 0? (t->dead? "dead" : "zombie") : "running",
+			t->status);
 		free(msg);
 		msg = tmp_msg;
 	}
-	list_for_each_entry_safe(t, tmp, &s->post_select_list, post_select_node) {
-		char *tmp_msg;
-//		if (t->pre_select)
-//			continue;
-		tmp_msg = make_message("%s%p\tpost\t%s\n", msg? msg : "", t, t->status);
-		free(msg);
-		msg = tmp_msg;
-	}
-	//PARA_DEBUG_LOG("task list:\n%s", msg);
 	return msg;
 }
 
@@ -243,7 +318,7 @@ void task_notify(struct task *t, int err)
  *
  * \sa \ref task_notify().
  */
-int task_get_notification(struct task *t)
+int task_get_notification(const struct task *t)
 {
 	return t->notification;
 }
@@ -261,9 +336,7 @@ void task_notify_all(struct sched *s, int err)
 {
 	struct task *t;
 
-	list_for_each_entry(t, &s->pre_select_list, pre_select_node)
-		task_notify(t, err);
-	list_for_each_entry(t, &s->post_select_list, post_select_node)
+	list_for_each_entry(t, &s->task_list, node)
 		task_notify(t, err);
 }