restart: make sure all tasks are in sys_restart

index 543b380..849bda5 100644
@@ -655,13 +655,35 @@ static struct ckpt_ctx *wait_checkpoint_ctx(void)
 static int do_ghost_task(void)
 {
 	struct ckpt_ctx *ctx;
+	int ret;
 
 	ctx = wait_checkpoint_ctx();
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	/*
+	 * Wait until coordinator task has completed prepare_descendants().
+	 * Note that prepare_descendants() wakes each task one a time
+	 * finishing the wait_checkpoint_ctx() above.  We may want to change
+	 * that flow so tasks only sleep once, but this works for now.
+	 */
+	ret = wait_event_interruptible(ctx->waitq,
+					atomic_read(&ctx->nr_running));
+	if (ret) {
+		ckpt_debug("wait for coordinator prep returned %d\n", ret);
+		return ret;
+	}
+	/*
+	 * Now each prepared task decrements nr_running, and, when that
+	 * hits zero, wakes the coordinator.  That way the coordinator
+	 * knows that all tasks are in sys_restart(0
+	 */
+	if (atomic_dec_and_test(&ctx->nr_running))
+		complete(&ctx->all_ready);
+
 	current->flags |= PF_RESTARTING;
 
+	/* Finally wait for all ghosts to be woken up - to die */
 	wait_event_interruptible(ctx->ghostq,
 				 all_tasks_activated(ctx) ||
 				 ckpt_test_ctx_error(ctx));

@@ -682,6 +704,26 @@ static int do_restore_task(void)
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	/*
+	 * Wait until coordinator task has completed prepare_descendants().
+	 * Note that prepare_descendants() wakes each task one a time
+	 * finishing the wait_checkpoint_ctx() above.  We may want to change
+	 * that flow so tasks only sleep once, but this works for now.
+	 */
+	ret = wait_event_interruptible(ctx->waitq,
+					atomic_read(&ctx->nr_running));
+	if (ret) {
+		ckpt_debug("wait for coordinator prep returned %d\n", ret);
+		return ret;
+	}
+	/*
+	 * Now each prepared task decrements nr_running, and, when that
+	 * hits zero, wakes the coordinator.  That way the coordinator
+	 * knows that all tasks are in sys_restart(0
+	 */
+	if (atomic_dec_and_test(&ctx->nr_running))
+		complete(&ctx->all_ready);
+
 	current->flags |= PF_RESTARTING;
 
 	/* wait for our turn, do the restore, and tell next task in line */

@@ -814,6 +856,7 @@ static int prepare_descendants(struct ckpt_ctx *ctx, struct task_struct *root)
 		ret = -ESRCH;
 
 	atomic_set(&ctx->nr_total, nr_pids);
+	atomic_set(&ctx->nr_running, nr_pids);
 	return ret;
 }
 

@@ -948,6 +991,16 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)
 		ret = prepare_descendants(ctx, ctx->root_task);
 		if (ret < 0)
 			goto out;
+
+		/* tell tasks we're ready for them to dec  ctx->nr_running */
+		wake_up_all(&ctx->waitq);
+		/* ctx->nr_running hit 0, all our tasks are in sys_restart */
+		ret = wait_for_completion_interruptible(&ctx->all_ready);
+		if (ret) {
+			ckpt_debug("waiting on all_ready returned %d\n", ret);
+			goto out;
+		}
+
 		/* wait for all other tasks to complete do_restore_task() */
 		ret = wait_all_tasks_finish(ctx);
 		if (ret < 0)

index 76a3fa9..d48e261 100644
@@ -233,6 +233,7 @@ static struct ckpt_ctx *ckpt_ctx_alloc(int fd, unsigned long uflags,
 	ctx->uflags = uflags;
 	ctx->kflags = kflags;
 	ctx->ktime_begin = ktime_get();
+	init_completion(&ctx->all_ready);
 
 	atomic_set(&ctx->refcount, 0);
 	INIT_LIST_HEAD(&ctx->pgarr_list);

index 9b7b4dd..2a854e0 100644
@@ -69,8 +69,10 @@ struct ckpt_ctx {
 	struct ckpt_pids *pids_arr;	/* array of all pids [restart] */
 	int nr_pids;			/* size of pids array */
 	atomic_t nr_total;		/* total tasks count (with ghosts) */
+	atomic_t nr_running;		/* countdown of tasks in sys_restart */
 	int active_pid;			/* (next) position in pids array */
 	struct completion complete;	/* completion for container root */
+	struct completion all_ready;	/* all tasks are in sys_restart */
 	wait_queue_head_t waitq;	/* waitqueue for restarting tasks */
 	wait_queue_head_t ghostq;	/* waitqueue for ghost tasks */
 	struct cred *realcred, *ecred;	/* tmp storage for cred at restart */

index 849bda5..1085ed5 100644
@@ -26,6 +26,98 @@
 #include <linux/checkpoint.h>
 #include <linux/checkpoint_hdr.h>
 
+#ifdef CONFIG_CHECKPOINT_DEBUG
+static struct ckpt_task_status *ckpt_debug_checkin(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+	s = kmalloc(sizeof(*s), GFP_KERNEL);
+	if (!s)
+		return NULL;
+	s->pid = current->pid;
+	s->error = 0;
+	s->flags = RESTART_DBG_WAITING;
+	if (current == ctx->root_task)
+		s->flags |= RESTART_DBG_ROOT;
+	list_add_tail(&s->list, &ctx->per_task_status);
+	return s;
+}
+
+static struct ckpt_task_status *getme(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s = NULL;
+	list_for_each_entry(s, &ctx->per_task_status, list) {
+		if (s->pid == current->pid)
+			break;
+	}
+	if (!s || s->pid != current->pid)
+		return NULL;
+	return s;
+}
+
+static void ckpt_debug_coord(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+
+	s = ckpt_debug_checkin(ctx);
+	if (s)
+		s->flags |= RESTART_DBG_COORD;
+}
+
+static void ckpt_debug_ghost(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+
+	s = ckpt_debug_checkin(ctx);
+	if (s)
+		s->flags |= RESTART_DBG_GHOST;
+}
+
+static void ckpt_debug_normal(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+
+	s = ckpt_debug_checkin(ctx);
+	if (s)
+		s->flags |= RESTART_DBG_NORMAL;
+}
+
+static void ckpt_debug_log_error(struct ckpt_ctx *ctx, int err)
+{
+	struct ckpt_task_status *s = getme(ctx);
+	if (!s)
+		return;
+	s->error = err;
+	s->flags &= ~RESTART_DBG_WAITING;
+	s->flags &= ~RESTART_DBG_RUNNING;
+	if (err)
+		s->flags |= RESTART_DBG_FAILED;
+	else
+		s->flags |= RESTART_DBG_SUCCESS;
+}
+
+static void ckpt_debug_log_running(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s = getme(ctx);
+	if (!s)
+		return;
+	s->flags &= ~RESTART_DBG_WAITING;
+	s->flags |= RESTART_DBG_RUNNING;
+}
+#else
+static inline void ckpt_debug_checkin(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_coord(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_ghost(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_normal(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_log_error(struct ckpt_ctx *ctx, int err)
+{}
+static inline void ckpt_debug_log_running(struct ckpt_ctx *ctx)
+{}
+#endif
+
 static int _ckpt_read_err(struct ckpt_ctx *ctx, struct ckpt_hdr *h)
 {
 	char *ptr;

@@ -661,6 +753,8 @@ static int do_ghost_task(void)
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	ckpt_debug_ghost(ctx);
+
 	/*
 	 * Wait until coordinator task has completed prepare_descendants().
 	 * Note that prepare_descendants() wakes each task one a time

@@ -681,6 +775,8 @@ static int do_ghost_task(void)
 	if (atomic_dec_and_test(&ctx->nr_running))
 		complete(&ctx->all_ready);
 
+	ckpt_debug_log_running(ctx);
+
 	current->flags |= PF_RESTARTING;
 
 	/* Finally wait for all ghosts to be woken up - to die */

@@ -688,6 +784,7 @@ static int do_ghost_task(void)
 				 all_tasks_activated(ctx) ||
 				 ckpt_test_ctx_error(ctx));
 
+	ckpt_debug_log_error(ctx, 0);
 	current->exit_signal = -1;
 	ckpt_ctx_put(ctx);
 	do_exit(0);

@@ -704,6 +801,8 @@ static int do_restore_task(void)
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	ckpt_debug_normal(ctx);
+
 	/*
 	 * Wait until coordinator task has completed prepare_descendants().
 	 * Note that prepare_descendants() wakes each task one a time

@@ -731,6 +830,8 @@ static int do_restore_task(void)
 	if (ret < 0)
 		goto out;
 
+	ckpt_debug_log_running(ctx);
+
 	zombie = restore_task(ctx);
 	if (zombie < 0) {
 		ret = zombie;

@@ -759,6 +860,7 @@ static int do_restore_task(void)
 	if (old_ctx)
 		ckpt_ctx_put(old_ctx);
 
+	ckpt_debug_log_error(ctx, ret);
 	/* if we're first to fail - notify others */
 	if (ret < 0 && !ckpt_test_ctx_error(ctx)) {
 		restore_notify_error(ctx, ret);

@@ -945,6 +1047,9 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)
 	struct ckpt_ctx *old_ctx;
 	int ret;
 
+	ckpt_debug_coord(ctx);
+	ckpt_debug_log_running(ctx);
+
 	ret = restore_read_header(ctx);
 	if (ret < 0)
 		return ret;

65 lines not on this page · raw message

index 2a854e0..1cda085 100644
@@ -76,10 +76,30 @@ struct ckpt_ctx {
 	wait_queue_head_t waitq;	/* waitqueue for restarting tasks */
 	wait_queue_head_t ghostq;	/* waitqueue for ghost tasks */
 	struct cred *realcred, *ecred;	/* tmp storage for cred at restart */
+#ifdef CONFIG_CHECKPOINT_DEBUG
+	struct list_head per_task_status;
+#endif
 
 	struct ckpt_stats stats;	/* statistics */
 };
 
+#ifdef CONFIG_CHECKPOINT_DEBUG
+struct ckpt_task_status {
+	pid_t pid;
+#define RESTART_DBG_ROOT	(1 << 0)
+#define RESTART_DBG_GHOST	(1 << 1)
+#define RESTART_DBG_COORD	(1 << 2)
+#define RESTART_DBG_NORMAL	(1 << 3)
+#define RESTART_DBG_WAITING	(1 << 4)
+#define RESTART_DBG_RUNNING	(1 << 5)
+#define RESTART_DBG_FAILED	(1 << 6)
+#define RESTART_DBG_SUCCESS	(1 << 7)
+	int flags;
+	int error;
+	struct list_head list;
+};
+#endif
+
 #endif /* __KERNEL__ */
 
 #endif /* _LINUX_CHECKPOINT_TYPES_H_ */

index b12c8bd..1f356c0 100644
@@ -26,6 +26,98 @@
 #include <linux/checkpoint.h>
 #include <linux/checkpoint_hdr.h>
 
+#ifdef CONFIG_CHECKPOINT_DEBUG
+static struct ckpt_task_status *ckpt_debug_checkin(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+	s = kmalloc(sizeof(*s), GFP_KERNEL);
+	if (!s)
+		return NULL;
+	s->pid = current->pid;
+	s->error = 0;
+	s->flags = RESTART_DBG_WAITING;
+	if (current == ctx->root_task)
+		s->flags |= RESTART_DBG_ROOT;
+	list_add_tail(&s->list, &ctx->per_task_status);
+	return s;
+}
+
+static struct ckpt_task_status *getme(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s = NULL;
+	list_for_each_entry(s, &ctx->per_task_status, list) {
+		if (s->pid == current->pid)
+			break;
+	}
+	if (!s || s->pid != current->pid)
+		return NULL;
+	return s;
+}
+
+static void ckpt_debug_coord(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+
+	s = ckpt_debug_checkin(ctx);
+	if (s)
+		s->flags |= RESTART_DBG_COORD;
+}
+
+static void ckpt_debug_ghost(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+
+	s = ckpt_debug_checkin(ctx);
+	if (s)
+		s->flags |= RESTART_DBG_GHOST;
+}
+
+static void ckpt_debug_normal(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s;
+
+	s = ckpt_debug_checkin(ctx);
+	if (s)
+		s->flags |= RESTART_DBG_NORMAL;
+}
+
+static void ckpt_debug_log_error(struct ckpt_ctx *ctx, int err)
+{
+	struct ckpt_task_status *s = getme(ctx);
+	if (!s)
+		return;
+	s->error = err;
+	s->flags &= ~RESTART_DBG_WAITING;
+	s->flags &= ~RESTART_DBG_RUNNING;
+	if (err)
+		s->flags |= RESTART_DBG_FAILED;
+	else
+		s->flags |= RESTART_DBG_SUCCESS;
+}
+
+static void ckpt_debug_log_running(struct ckpt_ctx *ctx)
+{
+	struct ckpt_task_status *s = getme(ctx);
+	if (!s)
+		return;
+	s->flags &= ~RESTART_DBG_WAITING;
+	s->flags |= RESTART_DBG_RUNNING;
+}
+#else
+static inline void ckpt_debug_checkin(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_coord(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_ghost(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_normal(struct ckpt_ctx *ctx)
+{}
+static inline void ckpt_debug_log_error(struct ckpt_ctx *ctx, int err)
+{}
+static inline void ckpt_debug_log_running(struct ckpt_ctx *ctx)
+{}
+#endif
+
 static int _ckpt_read_err(struct ckpt_ctx *ctx, struct ckpt_hdr *h)
 {
 	char *ptr;

@@ -680,11 +772,17 @@ static int do_ghost_task(void)
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	ckpt_debug_ghost(ctx);
+
+	ckpt_debug_log_running(ctx);
+
 	current->flags |= PF_RESTARTING;
 
 	ret = wait_event_interruptible(ctx->ghostq,
 				       all_tasks_activated(ctx) ||
 				       ckpt_test_ctx_error(ctx));
+
+	ckpt_debug_log_error(ctx, 0);
 	if (ret < 0)
 		restore_notify_error(ctx, ret);
 

@@ -752,6 +850,8 @@ static int do_restore_task(void)
 	if (IS_ERR(ctx))
 		return PTR_ERR(ctx);
 
+	ckpt_debug_normal(ctx);
+
 	current->flags |= PF_RESTARTING;
 
 	ret = wait_sync_threads();

@@ -767,6 +867,8 @@ static int do_restore_task(void)
 	if (ret < 0)
 		goto out;
 
+	ckpt_debug_log_running(ctx);
+
 	zombie = restore_task(ctx);
 	if (zombie < 0) {
 		ret = zombie;

@@ -791,6 +893,7 @@ static int do_restore_task(void)
 	restore_task_done(ctx);
 	ret = wait_task_sync(ctx);
  out:
+	ckpt_debug_log_error(ctx, ret);
 	if (ret < 0)
 		restore_notify_error(ctx, ret);
 

@@ -964,6 +1067,9 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)
 {
 	int ret;
 
+	ckpt_debug_coord(ctx);
+	ckpt_debug_log_running(ctx);
+
 	ret = restore_read_header(ctx);
 	ckpt_debug("restore header: %d\n", ret);
 	if (ret < 0)

73 lines not on this page · raw message

index b9393f4..198d4e9 100644
@@ -78,10 +78,30 @@ struct ckpt_ctx {
 	wait_queue_head_t waitq;	/* waitqueue for restarting tasks */
 	wait_queue_head_t ghostq;	/* waitqueue for ghost tasks */
 	struct cred *realcred, *ecred;	/* tmp storage for cred at restart */
+#ifdef CONFIG_CHECKPOINT_DEBUG
+	struct list_head per_task_status; /* list of status for each task */
+#endif
 
 	struct ckpt_stats stats;	/* statistics */
 };
 
+#ifdef CONFIG_CHECKPOINT_DEBUG
+struct ckpt_task_status {
+	pid_t pid;
+#define RESTART_DBG_ROOT	(1 << 0)
+#define RESTART_DBG_GHOST	(1 << 1)
+#define RESTART_DBG_COORD	(1 << 2)
+#define RESTART_DBG_NORMAL	(1 << 3)
+#define RESTART_DBG_WAITING	(1 << 4)
+#define RESTART_DBG_RUNNING	(1 << 5)
+#define RESTART_DBG_FAILED	(1 << 6)
+#define RESTART_DBG_SUCCESS	(1 << 7)
+	int flags;
+	int error;
+	struct list_head list;
+};
+#endif
+
 #endif /* __KERNEL__ */
 
 #endif /* _LINUX_CHECKPOINT_TYPES_H_ */

index 1085ed5..ebdd5ba 100644
@@ -662,6 +662,7 @@ static int restore_activate_next(struct ckpt_ctx *ctx)
 		rcu_read_unlock();
 
 		if (!task) {
+			ckpt_debug("could not find task %d\n", pid);
 			restore_notify_error(ctx, -ESRCH);
 			return -ESRCH;
 		}

@@ -685,6 +686,7 @@ static int wait_task_active(struct ckpt_ctx *ctx)
 	ckpt_debug("active %d < %d (ret %d)\n",
 		   ctx->active_pid, ctx->nr_pids, ret);
 	if (!ret && ckpt_test_ctx_error(ctx)) {
+		ckpt_debug("wait_event_interruptible returned %d\n", ret);
 		force_sig(SIGKILL, current);
 		ret = -EBUSY;
 	}

@@ -827,20 +829,25 @@ static int do_restore_task(void)
 
 	/* wait for our turn, do the restore, and tell next task in line */
 	ret = wait_task_active(ctx);
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("wait_task_active returned %d\n", ret);
 		goto out;
+	}
 
 	ckpt_debug_log_running(ctx);
 
 	zombie = restore_task(ctx);
 	if (zombie < 0) {
+		ckpt_debug("restore_task returned %d\n", ret);
 		ret = zombie;
 		goto out;
 	}
 
 	ret = restore_activate_next(ctx);
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("restore_activate_next returned %d\n", ret);
 		goto out;
+	}
 
 	/*
 	 * zombie: we're done here; do_exit() will notice the @ctx on

@@ -855,6 +862,8 @@ static int do_restore_task(void)
 
 	restore_task_done(ctx);
 	ret = wait_task_sync(ctx);
+	if (ret)
+		ckpt_debug("wait_task_sync returned %d\n", ret);
  out:
 	old_ctx = xchg(&current->checkpoint_ctx, NULL);
 	if (old_ctx)

@@ -970,12 +979,14 @@ static int wait_all_tasks_finish(struct ckpt_ctx *ctx)
 
 	BUG_ON(ctx->active_pid != -1);
 	ret = restore_activate_next(ctx);
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("restore_activate_next returned %d\n", ret);
 		return ret;
+	}
 
 	ret = wait_for_completion_interruptible(&ctx->complete);
 
-	ckpt_debug("final sync kflags %#lx\n", ctx->kflags);
+	ckpt_debug("final sync kflags %#lx (ret %d)\n", ctx->kflags, ret);
 	/*
 	 * Usually when restart fails, the restarting task will first
 	 * set @ctx->errno before waking us up. In the rare event that

@@ -1051,18 +1062,24 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)
 	ckpt_debug_log_running(ctx);
 
 	ret = restore_read_header(ctx);
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("restore_read_header returned %d\n", ret);
 		return ret;
+	}
 	ret = restore_read_tree(ctx);
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("restore_read_tree returned %d\n", ret);
 		return ret;
+	}
 
 	if ((ctx->uflags & RESTART_TASKSELF) && ctx->nr_pids != 1)
 		return -EINVAL;
 
 	ret = init_restart_ctx(ctx, pid);
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("init_restart_ctx returned %d\n", ret);
 		return ret;
+	}
 
 	/*
 	 * Populate own ->checkpoint_ctx: if an ancestor attempts to

@@ -1094,8 +1111,10 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)
 	} else {
 		/* prepare descendants' t->checkpoint_ctx point to coord */
 		ret = prepare_descendants(ctx, ctx->root_task);
-		if (ret < 0)
+		if (ret < 0) {
+			ckpt_debug("prepare_descendants returned %d", ret);
 			goto out;
+		}
 
 		/* tell tasks we're ready for them to dec  ctx->nr_running */
 		wake_up_all(&ctx->waitq);

@@ -1108,17 +1127,23 @@ static int do_restore_coord(struct ckpt_ctx *ctx, pid_t pid)
 
 		/* wait for all other tasks to complete do_restore_task() */
 		ret = wait_all_tasks_finish(ctx);
-		if (ret < 0)
+		if (ret < 0) {
+			ckpt_debug("wait_all_tasks_finish returned %d", ret);
 			goto out;
+		}
 	}
 
 	ret = deferqueue_run(ctx->deferqueue);  /* run deferred work */
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("deferqueue_run returned %d\n", ret);
 		goto out;
+	}
 
 	ret = restore_read_tail(ctx);
-	if (ret < 0)
+	if (ret < 0) {
+		ckpt_debug("restore_read_tail returned %d\n", ret);
 		goto out;
+	}
 
 	if (ctx->uflags & RESTART_FROZEN) {
 		ret = cgroup_freezer_make_frozen(ctx->root_task);

@@ -1202,6 +1227,9 @@ long do_restart(struct ckpt_ctx *ctx, pid_t pid, unsigned long flags)
 	else
 		ret = do_restore_task();
 
+	if (ret < 0)
+		ckpt_debug("ret %ld\n", ret);
+
 	/* restart(2) isn't idempotent: should not be auto-restarted */
 	if (ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
 	    ret == -ERESTARTNOHAND || ret == -ERESTART_RESTARTBLOCK)