diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index 1d32a07bb2d1fb..309516e6a96820 100644
--- a/fs/btrfs/async-thread.c
+++ b/fs/btrfs/async-thread.c
@@ -395,3 +395,11 @@ void btrfs_set_work_high_priority(struct btrfs_work *work)
 {
 	set_bit(WORK_HIGH_PRIO_BIT, &work->flags);
 }
+
+void btrfs_flush_workqueue(struct btrfs_workqueue *wq)
+{
+	if (wq->high)
+		flush_workqueue(wq->high->normal_wq);
+
+	flush_workqueue(wq->normal->normal_wq);
+}
diff --git a/fs/btrfs/async-thread.h b/fs/btrfs/async-thread.h
index a4434301d84dd5..3204daa51b9557 100644
--- a/fs/btrfs/async-thread.h
+++ b/fs/btrfs/async-thread.h
@@ -44,5 +44,6 @@ void btrfs_set_work_high_priority(struct btrfs_work *work);
 struct btrfs_fs_info * __pure btrfs_work_owner(const struct btrfs_work *work);
 struct btrfs_fs_info * __pure btrfs_workqueue_owner(const struct __btrfs_workqueue *wq);
 bool btrfs_workqueue_normal_congested(const struct btrfs_workqueue *wq);
+void btrfs_flush_workqueue(struct btrfs_workqueue *wq);
 
 #endif
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index f01f19c76cf594..483b54077d01f4 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -4067,6 +4067,19 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 		 */
 		btrfs_delete_unused_bgs(fs_info);
 
+		/*
+		 * There might be existing delayed inode workers still running
+		 * and holding an empty delayed inode item. We must wait for
+		 * them to complete first because they can create a transaction.
+		 * This happens when someone calls btrfs_balance_delayed_items()
+		 * and then a transaction commit runs the same delayed nodes
+		 * before any delayed worker has done something with the nodes.
+		 * We must wait for any worker here and not at transaction
+		 * commit time since that could cause a deadlock.
+		 * This is a very rare case.
+		 */
+		btrfs_flush_workqueue(fs_info->delayed_workers);
+
 		ret = btrfs_commit_super(fs_info);
 		if (ret)
 			btrfs_err(fs_info, "commit super ret %d", ret);