diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 947f04789389e9..ebb2d109e8bb2c 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -602,6 +602,9 @@ enum {
 	/* Indicate that we want the transaction kthread to commit right now. */
 	BTRFS_FS_COMMIT_TRANS,
 
+	/* Indicate we have half completed snapshot deletions pending. */
+	BTRFS_FS_UNFINISHED_DROPS,
+
 #if BITS_PER_LONG == 32
 	/* Indicate if we have error/warn message printed on 32bit systems */
 	BTRFS_FS_32BIT_ERROR,
@@ -1106,8 +1109,15 @@ enum {
 	BTRFS_ROOT_QGROUP_FLUSHING,
 	/* We started the orphan cleanup for this root. */
 	BTRFS_ROOT_ORPHAN_CLEANUP,
+	/* This root has a drop operation that was started previously. */
+	BTRFS_ROOT_UNFINISHED_DROP,
 };
 
+static inline void btrfs_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
+{
+	clear_and_wake_up_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
+}
+
 /*
  * Record swapped tree blocks of a subvolume tree for delayed subtree trace
  * code. For detail check comment in fs/btrfs/qgroup.c.
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 87a5addbedf6d5..48590a3807621a 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -3813,6 +3813,10 @@ int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_device
 
 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
 
+	/* Kick the cleaner thread so it'll start deleting snapshots. */
+	if (test_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags))
+		wake_up_process(fs_info->cleaner_kthread);
+
 clear_oneshot:
 	btrfs_clear_oneshot_options(fs_info);
 	return 0;
@@ -4538,6 +4542,12 @@ void __cold close_ctree(struct btrfs_fs_info *fs_info)
 	 */
 	kthread_park(fs_info->cleaner_kthread);
 
+	/*
+	 * If we had UNFINISHED_DROPS we could still be processing them, so
+	 * clear that bit and wake up relocation so it can stop.
+	 */
+	btrfs_wake_unfinished_drop(fs_info);
+
 	/* wait for the qgroup rescan worker to stop */
 	btrfs_qgroup_wait_for_completion(fs_info, false);
 
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index d89273c4b6b86e..96427b1ecac3ea 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -5622,6 +5622,7 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
 	int ret;
 	int level;
 	bool root_dropped = false;
+	bool unfinished_drop = false;
 
 	btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
 
@@ -5664,6 +5665,8 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
 	 * already dropped.
 	 */
 	set_bit(BTRFS_ROOT_DELETING, &root->state);
+	unfinished_drop = test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
+
 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
 		level = btrfs_header_level(root->node);
 		path->nodes[level] = btrfs_lock_root_node(root);
@@ -5838,6 +5841,13 @@ int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
 	kfree(wc);
 	btrfs_free_path(path);
 out:
+	/*
+	 * We were an unfinished drop root, check to see if there are any
+	 * pending, and if not clear and wake up any waiters.
+	 */
+	if (!err && unfinished_drop)
+		btrfs_maybe_wake_unfinished_drop(fs_info);
+
 	/*
 	 * So if we need to stop dropping the snapshot for whatever reason we
 	 * need to make sure to add it back to the dead root list so that we
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index f5465197996dec..9d8054839782e2 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -3960,6 +3960,19 @@ int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
 	int rw = 0;
 	int err = 0;
 
+	/*
+	 * This only gets set if we had a half-deleted snapshot on mount.  We
+	 * cannot allow relocation to start while we're still trying to clean up
+	 * these pending deletions.
+	 */
+	ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE);
+	if (ret)
+		return ret;
+
+	/* We may have been woken up by close_ctree, so bail if we're closing. */
+	if (btrfs_fs_closing(fs_info))
+		return -EINTR;
+
 	bg = btrfs_lookup_block_group(fs_info, group_start);
 	if (!bg)
 		return -ENOENT;
diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
index 3d68d2dcd83e3e..ca7426ef61c8a1 100644
--- a/fs/btrfs/root-tree.c
+++ b/fs/btrfs/root-tree.c
@@ -278,6 +278,21 @@ int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
 
 		WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
 		if (btrfs_root_refs(&root->root_item) == 0) {
+			struct btrfs_key drop_key;
+
+			btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
+			/*
+			 * If we have a non-zero drop_progress then we know we
+			 * made it partly through deleting this snapshot, and
+			 * thus we need to make sure we block any balance from
+			 * happening until this snapshot is completely dropped.
+			 */
+			if (drop_key.objectid != 0 || drop_key.type != 0 ||
+			    drop_key.offset != 0) {
+				set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
+				set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
+			}
+
 			set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
 			btrfs_add_dead_root(root);
 		}
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index c3cfdfd8de9b2d..f17bf3764ce86c 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -1319,6 +1319,32 @@ static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans)
 	return 0;
 }
 
+/*
+ * If we had a pending drop we need to see if there are any others left in our
+ * dead roots list, and if not clear our bit and wake any waiters.
+ */
+void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info)
+{
+	/*
+	 * We put the drop in progress roots at the front of the list, so if the
+	 * first entry doesn't have UNFINISHED_DROP set we can wake everybody
+	 * up.
+	 */
+	spin_lock(&fs_info->trans_lock);
+	if (!list_empty(&fs_info->dead_roots)) {
+		struct btrfs_root *root = list_first_entry(&fs_info->dead_roots,
+							   struct btrfs_root,
+							   root_list);
+		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) {
+			spin_unlock(&fs_info->trans_lock);
+			return;
+		}
+	}
+	spin_unlock(&fs_info->trans_lock);
+
+	btrfs_wake_unfinished_drop(fs_info);
+}
+
 /*
  * dead roots are old snapshots that need to be deleted.  This allocates
  * a dirty root struct and adds it into the list of dead roots that need to
@@ -1331,7 +1357,12 @@ void btrfs_add_dead_root(struct btrfs_root *root)
 	spin_lock(&fs_info->trans_lock);
 	if (list_empty(&root->root_list)) {
 		btrfs_grab_root(root);
-		list_add_tail(&root->root_list, &fs_info->dead_roots);
+
+		/* We want to process the partially complete drops first. */
+		if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state))
+			list_add(&root->root_list, &fs_info->dead_roots);
+		else
+			list_add_tail(&root->root_list, &fs_info->dead_roots);
 	}
 	spin_unlock(&fs_info->trans_lock);
 }
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index 9402d8d944846f..ba8a9826eb3756 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -216,6 +216,7 @@ int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid);
 
 void btrfs_add_dead_root(struct btrfs_root *root);
 int btrfs_defrag_root(struct btrfs_root *root);
+void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info);
 int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root);
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans);
 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans);