btrfs: add an ASSERT() to catch ordered extents without datasum

Inside btrfs_finish_one_ordered(), there are only very limited
situations where the OE has no checksum:

- The OE is completely truncated or error happened
  In that case no file extent is going to be inserted.

- The inode has NODATASUM flag

- The inode belongs to data reloc tree

Add an ASSERT() using the last two cases, which will help us to catch
problems described in commit 18de34d ("btrfs: truncate ordered
extent when skipping writeback past i_size"), and prevent future similar
cases.

Signed-off-by: Qu Wenruo <[email protected]>
Signed-off-by: David Sterba <[email protected]>

Author: adam900710

fs/btrfs/inode.c

@@ -3282,6 +3282,21 @@ int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
 		goto out;
 	}
 
+	/*
+	 * If we have no data checksum, either the OE is:
+	 * - Fully truncated
+	 *   Those ones won't reach here.
+	 *
+	 * - No data checksum
+	 *
+	 * - Belongs to data reloc inode
+	 *   Which doesn't have csum attached to OE, but cloned
+	 *   from original chunk.
+	 */
+	if (list_empty(&ordered_extent->list))
+		ASSERT(inode->flags & BTRFS_INODE_NODATASUM ||
+		       btrfs_is_data_reloc_root(inode->root));
+
 	ret = add_pending_csums(trans, &ordered_extent->list);
 	if (unlikely(ret)) {
 		btrfs_abort_transaction(trans, ret);