From glancing at the source code I believe this uses the same algorithm as df from coreutils. It outputs pretty much useless data for btrfs, which needs to be handled differently. See:

  $ btrfs fi df
  $ btrfs fi us