Hi!
I'm afraid that requesting the "random" pattern does generate just one chunk
of data to write (sized at blocks_at_once * block_size bytes).  The same
chunk is then written to subsequent locations -- rather than generating
another piece of pseudorandom data.

This has at least two downsides:
* compression will cause less data to be written
* it won't detect faked capacity (such as a popular scam with USB sticks)
  as the default chunk length (64KB) divides expected values of
  wraparound.

What I expected was badblocks generating a continuous random stream, then
comparing it with the stream after starting from the same seed.