Ugly workaround for an ugly bug.

Because my Toshiba USB3.0 HDD is broken beyond repair and it will go to sleep after about 20 seconds of inactivity. No amount of hdparm or sdparm can fix the thing, so the only alternative is writing something before the timeout expires.

Q: Can't you just ask Toshiba to fix it?
A: According to them, this is a "feature" to save power so there is nothing to fix. The fact that OS caches might still be dirty and some xHCI controllers might evict the device altogether from the bus pales in comparison of the few mW saved! Go green!

Q: What's the problem with the disk spinning down periodically? I mean, come on!
A: See the previous question: some xHCI controllers react to the device powering down by evicting it from the bus. To be honest, the device-side interface should NEVER power down, but apparently Toshiba is really hell-bent on saving power.

Q: Ok, I get it. But why the whole O_DIRECT|O_SYNC mess?
A: The OS cache can (and will) get in your way.

Q: It doesn't work with NTFS or FAT32!
A: INVALID/WONTFIX (sounds familiar, doesn't it?).

Same, boring old 2-clause BSD. See LICENSE for details.

Issues are not welcome (works for me), PRs are welcome iff they make sense and they don't break my use case.

If your PR gets rejected and you badly need your feature, feel free to fork.