My question, though, concerns the origin of the problem. There's
long been code to
completely and successfully flush all metadata before unmounting a
volume -- why
does that work, when this doesn't?
FreeBSD fixed this bug ( in 7.2 IIRC ). Pick up the fix from there.I used to always read that the various BSDs picked up bugfixes from each other,
but this doesn't seem to happen in practice. Why not?