Comment 31 for bug 1070539

On consideration I've decided to be more conservative with the Folsom patch
and not use O_DIRECT. John Griffith has seen that writing to the end of
snapshot volumes (the last 6MiB) can give EIO when O_DIRECT is in effect.
There were no issues seen when writing to standard volumes, as is the case here.
But since this is only a performance consideration, I'm thinking it's better
to be a bit conservative here, lest we trigger an issue on some kernels.