Fair enough.
Something I dislike about it, though, is that it leaves the RAM-backed
filesystems (ramfs, tmpfs, whatever) behaving visibly differently from
the others. Until now we've intentionally left them out of syncing and
dirty accounting, because it's useless overhead for them (one can argue
whether that's quite true of tmpfs overflowing out to swap, but that's
a different debate). So they won't be getting these faults on shared
writable, so their file times won't get updated in the same way.
But I guess that's an aesthetic consideration, of less significance
than bad backups - assuming not many people use backups of tmpfs.
Hugh
--
