> >I like Chris's idea of a generic "make sure the data is sync'd" ioctl
> >that the buffer cache and/or file system code can use.
> 
> No, that's really not the right thing to do I believe, *unless* you
> go to the effort of an ioctl that also *enables* device cacheing.
> I believe that block devices should not have this cache enabled 
> unless requested (and this should be enforced in the driver).

I agree.  However, I don't know enough about the cache operations on
SCSI and other drives to know if that's easily possible to accomplish
on all (or the vast majority of) drives.  You and others are much more
qualified in that regard.  8-) (It's worth noting that there are other
implications of not forcing sync writes through to the disk.
E.g. what does fsync() really force?)

Even running async with caches enabled, i'd _still_ expect 'sync' to
flush the written bits all the way to the platter.  If that property
can be obtained by forcing synchronous writes to hit the disk, great.
That'll solve the problem.  However, if that's not possible, some
explicit, device-specific 'sync' operation (e.g. an internal-use
IOCTL) might be necessary.



cgd