Re: [prepatch] address_space-based writeback

[Christoph Hellwig]

On Thu, Apr 11, 2002 at 04:55:36PM -0600, Andreas Dilger wrote:
> At one time Linus proposed having an array of dirty bits for a page,
> which would allow us to mark only parts of a page dirty (say down to
> the sector level).  I believe this was in the discussion about moving
> the block devices to the page cache around 2.4.10.

The early XFS code used to do this for kiobuf-based block I/O, but it
got dropped around 2.4.8 IIRC.  The new page->private handling from akpm
which seems to be merged in Linus' BK tree (Linus: please push it to
bkbits.net, thanks :)) can be used to do the same if and only if we don't
need the buffer_heads attached to the page.