Re: [PATCH, RFC] map multiple blocks at a time in mpage_readpage(s)

[Christoph Hellwig <hch@lst.de>]

On Tue, Oct 04, 2005 at 04:15:43PM -0700, Badari Pulavarty wrote:
> Do you really want mpage_readpage()/mpage_readpages() to take
> get_blocks() and get_block() to take as arguments ? Or it just
> a initial version and you have grand plans to clean it up ?
> Me & Suparna did a version which needed only get_blocks() 
> few months ago. I can try to dig it up, if you like.

I don't see how to get rid of it easily.  It's need for the confused
case that calls block_read_full_page, which needs a get_block()
callback.  Adding a failback_writepage callback would be a tad nicer
but not really solve the general issue.

> In fact, we want to do the same thing for writepages() also - so
> that the filesystem can do the block allocation in a single shot.
> (as part of delayed allocation).

I'm doing something like that internally for XFS right now.  Due to
thinks like delayed allocations, unwritten extents and probing ahead
unmapped blocks (for the mmap write case) we can't use the common
code currently - but I plan to get it closer to it so eventually we
could make the generic code handle all that.  It's just a long term
TODO list item, not a short term one.