Re: [PATCH 03/10] mm: Add support for a filesystem to control swapfiles

[Mel Gorman]

On Fri, Sep 09, 2011 at 09:00:08AM -0400, Christoph Hellwig wrote:
> On Fri, Sep 09, 2011 at 12:00:47PM +0100, Mel Gorman wrote:
> > Currently swapfiles are managed entirely by the core VM by using
> > ->bmap to allocate space and write to the blocks directly. This
> > patch adds address_space_operations methods that allow a filesystem
> > to optionally control the swapfile.
> > 
> >   int swap_activate(struct file *);
> >   int swap_deactivate(struct file *);
> >   int swap_writepage(struct file *, struct page *, struct writeback_control *);
> >   int swap_readpage(struct file *, struct page *);
> 
> Just as the last two dozen times this came up:
> 
> NAK
> 
> The right fix is to add a filesystem method to support direct-I/O on
> arbitrary kernel pages, instead of letting the wap abstraction leak into
> the filesystem.

Ok.

I confess I haven't investigated this direction at
all yet.  Is it correct that your previous objection was
http://linux.derkeiler.com/Mailing-Lists/Kernel/2009-10/msg00455.html
and the direct-IO patchset you were thinking of was
http://copilotco.com/mail-archives/linux-kernel.2009/msg87176.html ?

If so, are you suggesting that instead of swap_readpage and
swap_writepage I look into what is required for swap to use ->readpage
method and ->direct_IO aops?

-- 
Mel Gorman
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/