Re: [PATCH 1/3] Vectorize aio_read/aio_write methods

From: Christoph Hellwig <hch@lst.de>
To: Andrew Morton <akpm@osdl.org>
Cc: Christoph Hellwig <hch@lst.de>,
	pbadari@us.ibm.com, linux-kernel@vger.kernel.org, bcrl@kvack.org,
	cel@citi.umich.edu
Subject: Re: [PATCH 1/3] Vectorize aio_read/aio_write methods
Date: Tue, 9 May 2006 21:20:51 +0200	[thread overview]
Message-ID: <20060509192051.GA19378@lst.de> (raw)
In-Reply-To: <20060509121305.0840e770.akpm@osdl.org>

On Tue, May 09, 2006 at 12:13:05PM -0700, Andrew Morton wrote:
> > there's another patch ontop which I didn't bother to redo until this is
> > accepted which kills a lot more code.  After that filesystems only have
> > to implement one method each for all kinds of read/write calls.  Which
> > allows to both make the mm/filemap.c far less complex and actually
> > understandable aswell as for any filesystem that uses more complex
> > read/write variants than direct filemap.c calls.  In addition to these
> > simplification we also get a feature (async vectored I/O) for free.
> 
> Fair enough, thanks.  Simplifying filemap.c would be a win.
> 
> I'll crunch on these three patches in the normal fashion.  It'll be good if
> we can get the followup patch done within the next week or two so we can
> get it all tested at the same time.  Although from your description it
> doesn't sound like it'll be completely trivial...

That patch is lots of tirival and boring work.  If anyone wants to beat
me to it:

 - in any filesystem that implements the generic_file_aio_{read,write}
   directly remove these apply this patch to the file_operations
   vectors:

-	.read		= generic_file_read,
-	.write		= generic_file_write,
+	.read		= do_sync_read,
+	.write		= do_sync_write,

   Note that this does _not_ cause additional indirection for normal
   sys_read/sys_write calls because they call .aio_read/.aio_write
   directly.  It's only needed because we have various places in the
   tree that like to call .read/.write directly

 - in the filesystems that implement more or less trivial wrappers
   around  generic_file_read/generic_file_write to the
   aio_read/aio_write prototypes so they can set .read/.write as above

 - after that generic_file_read/generic_file_write/generic_file_read/
   generic_file_write_nolock should have no callers left and the code
   for read/write in mm/filemap.c can be collapsed into very few functions.
   What's left should be something like:

     - generic_file_aio_read
       (__generic_file_aio_read and generic_file_aio_read merged into one)
     - __generic_file_aio_write
       (basically the current __generic_file_aio_write_nolock)
     - generic_file_aio_write_nolock
     - generic_file_aio_write
       (small wrappers around __generic_file_aio_write)