2.6.21-rc5-git12 new aops patchset

2.6.21-rc5-git12 new aops patchset

[Nick Piggin]

http://www.kernel.org/pub/linux/kernel/people/npiggin/patches/new-aops/

Files are 2.6.21-rc5-git12*

This contains an update to recent git kernels from Mark, ext3 merge,
XFS fix from David Chinner, some FUSE fixes (now lightly tested and
working).

Documents perform_write a little better, and implements
simple_perform_write, and fuse_perform_write. However, note that
as per discussion, we may opt to take perform_write out of this
patchset, and instead work towards making generic_file_aio_write
easier to implement. I'm open to discussion, though.

Re: 2.6.21-rc5-git12 new aops patchset

[Badari Pulavarty]

On Thu, 2007-04-05 at 11:46 +0200, Nick Piggin wrote:
> http://www.kernel.org/pub/linux/kernel/people/npiggin/patches/new-aops/
> 
> Files are 2.6.21-rc5-git12*


ext3_ordered_write_end() did

+		ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
+		if (ret >= 0)
+			copied = ret;
	}
 	ret2 = ext3_journal_stop(handle);
 	if (!ret)
 		ret = ret2;
-	return ret;
+	return ret ? ret : copied;

If we do this, we are going to ignore ext3_journal_stop() errors
almost all the time (since "ret" will be mostly positive).

Instead can we do this ?

+		copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
+		if (copied < 0)
+			ret = copied;

Thanks,
Badari

Re: 2.6.21-rc5-git12 new aops patchset

[Nick Piggin]

On Thu, Apr 05, 2007 at 09:22:35AM -0700, Badari Pulavarty wrote:
> On Thu, 2007-04-05 at 11:46 +0200, Nick Piggin wrote:
> > http://www.kernel.org/pub/linux/kernel/people/npiggin/patches/new-aops/
> > 
> > Files are 2.6.21-rc5-git12*
> 
> 
> ext3_ordered_write_end() did
> 
> +		ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
> +		if (ret >= 0)
> +			copied = ret;
> 	}
>  	ret2 = ext3_journal_stop(handle);
>  	if (!ret)
>  		ret = ret2;
> -	return ret;
> +	return ret ? ret : copied;
> 
> If we do this, we are going to ignore ext3_journal_stop() errors
> almost all the time (since "ret" will be mostly positive).
> 
> Instead can we do this ?
> 
> +		copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
> +		if (copied < 0)
> +			ret = copied;

Definitely, thank you.

Nick

Re: 2.6.21-rc5-git12 new aops patchset

[Mark Fasheh]

On Thu, Apr 05, 2007 at 11:46:05AM +0200, Nick Piggin wrote:
> However, note that as per discussion, we may opt to take perform_write out
> of this patchset, and instead work towards making generic_file_aio_write
> easier to implement. I'm open to discussion, though.

I'd be fine with that - if the file systems can get at all the checks and
iov_iter stuff it lets them fine tune their write logic. Ocfs2 for instance
has very different locking for o_direct / buffered and so goes to some bit
of pain to avoid the fallback code in __generic_file_aio_write_nolock().
It'd be much simpler to just call some helpers for buffered writes and
change the code to never do that.
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh@oracle.com