Re: [PATCH] Introduce sys_splice() system call

[Jens Axboe <axboe@suse.de>]

On Thu, Mar 30 2006, Andrew Morton wrote:
> Jens Axboe <axboe@suse.de> wrote:
> >
> > > > +	ret = write_one_page(page, 0);
> >  > 
> >  > Still want to know why this is here??
> >  > 
> >  > > +out:
> >  > > +	if (ret < 0)
> >  > > +		unlock_page(page);
> >  > 
> >  > If write_one_page()'s call to ->writepage() failed, this will cause a
> >  > double unlock.
> > 
> >  Can probably be improved - can I drop write_one_page() and just unlock
> >  the page and regular cleaning will flush it out?
> > 
> 
> Of course - commit_write() will mark the page dirty and it'll get flushed
> back in the normal manner.  We don't need that set_page_dirty() in there
> either.

Ok, wasn't sure about the fact if it always dirtied the page for me.
I'll test it and get rid of it.

> But we do need some O_SYNC/S_ISSYNC handling...

Hrmpf so that isn't handled automagically. Noted.

-- 
Jens Axboe