Re: [PATCH] splice support #2

[Andrew Morton <akpm@osdl.org>]

Jens Axboe <axboe@suse.de> wrote:
>
> On Thu, Mar 30 2006, Jens Axboe wrote:
> > On Thu, Mar 30 2006, Andrew Morton wrote:
> > > Jens Axboe <axboe@suse.de> wrote:
> > > >
> > > > > > +static void *page_cache_pipe_buf_map(struct file *file,
> > > >  > > +				     struct pipe_inode_info *info,
> > > >  > > +				     struct pipe_buffer *buf)
> > > >  > > +{
> > > >  > > +	struct page *page = buf->page;
> > > >  > > +
> > > >  > > +	lock_page(page);
> > > >  > > +
> > > >  > > +	if (!PageUptodate(page)) {
> > > >  > 
> > > >  >  || page->mapping == NULL
> > > > 
> > > >  Fixed
> > > 
> > > Actually that wasn't right.  If the page was truncated then we shouldn't
> > > return -EIO to userspace.  We should return zero, as this is the first
> > > page.
> > 
> > _if_ this is the first page, then agree.
> > 
> > > Which means either returning an ERR_PTR here for -EIO or re-checking i_size
> > > in the caller.   Maybe the latter?
> > 
> > The latter sounds better, I'll add that.
> 
> Actually that's a little hard, since you don't know what state buf->page
> is in - it might be page cache, it might not be. So I opted for the
> ERR_PTR approach, just returning -ENODATA on !page->mapping.
> 

That's good, because the recheck-i_size idea was racy too.  If another
thread were to do a quick truncate-then-extend then there's a window where
we'd see a truncated page, return NULL from page_cache_pipe_buf_map() and
then by the time i_size is re-checked, i_size has grown again so we end up
deciding it must have been an IO error.

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