Re: [PATCH] mm: add replace_page_cache_page() function

[Hugh Dickins]

On Fri, 17 Dec 2010, Minchan Kim wrote:
> On Thu, Dec 16, 2010 at 12:59:58PM +0100, Miklos Szeredi wrote:
> > On Thu, 16 Dec 2010, Minchan Kim wrote:
> > > 
> > > Why do you release reference of old?
> > 
> > That's the page cache reference we release.  Just like we acquire the
> > page cache reference for "new" above.
> 
> I mean current page cache handling semantic and page reference counting semantic
> is separeated. For example, remove_from_page_cache doesn't drop the reference of page.
> That's because we need more works after drop the page from page cache.
> Look at shmem_writepage, truncate_complete_page.

I disagree with you there: I like the way Miklos made it symmetric,
I like the way delete_from_swap_cache drops the swap cache reference,
I dislike the way remove_from_page_cache does not - I did once try to
change that, but did a bad job, messed up reiserfs or reiser4 I forget
which, retreated in shame.

In both the examples you give, shmem_writepage and truncate_complete_page,
the caller has to be holding their own reference, in part because they
locked the page, and will need to unlock it before releasing their ref.
I think that would be true of any replace_page_cache_page caller.

> 
> You makes the general API and caller might need works before the old page 
> is free. So how about this?
> 
> err = replace_page_cache_page(oldpage, newpage, GFP_KERNEL);
> if (err) {
>         ...
> }
> 
> page_cache_release(oldpage); /* drop ref of page cache */
> 
> 
> > 
> > I suspect it's historic that page_cache_release() doesn't drop the
> > page cache ref.
> 
> Sorry I can't understand your words.

Me neither: I believe Miklos meant __remove_from_page_cache() rather
than page_cache_release() in that instance.

Hugh
--
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/