Race between try_to_swap_out() and swapin_readahead()

[Marcelo Tosatti]

Ok, what prevents this from happening: 

CPU0				CPU1

try_to_swap_out()
...

entry = get_swap_page();
if (!entry.val)
	goto out_unlock_restore;

				swapin_readahead()
				finds valid swap entry just allocated by try_to_swap_out()
				lookup_swap_cache() fails...
				new_page = alloc_page();
				lock_page(new_page);
				add_to_swap_cache(new_page, entry);
				rw_swap_page(READ, new_page);
			 	return new_page;

/* Add it to the swap cache and mark it dirty */
add_to_swap_cache(page, entry);
set_page_dirty(page);

----

We'll end up with two pages with the same index in the hash tables, and
one of them has crap read from swap.

I know the window is really really small, but anyway... 

Comments?