Re: page_launder() bug

[Linus Torvalds]

On Mon, 7 May 2001, David S. Miller wrote:
> 
> Marcelo Tosatti writes:
>  > I was wrong. The patch is indeed buggy because of the __GFP_IO thing.
> 
> What about the __GFP_IO thing?
> 
> Specifically, what protects the __GFP_IO thing from happening without
> my patch?

This:

                        /* First time through? Move it to the back of the list */
                        if (!launder_loop) {
                                list_del(page_lru);
                                list_add(page_lru, &inactive_dirty_list);
                                UnlockPage(page);
                                continue;
                        }

        if (can_get_io_locks && !launder_loop && free_shortage()) {
                launder_loop = 1;
		...

Notice? If "can_get_io_locks" is not true, we will never call
"writepage()" on the page, because if "can_get_io_locks" is false,
"launder_loop" will always be false too..

And "can_get_io_locks" depends on __GFP_IO.

		Linus