Re: page_launder() bug

[Marcelo Tosatti <marcelo@conectiva.com.br>]

On Mon, 7 May 2001, David S. Miller wrote:

> 
> Linus Torvalds writes:
>  > YOUR HEURISTIC IS WRONG!
> 
> Please start the conversation this way next time.
> 
>  > I call that a bug. You don't. Fine.
> 
> You made it sound like a data corrupter, a kernel crasher, and that
> any bug against a kernel with that patch indicates my patch caused it.
> There is an important distinction between "this is doing something
> silly" and "this will scramble your disk and crash the kernel".
> 
> The latter is the conclusion several people came to.
> 
> And I wanted a clarification on this, nothing more.
> 
> I wanted this clarification from you _BECAUSE_ the original posting in
> this thread saw data corruption which went away after reverting my
> patch.  But there is no possible connection between my patch and the
> crashes he saw.

Ugh, there is.

I just thought about this case:
  
We find a dead swap cache page, so dead_swap_page goes to 1.

We call swap_writepage(), but in the meantime the swapin readahead code   
got a reference on the swap map for the page.

We write the page out because "(swap_count(page) > 1)", and we may
not have __GFP_IO set in the gfp_mask. Boom.