Re: free_swap_and_cache() doubt

[Hugh Dickins]

On Sat, 19 Jan 2002 [e-mail] wrote:
> I am reading 2.4.16, let us assume following scenario
> 
> swap_map[offset] == 2;
> and page->count == 2; (before function execution)
> vm_swap is full (nr_swap_pages*2 > total_swap_pages);
> 
> first question is the above case possible.

Yes.

> if yes, then
> after execution of this function, we would have page->count == 1, i.e.
> mapped by some process, with good page->index and what we have is, the
> associated swap entry is already freed.
> 
> Am i wrong somewhere ??

The only place you are wrong is in using the word "good" of page->index
at the end, and indeed I think your point is that it's not good.  It is
not good, it is stale, but that's okay because page->mapping has been
set to NULL, and page->index has no meaning without page->mapping.

You'll notice that __free_pages_ok() contains many BUG() hurdles
(including check on page->mapping and redundant check on PageSwapCache)
but doesn't mind if page->index is non-zero.

But thanks for making me look at this function, there
is a bug there and I'll post a patch for it right now...

Hugh