Re: [PATCH 1/2] mm, page_alloc: Keep pcp count and list contents in sync if struct page is corrupted

[Mel Gorman]

On Fri, Dec 02, 2016 at 11:04:11AM +0100, Michal Hocko wrote:
> On Fri 02-12-16 17:30:07, Hillf Danton wrote:
> [...]
> > > >> @@ -2217,13 +2217,14 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order,
> > > >>  		else
> > > >>  			list_add_tail(&page->lru, list);
> > > >>  		list = &page->lru;
> > > >> +		alloced++;
> > > >>  		if (is_migrate_cma(get_pcppage_migratetype(page)))
> > > >>  			__mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
> > > >>  					      -(1 << order));
> > > >>  	}
> > > >>  	__mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
> > > >
> > > > Now i is a pure index, yes?
> > > 
> > > No, even if a page fails the check_pcp_refill() check and is not
> > > "allocated", it is also no longer a free page, so it's correct to
> > > subtract it from NR_FREE_PAGES.
> > > 
> > Yes, we can allocate free page   next time.
> 
> No we cannot. The page is gone from the free list. We have effectively
> leaked it.

And deliberately so. It's in an unknown state, possibly due to memory
corruption or a use-after free bug. The machine can continue limping on
with warnings in the kernel log but the VM stops going near the page
itself as much as possible.

-- 
Mel Gorman
SUSE Labs