Re: [PATCH 8/8] mm: memory-failure: fix race window when trying to get hugetlb folio

[Naoya Horiguchi <naoya.horiguchi@linux.dev>]

On Sat, Jul 08, 2023 at 04:57:44PM +0800, Miaohe Lin wrote:
> page_folio() is fetched before calling get_hwpoison_hugetlb_folio()
> without hugetlb_lock being held. So hugetlb page could be demoted
> before get_hwpoison_hugetlb_folio() holding hugetlb_lock but after
> page_folio() is fetched. So get_hwpoison_hugetlb_folio() will hold
> unexpected extra refcnt of hugetlb folio while leaving demoted page
> un-refcnted.

Very nice, thank you for finding the issue.

> 
> Fixes: 25182f05ffed ("mm,hwpoison: fix race with hugetlb page allocation")
> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> ---
>  mm/memory-failure.c | 18 ++++++++++++++----
>  1 file changed, 14 insertions(+), 4 deletions(-)
> 
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index 76d88d27cdbe..066bf57f2d22 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1388,8 +1388,14 @@ static int __get_hwpoison_page(struct page *page, unsigned long flags)
>  	bool hugetlb = false;
>  
>  	ret = get_hwpoison_hugetlb_folio(folio, &hugetlb, false);
> -	if (hugetlb)
> -		return ret;
> +	if (hugetlb) {
> +		if (folio == page_folio(page))
> +			return ret;

Some short comment about the race against demotion here is helpful.

Anyway, the patch looks good to me.

Acked-by: Naoya Horiguchi <naoya.horiguchi@nec.com>

> +		if (ret > 0) {
> +			folio_put(folio);
> +			folio = page_folio(page);
> +		}
> +	}
>  
>  	/*
>  	 * This check prevents from calling folio_try_get() for any
> @@ -1478,8 +1484,12 @@ static int __get_unpoison_page(struct page *page)
>  	bool hugetlb = false;
>  
>  	ret = get_hwpoison_hugetlb_folio(folio, &hugetlb, true);
> -	if (hugetlb)
> -		return ret;
> +	if (hugetlb) {
> +		if (folio == page_folio(page))
> +			return ret;
> +		if (ret > 0)
> +			folio_put(folio);
> +	}
>  
>  	/*
>  	 * PageHWPoisonTakenOff pages are not only marked as PG_hwpoison,
> -- 
> 2.33.0
> 
> 
>