Re: [PATCH] mm/hugetlb.c: fix reservation race when freeing surplus pages

[Michal Hocko]

On Mon 09-01-17 11:56:07, Mike Kravetz wrote:
> The routine return_unused_surplus_pages decrements the global
> reservation count, and frees any unused surplus pages that were
> backing the reservation.  Commit 7848a4bf51b3 ("mm/hugetlb.c:
> add cond_resched_lock() in return_unused_surplus_pages()") added
> a call to cond_resched_lock in the loop freeing the pages.  As
> a result, the hugetlb_lock could be dropped, and someone else
> could use the pages that will be freed in subsequent iterations
> of the loop.  This could result in inconsistent global hugetlb
> page state, application api failures (such as mmap) failures or
> application crashes.
> 
> When dropping the lock in return_unused_surplus_pages, make sure
> that the global reservation count (resv_huge_pages) remains
> sufficiently large to prevent someone else from claiming pages
> about to be freed.
> 
> Fixes: 7848a4bf51b3 ("mm/hugetlb.c: add cond_resched_lock() in return_unused_surplus_pages()")
> Reported-and-analyzed-by: Paul Cassella <cassella@xxxxxxxx>
> Suggested-by: Michal Hocko <mhocko@xxxxxxxxxx>
> Signed-off-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>

Looks good to me. I think we want also
Cc: stable # 3.15+

Paul, your Tested-by would be more than appreciated.

Thanks Mike!

> ---
>  mm/hugetlb.c | 37 ++++++++++++++++++++++++++++---------
>  1 file changed, 28 insertions(+), 9 deletions(-)
> 
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 418bf01..a1760fa 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -1773,23 +1773,32 @@ static int gather_surplus_pages(struct hstate *h, int delta)
>  }
>  
>  /*
> - * When releasing a hugetlb pool reservation, any surplus pages that were
> - * allocated to satisfy the reservation must be explicitly freed if they were
> - * never used.
> - * Called with hugetlb_lock held.
> + * This routine has two main purposes:
> + * 1) Decrement the reservation count (resv_huge_pages) by the value passed
> + *    in unused_resv_pages.  This corresponds to the prior adjustments made
> + *    to the associated reservation map.
> + * 2) Free any unused surplus pages that may have been allocated to satisfy
> + *    the reservation.  As many as unused_resv_pages may be freed.
> + *
> + * Called with hugetlb_lock held.  However, the lock could be dropped (and
> + * reacquired) during calls to cond_resched_lock.  Whenever dropping the lock,
> + * we must make sure nobody else can claim pages we are in the process of
> + * freeing.  Do this by ensuring resv_huge_page always is greater than the
> + * number of huge pages we plan to free when dropping the lock.
>   */
>  static void return_unused_surplus_pages(struct hstate *h,
>  					unsigned long unused_resv_pages)
>  {
>  	unsigned long nr_pages;
>  
> -	/* Uncommit the reservation */
> -	h->resv_huge_pages -= unused_resv_pages;
> -
>  	/* Cannot return gigantic pages currently */
>  	if (hstate_is_gigantic(h))
> -		return;
> +		goto out;
>  
> +	/*
> +	 * Part (or even all) of the reservation could have been backed
> +	 * by pre-allocated pages. Only free surplus pages.
> +	 */
>  	nr_pages = min(unused_resv_pages, h->surplus_huge_pages);
>  
>  	/*
> @@ -1799,12 +1808,22 @@ static void return_unused_surplus_pages(struct hstate *h,
>  	 * when the nodes with surplus pages have no free pages.
>  	 * free_pool_huge_page() will balance the the freed pages across the
>  	 * on-line nodes with memory and will handle the hstate accounting.
> +	 *
> +	 * Note that we decrement resv_huge_pages as we free the pages.  If
> +	 * we drop the lock, resv_huge_pages will still be sufficiently large
> +	 * to cover subsequent pages we may free.
>  	 */
>  	while (nr_pages--) {
> +		h->resv_huge_pages--;
> +		unused_resv_pages--;
>  		if (!free_pool_huge_page(h, &node_states[N_MEMORY], 1))
> -			break;
> +			goto out;
>  		cond_resched_lock(&hugetlb_lock);
>  	}
> +
> +out:
> +	/* Fully uncommit the reservation */
> +	h->resv_huge_pages -= unused_resv_pages;
>  }
>  
>  
> -- 
> 2.7.4

-- 
Michal Hocko
SUSE Labs