Re: [PATCH 1/3] hugetlbfs: extend hugetlb_vma_lock to private VMAs

[Mike Kravetz]

On 09/25/23 23:10, riel@xxxxxxxxxxx wrote:
> From: Rik van Riel <riel@xxxxxxxxxxx>
> 
> Extend the locking scheme used to protect shared hugetlb mappings
> from truncate vs page fault races, in order to protect private
> hugetlb mappings (with resv_map) against MADV_DONTNEED.
> 
> Add a read-write semaphore to the resv_map data structure, and
> use that from the hugetlb_vma_(un)lock_* functions, in preparation
> for closing the race between MADV_DONTNEED and page faults.
> 
> Signed-off-by: Rik van Riel <riel@xxxxxxxxxxx>
> Reviewed-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
> ---
>  include/linux/hugetlb.h |  6 ++++++
>  mm/hugetlb.c            | 41 +++++++++++++++++++++++++++++++++++++----
>  2 files changed, 43 insertions(+), 4 deletions(-)

My bad during the review of patch 2!

In reply to patch 1, I suggested the changes:

> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index f906c5fa4d09..8f3d5895fffc 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -372,6 +372,11 @@ static void __hugetlb_vma_unlock_write_free(struct vm_area_struct *vma)
>  		struct hugetlb_vma_lock *vma_lock = vma->vm_private_data;
>  
>  		__hugetlb_vma_unlock_write_put(vma_lock);
> +	} else if (__vma_private_lock(vma)) {
> +		struct resv_map *resv_map = vma_resv_map(vma);
> +
> +		/* no free for anon vmas, but still need to unlock */
> +		up_write(&resv_map->rw_sema);
>  	}
>  }

However, the check for 'if (__vma_private_lock(vma))' was dropped.

> @@ -345,6 +372,11 @@ static void __hugetlb_vma_unlock_write_free(struct vm_area_struct *vma)
>  		struct hugetlb_vma_lock *vma_lock = vma->vm_private_data;
>  
>  		__hugetlb_vma_unlock_write_put(vma_lock);
> +	} else {
> +		struct resv_map *resv_map = vma_resv_map(vma);
> +
> +		/* no free for anon vmas, but still need to unlock */
> +		up_write(&resv_map->rw_sema);
>  	}
>  }

So, the map_high_truncate_2 (2M: 32) libhugetlbfs test still BUGs with:

BUG: kernel NULL pointer dereference
-- 
Mike Kravetz