[1/8] mm: memory-failure: fix potential permanently locked hpage

Message ID 20230708085744.3599311-2-linmiaohe@huawei.com
State New
Headers
Series A few fixup and cleanup patches for memory-failure |

Commit Message

Miaohe Lin July 8, 2023, 8:57 a.m. UTC
  If hpage isn't Hugetlb page, MF_DELAYED is returned without unlock hpage
leading to hpage permanently locked. But this shouldn't trigger in the
real world because this PageHuge() check is just for potential problems.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
---
 mm/memory-failure.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
  

Comments

Naoya Horiguchi July 10, 2023, 7:56 a.m. UTC | #1
On Sat, Jul 08, 2023 at 04:57:37PM +0800, Miaohe Lin wrote:
> If hpage isn't Hugetlb page, MF_DELAYED is returned without unlock hpage
> leading to hpage permanently locked. But this shouldn't trigger in the
> real world because this PageHuge() check is just for potential problems.

Right, so this if-block is dead code, how about simply removing it?

Thanks,
Naoya Horiguchi

> 
> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> ---
>  mm/memory-failure.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/memory-failure.c b/mm/memory-failure.c
> index a6221a4bc5ea..d21ee27ad412 100644
> --- a/mm/memory-failure.c
> +++ b/mm/memory-failure.c
> @@ -1187,8 +1187,10 @@ static int me_huge_page(struct page_state *ps, struct page *p)
>  	struct address_space *mapping;
>  	bool extra_pins = false;
>  
> -	if (!PageHuge(hpage))
> +	if (!PageHuge(hpage)) {
> +		unlock_page(hpage);
>  		return MF_DELAYED;
> +	}
>  
>  	mapping = page_mapping(hpage);
>  	if (mapping) {
> -- 
> 2.33.0
> 
> 
>
  
Miaohe Lin July 10, 2023, 8:26 a.m. UTC | #2
On 2023/7/10 15:56, Naoya Horiguchi wrote:
> On Sat, Jul 08, 2023 at 04:57:37PM +0800, Miaohe Lin wrote:
>> If hpage isn't Hugetlb page, MF_DELAYED is returned without unlock hpage
>> leading to hpage permanently locked. But this shouldn't trigger in the
>> real world because this PageHuge() check is just for potential problems.
> 
> Right, so this if-block is dead code, how about simply removing it?

From the current code of view, I think this if-block is actually dead code and won't catch
anything too. So removing it makes sense to me. Will do it in v2.

Thanks Naoya.
  

Patch

diff --git a/mm/memory-failure.c b/mm/memory-failure.c
index a6221a4bc5ea..d21ee27ad412 100644
--- a/mm/memory-failure.c
+++ b/mm/memory-failure.c
@@ -1187,8 +1187,10 @@  static int me_huge_page(struct page_state *ps, struct page *p)
 	struct address_space *mapping;
 	bool extra_pins = false;
 
-	if (!PageHuge(hpage))
+	if (!PageHuge(hpage)) {
+		unlock_page(hpage);
 		return MF_DELAYED;
+	}
 
 	mapping = page_mapping(hpage);
 	if (mapping) {