Move shmem_recalc_inode() out of the loop in memfd_luo_retrieve_folios()
to improve performance when restoring large memfds.

Currently, shmem_recalc_inode() is called for each folio during restore,
which is O(n) expensive operations. This patch collects the number of
successfully added folios and calls shmem_recalc_inode() once after the
loop completes, reducing complexity to O(1).

Additionally, fix the error path to also call shmem_recalc_inode() for
the folios that were successfully added before the error occurred.

Signed-off-by: Chenghao Duan <duanchenghao@kylinos.cn>
---
 mm/memfd_luo.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/mm/memfd_luo.c b/mm/memfd_luo.c
index b8edb9f981d7..5ddd3657d8be 100644
--- a/mm/memfd_luo.c
+++ b/mm/memfd_luo.c
@@ -397,6 +397,7 @@ static int memfd_luo_retrieve_folios(struct file *file,
 	struct folio *folio;
 	int err = -EIO;
 	long i;
+	u64 nr_added = 0;
 
 	for (i = 0; i < nr_folios; i++) {
 		const struct memfd_luo_folio_ser *pfolio = &folios_ser[i];
@@ -448,12 +449,15 @@ static int memfd_luo_retrieve_folios(struct file *file,
 			goto unlock_folio;
 		}
 
-		shmem_recalc_inode(inode, 1, 0);
+		nr_added++;
 		folio_add_lru(folio);
 		folio_unlock(folio);
 		folio_put(folio);
 	}
 
+	if (nr_added)
+		shmem_recalc_inode(inode, nr_added, 0);
+
 	return 0;
 
 unlock_folio:
@@ -472,6 +476,9 @@ static int memfd_luo_retrieve_folios(struct file *file,
 			folio_put(folio);
 	}
 
+	if (nr_added)
+		shmem_recalc_inode(inode, nr_added, 0);
+
 	return err;
 }
 
-- 
2.25.1

On Thu, Mar 19 2026, Chenghao Duan wrote:

> Move shmem_recalc_inode() out of the loop in memfd_luo_retrieve_folios()
> to improve performance when restoring large memfds.
>
> Currently, shmem_recalc_inode() is called for each folio during restore,
> which is O(n) expensive operations. This patch collects the number of
> successfully added folios and calls shmem_recalc_inode() once after the
> loop completes, reducing complexity to O(1).
>
> Additionally, fix the error path to also call shmem_recalc_inode() for
> the folios that were successfully added before the error occurred.
>
> Signed-off-by: Chenghao Duan <duanchenghao@kylinos.cn>
> ---
>  mm/memfd_luo.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/mm/memfd_luo.c b/mm/memfd_luo.c
> index b8edb9f981d7..5ddd3657d8be 100644
> --- a/mm/memfd_luo.c
> +++ b/mm/memfd_luo.c
> @@ -397,6 +397,7 @@ static int memfd_luo_retrieve_folios(struct file *file,
>  	struct folio *folio;
>  	int err = -EIO;
>  	long i;
> +	u64 nr_added = 0;
>  
>  	for (i = 0; i < nr_folios; i++) {
>  		const struct memfd_luo_folio_ser *pfolio = &folios_ser[i];
> @@ -448,12 +449,15 @@ static int memfd_luo_retrieve_folios(struct file *file,
>  			goto unlock_folio;
>  		}
>  
> -		shmem_recalc_inode(inode, 1, 0);
> +		nr_added++;

https://sashiko.dev/#/patchset/20260319012845.29570-1-duanchenghao%40kylinos.cn

AI review picked up a real bug here:

    Since memfd files can use large folios, should nr_added track the
    number of pages instead of the number of folios?

    shmem_recalc_inode() expects the number of pages. Passing the number
    of folios might under-account blocks and bypass tmpfs limits or
    quotas.

    Also, shmem_inode_acct_blocks() earlier in the loop is hardcoded to
    1, which might have the same issue.

If THP is being used, we should account for nr_pages instead of
nr_folios. Can you please also add a fix for this with your series? Just
so we fix the bugs the code already has before refactoring it.

>  		folio_add_lru(folio);
>  		folio_unlock(folio);
>  		folio_put(folio);
>  	}
>  
> +	if (nr_added)
> +		shmem_recalc_inode(inode, nr_added, 0);

Nit: it is very very likely that nr_added > 0. And shmem_recalc_inode()
can deal with 0 nr_added. So please drop this if check and call it
directly.

Other than this, the patch LGTM. Thanks for working on this!

> +
>  	return 0;
>  
>  unlock_folio:
> @@ -472,6 +476,9 @@ static int memfd_luo_retrieve_folios(struct file *file,
>  			folio_put(folio);
>  	}
>  
> +	if (nr_added)
> +		shmem_recalc_inode(inode, nr_added, 0);
> +
>  	return err;
>  }

-- 
Regards,
Pratyush Yadav

On Wed, Mar 18, 2026 at 9:29 PM Chenghao Duan <duanchenghao@kylinos.cn> wrote:
>
> Move shmem_recalc_inode() out of the loop in memfd_luo_retrieve_folios()
> to improve performance when restoring large memfds.
>
> Currently, shmem_recalc_inode() is called for each folio during restore,
> which is O(n) expensive operations. This patch collects the number of
> successfully added folios and calls shmem_recalc_inode() once after the
> loop completes, reducing complexity to O(1).
>
> Additionally, fix the error path to also call shmem_recalc_inode() for
> the folios that were successfully added before the error occurred.
>
> Signed-off-by: Chenghao Duan <duanchenghao@kylinos.cn>
> ---
>  mm/memfd_luo.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/mm/memfd_luo.c b/mm/memfd_luo.c
> index b8edb9f981d7..5ddd3657d8be 100644
> --- a/mm/memfd_luo.c
> +++ b/mm/memfd_luo.c
> @@ -397,6 +397,7 @@ static int memfd_luo_retrieve_folios(struct file *file,
>         struct folio *folio;
>         int err = -EIO;
>         long i;
> +       u64 nr_added = 0;

nit: I perfer RCT for local variables order, but it is not followed in
this file anyway.

>
>         for (i = 0; i < nr_folios; i++) {
>                 const struct memfd_luo_folio_ser *pfolio = &folios_ser[i];
> @@ -448,12 +449,15 @@ static int memfd_luo_retrieve_folios(struct file *file,
>                         goto unlock_folio;
>                 }
>
> -               shmem_recalc_inode(inode, 1, 0);
> +               nr_added++;
>                 folio_add_lru(folio);
>                 folio_unlock(folio);
>                 folio_put(folio);
>         }
>
> +       if (nr_added)
> +               shmem_recalc_inode(inode, nr_added, 0);
> +
>         return 0;
>
>  unlock_folio:
> @@ -472,6 +476,9 @@ static int memfd_luo_retrieve_folios(struct file *file,
>                         folio_put(folio);
>         }
>
> +       if (nr_added)
> +               shmem_recalc_inode(inode, nr_added, 0);
> +

Reviewed-by: Pasha Tatashin <pasha.tatashin@soleen.com>

On Thu, Mar 19 2026, Pasha Tatashin wrote:

> On Wed, Mar 18, 2026 at 9:29 PM Chenghao Duan <duanchenghao@kylinos.cn> wrote:
>>
>> Move shmem_recalc_inode() out of the loop in memfd_luo_retrieve_folios()
>> to improve performance when restoring large memfds.
>>
>> Currently, shmem_recalc_inode() is called for each folio during restore,
>> which is O(n) expensive operations. This patch collects the number of
>> successfully added folios and calls shmem_recalc_inode() once after the
>> loop completes, reducing complexity to O(1).
>>
>> Additionally, fix the error path to also call shmem_recalc_inode() for
>> the folios that were successfully added before the error occurred.
>>
>> Signed-off-by: Chenghao Duan <duanchenghao@kylinos.cn>
>> ---
>>  mm/memfd_luo.c | 9 ++++++++-
>>  1 file changed, 8 insertions(+), 1 deletion(-)
>>
>> diff --git a/mm/memfd_luo.c b/mm/memfd_luo.c
>> index b8edb9f981d7..5ddd3657d8be 100644
>> --- a/mm/memfd_luo.c
>> +++ b/mm/memfd_luo.c
>> @@ -397,6 +397,7 @@ static int memfd_luo_retrieve_folios(struct file *file,
>>         struct folio *folio;
>>         int err = -EIO;
>>         long i;
>> +       u64 nr_added = 0;
>
> nit: I perfer RCT for local variables order, but it is not followed in
> this file anyway.

It is though, for the most part. I also prefer this so as much as I
could I followed it, but sometimes if you want to assign variables at
declaration, it isn't always possible.

Anyway, RCT would be nice to have indeed.

[...]

-- 
Regards,
Pratyush Yadav