[PATCH] fs: pack fields `i_state` and `i_fsnotify_mask`

[PATCH] fs: pack fields `i_state` and `i_fsnotify_mask`

[Max Kellermann]

Both were 32 bit fields followed by a 32 bit hole.  This reduces size
of `struct inode` by 8 bytes.

Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
---
 include/linux/fs.h | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 5b3a100aba77..9ab028dabc1b 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -813,7 +813,11 @@ struct inode {
 
 	/* Misc */
 	struct inode_state_flags i_state;
-	/* 32-bit hole */
+
+#ifdef CONFIG_FSNOTIFY
+	__u32			i_fsnotify_mask; /* all events this inode cares about */
+#endif
+
 	struct rw_semaphore	i_rwsem;
 
 	/* Timestamp for writeback list ordering and age checks (jiffies). */
@@ -865,8 +869,6 @@ struct inode {
 
 
 #ifdef CONFIG_FSNOTIFY
-	__u32			i_fsnotify_mask; /* all events this inode cares about */
-	/* 32-bit hole reserved for expanding i_fsnotify_mask */
 	struct fsnotify_mark_connector __rcu	*i_fsnotify_marks;
 #endif
 
-- 
2.53.0

Re: [PATCH] fs: pack fields `i_state` and `i_fsnotify_mask`

[Jan Kara]

On Wed 11-11-26 16:36:30, Max Kellermann wrote:
> Both were 32 bit fields followed by a 32 bit hole.  This reduces size
> of `struct inode` by 8 bytes.
> 
> Signed-off-by: Max Kellermann <max.kellermann@ionos.com>

When moving i_fsnotify_mask please move it along with i_fsnotify_marks. It
should better be in the same cacheline because these two are accessed on
some performance sensitive paths.

								Honza

> ---
>  include/linux/fs.h | 8 +++++---
>  1 file changed, 5 insertions(+), 3 deletions(-)
> 
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 5b3a100aba77..9ab028dabc1b 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -813,7 +813,11 @@ struct inode {
>  
>  	/* Misc */
>  	struct inode_state_flags i_state;
> -	/* 32-bit hole */
> +
> +#ifdef CONFIG_FSNOTIFY
> +	__u32			i_fsnotify_mask; /* all events this inode cares about */
> +#endif
> +
>  	struct rw_semaphore	i_rwsem;
>  
>  	/* Timestamp for writeback list ordering and age checks (jiffies). */
> @@ -865,8 +869,6 @@ struct inode {
>  
>  
>  #ifdef CONFIG_FSNOTIFY
> -	__u32			i_fsnotify_mask; /* all events this inode cares about */
> -	/* 32-bit hole reserved for expanding i_fsnotify_mask */
>  	struct fsnotify_mark_connector __rcu	*i_fsnotify_marks;
>  #endif
>  
> -- 
> 2.53.0
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH] fs: pack fields `i_state` and `i_fsnotify_mask`

[Max Kellermann]

On Thu, Nov 12, 2026 at 7:38 PM Jan Kara <jack@suse.cz> wrote:
> When moving i_fsnotify_mask please move it along with i_fsnotify_marks. It
> should better be in the same cacheline because these two are accessed on
> some performance sensitive paths.

Tried that, but:

        struct inode_state_flags   i_state;              /*   120     4 */
        __u32                      i_fsnotify_mask;      /*   124     4 */
        /* --- cacheline 2 boundary (128 bytes) --- */
        struct fsnotify_mark_connector * i_fsnotify_marks; /*   128     8 */

The cacheline happened to end after "i_state" (and now
"i_fsnotify_mask"). I could move "i_state" down instead, or move other
fields around - but you're a better expert here, please advise what
you believe is best.

-- 
Max Kellermann
Principal Architect
Hosting Technology

cm4all | Im Mediapark 6a | 50670 Köln | Germany
General information about the company can be found here:
https://www.cm4all.com/impressum
A member of the IONOS Group

Re: [PATCH] fs: pack fields `i_state` and `i_fsnotify_mask`

[Jan Kara]

On Fri 13-11-26 09:07:44, Max Kellermann wrote:
> On Thu, Nov 12, 2026 at 7:38 PM Jan Kara <jack@suse.cz> wrote:
> > When moving i_fsnotify_mask please move it along with i_fsnotify_marks. It
> > should better be in the same cacheline because these two are accessed on
> > some performance sensitive paths.
> 
> Tried that, but:
> 
>         struct inode_state_flags   i_state;              /*   120     4 */
>         __u32                      i_fsnotify_mask;      /*   124     4 */
>         /* --- cacheline 2 boundary (128 bytes) --- */
>         struct fsnotify_mark_connector * i_fsnotify_marks; /*   128     8 */
> 
> The cacheline happened to end after "i_state" (and now
> "i_fsnotify_mask"). I could move "i_state" down instead, or move other
> fields around - but you're a better expert here, please advise what
> you believe is best.

I have checked and none of the big filesystems (xfs, btrfs, ext4, tmpfs)
place struct inode to the beginning of their inode allocations. So where
exactly the cacheline boundary ends up being is kind of random anyway (plus
the allocations of the inode structures aren't cacheline aligned anyway).
So I'd just keep these two fields together to have a good chance of them
not being in different cachelines and leave the rest on $deity. ;)

								Honza

-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR