Re: [PATCH v2 2/4] fs/9p: drop inodes immediately on non-.L too

["Eric Van Hensbergen" <eric.vanhensbergen@linux.dev>]

Slowly parsing through these, thanks for the fixes.
This one has a bit of a problem in that we don't have a v9fs specific
drop_inode anymore (either in legacy or .L). The release of the fid should 
be handled by v9fs_dir_release in both versions of the protocol. 

I had convinced myself we could just use generic_drop_inode because we 
decoupled clunking fids from the dentry/inode structures and just handled
them directly in v9fs_dir_release -- so really by recovering the inode 
structure every time we were just forcing churn in the inode alloc/dealloc
routines that seemed unnecessary (even in the uncached mode).

Was your concern the performance of the client side lookup of the inode
based on qid or the server side based on fid and can you give me a bit
more information on what you are seeing?  Are you not seeing fid clunks
for certain conditions (ie. transient fids, etc.)?

    -eric


March 18, 2024 at 6:22 AM, "Joakim Sindholt" <opensource@zhasha.com> wrote:
> 
> Signed-off-by: Joakim Sindholt <opensource@zhasha.com>
> 
> ---
> 
>  fs/9p/vfs_super.c | 1 +
> 
>  1 file changed, 1 insertion(+)
> 
> diff --git a/fs/9p/vfs_super.c b/fs/9p/vfs_super.c
> 
> index 941f7d0e0bfa..23cc67f29af2 100644
> 
> --- a/fs/9p/vfs_super.c
> 
> +++ b/fs/9p/vfs_super.c
> 
> @@ -310,6 +310,7 @@ static const struct super_operations v9fs_super_ops = {
> 
>  .alloc_inode = v9fs_alloc_inode,
> 
>  .free_inode = v9fs_free_inode,
> 
>  .statfs = simple_statfs,
> 
> + .drop_inode = v9fs_drop_inode,
> 
>  .evict_inode = v9fs_evict_inode,
> 
>  .show_options = v9fs_show_options,
> 
>  .umount_begin = v9fs_umount_begin,
> 
> -- 
> 
> 2.43.2
>