Re: [PATCH 6/6] autofs4: don't take spinlock when not needed in autofs4_lookup_expiring

[Ian Kent <raven@themaw.net>]

On Thu, 2014-07-10 at 09:41 +1000, NeilBrown wrote:
> If the expiring_list is empty, we can avoid a costly spinlock
> in the rcu-walk path through authfs4_d_manage.
> 
> Signed-off-by: NeilBrown <neilb@suse.de>

I know it should be straight forward to say this is OK but I always
think twice and again about areas that are subject to race pressure,
such as the expire to mount pressure of this code.

After thinking about it for a while now I don't have any reason to think
this would be a problem. Perhaps later pressure testing will reveal
something I missed.

It occurs to me that autofs4_lookup_active() might benefit from a
similar addition. Multiple calls to ->lookup() made while the dentry is
still unhashed should have enforced ordering due to the directory
i_mutex so there shouldn't be a problem adding this. But perhaps you
haven't seen delays in that function.

Acked-by: Ian Kent <raven@themaw.net>

> ---
>  fs/autofs4/root.c |    2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/fs/autofs4/root.c b/fs/autofs4/root.c
> index 1ad119407e2f..774c2dab331b 100644
> --- a/fs/autofs4/root.c
> +++ b/fs/autofs4/root.c
> @@ -219,6 +219,8 @@ static struct dentry *autofs4_lookup_expiring(struct dentry *dentry,
>  	const unsigned char *str = name->name;
>  	struct list_head *p, *head;
>  
> +	if (list_empty(&sbi->expiring_list))
> +		return NULL;
>  	spin_lock(&sbi->lookup_lock);
>  	head = &sbi->expiring_list;
>  	list_for_each(p, head) {
> 
>