Re: [PATCH v5 12/12] nvmet-fc: use RCU list iterator for assoc_list

[Hannes Reinecke <hare@suse.de>]

On 1/31/24 16:51, Daniel Wagner wrote:
> The assoc_list is a RCU protected list, thus use the RCU flavor of list
> functions.
> 
> Let's use this opportunity and refactor this code and move the lookup
> into a helper and give it a descriptive name.
> 
> Signed-off-by: Daniel Wagner <dwagner@suse.de>
> ---
>   drivers/nvme/target/fc.c | 34 ++++++++++++++++++++++------------
>   1 file changed, 22 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/nvme/target/fc.c b/drivers/nvme/target/fc.c
> index 671d096745a5..fd229f310c93 100644
> --- a/drivers/nvme/target/fc.c
> +++ b/drivers/nvme/target/fc.c
> @@ -1114,14 +1114,27 @@ nvmet_fc_schedule_delete_assoc(struct nvmet_fc_tgt_assoc *assoc)
>   	queue_work(nvmet_wq, &assoc->del_work);
>   }
>   
> +static bool
> +nvmet_fc_assoc_exits(struct nvmet_fc_tgtport *tgtport, u64 association_id)
> +{
> +	struct nvmet_fc_tgt_assoc *a;
> +
> +	list_for_each_entry_rcu(a, &tgtport->assoc_list, a_list) {
> +		if (association_id == a->association_id)
> +			return true;
> +	}
> +
> +	return false;
> +}
> +
>   static struct nvmet_fc_tgt_assoc *
>   nvmet_fc_alloc_target_assoc(struct nvmet_fc_tgtport *tgtport, void *hosthandle)
>   {
> -	struct nvmet_fc_tgt_assoc *assoc, *tmpassoc;
> +	struct nvmet_fc_tgt_assoc *assoc;
>   	unsigned long flags;
> +	bool done;
>   	u64 ran;
>   	int idx;
> -	bool needrandom = true;
>   
>   	if (!tgtport->pe)
>   		return NULL;
> @@ -1145,24 +1158,21 @@ nvmet_fc_alloc_target_assoc(struct nvmet_fc_tgtport *tgtport, void *hosthandle)
>   	INIT_WORK(&assoc->del_work, nvmet_fc_delete_assoc_work);
>   	atomic_set(&assoc->terminating, 0);
>   
> -	while (needrandom) {
> +	done = false;
> +	do {
>   		get_random_bytes(&ran, sizeof(ran) - BYTES_FOR_QID);
>   		ran = ran << BYTES_FOR_QID_SHIFT;
>   
>   		spin_lock_irqsave(&tgtport->lock, flags);
> -		needrandom = false;
> -		list_for_each_entry(tmpassoc, &tgtport->assoc_list, a_list) {
> -			if (ran == tmpassoc->association_id) {
> -				needrandom = true;
> -				break;
> -			}
> -		}
> -		if (!needrandom) {
> +		rcu_read_lock();
> +		if (!nvmet_fc_assoc_exits(tgtport, ran)) {
>   			assoc->association_id = ran;
>   			list_add_tail_rcu(&assoc->a_list, &tgtport->assoc_list);
> +			done = true;
>   		}
> +		rcu_read_unlock();
>   		spin_unlock_irqrestore(&tgtport->lock, flags);

Odd. You already take the tgtport lock, so there really is no point in 
using rcu_read_lock() here.
Or does the tgtport lock not protect the assoc_list?

Cheers,

Hannes
-- 
Dr. Hannes Reinecke                Kernel Storage Architect
hare@suse.de                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), GF: Ivo Totev, Andrew McDonald,
Werner Knoblich