Re: [PATCH RESEND mptcp-next v5 1/8] mptcp: add struct mptcp_sched_ops

[Geliang Tang <geliang.tang@suse.com>]

On Thu, Mar 24, 2022 at 03:27:53PM +0100, Florian Westphal wrote:
> Geliang Tang <geliang.tang@suse.com> wrote:
> > This patch added struct mptcp_sched_ops. And define the scheduler
> > register, unregister and find functions.
> 
> ... but why are they pernet?  Makes no sense to me, so an
> explanation would help.  Or, remove the pernet ops.
> 
> All callers pass &init_net, so I don't think there is any reason
> for pernet data structures here.

v4 used global list instead of pernet list:

https://patchwork.kernel.org/project/mptcp/patch/02c1d2d2eee134713737fe5c5f73127e8c741589.1647942374.git.geliang.tang@suse.com/

Should I go back to v4?

> 
> > +struct mptcp_sched_ops *mptcp_sched_find(const struct net *net,
> > +					 const char *name)
> > +{
> > +	struct sched_pernet *pernet = sched_get_pernet(net);
> > +	struct mptcp_sched_ops *sched, *ret = NULL;
> > +
> > +	spin_lock(&pernet->lock);
> > +	list_for_each_entry_rcu(sched, &pernet->sched_list, list) {
> > +		if (!strcmp(sched->name, name)) {
> > +			ret = sched;
> > +			break;
> > +		}
> > +	}
> > +	spin_unlock(&pernet->lock);
> > +
> 
> Locking isn't needed here if caller holds rcu read lock.
> 
> > +int mptcp_register_scheduler(const struct net *net,
> > +			     struct mptcp_sched_ops *sched)
> > +{
> > +	struct sched_pernet *pernet = sched_get_pernet(net);
> > +
> > +	if (!sched->get_subflow)
> > +		return -EINVAL;
> > +
> > +	if (mptcp_sched_find(net, sched->name))
> > +		return -EEXIST;
> > +
> > +	spin_lock(&pernet->lock);
> > +	list_add_tail_rcu(&sched->list, &pernet->sched_list);
> > +	spin_unlock(&pernet->lock);
> 
> I would use global list/lock, not pernet registration.
> I don't see a reason why that is needed, its just duplicate info.
> 
> > +void mptcp_unregister_scheduler(const struct net *net,
> > +				struct mptcp_sched_ops *sched)
> > +{
> > +	struct sched_pernet *pernet = sched_get_pernet(net);
> > +
> > +	spin_lock(&pernet->lock);
> > +	list_del_rcu(&sched->list);
> > +	spin_unlock(&pernet->lock);
> > +
> > +	synchronize_rcu();
> 
> Why synchronize_rcu()?  Comment please why that is needed, no freeing
> happens?
> 
> > +	struct sched_pernet *pernet = sched_get_pernet(net);
> > +	struct mptcp_sched_ops *sched;
> > +
> > +	spin_lock(&pernet->lock);
> > +	list_for_each_entry_rcu(sched, &pernet->sched_list, list)
> > +		list_del_rcu(&sched->list);
> > +	spin_unlock(&pernet->lock);
> > +
> > +	synchronize_rcu();
> 
> Same, why is it needed?
>