Re: [RFC] nl80211: allow multiple active scheduled scan requests

[Johannes Berg <johannes@sipsolutions.net>]

> +	/*
> +	 * allow only one legacy scheduled scan if user-space
> +	 * does not indicate multiple scheduled scan support.
> +	 */
> +	if (!info->attrs[NL80211_ATTR_SCHED_SCAN_MULTI] &&
> +	    cfg80211_legacy_sched_scan_active(rdev))
>  		return -EINPROGRESS;

That probably doesn't go far enough - if legacy one is active then we
probably shouldn't allow a new MULTI one either (or abandon the legacy
one) so that older userspace doesn't get confused with multiple
notifications from sched scans it didn't start.
 
> +	if (rdev->sched_scan_req_count == rdev->wiphy.max_sched_scan_reqs)
> +		return -ENOSPC;

Do we really want to do the double-accounting, just to avoid counting
the list length here?

> +	/* leave request id zero for legacy request */

why? The ID would be ignored, so why special-case it?

> +static void cfg80211_del_sched_scan_req(struct
> cfg80211_registered_device *rdev,
> +					struct
> cfg80211_sched_scan_request *req)
> +{
> +	list_del_rcu(&req->list);
> +	kfree_rcu(req, rcu_head);
> +	synchronize_rcu();
> +	rdev->sched_scan_req_count--;
> +}

That's bogus - either you use kfree_rcu() or synchronize_rcu() (the
former is much better); combining both makes no sense.

> +bool cfg80211_legacy_sched_scan_active(struct
> cfg80211_registered_device *rdev)
> +{
> +	struct cfg80211_sched_scan_request *req;
> +
> +	req = list_first_or_null_rcu(&rdev->sched_scan_req_list,
> +				     struct
> cfg80211_sched_scan_request, list);
> +	/* request id 0 indicates legacy request in progress */
> +	return req && !req->reqid;
> +}

Ok, fair enough.

johannes