Re: [PATCH] cfg80211: unblock user hint when cfg80211_regdom is intersected
From: Luis R. Rodriguez <hidden>
Date: 2014-06-18 21:44:29
On Sat, Jun 14, 2014 at 1:42 PM, [off-list ref] wrote:
quoted hunk ↗ jump to hunk
From: Mihir Shete <redacted> User hints are always ignored if cfg80211_regdom is intersected. Add checks to make sure that user hints are processed if the previous hints are done processing. Signed-off-by: Mihir Shete <redacted> --- net/wireless/reg.c | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-)diff --git a/net/wireless/reg.c b/net/wireless/reg.c index b4973bc..2c4e2ba 100644 --- a/net/wireless/reg.c +++ b/net/wireless/reg.c@@ -396,6 +396,14 @@ static bool regdom_changes(const char *alpha2) return !alpha2_equal(r->alpha2, alpha2); } +static bool is_cfg80211_regdom_intersected(void) +{ + const struct ieee80211_regdomain *r = get_cfg80211_regdom(); + + if (!r) + return false; + return is_intersected_alpha2(r->alpha2); +}
Shouldn't this just be the last request->intersect too ?
quoted hunk ↗ jump to hunk
/* * The NL80211_REGDOM_SET_BY_USER regdom alpha2 is cached, this lets * you know if a valid regulatory hint with NL80211_REGDOM_SET_BY_USER@@ -1650,9 +1658,14 @@ __reg_process_hint_user(struct regulatory_request *user_request) */ if ((lr->initiator == NL80211_REGDOM_SET_BY_CORE || lr->initiator == NL80211_REGDOM_SET_BY_DRIVER || - lr->initiator == NL80211_REGDOM_SET_BY_USER) && - regdom_changes(lr->alpha2)) - return REG_REQ_IGNORE; + lr->initiator == NL80211_REGDOM_SET_BY_USER)) { + if (lr->intersect) { + if (!is_cfg80211_regdom_intersected()) + return REG_REQ_IGNORE;
Let's keep this in mind: When processing requests on the queue get_last_request() will return the last requests which was accepted and already processed. When reg_update_last_request() is issued last_request becomes the newly accepted request that we are about to process. If this seems rather confusing we should split this up into two, but so far its not clear to me this is needed given we serialize processing requests and have a last_request->pending too. __reg_process_hint_user() is evaluating whether or not we should apply the passed hint. Given all this, when would lr->intersect be true but the set regdomain not be intersected? That would seem like an issue to me. Also note that your commit log seems to claim to want to make a change only for user type hints but the above folds all your desired changes to all the checks it was checking for. Lastly, we should be allowing through user changes if and only if a core or driver hint was not issued, otherwise the intersection makes sense to upkeep, however I do see it making sense to allow a user regulatory if an intersection was done but *iff* we want to clear the old userspace setting, but note that to do that cleanly we'd have to assume an original state. We can do this without much overhead by resetting the regulatory core to the original state given that core and driver hints will be cached and be set once the reset is called. You have to consider when a reset makes sense though and be very very careful about this though otherwise you might end up in inconsistent states that make no sense at run time. Consider being allowed to do certain things, being associated, and then all of a sudden driving the regulatory core which would prohibit that current state. If you really want to enable this regardless of the state you may have to trigger disassoc, etc. Please think of all this. NACK for now. Luis