[PATCH v3] cfg80211: Reorder wiphy_register/unregister() notifications relevantly

[PATCH v3] cfg80211: Reorder wiphy_register/unregister() notifications relevantly

[Tomasz Bursztyka]

Currently it can send regulatory domain change notification before any
NEW_WIPHY notification. Moreover, if rfill_register() fails, calling
wiphy_unregister() will send a DEL_WIPHY though no NEW_WIPHY had been
sent previously.

Thus reordering so it properly notifies NEW_WIPHY before any other.

Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
---

Reusing "registered" marker to know whether it sent NEW_WIPHY yet or not.
Thus if wiphy_unregister() is called du to rfkill registration failure, it
won't send DEL_WIPHY.

 net/wireless/core.c | 33 +++++++++++++++++++--------------
 1 file changed, 19 insertions(+), 14 deletions(-)

diff --git a/net/wireless/core.c b/net/wireless/core.c
index afee5e0..08162e8 100644
--- a/net/wireless/core.c
+++ b/net/wireless/core.c
@@ -635,6 +635,21 @@ int wiphy_register(struct wiphy *wiphy)
 	if (IS_ERR(rdev->wiphy.debugfsdir))
 		rdev->wiphy.debugfsdir = NULL;
 
+	cfg80211_debugfs_rdev_add(rdev);
+	rtnl_unlock();
+
+	res = rfkill_register(rdev->rfkill);
+	if (res) {
+		rfkill_destroy(rdev->rfkill);
+		rdev->rfkill = NULL;
+		wiphy_unregister(&rdev->wiphy);
+		return res;
+	}
+
+	rtnl_lock();
+
+	nl80211_notify_wiphy(rdev, NL80211_CMD_NEW_WIPHY);
+
 	if (wiphy->regulatory_flags & REGULATORY_CUSTOM_REG) {
 		struct regulatory_request request;
 
@@ -646,21 +661,9 @@ int wiphy_register(struct wiphy *wiphy)
 		nl80211_send_reg_change_event(&request);
 	}
 
-	cfg80211_debugfs_rdev_add(rdev);
-
 	rdev->wiphy.registered = true;
 	rtnl_unlock();
 
-	res = rfkill_register(rdev->rfkill);
-	if (res) {
-		rfkill_destroy(rdev->rfkill);
-		rdev->rfkill = NULL;
-		wiphy_unregister(&rdev->wiphy);
-		return res;
-	}
-
-	nl80211_notify_wiphy(rdev, NL80211_CMD_NEW_WIPHY);
-
 	return 0;
 }
 EXPORT_SYMBOL(wiphy_register);
@@ -699,8 +702,10 @@ void wiphy_unregister(struct wiphy *wiphy)
 		rfkill_unregister(rdev->rfkill);
 
 	rtnl_lock();
-	nl80211_notify_wiphy(rdev, NL80211_CMD_DEL_WIPHY);
-	rdev->wiphy.registered = false;
+	if (rdev->wiphy.registered) {
+		nl80211_notify_wiphy(rdev, NL80211_CMD_DEL_WIPHY);
+		rdev->wiphy.registered = false;
+	}
 
 	WARN_ON(!list_empty(&rdev->wdev_list));
 
-- 
1.8.5.5

Re: [PATCH v3] cfg80211: Reorder wiphy_register/unregister() notifications relevantly

[Johannes Berg]

On Tue, 2014-08-19 at 12:35 +0300, Tomasz Bursztyka wrote:

> +	cfg80211_debugfs_rdev_add(rdev);
> +	rtnl_unlock();

I'm not convinced that it's safe to drop the rtnl here, when we only
later set wiphy.registered = true.

As is, the wiphy would end up on the list, but not registered - yet in
this period it would be reachable for example for nl80211 commands in
the unregistered state. This seems wrong.

Additionally, maybe add a brief note regarding the second part of the
patch (checking registered in unregister) to the commit log?

johannes

Re: [PATCH v3] cfg80211: Reorder wiphy_register/unregister() notifications relevantly

[Tomasz Bursztyka]

Hi Johannes,

>> >+	cfg80211_debugfs_rdev_add(rdev);
>> >+	rtnl_unlock();
> I'm not convinced that it's safe to drop the rtnl here, when we only
> later set wiphy.registered = true.
>
> As is, the wiphy would end up on the list, but not registered - yet in
> this period it would be reachable for example for nl80211 commands in
> the unregistered state. This seems wrong.

Ok, I counted on the fact the device is not yet advertized as being 
ready (the NEW_WIPHY below).
Will see how to avoid this early unlock.

> Additionally, maybe add a brief note regarding the second part of the
> patch (checking registered in unregister) to the commit log?

Sure.

Tomasz