On Thu, 2009-06-11 at 13:54 +0300, Jouni Malinen wrote:
> It looks like some programs (e.g., NM) are setting an empty SSID with
> SIOCSIWESSID in some cases. This seems to trigger mac80211 to try to
> associate with an invalid configuration (wildcard SSID) which will
> result in failing associations (or odd issues, potentially including
> kernel panic with some drivers) if the AP were to actually accept this
> anyway).
>
> Only start association process if the SSID is actually set. This
> speeds up connection with NM in number of cases and avoids sending out
> broken association request frames.
>
> Signed-off-by: Jouni Malinen <jouni.malinen@xxxxxxxxxxx>
>
> ---
> net/mac80211/mlme.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> It is unclear when this was broken, but I think we need to get this into
> 2.6.31. The kernel panic (hopefully due to cleanup changes after 2.6.30)
> can be worked around in the driver, I would assume, but it would be
> better if we do not get into this odd state with associations since it
> is not clear what various drivers will do with this kind of
> configuration.
>
>
> --- wireless-testing.orig/net/mac80211/mlme.c 2009-06-11 13:38:01.000000000 +0300
> +++ wireless-testing/net/mac80211/mlme.c 2009-06-11 13:45:44.000000000 +0300
> @@ -2436,6 +2436,14 @@ void ieee80211_sta_req_auth(struct ieee8
> if (WARN_ON(sdata->vif.type != NL80211_IFTYPE_STATION))
> return;
>
> + if (ifmgd->ssid_len == 0) {
> + /*
> + * Only allow association to be started if a valid SSID is
> + * configured.
> + */
> + return;
> + }
> +
> if ((ifmgd->flags & (IEEE80211_STA_BSSID_SET |
> IEEE80211_STA_AUTO_BSSID_SEL)) &&
> (ifmgd->flags & (IEEE80211_STA_SSID_SET |
This whole code is a little odd.. Shouldn't we disassoc if ssid_len ==
0? But then why don't we disassoc when the flags change etc? Looks like
this needs a good rewrite...
johannes
Attachment:
signature.asc
Description: This is a digitally signed message part
