cfg80211/mac80211 seems to be unwilling to change interface mode in
some cases. Make these less likely to cause problems by trying the
changes up to 10 times with 100 msec intervals.
In theory, this should not really be needed, but Windows 7 uses
Registrar mode to probe AP's WPS capabilities before trying to use
Enrollee and fails if the AP does not allow that probing to happen.
This allows the AP to start as an Enrollee and send M1, but refuse
to continue beyond that (M3 will not be sent if AP PIN is not known).
This makes it easier (and a bit faster) to handle multiple local
radios with FT. There is no need to depend on l2_packet in that
case since the frame can be delivered as a direct function call.
If the station does not include WSC IE in Association Request, it
is marked with WLAN_STA_MAYBE_WPS flag. We can update that to
WLAN_STA_WPS if the station uses either of the WPS identity strings.
This enables some workarounds for WPS stations.
This works around issues with EAP-Failure getting lost for some reason.
Instead of waiting up to 60 seconds on a timeout, 30 second timeout is
now used and whenever the provisioning step has been completed (either
successfully or unsuccessfully), this timeout is reduced to 2 seconds.
This is needed to avoid trying to reassociate based on new scan
results when using wpa_supplicant to control AP mode. This could
happen if something external triggered the driver to run a scan.
When SME is in wpa_supplicant (mac80211), we need to make sure that
the kernel code has valid BSS entry for the AP. In some cases it
seemed to be possible to end up not having current information in
cfg80211 or mac80211 which can result to association failures. Avoid
this by always running through the scan request before initial
connection attempt.
It is possible that l2_packet is not used with wpa_supplicant
in some cases, so better make sure we do not end up notifying
l2_packet code about authentications unless it was actually
initialized in the first place.
The pairwise cipher field is supposed to be little endian, but the
message building functions did not swap the bytes on big endian
hosts while the message processing functions did. Fix this by using
little endian byte order in both places.
Store list of all discovered BSSes in the ESS and on which frequencies
they have been seen. Use this information to dynamically generated the
list of channels for background scans.
It looks like 10.6.3 and 10.6.4 do not like to receive Network Key
with WPA passphrase while PSK format still works. Use peer information
from M1 to figure out whether the Enrollee is likely to be OS X and
if so, force PSK format to be used for Network Key.
Some frequent debug prints are of limited use and make debug output
difficult to read. Make them use a new debug level so that -dd
provides more readable output (-ddd can now be used to enable
the excessive debug prints).
The length of the prime was used incorrectly and this resulted
in WPS DH operation failing whenever the public key ended up having
leading zeros (i.e., about every 1/256th time).
The AP configuration may change after provisioning, so it is better
not to use the current security policy to prioritize results. Instead,
use WPS Selected Registrar attribute as the main sorting key and use
signal strength next without considering security policy or rate sets.
The non-WPS provisioning case remains as-is, i.e., this change applies
only when trying to find an AP for WPS provisioning.
Some deployed implementations do not include the mandatory Network
Key attribute when a WPS Credential is for an open network. Allow
this to improve interoperability since the actual key value is not
really needed for open networks.
Addition of the background scanning mechanism in commit
60b94c9819 moved the scan trigger
into a new function that was also incrementing the scan_runs
counter, but the removal of the previous scan_runs incrementation
was forgotten from that patch. This counter should only be updated
into a single location, so remove the old one. This improves AP
selection for WPS provisioning by not skipping some of the initial
scans.
Replace use of rfkill block event with rtnetlink ifdown. This makes
the design more robust since the rfkill event could have been for
another interface while the rtnetlink events are already filtered
based of ifindex. In addition, the new design handles other than
rfkill-triggered ifdown/ifup events, too. rfkill unblocked event
is still needed to try to set the interface back up. If the unblock
was for another interface, ifup will fail and the driver state is
not changed.
Some of the driver wrappers need to get set_ieee8021x() call before
they can configure keys. Reorder the operations in the reloading
of configuration case to match with that expectation.
