This test command is supposed to change the WPS version number in all
places immediately, so make sure that the IEs used in management
frames get updated immediately.
This is not supposed to happen, but in theory, the GO may end up
changing its channels at some point. While it would be reasonable
to use one of the common channels, GO may be forced to use another
channel. The most likely reason for this would be if the P2P client
is enforcing only a single channel to be used (e.g., to avoid
multi-channel concurrent operations).
The P2P group component is now tracking of associated stations
in the group and the Group Limit bit in the Group Capabilities
is updated based on whether there is room for new clients in
the group.
This driver capability limit was supposed to indicate limit on
how many stations can associate with us in the AP mode, not how
many P2P peers are stored in the peer table.
The workaround to ignore no ctrl::ack received for GO Negotiation
Confirmation frame was only supposed to be used when the frame was
actually transmitted and just the ack was not received. However, due
to the way the driver failure on transmitting the frame were reported,
this ended up getting applied for all failures in sending the GO
Negotiation Confirmation frame.
Improve this by providing a mechanism to indicate whether send_action
operations fail locally before the frame was actually transmitted or
because of not receiving ack frame after having transmitted the frame.
This is needed to work around cases where the pending interface gets
removed, e.g., due to failed GO Negotiation. In such case, we may still
accept new GO Negotiation to be completed, but the interface did not
get created without a separate p2p_connect command.
The current implementation is not ideal since the re-created interface
may, at least in theory, get incorrect interface addrees. Though, this
would likely require that the driver supported more than one P2P group
interface and that the previous one was already taken into use for
another group.
UPnP event subscriptions are not supposed to be dropped based on
whether events can be delivered quickly enough. Leave dropping to
happen only based on failed deliveries to avoid issues with a burst
of events kicking out still active subscribers.
Do not send more than five Probe Request WLANEvent notifications
per second. Even though the limit should in theory apply to all
WLANEvents, it is better not to drop EAP notifications because
of Probe Request frames and really, the only real reason for
event bursts is Probe Request frames.
Instead of queuing all events for a subscriber, only queue more
important events if delivery of event notifications starts failing.
This allows more time for the subscriber to recover since the maximum
queue length if not reached because of Probe Request frames only.
Instead of dropping the subscription on first failure, allow up to
10 failures before dropping. In addition, drop the callback URLs
one by one instead of full subscription if only one URL is failing.
This is a workaround for incorrect configuration (missing
virtual/physical identifier for config methods) for WPS 2.0 to
allow unmodified configuration from WPS 1.0 to be used while
enforcing compliant WPS 2.0 values.
If CONFIG_DRIVER_WEXT=y is not used in .config, but driver_wext.c
gets pull in to help another driver wrapper, rfkill code needs to
included to fix the build.
Listen to regulatory event messages from kernel and convert them to
internal driver event notifications indicated that the channel list
may have changed.
The AP operation with wpa_supplicant requires an additional callback
to get the needed event information from hostapd side so that
wpa_msg() can be called for wpa_s->parent if needed.
This makes it easier to figure out what could have failed in the
WPS protocol and potentially provide more information for the
user on how to resolve the issue.
The driver wrapper can now indicate whether the driver supports
concurrent operations on multiple channels (e.g., infra STA connection
on 5 GHz channel 36 and P2P group on 2.4 GHz channel 1). If not,
P2P_CONNECT commands will be rejected if they would require
multi-channel concurrency.
The new failure codes for P2P_CONNECT:
FAIL-CHANNEL-UNAVAILABLE:
The requested/needed channel is not currently available (i.e., user has
an option of disconnecting another interface to make the channel
available).
FAIL-CHANNEL-UNSUPPORTED:
The request channel is not available for P2P.
Allow only the expected P2P Interface Address as the BSSID for
the AP to avoid selecting incorrect BSS should there be another
device that is advertising active PBC mode before the target
P2P GO does.
At least D-Link DIR-600 and DIR-825 have been reported to include
an extra octet after the Network Key attribute within a Credential
attribute. This can happen at least when they are provisioning an
open network.
Add a workaround to detect this incorrectly encoded attribute and
to skip the extra octet when parsing such a Credential.
The new nl80211 API means we don't need to use monitor interfaces. This
means that the P2P implementation now requires a kernel that has support
for generic management frame (not just Action frame) transmission.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
This is backward compatible since older kernels will ignore the extra
attribute and only allow registration for Action frames.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When the P2P state machine requests a remain- on-channel, there's a
potential race where it can then request a stop before the r-o-c has
actually started, in which case the stop will not be processed. Fix
that.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
libnl has a bug, when binding more than two sockets and releasing one,
it will release the wrong address and then try to reuse it, which fails.
Therefore, we need to reimplement the socket address assignment logic
locally for libnl 1.1.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
The new DBus API helper function wpas_dbus_error_unknown_error
function can be called as a result of a failure within internal
getter calls, which will call this function with a NULL message
parameter. However, dbus_message_new_error looks very unkindly
(i.e, abort()) on a NULL message, so in this case, we should not
call it.
I've observed this course of events during a call to
wpas_dbus_getter_bss_wpa with a faileld parse of the IE parameter.
We got here through a call to fill_dict_with_properties which
explicitly calls getters with a NULL message parameter. Judging
from the way it is called, this could easily occur if an AP sends
out a malformed (or mis-received) probe response. I usually run
into this problem while driving through San Francisco, so I'm
exposed to any number of base stations along this path.
