Adam Langley [Tue, 6 Oct 2015 18:05:53 +0000 (11:05 -0700)]
Android: Fix keystore-backed keys with BoringSSL
The switch to BoringSSL broke keystore-backed keys because
wpa_supplicant was using the dynamic ENGINE loading to load
the keystore module.
The ENGINE-like functionality in BoringSSL is much simpler
and this change should enable it.
Dmitry Ivanov [Tue, 6 Oct 2015 13:26:58 +0000 (16:26 +0300)]
Wait longer for inactive client probe (empty data frame)
Some devices cannot respond to inactive client probe (empty data frame)
within one second. For example, iPhone may take up to 3 secs. This
becomes a significant problem when ap_max_inactivity is set to lower
value such as 10 secs. iPhone can lose Wi-Fi connection after ~1 min
of user inactivity.
Helmut Schaa [Wed, 16 Sep 2015 12:04:56 +0000 (14:04 +0200)]
hostapd: Force RADIUS socket renewal on RADIUS auth failures
On RADIUS auth/acct failures hostapd will try a new server if one is
available. Reuse the failover logic to force a socket renewal if only
one RADIUS server is configured.
This fixes problems when a route for the RADIUS server gets added after
the socket was "connected". The RADIUS socket is still sending the
RADIUS requests out using the previous route.
Signed-off-by: Helmut Schaa <helmut.schaa@googlemail.com>
Bob Copeland [Mon, 5 Oct 2015 12:16:40 +0000 (08:16 -0400)]
mesh: Generate proper AID for peer
IEEE Std 802.11-2012 13.3.1 states that the AID should be generated on
the local node for each peer. Previously, we were using the peer link ID
(generated by the peer) which may not be unique among all peers. Correct
this by reusing the AP AID generation code.
Michael Braun [Mon, 5 Oct 2015 14:14:27 +0000 (16:14 +0200)]
Remove WEP support from VLAN
Commit d66dcb0d0b584afdbaba00adcfe661741703094d ('WEP: Remove VLAN
support from hostapd') already removed VLAN support for WEP encryption,
so vlan_setup_encryption_dyn() is no longer needed.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
Michael Braun [Mon, 5 Oct 2015 14:14:26 +0000 (16:14 +0200)]
Fix init of group state machine for static VLANs
This ensures that group key is set as long as the interface exists.
Additionally, ifconfig_up is needed as wpa_group will enter
FATAL_FAILURE if the interface is still down. Also vlan_remove_dynamic()
is moved after wpa_auth_sta_deinit() so vlan_remove_dynamic() can check
it was the last user of the wpa_group.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
Jouni Malinen [Sun, 4 Oct 2015 17:45:20 +0000 (20:45 +0300)]
tests: Read monitor messages more frequently
These test cases left at least one of the attached monitor sockets
blocking for excessive time: ap_wpa2_eap_aka_ext,
ap_hs20_req_conn_capab_and_roaming_partner_preference,
ap_hs20_min_bandwidth_and_roaming_partner_preference, ap_wpa_ie_parsing.
Jouni Malinen [Sun, 4 Oct 2015 15:27:54 +0000 (18:27 +0300)]
tests: Speed up hostapd_oom_loop tests
At some point, these hostapd_oom_* test cases started to fail with
wpa_msg() allocation failure for the AP-ENABLED event. This resulted in
unnecessary long test execution (waiting 30 seconds for an event that
was dropped). Speed this up by using a shorter timeout.
Jouni Malinen [Sun, 4 Oct 2015 08:58:24 +0000 (11:58 +0300)]
Add forgotten list entry removal for control interface deinit
dl_list_del() must be called before freeing the list entries. Neither of
these cases caused problems because the full list data structure was
freed, but still, it is better to do this properly.
Jouni Malinen [Sun, 4 Oct 2015 08:53:26 +0000 (11:53 +0300)]
Throttle control interface event message bursts
Some operations like a new scan result processing can result in large
number of wpa_supplicant control interface messages being generated.
Especially with multiple control interface monitors, this could result
in hitting the output queue length maximum and event messages getting
dropped. In worst case, that could even result in hitting ten
consecutive sendto() errors which could result in an attached monitor
socket getting detached.
Avoid this type of issues by throttling monitor event transmission based
on the output queue length. If more than half of the maximum send buffer
is used, postpone sending of following event messages until the pending
output queue has dropped below the limit.
Jouni Malinen [Sat, 3 Oct 2015 22:42:42 +0000 (01:42 +0300)]
tests: Make it less likely to overflow wlan5 control iface socket
Number of test cases did not read all control interface socket events
from the dynamically added wlan5 interface. This could result in hitting
maximum socket TX queue length and failures in the following test cases.
Jouni Malinen [Sat, 3 Oct 2015 21:35:12 +0000 (00:35 +0300)]
tests: Fix connect_cmd_roam with old scan results
connect_cmd_roam did not force a new scan to find the second AP. This
could result in failures due to the ROAM command getting rejected, e.g.,
in the following test case sequence: wext_pmksa_cache connect_cmd_roam.
Johannes Berg [Tue, 8 Sep 2015 09:46:32 +0000 (12:46 +0300)]
hostapd: Add testing option to use only ECSA
Some APs don't include a CSA IE when an ECSA IE is generated,
and mac80211 used to fail following their channel switch. Add
a testing option to hostapd to allow reproducing the behavior.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Set channel and operating class in hostapd_fill_csa_settings()
The CSA channel and operating class values need to be set for all types
of channel switch (i.e., either if it's triggered by the control
interfaces or due to the GO-follows-STA flow). To do so, move the code
that sets them from the GO-follows-STA flow to the more generic
hostapd_fill_csa_settings() function.
Use ieee80211_freq_to_chan() when getting segment indices
The hostapd_hw_get_channel() function can't be used to convert center
frequencies to channel numbers, because the hw mode lists don't have all
the center frequencies. The hw mode lists have the main channel
frequencies and flags to indicate the channel topography.
For instance, channel 5805 with VHT80- has the channel center frequency
segment 0 at 5775. This segment is only indicated indirectly in the hw
mode list by the HOSTAPD_CHAN_VHT_50_30 flag. The hw mode list doesn't
have any elements with frequency 5775 to allow the conversion to a
channel number. Thus, we need to use ieee80211_freq_to_chan() instead.
Provide an offset to CSA counters in Probe Response frames
Pass to the driver a list of CSA counter offsets when sending Probe
Response frames during a CSA period. This allows the kernel to correctly
update the CSA/eCSA elements.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
nl80211: Specify CSA offsets in send_mlme() driver op
Some management frames contain CSA counters which should be updated by
kernel. Change driver op send_mlme() allowing to send a frame,
specifying an array of offsets to the CSA counters which should be
updated. For example, CSA offsets parameters should be specified when
sending Probe Response frames during CSA period.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Advertise Supported Operating Classes elements in Beacon/Probe Response
Advertise current operating class in Beacon and Probe Response frames.
This Supported Operating Classes element is required by the standard
when extended channel switch is supported. However, since this element
doesn't reflect correctly the sub-band spectrum breakdown and can't be
effectively used by clients, publish only the minimal required part
which is the current operating class.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Ilan Peer [Tue, 8 Sep 2015 09:46:24 +0000 (12:46 +0300)]
Handle channel switch notification for other interface types
Channel switch notification was handled only for AP/GO interfaces. As
the notification can be sent on other interface types as well, extend
the handling to handle other interface types.
Add new GO frequency move policy. The P2P_GO_FREQ_MOVE_SCM_ECSA prefers
SCM if all the clients advertise eCSA support and the candidate
frequency is one of the group common frequencies.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
P2P: Trigger channel selection correctly during CSA
Do not consider moving GOs to a new channel if one of them is in the
middle of CSA. In addition, call wpas_p2p_update_channel_list() after
EVENT_CH_SWITCH is handled.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Ilan Peer [Tue, 8 Sep 2015 09:46:17 +0000 (12:46 +0300)]
P2P: Relax wpas_p2p_init_go_params() for P2P GO CSA
wpas_p2p_move_go_csa() uses wpas_p2p_init_p2p_params() to select the
frequency to move to. However, it is possible that all the channels are
already used, so the selection of a new frequency would fail, although
the frequency used by the P2P GO should not be considered as used if it
is the only one using it.
To overcome this, allow the frequency selection to continue even if all
the frequencies are in use, but the frequency used by the P2P GO is not
used by any station interface.
Use channel switch mechanism to move a P2P GO to a new channel,
when required. In order to be able to reconfigure the GO channel,
split wpa_supplicant_conf_ap() function, so the frequency
configuration part can be reused to find additional CSA settings.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Extended channel switch provides an ability to switch between operating
classes and is required for P2P Devices by the P2P specification when
switching in 5 GHz.
When the operating class is provided for channel switch, the AP/P2P GO
will use eCSA IE in addition to the regular CSA IE both on 2.4 GHz and 5
GHz bands.
Transitions between different hw_modes are not supported.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Channel switch may be performed using both CSA and eCSA IEs together.
This may happen, for example with a P2P GO on band A with legacy
clients. Extend driver API to support up to 2 CSA counters.
This patch also includes the required implementation for nl80211.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
When building Beacon and Probe Response frames for the target channel,
consider bandwidth parameter for VHT channels. In addition, add support
for updating vht_oper_centr_freq_seg0_idx and
vht_oper_centr_freq_seg1_idx.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
When switching to a VHT channel with width greater than 20 MHz, add Wide
Bandwidth Channel Switch element. This element is added in Beacon and
Probe Response frames inside Channel Switch Wrapper element.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Use previously computed channel in cs_freq_params for Beacon frames
When CSA is started, hostapd_change_config_freq() computes the channel
from the provided frequency. Use this stored channel to add CSA IE in
Beacon frames, instead of recomputing the channel each time.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Jouni Malinen [Fri, 2 Oct 2015 16:16:04 +0000 (19:16 +0300)]
tests: Fix OCSP response for ap_wpa2_eap_ttls_ocsp_revoked
Due to a serial number mismatch, the correct "revoked" status was not
used; instead "unknown" was used. While the test case would not fail for
this, incorrect code path was checked.
The readline library depends on ncurses, so it should be set before
ncurses on the linker command line to be able to be statically linked
successfully.
Signed-off-by: David du Colombier <0intro@gmail.com> Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Jouni Malinen [Thu, 1 Oct 2015 15:51:04 +0000 (18:51 +0300)]
Fix TK configuration to the driver in EAPOL-Key 3/4 retry case
Commit 7d711541dced759b34313477d5d163e65c5b0131 ('Clear TK part of PTK
after driver key configuration') started clearing TK from memory
immediately after having configured it to the driver when processing
EAPOL-Key message 3/4. While this covered the most common case, it did
not take into account the possibility of the authenticator having to
retry EAPOL-Key message 3/4 in case the first EAPOL-Key message 4/4
response is lost. That case ended up trying to reinstall the same TK to
the driver, but the key was not available anymore.
Fix the EAPOL-Key message 3/4 retry case by configuring TK to the driver
only once. There was no need to try to set the same key after each
EAPOL-Key message 3/4 since TK could not change. If actual PTK rekeying
is used, the new TK will be configured once when processing the new
EAPOL-Key message 3/4 for the first time.
This commit contains the necessary changes to parallelize
distinct radio work operations which are different in type and
the band used, only when the underlying driver is capable of
supporting such simultaneous offchannel operations.
nl80211: Support vendor scan together with normal scan
Allow wpa_supplicant to use vendor scan (if supported by the driver)
together with the normal nl80211 scan and handling external scan events.
Since this results in possibility of concurrent scan operations, some of
the operations related to scan results need to check more carefully when
an event is relevant for a specific interface.
This commit contains the necessary changes for supporting the QCA vendor
scan implementation, i.e., sending the vendor scan command to underlying
driver and handling the vendor scan events from the driver.
Johannes Berg [Tue, 15 Sep 2015 15:09:22 +0000 (17:09 +0200)]
tests: Add test for hostapd cli ordering
When the 'SET wpa 2' command is executed last, it seems to somehow
reset parts of the settings, causing hostapd to beacon with the
pairwise cipher suite selector set to 00-0F-AC:0 (none/use-group).
This is not permitted and should be rejected; wpa_supplicant also
cannot connect.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Do not clear RSN parameters before full configuration
This fixes an issue where hostapd SET command is used to configure RSN
parameters and the wpa parameter is sent after the other parameters.
Previously, the default case here ended up clearing rsn_pairwise and
wpa_pairwise values and once wpa=2 was finally set, the cipher
configuration had already been lost.
Maneesh Jain [Thu, 18 Jun 2015 04:16:34 +0000 (09:46 +0530)]
D-Bus: Add InvitationReceived Signal
This is equivalent to the P2P_EVENT_INVITATION_RECEIVED signal on the
control interface. It can be used to sent the Invitation Received signal
to applications written using D-Bus.
Fix an iteration bug in get_shared_radio_freqs_data when building
freqs_data array. Only the last used-by flag was maintained instead of
making this a bitfield of all found uses.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
D-BUS: dev_passwd_id should be "q" because it uses DBUS_TYPE_UINT16
This corrects the type of dev_passwd_id in GONegotiationRequest event.
This field is packed as DBUS_TYPE_UINT16 but in
wpas_dbus_interface_signals it was "i" which is DBUS_TYPE_INT32.
Android: Set ctrl_iface client socket group (AID_WIFI) separately
Split chown() call in wpa_ctrl_open() and wpa_ctrl_open2() to allow the
group id to be set even if the process does not have privileges to
change the owner. This is needed for modules that need to communicate
with wpa_supplicant since without the group change, wpa_supplicant may
not have privileges to send the response to a control interface command.
Android: Avoid same per-iface and global ctrl socket
Android platform assigns the same socket id if the socket identifier in
conf->ctrl_interface and global->params.ctrl_interface (parameter for
android_get_control_socket) point to the same Android specific control
socket. This ends up having two eloop socket handlers registered for the
same file descriptor and thus, two attempt to receive and process each
command. This can result in unexpected failure, e.g., the prefix IFNAME=
for any command is valid for global socket handler, but results in
UNKNOWN COMMAND response from the per-interface ctrl socket handler).
Since it might be possible to end up with this type of invalid
configuration in OTA upgrade, compare the socket identifiers and do not
open the ctrl socket on the respective interface if both point to same.
This allows the Wi-Fi framework to use the global control interface.
Do not copy STA VHT capabilities if VHT is not enabled for AP
Previously, station's VHT information elements were copied and passed
regardless of the AP's VHT configuration. As a result, AP with VHT
disabled in configuration could have ended up transmitting packets in
VHT rates though AP is not advertising VHT support. Fix this by copying
the station's VHT capabilities only when AP supports VHT (both hardware
and configuration).
tests: WPA2-Enterprise interactive identity entry and ENABLE_NETWORK
This verifies that ENABLE_NETWORK does not trigger reconnection if
already connected. The previous commit fixed a case where it was
possible for that to happen.
Avoid reconnection on ENABLE_NETWORK if already connected
This was already the case for most command sequences, but it was
possible for wpa_s->reassociate to be set to 1 when CTRL-RSP-* commands
were used to set identity, password, or passphrase for EAP
authentication. In such cases, ENABLE_NETWORK issued after the
connection was completed could result in a new connection attempt
(likely reconnection back to the same BSS).
Fix this by checking whether an actual connection is already present
even if wpa_s->reassociate is set when processing the ENABLE_NETWORK
command.
Update AP WPA/RSN IE on all associations if driver can select BSS
It is possible for driver-based BSS selection to end up reassociating
back to the current AP. If wpa_supplicant preferred another BSS, it
would have updated the internal knowledge of the AP's WPA/RSN IE when
requesting a new connection. In the special case of existing association
and new association being with the same BSS that is different from the
wpa_supplicant preference, association event processing skipped the
WPA/RSN IE update. This could result in the following 4-way handshake
getting rejected due to incorrectly detected mismatch with AP's RSN/WPA
IE between Beacon/Probe Response frame and EAPOL-Key msg 3/4.
Fix this by updating the AP WPA/RSN IE on all association events when
driver-based BSS selection is used regardless of whether the BSSID
changes. This could also cover a theoretical case of the AP changing its
RSN/WPA IE at the very moment we try to reassociate back to the same
BSS.
Drop some control interface debug print verbosity for send operations
These prints were at DEBUG level (-d), but they can be very frequent, so
drop them to MSGDUMP (-dd). This allows the prints to be suppressed in
common debugging cases while still leaving them easily enablable to
debug control interface issues without having to enable excessive
debugging.
Reduce debug verbosity for read-only control interface commands
Commands like BSS and GET_NETWORK are used in some cases very frequently
and those can increase the amount of debug information from
wpa_supplicant without significant benefit. These were logged at the
DEBUG level (-d). Move logging of such read-only commands (i.e., no new
wpa_supplicant operation is started based on it) to EXCESSIVE level
(-ddd) which was already used for the PING command.
WPA: Do not print GTK in debug log unless requested
The GTK value received in RSN (WPA2) group rekeying did not use the
wpa_hexdump_key() version of debug printing that is conditional on -K
being included on the command line.
Use shorter variables for storing the attribute lengths and group these
variables together to allow compiler to pack them more efficiently. This
reduces the struct size from 960 bytes to 760 bytes in 64-bit builds.
This reduces stack use in number of functions.
Chen, Yi [Wed, 2 Sep 2015 15:55:01 +0000 (21:25 +0530)]
WPS: Fix num_probereq_cb clearing on DISABLE to avoid segfault
Reset hapd->num_probereq_cb to 0 on an interface deinit to avoid
unexpected behavior if the same interface is enabled again without fully
freeing the data structures. hostapd_register_probereq_cb() increments
hapd->num_probereq_cb by one and leaves all old values unchanged. In
this deinit+init case, that would result in the first entry in the list
having an uninitialized pointer and the next Probe Request frame
processing would likely cause the process to terminate on segmentation
fault.
This issue could be hit when hostapd was used with WPS enabled (non-zero
wps_state configuration parameter) and control interface command DISABLE
and ENABLE were used.