In highly congested network (BSSes almost on every channel
within ESS) we have hit a bug when wpa_supplicant become
completly irresponsive, infinite looping on while loop.
When probe_idx was equal 0 and we are not able to probe
new frequency, following condition were never fulfilled:
"if (!in_array(freqs, data->supp_freqs[idx]))"
Signed-hostap: Pawel Kulakowski <pawel.kulakowski@tieto.com>
Change the nl80211 driver in wpa_supplicant to correctly handle
connecting to a new AP through cfg80211 without SME capability. As
before, the driver will disconnect from the previously associated AP,
but now we attempt to immediately connect to our intended AP. This
prevents us from blacklisting the AP we were trying to connect to
because of a semantic mismatch between cfg80211 and wpa_supplicant. The
disconnect/connect patch generates a local disconnect nl80211 event
which we discard because we're already correctly tracking the pending
association request.
In detail:
cfg80211 does not support connecting to a new BSS while already
connected to another BSS, if the underlying driver doesn't support
separate authenticate and associate commands. wpa_supplicant is written
to expect that this is a supported operation, except for a little error
handling that disconnects from the current BSS when roaming fails and
relies on autoconnect logic to reconnect later. However, this failure to
connect is incorrectly attributed to the new AP we attempted to
associate with, rather than a local condition in cfg80211.
The combined effect of these two conditions is that full-mac drivers
accessible through cfg80211 but without SME capability take a long time
to roam across BSS's because wpa_supplicant will:
1) Fail to associate for local reasons
2) Disconnect and return that the association request failed
3) Blacklist the association target (incorrectly)
4) Do a scan
5) Pick a less desirable AP to associate with
Signed-hostap: Christoper Wiley <wiley@chromium.org>
Commit deca6eff74 added a redundant call
to l2_packet_get_own_addr. Use the information we already have in
atheros_init.
Signed-hostap: Baruch Siach <baruch@tkos.co.il>
If identity round limit is reached, EAP-SIM/AKA session is terminated.
This needs to free the allocated message.
Signed-hostap: Jouni Malinen <j@w1.fi>
There is no point in the hapd_iface == NULL validate after this pointer
has been dereferences, so move the code dereferencing hapd_iface after
the check.
Signed-hostap: Jouni Malinen <j@w1.fi>
The UFD (USB flash drive) configuration method was deprecated in WSC
2.0. Since this is not known to be used, remove the UFD implementation
from hostapd and wpa_supplicant to allow the WPS implementation to be
cleaned up. This removes the now unused OOB operations and ctrl_iface
commands that had already been deprecated by the new NFC operations.
Signed-hostap: Jouni Malinen <j@w1.fi>
The PSK generation done by pbkdf2_sha1() is one of the longest CPU time
users according to our profiling from boot to GO started.
So I have reduced some steps.
I could boot a GO by this command sequence.
-------------
add_net
set_network 0 ssid '"DIRECT-XX"'
set_network 0 psk
'"123456789012345678901234567890123456789012345678901234567890123"'
set_network 0 proto RSN
set_network 0 key_mgmt WPA-PSK
set_network 0 pairwise CCMP
set_network 0 auth_alg OPEN
set_network 0 mode 3
set_network 0 disabled 2
p2p_group_add persistent=0 freq=2412
-------------
By this sequence, pbkdf2_sha1() was called three times and the function
calculates the same value each time. Reduce number of calls to
pbkdf2_sha1() from 3 to 1 by caching the previous result.
Signed-hostap: Masashi Honma <masashi.honma at gmail.com>
There are a few instances where dbus handlers test the value
of errno to test whether strtoul completes successfully.
Since strtoul does not clear errno, and there's no strong
reason to suspect that errno is already clear, it is safer
to clear it right before calling strtoul. Also, any failure
in strtoul (setting errno non-zero) should be considered a
failure.
While testing using dbus-send, I found that a malformed
network path can cause a crash due to net_id being left
NULL. We should test for this before calling strtoul
on it.
Tested with:
dbus-send --system --dest=fi.w1.wpa_supplicant1 --print-reply \
/fi/w1/wpa_supplicant1/Interfaces/0 \
org.freedesktop.DBus.Properties.Get \
string:fi.w1.wpa_supplicant1.Interface string:Networks
dbus-send --system --dest=fi.w1.wpa_supplicant1 --print-reply \
/fi/w1/wpa_supplicant1/Interfaces/0 \
fi.w1.wpa_supplicant1.Interface.RemoveNetwork \
objpath:/fi/w1/wpa_supplicant1/Interfaces/0/Networks/0
dbus-send --system --dest=fi.w1.wpa_supplicant1 --print-reply \
/fi/w1/wpa_supplicant1/Interfaces/0 \
fi.w1.wpa_supplicant1.Interface.RemoveNetwork \
objpath:/fi/w1/wpa_supplicant1/Interfaces/0/Networks/0
dbus-send --system --dest=fi.w1.wpa_supplicant1 --print-reply \
/fi/w1/wpa_supplicant1/Interfaces/0 \
fi.w1.wpa_supplicant1.Interface.RemoveNetwork \
objpath:/fi/w1/wpa_supplicant1/Interfaces/0
Signed-hostap: Paul Stewart <pstew@chromium.org>
intended-for: hostap-1
This function is now unused after the last couple of commits that
removed the last uses, so remove this to keep code simpler since all
places that disassociate, can use deauthentication instead.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Even though the standard currently describes disassociation to be used
for RSN element mismatch between Beacon/Probe Response frames and
EAPOL-Key msg 3/4, this is unnecessary difference from other cases that
deauthenticate. In addition, there is no point in leaving the 802.11
Authentication in place in this case. To keep things simpler, use
deauthentication here to get rid of the only use of
wpa_sm_disassociate().
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
cfg80211/mac80211 may reject disassociation command if association has
not yet been formed. Use deauthentication in cases where it is possible
that we are associating at the moment the command is issued.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
wpa_supplicant_{disassociate,deauthenticate}() need to inform the driver
about decision to disconnect even if this happens during the time when
the driver is still trying to complete association. During that time,
wpa_s->bssid is not set, so the code in these functions needs to figure
out the correct BSSID based on that field or wpa_s->pending_bssid. In
addition, it is possible that the BSSID is not even known at
wpa_supplicant at this point in time when using drivers that perform BSS
selection internally. In those cases, the disconnect command needs to be
sent to the driver without the BSSID.
This fixes issues where the driver (or cfg80211 in particular) may be
left in mismatching state with wpa_supplicant when disconnection (e.g.,
due to a ctrl_iface command) happens between connection request and
association event.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
This helper script can now handle both reading of a NFC tag and
initiation of NFC connection handover if a peer NFC device is touched.
Signed-hostap: Jouni Malinen <j@w1.fi>
If the driver indicates the association (or authentication) was
rejected, wpa_supplicant should handle this connection failure similarly
to other cases. Previously, this was only handled with drivers that use
wpa_supplicant SME.
In case of cfg80211-based drivers, a rejected association was actually
already handled since cfg80211 generates a deauthentication event after
indicating connection failure. However, rejected authentication resulted
in wpa_supplicant waiting for authentication timeout to expire which is
unnecessary long wait.
Fix this by calling wpas_connection_failed() to use the common mechanism
to reschedule a new connection attempt with the previously attempted
BSSID blacklisted.
Signed-hostap: Jouni Malinen <j@w1.fi>
If WPS Registrar tries to provision a WPA/WPA2-Personal network without
including a valid Network Key, the network block cannot be used to
connect to the network. Reject such credential without adding the
network block. This makes wpa_supplicant send WSC_NACK as a response to
the invalid Credential and stop the provisioning process immediately
rather than only after trying unsuccessfully to connect to the network.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
The DNS Name is allowed to use or not use domain name compression. To
handle both cases, check human readable DNS Name match if binary
matching does not show a hit.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
There may be multiple Bonjour PTR matches for the same key, so extend
the P2P SD code for this to allow such entries to be added (i.e., do not
override previously added value, but add a new one). Similarly, return
multiple matches (one per Service TLV) for a query if it happens to
match more than a single configured Bonjour service.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
The new P2P_SET parameter disc_int can now be used to configure
discoverable interval for p2p_find operations. The format of the command
for setting the values is "P2P_SET disc_int <minDiscoverableInterval>
<maxDiscoverableInterval> <max TUs for discoverable interval>". The
first two parameters are given in units of 100 TUs (102.4 ms). The third
parameter can be used to further limit the interval into a specific TU
amount. If it is set to -1, no such additional limitation is enforced.
It should be noted that the P2P specification describes the random
Listen state interval to be in units of 100 TUs, so setting the max TU
value to anything else than -1 is not compliant with the specification
and should not be used in normal cases. The default parameters can be
set with "P2P_SET disc_int 1 3 -1".
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
If the driver rejects the scan request while handling an external
requested scan (e.g., SCAN command on ctrl_iface), wpa_s->scan_req gets
cleared in wpa_supplicant_scan(). This can results in issues when the
scheduled re-try of the scan in one seconds trigger another call to this
function. If ap_scan==2 mode is used, this would result in new
association attempt instead of a new scan. Avoid this by restoring value
of wpa_s->scan_req in case the scan trigger fails and a new scan attempt
is scheduled.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
wpas_p2p_pd_before_join_timeout() needs to clear the
pending_pd_before_join flag to match other uses of this flag prior to
calling wpas_p2p_join_start(). Without this, the flag could be left set
which can cause following P2P operations to behave in unexpected ways.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
The new p2p_no_group_iface=1 configuration parameter can now be used to
disable the default behavior of adding a separate interface for the P2P
group when driver support for concurrent interfaces is available.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 0d30cc240f forced
wpa_s->current_ssid to be cleared in wpa_supplicant_mark_disassoc()
which gets called from wpa_supplicant_event_disassoc(). This breaks the
P2P group idle mechanism for the case where p2p_group_idle is not set
(i.e., is the default 0) since wpas_p2p_group_idle_timeout() ignores the
timeout in that case if the interface is not recognized as a client
interface (which was based on wpa_s->current_ssid being set).
Fix this by making wpas_p2p_is_client() default to client case if
wpa_s->current_ssid is NULL. This is much more likely case since the P2P
GO mode operation would not really clear the pointer without explicit
request to disconnect.
Signed-hostap: Jouni Malinen <j@w1.fi>
The old WPS interface for using NFC has no known deployments and even
the binary libraries referenced here are not easily available anymore.
Since the new interface for using NFC with WPS covers the same
functionality, remove the old implementation to clean up WPS
implementation.
Signed-hostap: Jouni Malinen <j@w1.fi>
When WPS is used with NFC connection handover, the AP may indicate its
operating channel within the credential information. Use this
informatiom, if present, to speed up the scan process.
Signed-hostap: Jouni Malinen <j@w1.fi>
This commit adds new wpa_supplicant ctrl_iface commands to allow
external programs to go through NFC connection handover mechanism
with wpa_supplicant taking care of the WPS processing. This version
includes only the case where wpa_supplicant is operating as a
station/Enrollee.
Signed-hostap: Jouni Malinen <j@w1.fi>
The previous eapol_sm_notify_cached() implementation forced the port
to be authorized when receiving EAPOL-Key msg 1/4 that included a
matching PMKID in cases when PMKSA caching is used. This is too early
since the port should really be authorized only after the PTK has been
configured which is the case when PMKSA caching is not used.
Fix this by using the EAPOL supplicant PAE state machine to go through
the AUTHENTICATING and AUTHENTICATED states instead of forcing a jump
to AUTHENTICATED without performing full state machine steps. This can
be achieved simply by marking eapSuccess TRUE at least with the current
version of EAP and EAPOL state machines (the earlier commits in this
function seemed to indicate that this may have not been that easy in
the older versions due to the hacks needed here).
This addresses an issue with nl80211-based driver interface when the
driver depends on the STA Authorized flag being used to prevent
unprotected frames from being accepted (both TX and RX) prior to PTK
configuration.
Signed-hostap: Jouni Malinen <j@w1.fi>
intended-for: hostap-1
If the driver indicates support for multi-channel concurrency, change
the p2p_connect behavior to not force the current operating channel, but
instead, just mark it as preferred for GO Negotiation. This change
applies only for the case when the freq parameter is not used with the
p2p_connect command.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
There was an issue with EAPOL frame exchanges in a multi-BSS
configuration when a station switches between the BSSes controlled by
the same hostapd process. When processing the EAPOL packet, the array of
virtual APs (iface->bss) is searched looking for the station that sent
the packet in order to identify which signal context should be used
during processing. The first match of the station in its list gets used
in the ieee802_1x_receive() function. However, even after a station has
disassociated, it remains in the list of stations pending an inactivity
timeout. This leads to the wrong hapd context (one where the station had
already disassociated) being used in some cases (if the current/active
bss entry appears in the list after one where the station has just
disassociated from) for EAPOL processing.
Fix this by checking the WLAN_STA_ASSOC flag before assuming the right
hapd context was found for the given station.
Signed-hostap: David Bird <dbird@powercloudsystems.com>
intended-for: hostap-1
This allows the P2P_PEER command to be used to fetch the list of
secondary device types that each P2P peer has advertised.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
This uses the recent cfg80211 changes to allow SAE authentication to be
implemented with the nl80211 driver interface.
Signed-hostap: Jouni Malinen <j@w1.fi>
P2P includes two use cases where one of the devices is going to start a
group and likely change channels immediately after processing a frame.
This operation may be fast enough to make the device leave the current
channel before the peer has completed layer 2 retransmission of the
frame in case the ctrl::ack frame was lost. This can result in the peer
not getting TX status success notification.
For GO Negotiation Confirm frame, p2p_go_neg_conf_cb() has a workaround
that ignores the TX status failure and will continue with the group
formation with the assumption that the peer actually received the frame
even though we did not receive ctrl::ack. For Invitation Response frame
to re-invoke a persistent group, no such workaround is used in
p2p_invitation_resp_cb(). Consequently, TX status failure due to lost
ctrl::ack frame results in one of the peers not starting the group.
Increase the likelihood of layer 2 retransmission getting acknowledged
and ctrl::ack being received by waiting a short duration after having
processed the GO Negotiation Confirm and Invitation Response frames for
the re-invocation case. For the former, use 20 ms wait since this case
has been worked around in deployed devices. For the latter, use 50 ms
wait to get even higher likelihood of getting ctrl::ack through since
deployed devices (and the current wpa_supplicant implementation) do not
have a workaround to ignore TX status failure.
20 ms is long enough to include at least couple of retries and that
should increase likelihood of getting ctrl::ack through quite a bit. The
longer 50 ms wait is likely to include full set of layer 2 retries.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 458cb30191 broke LEAP since it
rejects EAP-Success packet that is used within LEAP and this frame does
not have a payload. Fix LEAP by relaxing the generic EAP packet
validation if LEAP has been negotiated.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Previously, WPS Registrar allowed multiple wildcard PINs to be
configured. This can get confusing since these PINs get assigned to any
Enrollee that does not have a specific PIN and as such, cannot really be
used with different PIN values in reasonable ways. To avoid confusion
with multiple enabled PINs, invalidate any previously configured
wildcard PIN whenever adding a new one.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Extend the wpa_cli wps_pin command to support specification of the PIN
expiration time in seconds similarly to hostapd_cli wps_pin command when
using wpa_supplicant for AP mode (including P2P GO).
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
By default, P2P is enabled globally for all virtual interfaces and this
makes wpa_supplicant include WSC and P2P IEs in Probe Request frames for
all scans even if this is for a non-P2P station connection to speed up
device discovery. If an interface is dedicated for non-P2P station mode
operations, it is now possible to disable addition of WSC and P2P IEs
into Probe Request frames with a per-interface p2p_disabled parameter.
This can be set either in the configuration file (p2p_disabled=1) or at
run time ("wpa_cli -i wlan0 set p2p_disabled 1"). Unlike the previous
mechanism ("wpa_cli p2p_set disabled 1"), the new parameter changes the
behavior only for the specified interface while other interfaces
continue to follow the global P2P enabled/disabled state.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
The bssid and priority parameters in a network block do not have any
effect on the validity of a PMKSA cache entry, so avoid flushing the
PMKSA cache when only these parameters are changed. This is mainly
to allow forced roaming or network selection changes without causing
a disconnection if the changes are done during RSN association that
used EAP.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Add a function to filter out known cases of disconnection during 4-way
handshake that are caused by something else than mismatch in PSK. This
commit adds the case where the local end determines a mismatch in
WPA/RSN element between Beacon/Probe Response frames and EAPOL-Key msg
3/4.
This can avoid some potentially confusing "WPA: 4-Way Handshake failed -
pre-shared key may be incorrect" ctrl_iface messages.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
While the existing code already addresses TLS Message Length validation
for both EAP-TLS peer and server side, this adds explicit checks and
rejection of invalid messages in the functions handling reassembly. This
does not change externally observable behavior in case of EAP server.
For EAP peer, this starts rejecting invalid messages instead of
addressing them by reallocating the buffer (i.e., ignoring TLS Message
Length in practice).
Signed-hostap: Jouni Malinen <j@w1.fi>
EAP-TLS/PEAP/TTLS/FAST server implementation did not validate TLS
Message Length value properly and could end up trying to store more
information into the message buffer than the allocated size if the first
fragment is longer than the indicated size. This could result in hostapd
process terminating in wpabuf length validation. Fix this by rejecting
messages that have invalid TLS Message Length value.
This would affect cases that use the internal EAP authentication server
in hostapd either directly with IEEE 802.1X or when using hostapd as a
RADIUS authentication server and when receiving an incorrectly
constructed EAP-TLS message. Cases where hostapd uses an external
authentication are not affected.
Thanks to Timo Warns for finding and reporting this issue.
Signed-hostap: Jouni Malinen <j@w1.fi>
intended-for: hostap-1
This replaces the previously used bogus test data in SAE messages with
the first real field. The actual SAE authentication mechanism is still
missing and the Scaler, Element, and Confirm fields are not included.
Signed-hostap: Jouni Malinen <j@w1.fi>
If wpa_s->current_ssid is not set (e.g., after disconnection that
did not result in immediate group removal), an incorrect group could
have been removed since the network block iteration here could select
the network block that is used to store persistent group credentials.
Fix this by verifying that disabled != 2 to avoid picking the network
block that could not have been the temporary P2P group.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
