There is a race condition in GO Negotiation Request frame sending and
processing that may end up with both devices sending GO Negotiation
Response. This response frame was previously accepted even if a response
had already been sent. This could result in two GO Negotiation Confirm
frames being exchanged and consequently, with two separate GO
Negotiations completing concurrently. These negotiations could result in
getting mismatching parameters (e.g., both device could believe it was
the GO).
Fix this by ignoring GO Negotiation Response from the peer if twe have
already sent a GO Negotiation Response frame and we have the higher P2P
Device Address. This is similar to the rule used to determine whether to
reply to GO Negotiation Request frame when Request was already sent,
i.e., the same direction of GO Negotiation is maintained here to enforce
that only the negotiation initiated by the device with smaller P2P
Device Address is completed.
Signed-hostap: Jouni Malinen <j@w1.fi>
intended-for: hostap-1
If both peers initiate GO Negotiation at about the same time, it is
possible for the GO Negotiation Request frame from the peer to be
received between the local attempt to send the GO Negotiation Request
and TX status event for that. This could result in both devices sending
GO Negotiation Response frames even though one of them should have
skipped this based which device uses a higher MAC address.
Resolve this race by incrementing go_neg_req_sent when p2p_send_action()
returns success instead of doing this from the TX status callback. If
the frame is not acknowledged, go_neg_req_sent is cleared in TX status
handler.
Signed-off-by: Neeraj Garg <neerajkg@broadcom.com>
Stop the connection attempt if GO Negotiation Confirm is not received
within 100 ms of the GO Negotiation Response getting acknowledged.
Previously, we would have continued trying to connect to the peer even
in this case which could result in confusing second GO Negotiation
Request frame and unnecessarily long wait before indicating failure.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 361cdf3400 changed the way the
group formation timeout is used on P2P client. However, it resulted in
clearing the timeout on the GO when the GO started beaconing. This is
not correct since the timeout is supposed to be maintained until at
least the completion of the WPS provisioning step. Fix this regression
by clearing the timeout here only in the case we are not GO in group
formation phase.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
It is possible for wpa_state to be WPA_DISCONNECTED when a new scan is
scheduled to be started in the future. If wpas_wps_cancel() gets called
(e.g., through control interface wps_cancel or a P2P group removal)
while in that state, the WPS operation (scan request and special network
block) were not removed. Fix this by clearing the WPS operations both in
WPA_SCANNING and WPA_DISCONNECTED states.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
intended-for: hostap-1
The previous version set scan_res_handler unconditionally and then
cleared it if scan request failed. This can result in incorrect clearing
of the handler to NULL for a previously started scan that has not yet
completed. This can make p2p_find command fail to use the
start-after-scan-completion mechanism in some cases. Fix this by setting
scan_res_handler properly after having verified that the driver command
for starting the scan was successful.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
intended-for: hostap-1
The GO Negotiation Confirm frame doesn't need to be sent with a wait
since we don't expect a response to it.
Signed-hostap: Johannes Berg <johannes.berg@intel.com>
Perform addition additional scan runs on the operating channel of the GO
(if known from previous scan results) and fall back to initiate the PD
for GO Negotiation if these additional scans do not detect the peer as
GO.
Signed-hostap: Sunil Dutt Undekari <duttus@codeaurora.org>
Concurrent Operation bit was not set for GO even if the device
supports concurrent operations. Make sure the Device Capability
value is consistent with other P2P use cases by using the value
determined in p2p_init().
Signed-hostap: Masashi Honma <masashi.honma@gmail.com>
Commit 361cdf3400 extended the use of
group formation timeout for the way handshake, but the registration was
done on the group_interface while the cancellation was done on the
parent interface. Fix the registration to set the eloop timeout on
parent to address potential issues when using a separate group
interface.
Signed-hostap: Nirav Shah <nirav.j2.shah@intel.com>
When looking for a BSS to eject due to too many entries, never
pick one that is in use. Otherwise, we run the risk of having
pointers to freed data.
Signed-hostap: Paul Stewart <pstew@chromium.org>
Send an "EAP" signal via the new DBus interface under various
conditions during EAP authentication:
- During method selection (ACK and NAK)
- During certificate verification
- While sending and receiving TLS alert messages
- EAP success and failure messages
This provides DBus callers a number of new tools:
- The ability to probe an AP for available EAP methods
(given an identity).
- The ability to identify why the remote certificate was
not verified.
- The ability to identify why the remote peer refused
a TLS connection.
Signed-hostap: Paul Stewart <pstew@chromium.org>
In certain cases like PBC session overlap it appears to be possible that
an interface is removed due to an event handled on it, this leads to
list corruption. Use safe iteration to prevent this issue.
Reported-by: Nirav Shah <nirav.j2.shah@intel.com>
Reported-by: Neeraj Kumar Garg <neerajkg@broadcom.com>
Signed-hostap: Johannes Berg <johannes.berg@intel.com>
intended-for: hostap-1
Use an empty IPv4 packet as the keep-alive packet for WNM BSS max idle
period mechanism. This is not really the best possible frame to use for
this, but for now, it can do until a more suitable frame is figured out
for this (e.g., special LLC header to indicate link test purposes).
Signed-hostap: Jouni Malinen <j@w1.fi>
When hostapd (or wpa_supplicant AP mode) limits the maximum number
of STA entries with a driver that implements SME, the driver needs
to be notified of the failed STA addition. Disassociate the STA if
the driver notifies of an association that would go beyond the
maximum number of STAs.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
If WNM is enabled for the build (CONFIG_WNM=y), add BSS max idle period
information to the (Re)Association Response frame from the AP and parse
this information on the station. For SME-in-wpa_supplicant case, add a
timer to handle periodic transmission of the keep-alive frame. The
actual request for the driver to transmit a frame is not yet
implemented.
Signed-hostap: Jouni Malinen <j@w1.fi>
Commit aa9bb7644b improved robustness
of p2p_connect-auto mechanism by using older scan results to help in
determination whether the peer was operating a GO. Improve this by
accepting new GO information from scan-for-WPS-provisioning results
even if the GO is not yet ready for WPS.
In addition, fix an issue where Provision Discovery exchange timeout
could have left offchannel TX operation in progress when the fallback
was used.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
When sending an offchannel frame (mainly, P2P Invitation Request), the
wait_time parameter was hardcoded to 0 for drivers that implement AP
mode SME. This is not correct and can cause problems for drivers that
support offloading of off-channel operations with driver/firmware based
AP SME.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
This is a workaround for interoperability issues with some deployed P2P
implementations that require a Provision Discovery exchange to be used
before GO Negotiation. The new provdisc parameter for the p2p_connect
command can be used to request this behavior without having to run a
separate p2p_prov_disc command.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Excessive debug level can now be used to dump the IEs received in scan
results from both Beacon and Probe Response frames. This makes it easier
to debug issues where the driver (or well, more likely cfg80211) does
not update the IEs based on Beacon frames if even a single Probe
Response frame has been received.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Previously, wpa_supplicant did not specify BSSID to any connection
request if the driver indicated that it will take care of BSS selection.
This is fine for most use cases, but can result to issues if the network
block has an explicit bssid parameter to select which BSS is to be used.
Fix this by setting BSSID and channel when the network block includes the
bssid parameter even if the driver indicates support for BSS selection.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
intended-for: hostap-1
If WPS commands are used with a specific BSSID instead of wildcard,
the BSSID that was supposed to be used only for the provisioning step
may end up getting copied to the network block that gets provisioned
based on the WPS credentials. Fix this by clearing ssid->bssid_set
when creating the network block by replacing the block used for WPS
provisioning.
This issue could show up with the provisioned network not getting
selected properly with APs that have multiple radios. Depending on
the driver, this could result in only a single one of the available
BSSes being available or the connection failing completely.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
intended-for: hostap-1
The nl80211 driver interface does not allow 128-bit WEP to be used
without a vendor specific cipher suite and no such suite is defined for
this purpose. Do not accept WEP key length 16 for nl80211 driver
interface forn ow. wext-interface can still try to use these for
backwards compatibility.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
I think this should fix the following valgrind complaint:
==1972== Source and destination overlap in memcpy(0x5181708, 0x5181708, 16)
==1972== at 0x4A073BA: memcpy (mc_replace_strmem.c:602)
==1972== by 0x45872B: wpas_wps_set_uuid (wps_supplicant.c:1116)
==1972== by 0x4599EC: wpas_wps_update_config (wps_supplicant.c:1747)
==1972== by 0x4C8DB0: wpa_supplicant_update_config (wpa_supplicant.c:3090)
==1972== by 0x4C3E5E: wpa_supplicant_reload_configuration (wpa_supplicant.c:746)
==1972== by 0x4B8B37: wpa_supplicant_ctrl_iface_process (ctrl_iface.c:4082)
==1972== by 0x4BA39C: wpa_supplicant_ctrl_iface_receive (ctrl_iface_unix.c:168)
==1972== by 0x4114D4: eloop_sock_table_dispatch_table (eloop.c:335)
==1972== by 0x411541: eloop_sock_table_dispatch (eloop.c:352)
==1972== by 0x41200D: eloop_run (eloop.c:766)
==1972== by 0x4C8B43: wpa_supplicant_run (wpa_supplicant.c:3010)
==1972== by 0x4D44AD: main (main.c:286)
Signed-hostap: Ben Greear <greearb@candelatech.com>
intended-for: hostap-1
The P2P implementation assumes that the first wpa_s interface instance
is used to manage P2P operations and the P2P module maintains a pointer
to this interface in msg_ctx. This can result in issues (e.g., use of
freed memory) when the management interface is removed. Fix this by
deinitializing global P2P data if the interface that created it is
removed. This will disable P2P until the next interface is added.
Signed-hostap: Jouni Malinen <j@w1.fi>
intended-for: hostap-1
If the AP indicates support for Hotspot 2.0, show this in the
ctrl_iface commands for BSS table and status.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
There are no plans to continue making the Windows binary package for
wpa_supplicant for 1.x or newer releases, so remove the scripts and
notes that are related to this.
Signed-hostap: Jouni Malinen <j@w1.fi>
If PD Request includes P2P Group ID, verify that the specified
group matches with a group we are currently operating. If no match
is found, reject the PD Request for join-a-group case.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
This can be used with P2P management operations that need to verify
whether the local device is operating a specific group based on
P2P Group ID attribute from a peer.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Previusly the peer was assumed to not be operating a GO if the BSS entry
for it was not updated in the single scan run started by
p2p_connect-auto. This is not very robust since a scan may miss the peer
if either a Probe Request or Probe Response frame is lost. Improve
robustness by assuming the peer is still operating the GO and starting
the join operation. If the GO is not found during PD-for-join or the
single-channel scans during the join, fall back to GO Negotiation.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 349b213cc8 added a separate
callback prov_disc_fail() for indicating PD failures, but it left the
Provision Discovery Response handler to call both callbacks in case the
peer rejected the PD. Commit f65a239ba4
added ctrl_iface event for PD failures. This combination can result in
two ctrl_iface events in the peer rejecting a PD case. Clean this up by
only indicating the failure event.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
Some drivers may accept the remain-on-channel command, but instead of
indicating start event for remain-on-channel, just indicate that the
operation has been canceled immediately. This could result in continuous
loop of search/listen states with very limited time to do anything else
in wpa_supplicant if the scan command is also completed quickly (e.g.,
if the driver is unable to scan other channels than the current
operating channel).
As a workaround, do not start the next step (search) in P2P device
discovery if this type of rejection of listen operation is detected.
This gives some more time for wpa_supplicant to handle whatever else
may be needed at to be done at the same time and reduces the amount
of CPU used in a loop that does not really work correctly from the
view point of being discoverable.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
intended-for: hostap-1
Do not try to associate with a network that has an invalid or incomplete
configuration because the association or at least data connection would
fail anyway. This commits adds a common function for checking whether a
network block is disabled to make it easier to check network blocks
without having to reject them during configuration file parsing (which
would prevent wpa_supplicant from starting). The only additional check
added in this commit is to verify the WEP key length. Similar checks for
other parameters can be added in future commits.
Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>
