When considering to roam to a different BSS and the connection
is an MLD connection, do not roam to a BSS which is already
included in the MLD connection.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
When a BSS is removed, the flow continues without actually
waiting for the AP to be stopped. This is racy in flows that
actually expect the AP to be stopped, e.g., test_ap_bss_add_remove().
Try to mitigate such cases by adding a short sleep after the
AP is removed.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
wpa_s->connect_without_scan could have been left pointing to invalid
network when a network profile was removed. It seems to be possible for
this to happen in some hwsim test case scenarios under specific timing,
but the exact reason for this is not clear. In any case, this pointer
needs to be cleared.
Signed-off-by: Jouni Malinen <j@w1.fi>
This is needed to avoid unexpected behavior if a previously executed
test case has left a BSS entry with Interworking emabled into the case.
Signed-off-by: Jouni Malinen <j@w1.fi>
The functions that determine the length of the RNR information and that
build the actual RNR need to use the same conditions for skipping BSSs.
Use a shared helper function for this to avoid having to maintain two
copies of the same implementation and the risking those getting out of
sync.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
RNR formation for co-location or MLO did not work as expected. Fix this.
For example, during co-location, if the BSS is also its ML partner
there is no need to include a separate TBTT for it.
Also, during co-location, if the BSS is not its partner but it is ML
capable, the TBTT length should be 16 bytes and it should include the
MLD Parameters for it in the RNR.
During co-location, for a given Neighbor AP (operating on a given
channel and op-class) if it has BSSs which are ML capable as well as
BSSs which are not, there should be two Neighbor AP Info present: one
indicating TBTT length as 13 bytes and one indicating TBTT info length
as 16 bytes.
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Include link ID and partner link details in the STATUS command output
for AP MLDs.
The details would be seen as below for an AP MLD interface:
$ hostapd_cli -i wlan0 status | grep link
num_links=1
link_id=0
link_addr=AA:BB:CC:DD:EE:FF
$ hostapd_cli -i wlan1 status | grep link
num_links=2
link_id=0
link_addr=AA:BB:CC:DD:EE:FF
partner_link[1]=AA:BB:CC:DD:EE:AA
Signed-off-by: Harshitha Prem <quic_hprem@quicinc.com>
Co-developed-by: Manish Dharanenthiran <quic_mdharane@quicinc.com>
Signed-off-by: Manish Dharanenthiran <quic_mdharane@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Whenever a BSS was set up,hostapd flushed all stations via the flush()
driver operation which maps to NL80211_CMD_DEL_STATION in the nl80211
interface. However, in case of MLO, a station could have been connected
to other links by the time this link is coming up. Since link ID was not
passed to flush(), all those stations entries were also removed in the
driver which is wrong.
Include the link ID along with the command in AP MLD so that the driver
can use this link ID and flush only the stations that use the passed
link ID as one of their links.
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
This is needed for MLO group rekeying.
Signed-off-by: Rameshkumar Sundaram <quic_ramess@quicinc.com>
Co-developed-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Group rekeying was not supported for ML stations when non-association
link initiates a group rekey. Support this by arming the group key rekey
timer on one of the affiliated links and whenever this timer fires,
rekey group keys on all the affiliated links.
Signed-off-by: Rameshkumar Sundaram <quic_ramess@quicinc.com>
Co-developed-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Calculate links specific MLO GTK/IGTK/BIGTK KDE lengths based on
corresponding cipher and key instead of taking length of one link and
multiplying it by no of associated links. This is needed since the group
ciphers might be different between the affiliated links.
Signed-off-by: Rameshkumar Sundaram <quic_ramess@quicinc.com>
Co-developed-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
This makes it easier to extend the design for MLO group rekeying.
Signed-off-by: Rameshkumar Sundaram <quic_ramess@quicinc.com>
Co-developed-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Adil Saeed Musthafa <quic_adilm@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Make the code more readable by using a define for the PN length to avoid
potential confusion of this 6 octet length with the MAC address length.
In addition, Use ETH_ALEN more consistently for the latter.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Add required ML specific members in struct wpa_authenticator and struct
wpa_state_machine to maintain self and partner link information.
Maintain state machine object in all associated link stations and
destroy/remove references from the same whenever link stations are
getting removed.
Increase the wpa_group object reference count for all links in which ML
station is getting associated and release the same whenever link
stations are getting removed.
Signed-off-by: Rameshkumar Sundaram <quic_ramess@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
All links were iterated over during processing ML info in Association
Request frame. However, the association link info will not be present in
the ML info and hence the following debug print is observed during ML
association (assoc link is 1):
MLD: No link match for link_id=1
Skip processing for the association link to avoid this.
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Whenever there is a beacon update for any one of the affiliated link,
all the other partner links' beacon should be refreshed.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Add link ID support into EAPOL RX handler so that the events can
be routed to the appropriate link BSSs.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Add link ID support into EAPOL TX status handler so that the events can
be routed to the appropriate link BSSs.
Check each BSS's other partner link BSS STA list as well in
hostapd_find_by_sta() to support this.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
hostapd_eapol_tx_status() function is used only in drv_callbacks.c.
However, it is defined in ieee802_11.c which is not really the correct
place for it.
Hence, move the function into drv_callbacks.c and make it static.
No functionality changes.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Control port TX status events were handled on drv's first BSS
only. However, to support multiple MLDs there is requirement to handle
this on a given BSS.
Use the passed BSS instead of always going with drv's first BSS.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Management frame TX status events were handled on drv's first BSS
only. However, to support multiple MLDs there is requirement to handle
this on a given BSS.
Use the passed BSS instead of always going with drv's first BSS.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Pass ctx in mlme_event_mgmt(). This will help in routing the event
properly to the link BSS.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
The get_hapd_bssid() function matched the given BSSID in all BSSs of its
own interface. However with MLO, there is requirement to check its own
partner BSS at least.
Compare the BSS's link partners as well and if the specified link ID
matches the link ID of the partner, return the BSS.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Modify necessary helper functions to support multiple BSS support for
MLO to make the changes scalable.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
AP MLD was added with an assumption of only a single BSS per link in the
hostapd configuration. This needs to be extended when a cohosted ML BSS
exist in the same configuration.
Extend the support for cohosted BSSs. This is required for MBSSID MLO
support as well.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Authenticator state machine ML info was set only when it was created.
However, if the association is tried again, the state machine will
already exist and hence the ML info will not be refreshed. This leads to
an issue where if in the subsequent association request, the MLD info is
different than the old info, validation of it will fail.
Fix this issue by refreshing the authenticator state machine's ML info
every time association request is handled.
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
This became unused, so remove the argument from this function, all its
callers, and from places that became unused with these changes.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
The nl80211 driver interface function mlme_event_mgmt_tx_status(),
filled in link_id only if the frame was the last transmitted on the
whole drv (driver) level. With co-hosted MLDs, there could be cases
where multiple frames are sent out by various interfaces (BSS) under the
same drv. Now while handling the TX status, only one interface will get
the proper link_id. Rest will get -1 and the event will be routed to the
first BSS always. If the frame was not sent from the first BSS this
leads to possibility of the frame getting dropped.
Hence to make the underlying link identification easier, modify
authentication and association frames to be always sent with the link
address as A1 and A3 for ease of TX status handling.
Signed-off-by: Sriram R <quic_srirrama@quicinc.com>
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
The changes in nl80211 to get rid of the libnl3-route dependency are not
sufficient to fully remove the depency from other parts of the code.
Revert the makefile related changes from that commit to avoid build
issues for cases where CONFIG_FULL_DYNAMIC_VLAN=y and
CONFIG_VLAN_NETLINK=y are used without CONFIG_DRIVER_MACSEC_LINUX=y
pulling in the needed library.
Fixes: a210fdb1c7 ("nl80211: Rewrite neigh code to not depend on libnl3-route")
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Introduce an attribute QCA_WLAN_VENDOR_ATTR_CONFIG_KEEP_ALIVE_INTERVAL
in QCA_NL80211_VENDOR_SUBCMD_SET_WIFI_CONFIGURATION to configure
station's keep-alive interval to the driver/firmware. This can be used
to resolve kickout issues from APs which kick out STAs before the BSS
maximum idle period expires.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Add a QCA vendor command for registering NL80211_CMD_GET_STATION
response as a unicast event when there is a NL80211_CMD_GET_STATION
request from any userspace module.
The driver will send the unicast events with the same netlink port ID
which is used by userspace application for sending the registration
command. If multiple registration commands are received with different
netlink port IDs, the driver will send unicast event with each netlink
port ID separately.
Userspace application can deregister the unicast events with disable
configuration. The registrations will be removed automatically by the
driver when the corresponding netlink socket is closed.
This will help avoid multiple NL80211_CMD_GET_STATION requests from
different userspace applications in short span. The userspace
application which registers for the unicast event can avoid sending
NL80211_CMD_GET_STATION request again if the response is available with
a recently received unicast event.
Signed-off-by: Veerendranath Jakkam <quic_vjakkam@quicinc.com>
Add support for TWT responder for AP operating in HT and VHT modes by
introducing a new configuration parameter ht_vht_twt_responder. When
this is enabled, TWT responder mode support in HT and VHT modes is
enabled if the driver supports this and is disabled otherwise.
Signed-off-by: Manaswini Paluri<quic_mpaluri@quicinc.com>
Add a feature flag to indicate driver support for TWT responder for AP
operating in HT and VHT modes.
Signed-off-by: Manaswini Paluri<quic_mpaluri@quicinc.com>
At least some of the previous versions have expired, so need to re-sign
these to avoid EAP test case failures. This contains updates from
running tests/hwsim/auth_server/update.sh.
Signed-off-by: Jouni Malinen <j@w1.fi>
i802_sta_deauth() already has the link_id passed to it in its arguments.
Use that to pass it down to send MLME handler as well.
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Function nl80211_link_add() created the link add netlink message on drv
basis which in turn always uses the drv's first BSS. To support link add
for various other interfaces, use the per-BSS function to create the
netlink message.
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
for_each_mld_link() macro used three nested for loops. Since now the
affliated links are linked together via a linked list, the logic can be
improved by using dl_list_for_each() macro instead which uses one for
loop.
Modify for_each_mld_link() macro to use dl_list_for_each() instead.
Signed-off-by: Aditya Kumar Singh <quic_adisi@quicinc.com>
Add a test behaving like an Asus RT-AC53 with firmware
3.0.0.4.380_10760-g21a5898, which (in some cases?) can have an ECSA
element stuck in the probe response, when the channel switch is long
finished.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Add a few tests to validate what happens with connections
while an AP is doing CSA:
- quiet to diff channel (shouldn't connect)
- quiet to same channel (shouldn't connect)
- non-quiet to diff channel (shouldn't connect)
- non-quiet to same channel (should connect)
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Add support for additional (vendor) elements to be added
to only Probe Response frames, for testing.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
If the build path is long, the contents of the _DIRS variable can be
very long, since it repeats the same directories very often. In some
cases, this has triggered an "Argument list too long" build error.
Reported-by: Robert Marko <robimarko@gmail.com>
Suggested-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
