wpa_group_update_count and wpa_pairwise_update_count can now be used to
set the GTK and PTK rekey retry limits (dot11RSNAConfigGroupUpdateCount
and dot11RSNAConfigPairwiseUpdateCount). Defaults set to current
hardcoded value (4).
Some stations may suffer from frequent deauthentications due to GTK
rekey failures: EAPOL 1/2 frame is not answered during the total timeout
period of currently ~3.5 seconds. For example, a Galaxy S6 with Android
6.0.1 appears to go into power save mode for up to 5 seconds. Increasing
wpa_group_update_count to 6 fixed this issue.
Signed-off-by: Günther Kelleter <guenther.kelleter@devolo.de>
Issue:
------
The test setup has 2 peers running MACsec in PSK mode, Peer A with
MAC address higher than MAC Address of peer B. Test sequence is
1. Peer B starts with actor_priority 255
2. Peer A starts with priority 16, becomes key server.
3. Peer A stops..
4. Peer A restarts with priority 255, but because of the stale values
participant->is_key_server(=TRUE) and participant->is_elected(=TRUE)
it continues to remain as Key Server.
5. For peer B, key server election happens and since it has lower MAC
address as compared to MAC address of A, it becomes the key server.
Now we have 2 key servers in CA and is not correct.
Root-cause & fix:
-----------------
When number of live peers become 0, the flags such lrx, ltx, orx,
otx, etc. need to be cleared. In MACsec PSK mode, these stale values
create problems while re-establishing CA.
Signed-off-by: Badrish Adiga H R <badrish.adigahr@gmail.com>
API ieee802_1x_mka_decode_dist_sak_body() wrongly puts
participant->to_use_sak to TRUE, if Distributed SAK Parameter Set of
length 0 is received. In MACsec PSK mode, this stale incorrect value can
create problems while re-establishing CA. In MACsec PSK mode, CA goes
down if interface goes down and ideally we should be able to
re-establish the CA once interface comes up.
Signed-off-by: Badrish Adiga H R <badrish.adigahr@gmail.com>
This patch hides a compiler warning:
osu_client.c: In function ‘cmd_osu_select’:
osu_client.c:2200:2: warning: ‘osu_count’ may be used uninitialized in this function [-Wmaybe-uninitialized]
for (i = 0; i < osu_count; i++) {
^
osu_count is actually initialized in parse_osu_providers() if non-NULL
value is returned.
Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
Some channels fail to be set, when falling back to 20 MHz, due to
remaining VHT info of center freq. As we are going to 20 MHz, reset the
VHT center frequency segment information as well.
Signed-off-by: Eduardo Abinader <eduardoabinader@gmail.com>
Relying on the group_capab from the P2P peer information can result in
improper information on whether the peer is currently operating as a GO.
However, there is a known implementation in Android that does this.
To reduce issues from this misuse in upper layer to try to determine
whether a specific peer is operationg a group, override the group_capab
value in P2P_PEER output with 0 if there are no BSS entries with the
peer P2P Device as a GO. This is not a perfect information since there
may not have been a recent scan on all channels, but this results in
less issues than trying to decide between new group formation and
joining an existing group based on stale or incorrect information.
Since no upper layer application is really supposed to use the
group_capab field value in P2P_PEER command, this change should not
cause any impact for properly design components and the possibility of
regressions is limited to cases that are already known to work
incorrectly in number of identifiable cases.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 8d1e693186 ('Use estimated
throughput to avoid signal based roaming decision') added a check for
the current BSS estimated throughput being significantly higher than the
selected BSS estimated throughput. However, this case for skipping a
roam used "return 1" which actually allows the roam. Fix this by
returning 0 in this case.
Signed-off-by: Jouni Malinen <j@w1.fi>
Following commit a70cd0db87 ('Don't
register for Beacon frames for IEEE 802.11ad AP'),
nl80211_get_wiphy_data_ap() is unconditionally called when starting AP.
This function tries to register for Beacon frames RX which fails for
some driver which don't support such registration and do not need it in
case the driver implements AP mode SME functionality.
Fix this by conditionally calling nl80211_get_wiphy_data_ap() like prior
to commit a70cd0db87.
Signed-off-by: Dedy Lansky <qca_dlansky@qca.qualcomm.com>
Commit 91d91abf6f ('FILS: DHCP relay for
HLP requests') added steps that are conditional on sta->fils_hlp_resp
being non-NULL. One of these cases within send_assoc_resp() was properly
protected from sta == NULL error case (that is now possible after a
recent DMG change), but the first one was not. A DMG error case in a
CONFIG_FILS=y build could have hit a NULL pointer dereference here. Fix
this by verifying sta != NULL more consistently.
Signed-off-by: Jouni Malinen <j@w1.fi>
To be more consistent with existing hwaddr_aton() and hexstr2bin()
callers, check the return values in this test command.
Signed-off-by: Jouni Malinen <j@w1.fi>
While the key derivation steps are not expected to fail, this was
already done on the AP side, so do the same in the STA side.
Signed-off-by: Jouni Malinen <j@w1.fi>
P802.11i/D3.0 described the Key Length as having value 16 for the group
key handshake. However, this was changed to 0 in the published IEEE Std
802.11i-2004 amendment (and still remains 0 in the current standard IEEE
Std 802.11-2016). We need to maintain the non-zero value for WPA (v1)
cases, but the RSN case can be changed to 0 to be closer to the current
standard.
Signed-off-by: Jouni Malinen <j@w1.fi>
While the IEEE Std 802.11-2016 is still indicating that GNonce would be
exchanged in EAPOL-Key messages (see, e.g., Figure 12-52 showing the
Send EAPOL-Key operation in the REKEYNEGOTIATING state or the sample
group key handshake in Figure 12-47), there are also examples of
describing this field as having value zero (e.g., 12.7.7.2 Group key
handshake message 1).
GNonce is used only with the Authenticator and the Supplicant does not
have any use for it, so it is better not to expose that internal value.
Hardcode the Key Nonce field to 0 in EAPOL-Key group message 1/2.
Signed-off-by: Jouni Malinen <j@w1.fi>
sm->Pair needs to be initialized to TRUE since unicast cipher is
supported and this is an ESS. However, the normal place for setting this
(WPA_PTK::INITIALIZE) is skipped with using FT protocol or FILS
authentication, so need to do that separately when forcing PTKINITDONE.
Signed-off-by: Jouni Malinen <j@w1.fi>
There is now one fewer direct allocation call in this function, so the
counters need to be updated to avoid test failures.
Signed-off-by: Jouni Malinen <j@w1.fi>
The local custom version of allocating and initializing struct
hostapd_data within wpa_supplicant_mesh_init() is problematic. This has
already missed couple of initialization steps that are required. Instead
of trying to remember to keep this up to date, use
hostapd_alloc_bss_data() so that there is only one place for this
initialization.
This is fixing a recent issue where FILS HLP started using
hapd->dhcp_server and expected that to be initialized to -1. For the
mesh case, that did not happen and when removing the interface, the FILS
HLP implementation ended up unregistering eloop socket for
hapd->dhcp_server (= 0). This could result in missing socket callbacks
for an arbitrary socket.
Signed-off-by: Jouni Malinen <j@w1.fi>
This wpa_drv_if_remove() call was previously modified to fix a different
issue, but that fix resulted in unconditional use of treed memory here
(wpa_supplicant_deinit_iface() frees wpa_s). Make a local copy of
wpa_s->parent to be able to use it after wpa_s is freed. The
mesh_if_created case has wpa_s->parent != wpa_s, so this should be
sufficient way of handling the wpa_drv_if_remove() call here.
Signed-off-by: Jouni Malinen <j@w1.fi>
Use the defines for vht_oper_chwidth values more consistently in
hostapd_set_freq_params() to make this more readable.
Signed-off-by: Jouni Malinen <j@w1.fi>
This commit introduces the vendor command to set the trace level
for the respective QCA host driver module.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The hostapd processing of the AES-SIV AAD was incorrect. The design for
the AAD changed between P802.11ai/D7.0 and D8.0 from a single vector
with concatenated data to separate vectors. The change in the
implementation had missed the change in the aes_siv_decrypt() call for
the num_elem parameter. This happened to work with the mac80211
implementation due to a similar error there.
Fix this by using the correct numbers of vectors in the SIV AAD so that
all the vectors get checked. The last vector was also 14 octets too long
due to incorrect starting pointer, so fix that as well. The changes here
are not backwards compatible, i.e., a similar fix in the Linux mac80211
is needed to make things interoperate again.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
While RFC 5295 uses "8" as the value to use in the length field in KDF
context when deriving EMSKname, it is clearer to use the macro defining
EMSKname as the value since the KDF design in RFC 5295 encodes the
length of the derived data in octets in that part of the context data.
This change is just making the implementation easier to understand while
not actually changing the behavior.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Unlike the EMSKname and rRK derivations, rIK derivation is actually
using the "optional data" component in the context data (see RFC 5295).
RFC 6696 defines that optional data to be the cryptosuite field for rIK.
This was missing from the previous implementation and that resulted in
incorrect rIK being derived.
In addition, the rIK Label string does not actually include the "EAP "
prefix in the way as the rRK Label in RFC 6696 does. This would also
have resulted in incorrect rIK value.
Fix rIK derivation by adding the cryptosuite value into the KDF context
data and fixing the label string. This change is not backwards
compatible and breaks all ERP use cases (including FILS shared key
authentication) with older (broken) and new (fixed)
hostapd/wpa_supplicant builds.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The new FILS-HLP-RX control interface event is now used to report
received FILS HLP responses from (Re)Association Response frame as a
response to the HLP requests configured with FILS_HLP_REQ_ADD.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The new dhcp_server configuration parameter can now be used to configure
hostapd to act as a DHCP relay for DHCPDISCOVER messages received as
FILS HLP requests. The dhcp_rapid_commit_proxy=1 parameter can be used
to configure hostapd to convert 4 message DHCP exchange into a 2 message
exchange in case the DHCP server does not support DHCP rapid commit
option.
The fils_hlp_wait_time parameter can be used to set the time hostapd
waits for an HLP response. This matches the dot11HLPWaitTime in IEEE Std
802.11ai-2016.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This is independent functionality from the core IEEE 802.11 management
handling and will increase significantly in size, so it is cleaner to
maintain this in a separate source code file.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Do not depend on undefined behavior with pointer arithmetic when
checking whether there is sufficient room for an option.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
These can be reused for other purposes than just the DHCP snoofing for
Proxy ARP. In addition, use more complete definition of the parameters
based on the current IANA registry.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Introduce definitions for QCA vendor specific subcommands and attributes
for fetching BSS transition status.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
With device_ap_sme disabled, ACL was checked upon authentication
request. In 802.11ad there is no authentication phase so need to check
ACL upon association.
Signed-off-by: Dedy Lansky <qca_dlansky@qca.qualcomm.com>
This verifies that the temporary STA entry timeout limit does not end up
breaking comeback_delay tests with values larger than five seconds.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Previously, the temporary STA session timeout was set to 5 seconds. If
gas_comeback_delay is configured to be longer than 5 seconds, GAS
Comeback Response frame can't include queried information as all pending
data has already been cleared due to session timeout. This commit
resolves the issue by setting session timeout to be larger than
gas_comeback_delay.
Signed-off-by: Daniel Kim <kimdan@qca.qualcomm.com>
This test case was triggering false failures with hostapd build that did
not include TEST_* commands.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This commit introduces QCA vendor commands and the corresponding
attributes to set/get NUD (Network Unreachability Detection) statistics.
The set NUD statistics configures the requisite parameters to the host
driver and thereby triggers the start/stop of collection of statistics.
The get stats fetches the statistics collected in the host driver.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
If vendor_scan_cookie is set to 0 after the scan_abort due to the scan
timeout ends in a cookie mismatch when processing the following
QCA_NL80211_VENDOR_SUBCMD_SCAN_DONE indication. This ends up considering
the scan results as being for an external scan and thus the current
ongoing scan is not removed from the radio_work. Hence, do not reset
this vendor_scan_cookie after the scan abort so that the scan completion
event gets processed properly and vendor_scan_cookie gets cleared at
that point.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
If the keychain holds additional certificates other than the end
certificate, read them into the certificate chain.
Signed-off-by: Paul Stewart <pstew@google.com>
We can only send module specific messages to syslog and not debug
messages printed with wpa_printf. Add an extra command line parameter
'-s' to allow it. The feature is enabled with compile flag
CONFIG_DEBUG_SYSLOG as for wpa_supplicant and behaves in the same manner
as the wpa_supplicant -s command line argument.
Signed-off-by: Wojciech Dubowik <Wojciech.Dubowik@neratec.com>
It was possible to hit this WPA_ASSERT when FST-MANAGER SESSION_REMOVE
command is exececuted when in not-associated state. In
CONFIG_EAPOL_TEST=y builds, this would result in the wpa_supplicant
process being terminated. Convert this WPA_ASSERT to a check that does
not terminate the process, but only rejects the command if wpa_s->bssid
does not match the da argument.
Signed-off-by: Jouni Malinen <j@w1.fi>
Increase the EAPOL RX frame timeout from 100 to 200 ms. This fixes lack
of optimization (i.e., first EAPOL frame dropped) in occasional roaming
and authentication cases on EAP networks if the kernel events can be
reordered and delayed a bit longer.
Signed-off-by: Tomoharu Hatano <tomoharu.hatano@sonymobile.com>
Instead of copying the struct wpa_auth_callbacks, just keep a pointer to
it, keep the context pointer separate, and let the user just provide a
static const structure. This reduces the attack surface of heap
overwrites, since the function pointers move elsewhere.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
