Radar detection event could have resulted in the driver interface
instance getting deinitialized and the related memory freed in the
middle of the loop. This was not an issue when the event was passed only
into a single interface, but it became an issue when the loop tried to
send it to all interfaces. If the driver were removed, that loop check
would have used freed memory. Avoid this by explicitly checking that the
driver interface instance is still valid.
Fixes: f136837202 ("nl80211: Pass wiphy events to all affected interfaces")
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
The TX status handler for DPP Authentication Confirm message might have
resulted in use-after-free if the start of a GAS query were to fail,
e.g., due to being somehow unable to transmit the initial request. Avoid
this by explicitly confirming that the authentication session was not
removed.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
The current connection (struct dpp_connection) might get removed during
the dpp_tcp_send_msg() call, so the code setting the
on_tcp_tx_complete_remove flag needs to check whether that happened to
avoid a potential use-after-free.
Fixes: 33cb47cf01 ("DPP: Fix connection result reporting when using TCP")
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
When removing the bootstrap info for the PB context, all the possible
pointers to that information needs to be cleared to avoid accesses to
freed memory.
Fixes: 37bccfcab8 ("DPP3: Push button bootstrap mechanism")
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
When the privacy protected itentity is used for EAP-SIM/AKA, the buffer
containing the identity was freed just before its use. Fix that by
reordering the operations.
Fixes: 881cb4198b ("EAP-SIM/AKA peer: Simplify identity selection for MK derivation")
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
nl_socket_get_cb() increases cb_refcnf for the cb that is bound to a
socket and as such, nl_cb_put() needs to be used with the returned cb
after having cloned it to avoid leaking memory due to cb_refcnt never
getting back to 0.
Fixes: da0d51fee7 ("nl80211: Use socket cb instead of global->nl_cb in send_and_recv()")
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
The OpenSSL 3.x version of crypto_ec_key_parse_priv using
OSSL_DECODER_CTX missed the call to free the context. Fix it to avoid a
memory leak.
Fixes: 4f4479ef9e ("OpenSSL: crypto_ec_key_parse_{priv,pub}() without EC_KEY API")
Signed-off-by: Jouni Malinen <j@w1.fi>
Use smaller variables when possible and reorder the variables to avoid
unnecessary padding. This drops struct mld_link_info size from 64 to 48
bytes and removes 240 bytes from struct sta_info.
Signed-off-by: Jouni Malinen <j@w1.fi>
This function has been used only within beacon.c and it got accidentally
converted to a global function.
Fixes: 6b5e00a80e ("AP: Use a struct for Probe Response generation in/out params")
Signed-off-by: Jouni Malinen <j@w1.fi>
Replace the fixed length maximum buffer size for STA profile with
dynamically allocated buffers for active links. This reduces struct
mld_link_info size by almost 16 kB and drops the per-STA information in
struct sta_info to a more reasonable size to avoid the almost 10x
increase from MLO support.
In addition, free the resp_sta_profile buffers as soon as the ML element
has bee generated for (Re)Association Response frame since those buffers
are not needed after that.
Signed-off-by: Jouni Malinen <j@w1.fi>
The singleton RADIUS client design did not address the deinit path
properly. Since hapd->radius could be shared with another links, the
pointer on all those other links needs to be cleared before freeing the
RADIUS client context. Without this, deinit path could have ended trying
to use freed memory when clearing STA entries from other links and
trying to flush any pending RADIUS client messages.
Fixes: a213fee11d ("AP: MLO: Make IEEE 802.1X SM, authserv, and RADIUS client singletons")
Signed-off-by: Jouni Malinen <j@w1.fi>
Allow the Timeout Interval Type field in the Timeout Interval element to
be overridden with a different value for testing purposes to be able to
bypass the association comeback processing in mac80211. This allows the
wpa_supplicant internal functionality to be tested.
Signed-off-by: Jouni Malinen <j@w1.fi>
- CMAC and GMAC modes
- 128-bit and 256-bit modes
- normal BIP and BIP using BCE
- test vectors with minimum and optional additional header elements in
S1G beacon frames
- S1G Beacon Compatibility element in some cases, no other beacon body
components
Signed-off-by: Henry Ptasinski <henry@e78com.com>
Signed-off-by: Andrew Pope <andrew.pope@morsemicro.com>
Signed-off-by: David Goodall <dave@morsemicro.com>
hostapd reads the list of Rx Key Holders from hostapd.conf file.
However, for systems where topology changes dynamically, the update
of RxKHs list is required without reloading the whole configuration.
Introduce a new source of RxKH definition with original syntax:
- rxkh_file - Path to a file containing a list of RxKHs.
In addition, add a control interface command RELOAD_RXKHS to
reload RxKHs definition from the file specified in `rxkh_file`.
This allows hostapd to properly distribute Rx keys even after topology
change (assuming rxkh_file is updated and reload_rxkhs command issued).
Syntax of rxkh_file is the same as extraction of r0kh and r1kh options
from original hostapd.conf file.
```
r0kh=ff:ff:ff:ff:ff:ff * 00112233445566778899aabbccddeeff
r0kh=ff:ff:ff:ff:ff:ff * 00112233445566778899aabbccddeeff
r1kh=00:00:00:00:00:00 00:00:00:00:00:00 00112233445566778899aabbccddeef
r1kh=00:00:00:00:00:00 00:00:00:00:00:00 00112233445566778899aabbccddeef
r1kh=00:00:00:00:00:00 00:00:00:00:00:00 00112233445566778899aabbccddeef
```
Signed-off-by: Dariusz Kopka <dariusz@plume.com>
Smartcard support uses the ENGINE API of OpenSSL, which has been
deprecated as of OpenSSL 3. Rather than migrating the code to the new API
or pretending that we do not support OpenSSL 3, accept that we use
deprecated functionality.
Signed-off-by: Jouke Witteveen <j.witteveen@gmail.com>
NL80211_WPA_VERSION_3 was only added in kernel 5.2 so it should not be
set for older kernel versions. There is no direct way to check if the
value is supported. However, we can use the new infrastructure to check
whether the kernel has the NL80211_ATTR_SAE_PASSWORD attribute. It is
related and was added at the same time.
Fixes: 6cc78b3945 ("nl80211: Set NL80211_WPA_VERSION_2 vs. _3 based on AKM")
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Now that we have a more advanced check for the differences within the
Key Data field, nonzero Key Data Length case can be determined to be
EAPOL-Key msg 4/4 if there is no RSNE in the Key Data field.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Allow the Key Data field to be encrypted in EAPOL-Key msg 2/4 and 4/4.
This is for testing purposes to enable a convenient mechanism for
testing Authenticator behavior with either potential future extensions
or unexpected Supplicant behavior.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Extend RSN authenticator to be able to process EAPOL-Key msg 2/4 and 4/4
messages in cases where the Key Data field is encrypted using AES key
wrapping (i.e., non-AEAD cipher). While there is not yet any defined
case where such encryption would be used in IEEE Std 802.11-2020,
extensions are considered to be added to use such constructions (e.g.,
in IEEE P802.11bh). As such, it is good to extend the parsing and
processing rules in the authenticator to be ready for such
functionality.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Allow additional elements and KDEs to be added to EAPOL-Key msg 2/4 and
4/4. This is for testing purposes to enable a convenient mechanism for
testing Authenticator behavior with either potential future extensions or
incorrect Supplicant behavior.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
The initial Authenticator implementation depended on the Key Data field
being empty for EAPOL-Key msg 4/4. This worked fine for years in
practice, but in theory, vendor specific elements or KDEs could have
been added inti EAPOL-Key msg 4/4 and that would have broken this
design. In addition, the MLD case did introduce a KDE into EAPOL-Key msg
4/4 and required changes here.
As an initial step to make this more robust for future extensions,
recognize a received EAPOL-Key message as msg 4/4 if it is for RSN
(i.e., not WPAv1), has Secure=1, contains an unencrypted Key Data field,
and does not include RSNE.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Allow additional elements and KDEs to be added to EAPOL-Key msg 1/4 and
3/4 and allow EAPOL-Key msg 3/4 Key Data field to be not encrypted.
These are for testing purposes to enable a convenient mechanism for
testing supplicant behavior with either potential future extensions or
incorrect Authenticator behavior.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
Handle the center frequency to channel mapping more cleanly by skipping
the cases where the center frequencies are not set and verifying that
the mapping succeeds when they are set.
Signed-off-by: Jouni Malinen <j@w1.fi>
seg0 and seg1 variables can have any initial value, but they may cause
the BSS to have a wrong configuration later on in cases where the
separate center frequencies are not set.
Signed-off-by: Baligh Gasmi <gasmibal@gmail.com>
When p2p->state == P2P_LISTEN_ONLY, the statement before it
'p2p->cfg->is_p2p_in_progress(p2p->cfg->cb_ctx)' will be true, too, so
this function will print a message "Operation in progress" and return;
the workaround to handle listen failure will be never reached.
I met such an error when the 'remain-on-channel' command failed, then
the function p2p_ext_listen_timeout() just printed 'Operation in
progress' again and again, and the listen operation was not started
anymore.
Fixes: 0f1034e388 ("P2P: Refrain from performing extended listen during P2P connection")
Signed-off-by: zhuhai <zhuhai.mail@163.com>
There are multiple redundant MIN macro declarations, some of which are
not protecting against side effects. Move it to common.h instead.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
The kernel will reject commands if newer attributes are included even
though they should not be. Add appropriate checks in set_ap().
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Older kernel versions may not support all attributes and may refuse
commands that include them. To avoid sending too new attributes query
the highest supported attribute. This allows adding appropriate checks
where needed.
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Previously, we would only pass the event to the first interface that
matches. However, one wiphy can have multiple interfaces and each one
needs to get the event delivered. Without this, it could e.g. happen
that a radar detection event is forwarded to p2p-dev-wlan0 and not to
the wlan0 interface which actually needs it.
As such, keep iterating if we are processing a wiphy match and send the
event to all affected BSSs.
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
The dpp_test DPP_TEST_STOP_AT_AUTH_CONF case is supposed to prevent the
GAS exchange from happening. It was possible to bypass that testing
functionality if the TX status event for the Authentication Confirm
message was lost or delayed long enough for the GAS Request to be
proessed first. This could happen at least with UML time travel.
Work around this by checking for the special dpp_test case within GAS
request handling.
Signed-off-by: Jouni Malinen <j@w1.fi>
This was done with spatch using the following semantic patch and minor
manual edits to clean up coding style and avoid compiler warnings in
driver_wext.c:
@@
expression a,b;
@@
- os_memcmp(a, b, ETH_ALEN) == 0
+ ether_addr_equal(a, b)
@@
expression a,b;
@@
- os_memcmp(a, b, ETH_ALEN) != 0
+ !ether_addr_equal(a, b)
@@
expression a,b;
@@
- !os_memcmp(a, b, ETH_ALEN)
+ ether_addr_equal(a, b)
Signed-off-by: Jouni Malinen <j@w1.fi>
Missing array referencing ended up only the first entry in the list to
be compared.
Fixes: 31fcea931d ("WPS 2.0: Add support for AuthorizedMACs attribute")
Signed-off-by: Jouni Malinen <j@w1.fi>
This makes a code more readable for the large number of cases that use
os_memcmp() to check whether two ethernet addresses are equal.
Signed-off-by: Jouni Malinen <j@w1.fi>
If CONFIG_FILS isn't set, the compiler complains about unused variables.
Fix it.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Fix commit 5545d995b3 which could use undefined symbol
testing_fail_alloc() in os_malloc() if CONFIG_WPA_TRACE=y is used
without CONFIG_WPA_TRACE_BFD=y.
Fixes: 5545d995b3 ("trace: Share common implementation for TEST_FAIL and TEST_ALLOC_FAIL")
Signed-off-by: Jintao Lin <jintaolin@chromium.org>
The AP MLD MAC address is meant to be preserved here, but since it's
reset to zeroes before attempting to authenticate in
nl80211_mark_disconnected(), we can't just point to the AP MLD MAC
address in drv->auth_ap_mld_addr. Fix it by using a copy.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
When a station tries to authenticate to an AP that is affiliated with an
AP MLD, we need to remove all station instances from all the APs.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
If the station is not an MLO station do not attempt to find the
association station and return false in the ML specific disconnection
processing.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
This avoids unnecessary code duplication and adds the sta->wpa_sm deinit
calls for deauthentication while it was previously done only in the
disassociation case.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Set the current value instead of hardcoded 1.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
cfg80211 requires the link ID to be specified for requests to start a
remain-on-channel operation during an ML association. This feels wrong
since the ROC operation is in most cases unrelated to the
association. However, that requirement has been in place since kernel
commit 7b0a0e3c3a88 ("wifi: cfg80211: do some rework towards MLO link
APIs") from April 2022, and as such, it looks necessary to have
wpa_supplicant work around this by specifying the currently used link ID
that would seem to match the ROC channel most closely.
Signed-off-by: Jouni Malinen <quic_jouni@quicinc.com>
