This allows the stations to be assigned to their own vif. It does not
need dynamic_vlan to be set. Make hostapd call ap_sta_set_vlan even if
!vlan_desc.notempty, so vlan_id can be assigned regardless.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
This makes vlan_newlink() and vlan_dellink() add tagged VLANs to AP_VLAN
interfaces as given by struct vlan_description.
hostapd_vlan_if_remove() is done in vlan_dellink() as tagged interfaces
need to be removed before the interface can be deleted and a DELLINK
message can be generated.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
To prepare for adding tagged VLAN support in vlan_init.c, vlan_newlink()
and vlan_dellink() are split into multiple functions. This reduces
indention and eases adding tagged VLANs as well.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
1. Add tagged VLAN to struct vlan_description
(compile limited number of tagged VLANs per description)
For k tagged VLANs, the first k entries in vlan_description.tagged
are used. They are sorted in ascending order. All other entries are
zero. This way os_memcmp() can find identical configurations.
2. Let tagged VLANs be parsed from RADIUS Access-Accept
3. Print VLAN %d+ with %d=untagged VID if tagged VLANs are set
4. Select an unused vlan_id > 4096 for new tagged VLAN configurations
5. Add EGRESS_VLAN RADIUS attribute parsing also for untagged VLANs
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
Separate uplink configuration (IEEE 802.1q VID) and grouping of stations
into AP_VLAN interfaces.
The int vlan_id will continue to identify the AP_VLAN interface the
station should be assigned to. Each AP_VLAN interface corresponds to an
instance of struct hostapd_vlan that is uniquely identified by int
vlan_id within an BSS.
New: Each station and struct hostapd_vlan holds a struct
vlan_description vlan_desc member that describes the uplink
configuration requested. Currently this is just an int untagged IEEE
802.1q VID, but can be extended to tagged VLANs and other settings
easily.
When the station was about to be assigned its vlan_id, vlan_desc and
vlan_id will now be set simultaneously by ap_sta_set_vlan(). So
sta->vlan_id can still be tested for whether the station needs to be
moved to an AP_VLAN interface.
To ease addition of tagged VLAN support, a member notempty is added to
struct vlan_description. Is is set to 1 if an untagged or tagged VLAN
assignment is requested and needs to be validated. The inverted form
allows os_zalloc() to initialize an empty description.
Though not depended on by the code, vlan_id assignment ensures:
* vlan_id = 0 will continue to mean no AP_VLAN interface
* vlan_id < 4096 will continue to mean vlan_id = untagged vlan id
with no per_sta_vif and no extra tagged vlan.
* vlan_id > 4096 will be used for per_sta_vif and/or tagged vlans.
This way struct wpa_group and drivers API do not need to be changed in
order to implement tagged VLANs or per_sta_vif support.
DYNAMIC_VLAN_* will refer to (struct vlan_description).notempty only,
thus grouping of the stations for per_sta_vif can be used with
DYNAMIC_VLAN_DISABLED, but not with CONFIG_NO_VLAN, as struct
hostapd_vlan is still used to manage AP_VLAN interfaces.
MAX_VLAN_ID will be checked in hostapd_vlan_valid and during setup of
VLAN interfaces and refer to IEEE 802.1q VID. VLAN_ID_WILDCARD will
continue to refer to int vlan_id.
Renaming vlan_id to vlan_desc when type changed from int to struct
vlan_description was avoided when vlan_id was also used in a way that
did not depend on its type (for example, when passed to another
function).
Output of "VLAN ID %d" continues to refer to int vlan_id, while "VLAN
%d" will refer to untagged IEEE 802.1q VID.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
Commit 8bcf8de827 ('OpenSSL: Fix memory
leak in PKCS12 additional certificate parsing') tried to fix a memory
leak in both the 1.0.2(and newer) and 1.0.1 branches of PKCS12 parsing.
However, the 1.0.1 case was not properly tested and freeing of the
certificate after a successful SSL_CTX_add_extra_chain_cert() call
resulted in use of freed memory when going through the TLS handshake.
Fix this by not freeing the certificate in that specific case.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
While '!func() == 0' here resulted in correct behavior, it is not clear
and clang is starting to warn about this (-Wlogical-not-parentheses).
Use 'func()' instead as the condition to clear this.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Ciuple of the nonce comparisons used a strange '!os_memcmp() == 0' to
check if the values were different. While this resulted in correct
behavior, the construction is not exactly clear and clang has started
warning about this (-Wlogical-not-parentheses). Clean this up by using
'os_mecmp() != 0'.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The use of the newer OpenSSL API in openssl_hmac_vector() removes one of
the memory allocations, so the TEST_ALLOC_FAIL here could not trigger.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The additional PKCS12 certificates were not freed properly in the loop
added in commit de2a7b796d ('OpenSSL: Use
connection certificate chain with PKCS#12 extra certs').
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 5c9a33702f ('OpenSSL: Clean up
crypto_hash_*() to use a single implementation') added a wrapper
function to allow the new OpenSSL API to be used with older OpenSSL
versions. However, the HMAC_CTX_free() wrapper was incorrectly skipping
the call to HMAC_CTX_cleanup() which is still needed to free the
resources OpenSSL allocated internally.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
SSL_library_init() does not work properly after EVP_cleanup() starting
from OpenSSL 1.1.0 pre release 3. The automated library init/deinit
functions in that pre release are supposed to handle all initialization
and deinitialiation, so comment out the explicit calls to these function
with OpenSSL 1.1.0 and newer.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 1eb87ae48d ('OpenSSL: Use
EVP_CIPHER_CTX_new() to work with OpenSSL 1.1.0') started using
EVP_CIPHER_CTX_new() to allocate EVP_CIPHER_CTX from heap instead of
using stack memory. This commit used incorrect EVP_CIPHER_CTX_reset()
function in number of cases when the allocated memory was supposed to be
freed instead of just reset for reuse. Fix this by using
EVP_CIPHER_CTX_free() properly.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
rfkill_init() uses realpath() which allocates memory and that memory was
not freed on the success path.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
If SSL_CTX_new(SSLv23_method()) fails, tls_init() error path did not
free the allocated struct tls_data instance.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
PBSS (Personal Basic Service Set) is a new BSS type for DMG
networks. It is similar to infrastructure BSS, having an AP-like
entity called PCP (PBSS Control Point), but it has few differences.
PBSS support is mandatory for IEEE 802.11ad devices.
Add a new "pbss" argument to network block. The argument is used
in the following scenarios:
1. When network has mode=2 (AP), when pbss flag is set will start
as a PCP instead of an AP.
2. When network has mode=0 (station), when pbss flag is set will
connect to PCP instead of AP.
The function wpa_scan_res_match() was modified to match BSS according to
the pbss flag in the network block (wpa_ssid structure). When pbss flag
is set it will match only PCPs, and when it is clear it will match only
APs.
Signed-off-by: Lior David <qca_liord@qca.qualcomm.com>
PBSS (Personal Basic Service Set) is a new BSS type for DMG
networks. It is similar to infrastructure BSS, having an AP-like
entity called PCP (PBSS Control Point), but it has few differences.
PBSS support is mandatory for IEEE 802.11ad devices.
Add a pbss flag to the relevant structures to support starting a PCP and
connecting to a PCP. Implement support in the nl80211 driver by using
the new PBSS flag attribute.
Signed-off-by: Lior David <qca_liord@qca.qualcomm.com>
NOTE: kqueue has to be closed and re-build after forking. epoll *should*
do the same, but it seems that wpa_supplicant doesn't need it at least.
I have re-worked a little bit of the epoll code (moved into a similar
kqueue function) so it's trivial to requeue epoll if needed in the
future.
Signed-off-by: Roy Marples <roy@marples.name>
This function can be used to re-build eloop socket tables after forking
for eloop implementations that need this.
Signed-off-by: Roy Marples <roy@marples.name>
It does more than intended; apart from denying messages to that
particular interface it also denies all messages non-qualified with an
interface globally. This blocks messages completely unrelated to
wpa_supplicant, such as NetworkManager communication with the VPN
plugins.
From the dbus-daemon manual:
Be careful with send_interface/receive_interface, because the
interface field in messages is optional. In particular, do NOT
specify <deny send_interface="org.foo.Bar"/>! This will cause
no-interface messages to be blocked for all services, which is almost
certainly not what you intended. Always use rules of the form: <deny
send_interface="org.foo.Bar" send_destination="org.foo.Service"/>
We can just safely remove those rules, since we're sufficiently
protected by the send_destination matches and method calls are
disallowed by default anyway.
Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
This verifies that num_plinks is decremented properly if a peer mesh STA
reconnects without closing the link explicitly.
Signed-off-by: Jouni Malinen <j@w1.fi>
When a mesh point reconnects by starting from Authentication frame
sequence, the plink count was not decremented from its last connection.
This resulted in leaking peer link count and causing wpa_supplicant to
reject the connection after max_peer_links (default: 99) reconnects.
This was reproduced by pre-configuring 2 mesh points with mesh
credentials. Boot both mesh points and make sure they connect to each
other. Then in a loop reboot one of the mesh points after it
successfully connects while leaving the other mesh point up and running.
After 99 iterations the supplicant on mesh point that is not rebooting
will reject the connection request from the other mesh point.
Fix this by decrementing num_plinks when freeing a STA entry that is
still in PLINK_ESTAB state.
Signed-off-by: Srinivasa Duvvuri <sduvvuri@chromium.org>
In cases where the bandwidth is not set when starting an AP/P2P GO,
the code tries to use 160 MHz or 80 MHz channels. As a result, the
AP/P2P GO configuration is set to use these channel widths even if
they are not available, which may results in failing to start the
AP/P2P GO.
Fix this by changing the AP/P2P GO configuration not to use VHT channels
when they are not available. In this case the AP/P2P GO will use a 40
MHz channel, if available, or a 20 MHz channel, if this is the maximum
available width.
Signed-off-by: Avraham Stern <avraham.stern@intel.com>
Setting a long off channel wait time for P2P Action frames when
we know we are already on the right channel may cause a delay in
sending the Action frame (because the driver may not be able to
satisfy the request for long wait time until previous off channel
requests are over). This may be crucial for P2P response frames
that must be sent within 100 milliseconds of receiving the request.
Fix this by adjusting P2P Action frame wait times as follows:
1. For GO Negotiation Response frame, shorten the wait time to 100 ms.
This is reasonable because the peer has just sent us the GO
Negotiation Request frame, so it is known to be on the right
channel and is probably ready to send us the GO Negotiation
Confirmation frame without delay.
2. For GO Negotiation Confirmation, P2P Invitation Response, and
Provision Discovery Response frames, there is no need for wait
time at all as this is the last frame in the exchange. So set
the wait time to 50 ms to ensure there is enough time to send the
frame.
Signed-off-by: Avraham Stern <avraham.stern@intel.com>
There is no need to maintain three copies of this functionality even if
it is currently implemented as a single function call.
Signed-off-by: Jouni Malinen <j@w1.fi>
Rework the Acct-Session-Id and Acct-Multi-Session-Id implementation to
give better global and temporal uniqueness. Previously, only 32-bits of
the Acct-Session-Id would contain random data, the other 32-bits would
be incremented. Previously, the Acct-Multi-Session-Id would not use
random data. Switch from two u32 variables to a single u64 for the
Acct-Session-Id and Acct-Multi-Session-Id. Do not increment, this serves
no legitimate purpose. Exclusively use os_get_random() to get quality
random numbers, do not use or mix in the time. Inherently take a
dependency on /dev/urandom working properly therefore. Remove the global
Acct-Session-Id and Acct-Multi-Session-Id values that serve no
legitimate purpose.
Signed-off-by: Nick Lowe <nick.lowe@lugatech.com>
These can get allocated within eapol_auth_alloc(), so it is more logical
to free them in eapol_auth_free() instead of ieee802_1x_free_station()
that ends up calling eapol_auth_free().
Signed-off-by: Jouni Malinen <j@w1.fi>
