If the RX frequency of the Probe Request frame is known, specify it when
sending the Probe Response frame. This is needed when the Probe Request
frame is received on another virtual interface, for example, when a GO
or P2PS client are discoverable on the group operating channel.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
Change send_mlme() API to allow sending management frames on a specific
channel, overriding the internal driver decision.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
Return P2P_PREQ_PROCESSED instead of P2P_PREQ_NOT_PROCESSED on
a successful Probe Request frame handling in p2p_reply_probe().
Verify a return value of p2p_reply_probe() in p2p_probe_req_rx()
and continue a pending invitation/connection flow only if the
Probe Request frame is from an expected P2P peer.
Signed-off-by: Max Stepanov <Max.Stepanov@intel.com>
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
In some cases, Probe Request frames can be received by a peer not only
on a listen channel. In this case an additional rx_freq parameter
explitly contains a Probe Request frame RX frequency. In case rx_freq is
set to 0, a Probe Request frame RX channel is assumed to be our own
listen channel (p2p->cfg->channel).
Signed-off-by: Max Stepanov <Max.Stepanov@intel.com>
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
CA country code was included mistakenly (copy-paste..) in cn_op_class_cc
while it was supposed to be included only in us_op_class_cc. In
practice, this did not result in incorrect operation due to the
us_op_class_cc list being checked first. Anyway, better fix
cn_op_class_cc to avoid confusion here.
Signed-off-by: Jouni Malinen <j@w1.fi>
This increases code coverage for gas.c testing to cover areas that
cannot be reached with pure hwsim test cases.
Signed-off-by: Jouni Malinen <j@w1.fi>
Currently, if multiple bss share are bridge and tagged vlan interface,
only the first instance of struct hostapd_vlan for this vlanid will have
the DVLAN_CLEAN_VLAN flag added. Thus, when this instance is removed,
the tagged vlan interface will be removed from bridge, thought other bss
might still need it. Similarily, the bridge will be left over, as the
does not have zero ports when the first instance of a struct
hostapd_vlan is freed.
This patch fixes this by having a global (per process) reference counter
for dynamic tagged vlan and dynamically created bridge interfaces, so
they are only removed after all local users are freed. (struct
hapd_interfaces *)->vlan_priv is used to hold src/ap/vlan_init.c global
per-process data like drv_priv does; right now this is only used for the
interface reference counting, but could get extended when needed. Then
possibly some vlan_global_init / vlan_global_deinit should be added, but
this is not required right now.
Additionally, vlan->configured is checked to avoid reference counter
decreasing before vlan_newlink increased them.
In order to avoid race conditions, vlan_dellink is called explicitly
after hostapd_vlan_if_remove. Otherwise there would be a short timeframe
between hostapd_vlan_if_remove and vlan_dellink during which the struct
hostapd_vlan still exists, so ap_sta_bind_vlan would try to attach
stations to it.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
Add operating class 125 (channels 149..169) to the list of P2P supported
channels. This allows the 5 GHz channels 161 and 169 to be used for P2P
GO when those channels are allowed for AP mode in the current regulatory
domain.
Signed-off-by: Amr BEN ABDESSALEM <amrx.ben.abdessalem@intel.com>
Instead of an int variable with magic values 0, 1, 2, use an enum that
gives clearer meaning to the values now that the original boolean type
global argument is not really a boolean anymore.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Commit 95b6bca66d ('Add rsn_pairwise bits
to set_ieee8021x() driver_ops') modified cipher configuration to use
unconditionally wpa_pairwise | rsn_pairwise. While that works for many
cases, it does not handle the case of dynamic configuration changes over
the control interface where wpa_pairwise or rsn_pairwise values may not
get cleared when the wpa parameter is modified. Fix this inconsistency
by configuring the driver with only the bits that are valid for the
currently enabled WPA/WPA2 version(s).
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Add GO Intent information of connecting device in GO Negotiation Request
event which will help applications to decide its own GO intent value in
advance and can avoid failure cases when both devices use GO Intent 15
depending on application requirement.
Signed-off-by: Mayank Haarit <mayank.h@samsung.com>
So that systems with bad clocks will send random session IDs,
instead of always ones starting at the same second.
If os_get_random() isn't available, use os_get_time(). But also
mix in now.tv_usec, so that the accounting session ID is more
likely to be globally and temporally unique.
Signed-off-by: Alan DeKok <aland@freeradius.org>
The OPENSSL_VERSION_NUMBER < 0x00909000L case of
openssl_get_keyblock_size() had not been kept in sync with the cleanup
changes.
Signed-off-by: Mayank Haarit <mayank.h@samsung.com>
Explicitly check for iface->current_mode before dereferencing it. While
this case may not happen in practice, it is better for the setup
functions to be more careful when doing the initial band selection.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
When device supports dual band operations with offloaded ACS, hw_mode
can now be set to any band (hw_mode=any) in order to allow ACS to select
the best channel from any band. After a channel is selected, the hw_mode
is updated for hostapd.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
While RFC 6124 does not define how Session-Id is constructed for
EAP-EKE, there seems to be consensus among the authors on the
construction. Use this Type | Nonce_P | Nonce_S construction based on
the following email:
From: Yaron Sheffer <yaronf.ietf at gmail.com>
To: ietf at ietf.org
Date: Wed, 17 Nov 2010 13:13:42 +0200
Expanding on my previous response, I suggest to resolve Bernard's
concern by adding the following text:
5.6 EAP Key Generation
EAP-EKE can be used for EAP key generation, as defined by [RFC 5247].
When used in this manner, the values required to establish the key
hierarchy are defined as follows:
- Peer-Id is the EAP-EKE ID_P value.
- Server-Id is the EAP-EKE ID_S value.
- Session-Id is the concatenated Type | Nonce_P | Nonce_S, where Type is
the method type defined for EAP-EKE in [Sec. 4.1], a single octet.
Thanks,
Yaron
Signed-off-by: Jouni Malinen <j@w1.fi>
The ACS code part of hostapd scans all the channels even if the channel
list is specified in the hostapd.conf. Limit the ACS scan channels to
the list specified in the config file.
Signed-off-by: Srinivasa Duvvuri<sduvvuri@chromium.org>
Do not add a P2P IE when a station interface is trying to associate
to an AP or P2P GO that publishes a P2P IE but does not include
a P2P manageability attribute.
This addresses an interoperability issue that was reported in
https://bugzilla.kernel.org/show_bug.cgi?id=96471, where a P2P GO
rejects association from a station interface without a specified
reason.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
Link Property query vendor command shall facilitate the information
of the Wi-Fi link. MAC address of the Wi-Fi peer is given as an input
for querying the link properties.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The cleanup code will handle this, but it is more robust to make sure
this is cleared to zero when allocating a new buffer.
Signed-off-by: Jouni Malinen <j@w1.fi>
The cleanup code will handle this, but it is more robust to make sure
this is cleared to zero when allocating a new buffer.
Signed-off-by: Jouni Malinen <j@w1.fi>
The L (Length) and M (More) flags needs to be cleared before deciding
whether the locally generated response requires fragmentation. This
fixes an issue where these flags from the server could have been invalid
for the following message. In some cases, this could have resulted in
triggering the wpabuf security check that would terminate the process
due to invalid buffer allocation.
Signed-off-by: Jouni Malinen <j@w1.fi>
The remaining number of bytes in the message could be smaller than the
Total-Length field size, so the length needs to be explicitly checked
prior to reading the field and decrementing the len variable. This could
have resulted in the remaining length becoming negative and interpreted
as a huge positive integer.
In addition, check that there is no already started fragment in progress
before allocating a new buffer for reassembling fragments. This avoid a
potential memory leak when processing invalid message.
Signed-off-by: Jouni Malinen <j@w1.fi>
The remaining number of bytes in the message could be smaller than the
Total-Length field size, so the length needs to be explicitly checked
prior to reading the field and decrementing the len variable. This could
have resulted in the remaining length becoming negative and interpreted
as a huge positive integer.
In addition, check that there is no already started fragment in progress
before allocating a new buffer for reassembling fragments. This avoid a
potential memory leak when processing invalid message.
Signed-off-by: Jouni Malinen <j@w1.fi>
The length of the received Commit and Confirm message payloads was not
checked before reading them. This could result in a buffer read
overflow when processing an invalid message.
Fix this by verifying that the payload is of expected length before
processing it. In addition, enforce correct state transition sequence to
make sure there is no unexpected behavior if receiving a Commit/Confirm
message before the previous exchanges have been completed.
Thanks to Kostya Kortchinsky of Google security team for discovering and
reporting this issue.
Signed-off-by: Jouni Malinen <j@w1.fi>
The length of the received Commit and Confirm message payloads was not
checked before reading them. This could result in a buffer read
overflow when processing an invalid message.
Fix this by verifying that the payload is of expected length before
processing it. In addition, enforce correct state transition sequence to
make sure there is no unexpected behavior if receiving a Commit/Confirm
message before the previous exchanges have been completed.
Thanks to Kostya Kortchinsky of Google security team for discovering and
reporting this issue.
Signed-off-by: Jouni Malinen <j@w1.fi>
The length of the WMM Action frame was not properly validated and the
length of the information elements (int left) could end up being
negative. This would result in reading significantly past the stack
buffer while parsing the IEs in ieee802_11_parse_elems() and while doing
so, resulting in segmentation fault.
This can result in an invalid frame being used for a denial of service
attack (hostapd process killed) against an AP with a driver that uses
hostapd for management frame processing (e.g., all mac80211-based
drivers).
Thanks to Kostya Kortchinsky of Google security team for discovering and
reporting this issue.
Signed-off-by: Jouni Malinen <j@w1.fi>
The debug information from httpread can be helpful in figuring out error
cases in general and as such, should be enabled by default. Get rid of
the hardcoded httpread_debug value that would require source code
changes to enable.
Signed-off-by: Jouni Malinen <j@w1.fi>
There is no need to continue processing a HTTP body when it becomes
clear that the end result would be over the maximum length.
Signed-off-by: Jouni Malinen <j@w1.fi>
Verify that ncopy parameter to memcpy is not negative. While this is not
supposed to be needed, it is a good additional protection against
unknown implementation issues.
Signed-off-by: Jouni Malinen <j@w1.fi>
strtoul() return value may end up overflowing the int h->chunk_size and
resulting in a negative value to be stored as the chunk_size. This could
result in the following memcpy operation using a very large length
argument which would result in a buffer overflow and segmentation fault.
This could have been used to cause a denial service by any device that
has been authorized for network access (either wireless or wired). This
would affect both the WPS UPnP functionality in a WPS AP (hostapd with
upnp_iface parameter set in the configuration) and WPS ER
(wpa_supplicant with WPS_ER_START control interface command used).
Validate the parsed chunk length value to avoid this. In addition to
rejecting negative values, we can also reject chunk size that would be
larger than the maximum configured body length.
Thanks to Kostya Kortchinsky of Google security team for discovering and
reporting this issue.
Signed-off-by: Jouni Malinen <j@w1.fi>
If the mp_init_multi() call had failed due to memory allocation failure,
mp_div() would have returned 1 instead of MP_MEM (-2). It looks like all
callers are checking the return value against MP_OKAY instead of <1
(etc.), so this does not seem to result in difference in behavior.
Anyway, it's best to fix the mp_div() return value for the MP_MEM error
case to avoid unexpected behavior.
Signed-off-by: Maks Naumov <maksqwe1@ukr.net>
In theory, the previous version could have resulted in reading one byte
beyond the end of the management frame RX buffer if the local driver
were to deliver a truncated Public Action frame for processing. In
practice, this did not seem to happen with mac80211-based drivers and
even if it were, the extra octet would be an uninitialized value in a
buffer rather than read beyond the end of the buffer.
Signed-off-by: Jouni Malinen <j@w1.fi>
This was previously checked as part of the eap_sim_parse_attr()
processing, but it is easier to review the code if there is an
additional explicit check for confirming that the Reserved field is
present since the pos variable is advanced beyond it.
Signed-off-by: Jouni Malinen <j@w1.fi>
Clean up eap_sake_parse_add_attr() design by passing in pointer to the
payload of the attribute instead of parsing these separately for each
attribute within the function.
Signed-off-by: Jouni Malinen <j@w1.fi>
The req_plen argument to eap_pax_process_std_1() and
eap_pax_process_std_3() could be smaller than sizeof(struct eap_pax_hdr)
since the main processing function was only verifying that there is
enough room for the ICV and then removed ICV length from the remaining
payload length.
In theory, this could have resulted in the size_t left parameter being
set to a negative value that would be interpreted as a huge positive
integer. That could then result in a small buffer read overflow and
process termination if MSGDUMP debug verbosity was in use.
In practice, it does not seem to be feasible to construct a short
message that would be able to pass the ICV validation (calculated using
HMAC-SHA1-128) even for the case where an empty password is used.
Anyway, the implementation should really check the length explicitly
instead of depending on implicit check through ICV validation.
Signed-off-by: Jouni Malinen <j@w1.fi>
The EAP-TLS-based helper functions can easily use struct wpabuf in more
places, so continue cleanup in that direction by replacing separate
pointer and length arguments with a single struct wpabuf argument.
Signed-off-by: Jouni Malinen <j@w1.fi>
This function is only using the Identifier field from the EAP request
header, so there is no need to pass it a pointer to the full message.
This makes it a bit easier to analyze the area that gets access to
unverified message payload.
Signed-off-by: Jouni Malinen <j@w1.fi>
MS-MPPE-Recv-Key generation in radius_msg_add_mppe_keys() used incorrect
function argument (send_key_len; should be recv_key_len) when allocating
a temporary buffer. Fix this by using the correct argument.
The only caller of the function uses the same length for both
send_key_len and recv_key_len, so this copy-paste error did not result
in any difference in the behavior.
Signed-off-by: Jouni Malinen <j@w1.fi>
Handling of WPS RF band for 60 GHz was missing. Add it in all relevant
places and also map "AES" as the cipher to GCMP instead of CCMP when
operating on the 60 GHz band.
Signed-off-by: Hamad Kadmany <qca_hkadmany@qca.qualcomm.com>
It was possible for the WPA_ALG_PMK algorithm in set_key() to result in
trying to configure a key with cipher suite 0. While this results in a
failure from cfg80211 or driver, this is not really desirable operation,
so add a check for cipher suite conversion result before issuing the
nl80211 command.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This can be used with Proxy ARP to allow multicast NAs to be forwarded
to associated STAs using link layer unicast delivery. This used to be
hardcoded to be enabled, but it is now disabled by default and can be
enabled with na_mcast_to_ucast=1. This functionality may not be desired
in all networks and most cases work without it, so the new
default-to-disabled is more appropriate.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Previously, struct wpa_group was created when the first station enters
the group and the struct wpa_group was not freed when all station left
the group. This causes a problem because wpa_group will enter
FATAL_FAILURE when a wpa_group is running while the AP_VLAN interface
has already been removed.
Fix this by adding a reference counter to struct wpa_group and free a
group if it is unused.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
If hostapd has bound a STA into a specific VLAN, the new vlan_id
parameter in the control interface STA command can now be used to check
which VLAN ID is in use.
Signed-off-by: Jouni Malinen <j@w1.fi>
All the system header files are supposed to be included before any other
internal header file apart from utils/includes.h.
Signed-off-by: Jouni Malinen <j@w1.fi>
By analysing objdump output some read only structures were found in
.data section. To help compiler further optimize code declare these
as const.
Signed-off-by: Mikael Kanstrup <mikael.kanstrup@sonymobile.com>
Now on an engine error we decode the error value and determine if the
issue is due to a true PIN error or not. If it is due to incorrrect PIN,
delete the PIN as usual, but if it isn't let the PIN be.
Signed-off-by: Mike Gerow <gerow@google.com>
Previously, mesh state machine transmits updated Commit Message when
receiving a Confirm Message in Committed state. According to the
standard, it should (re)send the latest Commit Message previously sent.
IEEE Std 802.11-2012, 11.3.8.6.4 Protocol instance behavior - Committed
state:
"Upon receipt of a Con event, ... If Sync is not greater than
dot11RSNASAESync, the protocol instance shall increment Sync, transmit
the last Commit Message sent to the peer, and set the t0
(retransmission) timer."
Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
This extends dynamic_vlan=required checks to apply for WPA-PSK with
macaddr_acl=2 (RADIUS) case.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
Previously, only P2P and mesh use cases enforced unique MAC address for
a dynamically added virtual interface. Extend this to cover normal
station mode interfaces since those can now be added with INTERFACE_ADD.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
There is no need to try to derive DH shared key with a peer that tries
to use too short or too long DH Public Key. Previously, such cases ended
up implicitly getting rejected by the DH operations failing to produce
matching results. That is unnecessarily, so simply reject the message
completely if it does not have a Public Key with valid length. Accept
couple of octets shorter value to be used to avoid interoperability
issues if there are implementations that do not use zero-padding
properly.
Signed-off-by: Jouni Malinen <j@w1.fi>
This enforces variable length strings Manufacturer, Model Name, Model
Number, and Serial Number to be within the maximum length defined in the
WSC specification. While none of the existing users for these within
hostapd/wpa_supplicant had problems with longer strings, it is good to
ensure the strings are not longer to avoid potential issues at higher
layer components.
Signed-off-by: Jouni Malinen <j@w1.fi>
Check the element length in the parser and remove the length field from
struct ieee802_11_elems since the element is of fixed length.
Signed-off-by: Jouni Malinen <j@w1.fi>
Check the element length in the parser and remove the length field from
struct ieee802_11_elems since the element is of fixed length.
Signed-off-by: Jouni Malinen <j@w1.fi>
Check the element length in the parser and remove the length field from
struct ieee802_11_elems since the element is of fixed length.
Signed-off-by: Jouni Malinen <j@w1.fi>
Check the element length in the parser and remove the length field from
struct ieee802_11_elems since the element is of fixed length.
Signed-off-by: Jouni Malinen <j@w1.fi>
Remove the length field from struct ieee802_11_elems since the only
allowed element length is five and that is checked by the parser.
Signed-off-by: Jouni Malinen <j@w1.fi>
Check the element length in the parser and remove the length field from
struct ieee802_11_elems since the only allowed element length is one.
Signed-off-by: Jouni Malinen <j@w1.fi>
Check the element length in the parser and remove the length field from
struct ieee802_11_elems since the only allowed element length is one.
Signed-off-by: Jouni Malinen <j@w1.fi>
All the existing users of these elements were already validating the
element length. However, it is clearer to validate this already at the
parser for extra layer of protection for any future changes.
Signed-off-by: Jouni Malinen <j@w1.fi>
Commit 88b32a99d3 ('FT: Add FT AP support
for drivers that manage MLME internally') added an alternative way of
processing the WMM TSPEC from RIC. However, that change did not seem to
include the same checks for WMM TSPEC element length that were used in
the original implementation for MLME-in-hostapd case. Fix this by
sharing the older implementation of copying the WMM TSPEC from RIC for
both cases.
It looks like the destination buffer for the response is sufficiently
long for the fixed length copy, but it may have been possible to trigger
a read beyond the end of the FTIE by about 50 bytes. Though, that seems
to be within the buffer received for RX buffer in the case that uses
this driver-based AP MLME design for FT.
Signed-off-by: Jouni Malinen <j@w1.fi>
This program can be used to run fuzzing tests for areas related to EAPOL
frame parsing and processing on the supplicant side.
Signed-off-by: Jouni Malinen <j@w1.fi>
This program can be used to run fuzzing tests for areas related to P2P
message parsing and processing. p2p-fuzzer allows data files to be used
to inject Probe Response and Action frames for processing by the P2P
module.
Signed-off-by: Jouni Malinen <j@w1.fi>
This makes it more obvious that the wps_parse_msg() bounds checking is
used to verify that there is sufficient space in the P2P buffer for the
secondary device types.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This modifies couple of code segments that replaced control characters
in strings with '_' to use a common helper function.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
While it looks like all the users of this parsed attribute were able to
handle longer SSID values, there is no valid use case for these and to
avoid any potential future issues, enforce maximum length (32 bytes) on
the SSID during parsing.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
While it looks like all the users of this parsed attribute were able to
handle longer Device Name values, there is no valid use case for these
and to avoid any potential issues in upper layer components, enforce
maximum length (32 bytes) on the Device Name during parsing.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
While none of the current users of msg.persistent_ssid{,_len} would have
issues with too long SSID value, it is safer to enforce bounds checking
on the SSID while parsing the attribute to avoid any potential issues in
the future.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The SSID element is defined to have a valid length range of 0-32. While
this length was supposed to validated by the users of the element
parser, there are not really any valid cases where the maximum length of
32 octet SSID would be exceeded and as such, the parser itself can
enforce the limit as an additional protection.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This fixes a possible memcpy overflow for P2P dev->oper_ssid in
p2p_add_device(). The length provided by the peer device (0..255 bytes)
was used without proper bounds checking and that could have resulted in
arbitrary data of up to 223 bytes being written beyond the end of the
dev->oper_ssid[] array (of which about 150 bytes would be beyond the
heap allocation) when processing a corrupted management frame for P2P
peer discovery purposes.
This could result in corrupted state in heap, unexpected program
behavior due to corrupted P2P peer device information, denial of service
due to process crash, exposure of memory contents during GO Negotiation,
and potentially arbitrary code execution.
Thanks to Google security team for reporting this issue and smart
hardware research group of Alibaba security team for discovering it.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The multi-SSID design that used a single beaconing BSSID with multiple
SSIDs was never completed in this repository, so there is no need to
maintain the per-STA ssid/ssid_probe pointers that could only point to
&hapd->conf->ssid. Save some memory and reduce code complexity by
removing this unused partial capability.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Commit fa0e715100 ('Use
tls_connection_prf() for all EAP TLS-based key derivation') copied some
pointer checks from the generic implementation to tls_openssl.c.
However, these are arrays and cannot be NULL in OpenSSL data. Remove the
unnecessary checks and add master_key_length check for completeness.
(CID 109619).
Signed-off-by: Jouni Malinen <j@w1.fi>
This makes it easier to interpret debug logs for cases where hostapd
control multiple interfaces.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This makes it easier to interpret debug logs in case hostapd controls
multiple interfaces and a STA roams between them.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Currently, vlan_remove_dynamic() is only called when the station VLAN ID
is changed (ap_sta_bind_vlan), but not when the station is freed. So
dynamic VLAN interfaces are not removed actually except within 1x
reauthentification VLAN ID change, although most of the code is already
there.
This patch fixes this by calling vlan_remove_dynamic() in ap_free_sta().
It cannot just use sta->vlan_id for this, as this might have been
changed without calling ap_sta_bind_vlan() (ap/ieee802_11.c:handle_auth
fetches from RADIUS cache for WPA-PSK), thus reference counting might
not have been updated. Additionally, reference counting might get wrong
due to old_vlanid = 0 being passed unconditionally, thus increasing the
reference counter multiple times.
So tracking the currently assigned (i.e., dynamic_vlan counter
increased) VLAN is done in a new variable sta->vlan_id_bound. Therefore,
the old_vlan_id argument of ap_sta_bind_vlan() is no longer needed and
setting the VLAN for the sta in driver happens unconditionally.
Additionally, vlan->dynamic_vlan is only incremented when it actually
is a dynamic VLAN.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
hostapd receives NEWLINK messages multiple times and thus does
configuration of the the vlan interface multiple times.
This is not required and leads to the following during cleanup in
test pmksa_cache_preauth_vlan_used:
1. run-test.py does: brctl delif brvlan1 wlan3.1
2. hostapd processes NEWLINK and does: brctl addif brvlan1 wlan3.1
3. run-test.py does: brctl delbr brvlan1
-> fails as wlan3.1 is still in the bridge
This patch fixes this by ignoring repeated NEWLINK messages.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
When hitting
> Failed to create interface wlan3.1: -23 (Too many open files in system)
> Try to remove and re-create wlan3.1
hostapd deletes the AP_VLAN interface and then recreates it. Thus the
kernel assigns the same ifidx to the new interfaces and sends DELLINK
and NEWLINK mesages.
As the DELLINK messages are processed after the struct hostapd_vlan is
added, hostapd deletes the struct hostapd_vlan entry, deconfigures the
AP_VLAN interface and leaves vlan_newlink nothing to find afterwards.
So this patch makes DELLINK messages to be ignored when the interface
exists.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
Previously, during RSN preauthentication, ap_sta_bind_vlan() was called,
which fails for non-zero sta->vlan_id as the station is not known to the
kernel driver.
Fix this by binding the station only if it is associated. If it is not
associated, ap_sta_bind_vlan() will be done later during association.
In addition, reject Access-Accept if the returned VLAN ID is not valid
in the current hostapd configuration.
Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
