Jouni Malinen [Sun, 22 Jan 2012 10:23:28 +0000 (12:23 +0200)]
Work around interop issue with WPA type EAPOL-Key 4/4 in WPA2 mode
Some deployed station implementations seem to send msg 4/4 with
incorrect type value in WPA2 mode. Add a workaround to ignore that issue
so that such stations can interoperate with hostapd authenticator. The
validation checks were added in commit f8e96eb6fd960a017793942cff0eb43b09f444c6.
Jouni Malinen [Sun, 8 Jan 2012 17:35:33 +0000 (09:35 -0800)]
P2P: Use Device ID attribute to filter Probe Request frames as GO
The Device ID attribute was already used in Listen state, but it was
ignored in GO role. Verify that there is a match with Device ID in
GO rule, too, before replying to the Probe Request frame.
Jouni Malinen [Mon, 2 Jan 2012 20:36:11 +0000 (22:36 +0200)]
Allow SNonce update after sending EAPOL-Key 3/4 if 1/4 was retransmitted
Some supplicant implementations (e.g., Windows XP WZC) update SNonce for
each EAPOL-Key 2/4. This breaks the workaround on accepting any of the
pending requests, so allow the SNonce to be updated even if we have
already sent out EAPOL-Key 3/4.
While the issue was made less likely to occur when the retransmit
timeout for the initial EAPOL-Key msg 1/4 was increased to 1000 ms,
this fixes the problem even if that timeout is not long enough.
Jouni Malinen [Mon, 2 Jan 2012 19:27:06 +0000 (21:27 +0200)]
Do not disconnect STA based on inactivity on driver failure
Now that we can use driver_nl80211.c with non-mac80211 drivers that
implement SME/MLME internally, we may not get inactivity time from the
driver. If that is the case, we need to skip disconnection based on
maximum inactivity timeout. This fixes some unexpected disconnection
cases with ath6kl in AP mode.
Jouni Malinen [Sun, 1 Jan 2012 16:53:12 +0000 (18:53 +0200)]
Do not trigger fast reconnection on locally generated deauth/disassoc
The deauthentication and disassociation events from nl80211 were being
processed identically regardless of whether the frame was generated by
the local STA or the AP. This resulted in fast reconnection mechanism
getting triggered even in the case where the disconnection was detected
locally (e.g., due to beacon loss) while this was supposed to happen
only in the case where the AP is sending an explicit Deauthentication
or Disassociation frame with a specific reason code.
Fix this by adding a new deauth/disassoc event variable to indicate
whether the event was generated locally.
Jouni Malinen [Sat, 31 Dec 2011 15:40:49 +0000 (17:40 +0200)]
dbus: Increase buffer size to fix Introspect XML
Commit e9c3c1afedd2d0c6a0939f40c40701af8c450e1f added a new D-Bus
method and that was enough to push the Introspect XML buffer over
the previously allocated 8000 bytes. Increase the buffer size to
make enough room for P2P interface. In addition, add a debug
message to indicate if an XML segment does not fit into the buffer
to make this types of failures somewhat easier to catch.
Jouni Malinen [Sat, 31 Dec 2011 11:42:23 +0000 (13:42 +0200)]
Clear EAPOL authWhile and heldWhile values when port is disabled
IEEE Std 802.1X-2004 does not clear authWhile and heldWhile in this
case, but doing so allows the timer tick to be stopped more quickly when
the port is not enabled. Since these variables are used only within HELD
and RECEIVE states, clearing them on initialization does not change
actual state machine behavior. This reduces some unnecessary operations
in port disabled state and cleans up the wpa_supplicant debug log after
disconnection.
Jouni Malinen [Sat, 31 Dec 2011 11:25:37 +0000 (13:25 +0200)]
Remove possible authentication timeout on connection failure
The authentication timeout could be triggered after the connection has
already been known to have failed. The event at that point can be
confusing, so better cancel the timeout when processing connection
failure.
Jouni Malinen [Sat, 31 Dec 2011 11:17:20 +0000 (13:17 +0200)]
WPS: Use single channel scan if AP channel already known
If the BSSID of the AP is specified in the WPS command, the target
AP is likely already in the BSS table and its operating channel is
known. Use this information to speed up connection by only scanning
the known channel.
Michał Górny [Thu, 29 Dec 2011 19:39:34 +0000 (21:39 +0200)]
Use correct (multi-user) target when installing systemd units
The 'network.target' is special (per systemd.special(7)), and is to be
brought up indirectly when network is actually configured (i.e. through
DHCP or static address settings).
Irrelevant of that, all services should be always installed in
multi-user.target.
Since wps_registrar_pbc_timeout is called to stop PBC, previously
registered wps_registrar_pbc_timeout must be canceled when canceling
the WPS operation.
Jouni Malinen [Thu, 29 Dec 2011 12:15:31 +0000 (14:15 +0200)]
nl80211: Fix NL80211_CMD_FRAME to not try offchannel without driver support
The offchanok parameter is hardcoded to one in number of paths and that
added NL80211_ATTR_OFFCHANNEL_TX_OK attribute to NL80211_CMD_FRAME
unconditional. cfg80211 rejects this with EINVAL if the driver does not
indicate support for offchannel TX. Fix this by not requesting
offchannel TX depending on driver capabilities. Remain-on-channel
operation was used for those cases anyway, so the additional attribute
was not really needed for these in the first place.
Jouni Malinen [Thu, 29 Dec 2011 11:28:17 +0000 (13:28 +0200)]
P2P: Stop remain-on-channel prior to starting join-a-group client
This fixes issues with drivers that do not handle concurrent
remain-on-channel and scan operations in a case where Provision
Discovery Response frame is not received to stop the Action frame
handshake.
P2P: Reject p2p_group_add if forced frequency is not acceptable
If the freq parameter is specified and we are already running legacy STA
on a different frequency with a driver that does not support
multi-channel concurrency, reject p2p_group_add. Same code already
exists in the path of P2P connection with go negotiation but is missing
for autonomous GO.
Jouni Malinen [Tue, 27 Dec 2011 18:32:29 +0000 (20:32 +0200)]
nl80211: Do not stop AP mode Probe Request reporting on Listen stop
When nl80211_setup_ap() has enabled Probe Request reporting, this must
not be disabled when P2P Listen state is stopped to avoid breaking AP
mode operations. This could happen, e.g., if a Probe Request frame was
received from a P2P device that the we are trying to invite to our group
(i.e., when operating in GO role). p2p_probe_req_rx() calls
p2p_invite_start() in this case and that ends up calling
p2p->cfg->stop_listen() which calls probe_req_report() driver op.
hostapd uses the poll method to check if the station is alive
after the station has been inactive for ap_max_inactivity seconds.
Make the poll mechanism configurable so that user can choose to
disconnect idle clients.
This can be especially useful when some devices/firmwares have
restrictions on the number of clients that can connect to the AP
and that limit is smaller than the total number of stations trying
to use the AP.
Jouni Malinen [Sat, 24 Dec 2011 11:50:59 +0000 (13:50 +0200)]
Increase initial group key handshake retransmit timeout to 500 ms
The 100 ms timeout on retransmitting group key message can be too short
for stations that are in power save mode or if there is a large number
of association stations. While the retransmission of the EAPOL-Key frame
should allow this to be recovered from, it is useful to avoid
unnecessary frames to save soem CPU and power.
Jouni Malinen [Fri, 23 Dec 2011 16:15:07 +0000 (18:15 +0200)]
nl80211: Fix mgmt RX for device SME in AP mode without monitor
Drivers that use device SME in AP mode may still need to be
subscribed for Action frame RX when monitor interface is not used.
This fixes number of P2P GO operations with ath6kl.
Jouni Malinen [Fri, 23 Dec 2011 16:13:01 +0000 (18:13 +0200)]
nl80211: Fix use_monitor driver capability for non-mac80211 drivers
Poll command may be enough for mac80211 to figure out whether monitor
interface is to be used, but this change did not take into account
non-mac80211 drivers that support AP mode without monitor interface.
For example, ath6kl needs to get use_monitor disabled.
Jouni Malinen [Thu, 22 Dec 2011 20:47:41 +0000 (22:47 +0200)]
P2P: Maintain a list of P2P Clients for persistent group on GO
Add a new persistent group network block field, p2p_client_list, to
maintain a list of P2P Clients that have connected to a persistent
group. This allows GO of a persistent group to figure out more easily
whether re-invocation of a persistent group can be used with a specific
peer device.
Jouni Malinen [Thu, 22 Dec 2011 19:45:20 +0000 (21:45 +0200)]
P2P: Show persistent group info on P2P_PEER data
Add "persistent=<network id>" line to P2P_PEER ctrl_iface data
if a persistent group credentials are available for this peer.
This makes it easier for external programs to figure out when
a persistent group could be re-invoked with this peer.
For now, this information is only available on the P2P client,
but similar information can be added for GO once a list of P2P
clients is maintained with the persistent group data.
Jouni Malinen [Thu, 22 Dec 2011 19:26:31 +0000 (21:26 +0200)]
P2P: Move public P2P_PEER info generation into ctrl_iface
The P2P module provides access to public peer data in struct
p2p_peer_info. Use this to build the P2P_PEER information in
ctrl_iface.c instead of providing such text format data from the P2P
module.
The internal data that was previously built in p2p_get_peer_info() as
part of the text format peer data is now available through a separate
p2p_get_peer_info_txt() function. This is still included in P2P_PEER
output to maintain backwards compatibility with external programs that
could have started to use this. However, it should be noted that this
data is not really supposed to be used for anything else apart from
debugging purposes and its format is subject to change.
Jouni Malinen [Thu, 22 Dec 2011 19:04:41 +0000 (21:04 +0200)]
P2P: Replace p2p_get_peer_info with p2p_peer_known when applicable
p2p_get_peer_info() was used in multiple places just to check whether a
specific peer is known. This was not the designed use for the function,
so introduce a simpler function for that purpose to make it obvious that
the p2p_get_peer_info() function is actually used only in ctrl_iface.c.
Jouni Malinen [Wed, 21 Dec 2011 11:36:58 +0000 (13:36 +0200)]
Process EVENT_SCHED_SCAN_STOPPED partially if interface is disabled
The internal sched_scanning state needs to be cleared on this event
even if the events happen to get ordered in a way that the interface
gets disabled just prior to EVENT_SCHED_SCAN_STOPPED event.
Jouni Malinen [Wed, 21 Dec 2011 11:36:13 +0000 (13:36 +0200)]
Ignore disassociation event in wpa_supplicant AP mode
If the driver was associated in station mode just before the AP mode was
started, the station mode disassociation event may end up getting
delivered to wpa_supplicant only after the AP mode has been started.
This can result in unexpected attempt to reassociate back to the network
that was previously used in station mode. Avoid this by ignoring the
disassociation event.
Jouni Malinen [Wed, 21 Dec 2011 11:35:18 +0000 (13:35 +0200)]
P2P: Do not allow pending listen command override connect
If p2p_listen is issued during a p2p_scan, a pending after-scan operation
is scheduled. However, since there is support for only a single pending
operation, this was able to override a previously scheduled pending
connect command. This can break some command sequences, so give higher
priority to pending connect operation.
Jouni Malinen [Wed, 21 Dec 2011 11:34:25 +0000 (13:34 +0200)]
P2P: Fix GO start on interface that has active station connection
The active station connection triggered wpa_supplicant AP mode startup
to try to update Beacon IEs before the AP mode was properly initialized
and that resulted in NULL pointer dereference in driver_nl80211.c. Fix
this by skipping the IE update before the AP mode is initialized.
Jouni Malinen [Wed, 30 Nov 2011 11:41:55 +0000 (13:41 +0200)]
Android: nl80211: Work around limited genl_ctrl_resolve()
Android ICS system/core/libnl_2 has very limited genl_ctrl_resolve()
implementation that cannot handle names other than nlctrl. Work
around that by implementing more complete genl_ctrl_resolve()
functionality within driver_nl80211.c for Android builds.
Jouni Malinen [Sun, 18 Dec 2011 20:21:22 +0000 (22:21 +0200)]
P2P: Use hardcoded idle timeout of 10 seconds in P2P client role
The p2p_group_idle configuration parameter is much more useful for
GO role, so use a separate hardcoded value of 10 seconds in P2P
client role. In practice, this means that the P2P client role will
automatically tear down the group when the GO tears down the group.
The 10 second timeout is enough to recover from temporary disconnections
without unnecessary tearing down the group if the GO is still present
and allows the client to connect.
Jouni Malinen [Sun, 18 Dec 2011 20:18:42 +0000 (22:18 +0200)]
P2P: Fix disconnect event not to increase idle timeout
When P2P client is processing a disconnection event, make sure the P2P
idle timeout does not get increased, i.e., set a new timeout only if no
timeout is in use. wpa_state changes between DISCONNECTED and SCANNING
can generate multiple calls to wpas_p2p_notif_disconnect() and
previously this was enough to force the idle timeout never hit in
practice when in P2P client role.
Jouni Malinen [Sun, 18 Dec 2011 19:48:25 +0000 (21:48 +0200)]
Lower RX_MGMT driver event debug level for Beacon frames
This event can be very frequent in AP mode when Beacon frames from
neighboring BSSes are delivered to user space. Drop the debug
message priority from DEBUG to EXCESSIVE for Beacon frames.
P2P: Reduce the idle time in Wait peer connect state
When waiting for go_neg frame from the peer in WAIT_PEER_CONNECT state,
I have observed that sometimes it takes 20 to 30 secs for successful GO
negotiation. I also found out that it is because of 1 second idle time,
in WAIT_PEER_CONNECT state. While it is good to have 1 second idle time
[for doing power-save or doing some other legacy STA Scan or some other
useful stuff], this makes GO Negotiation process slow.
We wait for 1 second idle and then listen for a random time between
100(min)-300(max) ms. Assume P1 is in WAIT_PEER_CONNECT state and P2 is
the one which is now to send go_neg frame. If P2 sends GO Negotiation
frame just at the boundary of 300 ms of P1 and assume that P2 takes
close to 600-800 ms for one iteration of sending go_neg request (one
iteration is GO Negotiation Request frame time + dwell time +
listen_time), P2 needs to transmit at least 16-18 Action frames for
hitting the listen time of P1.
Following patch reduces the idle time to 500 ms. Alternatively we can
increase the listen time interval to 500 ms just for WAIT_PEER_CONNECT
state.
P2P: Fix PROBE_REQ_ONLY flag use for Provision Discovery Request
Provision discovery from a known peer should actually check for
dev->flags & P2P_DEV_PROBE_REQ_ONLY. This is creating an issue of
updating the listen frequency of peer with the PD request frame
frequency. PD request frame will be sent by the peer on our local listen
frequency. This patch fixes that error. Suggested check has already been
implemented in the invitation req receive path.
Jouni Malinen [Sun, 18 Dec 2011 15:21:25 +0000 (17:21 +0200)]
P2P: Fix Provision Discovery channel for some join-GO cases
The Provision Discovery Request needs to be sent on the operating
channel of the GO and as such, the frequency from the BSS table
(scan results) need to override the frequency in the P2P peer
table that could be based on the Listen channel of the GO.
wpa_supplicant/dbus AP: Add support for WPS Internal Registrar
When in AP mode, wpa_supplicant is now enabling WPS (only Internal
Registrar). WPS.Start() call can be used to initiate WPS negotiation
similarly to how this is done in station mode.
Reinette Chatre [Sun, 18 Dec 2011 14:52:33 +0000 (16:52 +0200)]
P2P: Make GO negotiation peer and group information available over D-Bus
The GO negotiation response is very cryptic at the moment. For a success
message we only know on which interface the negotiation succeeded, not
which peer. For a failure we know the interface also and a status code
(number).
It will be very useful for clients to know upon receipt of such a message
which peer the negotiation occurred with.
Now that the peer information is available and the API is changed
already, the function composing the D-Bus message might as well include
all GO negotiation information. This is done with a dict to make things
easier on clients if this result information changes down the line.
Signed-hostap: Reinette Chatre <reinette.chatre@intel.com>
Signed-hostap: Johannes Berg <johannes.berg@intel.com>
Marek Kwaczynski [Sun, 18 Dec 2011 14:38:48 +0000 (16:38 +0200)]
P2P: Do not include own information in the peer table
When the station is connected to P2P GO after calling p2p_find command
the device sees itself. It is related to lack of filtering itself from
clients connected to P2P GO.
Jouni Malinen [Sun, 18 Dec 2011 12:44:03 +0000 (14:44 +0200)]
PCSC: Accept 0x67 (Wrong length) as a response to READ RECORD
It looks like some USIM cards respond with 0x67 (Wrong length) instead
of 0x6c to 00 b2 01 04 ff. This was getting rejected in
scard_get_record_len(). ETSI TS 102 221 is not very clear on this
detail, but it looks fine to accept the 0x67 error value, too, to learn
the record length.
The whole wpa_supplicant_ctrl_iface_ctrl_rsp_handle() function operates
on the ssid->eap field which exists only if IEEE8021X_EAPOL has been
defined. Therefore the whole function body needs to be enclosed within
an #ifdef/endif block.
Signed-hostap: Antonio Quartulli <ordex@autistici.org>
Jouni Malinen [Wed, 14 Dec 2011 23:06:02 +0000 (01:06 +0200)]
Android: Fix PNO start function conversion
The wpa_hexdump_ascii() call did not get converted properly and this
was missed becaused of it getting defined out from the build. Anyway,
this better use the correct variable names should that debug print
ever be enabled for Android.
Johannes Berg [Sun, 11 Dec 2011 17:57:50 +0000 (19:57 +0200)]
IBSS: fix RSN key initialisation
Antonio reported that RSN IBSS failed to work.
We traced it down to a GTK failure, and he then
bisected it to commit bdffdc5ddb0c838af4c90d11:
"AP: Reorder WPA/Beacon initialization".
The reason this commit broke it is that the state
machine's GInit variable is never set to false as
wpa_init_keys() never gets called, and thus new
keys are generated every time the state machine
executes.
Fix this by calling wpa_init_keys() when the new
group has been initialised.
Reported-by: Antonio Quartulli <ordex@autistici.org> Tested-by: Antonio Quartulli <ordex@autistici.org>
Signed-hostap: Johannes Berg <johannes.berg@intel.com>
Jouni Malinen [Sun, 11 Dec 2011 16:30:47 +0000 (18:30 +0200)]
Disassociate when starting WPS search
Previously, the WPS scans could have been done in associated state if we
happened to be associated when the request to use WPS was received. This
can slow down scanning and end up in unexpected state if no WPS
association is tried. Avoid these issues by disconnecting when WPS
search is started.
Jouni Malinen [Sun, 11 Dec 2011 16:06:34 +0000 (18:06 +0200)]
P2P: Fix 32-bit compiler warnings on service discovery reference
Convert core wpa_supplicant code to use u64 instead of void * for the
P2P service discovery reference. Use uintptr_t in type casts in
p2p_supplicant.c to handle the conversion without warnings.
Note: This needs to be revisited for 128-bit CPU where sizeof(void *)
could be larger than sizeof(u64).
P2P: Remove unexpected pending Provision Discovery Request in Search
A Pending Provision Discovery Request was sent in SEARCH phase after a
previous provision discovery timeout. Fix this by resetting the config
method of P2P device in the pending PD reset function. This avoids the
sending of a pending Provision Discovery Request during the next P2P
search.
Michael Braun [Sun, 11 Dec 2011 11:01:57 +0000 (13:01 +0200)]
Allow WPA passphrase to be fetched with RADIUS Tunnel-Password attribute
This allows per-device PSK to be configured for WPA-Personal using a
RADIUS authentication server. This uses RADIUS-based MAC address ACL
(macaddr_acl=2), i.e., Access-Request uses the MAC address of the
station as the User-Name and User-Password. The WPA passphrase is
returned in Tunnel-Password attribute in Access-Accept. This
functionality can be enabled with the new hostapd.conf parameter,
wpa_psk_radius.
Signed-hostap: Michael Braun <michael-dev@fami-braun.de>
Vitaly Wool [Sun, 11 Dec 2011 10:03:18 +0000 (12:03 +0200)]
Skip WPS PBC overlap detection if P2P address is the same
WPS overlap detection can detect false overlap if a P2P peer
changes UUID while authentication is ongoing. Changing UUID
is of course wrong but this is what some popular devices do
so we need to work around it in order to keep compatibility
with these devices. There already is a mechanism in WPS
registrar to skip overlap detection if P2P addresses of two
sessions match but it wasn't really triggered because the
address wasn't filled in in the caller function.
Let's fill in this address and also clean up WPS PBC sessions
on WSC process completion if UUID was changed.
Arik Nemtsov [Sat, 10 Dec 2011 14:50:00 +0000 (16:50 +0200)]
nl80211: Propagate Probe Response offload capabilities from kernel
Translate nl80211 flags to wpa_supplicant flags for Probe Response
offload support. The existence of the nl80211 PROBE_RESP_OFFLOAD_SUPPORT
attribute means Probe Response offload is supported. The value of the
attribute is a bitmap of supported protocols.
Jouni Malinen [Sat, 10 Dec 2011 10:56:42 +0000 (12:56 +0200)]
P2P: Clean up group formation on network block removal
If a P2P group network block is removed for any reason (e.g., wps_cancel
command) while the interface is in group formation, remove the group
formation timeout and indicate failure immediately. Previously, this
type of operations could end up leaving the timeout running and result
in somewhat unexpected group formation failure events later.
Jithu Jance [Sat, 10 Dec 2011 10:26:00 +0000 (12:26 +0200)]
P2P: Append P2P Device Address to AP-STA-DISCONNECTED event
Append "p2p_dev_addr" parameter to AP-STA-DISCONNECTED event for P2P
connections. In addition, for AP-STA-CONNECTED event during P2P
connection, the "dev_addr=" print is replaced with "p2p_dev_addr=" to
be more consistent with other events.
Johannes Berg [Sat, 10 Dec 2011 09:56:31 +0000 (11:56 +0200)]
nl80211: Store own address in BSS
Storing the address in the BSS instead of the DRV struct makes it usable
for hostapd and thus gets rid of the linux_get_ifhwaddr() call when
receiving a spurious frame.
Signed-hostap: Johannes Berg <johannes.berg@intel.com>
Jouni Malinen [Thu, 8 Dec 2011 22:15:04 +0000 (00:15 +0200)]
Add MSK dump mechanism into hostapd RADIUS server for testing
Testing code can now be enabled in the hostapd RADIUS server to dump
each derived MSK into a text file (e.g., to be used as an input to
wlantest). This functionality is not included in the default build
and can be enabled by adding the following line to hostapd/.config:
CFLAGS += -DCONFIG_RADIUS_TEST
The MSK dump file is specified with dump_msk_file parameter in
hostapd.conf (path to the dump file). If this variable is not set,
MSK dump mechanism is not enabled at run time.
Jouni Malinen [Tue, 6 Dec 2011 19:57:17 +0000 (21:57 +0200)]
P2P: Add group ifname to P2P-PROV-DISC-* events
If Provision Discovery Request is sent for GO role (i.e., P2P Group ID
attribute is included), add the group interface name to the control
interface event on the GO. This makes it easier to figure out which
ctrl_iface needs to be used for wps_pbc/wps_pin command to authorize
the joining P2P client.
Jithu Jance [Tue, 6 Dec 2011 19:28:02 +0000 (21:28 +0200)]
P2P: Add optional "join" argument for p2p_prov_disc command
This can be used to request Provision Discovery Request to be sent
for the purpose of joining a running group, e.g., to request the GO
to display a PIN that we can then use with p2p_connect join command.
Jouni Malinen [Tue, 6 Dec 2011 19:13:54 +0000 (21:13 +0200)]
Remove documentation for label option in p2p_connect
P2P use cases do not allow use of Label config method and the earlier
code for this has already been removed, but this documentation was not
updated at the same time.
Since the nl_cache is not used anymore, there is no need for maintaining
the struct nl80211_handles wrapper for struct nl_handle. Clean this up
by using nl_handle directly.