This patch adds epoll option for the eloop implementation. This can be
selected with the CONFIG_ELOOP_EPOLL=y build option.
[merit]
See Table1.
Table1. comparison table
+--------+--------+-----------+------------+-------------+
| | add fd | remove fd | prepare fd | dispatch fd |
+--------+--------+-----------+------------+-------------+
| select | O(1) | O(1) | O(N) | O(N) |
+--------+--------+-----------+------------+-------------+
| poll | O(1) | O(1) | O(N) | O(N) |
+--------+--------+-----------+------------+-------------+
| epoll | O(1) | O(1) | 0 | O(M) |
+--------+--------+-----------+------------+-------------+
"add fd" is addition of fd by eloop_sock_table_add_sock().
"remove fd" is removal of fd by eloop_sock_table_remove_sock().
"prepare fd" is preparation of fds before wait in eloop_run().
"dispatch fd" is dispatchment of fds by eloop_sock_table_dispatch().
"N" is all watching fds.
"M" is fds which could be dispatched after waiting.
As shown in Table1, epoll option has better performance on "prepare fd" column.
Because select/poll option requires setting fds before every select()/poll().
But epoll_wait() doesn't need it.
And epoll option has also better performance on "dispatch fd" column.
Because select/poll option needs to check all registered fds to find out
dispatchable fds. But epoll option doesn't require checking all registered fds.
Because epoll_wait() returns dispatchable fd set.
So epoll option is effective for GO/AP functionality.
[demerit]
The epoll option requires additional heap memory. In case of P2P GO, it is
about 8K bytes.
Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
The group object is already registered on DBus at that point, thus wpa_s
structure holds its path already.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Fixes a minor mistake: the p2p_info structure should be used here
instead of the peer_handler_args one.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
These signals delivers an array of key/value pairs, thus declaring those
as it should.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
When there is a pending WPS negotiation for P2P and signal interrupt is
triggered, invalid read occurs in wpas_wps_terminate_pending() if
wpas_p2p_disconnect() removed the interface. Inverting deinit order
solves the issue.
Signed-off-by: Eduardo Abinader <eduardo.abinader@openbossa.org>
The radio works for the interface get removed if interface is disabled.
This could have left P2P module in invalid state if the interface got
disabled during a p2p_find or p2p_listen operation. Clear the state in
such a case to avoid blocking following operations due to P2P module
assuming it is still in progress of doing something.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This new command can be used to simulate driver events without having to
go through the driver wrapper or kernel code for this. This enables more
testing coverage with hwsim.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
It looks like there was a possible sequence for wpa_s->scan_req to be
MANUAL_SCAN_REQ at the moment a GO is to be started. This could result
in the "Request scan (that will be skipped) to start GO" to actually not
skip the scan and end up stuck waiting for something external to trigger
a scan before the GO could be started. Fix this by clearing
wpa_s->scan_req when deciding to start the GO.
This issue could be hit at least by first enabling autoscan and then
issuing P2P_GROUP_ADD. Other sequences that set wpa_s->scan_req to
MANUAL_SCAN_REQ without going through wpa_supplicant_scan() to clear it
immediately could also have similar effect (and there is even a small
window for the wpa_supplicant_scan() call to happen only after the
P2P_GROUP_ADD command is processed, so this could potentially have
happened even with SCAN + P2P_GROUP_ADD).
Signed-off-by: Jouni Malinen <j@w1.fi>
Commit 41d5ce9e0b was intended to scan for
GO on the negotiated channel for few iterations, but it did not work
correctly due to incorrect operator being used. Fix this by requiring
both conditions to be met for the single channel scan.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
On receiving CHANNEL_LIST_CHANGED event from driver, verify that local
GO (if any) is operating in valid frequency. If not, we should remove
the group and reform on valid frequency. Indicate this similarly to the
avoid-frequency notification (i.e., a control interface message for
upper layers to react to this for now; potentially CSA later).
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
This command allows to copy network variable from one network to
another, e.g., to clone the psk field without having to extract it from
wpa_supplicant.
Signed-off-by: Dmitry Shmidt <dimitrysh@google.com>
Valgrind indicates reference to already freed memory if function
wpa_config_remove_network() is called prior to calling
wpa_supplicant_deauthenticate(), and this can lead to a crash.
Inverting the call order fixes the problem.
Signed-off-by: Hannu Mallat <hannu.mallat@jollamobile.com>
The new command line argument -e can be used to request the server to
send EAP-Key-Name in Access-Accept. If both the local EAP peer
implementation and server provide the EAP Session-Id, compare those
values and indicate in debug log whether a match was seen.
Signed-off-by: Jouni Malinen <j@w1.fi>
This extends Interworking network selection to enable FT-EAP as an
optional key_mgmt value to allow FT to be used instead of hardcoding
WPA2-Enterprise without FT.
Signed-off-by: Jouni Malinen <j@w1.fi>
No need to call ibss_rsn_deinit() again since it is already being called
by wpa_supplicant_mark_disassoc().
Signed-off-by: Eduardo Abinader <eduardo.abinader@openbossa.org>
Do not perform extended listen period operations when either a P2P
connection is in progress. This makes the connection more robust should
an extended listen timer trigger during such an operation.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
No need to use ENOBUFS within core wpa_supplicant, so just replace it
with -1 to work arounds MinGW build issues.
Signed-off-by: Jouni Malinen <j@w1.fi>
Add a new wowlan_triggers option to wpa_supplicant.conf. The triggers in
this key will be used to configure the kernel wowlan configuration.
For now, support only simple flags. More complex triggers can be added
later on.
Signed-off-by: Eliad Peller <eliadx.peller@intel.com>
This was supposed to end with a newline character so that parameters are
aligned in configuration file.
Signed-off-by: Sreenath Sharma <sreenats@broadcom.com>
The previous check for dl_list_len() or having an entry from the list is
sufficient, but some static analyzers cannot figure out that
dl_list_first() and dl_list_last() will return non-NULL in this type of
cases. Avoid invalid reports by explicitly checking for NULL.
Signed-off-by: Jouni Malinen <j@w1.fi>
The check based on last_scan_res_used is sufficient for making sure that
last_scan_res is allocated. However, it is a bit too complex for static
analyzers to notice, so add an explicit check to avoid bogus reports.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
It is possible for the disconnection event from the driver to not get
delivered when interface is disabled. To maintain consistent ctrl_iface
event behavior, indicate CTRL-EVENT-DISCONNECTED in such a case if we
were in connected state.
Signed-off-by: Jouni Malinen <j@w1.fi>
If two config files are merged together, the ssid and cred blocks may
not have been processed correctly since the tail pointers were not
updated to the last entry from the first configuration file.
Signed-off-by: Jouni Malinen <j@w1.fi>
Previously, the frequency conflict was handled only during the
association flow. However, some drivers, e.g., mac80211 based
drivers, will fail an authentication request in case that there
are no available channels for use (as they might be used by
other interfaces), and thus the frequency conflict resolution
is never called.
Fix this by calling frequency conflict resolution during
authentication (SME-in-wpa_supplicant) as well.
In addition, get the shared radio frequency from the wpa_s context in
both the SME-in-driver and SME-in-wpa_supplicant cases and not from the
driver.
Signed-off-by: Ilan Peer <ilan.peer@intel.com>
The normal scan counter is used to trigger a few normal scans before a
scheduled scan is issued. The reason for doing this is that we get
results faster and some drivers support more SSIDs per normal scan
than during sched scan. After 2 normal scans, we start using
sched_scan. But the problem is that when a connection succeeds, we do
not reset this counter, so we will keep using sched scans.
To fix this, now we reset the counter when we switch to WPA_COMPLETED
state.
Signed-off-by: Luciano Coelho <luciano.coelho@intel.com>
This avoids large number of undesired compiler warnings since Android
build system is added -Wextra.
Signed-off-by: Greg Hackmann <ghackmann@google.com>
This allows the specific SIM to be identified for authentication
purposes in multi-SIM devices. This SIM number represents the index of
the SIM slot. This SIM number shall be used for the authentication using
the respective SIM for the Wi-Fi connection to the corresponding
network.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The previous version prevented all use of wpas_global_ctrl_iface_set().
That's not desirable since there may be more global parameters added in
the future. Instead, try to use the global version first and redirect to
P2P interface only if the global version returns an error.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The P2P redirection for SAVE_CONFIG issued on the global control
interface was preventing wpas_global_ctrl_iface_save_config() from being
reached. The global version of SAVE_CONFIG was supposed to try to save
configuration files for all interface rather than just the P2P
management interface, so fix this by removing the unneeded and undesired
redirection.
Modify the global SAVE_CONFIG handler to return FAIL if no configuration
files were saved. This makes the behavior match with the per-interface
SAVE_CONFIG.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Rename the eap_proxy_*.mk file to eap_proxy_*.mak for non-Android builds
so that the same eap_proxy implementation can be more easily included in
the same directory for both cases.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
Previously, it was possible to remove the main interface and leave
behind dynamic P2P group interfaces. This would eventually result in
references freed memory, so it is not really suitable behavior. Instesd,
remove all the dynamic P2P group interfaces that were created through an
interface that is now removed.
Signed-off-by: Jouni Malinen <j@w1.fi>
If we receive a scan request via D-Bus and wpa_supplicant_scan_trigger
fails, return the failure to the caller over D-Bus.
Signed-hostap: mukesh agrawal <quiche@chromium.org>
Use eloop timeout to run the reconnect command after eapol_cb has
returned. This reduces stack use and can simplify the driver
command/event sequence by allowing the disconnection events from the
provisioning step to be processed before starting the actual data
connection.
Signed-off-by: Jouni Malinen <j@w1.fi>
If a P2P GO issues multiple credentials for some reason, clone the
p2p_group and temporary parameters to all those based on the initial
network block that gets used for the first credential.
Signed-off-by: Jouni Malinen <j@w1.fi>
This used to work, but it looks like the disabled-PSK-network check
ended up rejecting confiurations that configured a wildcard SSID with a
passphrase (instead of PSK).
Signed-off-by: Jouni Malinen <j@w1.fi>
