| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
For adapters that do not support LE and ones where LE hasn't been
enabled we shouldn't be trying to initiate background scanning. This
patch adds an extra check to the hci_update_background_scan() to ensure
that we bail out if HCI_LE_ENABLED is not set.
Since we do allow user space to feed the kernel with LE connection
parameters even when LE is not enabled we now need to also call
hci_update_background_scan() as soon as LE gets enabled so that scanning
gets started if necessary.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
If we have both LE scanning and advertising simultaneously enabled we
need a way to tell hci_connect_le() in which role to initiate a
connection. This patch adds a new parameter to the function to give it
the necessary information. For auto-connect and mgmt_pair_device we
always use master role, whereas for L2CAP users (in practice sockets) we
use slave role whenever HCI_ADVERTISING is set and master role
otherwise.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
The auth_type value which gets assigned to hci_conn->auth_type is
something that's only used for BR/EDR connections and is of no value for
LE connections. It makes therefore little sense to pass it to the
hci_connect_le() function. This patch removes the parameter from the
function.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Enabling passive scanning always when we're connectable aligns us with
the BR/EDR page scanning. This is also consistent with the fact that the
code dealing with passive scanning results will actively try to connect
any direct advertising event when we're connectable.
This patch implements the feature by adding the connectable condition to
hci_update_background_scan() checks for starting scanning and by calling
hci_update_background_scan() whenever the connectable state changes.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
If we have an active discovery going on we shouldn't do any changes to
LE scanning when hci_update_background_scan() is called (a call which
can happen for many different reasons). This patch fixes the issue by
returning from the function if the discovery state is anything else
except DISCOVERY_STOPPED.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
In a subsequent patch the hci_update_background_scan() function will
depend on being able to know the current discovery state. For this to be
possible we need to set the new state early in the function. Since we
also need to check what the old state was this patch introduces an extra
variable for tracking it.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
This reverts commit 277d916fc2e959c3f106904116bb4f7b1148d47a as it was
at least breaking iwlwifi by setting the IEEE80211_TX_CTL_NO_PS_BUFFER
flag in all kinds of interface modes, not only for AP mode where it is
appropriate.
To avoid reintroducing the original problem, explicitly check for probe
request frames in the multicast buffering code.
Cc: [email protected]
Fixes: 277d916fc2e9 ("mac80211: move "bufferable MMPDU" check to fix AP mode scan")
Signed-off-by: Johannes Berg <[email protected]>
|
|
Since the only caller of this function doesn't care about the return
value anymore let's just remove it.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Now that we have the process_adv_report() function doing the IRK lookup
and updating the bdaddr we don't need to do this anymore in mgmt.c in
the mgmt_device_found() function.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
If we have devices set as to be connected we should connect to them even
during normal discovery if we get a connectable advertising event. If we
also have HCI_CONNECTABLE set we should connect ADV_DIRECT_IND events
even to devices that we don't have in our pend_le_conns list. This patch
implements such behavior by passing the advertising report type to
check_pending_le_conn() and calls that function regardless of what type
of scanning we are doing.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
The hdev->send driver callback is mandatory to be provided by a driver
before calling hci_register_dev. So enforce it and return EINVAL in
case it is not available.
All existing drivers are providing this callback anyway, so this is
just an extra sanity check.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
The drivers are allowed to just return an error from hdev->send callback
and in that case the driver does not own the SKB. Which means that the
caller has to free the SKB.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
When we get an LE connection complete event we should restart background
scanning if there are any devices needing it. So far the code was only
making the decision based on whether the completed connection had any
stored parameters or not. This patch ensures that we trigger background
scanning always when necessary.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
When the driver sets HCI_QUIRK_RAW_DEVICE, the controller will be set
as unconfigured. However running the unconfigured init procecure is
not useful since raw-only devices are not allowed to change its
configuration. This change skips the init procedure and just allows
user channel operation for this device.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
When we establish connections as a consequence of receiving an
advertising report it makes no sense to wait the normal 20 second LE
connection timeout. This patch modifies the hci_connect_le function to
take an extra timeout value and uses a lower 2 second timeout for the
auto-connection case. This timeout is intentionally chosen to be just a
bit higher than the 1.28 second timeout that High Duty Cycle Advertising
uses.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
When an unconfigured controllers reaches the configured state, it is
important to change the HCI_RAW flag. It indicates to userspace that
the controller is fully operational.
External configuration allows to bring the controller back into an
unconfigured state. In that case make sure HCI_RAW flag is set again.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
This adds support for changing the public device address. This feature
is required by controllers that do not provide a public address and
have HCI_QUIRK_INVALID_BDADDR set.
Even if a controller has a public device address, this is useful when
an embedded system wants to use its own value. As long as the driver
provides the set_bdaddr callback, this allows changing the device
address before powering on the controller.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
When the external configuration triggers the switch to a configured
controller, it means the setup needs to be run. Controllers that start
out unconfigured have only run limited set of HCI commands. This is
not enough for complete operation and thus run the setup procedure
before announcing the new controller index.
This introduces HCI_CONFIG flag as companion to HCI_SETUP flag. The
HCI_SETUP flag is only used once for the initial setup procedure. And
during that procedure hdev->setup driver callback is called. With the
new HCI_CONFIG the switch from unconfigured to configured state is
triggering the same setup procedure just without hdev->setup. This
is required since bringing a controller back to unconfigured state
from configured state is possible.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
When calling Device Remove with BDADDR_ANY we should in a similar way
emit Device Removed events as we do when removing a single device. Since
we have to iterate the list and call device_removed() the dedicated
hci_conn_params_clear_enabled() is not really useful anymore. This patch
removes the helper function and does the event emission and list item
removal in a single loop.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
BR/EDR has the connection request and connection request rejection, but
LE doesn't have anything similar. We still request LE connections to
blocked devices to be disconnected but it's possible that ACL data slips
through before that. The check in l2cap_recv_frame really only needs to
be for LE and not BR/EDR because of this.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
During the setup phase of a controller, the Bluetooth address will be
read and to have that original address available for later use, store
it as setup address.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
For an unconfigured controller it is required to read at least the
local version information. If the set_bdaddr driver callback is
provideded, then also the local Bluetooth address will be read.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
The HCI event callbacks have grown over the last years and some
functions handle status checking different than others. For the
simple ones, check the status at the beginning and exit if an
error with the HCI command occured.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
When one or more of the missing configuration options change, then send
this even to all the other management interface clients.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
Move the mgmt_event function higher up in the code so that no forward
declaration is needed.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
The Set External Configuration management command allows for switching
between configured and unconfigured start if HCI_QURIK_EXTERNAL_CONFIG
is set by the transport driver.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
For the quirks that are allow to be set during setup callback, the
check needs to be modified so that they are applied even if no
setup callback provided by the driver.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
When a controller requires external configuration, then setting this
quirk will allow indicating this.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
Reading the controller configuration information is a valid management
command when the controller is unconfigured. Allow this command, but
return invalid index on all other commands.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
When a Bluetooth controller does not have a valid public Bluetooth
address, then allow the driver to indicate this. If the quirk is
set, the Bluetooth core will switch to unconfigured state first
and will allow userspace to configure the address before starting
the full initialization of the controller.
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
It's wasteful to use list_del_init (which re-initializes the list_head)
if we're just about to free the element and never use it again.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
If a found device is marked as blocked while doing passive LE scanning,
neither report it nor try to connect to it.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Instead of looking through the entire list of entries we can more
efficiently use the new hci_pend_le_action_lookup() function to look up
entries specifically in the pend_le_reports list. Since the search is
now limited to the right list we can also remove an unnecessary check
for list_empty() before the lookup.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
In some circumstances we need to look up entries in pend_le_conns and in
other in pend_le_reports. This patch converts the existing lookup
function for pend_le_conns to something that can be used for both lists.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Since there are no more users of this function we can simply go ahead
and remove it.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
This is the last place using hci_pend_le_conn_add() and we can more
just as simply manipulate the list directly here.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
When powering off (hci_dev_do_close) we should clear both the
pend_le_reports and pend_le_conns types of entries. When powering on
respectively we should populate both lists. This patch converts the
hci_pend_le_conns_clear() function into hci_pend_le_actions_clear()
(which can now be static) and converts the restart_le_auto_conns()
function into restart_le_actions().
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Now that there are no-longer any users of the hci_pend_le_conn_del()
function we can simply go ahead and remove it.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Since params->action is used for both the pend_le_conns and
pend_le_reports lists we can simplify the adding and deleting of the
lists considerably. For example, when deleting entries in most
situations we no-longer need to check the auto_connect value but can
directly proceed with calling list_del_init on param->action (which is
safe even if the entry is not part of any list).
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
When hci_conn_params_set() is called if the new auto_connect value is
the same as the old one we don't need to take any action. Simply return
success from the function in this case.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Now that pend_le_connections is a list of hci_conn_params entries we can
simply remove items from that list as we iterate through the global list
of le_conn_params. This also moves the responsibility of calling
hci_update_background_scan() to the functions that were previously
calling hci_pend_le_conns_clear(). The only user that's left for
hci_pend_le_conns_clear() is hci_dev_do_close() which anyway does not
need to call hci_update_background_scan().
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
To simplify manipulation and lookup of hci_conn_params entries of the
type HCI_AUTO_CONN_REPORT it makes sense to store them in their own
list. The new action list_head in hci_conn_params is used for this
purpose.
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
In preparation to store also HCI_AUTO_CONN_REPORT entries in a list it
makes sense to convert the existing pend_le_conn list head of
hci_conn_params into a more generically named "action". This makes sense
because a parameter entry will never participate in more than one action
list.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
If we're doing passive scanning we shouldn't proceed with any of the
code that deals with active scanning (pending reports, etc.). This patch
fixes a missing return statement for the passive scanning section in the
process_adv_report() function.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
The Read Controller Configuration Information command allows retrieving
details about possible configurations option. The supported options are
returned and also the missing options (if any).
Signed-off-by: Marcel Holtmann <[email protected]>
Signed-off-by: Johan Hedberg <[email protected]>
|
|
Since we only store entries with identity addresses in the
le_conn_params and pend_le_conns lists we can avoid unnecessary lookups
by checking for an identity address before diving into the lists
themselves.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
If a device is in the pend_le_conns list it cannot at the same time also
have the need to be notified through mgmt_device_found. By making
check_pending_le_conn return whether it found an entry or not we can
avoid unnecessary checks in process_adv_report().
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
Since the connection parameters are always a basis for adding entries to
hdev->pend_le_conns (so far of type bdaddr_list) it's simpler and more
efficient to have the parameters themselves be the entries in the
pend_le_conns list. We do this by adding another list_head to the
hci_conn_params struct.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
When processing passive scanning results we need the resolved identity
address both in check_pending_le_conn() as well as later in
process_adv_report(). Since process_adv_report() calls
check_pending_le_conn() we can simply resolve the IRK earlier in the
function and thereby eliminate a second lookup.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|
|
When calling Remove Device for an entry using HCI_AUTO_CONN_REPORT we
need to decrement the pend_le_reports value correspondingly. This patch
fixes one such missing action in the Remove Device command handler.
Signed-off-by: Johan Hedberg <[email protected]>
Signed-off-by: Marcel Holtmann <[email protected]>
|