- 18 Oct, 2013 2 commits
-
-
Marcel Holtmann authored
The blacklist debugfs should only be created together with the other entries after the setup procedure has been finished. Signed-off-by:
Marcel Holtmann <marcel@holtmann.org> Signed-off-by:
Johan Hedberg <johan.hedberg@intel.com>
-
Marcel Holtmann authored
The device blacklist is not taking care of the address type. Actually store the address type in the list entries and also use them when looking up addresses in the table. This is actually a serious bug. When adding a LE public address to the blacklist, then it would be blocking a device on BR/EDR. And this is not the expected behavior. Signed-off-by:
-
- 17 Oct, 2013 3 commits
-
-
Marcel Holtmann authored
For easier debugging of the current voice setting, expose the value in debugfs if the controller is BR/EDR capable. Signed-off-by:
-
Marcel Holtmann authored
For LE capable controllers, the static address can be configured. For debugging purposes expose the value in debugfs. Signed-off-by:
-
Marcel Holtmann authored
The auto_accept_delay debugfs entry is only valid for BR/EDR capable controllers that also support SSP. If SSP is not available or it is a LE-only single mode controller this value has no affect and so do not expose it. Since the value can be actually changed, switch the permissions to 0644 to clearly indicate that the value is indeed writeable. Signed-off-by:
-
- 16 Oct, 2013 18 commits
-
-
Marcel Holtmann authored
The inquiry_cache debugfs entry is only valid for BR/EDR capable controllers. In case of single mode LE-only controllers that entry is not valid. Move the creating of the debugfs entries to the end of controller init and only create the inquiry_cache entry if BR/EDR is actually supported. At the same time this avoids creating any debugfs entries for AMP controllers since none of the entries are valid there. Signed-off-by:
-
Marcel Holtmann authored
The L2CAP socket parameter for CID are actually provided in little endian. So convert our constants into little endian before comparing them. Signed-off-by:
-
Johan Hedberg authored
There is no need to use a timer since the entire Bluetooth subsystem runs using workqueues these days. Signed-off-by:
-
Johan Hedberg authored
Since the entire Bluetooth subsystem runs in workqueues these days there is no need to use a timer for deferring work. Signed-off-by:
-
Johan Hedberg authored
Right now we do not allow user space to use connection oriented channels on LE, and the only CID that can be used is the Attribute Protocol one. These restrictions went away together with the recent refactoring of the L2CAP code, but this patch puts them back to their appropriate places. Signed-off-by:
-
Johan Hedberg authored
This function should be operating on scan_rsp_data_len and scan_rsp_data and not the advertising data variables. Signed-off-by:
-
Marcel Holtmann authored
This seems to be a left-over. The module parameter enable_hs has been removed, but its extern declaration is still present. It is not needed anymore, so just remove it. Signed-off-by:
-
Johan Hedberg authored
The SMP CID is only defined for LE transports. Instead of returning an error from smp_sig_channel() in this case (which would cause a disconnection) just return 0 to ignore the data, which is consistent with the behavior for other unknown CIDs. Signed-off-by:
-
Johan Hedberg authored
The A2MP CID is only valid for BR/EDR transports. We should ignore A2MP data on non-BR/EDR links and refuse to create an amp_mgr object. Signed-off-by:
-
Johan Hedberg authored
Now that the only reason code this function can return is L2CAP_REJ_NOT_UNDERSTOOD we can just do the necessary assignment without needing a separate function at all. Signed-off-by:
-
Johan Hedberg authored
There is no command handler that would return an EMSGSIZE error, so just remove this mapping from the l2cap_err_to_reason function. Signed-off-by:
-
Johan Hedberg authored
When the reason code in the L2CAP command reject is "invalid CID" there should be four additional bytes of data in the PDU, namely the source and destination CIDs (which should be zero if one or both are not applicable). This patch fixes all occurrences of such errors to return the right kind of PDU. Signed-off-by:
-
Marcel Holtmann authored
The constants for advertising event types have been defined twice. So remove one copy of it. Signed-off-by:
-
Marcel Holtmann authored
Since there is update_scan_rsp_data, it is also better to use the clear name update_adv_data instead of update_ad. Signed-off-by:
-
Marcel Holtmann authored
The scan response data is a better place to store the device name since it has more space available and is also enforcing privacy. When the controller is advertising, the connectable setting decides if ADV_IND or ADV_NONCONN_IND is used. In case of ADV_IND, the remote side is allowed to request the scan response data. Same as with BR/EDR where either EIR is used or a remote name request. In non-connectable mode, the device name is not available since it is not allowed to request scan response data. Same as in BR/EDR where the device is non-discoverable and no name requests are answered. Signed-off-by:
-
Marcel Holtmann authored
On controller power on and when enabling LE functionality, make sure that also the scan response data is correctly set. Signed-off-by:
-
Marcel Holtmann authored
The scan response data needs to be stored in HCI device and so add a buffer for it and also ensure to clear it when resetting the controller. Signed-off-by:
-
Marcel Holtmann authored
Rename the create_ad function into create_adv_data to make it clear that it is used to create the advertising data. This is important since later on a function adding the scan response data will be added. Signed-off-by:
-
- 15 Oct, 2013 17 commits
-
-
Marcel Holtmann authored
The return value of mgmt_new_ltk() function is not used and so just change it to return void. Signed-off-by:
Gustavo Padovan <gustavo.padovan@collabora.co.uk>
-
Marcel Holtmann authored
The return value of mgmt_read_local_oob_data_reply_complete() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_set_local_name_complete() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_set_class_of_dev_complete() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_ssp_enable_complete() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_auth_enable_complete() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_auth_failed() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_pin_code_neg_reply_complete() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_pin_code_reply_complete() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The return value of mgmt_pin_code_request() function is not used and so just change it to return void. Signed-off-by:
-
Marcel Holtmann authored
The A2MP client for L2CAP channels needs to use l2cap_chan_no_resume empty stub function. Signed-off-by:
-
Gustavo Padovan authored
In this case the replacement by l2cap_chan_ready() doesn't change the code flow, the same operations will executed plus two others that have no effect: the use of the parent socket, that a non-oriented channel doesn't have and the reset of conf_state, which is also fine since the connection is ready at this point. Signed-off-by:
-
Gustavo Padovan authored
The wait_ack code has a heavy dependency on the socket data structures and, as of now, it won't be worthless change it to use non-socket structures as the only user of such feature is a socket. Signed-off-by:
-
Gustavo Padovan authored
We need to remove all direct access of struct sock from L2CAP core. This change is pretty simple and just add a new L2CAP channel callback to do the work in the L2CAP socket side. Signed-off-by:
-
Gustavo Padovan authored
As part of the work to remove struct sock from l2cap_core.c and make it more generic we remove in this commit the direct access to sk->sk_sndtimeo member. This objective of this change is purely remove sk usage from l2cap_core.c Now we have a new l2cap ops to get the current value of sk->sndtimeo. A l2cap_chan_no_get_sndtimeo was added for users of L2CAP that doesn't need to set a timeout. Signed-off-by:
-
Gustavo Padovan authored
l2cap_state_change_and_error() introduces the ability to update a l2cap_user with changes in channel's state and error code with just one call. The main reason for this is to avoid race conditions between and setting the state and then the error. Otherwise we would need to release the lock between both operations. This is another step of an ongoing work to make l2cap_core.c totally independent from l2cap's struct sock. Signed-off-by:
-
Gustavo Padovan authored
Instead of creating an new function pointer to report errors we are just reusing state_change for that and there is a simple reason for this, one place in the l2cap_core.c code needs, in a locked sk, set both the sk_state and sk_err. If we create two different functions for this we would need to release the lock between the two operation putting the socket in non desired state. The change is transparent to the l2cap_core.c code, user that only needs to set the state won't need any modification. This is another step of an ongoing work to make l2cap_core.c totally independent from l2cap's struct sock. Signed-off-by:
-
