Re: notification has not yet been cleaned in server side when ble is disconnected.

[Yunhan Wang <yunhanw@nestlabs.com>]

Hi, Luiz

On Tue, Oct 10, 2017 at 3:07 AM, Luiz Augusto von Dentz
<luiz.dentz@gmail.com> wrote:
> Hi Yunhan,
>
> On Tue, Oct 10, 2017 at 11:48 AM, Yunhan Wang <yunhanw@nestlabs.com> wrot=
e:
>> Hi=EF=BC=8C Luiz
>>
>> Here there is no pair operation in our end. Our rough idea is the below.
>> 1. ble scan
>> 2. ble connect from central to peripheral.
>> 3. central app writes data in characteristic A offered by peripheral
>> 4. central app subscribes characteristic B offered by peripheral app,
>> which is used to receive indication data from peripheral.
>>
>> I am using 2 btvirt, one is 00:AA:01:00:00:23(peripheral), another is
>> 00:AA:01:01:00:24 (central)
>>
>> For previous reproduced step:
>> 1. Assume server has one characteristic A that is subscribed by client
>> and sending indication,
>> 2. Then ble connection is disconnected by client directly or
>> connection disappear without unsubscribing A.
>> 3. Client tries to connect to ble successfully
>> 4. Client fails to subscribe that characteristic A although ble
>> connection is ok.
>> 5. Another retry to make Ble connection and subscribe to A can work.
>>
>> The below logs in bluetoothd are from step 2: ble disconnect seems not
>> clean anything related to gatt notification.
>>
>> src/gatt-database.c:send_notification_to_device() GATT server sending in=
dication
>> src/gatt-database.c:conf_cb() GATT server received confirmation
>> src/device.c:gatt_debug() Write Req - handle: 0x003f
>> src/gatt-database.c:send_notification_to_device() GATT server sending in=
dication
>> src/gatt-database.c:conf_cb() GATT server received confirmation
>> src/service.c:change_state() 0x925b20: device 00:AA:01:00:00:23
>> profile gap-profile state changed: connected -> disconnecting (0)
>> src/service.c:change_state() 0x925b20: device 00:AA:01:00:00:23
>> profile gap-profile state changed: disconnecting -> disconnected (0)
>> src/adapter.c:dev_disconnected() Device 00:AA:01:00:00:23 disconnected, =
reason 2
>> src/adapter.c:adapter_remove_connection()
>> plugins/policy.c:disconnect_cb() reason 2
>> src/adapter.c:bonding_attempt_complete() hci1 bdaddr 00:AA:01:00:00:23
>> type 1 status 0xe
>> src/device.c:device_bonding_complete() bonding (nil) status 0x0e
>> src/device.c:device_bonding_failed() status 14
>> src/adapter.c:resume_discovery()
>> src/device.c:att_disconnected_cb()
>> src/device.c:att_disconnected_cb() Connection reset by peer (104)
>> src/service.c:change_state() 0x9103e0: device 00:AA:01:01:00:24
>> profile gap-profile state changed: connected -> disconnecting (0)
>> src/service.c:change_state() 0x9103e0: device 00:AA:01:01:00:24
>> profile gap-profile state changed: disconnecting -> disconnected (0)
>> src/gatt-client.c:btd_gatt_client_disconnected() Device disconnected.
>> Cleaning up.
>> src/device.c:att_disconnected_cb() Automatic connection disabled
>> attrib/gattrib.c:g_attrib_unref() 0x91d190: g_attrib_unref=3D0
>> src/adapter.c:dev_disconnected() Device 00:AA:01:01:00:24 disconnected, =
reason 3
>> src/adapter.c:adapter_remove_connection()
>> plugins/policy.c:disconnect_cb() reason 3
>> src/adapter.c:bonding_attempt_complete() hci0 bdaddr 00:AA:01:01:00:24
>> type 1 status 0xe
>> src/device.c:device_bonding_complete() bonding (nil) status 0x0e
>> src/device.c:device_bonding_failed() status 14
>> src/device.c:att_disconnected_cb()
>> src/device.c:att_disconnected_cb() Connection reset by peer (104)
>> src/gatt-client.c:btd_gatt_client_disconnected() Device disconnected.
>> Cleaning up.
>> src/device.c:att_disconnected_cb() Automatic connection disabled
>> attrib/gattrib.c:g_attrib_unref() 0x925000: g_attrib_unref=3D0
>>
>> The below logs in bluetoothd are from step 4: Client fails to
>> subscribe that characteristic A although ble connection is ok. from
>> the log, it seems it is cleaning gatt notification in
>> https://git.kernel.org/pub/scm/bluetooth/bluez.git/tree/src/gatt-databas=
e.c#n2005
>
> Looks like we are no cleanup up the subscription no matter if paired
> or not which partially explain the problem here.
>
>> src/device.c:gatt_debug() start: 0x0001, end: 0x0005, uuid:
>> 00001800-0000-1000-8000-00805f9b34fb
>> src/device.c:gatt_debug() start: 0x0006, end: 0x0009, uuid:
>> 00001801-0000-1000-8000-00805f9b34fb
>> src/device.c:gatt_client_ready_cb() status: success, error: 10
>> src/gatt-client.c:create_services() Exporting objects for GATT
>> services: 00:AA:01:01:00:24
>> src/device.c:device_svc_resolved() /org/bluez/hci0/dev_00_AA_01_01_00_24=
 err 0
>> src/device.c:gatt_debug() Read Req - handle: 0x0003
>> src/gatt-database.c:gap_device_name_read_cb() GAP Device Name read reque=
st
>> profiles/gap/gas.c:read_device_name_cb() GAP Device Name: N0002
>> src/device.c:btd_device_device_set_name()
>> /org/bluez/hci1/dev_00_AA_01_00_00_23 N0002
>> src/device.c:gatt_debug() Write Req - handle: 0x0009
>> src/gatt-database.c:gatt_ccc_write_cb() CCC write called for handle: 0x0=
009
>> src/device.c:gatt_debug() Registered handler for "Service Changed": 1
>> src/device.c:gatt_debug() Read Req - handle: 0x0005
>> src/gatt-database.c:gap_appearance_read_cb() GAP Appearance read request
>> profiles/gap/gas.c:read_appearance_cb() GAP Appearance: 0x0000
>>
>> src/device.c:gatt_debug() Write Req - handle: 0x0039
>> src/gatt-client.c:notify_client_ref() owner :1.122201
>> src/device.c:gatt_debug() Write Req - handle: 0x0037
>> src/gatt-database.c:gatt_ccc_write_cb() CCC write called for handle: 0x0=
037
>> src/gatt-client.c:notify_client_unref() owner :1.122201
>> src/gatt-client.c:notify_client_free() owner :1.122201
>> src/device.c:gatt_debug() Write Req - handle: 0x0037
>> src/gatt-database.c:gatt_ccc_write_cb() CCC write called for handle: 0x0=
037
>> src/gatt-database.c:ccc_write_cb() External CCC write received with
>> value: 0x0000
>
> Well this is an unsubscribe value:0x0000 so by itself this is no
> wrong, please check with btmon if the remote is not unsubscribing
> before subscribing again.
>

Yes, normally the remote needs to unsubscribe the characterstic
gracefully, then everything would be fine, yes, i have done that in my
end. However, if something wrong happens in the remote or someone
moves the remote device far away from the peripheral, the client would
be disconnected abnormally without unsubscribing, and the peripheral
would fail to process the later ble connection if this connection is
subscribing the same characteristic.

I think we need cleanup functionality inside bluez when disconnect
happens, in addition it seems there is no grace way to cleanup
characteristic from application using any bluez dbus API.

thanks
best wishes
yunhan
>> src/gatt-database.c:pipe_hup() 0x921290 closed
>> src/gatt-database.c:proxy_removed_cb() Proxy removed - removing
>> service: /org/bluez/xxxx/xxx
>> src/gatt-database.c:gatt_db_service_removed() Local GATT service removed
>> src/gatt-database.c:send_notification_to_device() GATT server sending in=
dication
>> src/gatt-database.c:client_disconnect_cb() Client disconnected
>> src/advertising.c:client_disconnect_cb() Client disconnected
>> src/device.c:gatt_debug() Service Changed received - start: 0x0034 end: =
0x0039
>> src/gatt-database.c:conf_cb() GATT server received confirmation
>> src/device.c:gatt_debug() Read By Grp Type - start: 0x0034 end: 0x0039
>> src/device.c:gatt_debug() Primary service discovery failed. ATT ECODE: 0=
x0a
>> src/device.c:gatt_debug() Read By Grp Type - start: 0x0034 end: 0x0039
>> src/device.c:gatt_debug() Secondary service discovery failed. ATT ECODE:=
 0x0a
>> src/device.c:gatt_debug() Read By Type - start: 0x0000 end: 0x0000
>> src/device.c:gatt_debug() service disappeared: start 0x0034 end 0x0039
>> src/device.c:gatt_service_removed() start: 0x0034, end: 0x0039
>>
>> Any idea?
>>
>> Thanks
>> Best wishes
>> Yunhan
>>
>> On Tue, Oct 10, 2017 at 1:19 AM, Luiz Augusto von Dentz
>> <luiz.dentz@gmail.com> wrote:
>>> Hi Yunhan,
>>>
>>> On Tue, Oct 10, 2017 at 11:02 AM, Yunhan Wang <yunhanw@nestlabs.com> wr=
ote:
>>>> Hi, Luiz
>>>>
>>>> Please ignore the previous email.
>>>>
>>>> I am seeing a bluez gatt notification problem in gatt-database.c
>>>>
>>>> Here is scenario:
>>>>
>>>> 1. Assume server has one characteristic A that is subscribed by client
>>>> and sending notification,
>>>> 2. Then ble connection is disconnected by client directly or
>>>> connection disappear without unsubscribing A.
>>>> 3. Client tries to connect to ble successfully
>>>> 4. Client fails to subscribe that characteristic A although ble
>>>> connection is ok.
>>>> 5. Another retry to make Ble connection and subscribe to A can work.
>>>>
>>>> It seems ble disconnect at step 2 only disconnect ble connection and
>>>> does not cleanup notify_io and stop
>>>> notification in step 3 at code,
>>>> https://git.kernel.org/pub/scm/bluetooth/bluez.git/tree/src/gatt-datab=
ase.c
>>>>
>>>> It seems retry in step 4 would trigger ccc_write_cb and release
>>>> notify_io and stop notification so that step 5 would succeed.
>>>> partial log from step 4:
>>>> ccc_write_cb() External CCC write received with value: 0x0000
>>>> pipe_hup() 0x921290 closed
>>>>
>>>> Any idea?
>>>
>>> Do you have the full logs, this might be a bug since perhaps we
>>> cleanup the notification itself though if the device is paired the
>>> subscription should be persisted.
>>>
>>> --
>>> Luiz Augusto von Dentz
>
>
>
> --
> Luiz Augusto von Dentz