Re: Error setting UUIDs discovery filter on big endian systems

[Matt <mwtaylor@gmail.com>]

On 05/12/2018 00:27, Matt wrote:
> Hi Luiz,
>
> On 04/12/2018 19:53, Luiz Augusto von Dentz wrote:
>> Hi Matt,
>> On Mon, Dec 3, 2018 at 3:59 PM Matt <mwtaylor@gmail.com> wrote:
>>> Hi,
>>>
>>> When setting a UUIDs discovery filter I am receiving
>>> "org.bluez.Error.InProgress" on starting a scan (which isn't in 
>>> progress
>>> and will not start unless the discovery filter is cleared).
>>>
>>> I am using bluez 5.50 (on openWRT). The problem seems to only appear on
>>> big endian hardware (tested a couple of recent kernel and bluez
>>> versions), the same software compiled for and tested on little endian
>>> hardware works as expected. The issue can be reproduced in the 
>>> following
>>> way using bluetoothctl:
>>>
>>> # /etc/init.d/bluetoothd restart
>>> # bluetoothctl
>>> [bluetooth]# power on
>>> Changing power on succeeded
>>> [CHG] Controller 00:1A:7D:DA:71:13 Powered: yes
>>> [bluetooth]# menu scan
>>> [bluetooth]# uuids 0000180f-0000-1000-8000-00805f9b34fb
>>> [bluetooth]# back
>>> [bluetooth]# scan on
>>> SetDiscoveryFilter success
>>> Failed to start discovery: org.bluez.Error.InProgress
>>>
>>> Setting an rssi filter does work as expected on my big endian hardware,
>>> it is only the uuids filter that appears to show this problem. The 
>>> issue
>>> appears whether set using bluetoothctl or directly using DBus.
>> Do you have the bluetoothd logs when that happens? I wonder if it is
>> something with our string to UUID conversion.
>>
>>
> When performing the above sequence in bluetoothctl with 'bluetoothd 
> -d' I get the following in my log:
>
> daemon.debug bluetoothd[22991]: src/agent.c:agent_ref() 0xb03e60: ref=1
> daemon.debug bluetoothd[22991]: src/agent.c:register_agent() agent :1.29
> daemon.debug bluetoothd[22991]: src/adapter.c:property_set_mode() 
> sending Set Powered command for index 0
> daemon.debug bluetoothd[22991]: 
> src/adapter.c:property_set_mode_complete() Success (0x00)
> daemon.debug bluetoothd[22991]: src/adapter.c:new_settings_callback() 
> Settings: 0x00000ad1
> daemon.debug bluetoothd[22991]: src/adapter.c:settings_changed() 
> Changed settings: 0x00000001
> daemon.debug bluetoothd[22991]: src/adapter.c:adapter_start() adapter 
> /org/bluez/hci0 has been enabled
> daemon.debug bluetoothd[22991]: src/adapter.c:trigger_passive_scanning()
> daemon.debug bluetoothd[22991]: src/adapter.c:set_discovery_filter() 
> sender :1.29
> daemon.debug bluetoothd[22991]: 
> src/adapter.c:parse_discovery_filter_dict() filtered discovery params: 
> transport: 7 rssi: 32767 pathloss: 32767  duplicate data: false
> daemon.debug bluetoothd[22991]: src/adapter.c:set_discovery_filter() 
> successfully pre-set filter
> daemon.debug bluetoothd[22991]: src/adapter.c:start_discovery() sender 
> :1.29
> daemon.debug bluetoothd[22991]: src/adapter.c:update_discovery_filter()
> daemon.debug bluetoothd[22991]: 
> src/adapter.c:discovery_filter_to_mgmt_cp()
> daemon.debug bluetoothd[22991]: src/adapter.c:trigger_start_discovery()
> daemon.debug bluetoothd[22991]: src/adapter.c:cancel_passive_scanning()
> daemon.debug bluetoothd[22991]: src/adapter.c:start_discovery_timeout()
> daemon.debug bluetoothd[22991]: 
> src/adapter.c:start_discovery_timeout() 
> adapter->current_discovery_filter == 1
> daemon.debug bluetoothd[22991]: 
> src/adapter.c:start_discovery_timeout() sending 
> MGMT_OP_START_SERVICE_DISCOVERY 127, 7, 1
> daemon.debug bluetoothd[22991]: 
> src/adapter.c:start_discovery_complete() status 0x0d
> kern.err kernel: [709588.482104] Bluetooth: service_discovery: 
> expected 4100 bytes, got 20 bytes
> daemon.debug bluetoothd[22991]: src/agent.c:agent_disconnect() Agent 
> :1.29 disconnected
> daemon.debug bluetoothd[22991]: src/agent.c:agent_destroy() agent :1.29
> daemon.debug bluetoothd[22991]: src/agent.c:agent_unref() 0xb03e60: ref=0

If I instead try a uuid filter of 3 uuids in length I get this in the log:

bluetoothd[550]: src/adapter.c:start_discovery_timeout() sending 
MGMT_OP_START_SERVICE_DISCOVERY 127, 7, 3
kernel: [1077807.129187] Bluetooth: service_discovery: expected 12292 
bytes, got 52 bytes
bluetoothd[550]: src/adapter.c:start_discovery_complete() status 0x0d

So something about the size calculation is going wrong on big endian, 
from adapter.c (line 1607) the command is:

|

mgmt_send(adapter->mgmt,  MGMT_OP_START_SERVICE_DISCOVERY,
		adapter->dev_id,  sizeof(*sd_cp)  +  sd_cp->uuid_count  *  16,
		sd_cp,  start_discovery_complete,  adapter,  NULL);|

With 1 uuid the bytes expected is being returned as 4100 (should be 4 + 
1*16 = 20), with 3 uuids it is 12292 (should be 4 + 3*16 = 52). These 
are the numbers that would be returned if the endian of the uuid_count 
(or the number 16) was switched, i.e. 1 becomes 256, 3 becomes 768. I'm 
not sure why this should be happening.

I have naively tried changing `cp->uuid_count = uuid_count;` to 
`cp->uuid_count = htobs(uuid_count);` (line 2079), just to see what 
would happen by switching the endian, this stops that kernel error 
appearing but just creates an error further along.

bluetoothd[4333]: src/adapter.c:start_discovery_timeout() sending 
MGMT_OP_START_SERVICE_DISCOVERY -45, 7, 256
bluetoothd[4333]: src/adapter.c:start_discovery_complete() status 0x03
bluetoothd[4333]: Wrong size of start discovery return parameters

Perhaps there is some place that is just missing some endian switching 
that would fix this? Sorry, I am not experienced with bluez or related 
code to know where this could be but I would be happy to test any 
suggestions or help in some other way with fixing this issue if someone 
could advise.

Thanks,

Matt