Re: [PATCH v3] Bluetooth: Add HCI_AUTO_CONN_DIRECT_REPORT_IND

[Alfonso Acosta <fons@spotify.com>]

I was informally describing the HCI exchange between the device , not
the userspace calls. Sorry I didn't make that clear.

On Mon, Sep 29, 2014 at 4:35 PM, Johan Hedberg <johan.hedberg@gmail.com> wrote:
> Hi Alfonso,
>
> On Mon, Sep 29, 2014, Alfonso Acosta wrote:
>> * With 0x02, (HCI_AUTO_CONN_ALWAYS) there is auto-connection upon
>> ADV_DIRECT_IND but replacing the device's batteries  (power-on
>> ADV_IND) causes an infinite connection loop: connect->encryption
>> on->disconnect(due to "encryption on" failing)->connect .... Getting
>> the report data inside the  Device Connected event won't help because
>> it will never get to that point.
>
> Are you sure that you've mapped out the order of events correctly for
> this case? The way things should be going (based on going through the
> code - I haven't verified this in practice) on the user space side is:
>
> 1. mgmt_ev_device_connected (connected_callback() in adapter.c)
> 2. Connection indication on ATT server socket (src/attrib-server.c)
> 3. device_attach_attrib() called (due to step 2)
> 4. bt_io_set(io, ..., BT_IO_SEC_MEDIUM, ...);
>
> The last step above is what instructs the kernel to initiate encryption.
> If the mgmt_ev_device_connected would contain hints telling us to call
> mgmt_unpair + mgmt_pair instead of trying to elevate security your use
> case should be solvable.
>
> If the above order is not what's happening we might be having a race
> condition between the mgmt and ATT sockets. This race *should* be
> fixable by ensuring that the mgmt socket has higher priority in the
> mainloop than the ATT socket.  We used to have it like that in the past,
> but we might have lost it along the way because shared/mgmt.c that uses
> shared/io.h.
>
> As a quick hack to check if these sockets are indeed racing against each
> other during the same mainloop iteration you could try changing
> G_PRIORITY_DEFAULT to G_PRIORITY_HIGH in the io_set_read_handler()
> function in src/shared/io-glib.c (the ATT socket shouldn't be using that
> but shared/mgmt.c is).
>
> Johan



-- 
Alfonso Acosta

Embedded Systems Engineer at Spotify
Birger Jarlsgatan 61, Stockholm, Sweden
http://www.spotify.com