I fully agree with this. AF_BLUETOOTH is already in place, but it
only implements "BTPROTO_L2CAP" all of the GATT and
advertising/scanning related functionality is implemented over
"BTPROTO_HCI"
#define BTPROTO_L2CAP
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_L2CAP> 0
#define BTPROTO_HCI
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_HCI> 1
#define BTPROTO_SCO
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_SCO> 2
#define BTPROTO_RFCOMM
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_RFCOMM> 3
#define BTPROTO_BNEP
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_BNEP> 4
#define BTPROTO_CMTP
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_CMTP> 5
#define BTPROTO_HIDP
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_HIDP> 6
#define BTPROTO_AVDTP
<https://elixir.bootlin.com/linux/latest/C/ident/BTPROTO_AVDTP> 7
I have been thinking about how the socket interface would be
implemented. I believe that is one area were I could make a good
contribution to this effort.
Currently, there is very little use of the socket protocol under net/.
The only use of the protocol argument now is to verify that it has a
valid value.
'struct socket' remembers the protocol family (PF_BLUTOOTH), and the
socket type (SOCK_RAW), but not the protocol. So, the protocol field is
essentially forgotten after the socket is instantiated. In the
proposed Bluetooth usage, we would have to extend the usage of the
protocol in some fashion to accomplish what you suggest.
Two possibilities come to mind:
* Separate network drivers for each protocol. The supported
protocol(s?) would be passed to netdev_register() and would be used
to match a socket interface with a network device. In this case, the
BTPROTO_L2CAP and the BTPROTO_HCI support would be provided in
separately registered drivers (which might be two registrations of
the same driver exposing the two different "personalities" and two
different netdriver_s structures).
* A common network device for all protocols. This would require some
mechanism for passing the protocol on a packet-by-packet basis, this
would probably be a new field in both struct socket and struct
netdriver_s. The protocol would probably need to be written into the
netdriver_s structure instance along all other packet information.
That might be awkward under certain conditions.
