patacongo edited a comment on issue #1639:
URL:
https://github.com/apache/incubator-nuttx/issues/1639#issuecomment-682148993
> Do you think it would be reasonable to adapt this interface?
My preference would not be to make any major redesign of the networking.
Change struct sock_intf_s effects everything. I don't think that is called
for. I think it only requires extension of the existing interface.
The si_setup() method is the one that configures the socket for the address
family, socket type, and protocol. That is where the additional protocols and
channels would have to be handled. But I think it need only retain this
information in struct bluetooth_conn_s which part of the socket data structure
that holds information unique to an address family.
For Bluetooth sockets, none of this information matters until we get to the
network driver layer. I don't think that the selected Bluetooth protocol
necessarily has any impact on the network itself. However, when a frame is
sent or received, then it does effect the network driver and Bluetooth stack
significantly.
I think it is l2 layer where the additional protocols would have to be
implemented; I think the socket layer just has to verify and remember the
protocol and channel and coordinate this with the network driver / bluetooth
stack.
NOTE: I tend to refer everything below the logic under net/ as L2. The
architecture is odd for Bluetooth because the Bluetooth stack is not integrated
into the NuttX network stack. Normally, the network driver is at the L2 layer,
but not here. The network driver sits above the Bluetooth stack and the entire
Bluetooth stack is treated as the MAC sub-layer. This is a little unusual, but
I think think is a good partitioning.
User-Space
`- BSD Socket Layer
`- NuttX raw packet support (very minimal)
`- Network driver
`- Bluetooth stack
`- HCI driver
It was originally done that way to support 6LoWPAN sockets on top of L2CAP.
User-Space
`- BSD Socket Layer
`- IPv6
`- 6LoWPAN
`- Network driver
`- Bluetooth stack
`- HCI driver
struct sock_intf_s is the interface between the _user_ BSD sockets and the
per-address-family socket implementation. I don't think that should change. I
think the changes need to be in struct netdriver_s which is the interface
between the network and the Bluetooth network driver and stack.
It is only the Bluetooth network driver that can tap into the Bluetooth
stack to support each protocol; It is only the Bluetooth stack that can manage
the channel behavior.
Therefore, nothing much of substance has to change under net/. All of the
changes have to occur in the interface between net/ and the network drivers, in
the network driver, and the Bluetooth stack.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
