[lwip-users] Choosing an API and using callbacks

[Scott Miller]

Hi all,

I've got some questions about which API I ought to be using and how to 
properly receive data when using callbacks. Examples of the netconn API 
seem to be scarce.

Here's my situation: I'm redesigning some embedded projects that run 
under FreeRTOS and formerly used the SiLabs WGM110 WiFi module. The 
WGM110 has a version of lwIP running on the module and it provides a 
binary interface via UART or SPI to the host. I wrote a host-side driver 
for it that provides an API that follows the Berkeley Sockets API, more 
or less, at least for the sending side. The receive side has always been 
done through callbacks. The application code (for example, my HTTP 
client) registers a callback for a TCP connection and the callback is 
called from the network task when data has been received or when there's 
an error.

The WGM110 has been discontinued and I'm switching to the ATWINC3400, 
which is roughly similar but its internal network stack is too limited 
and I want to add support for wired Ethernet anyway, so I'm running the 
WINC3400 in bypass mode and running lwIP on the host.

The netconn callback seems similar to my own callback scheme, but the 
critical difference is that in my implementation the received packet is 
already available and a pointer to the buffer is passed in to the 
callback. It's zero-copy in a limited sense - the callback has to handle 
the packet immediately because it exists in one big DMA buffer. If the 
packet needs to be kept around it has to be copied out.

The issue I'm running into is that it looks like the callbacks for 
netconn happen in the TCP/IP task context, where you're not supposed to 
use netconn API calls. Is there a safe way to receive the data from 
within the callback? Can I mix raw API calls and use tcp_recv() with the 
PCB from the netconn struct? Should I be skipping netconn entirely? Do I 
need to have another task handle the recv calls and just pass it 
notifications from the lwIP callback?

My next concern is that at some point I'm going to have to implement TLS 
and I don't have a good grasp yet of how that works in lwIP. The WGM110 
had very limited support - it could handle a single outbound TLS 
connection and didn't support inbound at all. I see that the altcp API 
seems to be the way to go. Should I be using altcp everywhere that might 
potentially need TLS? Previously all I needed to do was set a single 
flag to enable TLS.

Are there any good, clear example projects I should be looking to as a 
guide?

Thanks,

Scott