Google SoC lwIP Report Week 3

by zhu48 | June 12, 2016

In week 3, I discovered a lot more about how data is being passed around.

After more in depth inspection of the memory, I realized that Winsock was passing on the correct IP address; I just wasn't aware of the data structure it was using. After modifying my driver to correctly extract the IP address from the IRP, my test server and client performed a successful TCP connection hand shake, with my client program exiting correctly. However, when I attempted to run the client program again, the operating system crashes.

To investigate this issue, I separated the client and server programs into their individual system calls to see what IRQ each system call sends out to the driver. This investigation revealed that I was handling socket binding wrong. On the bind call, I'm supposed to create a transport address. On the connect call, another IRQ creates a connection context. It is this connection context against which a connection to the server is made. In my code, I did not take into account that each TCP socket can have multiple connections associated with it.

The latter part of the week was spent looking at how trunk handles the two different IRQs - the one that create a transport address and the one that creates a connection context, and how I should link those IRQs to lwIP. I am not yet finished with this investigation. Next week, I will continue trying to finish socket creation and connection establishment, taking into account the fact that each TCP socket can hold multiple connections.