Re: [lwip-devel] ReactOS is now using lwIP for TCP

[address@hidden]

Cameron Gutman wrote:
> I encountered 2 issues that required changes to lwIP. One was a behavior
> change with regards to the TF_CLOSED flag in which we call TCP_EVENT_CLOSED
> before freeing the PCB to make the rest of our IP library aware that the
> PCB's port is no longer bound. I'm not sure whether you want to integrate
> that or not.
I'd be interested in what you need this for. TCP_EVENT_CLOSED() should 
have been called before already (when the FIN initiating the close has 
been received - moving from ESTABLISHED to CLOSE-WAIT). With that 
change, don't you get 2 close calls (i.e. recv-callback with 
pbuf==NULL)? I wouldn't have expected an "application" (to lwIP, that 
is) could need such an information.

However, if you do need it, we could integrate this to the lwIP sources. 
But if so, we should solve it in a different way (not calling 
TCP_EVENT_CLOSED twice for the same pcb).
> There were a few other changes that were made to lwIP but they all were
> temporary and reverted as far as I know. We should be on lwIP 1.4.0, but I'm
> not 100% sure it was synced properly since I didn't sync it myself or review
> the sync in depth.
I just did review the lwIP code from ReactOS svn (using 
svn://svn.reactos.org/reactos/trunk/reactos, is that correct?).
Unfortunately, it is *not* 1.4.0: although after compiling, only the 1st 
issue from the above should make a difference, there are some 
differences in source files which makes it difficult to have a quick 
look at what's been changed from the original lwIP sources.

Any chance to get this 1.4.0 merged in there? Anything I can help?

BTW: did anyone test this on a real machine or on a multicore processor? 
I would have expected lwIP to perform rather poor on a multicore (due to 
the single tcpip_thread).

Simon