Re: [lwip-users] Global reference to pbufs

[Enrico Murador - Research & Development - CET]

Hello,
I'm sorry, I would need some more information...

I would know wether it is ok to call tcp_close() everytime the receive callback is called with NULL data pointer,
and wether it is safe to call tcp_write() (and tcp_output()) for a pcb, outside of callbacks, regardless of the
pcb actual condition (e.g. when closing... Providing that the pcb is not deallocated, of course).

Thank you very much
Enrico

On 16/05/2011 14.41, Enrico Murador - Research & Development - CET wrote:

This is all a bit of an API violation, so I wouldn't do it that way.

Ok. I placed these "checks" since my first tries with lwip, after I found that I was trying to
close a pcb that was already deallocated... All the code for catching the pcb closure
was already there, but maybe there was some "hole" in my program flow...

The PCB should remain valid after you have accepted it until either (i)
the application closes it (and so you should already know about this);
or (ii) the stack closes it (e.g. because it receives a reset from the
other end) and it will then call your tcp_err() callback (I think, from
memory, so worth checking) to tell you about this.  If you use this
callback to update your application state then you should find that the
PCBs you are referencing are always valid.

Ok. So, there are three "catch-points": application close, receive callback with NULL data,
error callback...

Be careful about the context you are using to access these PCBs.  For
example, if you are closing connections from the accept callback (as
described above) this will be in the context of whatever is executing
the stack; you need to make sure that other contexts aren't also using
that PCB to send data.

All my TCP handling is placed in a single "thread" so, provided that I will not try to send data on a
"just deallocated" pcb reference (checking reference validity in the send routine), there should be no
problems, right?
This leads to the other question regarding tcp_write(): is it legal to call tcp_write() (and eventually
also tcp_output()) for a pcb, outside of callbacks, regardless of the pcb actual state?

Thank you
Enrico

--
Enrico Murador
R&D

CET Electronics via E. Fermi, 1 31050 Zenson di Piave (TV) Italia Phone: +39 0421 344100 Fax: +39 0421 464042 Web: http://www.cet-electronics.com/