[lwip-devel] [task #10088] Correctly implement close() vs. shutdown()

[Simon Goldschmidt]

Follow-up Comment #11, task #10088 (project lwip):

> I agree that, for native API apps, close is completely 
> equivalent to shutdown(SHUT_RDWR). In a standard BSD sockets 
> implementation, the two should be equivalent with respect to the 
> protocol stack; the only significant difference is that close 
> also deallocates the socket file descriptor.

The actual change to implement shutdown is rather simple (implemented with a
flag). However, there might be a problem for raw API apps to know when the PCB
is actually deallocated (and may not be referenced by the application any
more). We can implement this is in 2 ways:

a) Keep the current behaviour. This means that after successfully calling
tcp_shutdown(RDWR), the applicaiton may not reference the PCB any more and it
will be deallocated eventually (no more callbacks are made!).

b) Change the behaviour to *not* automatically deallocate the PCB unless
tcp_close() is called.

Solution a) has the disadvantage that when tcp_shutdown(WR) puts the PCB into
FIN_WAIT_1 or LAST_ACK, in which it will be closed eventually. This might not
be a problem for applications, but it violates TCP behaviour in that the
connection is closed too early (if it is idle).

    _______________________________________________________

Reply to this item at:

  <http://savannah.nongnu.org/task/?10088>

_______________________________________________
  Nachricht geschickt von/durch Savannah
  http://savannah.nongnu.org/