Re: [dcp] A new DCP Implementation

["Patrick R. McManus" <mcmanus@ducksong.com>]

[pat atducksong: Sat, May 18, 2002 at 10:59:16PM -0400]

> I've been working on a DCCP implementation for research
> purposes. Judging by the "release early and often" standard it is time
> to share what I've got. (Judging by any other sane standard of quality
> control it's far too early ;).


It has been a little while since I sent an update, so I thought I'd do
that. I haven't been able to invest a whole lot of time in DCCP, but
there has been some significant progress over the last 10 days or so:

* listen()
* poll() [and there was much rejoicing..]
* service names [1]
* payload allowed on both request and response [2]
* lookaside buffer cache allocations
* numerous crashing bugs
* a memory leak or two
* hashes on the lookup lists

The patches are still against linux's 2.4.18 vanilla kernel and are
always linked off of:  http://www.ducksong.com:81/dccp/

I want to mention the API I used for a couple of features, as it does
impact user space. Probably not the kind of thing apropos for the
specification, but it would be nice if different implementations had
complimentary apis - these aren't set in stone (heck the whole thing
is just a research implementation) so please comment.

[1] service names - I introduced a new structure
struct sockaddr_dcp
{
  struct sockaddr_in in;
  unsigned int service;
};

which may be passed to bind() or connect() in lieu of the more
traditional sockaddr_in. Its completely compatible with a ported UDP
ap that uses sockaddr_in though, the stack just assumes service name
of 0 in that case.

[2] handshake payloads.

Initially, the server needs a way to identify whether or not it should
accept data attached to a request or if it should answer with
option=discard (option 1).. if a client gets the discard option on the
response packet it will resend the data on with its ack. The server's
socket can be told to allow/disallow request data via a setsockopt

     setsockopt (s, SOL_DCCP, SO_ALLOW_HANDSHAKE_DATA, (void *)  &y, sizeof (int)));

similarly, setsockopt is used to "preload" data before calling
connect() or accept()..

     setsockopt (s, SOL_DCCP, SO_HANDSHAKE_DATA, (void *) buf, 225);

the data on a bound socket will apply to every accept() done off of
it.. though that data can be modified/canceled by subsequent
setsockopt() calls.

no actual data is returned by the accept() or connect() calls even if
the other side has sent some on the handshake - its just put into the
buffers for the next recv().

-P


_______________________________________________
dcp mailing list
dcp@ietf.org
https://www1.ietf.org/mailman/listinfo/dcp