Outbound port on sockets

[Grant Edwards]

On 2006-09-15, Sergei Organov <osv at javad.com> wrote:

>>>> It's not the issue here, but to specify the outgoing port
>>>> call bind(('', portnum)) before connect().

>> It's an interesting thing to know, but I've been doing TCP
>> stuff for many years and never run across a situation where
>> it's something I needed to do.  If somebody in this thread
>> actually does need to do it, I'd be curious bout why...
>
> Well, one of ftpd implementations I have here (C code from RTEMS) does
> this:
>
>       /* anchor socket to avoid multi-homing problems */
>       data_source = info->ctrl_addr;
>       data_source.sin_port = htons(20); /* ftp-data port */
>       if(bind(s, (struct sockaddr *)&data_source, sizeof(data_source)) < 0)
>         ERROR;
>       ...
>       if(connect(s,
>           (struct sockaddr *)&info->def_addr,
>           sizeof(struct sockaddr_in)) < 0
>       )
>         ERROR;
>
> I've no idea what "multi-homing problems" are, but maybe it gives you
> some hint?

I don't know what "multi-homing problems are either".
Apparently there must be some ftp clients that require the
source port for the data connection to be port 20.

The RFC is pretty vague. It does say the server and clinet but
must "support the use of the default data port [port 20]" or
something like that. But, it's not all all clear to me what
that is supposed to mean.  My reading is that they must support
the default port as the destination port for a data connection
untill it's been changed by receipt of a PORT command.

But, like I said, is very vague, and I suppose some client
implementor could have read it as the server must use the
default data port as the source port for a data connection.

-- 
Grant Edwards                   grante             Yow!  Ha ha  Ha ha Ha ha
                                  at               Ha Ha Ha Ha -- When will I
                               visi.com            EVER stop HAVING FUN?!!