Re: Invalid length of startup packet

On Tue, 28 May 2002, Tom Lane wrote:

> Ron Snyder <snyder@roguewave.com> writes:
> > If we portscan our database server and include the postgres port in the list
> > of scanned ports, we'll get that error in our logs.  You can also get that
> > message by attempting to telnet to your postgres port.  I'm a little
> > surprised that postgres doesn't list some source ip information with the
> > message.
>
> It would be fairly easy to rearrange the code in postmaster.c so that
> a log message like "attempted connection from host xxx port yyy" is
> emitted before we try to receive and interpret the startup packet.
> But I don't think we can guarantee that every error condition that might
> occur in that stretch of code could be modified to include host/port in
> the text of the error message.  So the "attempted connection" message
> would have to be logged unconditionally for every connection attempt.
> Does that seem useful?  Would you want it to be automatic when
> LOG_CONNECTIONS is on, or should it require some higher-than-normal
> debug setting?

I just looked at the ProcessStartupPacket code to see about doing a patch. I
noticed a few elog()s and thought that supplying the remote connection
information to cover all of these would be a bit of a hassle. Then I wondered
if it was necessary since I presumed all connections could be logged, if a
fatal error occurs the connection won't be listed by netstat by the time
someone looks, so all that's needed is that the connection log records the
information.

Obviously, that's not entirely usable if connections are being made and dropped
by applications. So how about elog() sticking the backend slot number into all
messages? Is that sort of information available in elog? (I'm just looking at
that now but can't say after only a quick scan and with my small knowledge of
the backend code)


--
Nigel J. Andrews
Director

---
Logictree Systems Limited
Computer Consultants