On Sun, 28 Dec 2003 13:43:26 +0100
mouss <[EMAIL PROTECTED]> wrote:
> Well, tcpdump (on both sides) will show what happened during the session 
> termination: was there any RST? is there a timeout issue? were FINs 
> acked or not?
> 
> Also, can you do your tests with other servers such as ftp.freebsd.org, 
> ftp.netbsd.org, ftp.free.fr?
> 
> This in turn will allow people here to better understand the "problem" 
> and see whether it can be "fixed" or not.
> 
> I've seen similar logs for http traffic (certainly related to the "stop" 
> button of the browser), but I didn't really check the details.
> 
> mouss
> 
> 
> fbsd_user wrote:
> 
> > Sorry but I do not agree with your opinion. Another reader posted he
> > gets the same results as I do, and I responded to him through this
> > list with more conversation about were the bug is and what the
> > circumstances are that causes it. Running an tcpdump is not going to
> > show me how the 'ipnat ftp/proxy' temp rules are launched and
> > removed with in the timing of the close session request, that's
> > happening internal to ipnat processing. My log shows me enough to
> > know where the problem is. Before one says right out of hand that
> > there is no bug, one should first try to reproduce it and use what
> > ever tools they think are necessary to disprove it, which may lead
> > to another person verifying it's a bug.
> > 
> 

Well, I know the phnomena that tcp sessions have often "troubles"
on closing. Probably for over five years. And that they are caused
by tcp implementation differences. You should have had the time to
think why results differ server by server or client by client, not
session by session.

BTW, you made other wrong observations:

o ipfilter and ipnat has nothing to do at the session clousre. Actually
  ipnat is more "conservative" due to fear of lost FIN etc. When you
  logged ipfilter, you could have logged filter and nat state transitions.

o active data connection is just another tcp connection. There is no
  difference from other tcp connections, sans direction. Also closing data
  connection doesn't mean closing control connection (consider ls for
  example), so that the 'troubles" on control connections you see are
  made by ipfilter or ipnat is unlikely.

o assumption ftp client always closes session is not true. The server
  could initiate close session when it knows EOF earlier than client.

o Your reading of Eric's message is just the opposite, whose observation
  matched what I made a few years ago.

o I did not say it's not a bug. I said you don't show strong evidence
  to support your strong word "bug". You merely submitted phaenomenal
  observations which are known for years.

o So, finally, you don't have different opinion from mine.



horio shoichi

Reply via email to