186

Unix/Linux 网络时间协议版本3 Network Time Protocol Version 3 (NTP) distribution for Unix systems

Received: from mail.eecis.udel.edu by whimsy.udel.edu id aa26947;
          20 Mar 1998 02:03 EST
Received: from green.images.com (green.images.com [198.207.178.4]) by orange.images.com (8.8.5/SCO5) with ESMTP id XAA16154 for <[e-mail]>; Thu, 19 Mar 1998 23:03:00 -0800 (PST)
Received: (from kamal@localhost) by green.images.com (8.8.5/SCO5) id XAA11828; Thu, 19 Mar 1998 23:02:55 -0800 (PST)
Message-ID: <19980319230254.52091@green>
Date: Thu, 19 Mar 1998 23:02:54 -0800
From: Kamal A Mostafa <[e-mail]>
To: [e-mail]
Cc: Kamal A Mostafa <[e-mail]>
Subject: xntp3-5.92 select error logging bug
References: <19980318231931.26036@green> <[e-mail]>
Mime-Version: 1.0
Content-Type: multipart/mixed; boundary=zhXaljGHf11kAtnf
X-Mailer: Mutt 0.89i
In-Reply-To: <[e-mail]>; from [e-mail] on Thu, Mar 19, 1998 at 08:01:07PM -0500


--zhXaljGHf11kAtnf
Content-Type: text/plain; charset=us-ascii

-----BEGIN PGP SIGNED MESSAGE-----

Hi Harlan-

I'm now replicating the TIMER_ENQUEUE core dump myself with 5.92 (instead of
relying on my colleague's reports) but it looks like for some reason 5.92d
doesn't exhibit the bug (even without my TIMER_ENQUEUE fix).  Oh Joy.  ;-|

However, while peeking at 'xntpd -d' output, I found (and fixed) a new
buglet.  It seems that on my systems, whenever I use -d, I get an error
message logged to syslog *once every second*!...

	select(): nfound=-1, error: Interrupted system call
	last message repeated 31 times
	...

Now that's a lotta syslog!  ;-}

The problem is some badly structured code in worker_thread() in ntpd.c.
Essentially, if you do not HAVE_SIGNALED_IO and you have set -d, then you
will log this completely normal event.

To make matters worse, somebody's mis-coded attempt to prevent this logging
for SYS_VXWORKS looks to me like it might break SYS_VWORKS altogether!  Have
a look at this...

	      else if (debug)
    #ifndef SYS_VXWORKS
		msyslog(LOG_DEBUG, "select(): nfound=%d, error: %m", nfound);
    #endif
    #else
	      wait_for_signal();
    #endif
	      if (alarm_flag)               /* alarmed? */
		  ...

Note that if you are SYS_VXWORKS, then you miss the semi-colon at the end of
the msyslog(), thus never closing the body of the "if (debug)"!  Since you
won't be doing the #else clause, you'll only end up trying the
"if (alarm_flag)" clause if you're in debug mode!

I have removed the offending junk (and added comments to the #else and #endif
lines) -- patch attached.

BTW, how do you prefer to receive patches?  Are the MIME attachements I've
been sending okay, or would you prefer uuencode'ings?


Okay, now that *that* nastiness is out of the way!...  I'm still not sure
that 5.92 and 5.92d are working properly for me in broadcastclient mode.  A
client system running 5.92d seems to receive packets from the broadcasting
server also running 5.92d, but it (the client) never appears to
clock_filter() or clock_update() or anything...  Just lots of this pattern:

[Note that the server is broadcasting every 16 seconds, not the default 64(?).]

    input_handler: fd=7 length 48 from 8493111e 132.147.17.30
    input_handler: fd=7 length 48 from 8493111e 132.147.17.30
    sendpkt(fd=5 132.147.17.30, 132.147.17.18, ttl=-8, 48)
    input_handler: fd=5 length 48 from 8493111e 132.147.17.30
    input_handler: fd=7 length 48 from 8493111e 132.147.17.30
    input_handler: fd=7 length 48 from 8493111e 132.147.17.30
    sendpkt(fd=5 132.147.17.30, 132.147.17.18, ttl=-8, 48)
    ...


Even after it's been running awhile, 'xntpdc -p' on the client just says:

         remote           local      st poll reach  delay   offset   disp
    ======================================================================
    =kestrel.pdev.sc 132.147.17.255  16   16    0  0.0000  0.000000 16.000


It would help me a bunch if I knew what code-path it was *supposed* to be
taking...  What does a "normal" -d log from an broadcast client look like?
Could you send me a sample?  What does 'xntpdc -p' say on such a client, for
that matter?

Another stupid question...  What's up with the sendpkt() lines?  I wouldn't
intuitively expect a broadcast client to *ever* send anything back to a
broadcasting server, but that -d output above shows that it does appear to be
sending something back to it's server periodically.  Why?  What is it
sending?



On Thu, Mar 19, 1998 at 08:01:07PM -0500, [e-mail] wrote:
> Kamal,
> 
> Try compiling xntpd with -DTIMERQUEUE_DEBUG, which should enable some
> debugging assertions in ntp_timer.c, if I'm not mistaken.
> 
> Harlan

Thanks much, I'll check it out.  Despite the fact that my core dump seems
to have gone away by itself with 5.92d, I really don't like the look of
that code in TIMER_ENCODE (and we still might want to de-macro-fy it for
debug purposes, anyway).

 -k

-----BEGIN PGP SIGNATURE-----
Version: 2.6.2

iQCVAwUBNRIUQYrzk78EeS5RAQHa/AQAt46LoFAK4pkSofnQIvVl4z0Moyv4jIMQ
Up7O/dYe+TuDoBtITuD6fXAjz13tEtzaHGtCUTx3AyjF8wJ7Qq4QXDwus96bzfRV
nmpJgHjOpmsZV8j01yQuvBmIaI2xliq93oN+ZTD+OOTpR2H04/7mWo0hBksiWi6a
l2rRHySNj7o=
=ZKu0
-----END PGP SIGNATURE-----

--zhXaljGHf11kAtnf
Content-Type: text/plain; charset=us-ascii
Content-Description: ntpd.c patch
Content-Disposition: attachment; filename="ntpd.patch"

*** ntpd.c.orig Thu Mar 19 20:47:49 1998
--- ntpd.c Thu Mar 19 21:39:42 1998
***************
*** 770,780 ****
  #endif /* SYS_WINNT */
  		   )
  	    msyslog(LOG_ERR, "select() error: %m");
! 	  else if (debug)
! #ifndef SYS_VXWORKS
! 	    msyslog(LOG_DEBUG, "select(): nfound=%d, error: %m", nfound);
! #endif
! #else
  	  wait_for_signal(); 
 #endif
  	  if (alarm_flag)		/* alarmed? */
--- 770,777 ----
  #endif /* SYS_WINNT */
  		   )
  	    msyslog(LOG_ERR, "select() error: %m");
! 
! #else /* HAVE_SIGNALED_IO */
  	  wait_for_signal();
  #endif /* HAVE_SIGNALED_IO */
  
***************
*** 776,782 ****
  #endif
  #else
  	  wait_for_signal();
! #endif
  	  if (alarm_flag)		/* alarmed? */
  	    {
  	      was_alarmed = 1;
--- 773,780 ----
  
  #else /* HAVE_SIGNALED_IO */
  	  wait_for_signal();
! #endif /* HAVE_SIGNALED_IO */
! 
  	  if (alarm_flag)		/* alarmed? */
  	    {
  	      was_alarmed = 1;

--zhXaljGHf11kAtnf--