Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uwm.edu!gem.mps.ohio-state.edu!ginosko!uunet!mcsun!ukc!acorn!john
From: john@acorn.co.uk (John Bowler)
Newsgroups: comp.windows.x
Subject: Re: Using xdm on IBM RTs running AOS 4.3/Sept 88
Keywords: xdm, IBM RT/PC, IBM RT 6152
Message-ID: <836@acorn.co.uk>
Date: 18 Sep 89 17:37:29 GMT
References: <EHRLICH.89Sep15084200@shire.cs.psu.edu>
Reply-To: john@acorn.UUCP (John Bowler)
Organization: Acorn Computers Ltd, Cambridge, UK
Lines: 167

In article <EHRLICH.89Sep15084200@shire.cs.psu.edu> ehrlich@cs.psu.edu (Daniel Ehrlich) writes:
>
>We have a number of IBM RT 6152 Academic Systems running AOS 4.3 (aka
>BSD 4.3).  It would be nice if we could coerce our users into using X
>windows on these machines.  Xdm looked like the way to go, but we have
>noticed that when a user logs out zombied processes tend to collect
>until the process table is full and the machine must be rebooted.
>
>Xdm is set up to restart, rather than reset, the server as the Xibm
>server tends to grow without bound if not restarted.  Has anyone else
>tried using xdm on RTs running AOS?

No - but I have suffered from zombied xdm's on a system where the server
shuts down instead of resetting when all the client connections are
closed.  (This was done for precisely the same reasons - to ensure that
a new X session didn't inherit an enormous server from a previous one).

The problem is that xdm suffers from at least one race condition when
the server dies - heavily loaded machines in particular are likely to
promote it.  The following (context) diffs are to display.c with patches
1-8 (neither patch 9 nor 10 affect it).  They fix the problem on Acorn's
system:-

*** /tmp/,SMSt1022767	Mon Feb 27 19:44:09 1989 (version with patches 1-8)
--- display.c	Thu Jun  1 19:31:11 1989
***************
*** 38,44 ****
  
  static jmp_buf	terminated;
  
! static CatchTerm (), someoneDied (), abortOpen (), StartServer ();
  static WaitForServer (), TerminateServer (), HupServer (), StartSession ();
  
  extern unsigned sleep ();
--- 38,44 ----
  
  static jmp_buf	terminated;
  
! static CatchTerm (), CatchChild (), abortOpen (), StartServer ();
  static WaitForServer (), TerminateServer (), HupServer (), StartSession ();
  
  extern unsigned sleep ();
***************
*** 58,70 ****
  	abort ();
  }
  
! static int	someoneDead;
  
  static
! someoneDied ()
  {
  	Debug ("someone died\n");
! 	someoneDead = 1;
  }
  
  ManageDisplay (d)
--- 58,77 ----
  	abort ();
  }
  
! /*
!  * This modification is to deal with a server (the ARM one) which
!  * insists on shutting down when the last client closes the display.
!  * This exposes an interesting race/bug in this code - WaitForServer
!  * fails to clean up (because the SIGCHLD handler is removed - below).
!  */
! static int	deaths;
! static int	burials;
  
  static
! CatchChild ()
  {
  	Debug ("someone died\n");
! 	++deaths;
  }
  
  ManageDisplay (d)
***************
*** 86,92 ****
  	}
  	(void) signal (SIGTERM, CatchTerm);
  	(void) signal (SIGHUP, CatchHup);
! 	(void) signal (SIGCHLD, someoneDied);
  	(void) signal (SIGPIPE, SIG_IGN);
  	/*
  	 * Step 4: Start server control program
--- 93,99 ----
  	}
  	(void) signal (SIGTERM, CatchTerm);
  	(void) signal (SIGHUP, CatchHup);
! 	(void) signal (SIGCHLD, CatchChild);
  	(void) signal (SIGPIPE, SIG_IGN);
  	/*
  	 * Step 4: Start server control program
***************
*** 95,101 ****
  		Debug ("aborting display %s\n", d->name);
  		exit (1);
  	}
- 	(void) signal (SIGCHLD, SIG_DFL);
  	/*
  	 * keep a session running on this display
  	 */
--- 102,107 ----
***************
*** 109,114 ****
--- 115,121 ----
  			else
  				continue;
  		}
+ 		++burials;
  		if (pid == sessionPid) {
  			Debug ("session died %s\n", d->name);
  			switch (waitVal (status)) {
***************
*** 266,274 ****
  				RegisterCloseOnFork (ConnectionNumber (dpy));
  				return 1;
  			}
! 			if (someoneDead) {
  				pid = wait ((waitType *) 0);
! 				if (pid == serverPid) {
  					Debug ("server died\n");
  					return 0;
  				}
--- 273,282 ----
  				RegisterCloseOnFork (ConnectionNumber (dpy));
  				return 1;
  			}
! 			if (burials < deaths) {
  				pid = wait ((waitType *) 0);
! 				if (pid != (-1)) ++burials;
! 				if (pid == serverPid || pid == (-1) && errno == ECHILD) {
  					Debug ("server died\n");
  					return 0;
  				}
***************
*** 285,292 ****
  	Debug ("giving up on server\n");
  	LogError ("server open failed for %s, giving up\n", d->name);
  	pid = 0;
! 	if (someoneDead)
! 		pid = wait ((waitType *) 0);
  	if (pid != serverPid)
  		TerminateServer (d, serverPid);
  	return 0;
--- 293,301 ----
  	Debug ("giving up on server\n");
  	LogError ("server open failed for %s, giving up\n", d->name);
  	pid = 0;
! 	while (burials < deaths && pid != serverPid)
! 		if ((pid = wait ((waitType *) 0)) != (-1)) ++burials;
! 		else break;
  	if (pid != serverPid)
  		TerminateServer (d, serverPid);
  	return 0;
***************
*** 338,343 ****
--- 347,353 ----
  				(void) alarm (d->openTimeout);
  				pid = wait ((waitType *) 0);
  				(void) alarm (0);
+ 				if (pid >= 0) ++burials;
  				if (pid == serverPid)
  					break;
  			}