Conference noted::dnu_osi

    Are we talking ULTRIX or OSF here?  On ULTRIX, dlogind does a
    "kill(0, SIGKILL)".  On OSF, it doesn't send any signal.  However,
    it does do a "revoke()" on the pty, which apparently telnetd 
    doesn't do.
    
    The man page for ps says:
    
    [Digital]  The system puts exiting child processes in the <defunct>
    state if their parent process is still running and has not caught the
    SIGCHLD signal or executed a wait() system call.
    
    I would expect the parent of these problem processes to be some
    shell, not dlogind.  In any event, dlogind does catch SIGCHLD
    signals and does a waitpid() in the handler routine.
    
    Hope this helps.
    
    vince
    

    Hi Vince,
    
    >> Are we talking ULTRIX or OSF here?  
    
    Sorry I forget to mention it was on Digital Unix v3.2x !
    
    >> On OSF, it doesn't send any signal.  However, it does 
    >> do a "revoke()" on the pty, which apparently telnetd
    >> doesn't do.
    
    Interesting !
    
    >> I would expect the parent of these problem processes to be some
    >> shell, not dlogind.  In any event, dlogind does catch SIGCHLD
    >> signals and does a waitpid() in the handler routine.
    
    You are correct, the parent process of the application is a shell
    (I think it's "sh" but I should check...)
    
    But if we know that dlogind does a revoke() on the pty (I don't
    know what exactly it means in term of child-parent relationship)
    how would you explain that the child of the shell (the applic.)
    goes into the <defunct> state if both the dlogind and the parent
    shell process do disappear ?
    
    I would expect the applic to become child of the "init" process
    and be "cleaned" by the "init"'s SIGCHLD signal handler !?
    
    As I mentioned in .0, it does work for LAT and telnet connections:
    all the processes are properly killed.
    So do you think it's an "init" bug with revoke() or could it be related
    to the applic programming ?
    
    Thanks for your help,
    
    					============================
    					Alain MARTIN/SSG Switzerland

A guess is that the revoke is preventing the SIGHUP from being sent to
the foreground process group.   dlogind should do a kill(0, SIGHUP)
before doing the revoke (and maybe a sleep(1) too.). 

    I'm willing to code up Matt's suggestion if the customer is
    willing to give it a try.  --  Vince
    

    I can ask the customer to test it if you do modify dlogind...
    
    Let me know when the fix is available.
    
    Cheers,				============================
    					Alain MARTIN/SSG Switzerland

    Hi,
    
    Copy over netrix::test/dlogind.note2121.Z and give it a try.  Don't
    forget dlogind needs to be set uid root.
    
    Vince
    

    Hi Vince,
    
    We tried the modified dlogind and 8-( the same...
    When the decnet connection close at the remote side, the dlogind
    disappear but the process launched at login time by .profile (a 
    korn shell script doing anything) becomes child of "init" 
    instead of being killed as it's father (dlogind).
    
    We discovered however that changing the user's login shell from
    ksh to csh does change the behaviour. In such a case all the
    descending processes of dlogind process do disappear.
    
    But using the same user account with ksh from telnet or lat connections
    does not show the symptom (all the descending processes of dlogind
    process do disappear too).
    
    Any idea ? 
    
    dlogind or ksh bug ????
    
    Sorry to be so late answering, but customer was not ready to test as
    the workaround we provided (connection through LAT) does work !
    
    Cheers,
    
    
    				Alain

    Hi,
    
    That's a good clue about the ksh.  We've run into differences between
    shells before.
    
    I can get a backgrounded process to hang around after dlogind exits,
    but it does not run out of control.  It just continues to function
    normally.
    
    Can you get a copy of the .profile file?  And what happens if you
    kill dlogind, rather than breaking the network connection?
    
    Vince
    

    Hi Vince,
    
    >> Can you get a copy of the .profile file?  
    
    I just used Digital Unix's template and launched a simple ksh
    script from it at the end (customer's one is running an applic.
    on top of oracle db) :
    
    #
    # *****************************************************************
    # *                                                               *
    # *    Copyright (c) Digital Equipment Corporation, 1991, 1995    *
    # *                                                               *
    # *   All Rights Reserved.  Unpublished rights  reserved  under   *
    # *   the copyright laws of the United States.                    *
    # *                                                               *
    # *   The software contained on this media  is  proprietary  to   *
    # *   and  embodies  the  confidential  technology  of  Digital   *
    # *   Equipment Corporation.  Possession, use,  duplication  or   *
    # *   dissemination of the software and media is authorized only  *
    # *   pursuant to a valid written license from Digital Equipment  *
    # *   Corporation.                                                *
    # *                                                               *
    # *   RESTRICTED RIGHTS LEGEND   Use, duplication, or disclosure  *
    # *   by the U.S. Government is subject to restrictions  as  set  *
    # *   forth in Subparagraph (c)(1)(ii)  of  DFARS  252.227-7013,  *
    # *   or  in  FAR 52.227-19, as applicable.                       *
    # *     
    # *****************************************************************
    #
    # HISTORY
    #
    # @(#)$RCSfile: .profile,v $ $Revision: 4.1.3.4 $ (DEC) $Date:
    1992/09/30 13:49:
    15 $
    #
    PATH=$HOME/bin:${PATH:-/usr/bin:.}
    export PATH
    stty dec
    tset -I -Q
    PS1="`hostname`> "
    MAIL=/usr/spool/mail/$USER
    ./sleep_600.ksh
    
    
    Here is my sleep_600.ksh
    
    #!/bin/ksh
    echo "Sleeping for 10 min."
    sleep 600
    
    >> And what happens if you kill dlogind, rather than breaking 
    >> the network connection?
    
    Exactly the same. When ksh is the user's login shell, ksh becomes
    child of init process instead of being killed by signal 9 (KILL).
    As if the ksh login shell hasn't received the SIGKILL signal.
    So the ksh process survives, all it's descending processes do 
    also survive.
    
    When we either break the connection at the remote side (VMS machine)
    for a csh login shell or we kill (SIGKILL) the dlogind manually , all 
    the descending processes of dlogind are killed. 
    That's what we expect !
    
    We tested your modified dlogind (netrix::test/dlogind.note2121.Z)
    on both v3.2c and v3.2g, but we haven't noticed any difference.
    
    
    Thanks again for your help, do you think it's time for IPMT
    (I understand you cannot work too long on unofficial requests,
    so we do...), but I'd like to know better where the problem is
    before filling up the IPMT form (dlogind or ksh bug???) ?
    
    Cheers,
    
    				Alain
    

    I think it is probably time to open an IPMT case.  Since things
    seem to work OK with lat & telnet, open the case against DECnet.
    If it turns out to be ksh after all we'll forward it up to OSG.
    
    Vince
    

    Hi Vince,
    
    I'll open an IPMT then against DECnet, but will wait until the
    onsite engineer comes back from holiday (next week) to collect
    all the necessary info.
    
    Thanks for your help.
    
    Alain