SUMMARY: shutdown broadcast

From: Daryl Crandall (daryl@dash.mitre.org)
Date: Wed Feb 27 1991 - 12:20:42 CST


ORIGINAL QUESTION:

I manage Sun workstations running mostly SunOS-4.0.3 on Sun3 & Sun4 hosts and
these workstations are located in several different states, NJ, VA, MA, DC
I thought I understood the machanisms that shutdown uses to notify
"appropriate" hosts that it is shutting down. However I'm now getting
complaints from remote users that they don't give a hoot about my machines
in VA shutting down when they are in NJ, and they want the shutdown messages
to stop.

I tried, but they still get the notices. The /etc/rmtab file was emptied, No
one was logged in on the other guy's machine. The only reference in the /etc
directory to the other machine(s) was in the the /etc/netgroup file. Is this
my problem?

Is there any way to tell shutdown to shut-up? (No pun intended.)

Does anybody have an alternative to shutdown other than /etc/halt?

A definitive list of exactly what shutdown uses to determine "appropriate"
hosts to notify would be appreciated.

        Daryl Crandall
        The Mitre Corporation
        daryl@mitre.org
        (703) 883-7278

#############################################################################
MY SUMMARY:

It looks like /etc/rmtab isn't queried during shutdown. The information is
maintained internally by rpc.mountd and the 'mountd' should be forced to
re-read an edited /etc/rmtab before attempting to shutdown.

Shutdown is intentionally noisy.

Shutdown can be simulated in a less noisy manner.
#############################################################################
RESPONDENT SUMMARIES:
#############################################################################
From: "Ric Anderson" <ric@cs.arizona.edu>

Shutdown calls rpc.mountd, to get a list of hosts who have mounted file
systems owned by the local machine. This list is based on rmtab when
the machine is booted (any line not beginning with a "#"), or when
rpc.mountd is restarted. rpc.mountd will fseek /etc/rmtab and plop a
"#" over the first character of the remote host name, if the remote
host issues a umount. Clobbering rmtab before shutdown runs does no
good, because mountd has the authoritative copy in core, and passes
that to shutdown.

However, if the remote host crashes, and doesn't remount the local fs,
then there will then be a bogus entry in /etc/rmtab.

I have a local version of shutdown (called ksys) which fixes the insane
babbling shutdown does, by reducing it to two messages (1 at 10 minutes
and 1 at 1 minute). It also reads rmtab directly, since the rpc call
failed more often than not on certain brain dead sequent operating
systems.

If you want a copy, I'll bundle it up and send it to you. You could
probably hack it to ignore certain hosts in rmtab if need be :-)

#############################################################################
From: Aydin Edguer <edguer@alpha.ces.cwru.edu>

shutdown does a few things.
shutdown creates /etc/nologin.
        touch /etc/nologin
shutdown syncs the system.
        sync;sync
shutdown sends a terminate signal to init.
        kill -TERM 1

shutdown -h halts the system rather than terminating init.
        halt

It uses an RPC to mountd to gather the list of hosts.
############################################################################
From: Rich Wales <wales@CS.UCLA.EDU>

"shutdown" notifies hosts which are using NFS to mount file systems on
the machine being halted. It uses the "rwall" RPC service to send the
warning to each such host.
############################################################################
From: kean@talon.UCS.ORST.EDU
 
My understanding is that shutdown looks at /etc/rmtab and /etc/xtab to
figure out who has mounted partitions off of this machine, and broadcasts
through the rwall/wall rpc service to those aforementioned machines. Clearing
both those files will stop the messages, but that's a bit drastic. You should
note that the rwall service is a security problem allowing any intelligent
person root access on any machine that runs rwall. I have it commented out
on all mine. The appropriate CERT bulletin can be found on cert.sei.cmu.edu
in the anonymous ftp area I believe.
############################################################################
From: Sam Horrocks <sam@john-bigboote.ICS.UCI.EDU>

"showmount" should show you exacly who will receive the messages. Clearing
rmtab won't help unless aftwerwards you re-start rpc.mountd. Shutdown
contacts rpc.mountd (the same way showmount does) to find out which hosts
have mounted partitiones. I don't think it has anything to do with
/etc/netgroup.
############################################################################
From: sundev!fletch!kevin@Sun.COM (Kevin Sheehan {Consulting Poster Child})

You might check /etc/sm* - if you are passing around NFS mounts, then
this may be used to notify folks as well.

sm* is used by rpc.lockd/rpc.statd to recover on reboots, it might
also be used when going down?
############################################################################
From: steve@umiacs.UMD.EDU (Steve D. Miller)

   It looks like shutdown gets its list of hosts not by looking in rmtab,
but by contacting mountd and doing a MOUNTPROC_DUMP. You might kill
off your mountd, clean out rmtab, and then restart it; that should flush
the bogus host entries.

   I don't see any way to get shutdown not to do any rwalls at all. I
at least think that's a bug, not a feature; I have those annoying messages...
############################################################################

Thanks also to:
From: peterg@murphy.com (Peter Gutmann)
From: David Beard <beard@spam.ua.oz.au>
From: "Anthony A. Datri" <datri@lovecraft.convex.com>
From: metzger@halley.watson.ibm.com (Perry E. Metzger)



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:11 CDT