Lock recovery doesn't work in debian statd/lockd ordering

Lock recovery doesn't work in debian statd/lockd ordering

[Neil Brown]

Hi Chip and NFSers.

 I just noticed that lock recovery doesn't work with a Debian NFS
 server.
 i.e.  server shuts down, restarts, and the client *Doesn't* reclaim
 the locks.

 The problem is that statd starts before lockd.

 What happens is that:
    statd starts
      it moves everything from /var/lib/nfs/sm to /var/lib/nfs/sm.bak
      it notifies each host listed in /var/lib/nfs/sm.bak that the
        server has restarted
    statd on the client gets the notification and tells it's local
        lockd to reclaim locks.
    lockd on client asked portmap on server for port number for lockd,
        but lockd hasn't started on the server so there isn't one.  So
        lockd on the client gives up (maybe it should persist, I'm not
        sure).
    Finally lockd on the server starts, but it is too late.


 I fixed this for myself by moving /etc/rc2.d/S19nfs-common to S21 so
 that it starts after nfsd and lockd which are started in
 S20nfs-kernel-server.

 There should be no problem with statd starting after lockd and lockd
 doesn't try to talk to statd until it gets the first lock request.

NeilBrown


-------------------------------------------------------
In remembrance
www.osdn.com/911/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: Lock recovery doesn't work in debian statd/lockd ordering

[Adrian Phillips]

>>>>> "Neil" == Neil Brown <neilb@cse.unsw.edu.au> writes:

    Neil>  I just noticed that lock recovery doesn't work with a
    Neil> Debian NFS server.  i.e.  server shuts down, restarts, and
    Neil> the client *Doesn't* reclaim the locks.

Um, which version ? On my woody box lockd is started in nfs-common
after statd still though :-

#!/bin/sh
#
# nfs-common    This shell script takes care of starting and stopping
#               common daemons required for NFS clients and servers.
#
# chkconfig: 345 20 80
# description: NFS is a popular protocol for file sharing across \
#	       TCP/IP networks. This service provides NFS file \
#	       locking functionality.
#

PREFIX=

NEED_LOCKD=yes
if test -f /proc/ksyms
then
    # We need to be conservative and run lockd,
    # unless we can prove that it isn't required.
    grep -q lockdctl /proc/ksyms || NEED_LOCKD=no
fi

[ -x $PREFIX/sbin/rpc.statd ] || exit 0
[ -x $PREFIX/sbin/rpc.lockd ] || [ "$NEED_LOCKD" = no ] || exit 0

# What is this?
DESC="NFS common utilities"

# See how we were called.
case "$1" in
  start)
	cd /	# daemons should have root dir as cwd
	printf "Starting $DESC:"
	printf " statd"
	start-stop-daemon --start --quiet \
	    --exec $PREFIX/sbin/rpc.statd
	if [ "$NEED_LOCKD" = yes ]
	then
	    printf " lockd"
	    start-stop-daemon --start --quiet \
		--exec $PREFIX/sbin/rpc.lockd
	fi
	echo "."
	;;

  stop)
	printf "Stopping $DESC:"
	if [ "$NEED_LOCKD" = yes ]
	then
	    printf " lockd"
	    start-stop-daemon --stop --oknodo --quiet \
		--exec $PREFIX/sbin/rpc.lockd
	fi
	printf " statd"
	start-stop-daemon --stop --oknodo --quiet \
	    --exec $PREFIX/sbin/rpc.statd
	echo "."
	;;

  restart | force-reload)
	$0 stop
	sleep 1
	$0 start
	;;

  *)
	echo "Usage: nfs-common {start|stop|restart}"
	exit 1
	;;
esac

exit 0

Sincerely,

Adrian Phillips

-- 
Your mouse has moved.
Windows NT must be restarted for the change to take effect.
Reboot now?  [OK]


-------------------------------------------------------
In remembrance
www.osdn.com/911/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: Lock recovery doesn't work in debian statd/lockd ordering

[Neil Brown]

On  September 12, a.phillips@met.no wrote:
> >>>>> "Neil" == Neil Brown <neilb@cse.unsw.edu.au> writes:
> 
>     Neil>  I just noticed that lock recovery doesn't work with a
>     Neil> Debian NFS server.  i.e.  server shuts down, restarts, and
>     Neil> the client *Doesn't* reclaim the locks.
> 
> Um, which version ? On my woody box lockd is started in nfs-common
> after statd still though :-

/sbin/rpc.lockd is an aberation that only applies to a small range of
kernels, about 2.2.14 to 2.2.17 I think.

In all other kernels, lockd is started on demand, either when nfsd
starts or when an nfs filesystem is mounted (without -o nolocks).

NeilBrown


-------------------------------------------------------
In remembrance
www.osdn.com/911/
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: Lock recovery doesn't work in debian statd/lockd ordering

[H. J. Lu]

On Thu, Sep 12, 2002 at 02:49:27PM +1000, Neil Brown wrote:
> On  September 12, a.phillips@met.no wrote:
> > >>>>> "Neil" == Neil Brown <neilb@cse.unsw.edu.au> writes:
> > 
> >     Neil>  I just noticed that lock recovery doesn't work with a
> >     Neil> Debian NFS server.  i.e.  server shuts down, restarts, and
> >     Neil> the client *Doesn't* reclaim the locks.
> > 
> > Um, which version ? On my woody box lockd is started in nfs-common
> > after statd still though :-
> 
> /sbin/rpc.lockd is an aberation that only applies to a small range of
> kernels, about 2.2.14 to 2.2.17 I think.
> 

Just for the record, I have a need to run the nlockmgr service at a
fixed port.  I have

nlockmgr        3003/tcp
nlockmgr        3003/udp

in my /etc/services. But I have to hard code it in my kernel. No, I
don't use kernel modules.  It will be nice to for rpc.lockd to passes
the port number to kernel. It doesn't have to start lockd.


H.J.


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: Lock recovery doesn't work in debian statd/lockd ordering

[Philippe Troin]

Neil Brown <neilb@cse.unsw.edu.au> writes:

> Hi Chip and NFSers.
> 
>  I just noticed that lock recovery doesn't work with a Debian NFS
>  server.
>  i.e.  server shuts down, restarts, and the client *Doesn't* reclaim
>  the locks.
> 
>  The problem is that statd starts before lockd.
> 
>  What happens is that:
>     statd starts
>       it moves everything from /var/lib/nfs/sm to /var/lib/nfs/sm.bak
>       it notifies each host listed in /var/lib/nfs/sm.bak that the
>         server has restarted
>     statd on the client gets the notification and tells it's local
>         lockd to reclaim locks.
>     lockd on client asked portmap on server for port number for lockd,
>         but lockd hasn't started on the server so there isn't one.  So
>         lockd on the client gives up (maybe it should persist, I'm not
>         sure).
>     Finally lockd on the server starts, but it is too late.
> 
> 
>  I fixed this for myself by moving /etc/rc2.d/S19nfs-common to S21 so
>  that it starts after nfsd and lockd which are started in
>  S20nfs-kernel-server.
> 
>  There should be no problem with statd starting after lockd and lockd
>  doesn't try to talk to statd until it gets the first lock request.

I've opened a debian bug (# 160800) about this.

  -> http://bugs.debian.org/160800

Phil.


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

Re: Lock recovery doesn't work in debian statd/lockd ordering

[Chip Salzenberg]

Since Neil is the bug reporter and no one has disagreed with his
analysis, I'm making the change he recommends for Debian unstable.
If it proves OK, I'll issue a fix for Debian stable.
-- 
Chip Salzenberg         - a.k.a.  -        <chip@pobox.com>
         "It furthers one to have somewhere to go."


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs

unsubcribe

[Will Stowe]

-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs