NFS server problem with debian client

[Cedric_8]

Hello,

I am running an NFS server on a HP-UX 11.11.

root> cat /etc/exports
/export/castorRoot -anon=0

root> cat /etc/rc.config.d/nfsconf
#********************** NFSCONF ******************************

# NFS configuration. See nfsd(1m), mount(1m), pcnfsd(1m)
#
# @(#)B.11.11_LR
#
# NFS_CLIENT: 1 if this node is an NFS client, 0 if not
# NFS_SERVER: 1 if this node is an NFS server, 0 if not
# Note: it is possible for one host to be a client, a server,
# both or neither! This system is an NFS client if you will
# be NFS mounting remote file systems; this system is a server
# if you will be exporting file systems to remote hosts.
# See Also: nfsd(1M), mount(1M).
# NUM_NFSD: Number of NFS deamons (nfsd) to start on an NFS server. Four
# has been chosen as optimal.
# NUM_NFSIOD: Number of NFS BIO daemons (biod) to start on an NFS client.
# Four has been chosen as optimal.
# PCNFS_SERVER: 1 if this node is a server for PC-NFS requests. This
# variable controls the startup of the pcnfsd(1M) server.
# See Also: pcnfsd(1M).
#
NFS_CLIENT=1
NFS_SERVER=1
NUM_NFSD=16
NUM_NFSIOD=16
PCNFS_SERVER=0
# export feature does not work in this file since files are being
# sourced into another file rc.config and this file is being sourced
# into the startup scripts.

#
# DAEMON OPTIONS
#
# LOCKD_OPTIONS: options to be passed to rpc.lockd when it is started.
# STATD_OPTIONS: options to be passed to rpc.statd when it is started.
# MOUNTD_OPTIONS: options to be passed to rpc.mountd when it is started.
#
LOCKD_OPTIONS=""
STATD_OPTIONS=""
MOUNTD_OPTIONS=""
#
# automount configuration
#
# AUTOMOUNT = 0 Do not start automount
# AUTOMOUNT = 1 Start Automount.
# AUTO_MASTER = filename of the master file passed to automount
# AUTO_OPTIONS = options passed to automount
#
AUTOMOUNT=1
AUTO_MASTER="/etc/auto_master"
AUTO_OPTIONS="-f $AUTO_MASTER"
#
# rpc.mountd configuration. See mountd(1m)
#
# START_MOUNTD: 1 if rpc.mountd should be started by a system startup script.
# 0 if /etc/inetd.conf has an entry for mountd.
# Note: rpc.mountd should be started from a system startup script,
# however, it can be started from either nfs.server or inetd, and
# MUST only be configured in one place.
#
START_MOUNTD=1
#
#autofs configuration. See autofs(1m)
#
#For the 11.0 Release line both AUTOFS and the old Automount
#are delivered. In order to invoke the AUTOFS instead of
#you must set the AUTOFS flag to 1.
#
#/usr/sbin/automount is now a script that sources in this file
#Depending on the variable AUTOFS, either AUTOFS or the old
#automount process will execute. The nfs.client start script
#will also use this variable to start the appropriate process
#during the boot sequence.
#AUTOFS= 0 - use the old automount process.
# 1 - use the new AutoFS.
#AUTOMOUNT_OPTIONS= - options to the AutoFS automount command
#AUTOMOUNTD_OPTIONS= - options to the AutoFS automountd daemon
#
#The AUTOMOUNT flag still needs to be set for either the old
#automount or new AutoFS to be started by the nfs.client script.
#
AUTOFS=0
AUTOMOUNT_OPTIONS=""
AUTOMOUNTD_OPTIONS=""


I have several redhat machines mounting the nfs export filesystem without any issue.

I recently installed a debian (lenny) machine, accessing the same NFS export.

Every machine has the same line in /etc/fstab :
nfsserver:/export/castorRoot /mnt/tcicodfs nfs soft,timeo=50,retry=10,rsize=8192,wsize=8192 0 0

I have a really weird behaviour. If I do a FTP connection to my debian machine and through that FTP connection I want to upload a file on /mnt/tcicodfs (which is NFS mounted), the NFS server on the HP-UX stops to answer correctly. For instance the command showmount on the machine fails and if I try to mount the NFS exported on a new machine, it fails with a RPC error message.
The FTP daemon is vsftpd on debian.

Ever heard of such a weird behaviour ?

I also have feelings that the NFS traffic is slower on the recent debian rather than on the old redhat I have (redhat 8, 9 and fedora core 4), are you aware of any known bug of tuning I could apply ?

Many thanks in advance.

[Dave Olker]

Hi Cedric,

I don't know much about the different variants of Linux but I do have a few questions.

1) Are you running the latest version of the nfs-utils package? They are constantly updating the NFS client stack for Linux, so that's one of the first things I check.

2) Are you using UDP or TCP? I think most Linux variants still default to UDP. You might want to try mounting with TCP and see if that helps.

3) I'm having difficulty following how NFS is involved here. Are you using ftp to access an NFS mounted directory on an NFS client and depositing the file there? In other words, you're using ftp and NFS together? Or am I misunderstanding the configuration? If this is the case, why not simply ftp directly to the NFS server? Or why not directly mount the NFS exported filesystem to the machine running ftp? Why transfer twice?

Again, if I've misunderstood your setup please explain it in greater detail.

Regards,

Dave

[Jim Keeble]

What exactly is the error message that you see, and where to you see it? Maybe a screen print of what you see whould be helpful.

[Cedric_8]

I know the setup is a little weird. On a LAN, servers are accessing via NFS a filesystem. NFS server is configured on HPUX (config already given in the 1st message).
Outside from this LAN, I have customers accessing the previous servers via FTP from internet and they need to access the filesystem exported by NFS but the ftp server is not running on the HPUX NFS server.
That's why I have FTP connections towards debian machines that are accessing NFS filesystem via this debian machine. A bit complicated I admit ...
Anyway, if ever the ftp client tried to upload a file on the filesystem mounted on the debian machine, the NFS server on the HPUX server cannot work preperly anymore.
By that I mean for instance:
- The command "showmount" on the HPUX hangs
- The commande "showmount -e nfsserver" gives an error like "RPC failed etc. (I can't reproduce the exact message because these are production servers and I don't want to mess around again).

This looks to me like a FTP debian bug or a HPUX NFS server bug.

Thanks for your help anyway.