NFS incompatibilties between AIX and HP-UX?

[Ralph Grothe]

Hi networkers,

a collegue of mine who runs a scheduled monitoring script that connects to our Unix servers (different brands: HP-UX, AIX, Solaris, Linux), and amongst other things checks filling of filesystems, experienced a strange Floating Point exception and subsequential core dumps whenever his script issued a "df -k".

This only happened after I had exported an Oracle 64Bit installation CDROM to a HP-UX 11.00 box that was mounted on an AIX 4.1 box.

The strange thing is that this only appears with -k flag.
A "df" without it does no core dump.

To me it seems like some idiosyncracy with block sizing between AIX and HP-UX?

The CDROM is globally exported with -ro option only and mounted like this on the HP-UX box:

#bdf -t nfs
Filesystem kbytes used avail %used Mounted on
aaa.bbb.ccc.ddd:/cdrom 0 0 0 0% /opt/ora_inst

While our DBAs need the CDROM for an Oracle upgrade I advised my colleague to change his df-flags to restrict lookups to the Veritas FS only (we safely can drop /stand as it usually stays fixed) by putting a
"df -kF vxfs" call in his script.

(mind you bdf can cope with the export from AIX, looks as if HP developers haven't been too keen to put much debugging effort in their standard Unix-fair df-port ;-)

Explanations are welcome.

Regards
Ralph

[Vincent Stedema]

Did you use "mount -F nfs" or "pfs_mount" to mount the remote filesystem? Normally, when you mount an Oracle CD on HP-UX, you need to use pfs_mount. Maybe you should give it a try...

I guess df -k coredumps because it apparently tries to divide 0 by 1024.

Vincent

[Ralph Grothe]

Hi Vincent,

this was the command line I issued to mount the CDROM:

# mount -F nfs -o ro,soft aaa.bbb.ccc.ddd:/cdrom /opt/ora_inst

(letters a-d represent IP of exporting AIX box)

[Vincent Stedema]

Apart from the core dumps, is the filesystem accessible for normal operation?

[Brian Hackley]

Hi,

I was just thinking that you might wan to pickup PHCO_22274, superceded byPHCO_23117 bdf(1):autofs:skip Size 31K 1Liner s700_800 11.00 bdf(1M) cumulative patch which fixes a problem with a buffer overflow. I don't know if that is the specific issue causing bdf to abort in your case, but you certainly might want to consider installing this patch since it does not cause a reboot and affects few other subsystems.

Hope this helps,
-> Brian Hackley

[Shannon Petry]

First, the user that said use pfs_mount is wrong. This should only be a concern with a local cdrom, and NOT a NFS filesystem. You did exactly right with the NFS.

Are you using/forcing NFS V3? This is the default in AIX, so you may want to reduce it to V2 or both. Dont know that it is causing any problems, but......

Also if the guys script uses df and not bdf, he should change his script. df on other system's does what bdf does on HP-UX. It is not so hard to look at the remote system, and issue commands appropriately. I do this alot just because of the difference in locations of binaries in the different systems.

I.E.
SYS_TYPE=`/bin/uname`
case $SYS_TYPE in
HP-UX)
DF=/usr/bin/bdf
RSH=/usr/bin/remsh
XDIR=/usr/bin/X11
;;
SunOS)
DF=/usr/sbin/df
XDIR=/usr/openwin/bin
RSH=/usr/bin/rsh
;;
esac

for HOST in host1 host2 ; do
$REMSH $HOST '$DF'
done

I know syntax is messy, but I think you get the idea.....

Regards,
Shannon

[Vincent Stedema]

Shannon,

pfs_mount can be used to mount a cd-rom from a remote server (read the man page...), but only if the remote system runs pfsd. PFS is strongly related to NFS.

I admit it won't work in this case, because AIX uses a different method to mount CD-ROM filesystems.

Regards,

Vincent

[Ralph Grothe]

Hi guys, many thanks for your helpful clues. :-)

Vincent, the mounted filesystem is accessible:

# find /opt/ora_inst|head
/opt/ora_inst
/opt/ora_inst/doc
/opt/ora_inst/doc/8iR3.pdf
/opt/ora_inst/doc/a86566.pdf
/opt/ora_inst/doc/images
/opt/ora_inst/doc/images/docbutton_h.gif
/opt/ora_inst/doc/images/docbutton.gif
/opt/ora_inst/doc/images/ds.gif
/opt/ora_inst/doc/images/ds_5fh.gif
/opt/ora_inst/doc/images/relnotebutton_h.gif

Funny thing is that the %used column of the bdf dump below for the AIX exportfs says 0% whereas another exported CDROM from an IRIX 6.5 Indy correctly states 100%

# bdf -t nfs
Filesystem kbytes used avail %used Mounted on
aaa.bbb.ccc.ddd:/cdrom 0 0 0 0% /opt/ora_inst
eee.fff.ggg.hhh:/CDROM 621752 621752 0 100% /opt/ora_doc

Brian, I will download the mentioned patches and dig through their README to see whether they address our little bug or are of any other use.

Shannon, if it were my script I would have done it in a similar way, by either checking the output from $(uname -s) or Perl's special variable $^O, and appropriatly use the bdf command for HP-UX.
By the way there is a wonderful CPAN module Proc::ProcessTable which gives you a single interface to all deviating kinds of the ps command (e.g. BSD vs. SysV)

I hope, you are not too disappointed about my poor rating of your answers, but none has come up yet with a fully satisfying explanation.