Hello,
I am running a system with 3 machines :
machine1: 2.2.13-7mdk #1 Wed Sep 15 18:02:18 CEST 1999 i686 unknown
machine2: 2.2.13-7mdksmp #1 SMP Wed Sep 15 16:38:50 CEST 1999 i686
unknown
machine3: 2.2.13-7mdk #1 Wed Sep 15 18:02:18 CEST 1999 i586 unknown
I have nfs mounts all possible ways. When anybody issues
a mv command on a directory that is physically on a nfs mounted
filesystem, the kernel hangs on a the nfs client side. The only way
out is to reset the machine.
example :
on machine1, /etc/exports contains:
/users machine2(rw)
on machine2, /etc/fstab contains
machine1:/users /users nfs defaults 1 1
logged on machine2, 'df /users' outputs
machine1:/users 17G 15G 1.4G 92% /users
if I type (as an unprivileged user, connected via telnet)
mv /users/snovae/toto /users/snovae/tata
where /users/snovae/toto is a directory, machine2 hangs, with a message
(at the console)
kernel: nfs_rmdir : snovae/toto inode <a_number> i_count=2, i_nlink=2
(I am not sure of the exact message since it does not go in any
log file.)
After that, machine2 still prompts for username on the console virtual
terminals but never prompts for a password. Remote connections are impossible. The
only thing to do is to press the reset button and wait for all fsck's
(we have more than 100Gb of disk per machine)
I cannot easily run tests on the machines, since they are heavily
used, and the reboot takes one hour after reset.
It does not seem to be connected to running a SMP or not SMP kernel.
The crash happens the same way between machine 1 and 3.
Is this a known problem ? Are there any fixes ? Any idea would be welcome.
Thanks a lot in advance,
-------------------
Pierre Astier , LPNHE, 4 place Jussieu F 75252 Paris Cedex 05
tel (33) 1 44 27 23 27 ---- fax (33) 1 44 27 46 38
