Wrong file readdir on NFS linux client

[antonietta]

I share directories on a Scientific Linux Cern linux cluster by using nfs tool.
SLC6 kernel 2.6.32-279.22.1.el6.x86_64
nfs-utils-1.2.3-26.el6.x86_64

After a server-client shutdown, a disagreeable event occurred.
By making ls list on a nfs client shared directory, duplicated entries for some files are shown.
Other files and sub-directories are non visible. The problem is not present on the nfs server.
The directory don't contain an enormous number of files:

On the server:
#ls |wc -l
330
#du -s
8855740

On the client:
#ls |wc -l
ls: reading directory .: Too many levels of symbolic links
120
n.b. symbolic links are not changed after the shutdown;
#du -s
not responding
#dmesg
some entries:
...
NFS: directory images/dp contains a readdir loop.Please contact your server vendor. The file: .. has duplicate cookie 683570819
NFS: directory images/dp contains a readdir loop.Please contact your server vendor. The file: .. has duplicate cookie 683570819
__ratelimit: 2 callbacks suppressed
NFS: directory images/green contains a readdir loop.Please contact your server vendor. The file: 57.png has duplicate cookie 1694199390
NFS: directory images/green contains a readdir loop.Please contact your server vendor. The file: 57.png has duplicate cookie 1694199390
...

I've tried to boot the system with the older kernel-2.6.32-279.el6.x86_64, but the bug is not
cleared. Any ideas?

Many thanks in advance
Antonietta

[Lakshmipathi]

hmm..that's quite interesting problem to solve. Can you give it try with tcpdump or wireshank on the client side and capture NFS requests/responses ? If possible post the dump file here too for analysis.
Before that try dumping caches on both server/client via command like #echo "3" > /proc/sys/vm/drop_caches

[Rubberman]

I would tend to agree with Lakshmipathi that it is likely a caching problem - have you tried rebooting? Also, you might want to post this problem to the SL user forums. Are you registered to post there? Not a problem to do. Just go to the SL user forum site and sign up: http://listserv.fnal.gov/scripts/wa....ic-linux-users

FWIW, I use SL extensively, though not in a cluster configuration, but my wife is a staff physicist in the Fermi Lab computer division that maintains SL.

[philix]

This looks like a NFS bug , which is fixed , NFS Client have an algorithm of looping into each Directory and Subdirectory during the changes in the directory and Client receives multiple cookies for mutliple request from server

Fix is

linux/kernel/git/torvalds/linux.git - Linux kernel source tree

Please upgrade your NFS client

Philix