Bug 225507

Description of problem:
I set up a small cluster of FC6 machines, with one server and 4 workstations. 
The server and 3 of the workstations are x86_64, one workstation is a Pentium 4.
 The server was configured to provide an NFS filesystem, with /etc/exports
defining anonuid and anongid as 4294967294, the same as the definition of
nfsnobody in /etc/passwd and /etc/group.

The Pentium 4 is a sacrificial machine for doing kernel experiments.  I gave one
of my users sudo permissions to let him build and install kernels.  When he used
 sudo to compile by mistake, the creation of a file owned by nfsnobody caused
every single workstation to freeze up.  Investigation turned up this line in
/var/log/messages on the server:

Jan 29 14:37:26 abbott rpc.idmapd[2758]: nfsdcb: id '-2' too big!

Although "service rpcidmapd status" showed that it was still running, it was
clear that idmapd wasn't functioning.  Forcing it to restart got the other
workstations running again.  I took the advice in the URL specified in this bug
report and changed nfsnobody to 65534, which has prevented the problem from
recurring.  However, contrary to what is said in the messages found at that URL,
there are no older Unix systems in my cluster; every machine is running FC6.

Version-Release number of selected component (if applicable):
nfs-utils-1.0.10-5.fc6

How reproducible:
Always.

Steps to Reproduce:
1. Set up an NFS server.  In /etc/exports, include
anonuid=4294967294,anongid=4294967294.
2. Create a file in an NFS mounted directory as a user unknown to the server
machine.
3. Observe that the client machines are now unable to perform operations on NFS
 mounted files.
  
Actual results:
The idmap service stops responding.

Expected results:
The idmap service should map the anonymous user/group to nfsnobody and proceed
normally.  Even if it issues an error, it should still proceed normally instead
of ceasing to respond.

Additional info: