Description of problem: When relocating an nfs service or restarting it in place during a write transaction the mounted filesystem returns EACCES. Version-Release number of selected component (if applicable): rgmanager-1.9.34-1 How reproducible: always Steps to Reproduce: 1.Setup an nfs service 2.mount it on another RHEL4 client 3.Start an NFS write 4.Use clusvcadm -R or -r ... Actual results: The write operation breaks with Permission denied, any further access to the mounted filesystem returns Permission denied. Expected results: Continuation of the write operation. Filsystem remains accessible Additional info: o Restarting/relocating when NFS is not busy is successful. o The network connection is GBit, the NFS protocol is 3, connection is via TCP o slow writes are successful (in restarting/relocating) o Relevant cluster.conf part: <service autostart="1" domain="nfs" name="homes-nfs"> <ip ref="XXX.XXX.XXX.XXX"/> <clusterfs ref="data"> <nfsexport name="home" path="/data/homes"> <nfsclient ref="clients"/> </nfsexport> </clusterfs> </service>
Assigning to NFS maintainer, but staying on CC list for now.
The issue here is that we're calling "exportfs -u" before the request queue is clear. Thus, clients get EPERM if they have pending requests on the queue, even if the IP address has already been torn down.
This should be fixed in current CVS and/or the RHCS4U3 packages when they come out. Setting to MODIFIED.
Although I can't test this anymore on RHEL4 I think this bug should be closed as fixed.
RHCS 4.3 and later.