If I stop knfsd via "/etc/rc.d/init.d/nfs stop", it will not restart when I issue the start command. I get the following output from the startup script: [root@jose /root]# /etc/rc.d/init.d/nfs start Starting NFS services: [ OK ] Starting NFS quotas: Cannot register service: RPC: Timed out [FAILED] Starting NFS mountd: Cannot register service: RPC: Timed out mountd: unable to register (mountd, 1, udp). [FAILED] Starting NFS daemon: nfssvc: Address already in use [FAILED] It hangs for two minutes trying to start the quotas, finally fails, and then again hangs trying to start mountd, again finally failing. As this happens in my logs I see: Dec 5 12:59:03 jose nfs: Starting NFS services: succeeded Dec 5 13:02:03 jose rquota[15674]: unable to register (RQUOTAPROG, RQUOTAVERS, udp). Dec 5 13:02:03 jose rpc.rquotad: Cannot register service: RPC: Timed out Dec 5 13:02:03 jose nfs: rpc.rquotad startup failed Dec 5 13:04:04 jose mountd[15689]: unable to register (mountd, 1, udp). Dec 5 13:04:04 jose rpc.mountd: Cannot register service: RPC: Timed out Dec 5 13:04:04 jose rpc.mountd: mountd: unable to register (mountd, 1, udp). Dec 5 13:04:04 jose nfs: rpc.mountd startup failed Dec 5 13:04:04 jose rpc.nfsd: nfssvc: Address already in use Dec 5 13:04:04 jose nfs: rpc.nfsd startup failed This happens for both 1.4.4-7 and 1.2.2-4 knfsd versions. It seems that the system is otherwise in order. For both versions of knfsd the services start fine at boot time, failing only when I've manually stopped the service and attempted to restart. The machines in question are 5.1, upgraded to 6.0. A friend of mine reports the same things with his 6.1 installations. The kernel support has been built as modules, which do unload if I stop the service after booting. But I reload them by hand. Loaded modules at the time of attempting to restart the knfsd service are: [root@jose /root]# lsmod Module Size Used by nfsd 150040 0 lockd 30760 0 [nfsd] sunrpc 51876 0 [nfsd lockd] ip_masq_raudio 2736 0 ip_masq_user 2376 0 (unused) Obviously bouncing the machine is a pretty annoying way to restart the service.
I'm not sure what the problem is, but I run three different machines, now all running under RH 6.1. One is a straight 6.1 install, one is updated from 5.2 direct and the third is updated from 5.1 via 6.0. On all three, I regularly stop and start nfs using the commands stated, and have never had any problems. Perhaps we can compare configurations?
I stand corrected...the problem seems to only appear on 6.0 machines.
This can happen when /var/lib/nfs/* gets stale information. If you ever mount a filesystem from a host that disappears (changes names), KNFS may not start anymore because it for some reason tries to contact or resolve that old hostname. This does not seem to be fixed in the lastest KNFSD beta releases. (I last checked a couple of months ago).
This is a DNS misconfiguration or problems with portmap/tcp_wrappers. The versions in rawhide do not exhibit this type of problem.