We have been using NFS to export files from a Linux box to Solaris for at least a year now and have not had any problems until we installed RH 6.0. Under the new version an attempt by the Sun to mount a Linux-exported directory fails and the following messages appear in the /var/log/messages file on the Linux system: Oct 6 18:00:18 mumble mountd[1386]: authenticated mount request from foobar:962 Oct 6 18:00:18 mumble mountd[1386]: getfh failed: Operation not permitted This is not a failure to include the kernel nfs server (Bugzilla #3102), since it happens even after I recompiled the kernel (2.2.5-22) with the knfsd code enabled. This also does not appear to be the problem that knfsd advertises NFS version 3 but fails to deliver it (Bugzilla #2554), since it occurs even when I include "vers=2" in the Sun's /etc/vfstab file. (Although a cursory look at the mountd source code suggests that getfh is part of the version 3 implementation.) I decided not to try the new improved knfsd from Raw Hide. WORKAROUND: I have been able to work around this problem by downloading nfs-server-2.2beta44-1.src.rpm from the RedHat 5.2 errata, compiling, and installing. There were a few extra steps needed to make the install work: I created an empty directory /var/state/nfs and empty file /var/lib/nfs/xtab by hand; moved exportfs and nfs.init to their target directories by hand; and added the appropriate links to rc?.d to start and stop automatically when the run level changes. Don't forget to recompile the kernel without knfsd and to remove the knfsd package. The old NFS daemon appears to work just fine, although I haven't really given it a hard workout yet.
If you decide to use this workaround, be sure to download the latest version of nfs-server from the RedHat errata site. The version on the 5.2 CDROM has some known security holes.
I experience this bug fairly frequently. I think the way to "fix" it is to stop NFS, edit /var/lib/nfs/xtab, restart NFS and pray that the "port in use" bug doesn't bite you. If the second bug DOES bite, reboot. It appears that linux NFS is stateful which is absolutely abominable.
It appears that the machine which is failing to mount should be turned off while you turn off NFS, edit the server's xtab and rmtab to remove references to the powered-off client, then reactivate NFS.
assigned to johnsonm
I am closing 5689 because I am unable to duplicate it on our cross mounts between Solaris 7 and Linux 7.3 machines. I am feeling that this was 'closed' during the change from 2.2->2.4 kernels or the various nfs changes from 2.4.7->2.4.19.