On 05/11/10 09:59 -0700, Michael wrote: > We use Solaris 10 on a couple x4500s in our Data Center. We recently > experienced a cooling issue which left these boxes in a hard shutdown state. > I rebooted and noticed our NFS system is acting funky. > > I use the ZFS "sharenfs" setting to share our ZFS volumes to different > clients. Well, if I do a showmount -e only one client appears, whereas before > the entire list of clients appeared. But now there's only one and only that > particular client may mount his filesystem. Others get "permission denied." > > So a bit of troubleshooting and I realized that this particular client is > just the last one that gets mounted. They're exported in alphabetical order, > so it's always the last one with the "sharenfs" setting. > > A bit more troubleshooting and the same phenomenon occurs when using > /etc/dfs/dfstab instead of ZFS. I run manual "share" commands and each one > supercedes the previous one. > > So, the upshot is: I am only able to export one filesystem and that's it. > Obviously some sort of corruption occurred due to the hard powerdown. Anyone > know where to look to fix this?
I'm not sure what went wrong during hard powerdown, but the most straightforward way, how to find out the root cause, is to inspect solaris kernel by mdb. So if you have a valid support contract for S10, you should raise a service request and provide kernel crash dump (reboot -d) for analyses. AFAIK information about exported file systems is maintained at two independent places (split-brain model): 1) In sharetab visible via /etc/dfs/sharetab 2) In internal kernel structures specific to nfssrv kernel module Could you compare content of /etc/dfs/sharetab with 'echo ::nfs_exptable | mdb -k'? Do you see the same exported filesystems in both outputs? You should see only one exported filesystem in /etc/dfs/sharetab as this is the source of information used by mountd when answering request from 'showmount -e'. best regards -jan _______________________________________________ nfs-discuss mailing list [email protected]
