Fucking opera crashed again while writing the article. Here the short version:
You try to mount an NFS export which resides on a xfs file system and the client fails with the message "Stale NFS file handle". A more verbose mount may show these messages:
mount.nfs: trying text-based options 'vers=3,addr=aaa.bbb.ccc.ddd'
mount.nfs: prog 100003, trying vers=3, prot=6
mount.nfs: trying aaa.bbb.ccc.ddd prog 100003 vers 3 prot TCP port 2049
mount.nfs: prog 100005, trying vers=3, prot=ddd
mount.nfs: trying aaa.bbb.ccc.ddd prog 100005 vers 3 prot UDP port 4002
mount.nfs: mount(2): Stale NFS file handle
mount.nfs: trying text-based options 'vers=3,addr=aaa.bbb.ccc.ddd'
mount.nfs: prog 100003, trying vers=3, prot=6
mount.nfs: trying aaa.bbb.ccc.ddd prog 100003 vers 3 prot TCP port 2049
mount.nfs: prog 100005, trying vers=3, prot=ddd
mount.nfs: trying aaa.bbb.ccc.ddd prog 100005 vers 3 prot UDP port 4002
mount.nfs: mount(2): Stale NFS file handle
The server syslog has entries like:
Sep 9 22:39:17 server rpc.mountd[11413]: authenticated mount request from aaa.bbb.ccc.ddd:809 for /srv/file/xxx (/srv/file/xxx)
Sep 9 22:39:19 server rpc.mountd[11413]: authenticated mount request from aaa.bbb.ccc.ddd:971 for /srv/file/xxx (/srv/file/xxx)
Sep 9 22:39:28 server rpc.mountd[11413]: authenticated mount request from aaa.bbb.ccc.ddd:913 for /srv/file/xxx (/srv/file/xxx)
Sep 9 22:40:20 server rpc.mountd[11413]: authenticated mount request from aaa.bbb.ccc.ddd:853 for /srv/file/xyz (/srv/file/xyz)
Before trying any stuff like turning off IPv6 or replacing portmapper with rpcbind as suggested by some folks, try changing the inode numbers of the exported directories as described in the mailing list entry "XFS, NFS and inode64 on 2.6.27" found via this forum entry.
To make this thing even more stupid, it is noted in the offical XFS FAQ, but in a way that you will not find it if you run into this problem. Second penalty point for XFS (first one is inability to shrink).
Recent comments