#79 NFS server causes fs ref count problem

[David Zafman]

From a single node cluster I export a chard mount via
NFS server. On a client I mount it, issue an "ls -lR."
After doing a service SSInfs stop, followed by a
filesystem umount, I get the following messages:

WARNING: inode busy: dev 8:19 (xc0673456) mode 10666
count 1

It must be either READDIR, GETATTR or LOOKUP. It
doesn't seem to happen unless NFS server is involved.

[Nobody/Anonymous]

Logged In: NO

I get similar messages on umount of an exported NFS chard
mount. And of course SSInfs is stopped. However dev 93:0
is not listed in my /proc/devices.

Nov 29 20:17:39 node1 kernel: inode busy: dev 93:01:1292080
(d20c2d00) mode 40711 count 1
Nov 29 20:17:39 node1 kernel: inode busy: dev 93:01:588962
(d20c2300) mode 40700 count 2
Nov 29 20:17:39 node1 kernel: inode busy: dev 93:01:883734
(f098c080) mode 40755 count 2
Nov 29 20:17:39 node1 kernel: inode busy: dev 93:01:261910
(d364f800) mode 40700 count 2
Nov 29 20:17:39 node1 kernel: inode busy: dev 93:01:33091
(ebd20580) mode 100664 count 1
Nov 29 20:17:39 node1 kernel: VFS: Busy inodes after
unmount. Self-destruct in 5 seconds. Have a nice day...