Fixed
Created: Jun 22, 2015
Updated: Dec 3, 2018
Resolved Date: Oct 27, 2015
Found In Version: 6.0.0.17,6.0.0.19
Fix Version: 6.0.0.24
Severity: Standard
Applicable for: Wind River Linux 6
Component/s: Kernel
"umount -f" fail to force the unmounting of an NFS directory when the server is unavailable. This means any NFS server failure will cause all WRL6 clients to "hang" if they have files open on the NFS mount.
In previous versions of WRL (at least WRL4) doing a forced unmount would allow the clients to become available again.
Configure a client machine such as the intel-x86-64 with standard kernel and glibc_std rootfs. No additional special configuration is needed.
Deploy and boot the target. Note that if you are using an NFS rootfs you will require two servers, one for the rootfs and a second one for the following test.
Log in to the client.
# mount <testsrv>:/exported/dir /mnt
# vi /mnt/text.txt
Enter some text and save the file, should work without problems.Add some more text but do not try to save yet...
On the server (I used an Ubuntu 14.04 machine for this)
$ sudo /etc/init.d/nfs-kernel-server stop
(Note, as previously mentioned, if using an nfs rootfs then that must be mounted from a different machine.)
Back on the client, try to save the edited text file. As the server is no longer responding the vi process will hang.
Open a second shell on the clinet (eg via ssh) and try to force the unmounting of the NFS area:
# umount -f /mnt
umount.nfs4: /mnt: device is busy
There is no way to force the unmount. All retries fail. Meanwhile, in the "vi" shell you may see messages such as this:
nfs: server <server> not responding, timed out
nfs: server <server> not responding, still trying
The vi process may start to respond after the timeout, but even trying to quit without saving will fail.
