umountnfs.sh should use "umount -i" to avoid CIFS umount timeout
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
sysvinit (Ubuntu) |
Invalid
|
Undecided
|
Unassigned |
Bug Description
Binary package hint: sysvinit
In certain cases at shutdown/reboot, the network is cut before CIFS filesystems are properly unmounted by umountnfs.sh. This results in a long timeout in the CIFS unmount that prevents shutdown/reboot from completing.
An example of this is when users mount CIFS shares and use network-manager default mode of operation, see (long) bug 211631 for more details.
We are fixing things to make sure network is normally shut down after network fs unmounts. However it would still be good to cover for network-lost corner cases by avoiding that timeout altogether. Steve French (CIFS upstream) analyzed the situation and we traced the timeout back to a statfs call, resulting from the umount.cifs call.
Unmounting using "umount -i" in umountnfs.sh would bypass the call to umount.cifs, which in turn would avoid the blocking statfs call, which would avoid the long timeout issue. As umountnfs.sh is always called by root it shouldn't cause an issue ?
Hm. Forget about it, even with -i there is a timeout, so there isn't much point in that patch.