We have an ONTAP 8.0.x filer here serving out a volume with an NTFS qtree via both CIFS and NFS. We're encountering the following situation:
1. An OSX user browses to the path via CIFS in Finder and then walks away/moves on to another task leaving the Finder window open on the path. - This generates a mandatory read lock on the folder (verified via lock status): CIFS path=\path(/path) host=10.28.102.30() owner=njarecha state=GRANTED mode=Read-denyN oplock=None
2. An NFS client comes along now (automated build process) and attempts delete the same folder. This appears to succeed from the NFS client's perspective. - We now see the following: CIFS path=\path(/path) host=10.28.102.30() owner=njarecha state=GRANTED mode=Read-denyN oplock=None CIFS path=\path(-) host= owner= state=GWAITING mode=DelOnClose
At this point, CIFS access to the path now is completely blocked. No one can navigate into the folder, nor view its ACLs (save likely for the OSX client still holding the read lock).
NFS clients can continue to access the contents of the folder (both read/write), but rmdir/rm operations on the folder itself fail with Permission denied.
The fix is to (a) close the Finder window or (b) clear the lock via 'lock break' on the filer.
I'll note that Windows Explorer doesn't appear to open a read lock when you only browse to a path -- not sure why Finder does.
We're looking for options here, but first I'm wondering if the filer's response here isn't ideal. Rather than completing closing off access to CIFS clients, shouldn't #2 above just fail from the get-go rather than generating the DelOnClose queued event (if a mandatory read lock exists on the resource)?
Other options we're considering:
- Modifying automated processes to avoid the "rmdir" command. - Sticking with only one access protocol (likely CIFS). - Modifying OSX Finder behavior to not generate and hold a CIFS read lock (not sure how easy this would be to do). - Finding some sort of tunable on the ONTAP side to minimize impact (e.g. ignore the read lock and let the NFS rmdir operation to succeed) at the potential elevated risk of reduced data integrity.
Anyone have any thoughts/experiences with something similar?
Thanks, Ray