When creating or updating a cache entry, we need to get an DFS
referral (get_dfs_referral), so avoid holding any locks during such
network operation.

To prevent that, do the following:
* change cache hashtable sync method from RCU sync to a read/write
  lock.
* use GFP_ATOMIC in memory allocations.
Signed-off-by: NPaulo Alcantara (SUSE) <pc@cjr.nz>
Signed-off-by: NSteve French <stfrench@microsoft.com>