When an AFS inode is allocated by afs_alloc_inode(), the allocated
afs_vnode struct isn't necessarily reset from the last time it was used as
an inode because the slab constructor is only invoked once when the memory
is obtained from the page allocator.

This means that information can leak from one inode to the next because
we're not calling kmem_cache_zalloc().  Some of the information isn't
reset, in particular the permit cache pointer.

Bring the clearances up to date.

Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: Marc Dionne <marc.dionne@auristor.com>