This looks a bad idea to reorder calls to achieve such a behvaiour.Are we all happy with this patch now?
I can't see why we fix shrink_dcache_parent() only, why shrink_dcache_anon() is totally missed?
First of all because anon-dentries don't have a parent. So they are not a real
problem in don't restarting the shrink_dcache_anon() if we waited for prunes.
Since I've reordered the calls to shrink_dcache_anon() and
shrink_dcache_parent() in generic_shutdown_super() they are handled as normal
dentries if they are pruned through shrink_dcache_memory() from the d_lru list.