Hi Daniel, It looks like mdcache_lru_cleanup_try_push() can be improved for the cases when there are ops executing parallely with an unexport request. In this case mdcache_lru_cleanup_try_push() is not able to drop the sentinel ref and mark the entry for cleanup. This is because one extra ref is taken by the op which is currently executing. The total number of refs are > 2. Once the current op execution finishes (after the execution of mdcache_unexport()) the sentinel ref remains and the entry is also not marked for cleanup even though no export is referring to it (entry->first_export_id == -1). Is it possible to modify mdcache_lru_cleanup_try_push() such that even if the refcnt is > 2 but if the entry->first_export_id == -1 (the current export which is going away was the only one referring to this entry), we should drop the sentinel ref and mark the entry for cleanup? This will ensure we cleanup the entries when the current op does put_ref and these are reused faster than other mdcache entries which are actually pointing to valid exports. Thanks, Ashish