MongoDB Server: unbounded memory growth from arbitrary namespace names in reads

MongoDB Server maintains an in-memory table of per-namespace statistics for operations that carry encrypted-field information. Entries are created for any namespace supplied in a request, with no limit on the number of entries and no way to remove entries that never corresponded to a real collection. A user with basic read privileges who repeatedly issues ordinary read commands referencing arbitrary namespace names may cause the server process to retain steadily increasing amounts of memory and to slow down on the affected code path; the memory is reclaimed only by restarting the process.