The function sd_journal_close first frees j->files and then indirectly calls detach_location which access j->files.

When allocating hashmaps from the pool, this likely goes undetected by valgrind and likely doesn't cause any crashes either.  This does crash when the hashmap is allocated with malloc, though.