Jeff Layton <jlayton@xxxxxxxxxx> wrote: > I don't think that can happen. An I_NEW inode hasn't been properly > hashed yet, so nothing should be able to find it until > unlock_new_inode() is called. That's not where the issue lies. I'm talking about *after* I_NEW has been cleared. Imagine you have a file that has hard links in several directories. Can simultaneous lookup on a number of those hard links result in you going through ceph_fill_inode() a number of times in parallel? David