On Wed, 14 Feb 2024 at 10:27, Christian Brauner <brauner@xxxxxxxxxx> wrote: > > Ok, that turned out to be simpler than I had evisioned - unless I made > horrible mistakes: Hmm. Could we do the same for nsfs? Also, quick question: > +void pid_init_pidfdfs(struct pid *pid) > +{ > + pid->ino = ++pidfdfs_ino; > +} As far as I can tell, the above is only safe because it is serialized by spin_lock_irq(&pidmap_lock); in the only caller. Can we please just not have a function at all, and just move it there, so that it's a whole lot more obvious that that inode generation really gets us a unique number? Am I missing something? Linus