> > + guard(rcu)();
> > + do {
> > + seq = read_seqcount_begin(&mnt_ns_tree_seqcount);
> > + node = rb_find_rcu(&mnt_ns_id, &mnt_ns_tree, mnt_ns_find);
> > + if (node)
> > + break;
> > + } while (read_seqcount_retry(&mnt_ns_tree_seqcount, seq));
>
> I don't really get the need for a seqcount here. Typically we use those
> when you could potentially find an object that's in some sort of
> inconsistent state.
>
> Is that the case here? It seems like you'd either find the object or
> not. It seems like either outcome is OK without the need to retry the
> search?
The sequence count is to detect concurrent changes to the rbtree. As a
change to the rbtree might cause us to miss whole subtrees as it could
be rotated. So it's not about the object but about the rbtree.
