On Wed, 11 Dec 2024 at 11:34, Christian Brauner <brauner@xxxxxxxxxx> wrote: > For that the caller has to exit or switch to another mount namespace. > But that can only happen when all notifications have been registered. > I may misunderstand what you mean though. Ah, umount can only be done by a task that is in the namespace of the mount. I cannot find a hole in that logic, but it does seem rather brittle. > Probably, although I'm still not too happy about it. Especially since > umount propagation can generate a lot more events then mount propagation > as it tries to be exhaustive. I guess we have to see. Would be nice to > have proper test for this. You mean performance test? Will try to think of something. Thanks, Miklos
