On Fri, Apr 09, 2021 at 02:46:08PM +0000, Al Viro wrote: > On Fri, Apr 09, 2021 at 04:39:01PM +0200, Christian Brauner wrote: > > On Fri, Apr 09, 2021 at 02:30:13PM +0000, Al Viro wrote: > > > On Fri, Apr 09, 2021 at 04:22:58PM +0300, Amir Goldstein wrote: > > > > > > > But we are actually going in cycles around the solution that we all want, > > > > but fear of rejection. It's time to try and solicit direct feedback from Al. > > > > > > > > Al, > > > > > > > > would you be ok with passing mnt arg to vfs_create() and friends, > > > > so that we can pass that to fsnotify_create() (and friends) in order to > > > > be able to report FAN_CREATE events to FAN_MARK_MOUNT listeners? > > > > > > I would very much prefer to avoid going that way. > > > > Fwif and if I understand this correctly then this would mostly matter > > for stacking filesystems or filesystems that already have access to the > > relevant struct vfsmount they need to talk about anyway. It's not about > > passing struct vfsmount into inode methods themselves which would be a > > bad idea. Meaning Amir's change would not require or cause vfsmounts > > to be made accessible where they arent't already. > > AFAICS, you are getting caught on mount marks semantics (or lack thereof). > It's _not_ "event happened to something visible here"; it's "event had > been requested by way that explicitly refers to that mount". Well yes, that's what mount marks are supposed to be afaict. I don't think that's necessarily wrong though, at the least it can be quite helpful.
