Re: [PATCH v4 04/10] fsnotify: send event with parent/name info to sb/mount/non-dir marks

[Amir Goldstein <amir73il@gmail.com>]

> > > +     const struct path *path = fsnotify_data_path(data, data_type);
> > > +     struct mount *mnt = path ? real_mount(path->mnt) : NULL;
> > >       struct inode *inode = d_inode(dentry);
> > >       struct dentry *parent;
> > > +     bool parent_watched = dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED;
> > > +     __u32 p_mask, test_mask, marks_mask = 0;
> > >       struct inode *p_inode;
> > >       int ret = 0;
> > >
> > > +     /*
> > > +      * Do inode/sb/mount care about parent and name info on non-dir?
> > > +      * Do they care about any event at all?
> > > +      */
> > > +     if (!inode->i_fsnotify_marks && !inode->i_sb->s_fsnotify_marks &&
> > > +         (!mnt || !mnt->mnt_fsnotify_marks)) {
> > > +             if (!parent_watched)
> > > +                     return 0;
> > > +     } else if (!(mask & FS_ISDIR) && !IS_ROOT(dentry)) {
> > > +             marks_mask |= fsnotify_want_parent(inode->i_fsnotify_mask);
> > > +             marks_mask |= fsnotify_want_parent(inode->i_sb->s_fsnotify_mask);
> > > +             if (mnt)
> > > +                     marks_mask |= fsnotify_want_parent(mnt->mnt_fsnotify_mask);
> > > +     }
> >
> > OK, so AFAIU at this point (mask & marks_mask) tells us whether we need to
> > grab parent because some mark needs parent into reported. Correct?
> >
> > Maybe I'd rename fsnotify_want_parent() (which seems like it returns bool)
> > to fsnotify_parent_needed_mask() or something like that. Also I'd hide all
> > those checks in a helper function like:
> >
> >         fsnotify_event_needs_parent(mnt, inode, mask)
> >
> > So we'd then have just something like:
> >         if (!inode->i_fsnotify_marks && !inode->i_sb->s_fsnotify_marks &&
> >             (!mnt || !mnt->mnt_fsnotify_marks) && !parent_watched)
> >                 return 0;
> >         if (!parent_watched && !fsnotify_event_needs_parent(mnt, inode, mask))
> >                 goto notify_child;
>
> OK, but we'd still need to store marks_mask for later in case event
> needs parent.
>

Nevermind, I understand now what you meant.
Changed the function according to your guidance and looks much better.
Fits now in one terminal screen :-)

>
> >         ...
> >
> > > +
> > >       parent = NULL;
> > > -     if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
> > > +     test_mask = mask & FS_EVENTS_POSS_TO_PARENT;
> > > +     if (!(marks_mask & test_mask) && !parent_watched)
> > >               goto notify_child;
> > >
> > > +     /* Does parent inode care about events on children? */
> > >       parent = dget_parent(dentry);
> > >       p_inode = parent->d_inode;
> > > +     p_mask = fsnotify_inode_watches_children(p_inode);
> > >
> > > -     if (unlikely(!fsnotify_inode_watches_children(p_inode))) {
> > > +     if (p_mask)
> > > +             marks_mask |= p_mask;
> > > +     else if (unlikely(parent_watched))
> > >               __fsnotify_update_child_dentry_flags(p_inode);
> > > -     } else if (p_inode->i_fsnotify_mask & mask & ALL_FSNOTIFY_EVENTS) {
> > > +
> > > +     if ((marks_mask & test_mask) && p_inode != inode) {
> >                                         ^^ this is effectively
> > !IS_ROOT(dentry), isn't it? But since you've checked that above can it ever
> > be that p_inode == inode?
> >
>
> True, but only if you add the hidden assumption (which should be true) that
> DCACHE_FSNOTIFY_PARENT_WATCHED cannot be set on an IS_ROOT
> dentry. Otherwise you can have parent_watched and not check IS_ROOT()
> so prefered defensive here, but I don't mind dropping it.
>

I dropped this and also dropped the IS_ROOT(dentry) check because
there is already a check in the inline fsnotify_parent().

FYI, pushed these changes to branch fsnotify_name.
Rebased/tested/pushed branch fanotify_name_fid on top.
There are no changes to following patches as they are all
fanotify patches.

Thanks,
Amir.