On Tue, Nov 20, 2018 at 1:32 PM Jan Kara <jack@xxxxxxx> wrote:
>
> On Wed 14-11-18 19:43:40, Amir Goldstein wrote:
> > Define a new data type to pass for event FSNOTIFY_EVENT_DENTRY
> > and use it whenever a dentry is available instead of passing
> > it's ->d_inode as data type FSNOTIFY_EVENT_INODE.
> >
> > None of the current fsnotify backends make use of the inode data
> > with data type FSNOTIFY_EVENT_INODE - only the data of type
> > FSNOTIFY_EVENT_PATH is ever used, so this change has no immediate
> > consequences.
> >
> > Soon, we are going to use the dentry data type to support more
> > events with fanotify backend.
> >
> > Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx>
>
> One general coment: Audit is using FSNOTIFY_EVENT_INODE and this change is
> going to break it. So it needs updating as well.
>
Ouch! I should really add the audit test suite to my testing routine...
> Also I'd prefer a more justified selection of dentry events than 'those
> where we can do it'. Because eventually that set will translate to events
> available to userspace in fanotify and that should be a reasonably
> consistent set. So here I suspect you target at directory modification
> events (you call them filename events in the next patch). So just define
> which FS_<foo> events these exactly are and convert exactly those event
> helpers to dentry ones...
>
It is not accurate that I target only "directory modification" events.
I also target FS_ATTRIB and in the future also FS_DELETE_SELF,
both applicable to files as well as directories.
But when I think about it... fanotify does not really need the dentry
information,
so I might just be able to do without FSNOTIFY_EVENT_DENTRY
and avert the questions about stability of dentry->d_inode
fanotify_dentry branch uses the dentry information in 3 occasions:
1. if (d_is_dir(dentry) in fanotify_group_event_mask()
This could be converted to S_ISDIR(inode->i_mode)
2. exportfs_encode_fh(dentry,... in fanotify_alloc_fid()
Can be converted to exportfs_encode_inode_fh(inode,...
3. statfs_by_dentry(dentry,... in fanotify_alloc_fid()
Can be converted to statfs_by_dentry(inode->i_sb->sb_root,...
I will leave the patch to annotate "directory change" events,
but the rest of this prep series can be discarded.
Does that sound reasonable?
Thanks,
Amir.
