On Tue, Mar 24, 2020 at 7:50 PM Jan Kara <jack@xxxxxxx> wrote:
>
> On Thu 19-03-20 17:10:18, Amir Goldstein wrote:
> > Breakup the union and make them both inherit from abstract fanotify_event.
> >
> > fanotify_path_event, fanotify_fid_event and fanotify_perm_event inherit
> > from fanotify_event.
> >
> > type field in abstract fanotify_event determines the concrete event type.
> >
> > fanotify_path_event, fanotify_fid_event and fanotify_perm_event are
> > allocated from separate memcache pools.
> >
> > The separation of struct fanotify_fid_hdr from the file handle that was
> > done for efficient packing of fanotify_event is no longer needed, so
> > re-group the file handle fields under struct fanotify_fh.
> >
> > The struct fanotify_fid, which served to group fsid and file handle for
> > the union is no longer needed so break it up.
> >
> > Rename fanotify_perm_event casting macro to FANOTIFY_PERM(), so that
> > FANOTIFY_PE() and FANOTIFY_FE() can be used as casting macros to
> > fanotify_path_event and fanotify_fid_event.
> >
> > Suggested-by: Jan Kara <jack@xxxxxxx>
> > Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx>
>
> So I was pondering about this commit. First I felt it should be split and
Oh yeh. The split makes things much clearer!
> second when splitting the commit I've realized I dislike how you rely on
> 'struct fanotify_event' being the first in events that inherit it. That is
> not well maintainable long term since over the time, hidden dependencies on
> this tend to develop (you already had like four in this patch) and then
> when you need to switch away from that in the future, you have a horrible
> time untangling the mess... I also wanted helpers like FANOTIFY_PE() to be
> inline functions to get type safety and realized you actually use
> FANOTIFY_PE() both for fsnotify_event and fanotify_event which is hacky as
Excellent! I avoided the FANOTIFY_E/fsn_event related cleanups, but now
code looks much better and safe.
> well. Finally, I've realized that fanotify was likely broken when
> generating overflow events (create_fd() was returning -EOVERFLOW which
> confused the caller - still need to write a testcase for that) and you
> silently fix that so I wanted that as separate commit as well.
I don't think you will find a test case.
Before the divorce patch, the meaning of fanotify_event_has_path() is:
event->fh_type == FILEID_ROOT;
but overflow event with NULL path has:
event->fh_type = FILEID_INVALID;
So -EOVERFLOW code in was not reachable.
Meaning that your patch "fanotify: Fix handling of overflow event" is
correct, but its commit message is wrong.
It also says: "by default fanotify event queues are unlimited",
but FAN_UNLIMITED_QUEUE is opt-in???
>
> All in all this commit ended up like three commits I'm attaching. I'd be
> happy if you could have a look through them but the final code isn't that
> different and LTP passes so I'm reasonably confident I didn't break
> anything.
The split and end result look very good.
After rebasing my fanotify_name branch on top of your changes, it also
fixed an error in FAN_REPORT_NAME test, which I was going to look
at later, so your cleanup paid off real fast :-)
Thanks,
Amir.
