On Wed, Feb 5, 2025 at 5:59 PM Jan Kara <jack@xxxxxxx> wrote:
>
> On Mon 03-02-25 23:32:05, Amir Goldstein wrote:
> > After introducing pre-content events, we had a regression related to
> > disabling huge faults on files that should never have pre-content events
> > enabled.
> >
> > This happened because the default f_mode of allocated files (0) does
> > not disable pre-content events.
> >
> > Pre-content events are disabled in file_set_fsnotify_mode_by_watchers()
> > but internal files may not get to call this helper.
> >
> > Initialize f_mode to disable permission and pre-content events for all
> > files and if needed they will be enabled for the callers of
> > file_set_fsnotify_mode_by_watchers().
> >
> > Fixes: 20bf82a898b6 ("mm: don't allow huge faults for files with pre content watches")
> > Reported-by: Alex Williamson <alex.williamson@xxxxxxxxxx>
> > Closes: https://lore.kernel.org/linux-fsdevel/20250131121703.1e4d00a7.alex.williamson@xxxxxxxxxx/
> > Tested-by: Alex Williamson <alex.williamson@xxxxxxxxxx>
> > Signed-off-by: Amir Goldstein <amir73il@xxxxxxxxx>
>
> Looks good. Feel free to add:
>
> Reviewed-by: Jan Kara <jack@xxxxxxx>
>
> What makes me somewhat uneasy is that this relies on the fact that
> file_set_fsnotify_mode_from_watchers() will override the
> FMODE_NONOTIFY_PERM (but it does not override FMODE_NONOTIFY).
> This seems a bit subtle and I was looking into if we could somehow simplify the fsnotify
> fmode initialization. But I didn't find anything that would be really
> simpler so let's keep what we have for now.
>
Yeh, I noticed that too.
I did not have a better idea :/
Thanks,
Amir.
