On Fri, Mar 19, 2021 at 11:35 AM Amir Goldstein [off-list ref] wrote:

One thing that we will probably need to do is use the RENAME_WHITEOUT
interface as the explicit way to create the shared whiteout instead of using
vfs_whiteout() for filesystems that support RENAME_WHITEOUT
(we check for RENAME_WHITEOUT support anyway).

The only thing that bothered me in moving from per-ovl-instance singleton
to per-ext4-singleton is what happens if someone tries to (say) chown -R
the upper layer or some other offline modification that was working up to
now and seemed to make sense.

Eek.

Surely, the ext4 singleton whiteout cannot allow modifications like that,
so what do we do about this? Let those scripts fail (if they exist) and
let their owners fix them to skip errors on whiteouts?

Might try that.  But the no-regressions rule means we'd have to change
that in case it breaks something.

Thanks,
Miklos

Thanks,
Amir.