On Wed, Nov 25, 2020 at 10:47:38AM -0800, Sargun Dhillon wrote:
> On Wed, Nov 25, 2020 at 10:43 AM Vivek Goyal <vgoyal@xxxxxxxxxx> wrote:
> >
> > On Wed, Nov 25, 2020 at 10:31:36AM -0800, Sargun Dhillon wrote:
> > > On Wed, Nov 25, 2020 at 10:17 AM Vivek Goyal <vgoyal@xxxxxxxxxx> wrote:
> > > >
> > > > On Wed, Nov 25, 2020 at 02:46:20AM -0800, Sargun Dhillon wrote:
> > > >
> > > > [..]
> > > > > @@ -1125,16 +1183,19 @@ static int ovl_workdir_cleanup_recurse(struct path *path, int level)
> > > > > if (p->len == 2 && p->name[1] == '.')
> > > > > continue;
> > > > > } else if (incompat) {
> > > > > - pr_err("overlay with incompat feature '%s' cannot be mounted\n",
> > > > > - p->name);
> > > > > - err = -EINVAL;
> > > > > - break;
> > > > > + err = ovl_check_incompat(ofs, p, path);
> > > > > + if (err < 0)
> > > > > + break;
> > > > > + /* Skip cleaning this */
> > > > > + if (err == 1)
> > > > > + continue;
> > > > > }
> > > >
> > > > Shouldn't we clean volatile/dirty on non-volatile mount. I did a
> > > > volatile mount followed by a non-volatile remount and I still
> > > > see work/incompat/volatile/dirty and "trusted.overlay.volatile" xattr
> > > > on "volatile" dir. I would expect that this will be all cleaned up
> > > > as soon as that upper/work is used for non-volatile mount.
> > > >
> > > >
> > >
> > > Amir pointed out this is incorrect behaviour earlier.
> > > You should be able to go:
> > > non-volatile -> volatile
> > > volatile -> volatile
> > >
> > > But never
> > > volatile -> non-volatile, since our mechanism is not bulletproof.
> >
> > Ok, so one needs to manually remove volatile/dirty to be able to
> > go from volatile to non-volatile.
> >
> > I am wondering what does this change mean in terms of user visible
> > behavior. So far, if somebody tried a remount of volatile overlay, it
> > will fail. After this change, it will most likely succeed. I am
> > hoping nobody relies on remount failure of volatile mount and
> > complain that user visible behavior changed after kernel upgrade.
> >
> > Thanks
> > Vivek
> >
> If I respin this shortly, can we get it in rc6, or do we want to wait
> until 5.11?
I think that trying to squeeze it in this late in cycle is probably
not a good idea. If above is a valid concern, then this feature probably
needs to be an opt-in.
Thanks
Vivek
