Re: [PATCH v2 4/5] overlay: test constant ino with nested overlay over samefs lower

[Amir Goldstein <amir73il@gmail.com>]

On Mon, Mar 16, 2020 at 2:29 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Mon, Dec 30, 2019 at 3:14 PM Amir Goldstein <amir73il@gmail.com> wrote:
> >
> > Also test that d_ino of readdir entries and i_ino from /proc/locks are
> > consistent with st_ino and that inode numbers persist after rename to
> > new parent, drop caches and mount cycle.
>
> overlay/070 and overlay/071 fail for me like this:
>
>      QA output created by 071
>     +flock: cannot open lock file
> /scratch/ovl-mnt/lowertestdir/blkdev: No such device or address
> ...
>
> I.e. there's no block dev with rdev=1/1.
>
> I don't see any other way to fix this, than to remove the device
> tests.

I ran into similar complain when I worked on generic/564.
Apparently, this is not the first test that uses rdev b/1/1 and c/1/1
so not sure how those tests work for everyone.
In generic/564 I used a loopdev as blockdev and /dev/zero as chardev.

> Why are these needed?  Is locking code in any way dependent on
> file type?
>

Not strictly needed.
See that they already skip file types fifo|socket|symlink.

But note that we are not testing locking, we are using /proc/locks
to get a peek at i_ino, so if we skip also blockdev and chardev, we
end up testing no special files at all for i_ino consistency.
Not the end of the world, but then again using loop dev and /dev/zero
would be quite trivial as well.

If it bothers you, I can post a fix.

Thanks,
Amir.