I'm okay with returning valid for directory mount point.

But the point that I'm trying to make here is that VFS reacts
differently when d_validate returns an error VS when it just returns
invalid:
https://elixir.bootlin.com/linux/latest/source/fs/namei.c#L1409

Notice how it calls d_invalidate only when there's no error. And
d_invalidate seems to have detach_mounts.
It is likely that the umount happens there.

I'm suggesting that we should return errors inside d_validate
handlers, rather than just 0 or 1.
Makes sense?

Regards,
Shyam

On Mon, Feb 1, 2021 at 4:01 PM Aurélien Aptel [off-list ref] wrote:

Shyam Prasad N [off-list ref] writes:

quoted

Going by the documentation of d_revalidate:

quoted

This function should return a positive value if the dentry is still
valid, and zero or a negative error code if it isn't.

In case of error, can we try returning the rc itself (rather than 0),
and see if VFS avoids a dentry put?
Because theoretically, the call execution has failed, and the dentry
is not found to be invalid.

AFAIK mount points are pinned, you cannot rm or mv them so it seems we
could make them always valid. I don't know if there are deeper and more
subtle implications.

The recent signal fixes are not fixing this issue.

Cheers,
--
Aurélien Aptel / SUSE Labs Samba Team
GPG: 1839 CB5F 9F5B FB9B AA97  8C99 03C8 A49B 521B D5D3
SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg, DE
GF: Felix Imendörffer, Mary Higgins, Sri Rasiah HRB 247165 (AG München)

-- 
-Shyam