On Wed, Jun 7, 2023 at 11:18 AM Luís Henriques <lhenriques@xxxxxxx> wrote:
>
> Xiubo Li <xiubli@xxxxxxxxxx> writes:
>
> > On 6/6/23 17:53, Luís Henriques wrote:
> >> xiubli@xxxxxxxxxx writes:
> >>
> >>> From: Xiubo Li <xiubli@xxxxxxxxxx>
> >>>
> >>> V2:
> >>> - Improve the code by switching to wait_for_completion_killable_timeout()
> >>> when umounting, at the same add one umount_timeout option.
> >> Instead of adding yet another (undocumented!) mount option, why not re-use
> >> the already existent 'mount_timeout' instead? It's already defined and
> >> kept in 'struct ceph_options', and the default value is defined with the
> >> same value you're using, in CEPH_MOUNT_TIMEOUT_DEFAULT.
> >
> > This is for mount purpose. Is that okay to use the in umount case ?
>
> Yeah, you're probably right. It's just that adding yet another knob for a
> corner case that probably will never be used and very few people will know
> about is never a good thing (IMO). Anyway, I think that at least this new
> mount option needs to be mentioned in 'Documentation/filesystems/ceph.rst'.
It's OK and in fact preferrable to stick to mount_timeout to avoid new
knobs. There is even a precedent for this: RBD uses mount_timeout both
when waiting for the latest osdmap (on "rbd map") and when tearing down
a watch (on "rbd unmap").
Thanks,
Ilya
