[lvm-devel] [LVM2 RFCv1 4/5] lib: locking: Parse PV list for IDM locking

[Leo Yan]

On Wed, Apr 28, 2021 at 02:39:27PM -0500, David Teigland wrote:
> On Sun, Apr 25, 2021 at 10:22:40AM +0800, Leo Yan wrote:
> > +static void _lockd_retrive_lv_pv_list(struct volume_group *vg,
> > +				      const char *lv_name,
> > +				      struct lvmlockd_pvs *lock_pvs)
> > +{
> 
> It looks like this wants a list of PVs (names) used by the LV.  Try
> iterating through all PVs in the VG and using the existing lv_is_on_pv()
> function to check if the LV is using that PV, e.g.
> 
> for each pv in vg->pvs,
> 	if (lv_is_on_pv(lv, pv))
> 		copy the pv name;

Using lv_is_on_pv() is much better.  I read a bit for the code, except
it checks the LV itself is on the PV or not, it also checks sub LVs
(like metadata, pool, etc).  Will fix it.

> You could pass the lv struct through to here, or use find_lv(vg, lv_name)
> to get it again here.
> 
> > @@ -251,7 +485,16 @@ static int _lockd_request(struct cmd_context *cmd,
> >  	if (vg_name && lv_name) {
> > -		reply = _lockd_send(req_name,
> > +		/*
> > +		 * For LV operation, the PV list must be passed for idm,
> > +		 * otherwise, IDM lock manager has no idea to send locking
> > +		 * request to which drives, so return failure.
> > +		 */
> > +		if (!lock_pvs)
> > +			return 1;
> > +
> > +		reply = _lockd_send_with_pvs(req_name,
> 
> Requires other lock managers to include lock_pvs?

No, will change code to only pass "lock_pvs" for IDM; for other
locking schemes, will assign NULL pointer to "lock_pvs".

> > +	/*
> > +	 * Create the VG's PV list when start the VG, the PV list
> > +	 * is passed to lvmlockd, and the the PVs path will be used
> > +	 * to send SCSI commands for idm locking scheme.
> > +	 */
> 
> if vg->lock_type is IDM before creating pv_list?

Yes, IIUC, when start the VG, "vg->lock_type" has been assigned to the
locking scheme which it's using; for IDM locking scheme, "vg->lock_type"
is "idm" before creating pv_list.

> > +	_lockd_retrive_vg_pv_list(vg, &lock_pvs);
> > +
> > +	reply = _lockd_send_with_pvs("start_vg",
> > +				&lock_pvs,
> 
> and probably NULL instead of &lock_pvs for non-IDM.

Will do.

Will follow up the comments in next patch set, thanks for reviewing!

Leo