[lvm-devel] [PATCH 8 of 13] LVM: add logic to allow mirrored log

[Takahiro Yasui]

Jonathan Brassow wrote:
>  static int _lv_update_log_type(struct cmd_context *cmd,
> @@ -657,21 +672,48 @@ static int _lv_update_log_type(struct cm
>  			       struct logical_volume *lv,
>  			       int log_count)
>  {
...
> +	log_lv = first_seg(original_lv)->log_lv;
> +
> +	/* Adding redundancy to the log */
> +	if (old_log_count < log_count) {
> +		log_error("Adding log redundancy not supported yet.");
> +		log_error("Try converting the log to 'core' first.");
> +		return_0;
>  	}
> -	return 1;
> +
> +	/* Reducing redundancy of the log */
> +	return remove_mirror_images(log_lv, log_count,
> +				    lp->pv_count ? lp->pvh : NULL, 1U);
>  }

Should we pass lp->pvh even if lp->pv_count is 0?

+	return remove_mirror_images(log_lv, log_count, lp->pvh, 1U);

Looking at remove_mirror_images(), the third agrument, removable_pvs,
is used to decide which mirror leg should be removed. When lp->pv_count
is 0, lp->pvh has the list of PVs in the VG (see lvconvert_single())
and the PV list needs to be passed to remove_mirror_images().

This issue is found during tests. When the primary leg of the mirrored
log (e.g. vg00-lv00_mlog_mimage_0) failed, lvconvert --repair vg00/lv00
didn't recover the vg00/lv00.

With the above fix, lvconvert --repair vg00/lv00 worked well when
the primary leg of the mirrored log failed.

Thanks,
Taka