LVM: New lvconvert option, '--replace'

PATCH Superseded Jonathan Brassow 2 participants 3 messages Nov 10, 2011 – Nov 14, 2011

lore

Share & tools

lore

Jonathan Brassow

author

2011-11-10 22:37 UTC #0 lore raw

PATCH 1/1 LVM: New lvconvert option, '--replace'

brassow

Support the ability to replace specific devices in a RAID array.

RAID is not like traditional LVM mirroring. LVM mirroring required failed devices to be removed or the logical volume would simply hang. RAID arrays can keep on running with failed devices. In fact, for RAID types other than RAID1, removing a device would mean substituting an error target or converting to a lower level RAID (e.g. RAID6 -> RAID5, or RAID4/5 to RAID0). Therefore, rather than removing a failed device unconditionally and potentially allocating a replacement, RAID allows the user to "replace" a device with a new one. This approach is a 1-step solution vs the current 2-step solution.

example> lvconvert --replace <dev_to_remove> vg/lv [possible_replacement_PVs]

'--replace' can be specified more than once.

eg> lvconvert --replace /dev/sdb1 --replace /dev/sdc1 vg/lv

lib/metadata/raid_manip.c +208 -0

Index: LVM2/lib/metadata/raid_manip.c
===================================================================
@@ -1428,3 +1428,211 @@ int lv_raid_reshape(struct logical_volum
 		  seg->segtype->name, new_segtype->name);
 	return 0;
 }
+
+/*
+ * lv_raid_replace
+ * @lv
+ * @replace_pvs
+ * @allocatable_pvs
+ *
+ * Replace the specified PVs.
+ */
+int lv_raid_replace(struct logical_volume *lv,
+		    struct dm_list *remove_pvs,
+		    struct dm_list *allocate_pvs)
+{
+	uint32_t s, sd, match_count = 0;
+	struct dm_list old_meta_lvs, old_data_lvs;
+	struct dm_list new_meta_lvs, new_data_lvs;
+	struct lv_segment *raid_seg = first_seg(lv);
+	struct lv_list *lvl;
+	char *tmp_names[raid_seg->area_count * 2];
+
+	dm_list_init(&old_meta_lvs);
+	dm_list_init(&old_data_lvs);
+	dm_list_init(&new_meta_lvs);
+	dm_list_init(&new_data_lvs);
+
+	/*
+	 * How many sub-LVs are being removed?
+	 */
+	for (s = 0; s < raid_seg->area_count; s++) {
+		if ((seg_type(raid_seg, s) == AREA_UNASSIGNED) ||
+		    (seg_metatype(raid_seg, s) == AREA_UNASSIGNED)) {
+			log_error("Unable to replace RAID images while the "
+				  "array has unassigned areas");
+			return 0;
+		}
+
+		if (_lv_is_on_pvs(seg_lv(raid_seg, s), remove_pvs) ||
+		    _lv_is_on_pvs(seg_metalv(raid_seg, s), remove_pvs))
+			match_count++;
+	}
+
+	if (!match_count) {
+		log_verbose("%s/%s does not contain devices specified"
+			    " for replacement", lv->vg->name, lv->name);
+		return 1;
+	} else if (match_count == raid_seg->area_count) {
+		log_error("Unable to remove all PVs from %s/%s at once.",
+			  lv->vg->name, lv->name);
+		return 0;
+	} else if (raid_seg->segtype->parity_devs &&
+		   (match_count > raid_seg->segtype->parity_devs)) {
+		log_error("Unable to replace more than %u PVs from (%s) %s/%s",
+			  raid_seg->segtype->parity_devs,
+			  raid_seg->segtype->name, lv->vg->name, lv->name);
+		return 0;
+	}
+
+	/*
+	 * Allocate the new image components first
+	 * - This makes it easy to avoid all currently used devs
+	 * - We can immediately tell if there is enough space
+	 *
+	 * - We need to change the LV names when we insert them.
+	 */
+	if (!_alloc_image_components(lv, allocate_pvs, match_count,
+				     &new_meta_lvs, &new_data_lvs)) {
+		log_error("Failed to allocate replacement images for %s/%s",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
+
+	/*
+	 * Remove the old images
+	 * - If we did this before the allocate, we wouldn't have to rename
+	 *   the allocated images, but it'd be much harder to avoid the right
+	 *   PVs during allocation.
+	 */
+	if (!_raid_extract_images(lv, raid_seg->area_count - match_count,
+				  remove_pvs, 0,
+				  &old_meta_lvs, &old_data_lvs)) {
+		log_error("Failed to remove the specified images from %s/%s",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
+
+	/*
+	 * Skip metadata operation normally done to clear the metadata sub-LVs.
+	 *
+	 * The LV_REBUILD flag is set on the new sub-LVs,
+	 * so they will be rebuilt and we don't need to clear the metadata dev.
+	 */
+
+	for (s = 0; s < raid_seg->area_count; s++) {
+		tmp_names[s] = NULL;
+		sd = s + raid_seg->area_count;
+		tmp_names[sd] = NULL;
+
+		if ((seg_type(raid_seg, s) == AREA_UNASSIGNED) &&
+		    (seg_metatype(raid_seg, s) == AREA_UNASSIGNED)) {
+			/* Adjust the new metadata LV name */
+			lvl = dm_list_item(dm_list_first(&new_meta_lvs),
+					   struct lv_list);
+			dm_list_del(&lvl->list);
+			tmp_names[s] = dm_pool_alloc(lv->vg->vgmem,
+						    strlen(lvl->lv->name) + 1);
+			if (!tmp_names[s])
+				return_0;
+			sprintf(tmp_names[s], "%s_rmeta_%u", lv->name, s);
+			if (!set_lv_segment_area_lv(raid_seg, s, lvl->lv, 0,
+						    lvl->lv->status)) {
+				log_error("Failed to add %s to %s",
+					  lvl->lv->name, lv->name);
+				return 0;
+			}
+			lv_set_hidden(lvl->lv);
+
+			/* Adjust the new data LV name */
+			lvl = dm_list_item(dm_list_first(&new_data_lvs),
+					   struct lv_list);
+			dm_list_del(&lvl->list);
+			tmp_names[sd] = dm_pool_alloc(lv->vg->vgmem,
+						     strlen(lvl->lv->name) + 1);
+			if (!tmp_names[sd])
+				return_0;
+			sprintf(tmp_names[sd], "%s_rimage_%u", lv->name, s);
+			if (!set_lv_segment_area_lv(raid_seg, s, lvl->lv, 0,
+						    lvl->lv->status)) {
+				log_error("Failed to add %s to %s",
+					  lvl->lv->name, lv->name);
+				return 0;
+			}
+			lv_set_hidden(lvl->lv);
+		}
+	}
+
+	if (!vg_write(lv->vg)) {
+		log_error("Failed to write changes to %s in %s",
+			  lv->name, lv->vg->name);
+		return 0;
+	}
+
+	if (!suspend_lv(lv->vg->cmd, lv)) {
+		log_error("Failed to suspend %s/%s before committing changes",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
+
+	if (!vg_commit(lv->vg)) {
+		log_error("Failed to commit changes to %s in %s",
+			  lv->name, lv->vg->name);
+		return 0;
+	}
+
+	if (!resume_lv(lv->vg->cmd, lv)) {
+		log_error("Failed to resume %s/%s after committing changes",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
+
+	sync_local_dev_names(lv->vg->cmd);
+	dm_list_iterate_items(lvl, &old_meta_lvs) {
+		if (!deactivate_lv(lv->vg->cmd, lvl->lv))
+			return_0;
+		if (!lv_remove(lvl->lv))
+			return_0;
+	}
+	dm_list_iterate_items(lvl, &old_data_lvs) {
+		if (!deactivate_lv(lv->vg->cmd, lvl->lv))
+			return_0;
+		if (!lv_remove(lvl->lv))
+			return_0;
+	}
+
+	/* Update new sub-LVs to correct name */
+	for (s = 0; s < raid_seg->area_count; s++) {
+		sd = s + raid_seg->area_count;
+		if (tmp_names[s] && tmp_names[sd]) {
+			seg_metalv(raid_seg, s)->name = tmp_names[s];
+			seg_lv(raid_seg, s)->name = tmp_names[sd];
+		}
+	}
+
+	if (!vg_write(lv->vg)) {
+		log_error("Failed to write changes to %s in %s",
+			  lv->name, lv->vg->name);
+		return 0;
+	}
+
+	if (!suspend_lv(lv->vg->cmd, lv)) {
+		log_error("Failed to suspend %s/%s before committing changes",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
+
+	if (!vg_commit(lv->vg)) {
+		log_error("Failed to commit changes to %s in %s",
+			  lv->name, lv->vg->name);
+		return 0;
+	}
+
+	if (!resume_lv(lv->vg->cmd, lv)) {
+		log_error("Failed to resume %s/%s after committing changes",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
+
+	return 1;
+}

tools/args.h +1 -0

Index: LVM2/tools/args.h
===================================================================
@@ -55,6 +55,7 @@ arg(corelog_ARG, '\0', "corelog", NULL, 
 arg(mirrorlog_ARG, '\0', "mirrorlog", string_arg, 0)
 arg(splitmirrors_ARG, '\0', "splitmirrors", int_arg, 0)
 arg(trackchanges_ARG, '\0', "trackchanges", NULL, 0)
+arg(replace_ARG, '\0', "replace", string_arg, ARG_GROUPABLE)
 arg(repair_ARG, '\0', "repair", NULL, 0)
 arg(use_policies_ARG, '\0', "use-policies", NULL, 0)
 arg(monitor_ARG, '\0', "monitor", yes_no_arg, 0)

tools/commands.h +3 -2

Index: LVM2/tools/commands.h
===================================================================
@@ -100,6 +100,7 @@ xx(lvconvert,
    "[-m|--mirrors Mirrors [{--mirrorlog {disk|core|mirrored}|--corelog}]]\n"
    "\t[--type SegmentType]\n"
    "\t[--repair [--use-policies]]\n"
+   "\t[--replace PhysicalVolume]\n"
    "\t[-R|--regionsize MirrorLogRegionSize]\n"
    "\t[--alloc AllocationPolicy]\n"
    "\t[-b|--background]\n"

@@ -141,8 +142,8 @@ xx(lvconvert,
 
    alloc_ARG, background_ARG, chunksize_ARG, corelog_ARG, interval_ARG,
    merge_ARG, mirrorlog_ARG, mirrors_ARG, name_ARG, noudevsync_ARG,
-   regionsize_ARG, repair_ARG, snapshot_ARG, splitmirrors_ARG, trackchanges_ARG,
-   type_ARG, stripes_long_ARG, stripesize_ARG, test_ARG,
+   regionsize_ARG, repair_ARG, replace_ARG, snapshot_ARG, splitmirrors_ARG,
+   trackchanges_ARG, type_ARG, stripes_long_ARG, stripesize_ARG, test_ARG,
    use_policies_ARG, yes_ARG, force_ARG, zero_ARG)
 
 xx(lvcreate,

tools/lvconvert.c +41 -2

85 lines not on this page · raw message

lib/metadata/metadata-exported.h +2 -0

Index: LVM2/lib/metadata/metadata-exported.h
===================================================================
@@ -784,6 +784,8 @@ int lv_raid_split_and_track(struct logic
 int lv_raid_merge(struct logical_volume *lv);
 int lv_raid_reshape(struct logical_volume *lv,
 		    const struct segment_type *new_segtype);
+int lv_raid_replace(struct logical_volume *lv, struct dm_list *remove_pvs,
+		    struct dm_list *allocate_pvs);
 
 /* --  metadata/raid_manip.c */

libdm/ioctl/libdm-iface.c +2 -2

Index: LVM2/libdm/ioctl/libdm-iface.c
===================================================================
@@ -1644,10 +1644,10 @@ static struct dm_ioctl *_do_dm_ioctl(str
 				    	    _cmd_data_v4[dmt->type].name,
 					    strerror(errno));
 			else
-				log_error("device-mapper: %s ioctl "
+				log_error("device-mapper: %s ioctl on %s "
 					  "failed: %s",
 					  _cmd_data_v4[dmt->type].name,
-					  strerror(errno));
+					  dmi->name, strerror(errno));
 
 			/*
 			 * It's sometimes worth retrying after EBUSY in case

man/lvconvert.8.in +24 -0

Index: LVM2/man/lvconvert.8.in
===================================================================
@@ -52,6 +52,14 @@ LogicalVolume[Path]...
 [\-\-version]
 LogicalVolume[Path] [PhysicalVolume[Path]...]
 
+.br
+.B lvconvert
+\-\-replace PhysicalVolume
+[\-h|\-?|\-\-help]
+[\-v|\-\-verbose]
+[\-\-version]
+LogicalVolume[Path] [PhysicalVolume[Path]...]
+
 .SH DESCRIPTION
 lvconvert is used to change the segment type (i.e. linear, mirror, etc) or
 characteristics of a logical volume.  For example, it can add or remove the

@@ -181,6 +189,14 @@ Use \-f if you do not want any replaceme
 viz. activation/mirror_log_fault_policy or
 activation/mirror_device_fault_policy.
 .br
+
+.TP
+.I \-\-replace PhysicalVolume
+Remove the specified device (PhysicalVolume) and replace it with one that is
+available in the volume group or from the specific list provided.  This option
+is only available to RAID segment types (e.g. "raid1", "raid5", etc).
+.br
+
 .SH Examples
 "lvconvert -m1 vg00/lvol1"
 .br

@@ -270,6 +286,14 @@ Merge an image that was detached tempora
 the '\-\-trackchanges' argument back into its original mirror and
 bring its contents back up-to-date.
 
+.br
+"lvconvert --replace /dev/sdb1 vg00/my_raid1 /dev/sdf1"
+.br
+Replace the physical volume "/dev/sdb1" in the RAID1 logical volume "my_raid1"
+with the specified physical volume "/dev/sdf1".  Had the argument "/dev/sdf1"
+been left out, lvconvert would attempt to find a suitable device from those
+available in the volume group.
+
 .SH SEE ALSO
 .BR lvm (8),
 .BR vgcreate (8),

Zdenek Kabelac

2011-11-11 15:39 UTC #1 lore raw

PATCH 1/1 LVM: New lvconvert option, '--replace'

Dne 10.11.2011 23:37, Jonathan Brassow napsal(a):

Quoted 19 lines

brassow
Support the ability to replace specific devices in a RAID array.
RAID is not like traditional LVM mirroring. LVM mirroring required failed devices to be removed or the logical volume would simply hang. RAID arrays can keep on running with failed devices. In fact, for RAID types other than RAID1, removing a device would mean substituting an error target or converting to a lower level RAID (e.g. RAID6 -> RAID5, or RAID4/5 to RAID0). Therefore, rather than removing a failed device unconditionally and potentially allocating a replacement, RAID allows the user to "replace" a device with a new one. This approach is a 1-step solution vs the current 2-step solution.
example> lvconvert --replace <dev_to_remove> vg/lv [possible_replacement_PVs]
'--replace' can be specified more than once.
eg> lvconvert --replace /dev/sdb1 --replace /dev/sdc1 vg/lv

lib/metadata/raid_manip.c int lv_raid_reshape(struct logical_volum · line 1538

Quoted 1 line

lib/metadata/raid_manip.c +1 -0
+			sprintf(tmp_names[s], "%s_rmeta_%u", lv->name, s);

if (dm_snprintf() < 0) error

Quoted 40 lines

lib/metadata/raid_manip.c +40 -0

+			if (!set_lv_segment_area_lv(raid_seg, s, lvl->lv, 0,
+						    lvl->lv->status)) {
+				log_error("Failed to add %s to %s",
+					  lvl->lv->name, lv->name);
+				return 0;
+			}
+			lv_set_hidden(lvl->lv);
+
+			/* Adjust the new data LV name */
+			lvl = dm_list_item(dm_list_first(&new_data_lvs),
+					   struct lv_list);
+			dm_list_del(&lvl->list);
+			tmp_names[sd] = dm_pool_alloc(lv->vg->vgmem,
+						     strlen(lvl->lv->name) + 1);
+			if (!tmp_names[sd])
+				return_0;
+			sprintf(tmp_names[sd], "%s_rimage_%u", lv->name, s);
+			if (!set_lv_segment_area_lv(raid_seg, s, lvl->lv, 0,
+						    lvl->lv->status)) {
+				log_error("Failed to add %s to %s",
+					  lvl->lv->name, lv->name);
+				return 0;
+			}
+			lv_set_hidden(lvl->lv);
+		}
+	}
+
+	if (!vg_write(lv->vg)) {
+		log_error("Failed to write changes to %s in %s",
+			  lv->name, lv->vg->name);
+		return 0;
+	}
+
+	if (!suspend_lv(lv->vg->cmd, lv)) {
+		log_error("Failed to suspend %s/%s before committing changes",
+			  lv->vg->name, lv->name);
+		return 0;
+	}
+
+	if (!vg_commit(lv->vg)) {

lib/metadata/raid_manip.c int lv_raid_reshape(struct logical_volum · line 1588

Quoted 10 lines

lib/metadata/raid_manip.c +10 -0

+		log_error("Failed to commit changes to %s in %s",
+			  lv->name, lv->vg->name);
+		return 0;
+	}
+
+	if (!resume_lv(lv->vg->cmd, lv)) {
+		log_error("Failed to resume %s/%s after committing changes",
+			  lv->vg->name, lv->name);
+		return 0;
+	}

I think its about the time to finaly make this sequence a common function :) (vg_write(), suspend(), vg_commit(), resume_lv())

Review

Quoted 2 lines

(quoted diff) +2 -0

+
+	sync_local_dev_names(lv->vg->cmd);

Hmm - you shouldn't need to do this
(It's been bug before, where it has not been
implicitely used in deactivate_lv())

So for now - unless you have trace which shows, there is still a problem - do not add them. (In fact some of those added, should be removed, since call of deactivate_lv() is doing approriate synchronization at the right place (i.e. on clmvd in clustered case).

lib/metadata/raid_manip.c int lv_raid_reshape(struct logical_volum · line 1602

Quoted 12 lines

lib/metadata/raid_manip.c +12 -0

+	dm_list_iterate_items(lvl, &old_meta_lvs) {
+		if (!deactivate_lv(lv->vg->cmd, lvl->lv))
+			return_0;
+		if (!lv_remove(lvl->lv))
+			return_0;
+	}
+	dm_list_iterate_items(lvl, &old_data_lvs) {
+		if (!deactivate_lv(lv->vg->cmd, lvl->lv))
+			return_0;
+		if (!lv_remove(lvl->lv))
+			return_0;
+	}

Zdenek