You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

532 lines
18 KiB

WHATS_NEW | 2 +
lib/metadata/raid_manip.c | 277 +++++++++++++++++++++++++++++++---
test/shell/lvconvert-raid-takeover.sh | 53 ++++++-
tools/lvconvert.c | 24 +--
4 files changed, 314 insertions(+), 42 deletions(-)
diff --git a/WHATS_NEW b/WHATS_NEW
index 5cbf4ec..6a0c311 100644
--- a/WHATS_NEW
+++ b/WHATS_NEW
@@ -1,5 +1,7 @@
Version 2.02.167 -
======================================
+ Add direct striped -> raid4 conversion
+ Fix raid4 parity image pair position on conversions from striped/raid0*
Disable lvconvert of thin pool to raid while active.
Version 2.02.166 - 26th September 2016
diff --git a/lib/metadata/raid_manip.c b/lib/metadata/raid_manip.c
index 5fc520e..e5fd195 100644
--- a/lib/metadata/raid_manip.c
+++ b/lib/metadata/raid_manip.c
@@ -2459,7 +2459,7 @@ static struct lv_segment *_convert_striped_to_raid0(struct logical_volume *lv,
0 /* chunk_size */,
0 /* seg->region_size */, 0u /* extents_copied */ ,
NULL /* pvmove_source_seg */))) {
- log_error("Failed to allocate new raid0 segement for LV %s.", display_lvname(lv));
+ log_error("Failed to allocate new raid0 segment for LV %s.", display_lvname(lv));
return NULL;
}
@@ -2519,42 +2519,51 @@ static struct possible_takeover_reshape_type _possible_takeover_reshape_types[]
{ .current_types = SEG_STRIPED_TARGET, /* linear, i.e. seg->area_count = 1 */
.possible_types = SEG_RAID1,
.current_areas = 1,
- .options = ALLOW_NONE },
+ .options = ALLOW_NONE }, /* FIXME: ALLOW_REGION_SIZE */
{ .current_types = SEG_STRIPED_TARGET, /* linear, i.e. seg->area_count = 1 */
.possible_types = SEG_RAID0|SEG_RAID0_META,
.current_areas = 1,
.options = ALLOW_STRIPE_SIZE },
- { .current_types = SEG_STRIPED_TARGET, /* striped, i.e. seg->area_count > 1 */
+ { .current_types = SEG_STRIPED_TARGET, /* striped -> raid0*, i.e. seg->area_count > 1 */
.possible_types = SEG_RAID0|SEG_RAID0_META,
.current_areas = ~0U,
.options = ALLOW_NONE },
+ { .current_types = SEG_STRIPED_TARGET, /* striped -> raid4 , i.e. seg->area_count > 1 */
+ .possible_types = SEG_RAID4,
+ .current_areas = ~0U,
+ .options = ALLOW_NONE }, /* FIXME: ALLOW_REGION_SIZE */
/* raid0* -> */
{ .current_types = SEG_RAID0|SEG_RAID0_META, /* seg->area_count = 1 */
.possible_types = SEG_RAID1,
.current_areas = 1,
+ .options = ALLOW_NONE }, /* FIXME: ALLOW_REGION_SIZE */
+ { .current_types = SEG_RAID0|SEG_RAID0_META, /* raid0* -> striped, i.e. seg->area_count > 1 */
+ .possible_types = SEG_STRIPED_TARGET,
+ .current_areas = ~0U,
.options = ALLOW_NONE },
- { .current_types = SEG_RAID0|SEG_RAID0_META, /* seg->area_count > 1 */
- .possible_types = SEG_RAID4,
+ { .current_types = SEG_RAID0|SEG_RAID0_META, /* raid0* -> raid0*, i.e. seg->area_count > 1 */
+ .possible_types = SEG_RAID0_META|SEG_RAID0,
.current_areas = ~0U,
.options = ALLOW_NONE },
- { .current_types = SEG_RAID0|SEG_RAID0_META, /* raid0 striped, i.e. seg->area_count > 0 */
+ { .current_types = SEG_RAID0|SEG_RAID0_META, /* raid0* -> raid4, i.e. seg->area_count > 1 */
+ .possible_types = SEG_RAID4,
+ .current_areas = ~0U,
+ .options = ALLOW_NONE }, /* FIXME: ALLOW_REGION_SIZE */
+ /* raid4 -> -> */
+ { .current_types = SEG_RAID4, /* raid4 ->striped/raid0*, i.e. seg->area_count > 1 */
.possible_types = SEG_STRIPED_TARGET|SEG_RAID0|SEG_RAID0_META,
.current_areas = ~0U,
.options = ALLOW_NONE },
- /* raid1 -> */
+ /* raid1 -> mirror */
{ .current_types = SEG_RAID1,
- .possible_types = SEG_RAID1|SEG_MIRROR,
+ .possible_types = SEG_MIRROR,
.current_areas = ~0U,
- .options = ALLOW_NONE },
+ .options = ALLOW_NONE }, /* FIXME: ALLOW_REGION_SIZE */
/* mirror -> raid1 with arbitrary number of legs */
{ .current_types = SEG_MIRROR,
- .possible_types = SEG_MIRROR|SEG_RAID1,
- .current_areas = ~0U,
- .options = ALLOW_NONE },
- { .current_types = SEG_RAID4,
- .possible_types = SEG_STRIPED_TARGET|SEG_RAID0|SEG_RAID0_META,
+ .possible_types = SEG_RAID1,
.current_areas = ~0U,
- .options = ALLOW_NONE },
+ .options = ALLOW_NONE }, /* FIXME: ALLOW_REGION_SIZE */
/* END */
{ .current_types = 0 }
@@ -2861,9 +2870,176 @@ static int _raid1_to_mirrored_wrapper(TAKEOVER_FN_ARGS)
allocate_pvs, 1, &removal_lvs);
}
+/*
+ * HM Helper: (raid0_meta -> raid4)
+ *
+ * To convert raid0_meta to raid4, which involves shifting the
+ * parity device to lv segment area 0 and thus changing MD
+ * array roles, detach the MetaLVs and reload as raid0 in
+ * order to wipe them then reattach and set back to raid0_meta.
+ */
+static int _clear_meta_lvs(struct logical_volume *lv)
+{
+ uint32_t s;
+ struct lv_segment *seg = first_seg(lv);
+ struct lv_segment_area *tmp_areas;
+ const struct segment_type *tmp_segtype;
+ struct dm_list meta_lvs;
+ struct lv_list *lvl_array, *lvl;
+
+ /* Reject non-raid0_meta segment types cautiously */
+ if (!seg_is_raid0_meta(seg) ||
+ !seg->meta_areas)
+ return_0;
+
+ if (!(lvl_array = dm_pool_alloc(lv->vg->vgmem, seg->area_count * sizeof(*lvl_array))))
+ return_0;
+
+ dm_list_init(&meta_lvs);
+ tmp_areas = seg->meta_areas;
+
+ /* Extract all MetaLVs listing them on @meta_lvs */
+ log_debug_metadata("Extracting all MetaLVs of %s to activate as raid0",
+ display_lvname(lv));
+ if (!_extract_image_component_sublist(seg, RAID_META, 0, seg->area_count, &meta_lvs, 0))
+ return_0;
+
+ /* Memorize meta areas and segtype to set again after initializing. */
+ seg->meta_areas = NULL;
+ tmp_segtype = seg->segtype;
+
+ if (!(seg->segtype = get_segtype_from_flag(lv->vg->cmd, SEG_RAID0)) ||
+ !lv_update_and_reload(lv))
+ return_0;
+
+ /*
+ * Now deactivate the MetaLVs before clearing, so
+ * that _clear_lvs() will activate them visible.
+ */
+ log_debug_metadata("Deactivating pulled out MetaLVs of %s before initializing.",
+ display_lvname(lv));
+ dm_list_iterate_items(lvl, &meta_lvs)
+ if (!deactivate_lv(lv->vg->cmd, lvl->lv))
+ return_0;
+
+ log_debug_metadata("Clearing allocated raid0_meta metadata LVs for conversion to raid4");
+ if (!_clear_lvs(&meta_lvs)) {
+ log_error("Failed to initialize metadata LVs");
+ return 0;
+ }
+
+ /* Set memorized meta areas and raid0_meta segtype */
+ seg->meta_areas = tmp_areas;
+ seg->segtype = tmp_segtype;
+
+ log_debug_metadata("Adding metadata LVs back into %s", display_lvname(lv));
+ s = 0;
+ dm_list_iterate_items(lvl, &meta_lvs) {
+ lv_set_hidden(lvl->lv);
+ if (!set_lv_segment_area_lv(seg, s++, lvl->lv, 0, RAID_META))
+ return 0;
+ }
+
+ return 1;
+}
+
+/*
+ * HM Helper: (raid0* <-> raid4)
+ *
+ * Rename SubLVs (pairs) allowing to shift names w/o collisions with active ones.
+ */
+#define SLV_COUNT 2
+static int _rename_area_lvs(struct logical_volume *lv, const char *suffix)
+{
+ uint32_t s;
+ size_t sz = strlen("rimage") + (suffix ? strlen(suffix) : 0) + 1;
+ char *sfx[SLV_COUNT] = { NULL, NULL };
+ struct lv_segment *seg = first_seg(lv);
+
+ /* Create _generate_raid_name() suffixes w/ or w/o passed in @suffix */
+ for (s = 0; s < SLV_COUNT; s++)
+ if (!(sfx[s] = dm_pool_alloc(lv->vg->cmd->mem, sz)) ||
+ dm_snprintf(sfx[s], sz, suffix ? "%s%s" : "%s", s ? "rmeta" : "rimage", suffix) < 0)
+ return_0;
+
+ /* Change names (temporarily) to be able to shift numerical name suffixes */
+ for (s = 0; s < seg->area_count; s++) {
+ if (!(seg_lv(seg, s)->name = _generate_raid_name(lv, sfx[0], s)))
+ return_0;
+ if (seg->meta_areas &&
+ !(seg_metalv(seg, s)->name = _generate_raid_name(lv, sfx[1], s)))
+ return_0;
+ }
+
+ for (s = 0; s < SLV_COUNT; s++)
+ dm_pool_free(lv->vg->cmd->mem, sfx[s]);
+
+ return 1;
+}
+
+/*
+ * HM Helper: (raid0* <-> raid4)
+ *
+ * Switch area LVs in lv segment @seg indexed by @s1 and @s2
+ */
+static void _switch_area_lvs(struct lv_segment *seg, uint32_t s1, uint32_t s2)
+{
+ struct logical_volume *lvt;
+
+ lvt = seg_lv(seg, s1);
+ seg_lv(seg, s1) = seg_lv(seg, s2);
+ seg_lv(seg, s2) = lvt;
+
+ /* Be cautious */
+ if (seg->meta_areas) {
+ lvt = seg_metalv(seg, s1);
+ seg_metalv(seg, s1) = seg_metalv(seg, s2);
+ seg_metalv(seg, s2) = lvt;
+ }
+}
+
+/*
+ * HM Helper:
+ *
+ * shift range of area LVs in @seg in range [ @s1, @s2 ] up if @s1 < @s2,
+ * else down bubbling the parity SubLVs up/down whilst shifting.
+ */
+static void _shift_area_lvs(struct lv_segment *seg, uint32_t s1, uint32_t s2)
+{
+ uint32_t s;
+
+ if (s1 < s2)
+ /* Forward shift n+1 -> n */
+ for (s = s1; s < s2; s++)
+ _switch_area_lvs(seg, s, s + 1);
+ else
+ /* Reverse shift n-1 -> n */
+ for (s = s1; s > s2; s--)
+ _switch_area_lvs(seg, s, s - 1);
+}
+
+/*
+ * Switch position of first and last area lv within
+ * @lv to move parity SubLVs from end to end.
+ *
+ * Direction depends on segment type raid4 / raid0_meta.
+ */
+static int _shift_parity_dev(struct lv_segment *seg)
+{
+ if (seg_is_raid0_meta(seg))
+ _shift_area_lvs(seg, seg->area_count - 1, 0);
+ else if (seg_is_raid4(seg))
+ _shift_area_lvs(seg, 0, seg->area_count - 1);
+ else
+ return 0;
+
+ return 1;
+}
+
/* raid45 -> raid0* / striped */
static int _raid456_to_raid0_or_striped_wrapper(TAKEOVER_FN_ARGS)
{
+ int rename_sublvs = 0;
struct lv_segment *seg = first_seg(lv);
struct dm_list removal_lvs;
@@ -2879,10 +3055,39 @@ static int _raid456_to_raid0_or_striped_wrapper(TAKEOVER_FN_ARGS)
if (!_raid_in_sync(lv))
return 0;
+ if (!yes && yes_no_prompt("Are you sure you want to convert \"%s\" LV %s to \"%s\" "
+ "type using all resilience? [y/n]: ",
+ lvseg_name(seg), display_lvname(lv), new_segtype->name) == 'n') {
+ log_error("Logical volume %s NOT converted to \"%s\"",
+ display_lvname(lv), new_segtype->name);
+ return 0;
+ }
+ if (sigint_caught())
+ return_0;
+
/* Archive metadata */
if (!archive(lv->vg))
return_0;
+ /*
+ * raid4 (which actually gets mapped to raid5/dedicated first parity disk)
+ * needs shifting of SubLVs to move the parity SubLV pair in the first area
+ * to the last one before conversion to raid0[_meta]/striped to allow for
+ * SubLV removal from the end of the areas arrays.
+ */
+ if (seg_is_raid4(seg)) {
+ /* Shift parity SubLV pair "PDD..." -> "DD...P" to be able to remove it off the end */
+ if (!_shift_parity_dev(seg))
+ return 0;
+
+ if (segtype_is_any_raid0(new_segtype) &&
+ !(rename_sublvs = _rename_area_lvs(lv, "_"))) {
+ log_error("Failed to rename %s LV %s MetaLVs", lvseg_name(seg), display_lvname(lv));
+ return 0;
+ }
+
+ }
+
/* Remove meta and data LVs requested */
if (!_lv_raid_change_image_count(lv, new_image_count, allocate_pvs, &removal_lvs, 0, 0))
return 0;
@@ -2902,7 +3107,19 @@ static int _raid456_to_raid0_or_striped_wrapper(TAKEOVER_FN_ARGS)
seg->region_size = 0;
- return _lv_update_reload_fns_reset_eliminate_lvs(lv, &removal_lvs);
+ if (!_lv_update_reload_fns_reset_eliminate_lvs(lv, &removal_lvs))
+ return_0;
+
+ if (rename_sublvs) {
+ if (!_rename_area_lvs(lv, NULL)) {
+ log_error("Failed to rename %s LV %s MetaLVs", lvseg_name(seg), display_lvname(lv));
+ return 0;
+ }
+ if (!lv_update_and_reload(lv))
+ return_0;
+ }
+
+ return 1;
}
static int _striped_to_raid0_wrapper(struct logical_volume *lv,
@@ -2930,6 +3147,9 @@ static int _striped_to_raid0_wrapper(struct logical_volume *lv,
static int _striped_or_raid0_to_raid45610_wrapper(TAKEOVER_FN_ARGS)
{
struct lv_segment *seg = first_seg(lv);
+ struct dm_list removal_lvs;
+
+ dm_list_init(&removal_lvs);
if (seg_is_raid10(seg))
return _takeover_unsupported_yet(lv, new_stripes, new_segtype);
@@ -2944,6 +3164,13 @@ static int _striped_or_raid0_to_raid45610_wrapper(TAKEOVER_FN_ARGS)
return 0;
}
+ /* FIXME: restricted to raid4 for the time being... */
+ if (!segtype_is_raid4(new_segtype)) {
+ /* Can't convert striped/raid0* to e.g. raid10_offset */
+ log_error("Can't convert %s to %s", display_lvname(lv), new_segtype->name);
+ return 0;
+ }
+
/* Archive metadata */
if (!archive(lv->vg))
return_0;
@@ -2961,7 +3188,10 @@ static int _striped_or_raid0_to_raid45610_wrapper(TAKEOVER_FN_ARGS)
log_debug_metadata("Adding metadata LVs to %s", display_lvname(lv));
if (!_raid0_add_or_remove_metadata_lvs(lv, 1 /* update_and_reload */, allocate_pvs, NULL))
return 0;
- }
+ /* raid0_meta -> raid4 needs clearing of MetaLVs in order to avoid raid disk role cahnge issues in the kernel */
+ } else if (segtype_is_raid4(new_segtype) &&
+ !_clear_meta_lvs(lv))
+ return 0;
/* Add the additional component LV pairs */
log_debug_metadata("Adding %" PRIu32 " component LV pair(s) to %s", new_image_count - lv_raid_image_count(lv),
@@ -2969,8 +3199,9 @@ static int _striped_or_raid0_to_raid45610_wrapper(TAKEOVER_FN_ARGS)
if (!_lv_raid_change_image_count(lv, new_image_count, allocate_pvs, NULL, 0, 1))
return 0;
- if (!segtype_is_raid4(new_segtype)) {
- /* Can't convert striped/raid0* to e.g. raid10_offset */
+ if (segtype_is_raid4(new_segtype) &&
+ (!_shift_parity_dev(seg) ||
+ !_rename_area_lvs(lv, "_"))) {
log_error("Can't convert %s to %s", display_lvname(lv), new_segtype->name);
return 0;
}
@@ -2987,6 +3218,14 @@ static int _striped_or_raid0_to_raid45610_wrapper(TAKEOVER_FN_ARGS)
if (!_lv_update_reload_fns_reset_eliminate_lvs(lv, NULL))
return_0;
+ if (segtype_is_raid4(new_segtype)) {
+ /* We had to rename SubLVs because of collision free sgifting, rename back... */
+ if (!_rename_area_lvs(lv, NULL))
+ return 0;
+ if (!lv_update_and_reload(lv))
+ return_0;
+ }
+
return 1;
}
diff --git a/test/shell/lvconvert-raid-takeover.sh b/test/shell/lvconvert-raid-takeover.sh
index 19a65d3..0140e22 100644
--- a/test/shell/lvconvert-raid-takeover.sh
+++ b/test/shell/lvconvert-raid-takeover.sh
@@ -78,22 +78,58 @@ aux wait_for_sync $vg $lv1
# Clean up
lvremove --yes $vg/$lv1
-# Create 3-way raid0
-lvcreate -y -aey --type raid0 -i 3 -L 64M -n $lv1 $vg
-check lv_field $vg/$lv1 segtype "raid0"
+# Create 3-way striped
+lvcreate -y -aey --type striped -i 3 -L 64M -n $lv1 $vg
+check lv_field $vg/$lv1 segtype "striped"
check lv_field $vg/$lv1 stripes 3
echo y | mkfs -t ext4 /dev/mapper/$vg-$lv1
fsck -fn /dev/mapper/$vg-$lv1
-# Convert raid0 -> raid4
+# Create 3-way raid0
+lvcreate -y -aey --type raid0 -i 3 -L 64M -n $lv2 $vg
+check lv_field $vg/$lv2 segtype "raid0"
+check lv_field $vg/$lv2 stripes 3
+echo y | mkfs -t ext4 /dev/mapper/$vg-$lv2
+fsck -fn /dev/mapper/$vg-$lv2
+
+# Create 3-way raid0_meta
+lvcreate -y -aey --type raid0_meta -i 3 -L 64M -n $lv3 $vg
+check lv_field $vg/$lv3 segtype "raid0_meta"
+check lv_field $vg/$lv3 stripes 3
+echo y | mkfs -t ext4 /dev/mapper/$vg-$lv3
+fsck -fn /dev/mapper/$vg-$lv3
+
+# Create 3-way raid4
+lvcreate -y -aey --type raid4 -i 3 -L 64M -n $lv4 $vg
+check lv_field $vg/$lv4 segtype "raid4"
+check lv_field $vg/$lv4 stripes 4
+echo y | mkfs -t ext4 /dev/mapper/$vg-$lv4
+fsck -fn /dev/mapper/$vg-$lv4
+aux wait_for_sync $vg $lv4
+fsck -fn /dev/mapper/$vg-$lv4
+
+# Convert raid4 -> striped (correct raid4 mapping test!)
+lvconvert -y --ty striped $vg/$lv4
+check lv_field $vg/$lv4 segtype "striped"
+check lv_field $vg/$lv4 stripes 3
+fsck -fn /dev/mapper/$vg-$lv4
+
+# Convert striped -> raid4
lvconvert -y --ty raid4 $vg/$lv1
-lvchange --refresh $vg/$lv1
check lv_field $vg/$lv1 segtype "raid4"
check lv_field $vg/$lv1 stripes 4
fsck -fn /dev/mapper/$vg-$lv1
aux wait_for_sync $vg $lv1
fsck -fn /dev/mapper/$vg-$lv1
+# Convert raid0 -> raid4
+lvconvert -y --ty raid4 $vg/$lv2
+check lv_field $vg/$lv2 segtype "raid4"
+check lv_field $vg/$lv2 stripes 4
+fsck -fn /dev/mapper/$vg-$lv2
+aux wait_for_sync $vg $lv2
+fsck -fn /dev/mapper/$vg-$lv2
+
# Convert raid4 -> raid0_meta
lvconvert -y --ty raid0_meta $vg/$lv1
check lv_field $vg/$lv1 segtype "raid0_meta"
@@ -116,11 +152,16 @@ fsck -fn /dev/mapper/$vg-$lv1
# Convert raid0 -> raid4
lvconvert -y --ty raid4 $vg/$lv1
-lvchange --refresh $vg/$lv1
check lv_field $vg/$lv1 segtype "raid4"
check lv_field $vg/$lv1 stripes 4
fsck -fn /dev/mapper/$vg-$lv1
aux wait_for_sync $vg $lv1
fsck -fn /dev/mapper/$vg-$lv1
+# Convert raid4 -> striped
+lvconvert -y --ty striped $vg/$lv1
+check lv_field $vg/$lv1 segtype "striped"
+check lv_field $vg/$lv1 stripes 3
+fsck -fn /dev/mapper/$vg-$lv1
+
vgremove -ff $vg
diff --git a/tools/lvconvert.c b/tools/lvconvert.c
index 0d2a4d1..541df72 100644
--- a/tools/lvconvert.c
+++ b/tools/lvconvert.c
@@ -1931,7 +1931,7 @@ static int _lvconvert_raid(struct logical_volume *lv, struct lvconvert_params *l
return 1;
}
goto try_new_takeover_or_reshape;
- } else if (!lp->repair && !lp->replace && (!*lp->type_str || seg->segtype == lp->segtype)) {
+ } else if (!lp->repair && !lp->replace && !*lp->type_str) {
log_error("Conversion operation not yet supported.");
return 0;
}
@@ -2017,28 +2017,18 @@ static int _lvconvert_raid(struct logical_volume *lv, struct lvconvert_params *l
return 1;
}
-
try_new_takeover_or_reshape:
-
/* FIXME This needs changing globally. */
if (!arg_is_set(cmd, stripes_long_ARG))
lp->stripes = 0;
- /* Only let raid4 through for now. */
- if (lp->type_str && lp->type_str[0] && lp->segtype != seg->segtype &&
- ((seg_is_raid4(seg) && seg_is_striped(lp) && lp->stripes > 1) ||
- (seg_is_striped(seg) && seg->area_count > 1 && seg_is_raid4(lp)))) {
- if (!lv_raid_convert(lv, lp->segtype, lp->yes, lp->force, lp->stripes, lp->stripe_size_supplied, lp->stripe_size,
- lp->region_size, lp->pvh))
- return_0;
-
- log_print_unless_silent("Logical volume %s successfully converted.",
- display_lvname(lv));
- return 1;
- }
+ if (!lv_raid_convert(lv, lp->segtype, lp->yes, lp->force, lp->stripes, lp->stripe_size_supplied, lp->stripe_size,
+ lp->region_size, lp->pvh))
+ return_0;
- log_error("Conversion operation not yet supported.");
- return 0;
+ log_print_unless_silent("Logical volume %s successfully converted.",
+ display_lvname(lv));
+ return 1;
}
static int _lvconvert_splitsnapshot(struct cmd_context *cmd, struct logical_volume *cow,