lvm2/make-mirror-legs-on-different-tag-pvs.patch

From 49a5da732c74b8b96bc123a541d590c4fd06d7af Mon Sep 17 00:00:00 2001
From: Guangliang Zhao <gzhao@suse.com>
Date: Mon, 12 Nov 2012 15:28:11 +0800
Subject: [PATCH 1/2] mirror: make mirror legs on different tag pvs

The mirror legs could be placed on the PVs(with different
tags) from the other mirror legs with this patch.

The lvcreate allocates the replica "randomly" when creating
a mirrored logical volume now, so it could happen that both
copies end up on the same array (or on the same site), clearly
undesired.

We can't control the fist allocation, because didn't restrict
the areas into parallel space. This patch add the limit when
finding areas for parallel space, so that mirror legs will
always be placed on pvs with differnt tags.

Signed-off-by: Guangliang Zhao <gzhao@suse.com>
---
 conf/example.conf.in         |   17 +++++++++++++++++
 lib/config/config_settings.h |    1 +
 lib/config/defaults.h        |    1 +
 lib/metadata/lv_manip.c      |   36 ++++++++++++++++++++++++++++++++++++
 man/lvcreate.8.in            |    4 ++++
 5 files changed, 59 insertions(+)

--- a/conf/example.conf.in
+++ b/conf/example.conf.in
@@ -315,6 +315,23 @@ allocation {
     #
     wipe_signatures_when_zeroing_new_lvs = 1
 
+    # Set to 1 to guarantee that mirror leg will always be placed on
+    # different PVs(with different tags) from the other mirror legs.
+    #
+    # If you want to enable this feature, the following conditions
+    # must be met:
+    # 1) The mirror_legs_require_separate_pvs must be set to 1, the
+    #    default value is 0.
+    # 2) The cling_tag_list must be activated.
+    # 3) The length of all pvs with same tag must greater than or equal
+    #    to the mirror's.
+
+    # This feature is only for the first allocation, on the other hand
+    # when creating new mirrored lvs.
+    # Please note that the commond may fail if the number of all tags
+    # on the pvs less than the mirror legs number.
+    mirror_legs_require_separate_pvs = 0
+
     # Set to 1 to guarantee that mirror logs will always be placed on 
     # different PVs from the mirror images.  This was the default
     # until version 2.02.85.
--- a/lib/config/config_settings.h
+++ b/lib/config/config_settings.h
@@ -119,6 +119,7 @@ cfg(allocation_maximise_cling_CFG, "maxi
 cfg(allocation_use_blkid_wiping_CFG, "use_blkid_wiping", allocation_CFG_SECTION, 0, CFG_TYPE_BOOL, 1, vsn(2, 2, 105), NULL)
 cfg(allocation_wipe_signatures_when_zeroing_new_lvs_CFG, "wipe_signatures_when_zeroing_new_lvs", allocation_CFG_SECTION, 0, CFG_TYPE_BOOL, 1, vsn(2, 2, 105), NULL)
 cfg(allocation_mirror_logs_require_separate_pvs_CFG, "mirror_logs_require_separate_pvs", allocation_CFG_SECTION, 0, CFG_TYPE_BOOL, DEFAULT_MIRROR_LOGS_REQUIRE_SEPARATE_PVS, vsn(2, 2, 85), NULL)
+cfg(allocation_mirror_legs_require_separate_pvs_CFG, "mirror_legs_require_separate_pvs", allocation_CFG_SECTION, 0, CFG_TYPE_BOOL, DEFAULT_MIRROR_LEGS_REQUIRE_SEPARATE_PVS, vsn(2, 2, 98), NULL)
 cfg(allocation_cache_pool_metadata_require_separate_pvs_CFG, "cache_pool_metadata_require_separate_pvs", allocation_CFG_SECTION, 0, CFG_TYPE_BOOL, DEFAULT_CACHE_POOL_METADATA_REQUIRE_SEPARATE_PVS, vsn(2, 2, 106), NULL)
 cfg(allocation_cache_pool_cachemode_CFG, "cache_pool_cachemode", allocation_CFG_SECTION, 0, CFG_TYPE_STRING, DEFAULT_CACHE_POOL_CACHEMODE, vsn(2, 2, 113), NULL)
 cfg_runtime(allocation_cache_pool_chunk_size_CFG, "cache_pool_chunk_size", allocation_CFG_SECTION, CFG_DEFAULT_UNDEFINED, CFG_TYPE_INT, vsn(2, 2, 106), NULL)
--- a/lib/config/defaults.h
+++ b/lib/config/defaults.h
@@ -125,6 +125,7 @@
 #define DEFAULT_MAX_LV 0
 #define DEFAULT_ALLOC_POLICY ALLOC_NORMAL
 #define DEFAULT_MIRROR_LOGS_REQUIRE_SEPARATE_PVS 0
+#define DEFAULT_MIRROR_LEGS_REQUIRE_SEPARATE_PVS 0
 #define DEFAULT_MAXIMISE_CLING 1
 #define DEFAULT_CLUSTERED 0
 
--- a/lib/metadata/lv_manip.c
+++ b/lib/metadata/lv_manip.c
@@ -1409,6 +1409,9 @@ struct alloc_handle {
 
 	unsigned maximise_cling;
 	unsigned mirror_logs_separate;	/* Force mirror logs on separate PVs? */
+	unsigned mirror_legs_separate;	/* Force mirror *legs* on separate PVs*/
+
+	const struct segment_type *segtype;
 
 	/*
 	 * RAID devices require a metadata area that accompanies each
@@ -1598,6 +1601,7 @@ static struct alloc_handle *_alloc_init(
 	ah->parity_count = parity_count;
 	ah->region_size = region_size;
 	ah->alloc = alloc;
+	ah->segtype = segtype;
 
 	/*
 	 * For the purposes of allocation, area_count and parity_count are
@@ -1609,6 +1613,7 @@ static struct alloc_handle *_alloc_init(
 	ah->area_multiple = _calc_area_multiple(segtype, area_count + parity_count, stripes);
 	//FIXME: s/mirror_logs_separate/metadata_separate/ so it can be used by others?
 	ah->mirror_logs_separate = find_config_tree_bool(cmd, allocation_mirror_logs_require_separate_pvs_CFG, NULL);
+ 	ah->mirror_legs_separate = find_config_tree_bool(cmd, allocation_mirror_legs_require_separate_pvs_CFG, NULL);
 
 	if (mirrors || stripes)
 		total_extents = new_extents;
@@ -2533,6 +2538,33 @@ static void _report_needed_allocation_sp
 			(metadata_count == 1) ? "" : "s",
 			metadata_size);
 }
+
+/*
+ * Return -1 if we don't need check tags, or there aren't any areas in alloc_status
+ * have the same tag with pva, the index otherwise.
+ */
+static int check_areas_separate_tags(struct alloc_handle *ah,
+				     struct alloc_state *alloc_state,
+				     unsigned ix_start,
+				     unsigned ix_end,
+				     struct pv_area *pva)
+{
+	int i;
+
+	if (!segtype_is_mirrored(ah->segtype) ||
+			alloc_state->allocated ||
+			!ah->mirror_legs_separate ||
+			!ah->cling_tag_list_cn)
+		return -1;
+
+	for (i = ix_start; i < ix_end; i++)
+		if(_pvs_have_matching_tag(ah->cling_tag_list_cn,
+					alloc_state->areas[i].pva->map->pv,
+					pva->map->pv))
+			return i;
+	return -1;
+}
+
 /*
  * Returns 1 regardless of whether any space was found, except on error.
  */
@@ -2663,6 +2695,10 @@ static int _find_some_parallel_space(str
 					continue;
 
 				case USE_AREA:
+					if(check_areas_separate_tags(ah, alloc_state, ix_offset,
+								     ix + ix_offset, pva) >= 0)
+						goto next_pv;
+
 					/*
 					 * Except with ALLOC_ANYWHERE, replace first area with this
 					 * one which is smaller but still big enough.
--- a/man/lvcreate.8.in
+++ b/man/lvcreate.8.in
@@ -284,6 +284,10 @@ will create a persistent log that is its
 Using \fIcore\fP means the mirror is regenerated by copying the data
 from the first device each time the logical volume is activated,
 like after every reboot.
+.br
+Every leg of the mirrored logical volume could be placed on the pvs with
+same tag, see details(mirror_legs_require_separate_pvs) in lvm.conf.
+
 .br
 Using \fImirrored\fP will create a persistent log that is itself mirrored.
 .TP