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

From a83a3ec7292197bd870cae570ed545267899bc5e Mon Sep 17 00:00:00 2001
From: Eric Ren <zren@suse.com>
Date: Tue, 4 Jul 2017 19:42:14 +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>
Refreshed-by: Eric Ren <zren@suse.com>
---
 conf/example.conf.in         | 17 +++++++++++++++++
 lib/config/config_settings.h |  5 ++++-
 lib/config/defaults.h        |  1 +
 lib/metadata/lv_manip.c      | 35 +++++++++++++++++++++++++++++++++++
 man/lvcreate.8_pregen        |  7 +++++++
 5 files changed, 64 insertions(+), 1 deletion(-)

diff --git a/conf/example.conf.in b/conf/example.conf.in
index b5b0e6a..947a3cf 100644
--- a/conf/example.conf.in
+++ b/conf/example.conf.in
@@ -372,6 +372,23 @@ allocation {
 	# or erased unless the --wipesignatures option is used directly.
 	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
+
 	# Configuration option allocation/mirror_logs_require_separate_pvs.
 	# Mirror logs and images will always use different PVs.
 	# The default setting changed in version 2.02.85.
diff --git a/lib/config/config_settings.h b/lib/config/config_settings.h
index 40b64ab..59eed26 100644
--- a/lib/config/config_settings.h
+++ b/lib/config/config_settings.h
@@ -474,7 +474,10 @@ cfg(allocation_raid_stripe_all_devices_CFG, "raid_stripe_all_devices", allocatio
 	"stripes to use.\n"
 	"This was the default behaviour until release 2.02.162.\n")
 
-cfg(allocation_cache_pool_metadata_require_separate_pvs_CFG, "cache_pool_metadata_require_separate_pvs", allocation_CFG_SECTION, CFG_PROFILABLE | CFG_PROFILABLE_METADATA, CFG_TYPE_BOOL, DEFAULT_CACHE_POOL_METADATA_REQUIRE_SEPARATE_PVS, vsn(2, 2, 106), NULL, 0, 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, 0, NULL,
+	"Mirror legs will always use different PVs.\n")
+
+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, 0, NULL,
 	"Cache pool metadata and data will always use different PVs.\n")
 
 cfg(allocation_cache_pool_cachemode_CFG, "cache_pool_cachemode", allocation_CFG_SECTION, CFG_PROFILABLE | CFG_PROFILABLE_METADATA | CFG_DEFAULT_COMMENTED, CFG_TYPE_STRING, DEFAULT_CACHE_MODE, vsn(2, 2, 113), NULL, vsn(2, 2, 128),
diff --git a/lib/config/defaults.h b/lib/config/defaults.h
index d9e19d9..e5134e4 100644
--- a/lib/config/defaults.h
+++ b/lib/config/defaults.h
@@ -161,6 +161,7 @@
 #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
 
diff --git a/lib/metadata/lv_manip.c b/lib/metadata/lv_manip.c
index f75cf51..8dc90b4 100644
--- a/lib/metadata/lv_manip.c
+++ b/lib/metadata/lv_manip.c
@@ -1622,6 +1622,9 @@ 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
@@ -2736,6 +2739,32 @@ static int _limit_to_one_area_per_tag(struct alloc_handle *ah, struct alloc_stat
 }
 
 /*
+ * 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, 0))
+			return i;
+	return -1;
+}
+
+/*
  * Returns 1 regardless of whether any space was found, except on error.
  */
 static int _find_some_parallel_space(struct alloc_handle *ah,
@@ -2865,6 +2894,10 @@ static int _find_some_parallel_space(struct alloc_handle *ah,
 					continue;
 
 				case USE_AREA:
+					if(check_areas_separate_tags(ah, alloc_state, alloc_state->num_positional_areas,
+								     ix + alloc_state->num_positional_areas, pva) >= 0)
+						goto next_pv;
+
 					/*
 					 * Except with ALLOC_ANYWHERE, replace first area with this
 					 * one which is smaller but still big enough.
@@ -3355,6 +3388,7 @@ static struct alloc_handle *_alloc_init(struct cmd_context *cmd,
 	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
@@ -3366,6 +3400,7 @@ static struct alloc_handle *_alloc_init(struct cmd_context *cmd,
 	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;
diff --git a/man/lvcreate.8_pregen b/man/lvcreate.8_pregen
index 6b05147..7c1beab 100644
--- a/man/lvcreate.8_pregen
+++ b/man/lvcreate.8_pregen
@@ -415,6 +415,13 @@ Create a raid10 LV.
 [ \fB-I\fP|\fB--stripesize\fP \fISize\fP[k|UNIT] ]
 .ad b
 .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 \fBmirrored\fP will create a persistent log that is itself mirrored.
+.
+.HP
 .ad l
 [ \fB-R\fP|\fB--regionsize\fP \fISize\fP[m|UNIT] ]
 .ad b
-- 
2.10.2