Illumos #4101, #4102, #4103, #4105, #4106

4101 metaslab_debug should allow for fine-grained control
4102 space_maps should store more information about themselves
4103 space map object blocksize should be increased
4105 removing a mirrored log device results in a leaked object
4106 asynchronously load metaslab
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Sebastien Roy <seb@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>

Prior to this patch, space_maps were preferred solely based on the
amount of free space left in each. Unfortunately, this heuristic didn't
contain any information about the make-up of that free space, which
meant we could keep preferring and loading a highly fragmented space map
that wouldn't actually have enough contiguous space to satisfy the
allocation; then unloading that space_map and repeating the process.

This change modifies the space_map's to store additional information
about the contiguous space in the space_map, so that we can use this
information to make a better decision about which space_map to load.
This requires reallocating all space_map objects to increase their
bonus buffer size sizes enough to fit the new metadata.

The above feature can be enabled via a new feature flag introduced by
this change: com.delphix:spacemap_histogram

In addition to the above, this patch allows the space_map block size to
be increase. Currently the block size is set to be 4K in size, which has
certain implications including the following:

    * 4K sector devices will not see any compression benefit
    * large space_maps require more metadata on-disk
    * large space_maps require more time to load (typically random reads)

Now the space_map block size can adjust as needed up to the maximum size
set via the space_map_max_blksz variable.

A bug was fixed which resulted in potentially leaking an object when
removing a mirrored log device. The previous logic for vdev_remove() did
not deal with removing top-level vdevs that are interior vdevs (i.e.
mirror) correctly. The problem would occur when removing a mirrored log
device, and result in the DTL space map object being leaked; because
top-level vdevs don't have DTL space map objects associated with them.

References:
  https://www.illumos.org/issues/4101
  https://www.illumos.org/issues/4102
  https://www.illumos.org/issues/4103
  https://www.illumos.org/issues/4105
  https://www.illumos.org/issues/4106
  https://github.com/illumos/illumos-gate/commit/0713e23

Porting notes:

A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also,
the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary.

Ported-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Prakash Surya <surya1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #2488

module/zfs/Makefile.in

@@ -35,6 +35,7 @@ $(MODULE)-objs += @top_srcdir@/module/zfs/gzip.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/lzjb.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/lz4.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/metaslab.o
+$(MODULE)-objs += @top_srcdir@/module/zfs/range_tree.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/refcount.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/rrwlock.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/sa.o
@@ -47,6 +48,7 @@ $(MODULE)-objs += @top_srcdir@/module/zfs/spa_history.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/spa_misc.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/spa_stats.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/space_map.o
+$(MODULE)-objs += @top_srcdir@/module/zfs/space_reftree.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/txg.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/uberblock.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/unique.o

module/zfs/dnode.c

@@ -1335,7 +1335,7 @@ dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx)
 	rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
 
 	/* Check for any allocated blocks beyond the first */
-	if (dn->dn_phys->dn_maxblkid != 0)
+	if (dn->dn_maxblkid != 0)
 		goto fail;
 
 	mutex_enter(&dn->dn_dbufs_mtx);

module/zfs/range_tree.c

@@ -0,0 +1,391 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright (c) 2013 by Delphix. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/spa.h>
+#include <sys/dmu.h>
+#include <sys/dnode.h>
+#include <sys/zio.h>
+#include <sys/range_tree.h>
+
+static kmem_cache_t *range_seg_cache;
+
+void
+range_tree_init(void)
+{
+	ASSERT(range_seg_cache == NULL);
+	range_seg_cache = kmem_cache_create("range_seg_cache",
+	    sizeof (range_seg_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
+}
+
+void
+range_tree_fini(void)
+{
+	kmem_cache_destroy(range_seg_cache);
+	range_seg_cache = NULL;
+}
+
+void
+range_tree_stat_verify(range_tree_t *rt)
+{
+	range_seg_t *rs;
+	uint64_t hist[RANGE_TREE_HISTOGRAM_SIZE] = { 0 };
+	int i;
+
+	for (rs = avl_first(&rt->rt_root); rs != NULL;
+	    rs = AVL_NEXT(&rt->rt_root, rs)) {
+		uint64_t size = rs->rs_end - rs->rs_start;
+		int idx	= highbit(size) - 1;
+
+		hist[idx]++;
+		ASSERT3U(hist[idx], !=, 0);
+	}
+
+	for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
+		if (hist[i] != rt->rt_histogram[i]) {
+			zfs_dbgmsg("i=%d, hist=%p, hist=%llu, rt_hist=%llu",
+			    i, hist, hist[i], rt->rt_histogram[i]);
+		}
+		VERIFY3U(hist[i], ==, rt->rt_histogram[i]);
+	}
+}
+
+static void
+range_tree_stat_incr(range_tree_t *rt, range_seg_t *rs)
+{
+	uint64_t size = rs->rs_end - rs->rs_start;
+	int idx = highbit(size) - 1;
+
+	ASSERT3U(idx, <,
+	    sizeof (rt->rt_histogram) / sizeof (*rt->rt_histogram));
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+	rt->rt_histogram[idx]++;
+	ASSERT3U(rt->rt_histogram[idx], !=, 0);
+}
+
+static void
+range_tree_stat_decr(range_tree_t *rt, range_seg_t *rs)
+{
+	uint64_t size = rs->rs_end - rs->rs_start;
+	int idx = highbit(size) - 1;
+
+	ASSERT3U(idx, <,
+	    sizeof (rt->rt_histogram) / sizeof (*rt->rt_histogram));
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+	ASSERT3U(rt->rt_histogram[idx], !=, 0);
+	rt->rt_histogram[idx]--;
+}
+
+/*
+ * NOTE: caller is responsible for all locking.
+ */
+static int
+range_tree_seg_compare(const void *x1, const void *x2)
+{
+	const range_seg_t *r1 = x1;
+	const range_seg_t *r2 = x2;
+
+	if (r1->rs_start < r2->rs_start) {
+		if (r1->rs_end > r2->rs_start)
+			return (0);
+		return (-1);
+	}
+	if (r1->rs_start > r2->rs_start) {
+		if (r1->rs_start < r2->rs_end)
+			return (0);
+		return (1);
+	}
+	return (0);
+}
+
+range_tree_t *
+range_tree_create(range_tree_ops_t *ops, void *arg, kmutex_t *lp)
+{
+	range_tree_t *rt;
+
+	rt = kmem_zalloc(sizeof (range_tree_t), KM_PUSHPAGE);
+
+	avl_create(&rt->rt_root, range_tree_seg_compare,
+	    sizeof (range_seg_t), offsetof(range_seg_t, rs_node));
+
+	rt->rt_lock = lp;
+	rt->rt_ops = ops;
+	rt->rt_arg = arg;
+
+	if (rt->rt_ops != NULL)
+		rt->rt_ops->rtop_create(rt, rt->rt_arg);
+
+	return (rt);
+}
+
+void
+range_tree_destroy(range_tree_t *rt)
+{
+	VERIFY0(rt->rt_space);
+
+	if (rt->rt_ops != NULL)
+		rt->rt_ops->rtop_destroy(rt, rt->rt_arg);
+
+	avl_destroy(&rt->rt_root);
+	kmem_free(rt, sizeof (*rt));
+}
+
+void
+range_tree_add(void *arg, uint64_t start, uint64_t size)
+{
+	range_tree_t *rt = arg;
+	avl_index_t where;
+	range_seg_t rsearch, *rs_before, *rs_after, *rs;
+	uint64_t end = start + size;
+	boolean_t merge_before, merge_after;
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+	VERIFY(size != 0);
+
+	rsearch.rs_start = start;
+	rsearch.rs_end = end;
+	rs = avl_find(&rt->rt_root, &rsearch, &where);
+
+	if (rs != NULL && rs->rs_start <= start && rs->rs_end >= end) {
+		zfs_panic_recover("zfs: allocating allocated segment"
+		    "(offset=%llu size=%llu)\n",
+		    (longlong_t)start, (longlong_t)size);
+		return;
+	}
+
+	/* Make sure we don't overlap with either of our neighbors */
+	VERIFY(rs == NULL);
+
+	rs_before = avl_nearest(&rt->rt_root, where, AVL_BEFORE);
+	rs_after = avl_nearest(&rt->rt_root, where, AVL_AFTER);
+
+	merge_before = (rs_before != NULL && rs_before->rs_end == start);
+	merge_after = (rs_after != NULL && rs_after->rs_start == end);
+
+	if (merge_before && merge_after) {
+		avl_remove(&rt->rt_root, rs_before);
+		if (rt->rt_ops != NULL) {
+			rt->rt_ops->rtop_remove(rt, rs_before, rt->rt_arg);
+			rt->rt_ops->rtop_remove(rt, rs_after, rt->rt_arg);
+		}
+
+		range_tree_stat_decr(rt, rs_before);
+		range_tree_stat_decr(rt, rs_after);
+
+		rs_after->rs_start = rs_before->rs_start;
+		kmem_cache_free(range_seg_cache, rs_before);
+		rs = rs_after;
+	} else if (merge_before) {
+		if (rt->rt_ops != NULL)
+			rt->rt_ops->rtop_remove(rt, rs_before, rt->rt_arg);
+
+		range_tree_stat_decr(rt, rs_before);
+
+		rs_before->rs_end = end;
+		rs = rs_before;
+	} else if (merge_after) {
+		if (rt->rt_ops != NULL)
+			rt->rt_ops->rtop_remove(rt, rs_after, rt->rt_arg);
+
+		range_tree_stat_decr(rt, rs_after);
+
+		rs_after->rs_start = start;
+		rs = rs_after;
+	} else {
+		rs = kmem_cache_alloc(range_seg_cache, KM_PUSHPAGE);
+		rs->rs_start = start;
+		rs->rs_end = end;
+		avl_insert(&rt->rt_root, rs, where);
+	}
+
+	if (rt->rt_ops != NULL)
+		rt->rt_ops->rtop_add(rt, rs, rt->rt_arg);
+
+	range_tree_stat_incr(rt, rs);
+	rt->rt_space += size;
+}
+
+void
+range_tree_remove(void *arg, uint64_t start, uint64_t size)
+{
+	range_tree_t *rt = arg;
+	avl_index_t where;
+	range_seg_t rsearch, *rs, *newseg;
+	uint64_t end = start + size;
+	boolean_t left_over, right_over;
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+	VERIFY3U(size, !=, 0);
+	VERIFY3U(size, <=, rt->rt_space);
+
+	rsearch.rs_start = start;
+	rsearch.rs_end = end;
+	rs = avl_find(&rt->rt_root, &rsearch, &where);
+
+	/* Make sure we completely overlap with someone */
+	if (rs == NULL) {
+		zfs_panic_recover("zfs: freeing free segment "
+		    "(offset=%llu size=%llu)",
+		    (longlong_t)start, (longlong_t)size);
+		return;
+	}
+	VERIFY3U(rs->rs_start, <=, start);
+	VERIFY3U(rs->rs_end, >=, end);
+
+	left_over = (rs->rs_start != start);
+	right_over = (rs->rs_end != end);
+
+	range_tree_stat_decr(rt, rs);
+
+	if (rt->rt_ops != NULL)
+		rt->rt_ops->rtop_remove(rt, rs, rt->rt_arg);
+
+	if (left_over && right_over) {
+		newseg = kmem_cache_alloc(range_seg_cache, KM_PUSHPAGE);
+		newseg->rs_start = end;
+		newseg->rs_end = rs->rs_end;
+		range_tree_stat_incr(rt, newseg);
+
+		rs->rs_end = start;
+
+		avl_insert_here(&rt->rt_root, newseg, rs, AVL_AFTER);
+		if (rt->rt_ops != NULL)
+			rt->rt_ops->rtop_add(rt, newseg, rt->rt_arg);
+	} else if (left_over) {
+		rs->rs_end = start;
+	} else if (right_over) {
+		rs->rs_start = end;
+	} else {
+		avl_remove(&rt->rt_root, rs);
+		kmem_cache_free(range_seg_cache, rs);
+		rs = NULL;
+	}
+
+	if (rs != NULL) {
+		range_tree_stat_incr(rt, rs);
+
+		if (rt->rt_ops != NULL)
+			rt->rt_ops->rtop_add(rt, rs, rt->rt_arg);
+	}
+
+	rt->rt_space -= size;
+}
+
+static range_seg_t *
+range_tree_find(range_tree_t *rt, uint64_t start, uint64_t size,
+    avl_index_t *wherep)
+{
+	range_seg_t rsearch, *rs;
+	uint64_t end = start + size;
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+	VERIFY(size != 0);
+
+	rsearch.rs_start = start;
+	rsearch.rs_end = end;
+	rs = avl_find(&rt->rt_root, &rsearch, wherep);
+
+	if (rs != NULL && rs->rs_start <= start && rs->rs_end >= end)
+		return (rs);
+	return (NULL);
+}
+
+void
+range_tree_verify(range_tree_t *rt, uint64_t off, uint64_t size)
+{
+	range_seg_t *rs;
+	avl_index_t where;
+
+	mutex_enter(rt->rt_lock);
+	rs = range_tree_find(rt, off, size, &where);
+	if (rs != NULL)
+		panic("freeing free block; rs=%p", (void *)rs);
+	mutex_exit(rt->rt_lock);
+}
+
+boolean_t
+range_tree_contains(range_tree_t *rt, uint64_t start, uint64_t size)
+{
+	avl_index_t where;
+
+	return (range_tree_find(rt, start, size, &where) != NULL);
+}
+
+void
+range_tree_swap(range_tree_t **rtsrc, range_tree_t **rtdst)
+{
+	range_tree_t *rt;
+
+	ASSERT(MUTEX_HELD((*rtsrc)->rt_lock));
+	ASSERT0(range_tree_space(*rtdst));
+	ASSERT0(avl_numnodes(&(*rtdst)->rt_root));
+
+	rt = *rtsrc;
+	*rtsrc = *rtdst;
+	*rtdst = rt;
+}
+
+void
+range_tree_vacate(range_tree_t *rt, range_tree_func_t *func, void *arg)
+{
+	range_seg_t *rs;
+	void *cookie = NULL;
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+
+	if (rt->rt_ops != NULL)
+		rt->rt_ops->rtop_vacate(rt, rt->rt_arg);
+
+	while ((rs = avl_destroy_nodes(&rt->rt_root, &cookie)) != NULL) {
+		if (func != NULL)
+			func(arg, rs->rs_start, rs->rs_end - rs->rs_start);
+		kmem_cache_free(range_seg_cache, rs);
+	}
+
+	bzero(rt->rt_histogram, sizeof (rt->rt_histogram));
+	rt->rt_space = 0;
+}
+
+void
+range_tree_walk(range_tree_t *rt, range_tree_func_t *func, void *arg)
+{
+	range_seg_t *rs;
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+
+	for (rs = avl_first(&rt->rt_root); rs; rs = AVL_NEXT(&rt->rt_root, rs))
+		func(arg, rs->rs_start, rs->rs_end - rs->rs_start);
+}
+
+uint64_t
+range_tree_space(range_tree_t *rt)
+{
+	return (rt->rt_space);
+}

module/zfs/spa.c

@@ -1259,6 +1259,15 @@ spa_unload(spa_t *spa)
 
 	bpobj_close(&spa->spa_deferred_bpobj);
 
+	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
+
+	/*
+	 * Close all vdevs.
+	 */
+	if (spa->spa_root_vdev)
+		vdev_free(spa->spa_root_vdev);
+	ASSERT(spa->spa_root_vdev == NULL);
+
 	/*
 	 * Close the dsl pool.
 	 */
@@ -1270,20 +1279,12 @@ spa_unload(spa_t *spa)
 
 	ddt_unload(spa);
 
-	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 
 	/*
 	 * Drop and purge level 2 cache
 	 */
 	spa_l2cache_drop(spa);
 
-	/*
-	 * Close all vdevs.
-	 */
-	if (spa->spa_root_vdev)
-		vdev_free(spa->spa_root_vdev);
-	ASSERT(spa->spa_root_vdev == NULL);
-
 	for (i = 0; i < spa->spa_spares.sav_count; i++)
 		vdev_free(spa->spa_spares.sav_vdevs[i]);
 	if (spa->spa_spares.sav_vdevs) {
@@ -4568,7 +4569,9 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
 	vdev_dirty(tvd, VDD_DTL, newvd, txg);
 
 	/*
-	 * Restart the resilver
+	 * Schedule the resilver to restart in the future. We do this to
+	 * ensure that dmu_sync-ed blocks have been stitched into the
+	 * respective datasets.
 	 */
 	dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
 
@@ -5193,7 +5196,7 @@ spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
 	ASSERT0(vd->vdev_stat.vs_alloc);
 	txg = spa_vdev_config_enter(spa);
 	vd->vdev_removing = B_TRUE;
-	vdev_dirty(vd, 0, NULL, txg);
+	vdev_dirty_leaves(vd, VDD_DTL, txg);
 	vdev_config_dirty(vd);
 	spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
 
@@ -5965,7 +5968,7 @@ spa_sync_props(void *arg, dmu_tx_t *tx)
 			ASSERT(zpool_prop_feature(nvpair_name(elem)));
 
 			fname = strchr(nvpair_name(elem), '@') + 1;
-			VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
+			VERIFY0(zfeature_lookup_name(fname, &feature));
 
 			spa_feature_enable(spa, feature, tx);
 			spa_history_log_internal(spa, "set", tx,
@@ -5973,7 +5976,7 @@ spa_sync_props(void *arg, dmu_tx_t *tx)
 			break;
 
 		case ZPOOL_PROP_VERSION:
-			VERIFY(nvpair_value_uint64(elem, &intval) == 0);
+			intval = fnvpair_value_uint64(elem);
 			/*
 			 * The version is synced seperatly before other
 			 * properties and should be correct by now.
@@ -5997,7 +6000,7 @@ spa_sync_props(void *arg, dmu_tx_t *tx)
 			 */
 			break;
 		case ZPOOL_PROP_COMMENT:
-			VERIFY(nvpair_value_string(elem, &strval) == 0);
+			strval = fnvpair_value_string(elem);
 			if (spa->spa_comment != NULL)
 				spa_strfree(spa->spa_comment);
 			spa->spa_comment = spa_strdup(strval);
@@ -6029,23 +6032,23 @@ spa_sync_props(void *arg, dmu_tx_t *tx)
 
 			if (nvpair_type(elem) == DATA_TYPE_STRING) {
 				ASSERT(proptype == PROP_TYPE_STRING);
-				VERIFY(nvpair_value_string(elem, &strval) == 0);
-				VERIFY(zap_update(mos,
+				strval = fnvpair_value_string(elem);
+				VERIFY0(zap_update(mos,
 				    spa->spa_pool_props_object, propname,
-				    1, strlen(strval) + 1, strval, tx) == 0);
+				    1, strlen(strval) + 1, strval, tx));
 				spa_history_log_internal(spa, "set", tx,
 				    "%s=%s", nvpair_name(elem), strval);
 			} else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
-				VERIFY(nvpair_value_uint64(elem, &intval) == 0);
+				intval = fnvpair_value_uint64(elem);
 
 				if (proptype == PROP_TYPE_INDEX) {
 					const char *unused;
-					VERIFY(zpool_prop_index_to_string(
-					    prop, intval, &unused) == 0);
+					VERIFY0(zpool_prop_index_to_string(
+					    prop, intval, &unused));
 				}
-				VERIFY(zap_update(mos,
+				VERIFY0(zap_update(mos,
 				    spa->spa_pool_props_object, propname,
-				    8, 1, &intval, tx) == 0);
+				    8, 1, &intval, tx));
 				spa_history_log_internal(spa, "set", tx,
 				    "%s=%lld", nvpair_name(elem), intval);
 			} else {

module/zfs/spa_misc.c

@@ -986,7 +986,7 @@ spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
 		txg_wait_synced(spa->spa_dsl_pool, txg);
 
 	if (vd != NULL) {
-		ASSERT(!vd->vdev_detached || vd->vdev_dtl_smo.smo_object == 0);
+		ASSERT(!vd->vdev_detached || vd->vdev_dtl_sm == NULL);
 		spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
 		vdev_free(vd);
 		spa_config_exit(spa, SCL_ALL, spa);
@@ -1655,7 +1655,7 @@ spa_init(int mode)
 	fm_init();
 	refcount_init();
 	unique_init();
-	space_map_init();
+	range_tree_init();
 	ddt_init();
 	zio_init();
 	dmu_init();
@@ -1682,7 +1682,7 @@ spa_fini(void)
 	dmu_fini();
 	zio_fini();
 	ddt_fini();
-	space_map_fini();
+	range_tree_fini();
 	unique_fini();
 	refcount_fini();
 	fm_fini();

module/zfs/space_reftree.c

@@ -0,0 +1,159 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * Copyright (c) 2013 by Delphix. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/range_tree.h>
+#include <sys/space_reftree.h>
+
+/*
+ * Space reference trees.
+ *
+ * A range tree is a collection of integers.  Every integer is either
+ * in the tree, or it's not.  A space reference tree generalizes
+ * the idea: it allows its members to have arbitrary reference counts,
+ * as opposed to the implicit reference count of 0 or 1 in a range tree.
+ * This representation comes in handy when computing the union or
+ * intersection of multiple space maps.  For example, the union of
+ * N range trees is the subset of the reference tree with refcnt >= 1.
+ * The intersection of N range trees is the subset with refcnt >= N.
+ *
+ * [It's very much like a Fourier transform.  Unions and intersections
+ * are hard to perform in the 'range tree domain', so we convert the trees
+ * into the 'reference count domain', where it's trivial, then invert.]
+ *
+ * vdev_dtl_reassess() uses computations of this form to determine
+ * DTL_MISSING and DTL_OUTAGE for interior vdevs -- e.g. a RAID-Z vdev
+ * has an outage wherever refcnt >= vdev_nparity + 1, and a mirror vdev
+ * has an outage wherever refcnt >= vdev_children.
+ */
+static int
+space_reftree_compare(const void *x1, const void *x2)
+{
+	const space_ref_t *sr1 = x1;
+	const space_ref_t *sr2 = x2;
+
+	if (sr1->sr_offset < sr2->sr_offset)
+		return (-1);
+	if (sr1->sr_offset > sr2->sr_offset)
+		return (1);
+
+	if (sr1 < sr2)
+		return (-1);
+	if (sr1 > sr2)
+		return (1);
+
+	return (0);
+}
+
+void
+space_reftree_create(avl_tree_t *t)
+{
+	avl_create(t, space_reftree_compare,
+	    sizeof (space_ref_t), offsetof(space_ref_t, sr_node));
+}
+
+void
+space_reftree_destroy(avl_tree_t *t)
+{
+	space_ref_t *sr;
+	void *cookie = NULL;
+
+	while ((sr = avl_destroy_nodes(t, &cookie)) != NULL)
+		kmem_free(sr, sizeof (*sr));
+
+	avl_destroy(t);
+}
+
+static void
+space_reftree_add_node(avl_tree_t *t, uint64_t offset, int64_t refcnt)
+{
+	space_ref_t *sr;
+
+	sr = kmem_alloc(sizeof (*sr), KM_PUSHPAGE);
+	sr->sr_offset = offset;
+	sr->sr_refcnt = refcnt;
+
+	avl_add(t, sr);
+}
+
+void
+space_reftree_add_seg(avl_tree_t *t, uint64_t start, uint64_t end,
+	int64_t refcnt)
+{
+	space_reftree_add_node(t, start, refcnt);
+	space_reftree_add_node(t, end, -refcnt);
+}
+
+/*
+ * Convert (or add) a range tree into a reference tree.
+ */
+void
+space_reftree_add_map(avl_tree_t *t, range_tree_t *rt, int64_t refcnt)
+{
+	range_seg_t *rs;
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+
+	for (rs = avl_first(&rt->rt_root); rs; rs = AVL_NEXT(&rt->rt_root, rs))
+		space_reftree_add_seg(t, rs->rs_start, rs->rs_end, refcnt);
+}
+
+/*
+ * Convert a reference tree into a range tree.  The range tree will contain
+ * all members of the reference tree for which refcnt >= minref.
+ */
+void
+space_reftree_generate_map(avl_tree_t *t, range_tree_t *rt, int64_t minref)
+{
+	uint64_t start = -1ULL;
+	int64_t refcnt = 0;
+	space_ref_t *sr;
+
+	ASSERT(MUTEX_HELD(rt->rt_lock));
+
+	range_tree_vacate(rt, NULL, NULL);
+
+	for (sr = avl_first(t); sr != NULL; sr = AVL_NEXT(t, sr)) {
+		refcnt += sr->sr_refcnt;
+		if (refcnt >= minref) {
+			if (start == -1ULL) {
+				start = sr->sr_offset;
+			}
+		} else {
+			if (start != -1ULL) {
+				uint64_t end = sr->sr_offset;
+				ASSERT(start <= end);
+				if (end > start)
+					range_tree_add(rt, start, end - start);
+				start = -1ULL;
+			}
+		}
+	}
+	ASSERT(refcnt == 0);
+	ASSERT(start == -1ULL);
+}

module/zfs/vdev.c

@@ -36,6 +36,7 @@
 #include <sys/metaslab.h>
 #include <sys/metaslab_impl.h>
 #include <sys/space_map.h>
+#include <sys/space_reftree.h>
 #include <sys/zio.h>
 #include <sys/zap.h>
 #include <sys/fs/zfs.h>
@@ -193,7 +194,7 @@ vdev_add_child(vdev_t *pvd, vdev_t *cvd)
 	pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
 	newsize = pvd->vdev_children * sizeof (vdev_t *);
 
-	newchild = kmem_zalloc(newsize, KM_PUSHPAGE);
+	newchild = kmem_alloc(newsize, KM_PUSHPAGE);
 	if (pvd->vdev_child != NULL) {
 		bcopy(pvd->vdev_child, newchild, oldsize);
 		kmem_free(pvd->vdev_child, oldsize);
@@ -263,7 +264,7 @@ vdev_compact_children(vdev_t *pvd)
 		if (pvd->vdev_child[c])
 			newc++;
 
-	newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_PUSHPAGE);
+	newchild = kmem_zalloc(newc * sizeof (vdev_t *), KM_PUSHPAGE);
 
 	for (c = newc = 0; c < oldc; c++) {
 		if ((cvd = pvd->vdev_child[c]) != NULL) {
@@ -324,7 +325,7 @@ vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
 	mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
 	for (t = 0; t < DTL_TYPES; t++) {
-		space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
+		vd->vdev_dtl[t] = range_tree_create(NULL, NULL,
 		    &vd->vdev_dtl_lock);
 	}
 	txg_list_create(&vd->vdev_ms_list,
@@ -510,7 +511,7 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 	    alloctype == VDEV_ALLOC_ROOTPOOL)) {
 		if (alloctype == VDEV_ALLOC_LOAD) {
 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
-			    &vd->vdev_dtl_smo.smo_object);
+			    &vd->vdev_dtl_object);
 			(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
 			    &vd->vdev_unspare);
 		}
@@ -633,9 +634,10 @@ vdev_free(vdev_t *vd)
 	txg_list_destroy(&vd->vdev_dtl_list);
 
 	mutex_enter(&vd->vdev_dtl_lock);
+	space_map_close(vd->vdev_dtl_sm);
 	for (t = 0; t < DTL_TYPES; t++) {
-		space_map_unload(&vd->vdev_dtl[t]);
-		space_map_destroy(&vd->vdev_dtl[t]);
+		range_tree_vacate(vd->vdev_dtl[t], NULL, NULL);
+		range_tree_destroy(vd->vdev_dtl[t]);
 	}
 	mutex_exit(&vd->vdev_dtl_lock);
 
@@ -859,27 +861,16 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 	vd->vdev_ms_count = newc;
 
 	for (m = oldc; m < newc; m++) {
-		space_map_obj_t smo = { 0, 0, 0 };
+		uint64_t object = 0;
+
 		if (txg == 0) {
-			uint64_t object = 0;
 			error = dmu_read(mos, vd->vdev_ms_array,
 			    m * sizeof (uint64_t), sizeof (uint64_t), &object,
 			    DMU_READ_PREFETCH);
 			if (error)
 				return (error);
-			if (object != 0) {
-				dmu_buf_t *db;
-				error = dmu_bonus_hold(mos, object, FTAG, &db);
-				if (error)
-					return (error);
-				ASSERT3U(db->db_size, >=, sizeof (smo));
-				bcopy(db->db_data, &smo, sizeof (smo));
-				ASSERT3U(smo.smo_object, ==, object);
-				dmu_buf_rele(db, FTAG);
-			}
 		}
-		vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
-		    m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
+		vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg);
 	}
 
 	if (txg == 0)
@@ -907,9 +898,12 @@ vdev_metaslab_fini(vdev_t *vd)
 
 	if (vd->vdev_ms != NULL) {
 		metaslab_group_passivate(vd->vdev_mg);
-		for (m = 0; m < count; m++)
-			if (vd->vdev_ms[m] != NULL)
-				metaslab_fini(vd->vdev_ms[m]);
+		for (m = 0; m < count; m++) {
+			metaslab_t *msp = vd->vdev_ms[m];
+
+			if (msp != NULL)
+				metaslab_fini(msp);
+		}
 		kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
 		vd->vdev_ms = NULL;
 	}
@@ -1572,9 +1566,10 @@ vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
 	}
 
 	/*
-	 * Recursively initialize all labels.
+	 * Recursively load DTLs and initialize all labels.
 	 */
-	if ((error = vdev_label_init(vd, txg, isreplacing ?
+	if ((error = vdev_dtl_load(vd)) != 0 ||
+	    (error = vdev_label_init(vd, txg, isreplacing ?
 	    VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
 		vdev_close(vd);
 		return (error);
@@ -1610,6 +1605,18 @@ vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
 	(void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
 }
 
+void
+vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg)
+{
+	int c;
+
+	for (c = 0; c < vd->vdev_children; c++)
+		vdev_dirty_leaves(vd->vdev_child[c], flags, txg);
+
+	if (vd->vdev_ops->vdev_op_leaf)
+		vdev_dirty(vd->vdev_top, flags, vd, txg);
+}
+
 /*
  * DTLs.
  *
@@ -1651,31 +1658,31 @@ vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
 void
 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
 {
-	space_map_t *sm = &vd->vdev_dtl[t];
+	range_tree_t *rt = vd->vdev_dtl[t];
 
 	ASSERT(t < DTL_TYPES);
 	ASSERT(vd != vd->vdev_spa->spa_root_vdev);
 	ASSERT(spa_writeable(vd->vdev_spa));
 
-	mutex_enter(sm->sm_lock);
-	if (!space_map_contains(sm, txg, size))
-		space_map_add(sm, txg, size);
-	mutex_exit(sm->sm_lock);
+	mutex_enter(rt->rt_lock);
+	if (!range_tree_contains(rt, txg, size))
+		range_tree_add(rt, txg, size);
+	mutex_exit(rt->rt_lock);
 }
 
 boolean_t
 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
 {
-	space_map_t *sm = &vd->vdev_dtl[t];
+	range_tree_t *rt = vd->vdev_dtl[t];
 	boolean_t dirty = B_FALSE;
 
 	ASSERT(t < DTL_TYPES);
 	ASSERT(vd != vd->vdev_spa->spa_root_vdev);
 
-	mutex_enter(sm->sm_lock);
-	if (sm->sm_space != 0)
-		dirty = space_map_contains(sm, txg, size);
-	mutex_exit(sm->sm_lock);
+	mutex_enter(rt->rt_lock);
+	if (range_tree_space(rt) != 0)
+		dirty = range_tree_contains(rt, txg, size);
+	mutex_exit(rt->rt_lock);
 
 	return (dirty);
 }
@@ -1683,12 +1690,12 @@ vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
 boolean_t
 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
 {
-	space_map_t *sm = &vd->vdev_dtl[t];
+	range_tree_t *rt = vd->vdev_dtl[t];
 	boolean_t empty;
 
-	mutex_enter(sm->sm_lock);
-	empty = (sm->sm_space == 0);
-	mutex_exit(sm->sm_lock);
+	mutex_enter(rt->rt_lock);
+	empty = (range_tree_space(rt) == 0);
+	mutex_exit(rt->rt_lock);
 
 	return (empty);
 }
@@ -1699,14 +1706,14 @@ vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
 static uint64_t
 vdev_dtl_min(vdev_t *vd)
 {
-	space_seg_t *ss;
+	range_seg_t *rs;
 
 	ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
-	ASSERT3U(vd->vdev_dtl[DTL_MISSING].sm_space, !=, 0);
+	ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
 	ASSERT0(vd->vdev_children);
 
-	ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
-	return (ss->ss_start - 1);
+	rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root);
+	return (rs->rs_start - 1);
 }
 
 /*
@@ -1715,14 +1722,14 @@ vdev_dtl_min(vdev_t *vd)
 static uint64_t
 vdev_dtl_max(vdev_t *vd)
 {
-	space_seg_t *ss;
+	range_seg_t *rs;
 
 	ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock));
-	ASSERT3U(vd->vdev_dtl[DTL_MISSING].sm_space, !=, 0);
+	ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0);
 	ASSERT0(vd->vdev_children);
 
-	ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
-	return (ss->ss_end);
+	rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root);
+	return (rs->rs_end);
 }
 
 /*
@@ -1743,7 +1750,7 @@ vdev_dtl_should_excise(vdev_t *vd)
 	ASSERT0(vd->vdev_children);
 
 	if (vd->vdev_resilver_txg == 0 ||
-	    vd->vdev_dtl[DTL_MISSING].sm_space == 0)
+	    range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0)
 		return (B_TRUE);
 
 	/*
@@ -1813,35 +1820,35 @@ vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
 			 * positive refcnt -- either 1 or 2.  We then convert
 			 * the reference tree into the new DTL_MISSING map.
 			 */
-			space_map_ref_create(&reftree);
-			space_map_ref_add_map(&reftree,
-			    &vd->vdev_dtl[DTL_MISSING], 1);
-			space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
-			space_map_ref_add_map(&reftree,
-			    &vd->vdev_dtl[DTL_SCRUB], 2);
-			space_map_ref_generate_map(&reftree,
-			    &vd->vdev_dtl[DTL_MISSING], 1);
-			space_map_ref_destroy(&reftree);
+			space_reftree_create(&reftree);
+			space_reftree_add_map(&reftree,
+			    vd->vdev_dtl[DTL_MISSING], 1);
+			space_reftree_add_seg(&reftree, 0, scrub_txg, -1);
+			space_reftree_add_map(&reftree,
+			    vd->vdev_dtl[DTL_SCRUB], 2);
+			space_reftree_generate_map(&reftree,
+			    vd->vdev_dtl[DTL_MISSING], 1);
+			space_reftree_destroy(&reftree);
 		}
-		space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
-		space_map_walk(&vd->vdev_dtl[DTL_MISSING],
-		    space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
+		range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
+		range_tree_walk(vd->vdev_dtl[DTL_MISSING],
+		    range_tree_add, vd->vdev_dtl[DTL_PARTIAL]);
 		if (scrub_done)
-			space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
-		space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
+			range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
+		range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
 		if (!vdev_readable(vd))
-			space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
+			range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
 		else
-			space_map_walk(&vd->vdev_dtl[DTL_MISSING],
-			    space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
+			range_tree_walk(vd->vdev_dtl[DTL_MISSING],
+			    range_tree_add, vd->vdev_dtl[DTL_OUTAGE]);
 
 		/*
 		 * If the vdev was resilvering and no longer has any
 		 * DTLs then reset its resilvering flag.
 		 */
 		if (vd->vdev_resilver_txg != 0 &&
-		    vd->vdev_dtl[DTL_MISSING].sm_space == 0 &&
-		    vd->vdev_dtl[DTL_OUTAGE].sm_space == 0)
+		    range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 &&
+		    range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0)
 			vd->vdev_resilver_txg = 0;
 
 		mutex_exit(&vd->vdev_dtl_lock);
@@ -1853,6 +1860,8 @@ vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
 
 	mutex_enter(&vd->vdev_dtl_lock);
 	for (t = 0; t < DTL_TYPES; t++) {
+		int c;
+
 		/* account for child's outage in parent's missing map */
 		int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
 		if (t == DTL_SCRUB)
@@ -1863,46 +1872,56 @@ vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
 			minref = vd->vdev_nparity + 1;	/* RAID-Z */
 		else
 			minref = vd->vdev_children;	/* any kind of mirror */
-		space_map_ref_create(&reftree);
+		space_reftree_create(&reftree);
 		for (c = 0; c < vd->vdev_children; c++) {
 			vdev_t *cvd = vd->vdev_child[c];
 			mutex_enter(&cvd->vdev_dtl_lock);
-			space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
+			space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1);
 			mutex_exit(&cvd->vdev_dtl_lock);
 		}
-		space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
-		space_map_ref_destroy(&reftree);
+		space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref);
+		space_reftree_destroy(&reftree);
 	}
 	mutex_exit(&vd->vdev_dtl_lock);
 }
 
-static int
+int
 vdev_dtl_load(vdev_t *vd)
 {
 	spa_t *spa = vd->vdev_spa;
-	space_map_obj_t *smo = &vd->vdev_dtl_smo;
 	objset_t *mos = spa->spa_meta_objset;
-	dmu_buf_t *db;
-	int error;
+	int error = 0;
+	int c;
 
-	ASSERT(vd->vdev_children == 0);
+	if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) {
+		ASSERT(!vd->vdev_ishole);
 
-	if (smo->smo_object == 0)
-		return (0);
+		error = space_map_open(&vd->vdev_dtl_sm, mos,
+		    vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock);
+		if (error)
+			return (error);
+		ASSERT(vd->vdev_dtl_sm != NULL);
 
-	ASSERT(!vd->vdev_ishole);
+		mutex_enter(&vd->vdev_dtl_lock);
+
+		/*
+		 * Now that we've opened the space_map we need to update
+		 * the in-core DTL.
+		 */
+		space_map_update(vd->vdev_dtl_sm);
+
+		error = space_map_load(vd->vdev_dtl_sm,
+		    vd->vdev_dtl[DTL_MISSING], SM_ALLOC);
+		mutex_exit(&vd->vdev_dtl_lock);
 
-	if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
 		return (error);
+	}
 
-	ASSERT3U(db->db_size, >=, sizeof (*smo));
-	bcopy(db->db_data, smo, sizeof (*smo));
-	dmu_buf_rele(db, FTAG);
-
-	mutex_enter(&vd->vdev_dtl_lock);
-	error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
-	    NULL, SM_ALLOC, smo, mos);
-	mutex_exit(&vd->vdev_dtl_lock);
+	for (c = 0; c < vd->vdev_children; c++) {
+		error = vdev_dtl_load(vd->vdev_child[c]);
+		if (error != 0)
+			break;
+	}
 
 	return (error);
 }
@@ -1911,64 +1930,74 @@ void
 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
 {
 	spa_t *spa = vd->vdev_spa;
-	space_map_obj_t *smo = &vd->vdev_dtl_smo;
-	space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
+	range_tree_t *rt = vd->vdev_dtl[DTL_MISSING];
 	objset_t *mos = spa->spa_meta_objset;
-	space_map_t smsync;
-	kmutex_t smlock;
-	dmu_buf_t *db;
+	range_tree_t *rtsync;
+	kmutex_t rtlock;
 	dmu_tx_t *tx;
+	uint64_t object = space_map_object(vd->vdev_dtl_sm);
 
 	ASSERT(!vd->vdev_ishole);
+	ASSERT(vd->vdev_ops->vdev_op_leaf);
 
 	tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
 
-	if (vd->vdev_detached) {
-		if (smo->smo_object != 0) {
-			VERIFY0(dmu_object_free(mos, smo->smo_object, tx));
-			smo->smo_object = 0;
-		}
+	if (vd->vdev_detached || vd->vdev_top->vdev_removing) {
+		mutex_enter(&vd->vdev_dtl_lock);
+		space_map_free(vd->vdev_dtl_sm, tx);
+		space_map_close(vd->vdev_dtl_sm);
+		vd->vdev_dtl_sm = NULL;
+		mutex_exit(&vd->vdev_dtl_lock);
 		dmu_tx_commit(tx);
 		return;
 	}
 
-	if (smo->smo_object == 0) {
-		ASSERT(smo->smo_objsize == 0);
-		ASSERT(smo->smo_alloc == 0);
-		smo->smo_object = dmu_object_alloc(mos,
-		    DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
-		    DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
-		ASSERT(smo->smo_object != 0);
+	if (vd->vdev_dtl_sm == NULL) {
+		uint64_t new_object;
+
+		new_object = space_map_alloc(mos, tx);
+		VERIFY3U(new_object, !=, 0);
+
+		VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object,
+		    0, -1ULL, 0, &vd->vdev_dtl_lock));
+		ASSERT(vd->vdev_dtl_sm != NULL);
+	}
+
+	mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL);
+
+	rtsync = range_tree_create(NULL, NULL, &rtlock);
+
+	mutex_enter(&rtlock);
+
+	mutex_enter(&vd->vdev_dtl_lock);
+	range_tree_walk(rt, range_tree_add, rtsync);
+	mutex_exit(&vd->vdev_dtl_lock);
+
+	space_map_truncate(vd->vdev_dtl_sm, tx);
+	space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx);
+	range_tree_vacate(rtsync, NULL, NULL);
+
+	range_tree_destroy(rtsync);
+
+	mutex_exit(&rtlock);
+	mutex_destroy(&rtlock);
+
+	/*
+	 * If the object for the space map has changed then dirty
+	 * the top level so that we update the config.
+	 */
+	if (object != space_map_object(vd->vdev_dtl_sm)) {
+		zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
+		    "new object %llu", txg, spa_name(spa), object,
+		    space_map_object(vd->vdev_dtl_sm));
 		vdev_config_dirty(vd->vdev_top);
 	}
 
-	mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
-
-	space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
-	    &smlock);
-
-	mutex_enter(&smlock);
+	dmu_tx_commit(tx);
 
 	mutex_enter(&vd->vdev_dtl_lock);
-	space_map_walk(sm, space_map_add, &smsync);
+	space_map_update(vd->vdev_dtl_sm);
 	mutex_exit(&vd->vdev_dtl_lock);
-
-	space_map_truncate(smo, mos, tx);
-	space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
-	space_map_vacate(&smsync, NULL, NULL);
-
-	space_map_destroy(&smsync);
-
-	mutex_exit(&smlock);
-	mutex_destroy(&smlock);
-
-	VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
-	dmu_buf_will_dirty(db, tx);
-	ASSERT3U(db->db_size, >=, sizeof (*smo));
-	bcopy(smo, db->db_data, sizeof (*smo));
-	dmu_buf_rele(db, FTAG);
-
-	dmu_tx_commit(tx);
 }
 
 /*
@@ -2018,7 +2047,7 @@ vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
 
 	if (vd->vdev_children == 0) {
 		mutex_enter(&vd->vdev_dtl_lock);
-		if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
+		if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 &&
 		    vdev_writeable(vd)) {
 
 			thismin = vdev_dtl_min(vd);
@@ -2126,29 +2155,25 @@ vdev_remove(vdev_t *vd, uint64_t txg)
 
 	tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
 
-	if (vd->vdev_dtl_smo.smo_object) {
-		ASSERT0(vd->vdev_dtl_smo.smo_alloc);
-		(void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
-		vd->vdev_dtl_smo.smo_object = 0;
-	}
-
 	if (vd->vdev_ms != NULL) {
 		for (m = 0; m < vd->vdev_ms_count; m++) {
 			metaslab_t *msp = vd->vdev_ms[m];
 
-			if (msp == NULL || msp->ms_smo.smo_object == 0)
+			if (msp == NULL || msp->ms_sm == NULL)
 				continue;
 
-			ASSERT0(msp->ms_smo.smo_alloc);
-			(void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
-			msp->ms_smo.smo_object = 0;
+			mutex_enter(&msp->ms_lock);
+			VERIFY0(space_map_allocated(msp->ms_sm));
+			space_map_free(msp->ms_sm, tx);
+			space_map_close(msp->ms_sm);
+			msp->ms_sm = NULL;
+			mutex_exit(&msp->ms_lock);
 		}
 	}
 
 	if (vd->vdev_ms_array) {
 		(void) dmu_object_free(mos, vd->vdev_ms_array, tx);
 		vd->vdev_ms_array = 0;
-		vd->vdev_ms_shift = 0;
 	}
 	dmu_tx_commit(tx);
 }

module/zfs/vdev_label.c

@@ -283,9 +283,10 @@ vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
 			    vd->vdev_removing);
 	}
 
-	if (vd->vdev_dtl_smo.smo_object != 0)
+	if (vd->vdev_dtl_sm != NULL) {
 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
-		    vd->vdev_dtl_smo.smo_object);
+		    space_map_object(vd->vdev_dtl_sm));
+	}
 
 	if (vd->vdev_crtxg)
 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);

module/zfs/zfeature.c

@@ -369,36 +369,46 @@ spa_feature_enable(spa_t *spa, zfeature_info_t *feature, dmu_tx_t *tx)
 	    spa->spa_feat_desc_obj, feature, FEATURE_ACTION_ENABLE, tx));
 }
 
-/*
- * If the specified feature has not yet been enabled, this function returns
- * ENOTSUP; otherwise, this function increments the feature's refcount (or
- * returns EOVERFLOW if the refcount cannot be incremented). This function must
- * be called from syncing context.
- */
 void
 spa_feature_incr(spa_t *spa, zfeature_info_t *feature, dmu_tx_t *tx)
 {
+	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT3U(spa_version(spa), >=, SPA_VERSION_FEATURES);
 	VERIFY3U(0, ==, feature_do_action(spa->spa_meta_objset,
 	    spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj,
 	    spa->spa_feat_desc_obj, feature, FEATURE_ACTION_INCR, tx));
 }
 
-/*
- * If the specified feature has not yet been enabled, this function returns
- * ENOTSUP; otherwise, this function decrements the feature's refcount (or
- * returns EOVERFLOW if the refcount is already 0). This function must
- * be called from syncing context.
- */
 void
 spa_feature_decr(spa_t *spa, zfeature_info_t *feature, dmu_tx_t *tx)
 {
+	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT3U(spa_version(spa), >=, SPA_VERSION_FEATURES);
 	VERIFY3U(0, ==, feature_do_action(spa->spa_meta_objset,
 	    spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj,
 	    spa->spa_feat_desc_obj, feature, FEATURE_ACTION_DECR, tx));
 }
 
+/*
+ * This interface is for debugging only. Normal consumers should use
+ * spa_feature_is_enabled/spa_feature_is_active.
+ */
+int
+spa_feature_get_refcount(spa_t *spa, zfeature_info_t *feature)
+{
+	int err;
+	uint64_t refcount = 0;
+
+	if (spa_version(spa) < SPA_VERSION_FEATURES)
+		return (B_FALSE);
+
+	err = feature_get_refcount(spa->spa_meta_objset,
+	    spa->spa_feat_for_read_obj, spa->spa_feat_for_write_obj,
+	    feature, &refcount);
+	ASSERT(err == 0 || err == ENOTSUP);
+	return (err == 0 ? refcount : 0);
+}
+
 boolean_t
 spa_feature_is_enabled(spa_t *spa, zfeature_info_t *feature)
 {

module/zfs/zfeature_common.c

@@ -20,7 +20,7 @@
  */
 
 /*
- * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2013 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  */
 
@@ -164,4 +164,7 @@ zpool_feature_init(void)
 	zfeature_register(SPA_FEATURE_LZ4_COMPRESS,
 	    "org.illumos:lz4_compress", "lz4_compress",
 	    "LZ4 compression algorithm support.", B_FALSE, B_FALSE, NULL);
+	zfeature_register(SPA_FEATURE_SPACEMAP_HISTOGRAM,
+	    "com.delphix:spacemap_histogram", "spacemap_histogram",
+	    "Spacemaps maintain space histograms.", B_TRUE, B_FALSE, NULL);
 }