OpenZFS 6736 - ZFS per-vdev ZAPs

6736 ZFS per-vdev ZAPs
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: John Kennedy <john.kennedy@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Don Brady <don.brady@intel.com>
Reviewed by: Dan McDonald <danmcd@omniti.com>

References:
  https://www.illumos.org/issues/6736
  https://github.com/openzfs/openzfs/commit/215198a

Ported-by: Don Brady <don.brady@intel.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #4515

configure.ac

@@ -266,6 +266,7 @@ AC_CONFIG_FILES([
 	tests/zfs-tests/tests/functional/threadsappend/Makefile
 	tests/zfs-tests/tests/functional/truncate/Makefile
 	tests/zfs-tests/tests/functional/userquota/Makefile
+	tests/zfs-tests/tests/functional/vdev_zaps/Makefile
 	tests/zfs-tests/tests/functional/write_dirs/Makefile
 	tests/zfs-tests/tests/functional/xattr/Makefile
 	tests/zfs-tests/tests/functional/zvol/Makefile

include/sys/dmu.h

@@ -315,6 +315,7 @@ typedef struct dmu_buf {
 #define	DMU_POOL_FREE_BPOBJ		"free_bpobj"
 #define	DMU_POOL_BPTREE_OBJ		"bptree_obj"
 #define	DMU_POOL_EMPTY_BPOBJ		"empty_bpobj"
+#define	DMU_POOL_VDEV_ZAP_MAP		"com.delphix:vdev_zap_map"
 
 /*
  * Allocate an object from this objset.  The range of object numbers

include/sys/fs/zfs.h

@@ -566,6 +566,9 @@ typedef struct zpool_rewind_policy {
 #define	ZPOOL_CONFIG_FEATURES_FOR_READ	"features_for_read"
 #define	ZPOOL_CONFIG_FEATURE_STATS	"feature_stats"	/* not stored on disk */
 #define	ZPOOL_CONFIG_ERRATA		"errata"	/* not stored on disk */
+#define	ZPOOL_CONFIG_VDEV_TOP_ZAP	"com.delphix:vdev_zap_top"
+#define	ZPOOL_CONFIG_VDEV_LEAF_ZAP	"com.delphix:vdev_zap_leaf"
+#define	ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS	"com.delphix:has_per_vdev_zaps"
 /*
  * The persistent vdev state is stored as separate values rather than a single
  * 'vdev_state' entry.  This is because a device can be in multiple states, such

include/sys/spa_impl.h

@@ -116,6 +116,12 @@ typedef struct spa_taskqs {
 	taskq_t **stqs_taskq;
 } spa_taskqs_t;
 
+typedef enum spa_all_vdev_zap_action {
+	AVZ_ACTION_NONE = 0,
+	AVZ_ACTION_DESTROY,	/* Destroy all per-vdev ZAPs and the AVZ. */
+	AVZ_ACTION_REBUILD	/* Populate the new AVZ, see spa_avz_rebuild */
+} spa_avz_action_t;
+
 struct spa {
 	/*
 	 * Fields protected by spa_namespace_lock.
@@ -251,6 +257,8 @@ struct spa {
 	uint64_t	spa_deadman_calls;	/* number of deadman calls */
 	hrtime_t	spa_sync_starttime;	/* starting time of spa_sync */
 	uint64_t	spa_deadman_synctime;	/* deadman expiration timer */
+	uint64_t	spa_all_vdev_zaps;	/* ZAP of per-vd ZAP obj #s */
+	spa_avz_action_t	spa_avz_action;	/* destroy/rebuild AVZ? */
 	uint64_t	spa_errata;		/* errata issues detected */
 	spa_stats_t	spa_stats;		/* assorted spa statistics */
 	hrtime_t	spa_ccw_fail_time;	/* Conf cache write fail time */

include/sys/vdev.h

@@ -70,6 +70,10 @@ extern void vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg,
 extern boolean_t vdev_dtl_required(vdev_t *vd);
 extern boolean_t vdev_resilver_needed(vdev_t *vd,
     uint64_t *minp, uint64_t *maxp);
+extern void vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj,
+    dmu_tx_t *tx);
+extern uint64_t vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx);
+extern void vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx);
 
 extern void vdev_hold(vdev_t *);
 extern void vdev_rele(vdev_t *);
@@ -133,7 +137,8 @@ extern void vdev_state_clean(vdev_t *vd);
 typedef enum vdev_config_flag {
 	VDEV_CONFIG_SPARE = 1 << 0,
 	VDEV_CONFIG_L2CACHE = 1 << 1,
-	VDEV_CONFIG_REMOVING = 1 << 2
+	VDEV_CONFIG_REMOVING = 1 << 2,
+	VDEV_CONFIG_MOS = 1 << 3
 } vdev_config_flag_t;
 
 extern void vdev_top_config_generate(spa_t *spa, nvlist_t *config);

include/sys/vdev_impl.h

@@ -177,6 +177,7 @@ struct vdev {
 	uint64_t	vdev_islog;	/* is an intent log device	*/
 	uint64_t	vdev_removing;	/* device is being removed?	*/
 	boolean_t	vdev_ishole;	/* is a hole in the namespace 	*/
+	uint64_t	vdev_top_zap;
 
 	/*
 	 * Leaf vdev state.
@@ -215,6 +216,7 @@ struct vdev {
 	spa_aux_vdev_t	*vdev_aux;	/* for l2cache and spares vdevs	*/
 	zio_t		*vdev_probe_zio; /* root of current probe	*/
 	vdev_aux_t	vdev_label_aux;	/* on-disk aux state		*/
+	uint64_t	vdev_leaf_zap;
 
 	/*
 	 * For DTrace to work in userland (libzpool) context, these fields must

module/zfs/spa.c

@@ -1674,6 +1674,21 @@ spa_check_removed(vdev_t *vd)
 	}
 }
 
+static void
+spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
+{
+	uint64_t i;
+
+	ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
+
+	vd->vdev_top_zap = mvd->vdev_top_zap;
+	vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
+
+	for (i = 0; i < vd->vdev_children; i++) {
+		spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
+	}
+}
+
 /*
  * Validate the current config against the MOS config
  */
@@ -1778,16 +1793,25 @@ spa_config_valid(spa_t *spa, nvlist_t *config)
 			spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 
 			vdev_reopen(rvd);
-		} else if (mtvd->vdev_islog) {
+		} else {
+			if (mtvd->vdev_islog) {
 				/*
-			 * Load the slog device's state from the MOS config
+				 * Load the slog device's state from the MOS
-			 * since it's possible that the label does not
+				 * config since it's possible that the label
-			 * contain the most up-to-date information.
+				 * does not contain the most up-to-date
+				 * information.
 				 */
 				vdev_load_log_state(tvd, mtvd);
 				vdev_reopen(tvd);
 			}
+
+			/*
+			 * Per-vdev ZAP info is stored exclusively in the MOS.
+			 */
+			spa_config_valid_zaps(tvd, mtvd);
 		}
+	}
+
 	vdev_free(mrvd);
 	spa_config_exit(spa, SCL_ALL, FTAG);
 
@@ -2214,6 +2238,36 @@ spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
 	return (error);
 }
 
+/*
+ * Count the number of per-vdev ZAPs associated with all of the vdevs in the
+ * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
+ * spa's per-vdev ZAP list.
+ */
+static uint64_t
+vdev_count_verify_zaps(vdev_t *vd)
+{
+	spa_t *spa = vd->vdev_spa;
+	uint64_t total = 0;
+	uint64_t i;
+
+	if (vd->vdev_top_zap != 0) {
+		total++;
+		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
+		    spa->spa_all_vdev_zaps, vd->vdev_top_zap));
+	}
+	if (vd->vdev_leaf_zap != 0) {
+		total++;
+		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
+		    spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
+	}
+
+	for (i = 0; i < vd->vdev_children; i++) {
+		total += vdev_count_verify_zaps(vd->vdev_child[i]);
+	}
+
+	return (total);
+}
+
 /*
  * Load an existing storage pool, using the pool's builtin spa_config as a
  * source of configuration information.
@@ -2234,6 +2288,7 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
 	int parse, i;
 	uint64_t obj;
 	boolean_t missing_feat_write = B_FALSE;
+	nvlist_t *mos_config;
 
 	/*
 	 * If this is an untrusted config, access the pool in read-only mode.
@@ -2632,6 +2687,38 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
 	if (error != 0 && error != ENOENT)
 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
 
+	/*
+	 * Load the per-vdev ZAP map. If we have an older pool, this will not
+	 * be present; in this case, defer its creation to a later time to
+	 * avoid dirtying the MOS this early / out of sync context. See
+	 * spa_sync_config_object.
+	 */
+
+	/* The sentinel is only available in the MOS config. */
+	if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
+		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
+
+	error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
+	    &spa->spa_all_vdev_zaps);
+
+	if (error != ENOENT && error != 0) {
+		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
+	} else if (error == 0 && !nvlist_exists(mos_config,
+	    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
+		/*
+		 * An older version of ZFS overwrote the sentinel value, so
+		 * we have orphaned per-vdev ZAPs in the MOS. Defer their
+		 * destruction to later; see spa_sync_config_object.
+		 */
+		spa->spa_avz_action = AVZ_ACTION_DESTROY;
+		/*
+		 * We're assuming that no vdevs have had their ZAPs created
+		 * before this. Better be sure of it.
+		 */
+		ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
+	}
+	nvlist_free(mos_config);
+
 	/*
 	 * If we're assembling the pool from the split-off vdevs of
 	 * an existing pool, we don't want to attach the spares & cache
@@ -5170,6 +5257,16 @@ spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
 		    vml[c]->vdev_top->vdev_asize) == 0);
 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
 		    vml[c]->vdev_top->vdev_ashift) == 0);
+
+		/* transfer per-vdev ZAPs */
+		ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
+		VERIFY0(nvlist_add_uint64(child[c],
+		    ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
+
+		ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
+		VERIFY0(nvlist_add_uint64(child[c],
+		    ZPOOL_CONFIG_VDEV_TOP_ZAP,
+		    vml[c]->vdev_parent->vdev_top_zap));
 	}
 
 	if (error != 0) {
@@ -5211,11 +5308,13 @@ spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
 	    spa->spa_config_txg) == 0);
 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 	    spa_generate_guid(NULL)) == 0);
+	VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
 	(void) nvlist_lookup_string(props,
 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
 
 	/* add the new pool to the namespace */
 	newspa = spa_add(newname, config, altroot);
+	newspa->spa_avz_action = AVZ_ACTION_REBUILD;
 	newspa->spa_config_txg = spa->spa_config_txg;
 	spa_set_log_state(newspa, SPA_LOG_CLEAR);
 
@@ -5273,9 +5372,11 @@ spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
 			if (error == 0)
 				spa_history_log_internal(spa, "detach", tx,
 				    "vdev=%s", vml[c]->vdev_path);
+
 			vdev_free(vml[c]);
 		}
 	}
+	spa->spa_avz_action = AVZ_ACTION_REBUILD;
 	vdev_config_dirty(spa->spa_root_vdev);
 	spa->spa_config_splitting = NULL;
 	nvlist_free(nvl);
@@ -6109,16 +6210,120 @@ spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
 	sav->sav_sync = B_FALSE;
 }
 
+/*
+ * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
+ * The all-vdev ZAP must be empty.
+ */
+static void
+spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
+{
+	spa_t *spa = vd->vdev_spa;
+	uint64_t i;
+
+	if (vd->vdev_top_zap != 0) {
+		VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
+		    vd->vdev_top_zap, tx));
+	}
+	if (vd->vdev_leaf_zap != 0) {
+		VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
+		    vd->vdev_leaf_zap, tx));
+	}
+	for (i = 0; i < vd->vdev_children; i++) {
+		spa_avz_build(vd->vdev_child[i], avz, tx);
+	}
+}
+
 static void
 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
 {
 	nvlist_t *config;
 
-	if (list_is_empty(&spa->spa_config_dirty_list))
+	/*
+	 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
+	 * its config may not be dirty but we still need to build per-vdev ZAPs.
+	 * Similarly, if the pool is being assembled (e.g. after a split), we
+	 * need to rebuild the AVZ although the config may not be dirty.
+	 */
+	if (list_is_empty(&spa->spa_config_dirty_list) &&
+	    spa->spa_avz_action == AVZ_ACTION_NONE)
 		return;
 
 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
 
+	ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
+	    spa->spa_all_vdev_zaps != 0);
+
+	if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
+		zap_cursor_t zc;
+		zap_attribute_t za;
+
+		/* Make and build the new AVZ */
+		uint64_t new_avz = zap_create(spa->spa_meta_objset,
+		    DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
+		spa_avz_build(spa->spa_root_vdev, new_avz, tx);
+
+		/* Diff old AVZ with new one */
+		for (zap_cursor_init(&zc, spa->spa_meta_objset,
+		    spa->spa_all_vdev_zaps);
+		    zap_cursor_retrieve(&zc, &za) == 0;
+		    zap_cursor_advance(&zc)) {
+			uint64_t vdzap = za.za_first_integer;
+			if (zap_lookup_int(spa->spa_meta_objset, new_avz,
+			    vdzap) == ENOENT) {
+				/*
+				 * ZAP is listed in old AVZ but not in new one;
+				 * destroy it
+				 */
+				VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
+				    tx));
+			}
+		}
+
+		zap_cursor_fini(&zc);
+
+		/* Destroy the old AVZ */
+		VERIFY0(zap_destroy(spa->spa_meta_objset,
+		    spa->spa_all_vdev_zaps, tx));
+
+		/* Replace the old AVZ in the dir obj with the new one */
+		VERIFY0(zap_update(spa->spa_meta_objset,
+		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
+		    sizeof (new_avz), 1, &new_avz, tx));
+
+		spa->spa_all_vdev_zaps = new_avz;
+	} else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
+		zap_cursor_t zc;
+		zap_attribute_t za;
+
+		/* Walk through the AVZ and destroy all listed ZAPs */
+		for (zap_cursor_init(&zc, spa->spa_meta_objset,
+		    spa->spa_all_vdev_zaps);
+		    zap_cursor_retrieve(&zc, &za) == 0;
+		    zap_cursor_advance(&zc)) {
+			uint64_t zap = za.za_first_integer;
+			VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
+		}
+
+		zap_cursor_fini(&zc);
+
+		/* Destroy and unlink the AVZ itself */
+		VERIFY0(zap_destroy(spa->spa_meta_objset,
+		    spa->spa_all_vdev_zaps, tx));
+		VERIFY0(zap_remove(spa->spa_meta_objset,
+		    DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
+		spa->spa_all_vdev_zaps = 0;
+	}
+
+	if (spa->spa_all_vdev_zaps == 0) {
+		spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
+		    DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
+		    DMU_POOL_VDEV_ZAP_MAP, tx);
+	}
+	spa->spa_avz_action = AVZ_ACTION_NONE;
+
+	/* Create ZAPs for vdevs that don't have them. */
+	vdev_construct_zaps(spa->spa_root_vdev, tx);
+
 	config = spa_config_generate(spa, spa->spa_root_vdev,
 	    dmu_tx_get_txg(tx), B_FALSE);
 
@@ -6509,6 +6714,21 @@ spa_sync(spa_t *spa, uint64_t txg)
 
 	} while (dmu_objset_is_dirty(mos, txg));
 
+	if (!list_is_empty(&spa->spa_config_dirty_list)) {
+		/*
+		 * Make sure that the number of ZAPs for all the vdevs matches
+		 * the number of ZAPs in the per-vdev ZAP list. This only gets
+		 * called if the config is dirty; otherwise there may be
+		 * outstanding AVZ operations that weren't completed in
+		 * spa_sync_config_object.
+		 */
+		uint64_t all_vdev_zap_entry_count;
+		ASSERT0(zap_count(spa->spa_meta_objset,
+		    spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
+		ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
+		    all_vdev_zap_entry_count);
+	}
+
 	/*
 	 * Rewrite the vdev configuration (which includes the uberblock)
 	 * to commit the transaction group.

module/zfs/spa_config.c

@@ -22,7 +22,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  */
 
 #include <sys/spa.h>
@@ -129,7 +129,7 @@ spa_config_load(void)
 		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
 			continue;
 
-		VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
+		child = fnvpair_value_nvlist(nvpair);
 
 		if (spa_lookup(nvpair_name(nvpair)) != NULL)
 			continue;
@@ -167,14 +167,9 @@ spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 	/*
 	 * Pack the configuration into a buffer.
 	 */
-	VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
+	buf = fnvlist_pack(nvl, &buflen);
-
-	buf = vmem_alloc(buflen, KM_SLEEP);
 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
 
-	VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
-	    KM_SLEEP) == 0);
-
 #if defined(__linux__) && defined(_KERNEL)
 	/*
 	 * Write the configuration to disk.  Due to the complexity involved
@@ -216,7 +211,7 @@ spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 	(void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
 #endif
 
-	vmem_free(buf, buflen);
+	fnvlist_pack_free(buf, buflen);
 	kmem_free(temp, MAXPATHLEN);
 	return (err);
 }
@@ -282,17 +277,15 @@ spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 			}
 
 			if (nvl == NULL)
-				VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
+				nvl = fnvlist_alloc();
-				    KM_SLEEP) == 0);
 
-			if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
+			if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME)
-				VERIFY0(nvlist_lookup_string(spa->spa_config,
+				pool_name = fnvlist_lookup_string(
-					ZPOOL_CONFIG_POOL_NAME, &pool_name));
+				    spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
-			} else
+			else
 				pool_name = spa_name(spa);
 
-			VERIFY(nvlist_add_nvlist(nvl, pool_name,
+			fnvlist_add_nvlist(nvl, pool_name, spa->spa_config);
-			    spa->spa_config) == 0);
 			mutex_exit(&spa->spa_props_lock);
 		}
 
@@ -354,15 +347,15 @@ spa_all_configs(uint64_t *generation)
 	if (*generation == spa_config_generation)
 		return (NULL);
 
-	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
+	pools = fnvlist_alloc();
 
 	mutex_enter(&spa_namespace_lock);
 	while ((spa = spa_next(spa)) != NULL) {
 		if (INGLOBALZONE(curproc) ||
 		    zone_dataset_visible(spa_name(spa), NULL)) {
 			mutex_enter(&spa->spa_props_lock);
-			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
+			fnvlist_add_nvlist(pools, spa_name(spa),
-			    spa->spa_config) == 0);
+			    spa->spa_config);
 			mutex_exit(&spa->spa_props_lock);
 		}
 	}
@@ -396,6 +389,7 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 	boolean_t locked = B_FALSE;
 	uint64_t split_guid;
 	char *pool_name;
+	int config_gen_flags = 0;
 
 	if (vd == NULL) {
 		vd = rvd;
@@ -428,23 +422,17 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 	} else
 		pool_name = spa_name(spa);
 
-	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
+	config = fnvlist_alloc();
-
-	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
-	    spa_version(spa)) == 0);
-	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
-	    pool_name) == 0);
-	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
-	    spa_state(spa)) == 0);
-	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
-	    txg) == 0);
-	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
-	    spa_guid(spa)) == 0);
-	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
-	    spa->spa_errata) == 0);
-	VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
-	    ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
 
+	fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
+	fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, spa_name(spa));
+	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
+	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
+	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
+	fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
+	if (spa->spa_comment != NULL)
+		fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
+		    spa->spa_comment);
 
 #ifdef	_KERNEL
 	hostid = zone_get_hostid(NULL);
@@ -455,24 +443,21 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 	 */
 	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
 #endif	/* _KERNEL */
-	if (hostid != 0) {
+	if (hostid != 0)
-		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
+		fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
-		    hostid) == 0);
+	fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename);
-	}
-	VERIFY0(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
-	    utsname()->nodename));
 
 	if (vd != rvd) {
-		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
+		fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
-		    vd->vdev_top->vdev_guid) == 0);
+		    vd->vdev_top->vdev_guid);
-		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
+		fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
-		    vd->vdev_guid) == 0);
+		    vd->vdev_guid);
 		if (vd->vdev_isspare)
-			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
+			fnvlist_add_uint64(config,
-			    1ULL) == 0);
+			    ZPOOL_CONFIG_IS_SPARE, 1ULL);
 		if (vd->vdev_islog)
-			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
+			fnvlist_add_uint64(config,
-			    1ULL) == 0);
+			    ZPOOL_CONFIG_IS_LOG, 1ULL);
 		vd = vd->vdev_top;		/* label contains top config */
 	} else {
 		/*
@@ -480,8 +465,12 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 		 * in the mos config, and not in the vdev labels
 		 */
 		if (spa->spa_config_splitting != NULL)
-			VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
+			fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
-			    spa->spa_config_splitting) == 0);
+			    spa->spa_config_splitting);
+
+		fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
+
+		config_gen_flags |= VDEV_CONFIG_MOS;
 	}
 
 	/*
@@ -496,19 +485,18 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 	if (spa->spa_config_splitting != NULL &&
 	    nvlist_lookup_uint64(spa->spa_config_splitting,
 	    ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
-		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
+		fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid);
-		    split_guid) == 0);
 	}
 
-	nvroot = vdev_config_generate(spa, vd, getstats, 0);
+	nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
-	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
+	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
 	nvlist_free(nvroot);
 
 	/*
 	 * Store what's necessary for reading the MOS in the label.
 	 */
-	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
+	fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
-	    spa->spa_label_features) == 0);
+	    spa->spa_label_features);
 
 	if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
 		ddt_histogram_t *ddh;
@@ -517,23 +505,23 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 
 		ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
 		ddt_get_dedup_histogram(spa, ddh);
-		VERIFY(nvlist_add_uint64_array(config,
+		fnvlist_add_uint64_array(config,
 		    ZPOOL_CONFIG_DDT_HISTOGRAM,
-		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
+		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
 		kmem_free(ddh, sizeof (ddt_histogram_t));
 
 		ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
 		ddt_get_dedup_object_stats(spa, ddo);
-		VERIFY(nvlist_add_uint64_array(config,
+		fnvlist_add_uint64_array(config,
 		    ZPOOL_CONFIG_DDT_OBJ_STATS,
-		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
+		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
 		kmem_free(ddo, sizeof (ddt_object_t));
 
 		dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
 		ddt_get_dedup_stats(spa, dds);
-		VERIFY(nvlist_add_uint64_array(config,
+		fnvlist_add_uint64_array(config,
 		    ZPOOL_CONFIG_DDT_STATS,
-		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
+		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
 		kmem_free(dds, sizeof (ddt_stat_t));
 	}
 

module/zfs/vdev.c

@@ -519,6 +519,10 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 		    &vd->vdev_asize);
 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
 		    &vd->vdev_removing);
+		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
+		    &vd->vdev_top_zap);
+	} else {
+		ASSERT0(vd->vdev_top_zap);
 	}
 
 	if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
@@ -530,9 +534,18 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 		    spa_log_class(spa) : spa_normal_class(spa), vd);
 	}
 
+	if (vd->vdev_ops->vdev_op_leaf &&
+	    (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
+		(void) nvlist_lookup_uint64(nv,
+		    ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap);
+	} else {
+		ASSERT0(vd->vdev_leaf_zap);
+	}
+
 	/*
 	 * If we're a leaf vdev, try to load the DTL object and other state.
 	 */
+
 	if (vd->vdev_ops->vdev_op_leaf &&
 	    (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
 	    alloctype == VDEV_ALLOC_ROOTPOOL)) {
@@ -694,10 +707,12 @@ vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
 	tvd->vdev_ms_array = svd->vdev_ms_array;
 	tvd->vdev_ms_shift = svd->vdev_ms_shift;
 	tvd->vdev_ms_count = svd->vdev_ms_count;
+	tvd->vdev_top_zap = svd->vdev_top_zap;
 
 	svd->vdev_ms_array = 0;
 	svd->vdev_ms_shift = 0;
 	svd->vdev_ms_count = 0;
+	svd->vdev_top_zap = 0;
 
 	if (tvd->vdev_mg)
 		ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
@@ -1976,6 +1991,51 @@ vdev_dtl_load(vdev_t *vd)
 	return (error);
 }
 
+void
+vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
+{
+	spa_t *spa = vd->vdev_spa;
+
+	VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx));
+	VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
+	    zapobj, tx));
+}
+
+uint64_t
+vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx)
+{
+	spa_t *spa = vd->vdev_spa;
+	uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA,
+	    DMU_OT_NONE, 0, tx);
+
+	ASSERT(zap != 0);
+	VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps,
+	    zap, tx));
+
+	return (zap);
+}
+
+void
+vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
+{
+	uint64_t i;
+
+	if (vd->vdev_ops != &vdev_hole_ops &&
+	    vd->vdev_ops != &vdev_missing_ops &&
+	    vd->vdev_ops != &vdev_root_ops &&
+	    !vd->vdev_top->vdev_removing) {
+		if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) {
+			vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx);
+		}
+		if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
+			vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
+		}
+	}
+	for (i = 0; i < vd->vdev_children; i++) {
+		vdev_construct_zaps(vd->vdev_child[i], tx);
+	}
+}
+
 void
 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
 {
@@ -1998,6 +2058,18 @@ vdev_dtl_sync(vdev_t *vd, uint64_t txg)
 		space_map_close(vd->vdev_dtl_sm);
 		vd->vdev_dtl_sm = NULL;
 		mutex_exit(&vd->vdev_dtl_lock);
+
+		/*
+		 * We only destroy the leaf ZAP for detached leaves or for
+		 * removed log devices. Removed data devices handle leaf ZAP
+		 * cleanup later, once cancellation is no longer possible.
+		 */
+		if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached ||
+		    vd->vdev_top->vdev_islog)) {
+			vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx);
+			vd->vdev_leaf_zap = 0;
+		}
+
 		dmu_tx_commit(tx);
 		return;
 	}
@@ -2204,6 +2276,8 @@ vdev_remove(vdev_t *vd, uint64_t txg)
 	int m, i;
 
 	tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
+	ASSERT(vd == vd->vdev_top);
+	ASSERT3U(txg, ==, spa_syncing_txg(spa));
 
 	if (vd->vdev_ms != NULL) {
 		metaslab_group_t *mg = vd->vdev_mg;
@@ -2245,6 +2319,11 @@ vdev_remove(vdev_t *vd, uint64_t txg)
 		(void) dmu_object_free(mos, vd->vdev_ms_array, tx);
 		vd->vdev_ms_array = 0;
 	}
+
+	if (vd->vdev_islog && vd->vdev_top_zap != 0) {
+		vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx);
+		vd->vdev_top_zap = 0;
+	}
 	dmu_tx_commit(tx);
 }
 

module/zfs/vdev_label.c

@@ -291,6 +291,20 @@ vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
 	if (vd->vdev_crtxg)
 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
 
+	if (flags & VDEV_CONFIG_MOS) {
+		if (vd->vdev_leaf_zap != 0) {
+			ASSERT(vd->vdev_ops->vdev_op_leaf);
+			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
+			    vd->vdev_leaf_zap);
+		}
+
+		if (vd->vdev_top_zap != 0) {
+			ASSERT(vd == vd->vdev_top);
+			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
+			    vd->vdev_top_zap);
+		}
+	}
+
 	if (getstats) {
 		vdev_stat_t vs;
 		pool_scan_stat_t ps;

module/zfs/zap.c

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
 
@@ -965,8 +965,8 @@ zap_create_link(objset_t *os, dmu_object_type_t ot, uint64_t parent_obj,
 	uint64_t new_obj;
 
 	VERIFY((new_obj = zap_create(os, ot, DMU_OT_NONE, 0, tx)) > 0);
-	VERIFY(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
+	VERIFY0(zap_add(os, parent_obj, name, sizeof (uint64_t), 1, &new_obj,
-	    tx) == 0);
+	    tx));
 
 	return (new_obj);
 }

tests/runfiles/linux.run

@@ -614,6 +614,13 @@ tests = [
     'userquota_008_pos', 'userquota_009_pos',
     'userquota_011_pos', 'userquota_012_neg']
 
+# DISABLED:
+# vdev_zaps_007_pos -- fails due to a pre-existing issue with zpool split
+#
+[tests/functional/vdev_zaps]
+tests = ['vdev_zaps_001_pos', 'vdev_zaps_002_pos', 'vdev_zaps_003_pos',
+    'vdev_zaps_004_pos', 'vdev_zaps_005_pos', 'vdev_zaps_006_pos']
+
 # DISABLED:
 # write_dirs_002_pos - needs investigation
 [tests/functional/write_dirs]

tests/zfs-tests/tests/functional/Makefile.am

@@ -50,6 +50,7 @@ SUBDIRS = \
 	threadsappend \
 	truncate \
 	userquota \
+	vdev_zaps \
 	write_dirs \
 	xattr \
 	zvol

tests/zfs-tests/tests/functional/vdev_zaps/Makefile.am

@@ -0,0 +1,12 @@
+pkgdatadir = $(datadir)/@PACKAGE@/zfs-tests/tests/functional/vdev_zaps
+dist_pkgdata_SCRIPTS = \
+	setup.ksh \
+	cleanup.ksh \
+	vdev_zaps_001_pos.ksh \
+	vdev_zaps_002_pos.ksh \
+	vdev_zaps_003_pos.ksh \
+	vdev_zaps_004_pos.ksh \
+	vdev_zaps_005_pos.ksh \
+	vdev_zaps_006_pos.ksh \
+	vdev_zaps_007_pos.ksh \
+	vdev_zaps.kshlib

tests/zfs-tests/tests/functional/vdev_zaps/cleanup.ksh

@@ -0,0 +1,20 @@
+#!/bin/ksh -p
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+default_cleanup

tests/zfs-tests/tests/functional/vdev_zaps/setup.ksh

@@ -0,0 +1,21 @@
+#!/bin/ksh -p
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+rm -rf $TESTDIR || log_fail Could not remove $TESTDIR
+mkdir -p $TESTDIR || log_fail Could not create $TESTDIR

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps.kshlib

@@ -0,0 +1,114 @@
+#
+# CDDL HEADER START
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+function get_conf_section # regex conf
+{
+        typeset dsk_line next_vd_line conf section
+        typeset regex="$1"
+        typeset conf="$2"
+
+        dsk_line=$(grep -n "$regex" "$conf" | awk -F: '{print $1}')
+        if [[ -z "$dsk_line" ]]; then
+                return
+        fi
+        next_vd_line=$(tail -n +$dsk_line "$conf" | \
+                grep -n "children\[" | awk -F: '{print $1}' | head -n 1)
+
+        if [[ -n "$next_vd_line" ]]; then
+                section=$(cat "$conf" | sed "1,${dsk_line}d" | head -n \
+                        $(($next_vd_line - 2)))
+
+        else
+                section=$(tail -n +$dsk_line "$conf")
+        fi
+        echo "$section"
+}
+
+function get_leaf_vd_zap # dsk conf
+{
+        typeset section=$(get_conf_section "$1" "$2")
+        echo "$section" | egrep \
+                "com.delphix:vdev_zap_leaf: [0-9]+" | awk '{print $2}'
+}
+
+function get_top_vd_zap # dsk conf
+{
+        typeset section=$(get_conf_section "$1" "$2")
+        echo "$section" | egrep \
+                "com.delphix:vdev_zap_top: [0-9]+" | awk '{print $2}'
+}
+
+function assert_has_sentinel # conf
+{
+        res=$(grep "com.delphix:has_per_vdev_zaps" "$1")
+        [[ -z "$res" ]] && log_fail "Pool missing ZAP feature sentinel value"
+}
+
+function assert_zap_common # pool vd lvl zapobj
+{
+        typeset pool=$1
+        typeset vd="$2"
+        typeset lvl=$3
+        typeset zapobj=$4
+
+        if [[ -z "$zapobj" ]]; then
+                log_fail "$vd on $pool has no $lvl ZAP in config"
+        elif [[ -z "$(zdb -d $pool $zapobj | grep 'zap')" ]]; then
+                log_fail "$vd on $pool has no $lvl ZAP in MOS"
+        fi
+}
+
+function assert_top_zap # pool vd conf
+{
+        typeset pool=$1
+        typeset vd="$2"
+        typeset conf=$3
+
+        top_zap=$(get_top_vd_zap "$vd" $conf)
+        assert_zap_common $pool "$vd" "top" $top_zap
+}
+
+function assert_leaf_zap # pool vd conf
+{
+        typeset pool=$1
+        typeset vd="$2"
+        typeset conf=$3
+
+        leaf_zap=$(get_leaf_vd_zap "$vd" $conf)
+        assert_zap_common $pool "$vd" "leaf" $leaf_zap
+}
+
+#
+# Code common to setup/teardown for each test.
+#
+
+function cleanup
+{
+        if datasetexists $TESTPOOL ; then
+                log_must zpool destroy -f $TESTPOOL
+        fi
+        if [[ -e $conf ]]; then
+                log_must $RM -f "$conf"
+        fi
+        if [[ -e $POOL2 ]]; then
+                log_must zpool destroy -f $POOL2
+        fi
+}
+
+log_onexit cleanup

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps_001_pos.ksh

@@ -0,0 +1,42 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+#
+# Description:
+# Verify that per-vdev ZAPs are created with one vdev.
+#
+# Strategy:
+# 1. Create a pool with one disk.
+# 2. Verify that the disk has a top and leaf ZAP in its config and the MOS.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/vdev_zaps/vdev_zaps.kshlib
+
+log_assert "Per-vdev ZAPs are created on pool creation with one disk."
+
+DISK=${DISKS%% *}
+
+log_must zpool create -f $TESTPOOL $DISK
+conf="$TESTDIR/vz001"
+log_must zdb -PC $TESTPOOL > $conf
+
+assert_top_zap $TESTPOOL $DISK "$conf"
+assert_leaf_zap $TESTPOOL $DISK "$conf"
+assert_has_sentinel "$conf"
+
+log_pass "Per-vdev ZAPs are created in a one-disk pool."

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps_002_pos.ksh

@@ -0,0 +1,44 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+#
+# Description:
+# Verify that per-vdev ZAPs are created with multiple vdevs.
+#
+# Strategy:
+# 1. Create a pool with multiple disks.
+# 2. Verify that each has both a top and leaf zap.
+# 3. Verify that each of those ZAPs exists in the MOS.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/vdev_zaps/vdev_zaps.kshlib
+
+log_assert "Per-vdev ZAPs are created on pool creation with many disks."
+
+log_must zpool create -f $TESTPOOL $DISKS
+
+conf="$TESTDIR/vz002"
+log_must zdb -PC $TESTPOOL > $conf
+
+assert_has_sentinel "$conf"
+for DISK in $DISKS; do
+	assert_top_zap $TESTPOOL $DISK "$conf"
+	assert_leaf_zap $TESTPOOL $DISK "$conf"
+done
+
+log_pass

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps_003_pos.ksh

@@ -0,0 +1,47 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+#
+# Description:
+# Verify that per-vdev ZAPs are created with multi-level vdev tree.
+#
+# Strategy:
+# 1. Create a pool with a multi-disk mirror.
+# 2. Verify that mirror has top ZAP but no leaf ZAP.
+# 3. Verify that each disk has a leaf ZAP but no top ZAP.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/vdev_zaps/vdev_zaps.kshlib
+
+log_assert "Per-vdev ZAPs are created on pool creation with multi-level vdev "\
+        "trees."
+
+log_must zpool create -f $TESTPOOL mirror $DISKS
+
+conf="$TESTDIR/vz003"
+log_must zdb -PC $TESTPOOL > $conf
+
+assert_has_sentinel "$conf"
+assert_top_zap $TESTPOOL "type: 'mirror'" "$conf"
+for DISK in $DISKS; do
+	assert_leaf_zap $TESTPOOL $DISK "$conf"
+        top_zap=$(get_top_vd_zap $DISK "$conf")
+        [[ -n "$top_zap" ]] && log_fail "Leaf vdev $DISK has top-level ZAP."
+done
+
+log_pass

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps_004_pos.ksh

@@ -0,0 +1,94 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+#
+# Description:
+# Verify that per-vdev ZAPs are properly transferred on attach/detach.
+#
+# Strategy:
+# 1. Create a pool with one disk. Verify that it has a top and leaf ZAP.
+# 2. Attach a disk.
+# 3. Verify that top-level and leaf-level ZAPs were transferred properly.
+# 4. Verify that the newly-attached disk has a leaf ZAP.
+# 5. Detach the original disk.
+# 6. Verify that top-level and leaf-level ZAPs were transferred properly.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/vdev_zaps/vdev_zaps.kshlib
+
+log_assert "Per-vdev ZAPs are transferred properly on attach/detach"
+
+DISK=${DISKS%% *}
+log_must zpool create -f $TESTPOOL $DISK
+
+# Make the pool.
+conf="$TESTDIR/vz004"
+log_must zdb -PC $TESTPOOL > $conf
+assert_has_sentinel "$conf"
+orig_top=$(get_top_vd_zap $DISK $conf)
+orig_leaf=$(get_leaf_vd_zap $DISK $conf)
+assert_zap_common $TESTPOOL $DISK "top" $orig_top
+
+#
+# Attach a disk.
+#
+
+disk2=$(echo $DISKS | awk '{print $2}')
+log_must zpool attach $TESTPOOL $DISK $disk2
+log_must zdb -PC $TESTPOOL > $conf
+
+# Ensure top-level ZAP was transferred successfully.
+new_top=$(get_top_vd_zap "type: 'mirror'" $conf)
+if [[ "$new_top" -ne "$orig_top" ]]; then
+        log_fail "Top-level ZAP wasn't transferred successfully on attach."
+fi
+
+# Ensure leaf ZAP of original disk was transferred successfully.
+new_leaf=$(get_leaf_vd_zap $DISK $conf)
+if [[ "$new_leaf" -ne "$orig_leaf" ]]; then
+        log_fail "$DISK used to have leaf-level ZAP $orig_leaf, now has "\
+                "$new_leaf"
+fi
+# Ensure original disk no longer has top-level ZAP.
+dsk1_top=$(get_top_vd_zap $DISK $conf)
+[[ -n "$dsk1_top" ]] && log_fail "$DISK has top-level ZAP, but is only leaf."
+
+# Ensure attached disk got a leaf-level ZAP but not a top-level ZAP.
+dsk2_top=$(get_top_vd_zap $disk2 $conf)
+dsk2_leaf=$(get_leaf_vd_zap $disk2 $conf)
+[[ -n "$dsk2_top" ]] && log_fail "Attached disk $disk2 has top ZAP."
+[[ -z "$dsk2_leaf" ]] && log_fail "Attached disk $disk2 has no leaf ZAP."
+
+#
+# Detach original disk.
+#
+
+log_must zpool detach $TESTPOOL $DISK
+log_must zdb -PC $TESTPOOL > $conf
+
+final_top=$(get_top_vd_zap $disk2 $conf)
+final_leaf=$(get_leaf_vd_zap $disk2 $conf)
+# Make sure top ZAP was successfully transferred.
+[[ "$final_top" -ne "$orig_top" ]] && log_fail "Lost top-level ZAP when "\
+        "promoting $disk2 (expected $orig_top, found $final_top)"
+
+# Make sure leaf ZAP was successfully transferred.
+[[ "$final_leaf" -ne "$dsk2_leaf" ]] && log_fail "$disk2 lost its leaf ZAP "\
+        "on promotion (expected $dsk2_leaf, got $final_leaf)"
+
+log_pass

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps_005_pos.ksh

@@ -0,0 +1,62 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+#
+# Description:
+# Verify that per-vdev ZAPs persist when the pool is exported and imported.
+#
+# Strategy:
+# 1. Create a pool with a disk.
+# 2. Export the pool and re-import it.
+# 3. Verify that the ZAPs aren't different.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/vdev_zaps/vdev_zaps.kshlib
+
+log_assert "Per-vdev ZAPs persist across export/import."
+
+DISK=${DISKS%% *}
+log_must zpool create -f $TESTPOOL $DISK
+
+# Make the pool.
+conf="$TESTDIR/vz005"
+log_must zdb -PC $TESTPOOL > $conf
+assert_has_sentinel "$conf"
+orig_top=$(get_top_vd_zap $DISK $conf)
+orig_leaf=$(get_leaf_vd_zap $DISK $conf)
+assert_zap_common $TESTPOOL $DISK "top" $orig_top
+assert_zap_common $TESTPOOL $DISK "leaf" $orig_leaf
+
+# Export the pool.
+log_must zpool export $TESTPOOL
+
+# Import the pool.
+log_must zpool import $TESTPOOL
+
+# Verify that ZAPs persisted.
+log_must zdb -PC $TESTPOOL > $conf
+
+new_top=$(get_top_vd_zap $DISK $conf)
+new_leaf=$(get_leaf_vd_zap $DISK $conf)
+
+[[ "$new_top" -ne "$orig_top" ]] && log_fail "Top ZAP ($new_top) after "\
+        "import does not match top ZAP before export ($orig_top)"
+[[ "$new_leaf" -ne "$orig_leaf" ]] && log_fail "Leaf ZAP ($new_leaf) after "\
+        "import does not match leaf ZAP before export ($orig_leaf)"
+
+log_pass

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps_006_pos.ksh

@@ -0,0 +1,46 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015, 2016 by Delphix. All rights reserved.
+#
+
+#
+# Description:
+# Verify that top-level per-vdev ZAPs are created for added devices
+#
+# Strategy:
+# 1. Create a pool with one disk.
+# 2. Add a disk.
+# 3. Verify its ZAPs were created.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/vdev_zaps/vdev_zaps.kshlib
+
+DISK_ARR=($DISKS)
+DISK=${DISK_ARR[0]}
+log_must zpool create -f $TESTPOOL $DISK
+
+log_assert "Per-vdev ZAPs are created for added vdevs."
+
+log_must zpool add -f $TESTPOOL ${DISK_ARR[1]}
+conf="$TESTDIR/vz006"
+log_must zdb -PC $TESTPOOL > $conf
+
+assert_has_sentinel "$conf"
+orig_top=$(get_top_vd_zap ${DISK_ARR[1]} $conf)
+assert_zap_common $TESTPOOL ${DISK_ARR[1]} "top" $orig_top
+assert_leaf_zap $TESTPOOL ${DISK_ARR[1]} "$conf"
+
+log_pass

tests/zfs-tests/tests/functional/vdev_zaps/vdev_zaps_007_pos.ksh

@@ -0,0 +1,74 @@
+#!/bin/ksh
+
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+
+#
+# Copyright (c) 2015 by Delphix. All rights reserved.
+#
+
+#
+# Description:
+# Verify that ZAPs are handled properly during mirror pool splitting.
+#
+# Strategy:
+# 1. Create a pool with a two-way mirror.
+# 2. Split the pool.
+# 3. Verify that the ZAPs in the old pool persisted.
+# 4. Import the new pool.
+# 5. Verify that the ZAPs in the new pool persisted.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/vdev_zaps/vdev_zaps.kshlib
+
+DISK_ARR=($DISKS)
+POOL2=${TESTPOOL}2
+log_must zpool create -f $TESTPOOL mirror ${DISK_ARR[0]} ${DISK_ARR[1]}
+
+log_assert "Per-vdev ZAPs persist correctly on the original pool after split."
+conf="$TESTDIR/vz007"
+log_must zdb -PC $TESTPOOL > $conf
+
+assert_has_sentinel "$conf"
+orig_top=$(get_top_vd_zap "type: 'mirror'" $conf)
+orig_leaf0=$(get_leaf_vd_zap ${DISK_ARR[0]} $conf)
+orig_leaf1=$(get_leaf_vd_zap ${DISK_ARR[1]} $conf)
+assert_zap_common $TESTPOOL "type: 'mirror'" "top" $orig_top
+assert_zap_common $TESTPOOL ${DISK_ARR[0]} "leaf" $orig_leaf0
+assert_zap_common $TESTPOOL ${DISK_ARR[1]} "leaf" $orig_leaf1
+
+log_must zpool split $TESTPOOL $POOL2 ${DISK_ARR[1]}
+
+# Make sure old pool's ZAPs are consistent.
+log_must zdb -PC $TESTPOOL > $conf
+new_leaf0=$(get_leaf_vd_zap ${DISK_ARR[0]} $conf)
+new_top_s0=$(get_top_vd_zap ${DISK_ARR[0]} $conf)
+
+[[ "$new_leaf0" -ne "$orig_leaf0" ]] && log_fail "Leaf ZAP in original pool "\
+        "didn't persist (expected $orig_leaf0, got $new_leaf0)"
+[[ "$new_top_s0" -ne "$orig_top" ]] && log_fail "Top ZAP in original pool "\
+        "didn't persist (expected $orig_top, got $new_top_s0)"
+
+log_assert "Per-vdev ZAPs persist on the new pool after import."
+
+# Import the split pool.
+log_must zpool import $POOL2
+log_must zdb -PC $TESTPOOL > $conf
+
+new_leaf1=$(get_leaf_vd_zap ${DISK_ARR[1]} $conf)
+new_top_s1=$(get_top_vd_zap ${DISK_ARR[1]} $conf)
+[[ "$new_leaf1" -ne "$orig_leaf1" ]] && log_fail "Leaf ZAP in new pool "\
+        "didn't persist (expected $orig_leaf1, got $new_leaf1)"
+[[ "$new_top_s1" -ne "$orig_top" ]] && log_fail "Top ZAP in new pool "\
+        "didn't persist (expected $orig_top, got $new_top_s1)"
+
+log_pass