diff --git a/stand/libsa/zfs/zfsimpl.c b/stand/libsa/zfs/zfsimpl.c
--- a/stand/libsa/zfs/zfsimpl.c
+++ b/stand/libsa/zfs/zfsimpl.c
@@ -1267,6 +1267,19 @@
 	return (rc);
 }
 
+static void
+vdev_free(struct vdev *vdev)
+{
+	struct vdev *kid, *safe;
+
+	STAILQ_FOREACH_SAFE(kid, &vdev->v_children, v_childlink, safe)
+		vdev_free(kid);
+	/* See comment in vdev_create().  Root vdev isn't linked. */
+	if (vdev->v_read != NULL)
+		STAILQ_REMOVE(&zfs_vdevs, vdev, vdev, v_alllink);
+	free(vdev);
+}
+
 static int
 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
 {
@@ -1373,13 +1386,6 @@
 	spa->spa_uberblock = &spa->spa_uberblock_master;
 	spa->spa_mos = &spa->spa_mos_master;
 	spa->spa_guid = guid;
-	spa->spa_root_vdev = vdev_create(guid, NULL);
-	if (spa->spa_root_vdev == NULL) {
-		free(spa->spa_name);
-		free(spa);
-		return (NULL);
-	}
-	spa->spa_root_vdev->v_name = strdup("root");
 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
 
 	return (spa);
@@ -2006,7 +2012,7 @@
 	vdev_t *vdev;
 	nvlist_t *nvl;
 	uint64_t val;
-	uint64_t guid, vdev_children;
+	uint64_t guid;
 	uint64_t pool_txg, pool_guid;
 	const char *pool_name;
 	int rc, namelen;
@@ -2067,7 +2073,9 @@
 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
 	    NULL, &pool_guid, NULL) != 0 ||
 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
-	    NULL, &pool_name, &namelen) != 0) {
+	    NULL, &pool_name, &namelen) != 0 ||
+	    nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
+	    NULL, &guid, NULL) != 0) {
 		/*
 		 * Cache and spare devices end up here - just ignore
 		 * them.
@@ -2083,8 +2091,6 @@
 	if (spa == NULL) {
 		char *name;
 
-		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
-		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
 		name = malloc(namelen + 1);
 		if (name == NULL) {
 			nvlist_destroy(nvl);
@@ -2098,10 +2104,22 @@
 			nvlist_destroy(nvl);
 			return (ENOMEM);
 		}
-		spa->spa_root_vdev->v_nchildren = vdev_children;
+	} else if (pool_txg > spa->spa_txg) {
+		printf("ZFS: pool %s ignoring stale config:txg 0x%jx:0x%jx, "
+		    "new config:txg 0x%jx:0x%jx\n", spa->spa_name,
+		    spa->spa_root_vdev->v_guid, spa->spa_txg,
+		    guid, pool_txg);
+		vdev_free(spa->spa_root_vdev);
+		spa->spa_root_vdev = NULL;
 	}
-	if (pool_txg > spa->spa_txg)
+
+	if (spa->spa_root_vdev == NULL) {
+		spa->spa_root_vdev = vdev_create(pool_guid, NULL);
+		if (spa->spa_root_vdev == NULL)
+			return (ENOMEM);
+		spa->spa_root_vdev->v_name = spa->spa_name;
 		spa->spa_txg = pool_txg;
+	}
 
 	/*
 	 * Get the vdev tree and create our in-core copy of it.
@@ -2109,11 +2127,6 @@
 	 * be some kind of alias (overlapping slices, dangerously dedicated
 	 * disks etc).
 	 */
-	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
-	    NULL, &guid, NULL) != 0) {
-		nvlist_destroy(nvl);
-		return (EIO);
-	}
 	vdev = vdev_find(guid);
 	/* Has this vdev already been inited? */
 	if (vdev && vdev->v_phys_read) {
@@ -3541,8 +3554,10 @@
 		return (EIO);
 	}
 	rc = load_nvlist(spa, config_object, &nvlist);
-	if (rc != 0)
+	if (rc != 0) {
+		printf("ZFS: failed to load pool %s nvlist\n", spa->spa_name);
 		return (rc);
+	}
 
 	rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT,
 	    sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t),