Pool allocation classes

Allocation Classes add the ability to have allocation classes in a
pool that are dedicated to serving specific block categories, such
as DDT data, metadata, and small file blocks. A pool can opt-in to
this feature by adding a 'special' or 'dedup' top-level VDEV.

Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed-by: Richard Laager <rlaager@wiktel.com>
Reviewed-by: Alek Pinchuk <apinchuk@datto.com>
Reviewed-by: Håkan Johansson <f96hajo@chalmers.se>
Reviewed-by: Andreas Dilger <andreas.dilger@chamcloud.com>
Reviewed-by: DHE <git@dehacked.net>
Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed-by: Gregor Kopka <gregor@kopka.net>
Reviewed-by: Kash Pande <kash@tripleback.net>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Signed-off-by: Don Brady <don.brady@delphix.com>
Closes #5182

cmd/zdb/zdb.c

@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2014 Integros [integros.com]
  * Copyright 2016 Nexenta Systems, Inc.
  * Copyright (c) 2017, 2018 Lawrence Livermore National Security, LLC.
@@ -932,13 +932,23 @@ dump_metaslab(metaslab_t *msp)
 static void
 print_vdev_metaslab_header(vdev_t *vd)
 {
-	(void) printf("\tvdev %10llu\n\t%-10s%5llu   %-19s   %-15s   %-10s\n",
+	vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
-	    (u_longlong_t)vd->vdev_id,
+	const char *bias_str;
+
+	bias_str = (alloc_bias == VDEV_BIAS_LOG || vd->vdev_islog) ?
+	    VDEV_ALLOC_BIAS_LOG :
+	    (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
+	    (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP :
+	    vd->vdev_islog ? "log" : "";
+
+	(void) printf("\tvdev %10llu   %s\n"
+	    "\t%-10s%5llu   %-19s   %-15s   %-12s\n",
+	    (u_longlong_t)vd->vdev_id, bias_str,
 	    "metaslabs", (u_longlong_t)vd->vdev_ms_count,
 	    "offset", "spacemap", "free");
-	(void) printf("\t%15s   %19s   %15s   %10s\n",
+	(void) printf("\t%15s   %19s   %15s   %12s\n",
 	    "---------------", "-------------------",
-	    "---------------", "-------------");
+	    "---------------", "------------");
 }
 
 static void
@@ -954,7 +964,7 @@ dump_metaslab_groups(spa_t *spa)
 		vdev_t *tvd = rvd->vdev_child[c];
 		metaslab_group_t *mg = tvd->vdev_mg;
 
-		if (mg->mg_class != mc)
+		if (mg == NULL || mg->mg_class != mc)
 			continue;
 
 		metaslab_group_histogram_verify(mg);
@@ -3158,6 +3168,7 @@ typedef struct zdb_blkstats {
 	uint64_t zb_count;
 	uint64_t zb_gangs;
 	uint64_t zb_ditto_samevdev;
+	uint64_t zb_ditto_same_ms;
 	uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
 } zdb_blkstats_t;
 
@@ -3197,6 +3208,16 @@ typedef struct zdb_cb {
 	uint32_t	**zcb_vd_obsolete_counts;
 } zdb_cb_t;
 
+/* test if two DVA offsets from same vdev are within the same metaslab */
+static boolean_t
+same_metaslab(spa_t *spa, uint64_t vdev, uint64_t off1, uint64_t off2)
+{
+	vdev_t *vd = vdev_lookup_top(spa, vdev);
+	uint64_t ms_shift = vd->vdev_ms_shift;
+
+	return ((off1 >> ms_shift) == (off2 >> ms_shift));
+}
+
 static void
 zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
     dmu_object_type_t type)
@@ -3209,6 +3230,8 @@ zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
 	if (zilog && zil_bp_tree_add(zilog, bp) != 0)
 		return;
 
+	spa_config_enter(zcb->zcb_spa, SCL_CONFIG, FTAG, RW_READER);
+
 	for (i = 0; i < 4; i++) {
 		int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
 		int t = (i & 1) ? type : ZDB_OT_TOTAL;
@@ -3234,8 +3257,15 @@ zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
 		switch (BP_GET_NDVAS(bp)) {
 		case 2:
 			if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
-			    DVA_GET_VDEV(&bp->blk_dva[1]))
+			    DVA_GET_VDEV(&bp->blk_dva[1])) {
 				zb->zb_ditto_samevdev++;
+
+				if (same_metaslab(zcb->zcb_spa,
+				    DVA_GET_VDEV(&bp->blk_dva[0]),
+				    DVA_GET_OFFSET(&bp->blk_dva[0]),
+				    DVA_GET_OFFSET(&bp->blk_dva[1])))
+					zb->zb_ditto_same_ms++;
+			}
 			break;
 		case 3:
 			equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
@@ -3244,13 +3274,37 @@ zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
 			    DVA_GET_VDEV(&bp->blk_dva[2])) +
 			    (DVA_GET_VDEV(&bp->blk_dva[1]) ==
 			    DVA_GET_VDEV(&bp->blk_dva[2]));
-			if (equal != 0)
+			if (equal != 0) {
 				zb->zb_ditto_samevdev++;
+
+				if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
+				    DVA_GET_VDEV(&bp->blk_dva[1]) &&
+				    same_metaslab(zcb->zcb_spa,
+				    DVA_GET_VDEV(&bp->blk_dva[0]),
+				    DVA_GET_OFFSET(&bp->blk_dva[0]),
+				    DVA_GET_OFFSET(&bp->blk_dva[1])))
+					zb->zb_ditto_same_ms++;
+				else if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
+				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
+				    same_metaslab(zcb->zcb_spa,
+				    DVA_GET_VDEV(&bp->blk_dva[0]),
+				    DVA_GET_OFFSET(&bp->blk_dva[0]),
+				    DVA_GET_OFFSET(&bp->blk_dva[2])))
+					zb->zb_ditto_same_ms++;
+				else if (DVA_GET_VDEV(&bp->blk_dva[1]) ==
+				    DVA_GET_VDEV(&bp->blk_dva[2]) &&
+				    same_metaslab(zcb->zcb_spa,
+				    DVA_GET_VDEV(&bp->blk_dva[1]),
+				    DVA_GET_OFFSET(&bp->blk_dva[1]),
+				    DVA_GET_OFFSET(&bp->blk_dva[2])))
+					zb->zb_ditto_same_ms++;
+			}
 			break;
 		}
-
 	}
 
+	spa_config_exit(zcb->zcb_spa, SCL_CONFIG, FTAG);
+
 	if (BP_IS_EMBEDDED(bp)) {
 		zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
 		zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
@@ -4103,6 +4157,8 @@ dump_block_stats(spa_t *spa)
 		flags |= TRAVERSE_PREFETCH_DATA;
 
 	zcb.zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
+	zcb.zcb_totalasize += metaslab_class_get_alloc(spa_special_class(spa));
+	zcb.zcb_totalasize += metaslab_class_get_alloc(spa_dedup_class(spa));
 	zcb.zcb_start = zcb.zcb_lastprint = gethrtime();
 	err = traverse_pool(spa, 0, flags, zdb_blkptr_cb, &zcb);
 
@@ -4147,7 +4203,10 @@ dump_block_stats(spa_t *spa)
 	norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
 	norm_space = metaslab_class_get_space(spa_normal_class(spa));
 
-	total_alloc = norm_alloc + metaslab_class_get_alloc(spa_log_class(spa));
+	total_alloc = norm_alloc +
+	    metaslab_class_get_alloc(spa_log_class(spa)) +
+	    metaslab_class_get_alloc(spa_special_class(spa)) +
+	    metaslab_class_get_alloc(spa_dedup_class(spa));
 	total_found = tzb->zb_asize - zcb.zcb_dedup_asize +
 	    zcb.zcb_removing_size + zcb.zcb_checkpoint_size;
 
@@ -4169,31 +4228,50 @@ dump_block_stats(spa_t *spa)
 		return (2);
 
 	(void) printf("\n");
-	(void) printf("\tbp count:      %10llu\n",
+	(void) printf("\t%-16s %14llu\n", "bp count:",
 	    (u_longlong_t)tzb->zb_count);
-	(void) printf("\tganged count:  %10llu\n",
+	(void) printf("\t%-16s %14llu\n", "ganged count:",
 	    (longlong_t)tzb->zb_gangs);
-	(void) printf("\tbp logical:    %10llu      avg: %6llu\n",
+	(void) printf("\t%-16s %14llu      avg: %6llu\n", "bp logical:",
 	    (u_longlong_t)tzb->zb_lsize,
 	    (u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
-	(void) printf("\tbp physical:   %10llu      avg:"
+	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
-	    " %6llu     compression: %6.2f\n",
+	    "bp physical:", (u_longlong_t)tzb->zb_psize,
-	    (u_longlong_t)tzb->zb_psize,
 	    (u_longlong_t)(tzb->zb_psize / tzb->zb_count),
 	    (double)tzb->zb_lsize / tzb->zb_psize);
-	(void) printf("\tbp allocated:  %10llu      avg:"
+	(void) printf("\t%-16s %14llu      avg: %6llu     compression: %6.2f\n",
-	    " %6llu     compression: %6.2f\n",
+	    "bp allocated:", (u_longlong_t)tzb->zb_asize,
-	    (u_longlong_t)tzb->zb_asize,
 	    (u_longlong_t)(tzb->zb_asize / tzb->zb_count),
 	    (double)tzb->zb_lsize / tzb->zb_asize);
-	(void) printf("\tbp deduped:    %10llu    ref>1:"
+	(void) printf("\t%-16s %14llu    ref>1: %6llu   deduplication: %6.2f\n",
-	    " %6llu   deduplication: %6.2f\n",
+	    "bp deduped:", (u_longlong_t)zcb.zcb_dedup_asize,
-	    (u_longlong_t)zcb.zcb_dedup_asize,
 	    (u_longlong_t)zcb.zcb_dedup_blocks,
 	    (double)zcb.zcb_dedup_asize / tzb->zb_asize + 1.0);
-	(void) printf("\tSPA allocated: %10llu     used: %5.2f%%\n",
+	(void) printf("\t%-16s %14llu     used: %5.2f%%\n", "Normal class:",
 	    (u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
 
+	if (spa_special_class(spa)->mc_rotor != NULL) {
+		uint64_t alloc = metaslab_class_get_alloc(
+		    spa_special_class(spa));
+		uint64_t space = metaslab_class_get_space(
+		    spa_special_class(spa));
+
+		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
+		    "Special class", (u_longlong_t)alloc,
+		    100.0 * alloc / space);
+	}
+
+	if (spa_dedup_class(spa)->mc_rotor != NULL) {
+		uint64_t alloc = metaslab_class_get_alloc(
+		    spa_dedup_class(spa));
+		uint64_t space = metaslab_class_get_space(
+		    spa_dedup_class(spa));
+
+		(void) printf("\t%-16s %14llu     used: %5.2f%%\n",
+		    "Dedup class", (u_longlong_t)alloc,
+		    100.0 * alloc / space);
+	}
+
 	for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
 		if (zcb.zcb_embedded_blocks[i] == 0)
 			continue;
@@ -4215,6 +4293,10 @@ dump_block_stats(spa_t *spa)
 		(void) printf("\tDittoed blocks on same vdev: %llu\n",
 		    (longlong_t)tzb->zb_ditto_samevdev);
 	}
+	if (tzb->zb_ditto_same_ms != 0) {
+		(void) printf("\tDittoed blocks in same metaslab: %llu\n",
+		    (longlong_t)tzb->zb_ditto_same_ms);
+	}
 
 	for (uint64_t v = 0; v < spa->spa_root_vdev->vdev_children; v++) {
 		vdev_t *vd = spa->spa_root_vdev->vdev_child[v];

cmd/zpool/zpool_main.c

@@ -22,13 +22,14 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
- * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2012 by Frederik Wessels. All rights reserved.
  * Copyright (c) 2012 by Cyril Plisko. All rights reserved.
  * Copyright (c) 2013 by Prasad Joshi (sTec). All rights reserved.
  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>.
  * Copyright (c) 2017 Datto Inc.
  * Copyright (c) 2017 Open-E, Inc. All Rights Reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #include <assert.h>
@@ -291,6 +292,8 @@ static zpool_command_t command_table[] = {
 
 #define	NCOMMAND	(ARRAY_SIZE(command_table))
 
+#define	VDEV_ALLOC_CLASS_LOGS	"logs"
+
 static zpool_command_t *current_command;
 static char history_str[HIS_MAX_RECORD_LEN];
 static boolean_t log_history = B_TRUE;
@@ -393,7 +396,7 @@ print_prop_cb(int prop, void *cb)
 {
 	FILE *fp = cb;
 
-	(void) fprintf(fp, "\t%-15s  ", zpool_prop_to_name(prop));
+	(void) fprintf(fp, "\t%-19s  ", zpool_prop_to_name(prop));
 
 	if (zpool_prop_readonly(prop))
 		(void) fprintf(fp, "  NO   ");
@@ -445,14 +448,14 @@ usage(boolean_t requested)
 		(void) fprintf(fp,
 		    gettext("\nthe following properties are supported:\n"));
 
-		(void) fprintf(fp, "\n\t%-15s  %s   %s\n\n",
+		(void) fprintf(fp, "\n\t%-19s  %s   %s\n\n",
 		    "PROPERTY", "EDIT", "VALUES");
 
 		/* Iterate over all properties */
 		(void) zprop_iter(print_prop_cb, fp, B_FALSE, B_TRUE,
 		    ZFS_TYPE_POOL);
 
-		(void) fprintf(fp, "\t%-15s   ", "feature@...");
+		(void) fprintf(fp, "\t%-19s   ", "feature@...");
 		(void) fprintf(fp, "YES   disabled | enabled | active\n");
 
 		(void) fprintf(fp, gettext("\nThe feature@ properties must be "
@@ -470,32 +473,45 @@ usage(boolean_t requested)
 	exit(requested ? 0 : 2);
 }
 
-void
+/*
+ * print a pool vdev config for dry runs
+ */
+static void
 print_vdev_tree(zpool_handle_t *zhp, const char *name, nvlist_t *nv, int indent,
-    boolean_t print_logs, int name_flags)
+    const char *match, int name_flags)
 {
 	nvlist_t **child;
 	uint_t c, children;
 	char *vname;
-
+	boolean_t printed = B_FALSE;
-	if (name != NULL)
-		(void) printf("\t%*s%s\n", indent, "", name);
 
 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
-	    &child, &children) != 0)
+	    &child, &children) != 0) {
+		if (name != NULL)
+			(void) printf("\t%*s%s\n", indent, "", name);
 		return;
+	}
 
 	for (c = 0; c < children; c++) {
 		uint64_t is_log = B_FALSE;
+		char *class = "";
 
 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 		    &is_log);
-		if ((is_log && !print_logs) || (!is_log && print_logs))
+		if (is_log)
+			class = VDEV_ALLOC_BIAS_LOG;
+		(void) nvlist_lookup_string(child[c],
+		    ZPOOL_CONFIG_ALLOCATION_BIAS, &class);
+		if (strcmp(match, class) != 0)
 			continue;
 
+		if (!printed && name != NULL) {
+			(void) printf("\t%*s%s\n", indent, "", name);
+			printed = B_TRUE;
+		}
 		vname = zpool_vdev_name(g_zfs, zhp, child[c], name_flags);
-		print_vdev_tree(zhp, vname, child[c], indent + 2,
+		print_vdev_tree(zhp, vname, child[c], indent + 2, "",
-		    B_FALSE, name_flags);
+		    name_flags);
 		free(vname);
 	}
 }
@@ -734,20 +750,25 @@ zpool_do_add(int argc, char **argv)
 		    "configuration:\n"), zpool_get_name(zhp));
 
 		/* print original main pool and new tree */
-		print_vdev_tree(zhp, poolname, poolnvroot, 0, B_FALSE,
+		print_vdev_tree(zhp, poolname, poolnvroot, 0, "",
-		    name_flags);
+		    name_flags | VDEV_NAME_TYPE_ID);
-		print_vdev_tree(zhp, NULL, nvroot, 0, B_FALSE, name_flags);
+		print_vdev_tree(zhp, NULL, nvroot, 0, "", name_flags);
 
-		/* Do the same for the logs */
+		/* print other classes: 'dedup', 'special', and 'log' */
-		if (num_logs(poolnvroot) > 0) {
+		print_vdev_tree(zhp, "dedup", poolnvroot, 0,
-			print_vdev_tree(zhp, "logs", poolnvroot, 0, B_TRUE,
+		    VDEV_ALLOC_BIAS_DEDUP, name_flags);
+		print_vdev_tree(zhp, NULL, nvroot, 0, VDEV_ALLOC_BIAS_DEDUP,
 		    name_flags);
-			print_vdev_tree(zhp, NULL, nvroot, 0, B_TRUE,
+
+		print_vdev_tree(zhp, "special", poolnvroot, 0,
+		    VDEV_ALLOC_BIAS_SPECIAL, name_flags);
+		print_vdev_tree(zhp, NULL, nvroot, 0, VDEV_ALLOC_BIAS_SPECIAL,
 		    name_flags);
-		} else if (num_logs(nvroot) > 0) {
+
-			print_vdev_tree(zhp, "logs", nvroot, 0, B_TRUE,
+		print_vdev_tree(zhp, "logs", poolnvroot, 0, VDEV_ALLOC_BIAS_LOG,
+		    name_flags);
+		print_vdev_tree(zhp, NULL, nvroot, 0, VDEV_ALLOC_BIAS_LOG,
 		    name_flags);
-		}
 
 		/* Do the same for the caches */
 		if (nvlist_lookup_nvlist_array(poolnvroot, ZPOOL_CONFIG_L2CACHE,
@@ -1305,9 +1326,13 @@ zpool_do_create(int argc, char **argv)
 		(void) printf(gettext("would create '%s' with the "
 		    "following layout:\n\n"), poolname);
 
-		print_vdev_tree(NULL, poolname, nvroot, 0, B_FALSE, 0);
+		print_vdev_tree(NULL, poolname, nvroot, 0, "", 0);
-		if (num_logs(nvroot) > 0)
+		print_vdev_tree(NULL, "dedup", nvroot, 0,
-			print_vdev_tree(NULL, "logs", nvroot, 0, B_TRUE, 0);
+		    VDEV_ALLOC_BIAS_DEDUP, 0);
+		print_vdev_tree(NULL, "special", nvroot, 0,
+		    VDEV_ALLOC_BIAS_SPECIAL, 0);
+		print_vdev_tree(NULL, "logs", nvroot, 0,
+		    VDEV_ALLOC_BIAS_LOG, 0);
 
 		ret = 0;
 	} else {
@@ -1865,6 +1890,10 @@ print_status_config(zpool_handle_t *zhp, status_cbdata_t *cb, const char *name,
 		    &ishole);
 		if (islog || ishole)
 			continue;
+		/* Only print normal classes here */
+		if (nvlist_exists(child[c], ZPOOL_CONFIG_ALLOCATION_BIAS))
+			continue;
+
 		vname = zpool_vdev_name(g_zfs, zhp, child[c],
 		    cb->cb_name_flags | VDEV_NAME_TYPE_ID);
 		print_status_config(zhp, cb, vname, child[c], depth + 2,
@@ -1951,6 +1980,8 @@ print_import_config(status_cbdata_t *cb, const char *name, nvlist_t *nv,
 		    &is_log);
 		if (is_log)
 			continue;
+		if (nvlist_exists(child[c], ZPOOL_CONFIG_ALLOCATION_BIAS))
+			continue;
 
 		vname = zpool_vdev_name(g_zfs, NULL, child[c],
 		    cb->cb_name_flags | VDEV_NAME_TYPE_ID);
@@ -1982,34 +2013,56 @@ print_import_config(status_cbdata_t *cb, const char *name, nvlist_t *nv,
 }
 
 /*
- * Print log vdevs.
+ * Print specialized class vdevs.
- * Logs are recorded as top level vdevs in the main pool child array
+ *
- * but with "is_log" set to 1. We use either print_status_config() or
+ * These are recorded as top level vdevs in the main pool child array
- * print_import_config() to print the top level logs then any log
+ * but with "is_log" set to 1 or an "alloc_bias" string. We use either
- * children (eg mirrored slogs) are printed recursively - which
+ * print_status_config() or print_import_config() to print the top level
- * works because only the top level vdev is marked "is_log"
+ * class vdevs then any of their children (eg mirrored slogs) are printed
+ * recursively - which works because only the top level vdev is marked.
  */
 static void
-print_logs(zpool_handle_t *zhp, status_cbdata_t *cb, nvlist_t *nv)
+print_class_vdevs(zpool_handle_t *zhp, status_cbdata_t *cb, nvlist_t *nv,
+    const char *class)
 {
 	uint_t c, children;
 	nvlist_t **child;
+	boolean_t printed = B_FALSE;
+
+	assert(zhp != NULL || !cb->cb_verbose);
 
 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, &child,
 	    &children) != 0)
 		return;
 
-	(void) printf(gettext("\tlogs\n"));
-
 	for (c = 0; c < children; c++) {
 		uint64_t is_log = B_FALSE;
-		char *name;
+		char *bias = NULL;
+		char *type = NULL;
 
 		(void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 		    &is_log);
-		if (!is_log)
+
+		if (is_log) {
+			bias = VDEV_ALLOC_CLASS_LOGS;
+		} else {
+			(void) nvlist_lookup_string(child[c],
+			    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias);
+			(void) nvlist_lookup_string(child[c],
+			    ZPOOL_CONFIG_TYPE, &type);
+		}
+
+		if (bias == NULL || strcmp(bias, class) != 0)
 			continue;
-		name = zpool_vdev_name(g_zfs, zhp, child[c],
+		if (!is_log && strcmp(type, VDEV_TYPE_INDIRECT) == 0)
+			continue;
+
+		if (!printed) {
+			(void) printf("\t%s\t\n", gettext(class));
+			printed = B_TRUE;
+		}
+
+		char *name = zpool_vdev_name(g_zfs, zhp, child[c],
 		    cb->cb_name_flags | VDEV_NAME_TYPE_ID);
 		if (cb->cb_print_status)
 			print_status_config(zhp, cb, name, child[c], 2,
@@ -2313,8 +2366,10 @@ show_import(nvlist_t *config)
 		cb.cb_namewidth = 10;
 
 	print_import_config(&cb, name, nvroot, 0);
-	if (num_logs(nvroot) > 0)
+
-		print_logs(NULL, &cb, nvroot);
+	print_class_vdevs(NULL, &cb, nvroot, VDEV_ALLOC_BIAS_DEDUP);
+	print_class_vdevs(NULL, &cb, nvroot, VDEV_ALLOC_BIAS_SPECIAL);
+	print_class_vdevs(NULL, &cb, nvroot, VDEV_ALLOC_CLASS_LOGS);
 
 	if (reason == ZPOOL_STATUS_BAD_GUID_SUM) {
 		(void) printf(gettext("\n\tAdditional devices are known to "
@@ -3810,6 +3865,12 @@ print_iostat_default(vdev_stat_t *vs, iostat_cbdata_t *cb, double scale)
 	    format, column_width, cb->cb_scripted);
 }
 
+static const char *class_name[] = {
+	VDEV_ALLOC_BIAS_DEDUP,
+	VDEV_ALLOC_BIAS_SPECIAL,
+	VDEV_ALLOC_CLASS_LOGS
+};
+
 /*
  * Print out all the statistics for the given vdev.  This can either be the
  * toplevel configuration, or called recursively.  If 'name' is NULL, then this
@@ -3817,7 +3878,7 @@ print_iostat_default(vdev_stat_t *vs, iostat_cbdata_t *cb, double scale)
  *
  * Returns the number of stat lines printed.
  */
-unsigned int
+static unsigned int
 print_vdev_stats(zpool_handle_t *zhp, const char *name, nvlist_t *oldnv,
     nvlist_t *newnv, iostat_cbdata_t *cb, int depth)
 {
@@ -3945,6 +4006,9 @@ children:
 		children = MIN(oldchildren, children);
 	}
 
+	/*
+	 * print normal top-level devices
+	 */
 	for (c = 0; c < children; c++) {
 		uint64_t ishole = B_FALSE, islog = B_FALSE;
 
@@ -3957,6 +4021,9 @@ children:
 		if (ishole || islog)
 			continue;
 
+		if (nvlist_exists(newchild[c], ZPOOL_CONFIG_ALLOCATION_BIAS))
+			continue;
+
 		vname = zpool_vdev_name(g_zfs, zhp, newchild[c],
 		    cb->cb_name_flags);
 		ret += print_vdev_stats(zhp, vname, oldnv ? oldchild[c] : NULL,
@@ -3965,33 +4032,49 @@ children:
 	}
 
 	/*
-	 * Log device section
+	 * print all other top-level devices
 	 */
-
+	for (uint_t n = 0; n < 3; n++) {
-	if (num_logs(newnv) > 0) {
+		boolean_t printed = B_FALSE;
-		if ((!(cb->cb_flags & IOS_ANYHISTO_M)) && !cb->cb_scripted &&
-		    !cb->cb_vdev_names) {
-			print_iostat_dashes(cb, 0, "logs");
-		}
-		printf("\n");
 
 		for (c = 0; c < children; c++) {
 			uint64_t islog = B_FALSE;
+			char *bias = NULL;
+			char *type = NULL;
+
 			(void) nvlist_lookup_uint64(newchild[c],
 			    ZPOOL_CONFIG_IS_LOG, &islog);
-
 			if (islog) {
+				bias = VDEV_ALLOC_CLASS_LOGS;
+			} else {
+				(void) nvlist_lookup_string(newchild[c],
+				    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias);
+				(void) nvlist_lookup_string(newchild[c],
+				    ZPOOL_CONFIG_TYPE, &type);
+			}
+			if (bias == NULL || strcmp(bias, class_name[n]) != 0)
+				continue;
+			if (!islog && strcmp(type, VDEV_TYPE_INDIRECT) == 0)
+				continue;
+
+			if (!printed) {
+				if ((!(cb->cb_flags & IOS_ANYHISTO_M)) &&
+				    !cb->cb_scripted && !cb->cb_vdev_names) {
+					print_iostat_dashes(cb, 0,
+					    class_name[n]);
+				}
+				printf("\n");
+				printed = B_TRUE;
+			}
+
 			vname = zpool_vdev_name(g_zfs, zhp, newchild[c],
 			    cb->cb_name_flags);
 			ret += print_vdev_stats(zhp, vname, oldnv ?
-				    oldchild[c] : NULL, newchild[c],
+			    oldchild[c] : NULL, newchild[c], cb, depth + 2);
-				    cb, depth + 2);
 			free(vname);
 		}
 	}
 
-	}
-
 	/*
 	 * Include level 2 ARC devices in iostat output
 	 */
@@ -4937,6 +5020,7 @@ typedef struct list_cbdata {
 	boolean_t	cb_literal;
 } list_cbdata_t;
 
+
 /*
  * Given a list of columns to display, output appropriate headers for each one.
  */
@@ -4991,7 +5075,7 @@ print_header(list_cbdata_t *cb)
 
 /*
  * Given a pool and a list of properties, print out all the properties according
- * to the described layout.
+ * to the described layout. Used by zpool_do_list().
  */
 static void
 print_pool(zpool_handle_t *zhp, list_cbdata_t *cb)
@@ -5087,12 +5171,14 @@ print_one_column(zpool_prop_t prop, uint64_t value, boolean_t scripted,
 		}
 		break;
 	case ZPOOL_PROP_CAPACITY:
+		/* capacity value is in parts-per-10,000 (aka permyriad) */
 		if (format == ZFS_NICENUM_RAW)
 			(void) snprintf(propval, sizeof (propval), "%llu",
-			    (unsigned long long)value);
+			    (unsigned long long)value / 100);
 		else
-			(void) snprintf(propval, sizeof (propval), "%llu%%",
+			(void) snprintf(propval, sizeof (propval),
-			    (unsigned long long)value);
+			    value < 1000 ? "%1.2f%%" : value < 10000 ?
+			    "%2.1f%%" : "%3.0f%%", value / 100.0);
 		break;
 	default:
 		zfs_nicenum_format(value, propval, sizeof (propval), format);
@@ -5107,6 +5193,10 @@ print_one_column(zpool_prop_t prop, uint64_t value, boolean_t scripted,
 		(void) printf("  %*s", (int)width, propval);
 }
 
+/*
+ * print static default line per vdev
+ * not compatible with '-o' <proplist> option
+ */
 void
 print_list_stats(zpool_handle_t *zhp, const char *name, nvlist_t *nv,
     list_cbdata_t *cb, int depth)
@@ -5117,7 +5207,6 @@ print_list_stats(zpool_handle_t *zhp, const char *name, nvlist_t *nv,
 	char *vname;
 	boolean_t scripted = cb->cb_scripted;
 	uint64_t islog = B_FALSE;
-	boolean_t haslog = B_FALSE;
 	char *dashes = "%-*s      -      -      -         -      -      -\n";
 
 	verify(nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
@@ -5165,7 +5254,7 @@ print_list_stats(zpool_handle_t *zhp, const char *name, nvlist_t *nv,
 		    (vs->vs_fragmentation != ZFS_FRAG_INVALID && toplevel),
 		    format);
 		cap = (vs->vs_space == 0) ? 0 :
-		    (vs->vs_alloc * 100 / vs->vs_space);
+		    (vs->vs_alloc * 10000 / vs->vs_space);
 		print_one_column(ZPOOL_PROP_CAPACITY, cap, scripted, toplevel,
 		    format);
 		(void) printf("\n");
@@ -5175,6 +5264,7 @@ print_list_stats(zpool_handle_t *zhp, const char *name, nvlist_t *nv,
 	    &child, &children) != 0)
 		return;
 
+	/* list the normal vdevs first */
 	for (c = 0; c < children; c++) {
 		uint64_t ishole = B_FALSE;
 
@@ -5183,10 +5273,11 @@ print_list_stats(zpool_handle_t *zhp, const char *name, nvlist_t *nv,
 			continue;
 
 		if (nvlist_lookup_uint64(child[c],
-		    ZPOOL_CONFIG_IS_LOG, &islog) == 0 && islog) {
+		    ZPOOL_CONFIG_IS_LOG, &islog) == 0 && islog)
-			haslog = B_TRUE;
+			continue;
+
+		if (nvlist_exists(child[c], ZPOOL_CONFIG_ALLOCATION_BIAS))
 			continue;
-		}
 
 		vname = zpool_vdev_name(g_zfs, zhp, child[c],
 		    cb->cb_name_flags);
@@ -5194,13 +5285,34 @@ print_list_stats(zpool_handle_t *zhp, const char *name, nvlist_t *nv,
 		free(vname);
 	}
 
-	if (haslog == B_TRUE) {
+	/* list the classes: 'logs', 'dedup', and 'special' */
-		/* LINTED E_SEC_PRINTF_VAR_FMT */
+	for (uint_t n = 0; n < 3; n++) {
-		(void) printf(dashes, cb->cb_namewidth, "log");
+		boolean_t printed = B_FALSE;
+
 		for (c = 0; c < children; c++) {
+			char *bias = NULL;
+			char *type = NULL;
+
 			if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
-			    &islog) != 0 || !islog)
+			    &islog) == 0 && islog) {
+				bias = VDEV_ALLOC_CLASS_LOGS;
+			} else {
+				(void) nvlist_lookup_string(child[c],
+				    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias);
+				(void) nvlist_lookup_string(child[c],
+				    ZPOOL_CONFIG_TYPE, &type);
+			}
+			if (bias == NULL || strcmp(bias, class_name[n]) != 0)
 				continue;
+			if (!islog && strcmp(type, VDEV_TYPE_INDIRECT) == 0)
+				continue;
+
+			if (!printed) {
+				/* LINTED E_SEC_PRINTF_VAR_FMT */
+				(void) printf(dashes, cb->cb_namewidth,
+				    class_name[n]);
+				printed = B_TRUE;
+			}
 			vname = zpool_vdev_name(g_zfs, zhp, child[c],
 			    cb->cb_name_flags);
 			print_list_stats(zhp, vname, child[c], cb, depth + 2);
@@ -5233,7 +5345,6 @@ print_list_stats(zpool_handle_t *zhp, const char *name, nvlist_t *nv,
 	}
 }
 
-
 /*
  * Generic callback function to list a pool.
  */
@@ -5246,12 +5357,20 @@ list_callback(zpool_handle_t *zhp, void *data)
 
 	config = zpool_get_config(zhp, NULL);
 
-	print_pool(zhp, cbp);
+	if (cbp->cb_verbose) {
-	if (!cbp->cb_verbose)
+		config = zpool_get_config(zhp, NULL);
-		return (0);
 
 		verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
 		    &nvroot) == 0);
+	}
+
+	if (cbp->cb_verbose)
+		cbp->cb_namewidth = max_width(zhp, nvroot, 0, 0,
+		    cbp->cb_name_flags);
+
+	print_pool(zhp, cbp);
+
+	if (cbp->cb_verbose)
 		print_list_stats(zhp, NULL, nvroot, cbp, 0);
 
 	return (0);
@@ -5315,6 +5434,7 @@ zpool_do_list(int argc, char **argv)
 			break;
 		case 'v':
 			cb.cb_verbose = B_TRUE;
+			cb.cb_namewidth = 8;	/* 8 until precalc is avail */
 			break;
 		case ':':
 			(void) fprintf(stderr, gettext("missing argument for "
@@ -5716,7 +5836,7 @@ zpool_do_split(int argc, char **argv)
 		if (flags.dryrun) {
 			(void) printf(gettext("would create '%s' with the "
 			    "following layout:\n\n"), newpool);
-			print_vdev_tree(NULL, newpool, config, 0, B_FALSE,
+			print_vdev_tree(NULL, newpool, config, 0, "",
 			    flags.name_flags);
 		}
 	}
@@ -6965,11 +7085,14 @@ status_callback(zpool_handle_t *zhp, void *data)
 			print_cmd_columns(cbp->vcdl, 0);
 
 		printf("\n");
+
 		print_status_config(zhp, cbp, zpool_get_name(zhp), nvroot, 0,
 		    B_FALSE);
 
-		if (num_logs(nvroot) > 0)
+		print_class_vdevs(zhp, cbp, nvroot, VDEV_ALLOC_BIAS_DEDUP);
-			print_logs(zhp, cbp, nvroot);
+		print_class_vdevs(zhp, cbp, nvroot, VDEV_ALLOC_BIAS_SPECIAL);
+		print_class_vdevs(zhp, cbp, nvroot, VDEV_ALLOC_CLASS_LOGS);
+
 		if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
 		    &l2cache, &nl2cache) == 0)
 			print_l2cache(zhp, cbp, l2cache, nl2cache);

cmd/zpool/zpool_vdev.c

@@ -21,8 +21,8 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2013, 2018 by Delphix. All rights reserved.
- * Copyright (c) 2016 Intel Corporation.
+ * Copyright (c) 2016, 2017 Intel Corporation.
  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>.
  */
 
@@ -684,6 +684,9 @@ make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
 	verify(nvlist_add_string(vdev, ZPOOL_CONFIG_PATH, path) == 0);
 	verify(nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE, type) == 0);
 	verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_LOG, is_log) == 0);
+	if (is_log)
+		verify(nvlist_add_string(vdev, ZPOOL_CONFIG_ALLOCATION_BIAS,
+		    VDEV_ALLOC_BIAS_LOG) == 0);
 	if (strcmp(type, VDEV_TYPE_DISK) == 0)
 		verify(nvlist_add_uint64(vdev, ZPOOL_CONFIG_WHOLE_DISK,
 		    (uint64_t)wholedisk) == 0);
@@ -742,6 +745,9 @@ make_leaf_vdev(nvlist_t *props, const char *arg, uint64_t is_log)
  *
  * 	Otherwise, make sure that the current spec (if there is one) and the new
  * 	spec have consistent replication levels.
+ *
+ *	If there is no current spec (create), make sure new spec has at least
+ *	one general purpose vdev.
  */
 typedef struct replication_level {
 	char *zprl_type;
@@ -965,7 +971,7 @@ get_replication(nvlist_t *nvroot, boolean_t fatal)
 
 		/*
 		 * At this point, we have the replication of the last toplevel
-		 * vdev in 'rep'.  Compare it to 'lastrep' to see if its
+		 * vdev in 'rep'.  Compare it to 'lastrep' to see if it is
 		 * different.
 		 */
 		if (lastrep.zprl_type != NULL) {
@@ -1466,6 +1472,13 @@ is_grouping(const char *type, int *mindev, int *maxdev)
 		return (VDEV_TYPE_LOG);
 	}
 
+	if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0 ||
+	    strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) {
+		if (mindev != NULL)
+			*mindev = 1;
+		return (type);
+	}
+
 	if (strcmp(type, "cache") == 0) {
 		if (mindev != NULL)
 			*mindev = 1;
@@ -1487,7 +1500,7 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 	nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
 	int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
 	const char *type;
-	uint64_t is_log;
+	uint64_t is_log, is_special, is_dedup;
 	boolean_t seen_logs;
 
 	top = NULL;
@@ -1497,7 +1510,7 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 	nspares = 0;
 	nlogs = 0;
 	nl2cache = 0;
-	is_log = B_FALSE;
+	is_log = is_special = is_dedup = B_FALSE;
 	seen_logs = B_FALSE;
 	nvroot = NULL;
 
@@ -1520,7 +1533,7 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 					    "specified only once\n"));
 					goto spec_out;
 				}
-				is_log = B_FALSE;
+				is_log = is_special = is_dedup = B_FALSE;
 			}
 
 			if (strcmp(type, VDEV_TYPE_LOG) == 0) {
@@ -1533,6 +1546,8 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 				}
 				seen_logs = B_TRUE;
 				is_log = B_TRUE;
+				is_special = B_FALSE;
+				is_dedup = B_FALSE;
 				argc--;
 				argv++;
 				/*
@@ -1542,6 +1557,24 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 				continue;
 			}
 
+			if (strcmp(type, VDEV_ALLOC_BIAS_SPECIAL) == 0) {
+				is_special = B_TRUE;
+				is_log = B_FALSE;
+				is_dedup = B_FALSE;
+				argc--;
+				argv++;
+				continue;
+			}
+
+			if (strcmp(type, VDEV_ALLOC_BIAS_DEDUP) == 0) {
+				is_dedup = B_TRUE;
+				is_log = B_FALSE;
+				is_special = B_FALSE;
+				argc--;
+				argv++;
+				continue;
+			}
+
 			if (strcmp(type, VDEV_TYPE_L2CACHE) == 0) {
 				if (l2cache != NULL) {
 					(void) fprintf(stderr,
@@ -1550,15 +1583,16 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 					    "specified only once\n"));
 					goto spec_out;
 				}
-				is_log = B_FALSE;
+				is_log = is_special = is_dedup = B_FALSE;
 			}
 
-			if (is_log) {
+			if (is_log || is_special || is_dedup) {
 				if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
 					(void) fprintf(stderr,
 					    gettext("invalid vdev "
-					    "specification: unsupported 'log' "
+					    "specification: unsupported '%s' "
-					    "device: %s\n"), type);
+					    "device: %s\n"), is_log ? "log" :
+					    "special", type);
 					goto spec_out;
 				}
 				nlogs++;
@@ -1615,12 +1649,27 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 				nl2cache = children;
 				continue;
 			} else {
+				/* create a top-level vdev with children */
 				verify(nvlist_alloc(&nv, NV_UNIQUE_NAME,
 				    0) == 0);
 				verify(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
 				    type) == 0);
 				verify(nvlist_add_uint64(nv,
 				    ZPOOL_CONFIG_IS_LOG, is_log) == 0);
+				if (is_log)
+					verify(nvlist_add_string(nv,
+					    ZPOOL_CONFIG_ALLOCATION_BIAS,
+					    VDEV_ALLOC_BIAS_LOG) == 0);
+				if (is_special) {
+					verify(nvlist_add_string(nv,
+					    ZPOOL_CONFIG_ALLOCATION_BIAS,
+					    VDEV_ALLOC_BIAS_SPECIAL) == 0);
+				}
+				if (is_dedup) {
+					verify(nvlist_add_string(nv,
+					    ZPOOL_CONFIG_ALLOCATION_BIAS,
+					    VDEV_ALLOC_BIAS_DEDUP) == 0);
+				}
 				if (strcmp(type, VDEV_TYPE_RAIDZ) == 0) {
 					verify(nvlist_add_uint64(nv,
 					    ZPOOL_CONFIG_NPARITY,
@@ -1645,6 +1694,16 @@ construct_spec(nvlist_t *props, int argc, char **argv)
 
 			if (is_log)
 				nlogs++;
+			if (is_special) {
+				verify(nvlist_add_string(nv,
+				    ZPOOL_CONFIG_ALLOCATION_BIAS,
+				    VDEV_ALLOC_BIAS_SPECIAL) == 0);
+			}
+			if (is_dedup) {
+				verify(nvlist_add_string(nv,
+				    ZPOOL_CONFIG_ALLOCATION_BIAS,
+				    VDEV_ALLOC_BIAS_DEDUP) == 0);
+			}
 			argc--;
 			argv++;
 		}
@@ -1745,6 +1804,30 @@ split_mirror_vdev(zpool_handle_t *zhp, char *newname, nvlist_t *props,
 	return (newroot);
 }
 
+static int
+num_normal_vdevs(nvlist_t *nvroot)
+{
+	nvlist_t **top;
+	uint_t t, toplevels, normal = 0;
+
+	verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
+	    &top, &toplevels) == 0);
+
+	for (t = 0; t < toplevels; t++) {
+		uint64_t log = B_FALSE;
+
+		(void) nvlist_lookup_uint64(top[t], ZPOOL_CONFIG_IS_LOG, &log);
+		if (log)
+			continue;
+		if (nvlist_exists(top[t], ZPOOL_CONFIG_ALLOCATION_BIAS))
+			continue;
+
+		normal++;
+	}
+
+	return (normal);
+}
+
 /*
  * Get and validate the contents of the given vdev specification.  This ensures
  * that the nvlist returned is well-formed, that all the devices exist, and that
@@ -1797,6 +1880,16 @@ make_root_vdev(zpool_handle_t *zhp, nvlist_t *props, int force, int check_rep,
 		return (NULL);
 	}
 
+	/*
+	 * On pool create the new vdev spec must have one normal vdev.
+	 */
+	if (poolconfig == NULL && num_normal_vdevs(newroot) == 0) {
+		vdev_error(gettext("at least one general top-level vdev must "
+		    "be specified\n"));
+		nvlist_free(newroot);
+		return (NULL);
+	}
+
 	/*
 	 * Run through the vdev specification and label any whole disks found.
 	 */

cmd/ztest/ztest.c

@@ -20,11 +20,12 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  * Copyright (c) 2014 Integros [integros.com]
  * Copyright 2017 Joyent, Inc.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 /*
@@ -149,6 +150,12 @@ typedef struct ztest_shared_hdr {
 
 static ztest_shared_hdr_t *ztest_shared_hdr;
 
+enum ztest_class_state {
+	ZTEST_VDEV_CLASS_OFF,
+	ZTEST_VDEV_CLASS_ON,
+	ZTEST_VDEV_CLASS_RND
+};
+
 typedef struct ztest_shared_opts {
 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
@@ -171,6 +178,7 @@ typedef struct ztest_shared_opts {
 	uint64_t zo_maxloops;
 	uint64_t zo_metaslab_force_ganging;
 	int zo_mmp_test;
+	int zo_special_vdevs;
 } ztest_shared_opts_t;
 
 static const ztest_shared_opts_t ztest_opts_defaults = {
@@ -194,6 +202,7 @@ static const ztest_shared_opts_t ztest_opts_defaults = {
 	.zo_time = 300,			/* 5 minutes */
 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
 	.zo_metaslab_force_ganging = 32 << 10,
+	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
 };
 
 extern uint64_t metaslab_force_ganging;
@@ -342,6 +351,7 @@ ztest_func_t ztest_dsl_dataset_promote_busy;
 ztest_func_t ztest_vdev_attach_detach;
 ztest_func_t ztest_vdev_LUN_growth;
 ztest_func_t ztest_vdev_add_remove;
+ztest_func_t ztest_vdev_class_add;
 ztest_func_t ztest_vdev_aux_add_remove;
 ztest_func_t ztest_split_pool;
 ztest_func_t ztest_reguid;
@@ -398,6 +408,7 @@ ztest_info_t ztest_info[] = {
 	ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
 	ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
+	ZTI_INIT(ztest_vdev_class_add, 1, &ztest_opts.zo_vdevtime),
 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
 	ZTI_INIT(ztest_remap_blocks, 1, &zopt_sometimes),
@@ -666,6 +677,7 @@ usage(boolean_t requested)
 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
+	    "\t[-C vdev class state (default: random)] special=on|off|random\n"
 	    "\t[-o variable=value] ... set global variable to an unsigned\n"
 	    "\t    32-bit integer value\n"
 	    "\t[-G dump zfs_dbgmsg buffer before exiting due to an error\n"
@@ -691,6 +703,46 @@ usage(boolean_t requested)
 	exit(requested ? 0 : 1);
 }
 
+
+static void
+ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
+{
+	char name[32];
+	char *value;
+	int state = ZTEST_VDEV_CLASS_RND;
+
+	(void) strlcpy(name, input, sizeof (name));
+
+	value = strchr(name, '=');
+	if (value == NULL) {
+		(void) fprintf(stderr, "missing value in property=value "
+		    "'-C' argument (%s)\n", input);
+		usage(B_FALSE);
+	}
+	*(value) = '\0';
+	value++;
+
+	if (strcmp(value, "on") == 0) {
+		state = ZTEST_VDEV_CLASS_ON;
+	} else if (strcmp(value, "off") == 0) {
+		state = ZTEST_VDEV_CLASS_OFF;
+	} else if (strcmp(value, "random") == 0) {
+		state = ZTEST_VDEV_CLASS_RND;
+	} else {
+		(void) fprintf(stderr, "invalid property value '%s'\n", value);
+		usage(B_FALSE);
+	}
+
+	if (strcmp(name, "special") == 0) {
+		zo->zo_special_vdevs = state;
+	} else {
+		(void) fprintf(stderr, "invalid property name '%s'\n", name);
+		usage(B_FALSE);
+	}
+	if (zo->zo_verbose >= 3)
+		(void) printf("%s vdev state is '%s'\n", name, value);
+}
+
 static void
 process_options(int argc, char **argv)
 {
@@ -704,7 +756,7 @@ process_options(int argc, char **argv)
 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
 
 	while ((opt = getopt(argc, argv,
-	    "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:o:G")) != EOF) {
+	    "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:C:o:G")) != EOF) {
 		value = 0;
 		switch (opt) {
 		case 'v':
@@ -795,6 +847,9 @@ process_options(int argc, char **argv)
 		case 'B':
 			(void) strlcpy(altdir, optarg, sizeof (altdir));
 			break;
+		case 'C':
+			ztest_parse_name_value(optarg, zo);
+			break;
 		case 'o':
 			if (set_global_var(optarg) != 0)
 				usage(B_FALSE);
@@ -1022,13 +1077,16 @@ make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
 
 static nvlist_t *
 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
-    int log, int r, int m, int t)
+    const char *class, int r, int m, int t)
 {
 	nvlist_t *root, **child;
 	int c;
+	boolean_t log;
 
 	ASSERT(t > 0);
 
+	log = (class != NULL && strcmp(class, "log") == 0);
+
 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
 
 	for (c = 0; c < t; c++) {
@@ -1036,6 +1094,12 @@ make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
 		    r, m);
 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
 		    log) == 0);
+
+		if (class != NULL && class[0] != '\0') {
+			ASSERT(m > 1 || log);   /* expecting a mirror */
+			VERIFY(nvlist_add_string(child[c],
+			    ZPOOL_CONFIG_ALLOCATION_BIAS, class) == 0);
+		}
 	}
 
 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
@@ -1075,6 +1139,8 @@ ztest_random_spa_version(uint64_t initial_version)
 static int
 ztest_random_blocksize(void)
 {
+	ASSERT(ztest_spa->spa_max_ashift != 0);
+
 	/*
 	 * Choose a block size >= the ashift.
 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
@@ -2722,7 +2788,7 @@ ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
 	/*
 	 * Attempt to create using a bad file.
 	 */
-	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
+	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
 	VERIFY3U(ENOENT, ==,
 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
 	nvlist_free(nvroot);
@@ -2730,7 +2796,7 @@ ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
 	/*
 	 * Attempt to create using a bad mirror.
 	 */
-	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
+	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
 	VERIFY3U(ENOENT, ==,
 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
 	nvlist_free(nvroot);
@@ -2740,7 +2806,7 @@ ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
 	 * what's in the nvroot; we should fail with EEXIST.
 	 */
 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
-	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
+	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
 	VERIFY3U(EEXIST, ==,
 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
 	nvlist_free(nvroot);
@@ -2816,7 +2882,7 @@ ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
 	(void) spa_destroy(name);
 
 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
-	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
+	    NULL, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
 
 	/*
 	 * If we're configuring a RAIDZ device then make sure that the
@@ -2990,10 +3056,16 @@ ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
 	 * If we have slogs then remove them 1/4 of the time.
 	 */
 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
+		metaslab_group_t *mg;
+
 		/*
-		 * Grab the guid from the head of the log class rotor.
+		 * find the first real slog in log allocation class
 		 */
-		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
+		mg =  spa_log_class(spa)->mc_rotor;
+		while (!mg->mg_vd->vdev_islog)
+			mg = mg->mg_next;
+
+		guid = mg->mg_vd->vdev_guid;
 
 		spa_config_exit(spa, SCL_VDEV, FTAG);
 
@@ -3024,12 +3096,11 @@ ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
 		spa_config_exit(spa, SCL_VDEV, FTAG);
 
 		/*
-		 * Make 1/4 of the devices be log devices.
+		 * Make 1/4 of the devices be log devices
 		 */
 		nvroot = make_vdev_root(NULL, NULL, NULL,
-		    ztest_opts.zo_vdev_size, 0,
+		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
-		    ztest_random(4) == 0, ztest_opts.zo_raidz,
+		    "log" : NULL, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
-		    zs->zs_mirrors, 1);
 
 		error = spa_vdev_add(spa, nvroot);
 		nvlist_free(nvroot);
@@ -3048,6 +3119,83 @@ ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
 	mutex_exit(&ztest_vdev_lock);
 }
 
+/* ARGSUSED */
+void
+ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
+{
+	ztest_shared_t *zs = ztest_shared;
+	spa_t *spa = ztest_spa;
+	uint64_t leaves;
+	nvlist_t *nvroot;
+	const char *class = (ztest_random(2) == 0) ?
+	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
+	int error;
+
+	/*
+	 * By default add a special vdev 50% of the time
+	 */
+	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
+	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
+	    ztest_random(2) == 0)) {
+		return;
+	}
+
+	mutex_enter(&ztest_vdev_lock);
+
+	/* Only test with mirrors */
+	if (zs->zs_mirrors < 2) {
+		mutex_exit(&ztest_vdev_lock);
+		return;
+	}
+
+	/* requires feature@allocation_classes */
+	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
+		mutex_exit(&ztest_vdev_lock);
+		return;
+	}
+
+	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
+
+	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
+	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
+	spa_config_exit(spa, SCL_VDEV, FTAG);
+
+	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
+	    class, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
+
+	error = spa_vdev_add(spa, nvroot);
+	nvlist_free(nvroot);
+
+	if (error == ENOSPC)
+		ztest_record_enospc("spa_vdev_add");
+	else if (error != 0)
+		fatal(0, "spa_vdev_add() = %d", error);
+
+	/*
+	 * 50% of the time allow small blocks in the special class
+	 */
+	if (error == 0 &&
+	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
+		if (ztest_opts.zo_verbose >= 3)
+			(void) printf("Enabling special VDEV small blocks\n");
+		(void) ztest_dsl_prop_set_uint64(zd->zd_name,
+		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
+	}
+
+	mutex_exit(&ztest_vdev_lock);
+
+	if (ztest_opts.zo_verbose >= 3) {
+		metaslab_class_t *mc;
+
+		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
+			mc = spa_special_class(spa);
+		else
+			mc = spa_dedup_class(spa);
+		(void) printf("Added a %s mirrored vdev (of %d)\n",
+		    class, (int)mc->mc_groups);
+	}
+}
+
 /*
  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
  */
@@ -3114,7 +3262,7 @@ ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
 		 * Add a new device.
 		 */
 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
-		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
+		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
 		error = spa_vdev_add(spa, nvroot);
 
 		switch (error) {
@@ -3316,11 +3464,15 @@ ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
 	 * Locate this vdev.
 	 */
 	oldvd = rvd->vdev_child[top];
+
+	/* pick a child from the mirror */
 	if (zs->zs_mirrors >= 1) {
 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
 	}
+
+	/* pick a child out of the raidz group */
 	if (ztest_opts.zo_raidz > 1) {
 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
@@ -3422,7 +3574,7 @@ ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
 	 * Build the nvlist describing newpath.
 	 */
 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
-	    ashift, 0, 0, 0, 1);
+	    ashift, NULL, 0, 0, 1);
 
 	error = spa_vdev_attach(spa, oldguid, root, replacing);
 
@@ -3688,7 +3840,7 @@ ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
 		return;
 	}
 	ASSERT(psize > 0);
-	newsize = psize + psize / 8;
+	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
 	ASSERT3U(newsize, >, psize);
 
 	if (ztest_opts.zo_verbose >= 6) {
@@ -7027,6 +7179,7 @@ make_random_props(void)
 	nvlist_t *props;
 
 	VERIFY0(nvlist_alloc(&props, NV_UNIQUE_NAME, 0));
+
 	if (ztest_random(2) == 0)
 		return (props);
 
@@ -7113,7 +7266,7 @@ ztest_init(ztest_shared_t *zs)
 	zs->zs_splits = 0;
 	zs->zs_mirrors = ztest_opts.zo_mirrors;
 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
-	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
+	    NULL, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
 	props = make_random_props();
 
 	/*

configure.ac

@@ -319,6 +319,7 @@ AC_CONFIG_FILES([
 	tests/zfs-tests/tests/functional/snapshot/Makefile
 	tests/zfs-tests/tests/functional/snapused/Makefile
 	tests/zfs-tests/tests/functional/sparse/Makefile
+	tests/zfs-tests/tests/functional/alloc_class/Makefile
 	tests/zfs-tests/tests/functional/threadsappend/Makefile
 	tests/zfs-tests/tests/functional/tmpfile/Makefile
 	tests/zfs-tests/tests/functional/truncate/Makefile

include/sys/dmu.h

@@ -20,12 +20,13 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright 2014 HybridCluster. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  * Copyright 2013 Saso Kiselkov. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -136,6 +137,16 @@ typedef enum dmu_object_byteswap {
 	((ot) & DMU_OT_METADATA) : \
 	DMU_OT_IS_METADATA_IMPL(ot))
 
+#define	DMU_OT_IS_DDT(ot) \
+	((ot) == DMU_OT_DDT_ZAP)
+
+#define	DMU_OT_IS_ZIL(ot) \
+	((ot) == DMU_OT_INTENT_LOG)
+
+/* Note: ztest uses DMU_OT_UINT64_OTHER as a proxy for file blocks */
+#define	DMU_OT_IS_FILE(ot) \
+	((ot) == DMU_OT_PLAIN_FILE_CONTENTS || (ot) == DMU_OT_UINT64_OTHER)
+
 #define	DMU_OT_IS_ENCRYPTED(ot) (((ot) & DMU_OT_NEWTYPE) ? \
 	((ot) & DMU_OT_ENCRYPTED) : \
 	DMU_OT_IS_ENCRYPTED_IMPL(ot))
@@ -226,7 +237,7 @@ typedef enum dmu_object_type {
 	 * values.
 	 *
 	 * The DMU_OTN_* types do not have entries in the dmu_ot table,
-	 * use the DMU_OT_IS_METDATA() and DMU_OT_BYTESWAP() macros instead
+	 * use the DMU_OT_IS_METADATA() and DMU_OT_BYTESWAP() macros instead
 	 * of indexing into dmu_ot directly (this works for both DMU_OT_* types
 	 * and DMU_OTN_* types).
 	 */

include/sys/dmu_objset.h

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  */
@@ -137,6 +137,11 @@ struct objset {
 	uint64_t os_normalization;
 	uint64_t os_utf8only;
 	uint64_t os_casesensitivity;
+	/*
+	 * The largest zpl file block allowed in special class.
+	 * cached here instead of zfsvfs for easier access.
+	 */
+	int os_zpl_special_smallblock;
 
 	/*
 	 * Pointer is constant; the blkptr it points to is protected by

include/sys/fs/zfs.h

@@ -26,6 +26,7 @@
  * Copyright (c) 2013, 2017 Joyent, Inc. All rights reserved.
  * Copyright (c) 2014 Integros [integros.com]
  * Copyright (c) 2017 Datto Inc.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 /* Portions Copyright 2010 Robert Milkowski */
@@ -181,6 +182,7 @@ typedef enum {
 	ZFS_PROP_KEY_GUID,
 	ZFS_PROP_KEYSTATUS,
 	ZFS_PROP_REMAPTXG,		/* not exposed to the user */
+	ZFS_PROP_SPECIAL_SMALL_BLOCKS,
 	ZFS_NUM_PROPS
 } zfs_prop_t;
 
@@ -713,6 +715,7 @@ typedef struct zpool_load_policy {
 #define	ZPOOL_CONFIG_MMP_TXG		"mmp_txg"	/* not stored on disk */
 #define	ZPOOL_CONFIG_MMP_HOSTNAME	"mmp_hostname"	/* not stored on disk */
 #define	ZPOOL_CONFIG_MMP_HOSTID		"mmp_hostid"	/* not stored on disk */
+#define	ZPOOL_CONFIG_ALLOCATION_BIAS	"alloc_bias"	/* not stored on disk */
 
 /*
  * The persistent vdev state is stored as separate values rather than a single
@@ -759,6 +762,14 @@ typedef struct zpool_load_policy {
 #define	VDEV_TOP_ZAP_POOL_CHECKPOINT_SM \
 	"com.delphix:pool_checkpoint_sm"
 
+#define	VDEV_TOP_ZAP_ALLOCATION_BIAS \
+	"org.zfsonlinux:allocation_bias"
+
+/* vdev metaslab allocation bias */
+#define	VDEV_ALLOC_BIAS_LOG		"log"
+#define	VDEV_ALLOC_BIAS_SPECIAL		"special"
+#define	VDEV_ALLOC_BIAS_DEDUP		"dedup"
+
 /*
  * This is needed in userland to report the minimum necessary device size.
  */

include/sys/metaslab.h

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef _SYS_METASLAB_H
@@ -57,13 +58,17 @@ void metaslab_sync_done(metaslab_t *, uint64_t);
 void metaslab_sync_reassess(metaslab_group_t *);
 uint64_t metaslab_block_maxsize(metaslab_t *);
 
+/*
+ * metaslab alloc flags
+ */
 #define	METASLAB_HINTBP_FAVOR		0x0
 #define	METASLAB_HINTBP_AVOID		0x1
 #define	METASLAB_GANG_HEADER		0x2
 #define	METASLAB_GANG_CHILD		0x4
 #define	METASLAB_ASYNC_ALLOC		0x8
 #define	METASLAB_DONT_THROTTLE		0x10
-#define	METASLAB_FASTWRITE		0x20
+#define	METASLAB_MUST_RESERVE		0x20
+#define	METASLAB_FASTWRITE		0x40
 
 int metaslab_alloc(spa_t *, metaslab_class_t *, uint64_t,
     blkptr_t *, int, uint64_t, blkptr_t *, int, zio_alloc_list_t *, zio_t *,
@@ -96,8 +101,6 @@ boolean_t metaslab_class_throttle_reserve(metaslab_class_t *, int, int,
     zio_t *, int);
 void metaslab_class_throttle_unreserve(metaslab_class_t *, int, int, zio_t *);
 
-void metaslab_class_space_update(metaslab_class_t *, int64_t, int64_t,
-    int64_t, int64_t);
 uint64_t metaslab_class_get_alloc(metaslab_class_t *);
 uint64_t metaslab_class_get_space(metaslab_class_t *);
 uint64_t metaslab_class_get_dspace(metaslab_class_t *);

include/sys/spa.h

@@ -20,13 +20,14 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  * Copyright 2013 Saso Kiselkov. All rights reserved.
  * Copyright (c) 2014 Integros [integros.com]
  * Copyright 2017 Joyent, Inc.
  * Copyright (c) 2017 Datto Inc.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef _SYS_SPA_H
@@ -976,6 +977,11 @@ extern uint64_t spa_version(spa_t *spa);
 extern boolean_t spa_deflate(spa_t *spa);
 extern metaslab_class_t *spa_normal_class(spa_t *spa);
 extern metaslab_class_t *spa_log_class(spa_t *spa);
+extern metaslab_class_t *spa_special_class(spa_t *spa);
+extern metaslab_class_t *spa_dedup_class(spa_t *spa);
+extern metaslab_class_t *spa_preferred_class(spa_t *spa, uint64_t size,
+    dmu_object_type_t objtype, uint_t level, uint_t special_smallblk);
+
 extern void spa_evicting_os_register(spa_t *, objset_t *os);
 extern void spa_evicting_os_deregister(spa_t *, objset_t *os);
 extern void spa_evicting_os_wait(spa_t *spa);
@@ -1040,6 +1046,7 @@ extern void spa_set_missing_tvds(spa_t *spa, uint64_t missing);
 extern boolean_t spa_top_vdevs_spacemap_addressable(spa_t *spa);
 extern boolean_t spa_multihost(spa_t *spa);
 extern unsigned long spa_get_hostid(void);
+extern void spa_activate_allocation_classes(spa_t *, dmu_tx_t *);
 
 extern int spa_mode(spa_t *spa);
 extern uint64_t zfs_strtonum(const char *str, char **nptr);

include/sys/spa_impl.h

@@ -20,12 +20,13 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  * Copyright 2013 Saso Kiselkov. All rights reserved.
  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
  * Copyright (c) 2017 Datto Inc.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef _SYS_SPA_IMPL_H
@@ -220,6 +221,8 @@ struct spa {
 	boolean_t	spa_is_initializing;	/* true while opening pool */
 	metaslab_class_t *spa_normal_class;	/* normal data class */
 	metaslab_class_t *spa_log_class;	/* intent log data class */
+	metaslab_class_t *spa_special_class;	/* special allocation class */
+	metaslab_class_t *spa_dedup_class;	/* dedup allocation class */
 	uint64_t	spa_first_txg;		/* first txg after spa_open() */
 	uint64_t	spa_final_txg;		/* txg of export/destroy */
 	uint64_t	spa_freeze_txg;		/* freeze pool at this txg */

include/sys/vdev.h

@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef _SYS_VDEV_H
@@ -108,6 +109,8 @@ extern boolean_t vdev_children_are_offline(vdev_t *vd);
 extern void vdev_space_update(vdev_t *vd,
     int64_t alloc_delta, int64_t defer_delta, int64_t space_delta);
 
+extern int64_t vdev_deflated_space(vdev_t *vd, int64_t space);
+
 extern uint64_t vdev_psize_to_asize(vdev_t *vd, uint64_t psize);
 
 extern int vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux);

include/sys/vdev_impl.h

@@ -21,6 +21,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef _SYS_VDEV_IMPL_H
@@ -140,6 +141,14 @@ struct vdev_queue {
 	kmutex_t	vq_lock;
 };
 
+typedef enum vdev_alloc_bias {
+	VDEV_BIAS_NONE,
+	VDEV_BIAS_LOG,		/* dedicated to ZIL data (SLOG) */
+	VDEV_BIAS_SPECIAL,	/* dedicated to ddt, metadata, and small blks */
+	VDEV_BIAS_DEDUP		/* dedicated to dedup metadata */
+} vdev_alloc_bias_t;
+
+
 /*
  * On-disk indirect vdev state.
  *
@@ -236,6 +245,7 @@ struct vdev {
 	boolean_t	vdev_ishole;	/* is a hole in the namespace	*/
 	kmutex_t	vdev_queue_lock; /* protects vdev_queue_depth	*/
 	uint64_t	vdev_top_zap;
+	vdev_alloc_bias_t vdev_alloc_bias; /* metaslab allocation bias	*/
 
 	/* pool checkpoint related */
 	space_map_t	*vdev_checkpoint_sm;	/* contains reserved blocks */

include/sys/zio.h

@@ -353,6 +353,7 @@ typedef struct zio_prop {
 	uint8_t			zp_salt[ZIO_DATA_SALT_LEN];
 	uint8_t			zp_iv[ZIO_DATA_IV_LEN];
 	uint8_t			zp_mac[ZIO_DATA_MAC_LEN];
+	uint32_t		zp_zpl_smallblk;
 } zio_prop_t;
 
 typedef struct zio_cksum_report zio_cksum_report_t;
@@ -473,6 +474,7 @@ struct zio {
 	vdev_t		*io_vd;
 	void		*io_vsd;
 	const zio_vsd_ops_t *io_vsd_ops;
+	metaslab_class_t *io_metaslab_class;	/* dva throttle class */
 
 	uint64_t	io_offset;
 	hrtime_t	io_timestamp;	/* submitted at */

include/zfeature_common.h

@@ -23,6 +23,7 @@
  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef _ZFEATURE_COMMON_H
@@ -63,6 +64,7 @@ typedef enum spa_feature {
 	SPA_FEATURE_OBSOLETE_COUNTS,
 	SPA_FEATURE_POOL_CHECKPOINT,
 	SPA_FEATURE_SPACEMAP_V2,
+	SPA_FEATURE_ALLOCATION_CLASSES,
 	SPA_FEATURES
 } spa_feature_t;
 

lib/libzfs/libzfs_dataset.c

@@ -1195,6 +1195,36 @@ zfs_valid_proplist(libzfs_handle_t *hdl, zfs_type_t type, nvlist_t *nvl,
 			}
 			break;
 		}
+
+		case ZFS_PROP_SPECIAL_SMALL_BLOCKS:
+			if (zpool_hdl != NULL) {
+				char state[64] = "";
+
+				/*
+				 * Issue a warning but do not fail so that
+				 * tests for setable properties succeed.
+				 */
+				if (zpool_prop_get_feature(zpool_hdl,
+				    "feature@allocation_classes", state,
+				    sizeof (state)) != 0 ||
+				    strcmp(state, ZFS_FEATURE_ACTIVE) != 0) {
+					(void) fprintf(stderr, gettext(
+					    "%s: property requires a special "
+					    "device in the pool\n"), propname);
+				}
+			}
+			if (intval != 0 &&
+			    (intval < SPA_MINBLOCKSIZE ||
+			    intval > SPA_OLD_MAXBLOCKSIZE || !ISP2(intval))) {
+				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
+				    "invalid '%s=%d' property: must be zero or "
+				    "a power of 2 from 512B to 128K"), propname,
+				    intval);
+				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
+				goto error;
+			}
+			break;
+
 		case ZFS_PROP_MLSLABEL:
 		{
 #ifdef HAVE_MLSLABEL

lib/libzfs/libzfs_pool.c

@@ -26,6 +26,7 @@
  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
  * Copyright (c) 2018 Datto Inc.
  * Copyright (c) 2017 Open-E, Inc. All Rights Reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #include <ctype.h>
@@ -720,6 +721,7 @@ zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 				goto error;
 			}
 			break;
+
 		default:
 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 			    "property '%s'(%d) not defined"), propname, prop);
@@ -1172,6 +1174,30 @@ zpool_get_state(zpool_handle_t *zhp)
 	return (zhp->zpool_state);
 }
 
+/*
+ * Check if vdev list contains a special vdev
+ */
+static boolean_t
+zpool_has_special_vdev(nvlist_t *nvroot)
+{
+	nvlist_t **child;
+	uint_t children;
+
+	if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &child,
+	    &children) == 0) {
+		for (uint_t c = 0; c < children; c++) {
+			char *bias;
+
+			if (nvlist_lookup_string(child[c],
+			    ZPOOL_CONFIG_ALLOCATION_BIAS, &bias) == 0 &&
+			    strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) {
+				return (B_TRUE);
+			}
+		}
+	}
+	return (B_FALSE);
+}
+
 /*
  * Create the named pool, using the provided vdev list.  It is assumed
  * that the consumer has already validated the contents of the nvlist, so we
@@ -1220,6 +1246,17 @@ zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
 		    fsprops, zoned, NULL, NULL, B_TRUE, msg)) == NULL) {
 			goto create_failed;
 		}
+
+		if (nvlist_exists(zc_fsprops,
+		    zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS)) &&
+		    !zpool_has_special_vdev(nvroot)) {
+			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
+			    "%s property requires a special vdev"),
+			    zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS));
+			(void) zfs_error(hdl, EZFS_BADPROP, msg);
+			goto create_failed;
+		}
+
 		if (!zc_props &&
 		    (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
 			goto create_failed;

lib/libzpool/util.c

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2016 by Delphix. All rights reserved.
  * Copyright 2017 Jason King
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #include <assert.h>
@@ -150,7 +151,6 @@ show_vdev_stats(const char *desc, const char *ctype, nvlist_t *nv, int indent)
 	uint_t c, children;
 	char used[6], avail[6];
 	char rops[6], wops[6], rbytes[6], wbytes[6], rerr[6], werr[6], cerr[6];
-	char *prefix = "";
 
 	v0 = umem_zalloc(sizeof (*v0), UMEM_NOFAIL);
 
@@ -162,15 +162,24 @@ show_vdev_stats(const char *desc, const char *ctype, nvlist_t *nv, int indent)
 	}
 
 	if (desc != NULL) {
+		char *suffix = "", *bias = NULL;
+		char bias_suffix[32];
+
 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &is_log);
-
+		(void) nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
-		if (is_log)
+		    &bias);
-			prefix = "log ";
-
 		if (nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
 		    (uint64_t **)&vs, &c) != 0)
 			vs = v0;
 
+		if (bias != NULL) {
+			(void) snprintf(bias_suffix, sizeof (bias_suffix),
+			    " (%s)", bias);
+			suffix = bias_suffix;
+		} else if (is_log) {
+			suffix = " (log)";
+		}
+
 		sec = MAX(1, vs->vs_timestamp / NANOSEC);
 
 		nicenum(vs->vs_alloc, used, sizeof (used));
@@ -187,9 +196,9 @@ show_vdev_stats(const char *desc, const char *ctype, nvlist_t *nv, int indent)
 
 		(void) printf("%*s%s%*s%*s%*s %5s %5s %5s %5s %5s %5s %5s\n",
 		    indent, "",
-		    prefix,
-		    (int)(indent+strlen(prefix)-25-(vs->vs_space ? 0 : 12)),
 		    desc,
+		    (int)(indent+strlen(desc)-25-(vs->vs_space ? 0 : 12)),
+		    suffix,
 		    vs->vs_space ? 6 : 0, vs->vs_space ? used : "",
 		    vs->vs_space ? 6 : 0, vs->vs_space ? avail : "",
 		    rops, wops, rbytes, wbytes, rerr, werr, cerr);

man/man5/zfs-module-parameters.5

@@ -1719,6 +1719,29 @@ otherwise would under the old \fBzfs_mg_alloc_failures\fR facility.
 Default value: \fB0\fR.
 .RE
 
+.sp
+.ne 2
+.na
+\fBzfs_ddt_data_is_special\fR (int)
+.ad
+.RS 12n
+If enabled, ZFS will place DDT data into the special allocation class.
+.sp
+Default value: \fB1\fR.
+.RE
+
+.sp
+.ne 2
+.na
+\fBzfs_user_indirect_is_special\fR (int)
+.ad
+.RS 12n
+If enabled, ZFS will place user data (both file and zvol) indirect blocks
+into the special allocation class.
+.sp
+Default value: \fB1\fR.
+.RE
+
 .sp
 .ne 2
 .na

man/man5/zpool-features.5

@@ -758,5 +758,27 @@ for the filesystems containing a large number of files.
 
 .RE
 
+.sp
+.ne 2
+.na
+\fB\fBallocation_classes\fR\fR
+.ad
+.RS 4n
+.TS
+l l .
+GUID	com.intel:allocation_classes
+READ\-ONLY COMPATIBLE	yes
+DEPENDENCIES	none
+.TE
+
+This feature enables support for separate allocation classes.
+
+This feature becomes \fBactive\fR when a dedicated allocation class vdev
+(dedup or special) is created with zpool create or zpool add. With device
+removal, it can be returned to the \fBenabled\fR state if all the top-level
+vdevs from an allocation class are removed.
+
+.RE
+
 .SH "SEE ALSO"
 \fBzpool\fR(8)

man/man8/zfs.8

@@ -1522,6 +1522,16 @@ This feature must be enabled to be used
 .Po see
 .Xr zpool-features 5
 .Pc .
+.It Sy special_small_blocks Ns = Ns Em size
+This value represents the threshold block size for including small file
+blocks into the special allocation class. Valid values are zero or a
+power of two from 512B up to 128K. The default size is 0 which means no
+small file blocks will be allocated in the special class.
+.Pp
+Before setting this property, a special class vdev must be added to the
+pool. See
+.Xr zpool 8
+for more details on the special allocation class.
 .It Sy mountpoint Ns = Ns Pa path Ns | Ns Sy none Ns | Ns Sy legacy
 Controls the mount point used for this file system.
 See the

man/man8/zpool.8

@@ -278,6 +278,21 @@ However, raidz vdev types are not supported for the intent log.
 For more information, see the
 .Sx Intent Log
 section.
+.It Sy dedup
+A device dedicated solely for allocating dedup data.
+The redundancy of this device should match the redundancy of the other normal
+devices in the pool. If more than one dedup device is specified, then
+allocations are load-balanced between devices.
+.It Sy special
+A device dedicated solely for allocating various kinds of internal metadata,
+and optionally small file data.
+The redundancy of this device should match the redundancy of the other normal
+devices in the pool. If more than one special device is specified, then
+allocations are load-balanced between devices.
+.Pp
+For more information on special allocations, see the
+.Sx Special Allocation Class
+section.
 .It Sy cache
 A device used to cache storage pool data.
 A cache device cannot be configured as a mirror or raidz group.
@@ -520,6 +535,29 @@ zfs properties) may be unenforceable while a checkpoint exists, because the
 checkpoint is allowed to consume the dataset's reservation.
 Finally, data that is part of the checkpoint but has been freed in the
 current state of the pool won't be scanned during a scrub.
+.Ss Special Allocation Class
+The allocations in the special class are dedicated to specific block types.
+By default this includes all metadata, the indirect blocks of user data, and
+any dedup data.  The class can also be provisioned to accept a limited
+percentage of small file data blocks.
+.Pp
+A pool must always have at least one general (non-specified) vdev before
+other devices can be assigned to the special class. If the special class
+becomes full, then allocations intended for it will spill back into the
+normal class.
+.Pp
+Dedup data can be excluded from the special class by setting the
+.Sy zfs_ddt_data_is_special
+zfs module parameter to false (0).
+.Pp
+Inclusion of small file blocks in the special class is opt-in. Each dataset
+can control the size of small file blocks allowed in the special class by
+setting the
+.Sy special_small_blocks
+dataset property. It defaults to zero so you must opt-in by setting it to a
+non-zero value. See
+.Xr zfs 8
+for more info on setting this property.
 .Ss Properties
 Each pool has several properties associated with it.
 Some properties are read-only statistics while others are configurable and

module/zcommon/zfeature_common.c

@@ -20,10 +20,11 @@
  */
 
 /*
- * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  * Copyright (c) 2014, Nexenta Systems, Inc. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #ifndef _KERNEL
@@ -408,6 +409,13 @@ zpool_feature_init(void)
 	    ZFEATURE_FLAG_READONLY_COMPAT | ZFEATURE_FLAG_PER_DATASET,
 	    project_quota_deps);
 	}
+
+	{
+	zfeature_register(SPA_FEATURE_ALLOCATION_CLASSES,
+	    "org.zfsonlinux:allocation_classes", "allocation_classes",
+	    "Support for separate allocation classes.",
+	    ZFEATURE_FLAG_READONLY_COMPAT, NULL);
+	}
 }
 
 #if defined(_KERNEL)

module/zcommon/zfs_prop.c

@@ -538,6 +538,9 @@ zfs_prop_init(void)
 	zprop_register_number(ZFS_PROP_RECORDSIZE, "recordsize",
 	    SPA_OLD_MAXBLOCKSIZE, PROP_INHERIT,
 	    ZFS_TYPE_FILESYSTEM, "512 to 1M, power of 2", "RECSIZE");
+	zprop_register_number(ZFS_PROP_SPECIAL_SMALL_BLOCKS,
+	    "special_small_blocks", 0, PROP_INHERIT, ZFS_TYPE_FILESYSTEM,
+	    "zero or 512 to 128K, power of 2", "SPECIAL_SMALL_BLOCKS");
 
 	/* hidden properties */
 	zprop_register_hidden(ZFS_PROP_NUMCLONES, "numclones", PROP_TYPE_NUMBER,

module/zfs/dmu.c

@@ -2281,6 +2281,8 @@ dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
 	bzero(zp->zp_salt, ZIO_DATA_SALT_LEN);
 	bzero(zp->zp_iv, ZIO_DATA_IV_LEN);
 	bzero(zp->zp_mac, ZIO_DATA_MAC_LEN);
+	zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ?
+	    os->os_zpl_special_smallblock : 0;
 
 	ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT);
 }

module/zfs/dmu_objset.c

@@ -314,6 +314,20 @@ dnodesize_changed_cb(void *arg, uint64_t newval)
 	}
 }
 
+static void
+smallblk_changed_cb(void *arg, uint64_t newval)
+{
+	objset_t *os = arg;
+
+	/*
+	 * Inheritance and range checking should have been done by now.
+	 */
+	ASSERT(newval <= SPA_OLD_MAXBLOCKSIZE);
+	ASSERT(ISP2(newval));
+
+	os->os_zpl_special_smallblock = newval;
+}
+
 static void
 logbias_changed_cb(void *arg, uint64_t newval)
 {
@@ -556,6 +570,12 @@ dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
 				    zfs_prop_to_name(ZFS_PROP_DNODESIZE),
 				    dnodesize_changed_cb, os);
 			}
+			if (err == 0) {
+				err = dsl_prop_register(ds,
+				    zfs_prop_to_name(
+				    ZFS_PROP_SPECIAL_SMALL_BLOCKS),
+				    smallblk_changed_cb, os);
+			}
 		}
 		if (needlock)
 			dsl_pool_config_exit(dmu_objset_pool(os), FTAG);

module/zfs/metaslab.c

@@ -20,8 +20,9 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #include <sys/zfs_context.h>
@@ -300,7 +301,7 @@ metaslab_class_validate(metaslab_class_t *mc)
 	return (0);
 }
 
-void
+static void
 metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta,
     int64_t defer_delta, int64_t space_delta, int64_t dspace_delta)
 {
@@ -337,7 +338,8 @@ metaslab_class_get_dspace(metaslab_class_t *mc)
 void
 metaslab_class_histogram_verify(metaslab_class_t *mc)
 {
-	vdev_t *rvd = mc->mc_spa->spa_root_vdev;
+	spa_t *spa = mc->mc_spa;
+	vdev_t *rvd = spa->spa_root_vdev;
 	uint64_t *mc_hist;
 	int i;
 
@@ -834,7 +836,8 @@ metaslab_group_histogram_verify(metaslab_group_t *mg)
 	for (int m = 0; m < vd->vdev_ms_count; m++) {
 		metaslab_t *msp = vd->vdev_ms[m];
 
-		if (msp->ms_sm == NULL)
+		/* skip if not active or not a member */
+		if (msp->ms_sm == NULL || msp->ms_group != mg)
 			continue;
 
 		for (i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++)
@@ -967,12 +970,14 @@ metaslab_group_fragmentation(metaslab_group_t *mg)
 
 		if (msp->ms_fragmentation == ZFS_FRAG_INVALID)
 			continue;
+		if (msp->ms_group != mg)
+			continue;
 
 		valid_ms++;
 		fragmentation += msp->ms_fragmentation;
 	}
 
-	if (valid_ms <= vd->vdev_ms_count / 2)
+	if (valid_ms <= mg->mg_vd->vdev_ms_count / 2)
 		return (ZFS_FRAG_INVALID);
 
 	fragmentation /= valid_ms;
@@ -1003,7 +1008,10 @@ metaslab_group_allocatable(metaslab_group_t *mg, metaslab_group_t *rotor,
 	 * groups to select from. Otherwise, we always consider it eligible
 	 * for allocations.
 	 */
-	if (mc != spa_normal_class(spa) || mc->mc_groups <= 1)
+	if ((mc != spa_normal_class(spa) &&
+	    mc != spa_special_class(spa) &&
+	    mc != spa_dedup_class(spa)) ||
+	    mc->mc_groups <= 1)
 		return (B_TRUE);
 
 	/*
@@ -1466,12 +1474,26 @@ metaslab_unload(metaslab_t *msp)
 	msp->ms_max_size = 0;
 }
 
+static void
+metaslab_space_update(vdev_t *vd, metaslab_class_t *mc, int64_t alloc_delta,
+    int64_t defer_delta, int64_t space_delta)
+{
+	vdev_space_update(vd, alloc_delta, defer_delta, space_delta);
+
+	ASSERT3P(vd->vdev_spa->spa_root_vdev, ==, vd->vdev_parent);
+	ASSERT(vd->vdev_ms_count != 0);
+
+	metaslab_class_space_update(mc, alloc_delta, defer_delta, space_delta,
+	    vdev_deflated_space(vd, space_delta));
+}
+
 int
 metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg,
     metaslab_t **msp)
 {
 	vdev_t *vd = mg->mg_vd;
-	objset_t *mos = vd->vdev_spa->spa_meta_objset;
+	spa_t *spa = vd->vdev_spa;
+	objset_t *mos = spa->spa_meta_objset;
 	metaslab_t *ms;
 	int error;
 
@@ -1528,8 +1550,8 @@ metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg,
 
 	/*
 	 * If metaslab_debug_load is set and we're initializing a metaslab
-	 * that has an allocated space map object then load the its space
+	 * that has an allocated space map object then load the space map
-	 * map so that can verify frees.
+	 * so that we can verify frees.
 	 */
 	if (metaslab_debug_load && ms->ms_sm != NULL) {
 		mutex_enter(&ms->ms_lock);
@@ -1551,16 +1573,19 @@ void
 metaslab_fini(metaslab_t *msp)
 {
 	metaslab_group_t *mg = msp->ms_group;
+	vdev_t *vd = mg->mg_vd;
 
 	metaslab_group_remove(mg, msp);
 
 	mutex_enter(&msp->ms_lock);
 	VERIFY(msp->ms_group == NULL);
-	vdev_space_update(mg->mg_vd, -space_map_allocated(msp->ms_sm),
+	metaslab_space_update(vd, mg->mg_class,
-	    0, -msp->ms_size);
+	    -space_map_allocated(msp->ms_sm), 0, -msp->ms_size);
+
 	space_map_close(msp->ms_sm);
 
 	metaslab_unload(msp);
+
 	range_tree_destroy(msp->ms_allocatable);
 	range_tree_destroy(msp->ms_freeing);
 	range_tree_destroy(msp->ms_freed);
@@ -2583,7 +2608,7 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
 		ASSERT3P(msp->ms_checkpointing, ==, NULL);
 		msp->ms_checkpointing = range_tree_create(NULL, NULL);
 
-		vdev_space_update(vd, 0, 0, msp->ms_size);
+		metaslab_space_update(vd, mg->mg_class, 0, 0, msp->ms_size);
 	}
 	ASSERT0(range_tree_space(msp->ms_freeing));
 	ASSERT0(range_tree_space(msp->ms_checkpointing));
@@ -2605,7 +2630,8 @@ metaslab_sync_done(metaslab_t *msp, uint64_t txg)
 		defer_delta -= range_tree_space(*defer_tree);
 	}
 
-	vdev_space_update(vd, alloc_delta + defer_delta, defer_delta, 0);
+	metaslab_space_update(vd, mg->mg_class, alloc_delta + defer_delta,
+	    defer_delta, 0);
 
 	/*
 	 * If there's a metaslab_load() in progress, wait for it to complete
@@ -2704,21 +2730,25 @@ metaslab_sync_reassess(metaslab_group_t *mg)
 	spa_config_exit(spa, SCL_ALLOC, FTAG);
 }
 
-static uint64_t
+/*
-metaslab_distance(metaslab_t *msp, dva_t *dva)
+ * When writing a ditto block (i.e. more than one DVA for a given BP) on
+ * the same vdev as an existing DVA of this BP, then try to allocate it
+ * on a different metaslab than existing DVAs (i.e. a unique metaslab).
+ */
+static boolean_t
+metaslab_is_unique(metaslab_t *msp, dva_t *dva)
 {
-	uint64_t ms_shift = msp->ms_group->mg_vd->vdev_ms_shift;
+	uint64_t dva_ms_id;
-	uint64_t offset = DVA_GET_OFFSET(dva) >> ms_shift;
+
-	uint64_t start = msp->ms_id;
+	if (DVA_GET_ASIZE(dva) == 0)
+		return (B_TRUE);
 
 	if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva))
-		return (1ULL << 63);
+		return (B_TRUE);
 
-	if (offset < start)
+	dva_ms_id = DVA_GET_OFFSET(dva) >> msp->ms_group->mg_vd->vdev_ms_shift;
-		return ((start - offset) << ms_shift);
+
-	if (offset > start)
+	return (msp->ms_id != dva_ms_id);
-		return ((offset - start) << ms_shift);
-	return (0);
 }
 
 /*
@@ -2978,7 +3008,7 @@ metaslab_block_alloc(metaslab_t *msp, uint64_t size, uint64_t txg)
  */
 static metaslab_t *
 find_valid_metaslab(metaslab_group_t *mg, uint64_t activation_weight,
-    dva_t *dva, int d, uint64_t min_distance, uint64_t asize, int allocator,
+    dva_t *dva, int d, boolean_t want_unique, uint64_t asize, int allocator,
     zio_alloc_list_t *zal, metaslab_t *search, boolean_t *was_active)
 {
 	avl_index_t idx;
@@ -3012,13 +3042,10 @@ find_valid_metaslab(metaslab_group_t *mg, uint64_t activation_weight,
 		if (activation_weight == METASLAB_WEIGHT_PRIMARY || *was_active)
 			break;
 
-		uint64_t target_distance = min_distance
-		    + (space_map_allocated(msp->ms_sm) != 0 ? 0 :
-		    min_distance >> 1);
-
 		for (i = 0; i < d; i++) {
-			if (metaslab_distance(msp, &dva[i]) < target_distance)
+			if (want_unique &&
-				break;
+			    !metaslab_is_unique(msp, &dva[i]))
+				break;  /* try another metaslab */
 		}
 		if (i == d)
 			break;
@@ -3036,8 +3063,8 @@ find_valid_metaslab(metaslab_group_t *mg, uint64_t activation_weight,
 /* ARGSUSED */
 static uint64_t
 metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal,
-    uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d,
+    uint64_t asize, uint64_t txg, boolean_t want_unique, dva_t *dva,
-    int allocator)
+    int d, int allocator)
 {
 	metaslab_t *msp = NULL;
 	uint64_t offset = -1ULL;
@@ -3091,7 +3118,7 @@ metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal,
 			was_active = B_TRUE;
 		} else {
 			msp = find_valid_metaslab(mg, activation_weight, dva, d,
-			    min_distance, asize, allocator, zal, search,
+			    want_unique, asize, allocator, zal, search,
 			    &was_active);
 		}
 
@@ -3221,6 +3248,7 @@ next:
 		 * metaslab.
 		 */
 		ASSERT(!metaslab_should_allocate(msp, asize));
+
 		mutex_exit(&msp->ms_lock);
 	}
 	mutex_exit(&msp->ms_lock);
@@ -3230,14 +3258,14 @@ next:
 
 static uint64_t
 metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal,
-    uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d,
+    uint64_t asize, uint64_t txg, boolean_t want_unique, dva_t *dva,
-    int allocator)
+    int d, int allocator)
 {
 	uint64_t offset;
 	ASSERT(mg->mg_initialized);
 
-	offset = metaslab_group_alloc_normal(mg, zal, asize, txg,
+	offset = metaslab_group_alloc_normal(mg, zal, asize, txg, want_unique,
-	    min_distance, dva, d, allocator);
+	    dva, d, allocator);
 
 	mutex_enter(&mg->mg_lock);
 	if (offset == -1ULL) {
@@ -3264,14 +3292,6 @@ metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal,
 	return (offset);
 }
 
-/*
- * If we have to write a ditto block (i.e. more than one DVA for a given BP)
- * on the same vdev as an existing DVA of this BP, then try to allocate it
- * at least (vdev_asize / (2 ^ ditto_same_vdev_distance_shift)) away from the
- * existing DVAs.
- */
-int ditto_same_vdev_distance_shift = 3;
-
 /*
  * Allocate a block for the specified i/o.
  */
@@ -3288,6 +3308,7 @@ metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
 
 	/*
 	 * For testing, make some blocks above a certain size be gang blocks.
+	 * This will also test spilling from special to normal.
 	 */
 	if (psize >= metaslab_force_ganging && (ddi_get_lbolt() & 3) == 0) {
 		metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG,
@@ -3348,6 +3369,7 @@ metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
 		} while ((fast_mg = fast_mg->mg_next) != mc->mc_rotor);
 
 	} else {
+		ASSERT(mc->mc_rotor != NULL);
 		mg = mc->mc_rotor;
 	}
 
@@ -3412,25 +3434,17 @@ top:
 
 		ASSERT(mg->mg_class == mc);
 
-		/*
-		 * If we don't need to try hard, then require that the
-		 * block be 1/8th of the device away from any other DVAs
-		 * in this BP.  If we are trying hard, allow any offset
-		 * to be used (distance=0).
-		 */
-		uint64_t distance = 0;
-		if (!try_hard) {
-			distance = vd->vdev_asize >>
-			    ditto_same_vdev_distance_shift;
-			if (distance <= (1ULL << vd->vdev_ms_shift))
-				distance = 0;
-		}
-
 		uint64_t asize = vdev_psize_to_asize(vd, psize);
 		ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0);
 
+		/*
+		 * If we don't need to try hard, then require that the
+		 * block be on an different metaslab from any other DVAs
+		 * in this BP (unique=true).  If we are trying hard, then
+		 * allow any metaslab to be used (unique=false).
+		 */
 		uint64_t offset = metaslab_group_alloc(mg, zal, asize, txg,
-		    distance, dva, d, allocator);
+		    !try_hard, dva, d, allocator);
 
 		if (offset != -1ULL) {
 			/*
@@ -3830,7 +3844,8 @@ metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, int allocator,
 	if (reserved_slots < max)
 		available_slots = max - reserved_slots;
 
-	if (slots <= available_slots || GANG_ALLOCATION(flags)) {
+	if (slots <= available_slots || GANG_ALLOCATION(flags) ||
+	    flags & METASLAB_MUST_RESERVE) {
 		/*
 		 * We reserve the slots individually so that we can unreserve
 		 * them individually when an I/O completes.
@@ -4107,9 +4122,11 @@ metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg)
 
 	spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
 
-	for (int d = 0; d < ndvas; d++)
+	for (int d = 0; d < ndvas; d++) {
-		if ((error = metaslab_claim_dva(spa, &dva[d], txg)) != 0)
+		error = metaslab_claim_dva(spa, &dva[d], txg);
+		if (error != 0)
 			break;
+	}
 
 	spa_config_exit(spa, SCL_ALLOC, FTAG);
 
@@ -4235,7 +4252,7 @@ metaslab_check_free(spa_t *spa, const blkptr_t *bp)
 }
 
 #if defined(_KERNEL)
-/* CSTYLED */
+/* BEGIN CSTYLED */
 module_param(metaslab_aliquot, ulong, 0644);
 MODULE_PARM_DESC(metaslab_aliquot,
 	"allocation granularity (a.k.a. stripe size)");
@@ -4284,8 +4301,9 @@ module_param(zfs_metaslab_switch_threshold, int, 0644);
 MODULE_PARM_DESC(zfs_metaslab_switch_threshold,
 	"segment-based metaslab selection maximum buckets before switching");
 
-/* CSTYLED */
 module_param(metaslab_force_ganging, ulong, 0644);
 MODULE_PARM_DESC(metaslab_force_ganging,
 	"blocks larger than this size are forced to be gang blocks");
+/* END CSTYLED */
+
 #endif

module/zfs/spa.c

@@ -31,6 +31,7 @@
  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
  * Copyright (c) 2017 Datto Inc.
  * Copyright 2017 Joyent, Inc.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 /*
@@ -272,8 +273,14 @@ spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
 	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
 
 	if (rvd != NULL) {
-		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
+		alloc = metaslab_class_get_alloc(mc);
-		size = metaslab_class_get_space(spa_normal_class(spa));
+		alloc += metaslab_class_get_alloc(spa_special_class(spa));
+		alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
+
+		size = metaslab_class_get_space(mc);
+		size += metaslab_class_get_space(spa_special_class(spa));
+		size += metaslab_class_get_space(spa_dedup_class(spa));
+
 		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
 		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
@@ -1173,6 +1180,8 @@ spa_activate(spa_t *spa, int mode)
 
 	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
 	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
+	spa->spa_special_class = metaslab_class_create(spa, zfs_metaslab_ops);
+	spa->spa_dedup_class = metaslab_class_create(spa, zfs_metaslab_ops);
 
 	/* Try to create a covering process */
 	mutex_enter(&spa->spa_proc_lock);
@@ -1320,6 +1329,12 @@ spa_deactivate(spa_t *spa)
 	metaslab_class_destroy(spa->spa_log_class);
 	spa->spa_log_class = NULL;
 
+	metaslab_class_destroy(spa->spa_special_class);
+	spa->spa_special_class = NULL;
+
+	metaslab_class_destroy(spa->spa_dedup_class);
+	spa->spa_dedup_class = NULL;
+
 	/*
 	 * If this was part of an import or the open otherwise failed, we may
 	 * still have errors left in the queues.  Empty them just in case.
@@ -4988,7 +5003,8 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 	char *poolname;
 	nvlist_t *nvl;
 
-	if (nvlist_lookup_string(props, "tname", &poolname) != 0)
+	if (props == NULL ||
+	    nvlist_lookup_string(props, "tname", &poolname) != 0)
 		poolname = (char *)pool;
 
 	/*
@@ -5092,9 +5108,15 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 	    (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
 	    (error = spa_validate_aux(spa, nvroot, txg,
 	    VDEV_ALLOC_ADD)) == 0) {
-		for (int c = 0; c < rvd->vdev_children; c++) {
+		/*
-			vdev_metaslab_set_size(rvd->vdev_child[c]);
+		 * instantiate the metaslab groups (this will dirty the vdevs)
-			vdev_expand(rvd->vdev_child[c], txg);
+		 * we can no longer error exit past this point
+		 */
+		for (int c = 0; error == 0 && c < rvd->vdev_children; c++) {
+			vdev_t *vd = rvd->vdev_child[c];
+
+			vdev_metaslab_set_size(vd);
+			vdev_expand(vd, txg);
 		}
 	}
 
@@ -6940,8 +6962,14 @@ spa_async_thread(void *arg)
 
 		mutex_enter(&spa_namespace_lock);
 		old_space = metaslab_class_get_space(spa_normal_class(spa));
+		old_space += metaslab_class_get_space(spa_special_class(spa));
+		old_space += metaslab_class_get_space(spa_dedup_class(spa));
+
 		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
+
 		new_space = metaslab_class_get_space(spa_normal_class(spa));
+		new_space += metaslab_class_get_space(spa_special_class(spa));
+		new_space += metaslab_class_get_space(spa_dedup_class(spa));
 		mutex_exit(&spa_namespace_lock);
 
 		/*
@@ -7630,6 +7658,9 @@ spa_sync(spa_t *spa, uint64_t txg)
 	dsl_pool_t *dp = spa->spa_dsl_pool;
 	objset_t *mos = spa->spa_meta_objset;
 	bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
+	metaslab_class_t *normal = spa_normal_class(spa);
+	metaslab_class_t *special = spa_special_class(spa);
+	metaslab_class_t *dedup = spa_dedup_class(spa);
 	vdev_t *rvd = spa->spa_root_vdev;
 	vdev_t *vd;
 	dmu_tx_t *tx;
@@ -7723,9 +7754,13 @@ spa_sync(spa_t *spa, uint64_t txg)
 	for (int c = 0; c < rvd->vdev_children; c++) {
 		vdev_t *tvd = rvd->vdev_child[c];
 		metaslab_group_t *mg = tvd->vdev_mg;
+		metaslab_class_t *mc;
 
-		if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
+		if (mg == NULL || !metaslab_group_initialized(mg))
-		    !metaslab_group_initialized(mg))
+			continue;
+
+		mc = mg->mg_class;
+		if (mc != normal && mc != special && mc != dedup)
 			continue;
 
 		/*
@@ -7743,12 +7778,18 @@ spa_sync(spa_t *spa, uint64_t txg)
 		}
 		slots_per_allocator += zfs_vdev_def_queue_depth;
 	}
-	metaslab_class_t *mc = spa_normal_class(spa);
+
 	for (int i = 0; i < spa->spa_alloc_count; i++) {
-		ASSERT0(refcount_count(&mc->mc_alloc_slots[i]));
+		ASSERT0(refcount_count(&normal->mc_alloc_slots[i]));
-		mc->mc_alloc_max_slots[i] = slots_per_allocator;
+		ASSERT0(refcount_count(&special->mc_alloc_slots[i]));
+		ASSERT0(refcount_count(&dedup->mc_alloc_slots[i]));
+		normal->mc_alloc_max_slots[i] = slots_per_allocator;
+		special->mc_alloc_max_slots[i] = slots_per_allocator;
+		dedup->mc_alloc_max_slots[i] = slots_per_allocator;
 	}
-	mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
+	normal->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
+	special->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
+	dedup->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
 
 	for (int c = 0; c < rvd->vdev_children; c++) {
 		vdev_t *vd = rvd->vdev_child[c];

module/zfs/spa_misc.c

@@ -25,6 +25,7 @@
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  * Copyright 2013 Saso Kiselkov. All rights reserved.
  * Copyright (c) 2017 Datto Inc.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #include <sys/zfs_context.h>
@@ -408,6 +409,19 @@ spa_load_note(spa_t *spa, const char *fmt, ...)
 	    spa->spa_trust_config ? "trusted" : "untrusted", buf);
 }
 
+/*
+ * By default dedup and user data indirects land in the special class
+ */
+int zfs_ddt_data_is_special = B_TRUE;
+int zfs_user_indirect_is_special = B_TRUE;
+
+/*
+ * The percentage of special class final space reserved for metadata only.
+ * Once we allocate 100 - zfs_special_class_metadata_reserve_pct we only
+ * let metadata into the class.
+ */
+int zfs_special_class_metadata_reserve_pct = 25;
+
 /*
  * ==========================================================================
  * SPA config locking
@@ -1159,6 +1173,8 @@ spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error, char *tag)
 	 */
 	ASSERT(metaslab_class_validate(spa_normal_class(spa)) == 0);
 	ASSERT(metaslab_class_validate(spa_log_class(spa)) == 0);
+	ASSERT(metaslab_class_validate(spa_special_class(spa)) == 0);
+	ASSERT(metaslab_class_validate(spa_dedup_class(spa)) == 0);
 
 	spa_config_exit(spa, SCL_ALL, spa);
 
@@ -1554,6 +1570,16 @@ zfs_strtonum(const char *str, char **nptr)
 	return (val);
 }
 
+void
+spa_activate_allocation_classes(spa_t *spa, dmu_tx_t *tx)
+{
+	/*
+	 * We bump the feature refcount for each special vdev added to the pool
+	 */
+	ASSERT(spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES));
+	spa_feature_incr(spa, SPA_FEATURE_ALLOCATION_CLASSES, tx);
+}
+
 /*
  * ==========================================================================
  * Accessor functions
@@ -1811,6 +1837,79 @@ spa_log_class(spa_t *spa)
 	return (spa->spa_log_class);
 }
 
+metaslab_class_t *
+spa_special_class(spa_t *spa)
+{
+	return (spa->spa_special_class);
+}
+
+metaslab_class_t *
+spa_dedup_class(spa_t *spa)
+{
+	return (spa->spa_dedup_class);
+}
+
+/*
+ * Locate an appropriate allocation class
+ */
+metaslab_class_t *
+spa_preferred_class(spa_t *spa, uint64_t size, dmu_object_type_t objtype,
+    uint_t level, uint_t special_smallblk)
+{
+	if (DMU_OT_IS_ZIL(objtype)) {
+		if (spa->spa_log_class->mc_groups != 0)
+			return (spa_log_class(spa));
+		else
+			return (spa_normal_class(spa));
+	}
+
+	boolean_t has_special_class = spa->spa_special_class->mc_groups != 0;
+
+	if (DMU_OT_IS_DDT(objtype)) {
+		if (spa->spa_dedup_class->mc_groups != 0)
+			return (spa_dedup_class(spa));
+		else if (has_special_class && zfs_ddt_data_is_special)
+			return (spa_special_class(spa));
+		else
+			return (spa_normal_class(spa));
+	}
+
+	/* Indirect blocks for user data can land in special if allowed */
+	if (level > 0 && (DMU_OT_IS_FILE(objtype) || objtype == DMU_OT_ZVOL)) {
+		if (has_special_class && zfs_user_indirect_is_special)
+			return (spa_special_class(spa));
+		else
+			return (spa_normal_class(spa));
+	}
+
+	if (DMU_OT_IS_METADATA(objtype) || level > 0) {
+		if (has_special_class)
+			return (spa_special_class(spa));
+		else
+			return (spa_normal_class(spa));
+	}
+
+	/*
+	 * Allow small file blocks in special class in some cases (like
+	 * for the dRAID vdev feature). But always leave a reserve of
+	 * zfs_special_class_metadata_reserve_pct exclusively for metadata.
+	 */
+	if (DMU_OT_IS_FILE(objtype) &&
+	    has_special_class && size < special_smallblk) {
+		metaslab_class_t *special = spa_special_class(spa);
+		uint64_t alloc = metaslab_class_get_alloc(special);
+		uint64_t space = metaslab_class_get_space(special);
+		uint64_t limit =
+		    (space * (100 - zfs_special_class_metadata_reserve_pct))
+		    / 100;
+
+		if (alloc < limit)
+			return (special);
+	}
+
+	return (spa_normal_class(spa));
+}
+
 void
 spa_evicting_os_register(spa_t *spa, objset_t *os)
 {
@@ -2500,6 +2599,8 @@ EXPORT_SYMBOL(spa_update_dspace);
 EXPORT_SYMBOL(spa_deflate);
 EXPORT_SYMBOL(spa_normal_class);
 EXPORT_SYMBOL(spa_log_class);
+EXPORT_SYMBOL(spa_special_class);
+EXPORT_SYMBOL(spa_preferred_class);
 EXPORT_SYMBOL(spa_max_replication);
 EXPORT_SYMBOL(spa_prev_software_version);
 EXPORT_SYMBOL(spa_get_failmode);
@@ -2579,5 +2680,13 @@ MODULE_PARM_DESC(spa_asize_inflation,
 
 module_param(spa_slop_shift, int, 0644);
 MODULE_PARM_DESC(spa_slop_shift, "Reserved free space in pool");
+
+module_param(zfs_ddt_data_is_special, int, 0644);
+MODULE_PARM_DESC(zfs_ddt_data_is_special,
+	"Place DDT data into the special class");
+
+module_param(zfs_user_indirect_is_special, int, 0644);
+MODULE_PARM_DESC(zfs_user_indirect_is_special,
+	"Place user data indirect blocks into the special class");
 /* END CSTYLED */
 #endif

module/zfs/vdev.c

@@ -26,6 +26,7 @@
  * Copyright (c) 2014 Integros [integros.com]
  * Copyright 2016 Toomas Soome <tsoome@me.com>
  * Copyright 2017 Joyent, Inc.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #include <sys/zfs_context.h>
@@ -204,6 +205,25 @@ vdev_getops(const char *type)
 	return (ops);
 }
 
+/*
+ * Derive the enumerated alloction bias from string input.
+ * String origin is either the per-vdev zap or zpool(1M).
+ */
+static vdev_alloc_bias_t
+vdev_derive_alloc_bias(const char *bias)
+{
+	vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
+
+	if (strcmp(bias, VDEV_ALLOC_BIAS_LOG) == 0)
+		alloc_bias = VDEV_BIAS_LOG;
+	else if (strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0)
+		alloc_bias = VDEV_BIAS_SPECIAL;
+	else if (strcmp(bias, VDEV_ALLOC_BIAS_DEDUP) == 0)
+		alloc_bias = VDEV_BIAS_DEDUP;
+
+	return (alloc_bias);
+}
+
 /*
  * Default asize function: return the MAX of psize with the asize of
  * all children.  This is what's used by anything other than RAID-Z.
@@ -528,6 +548,8 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 	vdev_indirect_config_t *vic;
 	char *tmp = NULL;
 	int rc;
+	vdev_alloc_bias_t alloc_bias = VDEV_BIAS_NONE;
+	boolean_t top_level = (parent && !parent->vdev_parent);
 
 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 
@@ -614,11 +636,32 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 	}
 	ASSERT(nparity != -1ULL);
 
+	/*
+	 * If creating a top-level vdev, check for allocation classes input
+	 */
+	if (top_level && alloctype == VDEV_ALLOC_ADD) {
+		char *bias;
+
+		if (nvlist_lookup_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
+		    &bias) == 0) {
+			alloc_bias = vdev_derive_alloc_bias(bias);
+
+			/* spa_vdev_add() expects feature to be enabled */
+			if (spa->spa_load_state != SPA_LOAD_CREATE &&
+			    !spa_feature_is_enabled(spa,
+			    SPA_FEATURE_ALLOCATION_CLASSES)) {
+				return (SET_ERROR(ENOTSUP));
+			}
+		}
+	}
+
 	vd = vdev_alloc_common(spa, id, guid, ops);
 	vic = &vd->vdev_indirect_config;
 
 	vd->vdev_islog = islog;
 	vd->vdev_nparity = nparity;
+	if (top_level && alloc_bias != VDEV_BIAS_NONE)
+		vd->vdev_alloc_bias = alloc_bias;
 
 	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
 		vd->vdev_path = spa_strdup(vd->vdev_path);
@@ -687,7 +730,7 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 	/*
 	 * If we're a top-level vdev, try to load the allocation parameters.
 	 */
-	if (parent && !parent->vdev_parent &&
+	if (top_level &&
 	    (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
 		(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
 		    &vd->vdev_ms_array);
@@ -703,14 +746,12 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 		ASSERT0(vd->vdev_top_zap);
 	}
 
-	if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
+	if (top_level && alloctype != VDEV_ALLOC_ATTACH) {
 		ASSERT(alloctype == VDEV_ALLOC_LOAD ||
 		    alloctype == VDEV_ALLOC_ADD ||
 		    alloctype == VDEV_ALLOC_SPLIT ||
 		    alloctype == VDEV_ALLOC_ROOTPOOL);
-		vd->vdev_mg = metaslab_group_create(islog ?
+		/* Note: metaslab_group_create() is now deferred */
-		    spa_log_class(spa) : spa_normal_class(spa), vd,
-		    spa->spa_alloc_count);
 	}
 
 	if (vd->vdev_ops->vdev_op_leaf &&
@@ -952,6 +993,9 @@ vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
 	tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm;
 	svd->vdev_checkpoint_sm = NULL;
 
+	tvd->vdev_alloc_bias = svd->vdev_alloc_bias;
+	svd->vdev_alloc_bias = VDEV_BIAS_NONE;
+
 	tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
 	tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
 	tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
@@ -1114,6 +1158,55 @@ vdev_remove_parent(vdev_t *cvd)
 	vdev_free(mvd);
 }
 
+static void
+vdev_metaslab_group_create(vdev_t *vd)
+{
+	spa_t *spa = vd->vdev_spa;
+
+	/*
+	 * metaslab_group_create was delayed until allocation bias was available
+	 */
+	if (vd->vdev_mg == NULL) {
+		metaslab_class_t *mc;
+
+		if (vd->vdev_islog && vd->vdev_alloc_bias == VDEV_BIAS_NONE)
+			vd->vdev_alloc_bias = VDEV_BIAS_LOG;
+
+		ASSERT3U(vd->vdev_islog, ==,
+		    (vd->vdev_alloc_bias == VDEV_BIAS_LOG));
+
+		switch (vd->vdev_alloc_bias) {
+		case VDEV_BIAS_LOG:
+			mc = spa_log_class(spa);
+			break;
+		case VDEV_BIAS_SPECIAL:
+			mc = spa_special_class(spa);
+			break;
+		case VDEV_BIAS_DEDUP:
+			mc = spa_dedup_class(spa);
+			break;
+		default:
+			mc = spa_normal_class(spa);
+		}
+
+		vd->vdev_mg = metaslab_group_create(mc, vd,
+		    spa->spa_alloc_count);
+
+		/*
+		 * The spa ashift values currently only reflect the
+		 * general vdev classes. Class destination is late
+		 * binding so ashift checking had to wait until now
+		 */
+		if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
+		    mc == spa_normal_class(spa) && vd->vdev_aux == NULL) {
+			if (vd->vdev_ashift > spa->spa_max_ashift)
+				spa->spa_max_ashift = vd->vdev_ashift;
+			if (vd->vdev_ashift < spa->spa_min_ashift)
+				spa->spa_min_ashift = vd->vdev_ashift;
+		}
+	}
+}
+
 int
 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 {
@@ -1124,6 +1217,7 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 	uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
 	metaslab_t **mspp;
 	int error;
+	boolean_t expanding = (oldc != 0);
 
 	ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
 
@@ -1139,7 +1233,7 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 
 	mspp = vmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
 
-	if (oldc != 0) {
+	if (expanding) {
 		bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
 		vmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
 	}
@@ -1165,6 +1259,17 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 			}
 		}
 
+#ifndef _KERNEL
+		/*
+		 * To accomodate zdb_leak_init() fake indirect
+		 * metaslabs, we allocate a metaslab group for
+		 * indirect vdevs which normally don't have one.
+		 */
+		if (vd->vdev_mg == NULL) {
+			ASSERT0(vdev_is_concrete(vd));
+			vdev_metaslab_group_create(vd);
+		}
+#endif
 		error = metaslab_init(vd->vdev_mg, m, object, txg,
 		    &(vd->vdev_ms[m]));
 		if (error != 0) {
@@ -1182,8 +1287,9 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 	 * the metaslabs since we want to ensure that no new
 	 * allocations are performed on this device.
 	 */
-	if (oldc == 0 && !vd->vdev_removing)
+	if (!expanding && !vd->vdev_removing) {
 		metaslab_group_activate(vd->vdev_mg);
+	}
 
 	if (txg == 0)
 		spa_config_exit(spa, SCL_ALLOC, FTAG);
@@ -1673,9 +1779,13 @@ vdev_open(vdev_t *vd)
 
 	/*
 	 * Track the min and max ashift values for normal data devices.
+	 *
+	 * DJB - TBD these should perhaps be tracked per allocation class
+	 * (e.g. spa_min_ashift is used to round up post compression buffers)
 	 */
 	if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
-	    !vd->vdev_islog && vd->vdev_aux == NULL) {
+	    vd->vdev_alloc_bias == VDEV_BIAS_NONE &&
+	    vd->vdev_aux == NULL) {
 		if (vd->vdev_ashift > spa->spa_max_ashift)
 			spa->spa_max_ashift = vd->vdev_ashift;
 		if (vd->vdev_ashift < spa->spa_min_ashift)
@@ -2561,6 +2671,30 @@ vdev_dtl_load(vdev_t *vd)
 	return (error);
 }
 
+static void
+vdev_zap_allocation_data(vdev_t *vd, dmu_tx_t *tx)
+{
+	spa_t *spa = vd->vdev_spa;
+	objset_t *mos = spa->spa_meta_objset;
+	vdev_alloc_bias_t alloc_bias = vd->vdev_alloc_bias;
+	const char *string;
+
+	ASSERT(alloc_bias != VDEV_BIAS_NONE);
+
+	string =
+	    (alloc_bias == VDEV_BIAS_LOG) ? VDEV_ALLOC_BIAS_LOG :
+	    (alloc_bias == VDEV_BIAS_SPECIAL) ? VDEV_ALLOC_BIAS_SPECIAL :
+	    (alloc_bias == VDEV_BIAS_DEDUP) ? VDEV_ALLOC_BIAS_DEDUP : NULL;
+
+	ASSERT(string != NULL);
+	VERIFY0(zap_add(mos, vd->vdev_top_zap, VDEV_TOP_ZAP_ALLOCATION_BIAS,
+	    1, strlen(string) + 1, string, tx));
+
+	if (alloc_bias == VDEV_BIAS_SPECIAL || alloc_bias == VDEV_BIAS_DEDUP) {
+		spa_activate_allocation_classes(spa, tx);
+	}
+}
+
 void
 vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx)
 {
@@ -2597,8 +2731,11 @@ vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx)
 		}
 		if (vd == vd->vdev_top && vd->vdev_top_zap == 0) {
 			vd->vdev_top_zap = vdev_create_link_zap(vd, tx);
+			if (vd->vdev_alloc_bias != VDEV_BIAS_NONE)
+				vdev_zap_allocation_data(vd, tx);
 		}
 	}
+
 	for (uint64_t i = 0; i < vd->vdev_children; i++) {
 		vdev_construct_zaps(vd->vdev_child[i], tx);
 	}
@@ -2801,10 +2938,27 @@ vdev_load(vdev_t *vd)
 
 	vdev_set_deflate_ratio(vd);
 
+	/*
+	 * On spa_load path, grab the allocation bias from our zap
+	 */
+	if (vd == vd->vdev_top && vd->vdev_top_zap != 0) {
+		spa_t *spa = vd->vdev_spa;
+		char bias_str[64];
+
+		if (zap_lookup(spa->spa_meta_objset, vd->vdev_top_zap,
+		    VDEV_TOP_ZAP_ALLOCATION_BIAS, 1, sizeof (bias_str),
+		    bias_str) == 0) {
+			ASSERT(vd->vdev_alloc_bias == VDEV_BIAS_NONE);
+			vd->vdev_alloc_bias = vdev_derive_alloc_bias(bias_str);
+		}
+	}
+
 	/*
 	 * If this is a top-level vdev, initialize its metaslabs.
 	 */
 	if (vd == vd->vdev_top && vdev_is_concrete(vd)) {
+		vdev_metaslab_group_create(vd);
+
 		if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
 			vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
 			    VDEV_AUX_CORRUPT_DATA);
@@ -2999,6 +3153,7 @@ vdev_remove_empty(vdev_t *vd, uint64_t txg)
 
 		metaslab_group_histogram_verify(mg);
 		metaslab_class_histogram_verify(mg->mg_class);
+
 		for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
 			ASSERT0(mg->mg_histogram[i]);
 	}
@@ -3673,7 +3828,8 @@ vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
 		}
 		if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
 		    vdev_is_concrete(vd)) {
-			vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
+			vs->vs_fragmentation = (vd->vdev_mg != NULL) ?
+			    vd->vdev_mg->mg_fragmentation : 0;
 		}
 	}
 
@@ -3878,19 +4034,25 @@ vdev_stat_update(zio_t *zio, uint64_t psize)
 	}
 }
 
+int64_t
+vdev_deflated_space(vdev_t *vd, int64_t space)
+{
+	ASSERT((space & (SPA_MINBLOCKSIZE-1)) == 0);
+	ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
+
+	return ((space >> SPA_MINBLOCKSHIFT) * vd->vdev_deflate_ratio);
+}
+
 /*
- * Update the in-core space usage stats for this vdev, its metaslab class,
+ * Update the in-core space usage stats for this vdev and the root vdev.
- * and the root vdev.
  */
 void
 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
     int64_t space_delta)
 {
-	int64_t dspace_delta = space_delta;
+	int64_t dspace_delta;
 	spa_t *spa = vd->vdev_spa;
 	vdev_t *rvd = spa->spa_root_vdev;
-	metaslab_group_t *mg = vd->vdev_mg;
-	metaslab_class_t *mc = mg ? mg->mg_class : NULL;
 
 	ASSERT(vd == vd->vdev_top);
 
@@ -3900,10 +4062,7 @@ vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
 	 * because the root vdev's psize-to-asize is simply the max of its
 	 * childrens', thus not accurate enough for us.
 	 */
-	ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
+	dspace_delta = vdev_deflated_space(vd, space_delta);
-	ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
-	dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
-	    vd->vdev_deflate_ratio;
 
 	mutex_enter(&vd->vdev_stat_lock);
 	vd->vdev_stat.vs_alloc += alloc_delta;
@@ -3911,21 +4070,15 @@ vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
 	vd->vdev_stat.vs_dspace += dspace_delta;
 	mutex_exit(&vd->vdev_stat_lock);
 
-	if (mc == spa_normal_class(spa)) {
+	/* every class but log contributes to root space stats */
+	if (vd->vdev_mg != NULL && !vd->vdev_islog) {
 		mutex_enter(&rvd->vdev_stat_lock);
 		rvd->vdev_stat.vs_alloc += alloc_delta;
 		rvd->vdev_stat.vs_space += space_delta;
 		rvd->vdev_stat.vs_dspace += dspace_delta;
 		mutex_exit(&rvd->vdev_stat_lock);
 	}
-
+	/* Note: metaslab_class_space_update moved to metaslab_space_update */
-	if (mc != NULL) {
-		ASSERT(rvd == vd->vdev_parent);
-		ASSERT(vd->vdev_ms_count != 0);
-
-		metaslab_class_space_update(mc,
-		    alloc_delta, defer_delta, space_delta, dspace_delta);
-	}
 }
 
 /*
@@ -4354,7 +4507,9 @@ vdev_expand(vdev_t *vd, uint64_t txg)
 
 	vdev_set_deflate_ratio(vd);
 
-	if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
+	if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
+	    vdev_is_concrete(vd)) {
+		vdev_metaslab_group_create(vd);
 		VERIFY(vdev_metaslab_init(vd, txg) == 0);
 		vdev_config_dirty(vd);
 	}

module/zfs/vdev_label.c

@@ -22,6 +22,8 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
+ * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 /*
@@ -463,6 +465,28 @@ vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
 			    vd->vdev_removing);
 		}
+
+		/* zpool command expects alloc class data */
+		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
+			const char *bias = NULL;
+
+			switch (vd->vdev_alloc_bias) {
+			case VDEV_BIAS_LOG:
+				bias = VDEV_ALLOC_BIAS_LOG;
+				break;
+			case VDEV_BIAS_SPECIAL:
+				bias = VDEV_ALLOC_BIAS_SPECIAL;
+				break;
+			case VDEV_BIAS_DEDUP:
+				bias = VDEV_ALLOC_BIAS_DEDUP;
+				break;
+			default:
+				ASSERT3U(vd->vdev_alloc_bias, ==,
+				    VDEV_BIAS_NONE);
+			}
+			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
+			    bias);
+		}
 	}
 
 	if (vd->vdev_dtl_sm != NULL) {

module/zfs/vdev_removal.c

@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -944,8 +944,18 @@ spa_vdev_copy_segment(vdev_t *vd, range_tree_t *segs,
 	}
 	ASSERT3U(size, <=, maxalloc);
 
-	int error = metaslab_alloc_dva(spa, mg->mg_class, size,
+	/*
+	 * An allocation class might not have any remaining vdevs or space
+	 */
+	metaslab_class_t *mc = mg->mg_class;
+	if (mc != spa_normal_class(spa) && mc->mc_groups <= 1)
+		mc = spa_normal_class(spa);
+	int error = metaslab_alloc_dva(spa, mc, size, &dst, 0, NULL, txg, 0,
+	    zal, 0);
+	if (error == ENOSPC && mc != spa_normal_class(spa)) {
+		error = metaslab_alloc_dva(spa, spa_normal_class(spa), size,
 		    &dst, 0, NULL, txg, 0, zal, 0);
+	}
 	if (error != 0)
 		return (error);
 
@@ -1853,15 +1863,31 @@ spa_vdev_remove_top_check(vdev_t *vd)
 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_DEVICE_REMOVAL))
 		return (SET_ERROR(ENOTSUP));
 
+	/* available space in the pool's normal class */
+	uint64_t available = dsl_dir_space_available(
+	    spa->spa_dsl_pool->dp_root_dir, NULL, 0, B_TRUE);
+
+	metaslab_class_t *mc = vd->vdev_mg->mg_class;
+
+	/*
+	 * When removing a vdev from an allocation class that has
+	 * remaining vdevs, include available space from the class.
+	 */
+	if (mc != spa_normal_class(spa) && mc->mc_groups > 1) {
+		uint64_t class_avail = metaslab_class_get_space(mc) -
+		    metaslab_class_get_alloc(mc);
+
+		/* add class space, adjusted for overhead */
+		available += (class_avail * 94) / 100;
+	}
+
 	/*
 	 * There has to be enough free space to remove the
 	 * device and leave double the "slop" space (i.e. we
 	 * must leave at least 3% of the pool free, in addition to
 	 * the normal slop space).
 	 */
-	if (dsl_dir_space_available(spa->spa_dsl_pool->dp_root_dir,
+	if (available < vd->vdev_stat.vs_dspace + spa_get_slop_space(spa)) {
-	    NULL, 0, B_TRUE) <
-	    vd->vdev_stat.vs_dspace + spa_get_slop_space(spa)) {
 		return (SET_ERROR(ENOSPC));
 	}
 

module/zfs/zfs_debug.c

@@ -133,11 +133,15 @@ zfs_dbgmsg_fini(void)
 {
 	if (zfs_dbgmsg_kstat)
 		kstat_delete(zfs_dbgmsg_kstat);
-
+	/*
+	 * TODO - decide how to make this permanent
+	 */
+#ifdef _KERNEL
 	mutex_enter(&zfs_dbgmsgs_lock);
 	zfs_dbgmsg_purge(0);
 	mutex_exit(&zfs_dbgmsgs_lock);
 	mutex_destroy(&zfs_dbgmsgs_lock);
+#endif
 }
 
 void

module/zfs/zfs_ioctl.c

@@ -4191,6 +4191,15 @@ zfs_check_settable(const char *dsname, nvpair_t *pair, cred_t *cr)
 		}
 		break;
 
+	case ZFS_PROP_SPECIAL_SMALL_BLOCKS:
+		/*
+		 * This property could require the allocation classes
+		 * feature to be active for setting, however we allow
+		 * it so that tests of settable properties succeed.
+		 * The CLI will issue a warning in this case.
+		 */
+		break;
+
 	case ZFS_PROP_SHARESMB:
 		if (zpl_earlier_version(dsname, ZPL_VERSION_FUID))
 			return (SET_ERROR(ENOTSUP));

module/zfs/zio.c

@@ -22,6 +22,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
+ * Copyright (c) 2017, Intel Corporation.
  */
 
 #include <sys/sysmacros.h>
@@ -825,6 +826,8 @@ zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 		zio->io_bookmark = *zb;
 
 	if (pio != NULL) {
+		if (zio->io_metaslab_class == NULL)
+			zio->io_metaslab_class = pio->io_metaslab_class;
 		if (zio->io_logical == NULL)
 			zio->io_logical = pio->io_logical;
 		if (zio->io_child_type == ZIO_CHILD_GANG)
@@ -1315,9 +1318,8 @@ zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset,
 	 */
 	if (flags & ZIO_FLAG_IO_ALLOCATING &&
 	    (vd != vd->vdev_top || (flags & ZIO_FLAG_IO_RETRY))) {
-		ASSERTV(metaslab_class_t *mc = spa_normal_class(pio->io_spa));
+		ASSERT(pio->io_metaslab_class != NULL);
-
+		ASSERT(pio->io_metaslab_class->mc_alloc_throttle_enabled);
-		ASSERT(mc->mc_alloc_throttle_enabled);
 		ASSERT(type == ZIO_TYPE_WRITE);
 		ASSERT(priority == ZIO_PRIORITY_ASYNC_WRITE);
 		ASSERT(!(flags & ZIO_FLAG_IO_REPAIR));
@@ -1644,8 +1646,9 @@ zio_write_compress(zio_t *zio)
 	if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg &&
 	    BP_GET_PSIZE(bp) == psize &&
 	    pass >= zfs_sync_pass_rewrite) {
-		ASSERT(psize != 0);
+		VERIFY3U(psize, !=, 0);
 		enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
+
 		zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages;
 		zio->io_flags |= ZIO_FLAG_IO_REWRITE;
 	} else {
@@ -3266,7 +3269,7 @@ zio_io_to_allocate(spa_t *spa, int allocator)
 	 * reserve then we throttle.
 	 */
 	ASSERT3U(zio->io_allocator, ==, allocator);
-	if (!metaslab_class_throttle_reserve(spa_normal_class(spa),
+	if (!metaslab_class_throttle_reserve(zio->io_metaslab_class,
 	    zio->io_prop.zp_copies, zio->io_allocator, zio, 0)) {
 		return (NULL);
 	}
@@ -3282,9 +3285,14 @@ zio_dva_throttle(zio_t *zio)
 {
 	spa_t *spa = zio->io_spa;
 	zio_t *nio;
+	metaslab_class_t *mc;
+
+	/* locate an appropriate allocation class */
+	mc = spa_preferred_class(spa, zio->io_size, zio->io_prop.zp_type,
+	    zio->io_prop.zp_level, zio->io_prop.zp_zpl_smallblk);
 
 	if (zio->io_priority == ZIO_PRIORITY_SYNC_WRITE ||
-	    !spa_normal_class(zio->io_spa)->mc_alloc_throttle_enabled ||
+	    !mc->mc_alloc_throttle_enabled ||
 	    zio->io_child_type == ZIO_CHILD_GANG ||
 	    zio->io_flags & ZIO_FLAG_NODATA) {
 		return (zio);
@@ -3306,17 +3314,15 @@ zio_dva_throttle(zio_t *zio)
 	zio->io_allocator = cityhash4(bm->zb_objset, bm->zb_object,
 	    bm->zb_level, bm->zb_blkid >> 20) % spa->spa_alloc_count;
 	mutex_enter(&spa->spa_alloc_locks[zio->io_allocator]);
-
 	ASSERT(zio->io_type == ZIO_TYPE_WRITE);
+	zio->io_metaslab_class = mc;
 	avl_add(&spa->spa_alloc_trees[zio->io_allocator], zio);
-
+	nio = zio_io_to_allocate(spa, zio->io_allocator);
-	nio = zio_io_to_allocate(zio->io_spa, zio->io_allocator);
 	mutex_exit(&spa->spa_alloc_locks[zio->io_allocator]);
-
 	return (nio);
 }
 
-void
+static void
 zio_allocate_dispatch(spa_t *spa, int allocator)
 {
 	zio_t *zio;
@@ -3336,7 +3342,7 @@ static zio_t *
 zio_dva_allocate(zio_t *zio)
 {
 	spa_t *spa = zio->io_spa;
-	metaslab_class_t *mc = spa_normal_class(spa);
+	metaslab_class_t *mc;
 	blkptr_t *bp = zio->io_bp;
 	int error;
 	int flags = 0;
@@ -3360,10 +3366,50 @@ zio_dva_allocate(zio_t *zio)
 	if (zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE)
 		flags |= METASLAB_ASYNC_ALLOC;
 
+	/*
+	 * if not already chosen, locate an appropriate allocation class
+	 */
+	mc = zio->io_metaslab_class;
+	if (mc == NULL) {
+		mc = spa_preferred_class(spa, zio->io_size,
+		    zio->io_prop.zp_type, zio->io_prop.zp_level,
+		    zio->io_prop.zp_zpl_smallblk);
+		zio->io_metaslab_class = mc;
+	}
+
 	error = metaslab_alloc(spa, mc, zio->io_size, bp,
 	    zio->io_prop.zp_copies, zio->io_txg, NULL, flags,
 	    &zio->io_alloc_list, zio, zio->io_allocator);
 
+	/*
+	 * Fallback to normal class when an alloc class is full
+	 */
+	if (error == ENOSPC && mc != spa_normal_class(spa)) {
+		/*
+		 * If throttling, transfer reservation over to normal class.
+		 * The io_allocator slot can remain the same even though we
+		 * are switching classes.
+		 */
+		if (mc->mc_alloc_throttle_enabled &&
+		    (zio->io_flags & ZIO_FLAG_IO_ALLOCATING)) {
+			metaslab_class_throttle_unreserve(mc,
+			    zio->io_prop.zp_copies, zio->io_allocator, zio);
+			zio->io_flags &= ~ZIO_FLAG_IO_ALLOCATING;
+
+			mc = spa_normal_class(spa);
+			VERIFY(metaslab_class_throttle_reserve(mc,
+			    zio->io_prop.zp_copies, zio->io_allocator, zio,
+			    flags | METASLAB_MUST_RESERVE));
+		} else {
+			mc = spa_normal_class(spa);
+		}
+		zio->io_metaslab_class = mc;
+
+		error = metaslab_alloc(spa, mc, zio->io_size, bp,
+		    zio->io_prop.zp_copies, zio->io_txg, NULL, flags,
+		    &zio->io_alloc_list, zio, zio->io_allocator);
+	}
+
 	if (error != 0) {
 		zfs_dbgmsg("%s: metaslab allocation failure: zio %p, "
 		    "size %llu, error %d", spa_name(spa), zio, zio->io_size,
@@ -3431,6 +3477,15 @@ zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, blkptr_t *new_bp,
 	ASSERT(txg > spa_syncing_txg(spa));
 
 	metaslab_trace_init(&io_alloc_list);
+
+	/*
+	 * Block pointer fields are useful to metaslabs for stats and debugging.
+	 * Fill in the obvious ones before calling into metaslab_alloc().
+	 */
+	BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG);
+	BP_SET_PSIZE(new_bp, size);
+	BP_SET_LEVEL(new_bp, 0);
+
 	/*
 	 * When allocating a zil block, we don't have information about
 	 * the final destination of the block except the objset it's part
@@ -4144,13 +4199,15 @@ zio_ready(zio_t *zio)
 		if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
 			ASSERT(IO_IS_ALLOCATING(zio));
 			ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
+			ASSERT(zio->io_metaslab_class != NULL);
+
 			/*
 			 * We were unable to allocate anything, unreserve and
 			 * issue the next I/O to allocate.
 			 */
 			metaslab_class_throttle_unreserve(
-			    spa_normal_class(zio->io_spa),
+			    zio->io_metaslab_class, zio->io_prop.zp_copies,
-			    zio->io_prop.zp_copies, zio->io_allocator, zio);
+			    zio->io_allocator, zio);
 			zio_allocate_dispatch(zio->io_spa, zio->io_allocator);
 		}
 	}
@@ -4233,14 +4290,15 @@ zio_dva_throttle_done(zio_t *zio)
 	ASSERT(zio->io_logical != NULL);
 	ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REPAIR));
 	ASSERT0(zio->io_flags & ZIO_FLAG_NOPWRITE);
+	ASSERT(zio->io_metaslab_class != NULL);
 
 	mutex_enter(&pio->io_lock);
 	metaslab_group_alloc_decrement(zio->io_spa, vd->vdev_id, pio, flags,
 	    pio->io_allocator, B_TRUE);
 	mutex_exit(&pio->io_lock);
 
-	metaslab_class_throttle_unreserve(spa_normal_class(zio->io_spa),
+	metaslab_class_throttle_unreserve(zio->io_metaslab_class, 1,
-	    1, pio->io_allocator, pio);
+	    pio->io_allocator, pio);
 
 	/*
 	 * Call into the pipeline to see if there is more work that
@@ -4259,7 +4317,6 @@ zio_done(zio_t *zio)
 	 */
 	const uint64_t psize = zio->io_size;
 	zio_t *pio, *pio_next;
-	ASSERTV(metaslab_class_t *mc = spa_normal_class(zio->io_spa));
 	zio_link_t *zl = NULL;
 
 	/*
@@ -4278,7 +4335,8 @@ zio_done(zio_t *zio)
 	 */
 	if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING &&
 	    zio->io_child_type == ZIO_CHILD_VDEV) {
-		ASSERT(mc->mc_alloc_throttle_enabled);
+		ASSERT(zio->io_metaslab_class != NULL);
+		ASSERT(zio->io_metaslab_class->mc_alloc_throttle_enabled);
 		zio_dva_throttle_done(zio);
 	}
 
@@ -4290,9 +4348,11 @@ zio_done(zio_t *zio)
 		ASSERT(zio->io_type == ZIO_TYPE_WRITE);
 		ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
 		ASSERT(zio->io_bp != NULL);
+
 		metaslab_group_alloc_verify(zio->io_spa, zio->io_bp, zio,
 		    zio->io_allocator);
-		VERIFY(refcount_not_held(&mc->mc_alloc_slots[zio->io_allocator],
+		VERIFY(refcount_not_held(
+		    &zio->io_metaslab_class->mc_alloc_slots[zio->io_allocator],
 		    zio));
 	}
 

tests/runfiles/linux.run

@@ -24,6 +24,14 @@ tags = ['functional']
 tests = ['posix_001_pos', 'posix_002_pos', 'posix_003_pos']
 tags = ['functional', 'acl', 'posix']
 
+[tests/functional/alloc_class]
+tests = ['alloc_class_001_pos', 'alloc_class_002_neg', 'alloc_class_003_pos',
+    'alloc_class_004_pos', 'alloc_class_005_pos', 'alloc_class_006_pos',
+    'alloc_class_007_pos', 'alloc_class_008_pos', 'alloc_class_009_pos',
+    'alloc_class_010_pos', 'alloc_class_011_neg', 'alloc_class_012_pos',
+    'alloc_class_013_pos']
+tags = ['functional', 'alloc_class']
+
 [tests/functional/arc]
 tests = ['dbufstats_001_pos', 'dbufstats_002_pos']
 tags = ['functional', 'arc']

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

@@ -1,5 +1,6 @@
 SUBDIRS = \
 	acl \
+	alloc_class \
 	arc \
 	atime \
 	bootfs \

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

@@ -0,0 +1,19 @@
+pkgdatadir = $(datadir)/@PACKAGE@/zfs-tests/tests/functional/alloc_class
+dist_pkgdata_SCRIPTS = \
+	alloc_class.cfg \
+	alloc_class.kshlib \
+	setup.ksh \
+	cleanup.ksh \
+	alloc_class_001_pos.ksh \
+	alloc_class_002_neg.ksh \
+	alloc_class_003_pos.ksh \
+	alloc_class_004_pos.ksh \
+	alloc_class_005_pos.ksh \
+	alloc_class_006_pos.ksh \
+	alloc_class_007_pos.ksh \
+	alloc_class_008_pos.ksh \
+	alloc_class_009_pos.ksh \
+	alloc_class_010_pos.ksh \
+	alloc_class_011_neg.ksh \
+	alloc_class_012_pos.ksh \
+	alloc_class_013_pos.ksh

tests/zfs-tests/tests/functional/alloc_class/alloc_class.cfg

@@ -0,0 +1,31 @@
+#
+# 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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+
+export ZPOOL_DISK0="$TEST_BASE_DIR/device-0"
+export ZPOOL_DISK1="$TEST_BASE_DIR/device-1"
+export ZPOOL_DISK2="$TEST_BASE_DIR/device-2"
+export ZPOOL_DISKS="${ZPOOL_DISK0} ${ZPOOL_DISK1} ${ZPOOL_DISK2}"
+
+export CLASS_DISK0="$TEST_BASE_DIR/device-3"
+export CLASS_DISK1="$TEST_BASE_DIR/device-4"
+export CLASS_DISK2="$TEST_BASE_DIR/device-5"
+export CLASS_DISK3="$TEST_BASE_DIR/device-6"
+export CLASS_DISKS="${CLASS_DISK0} ${CLASS_DISK1} ${CLASS_DISK2} ${CLASS_DISK3}"
+
+export ZPOOL_DEVSIZE=1G
+export CLASS_DEVSIZE=512M

tests/zfs-tests/tests/functional/alloc_class/alloc_class.kshlib

@@ -0,0 +1,67 @@
+#
+# 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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.cfg
+
+function disk_setup
+{
+	truncate -s $ZPOOL_DEVSIZE $ZPOOL_DISKS
+	truncate -s $CLASS_DEVSIZE $CLASS_DISKS
+}
+
+function disk_cleanup
+{
+	rm -f $ZPOOL_DEVSIZE $ZPOOL_DISKS 2> /dev/null
+	rm -f $CLASS_DEVSIZE $CLASS_DISKS 2> /dev/null
+}
+
+function cleanup
+{
+	if datasetexists $TESTPOOL ; then
+		zpool destroy -f $TESTPOOL 2> /dev/null
+	fi
+
+	disk_cleanup
+}
+
+#
+# Try zpool status/iostat for given pool
+#
+# $1 pool
+#
+function display_status
+{
+	typeset pool=$1
+
+	typeset -i ret=0
+	zpool status -xv $pool > /dev/null 2>&1
+	ret=$?
+
+	zpool iostat > /dev/null 2>&1
+	((ret |= $?))
+
+	typeset mntpnt=$(get_prop mountpoint $pool)
+	dd if=/dev/random of=$mntpnt/testfile.$$ &
+	typeset pid=$!
+
+	zpool iostat -v 1 3 > /dev/null
+	((ret |= $?))
+
+	kill -9 $pid
+
+	return $ret
+}

tests/zfs-tests/tests/functional/alloc_class/alloc_class_001_pos.ksh

@@ -0,0 +1,39 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Creating a pool with a special device succeeds.
+#
+
+verify_runnable "global"
+
+claim="Creating a pool with a special device succeeds."
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+log_must zpool create $TESTPOOL raidz $ZPOOL_DISKS special mirror \
+    $CLASS_DISK0 $CLASS_DISK1
+log_must display_status "$TESTPOOL"
+log_must zpool destroy -f "$TESTPOOL"
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_002_neg.ksh

@@ -0,0 +1,48 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Creating a pool fails when an invalid special device type is specified.
+#
+claim="Creating a pool with an invalid special device type fails."
+
+verify_runnable "global"
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+log_mustnot zpool create $TESTPOOL raidz $ZPOOL_DISKS special $CLASS_DISK0
+log_mustnot display_status $TESTPOOL
+log_mustnot zpool destroy -f $TESTPOOL
+
+log_mustnot zpool create $TESTPOOL $ZPOOL_DISKS special mirror \
+    $CLASS_DISK0 $CLASS_DISK1
+log_mustnot display_status $TESTPOOL
+log_mustnot zpool destroy -f $TESTPOOL
+
+log_mustnot zpool create $TESTPOOL raidz $ZPOOL_DISKS special raidz \
+    $CLASS_DISK0 $CLASS_DISK1 $CLASS_DISK2
+log_mustnot display_status $TESTPOOL
+log_mustnot zpool destroy -f $TESTPOOL
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_003_pos.ksh

@@ -0,0 +1,57 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Adding a special device to a normal pool succeeds.
+#
+claim="Adding a special device to a normal pool succeeds."
+
+verify_runnable "global"
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+for type in "" "mirror" "raidz"
+do
+	log_must zpool create $TESTPOOL $type $ZPOOL_DISKS
+
+	if [ "$type" = "mirror" ]; then
+		log_must zpool add $TESTPOOL special mirror \
+		    $CLASS_DISK0 $CLASS_DISK1 $CLASS_DISK2
+		log_must zpool iostat -H $TESTPOOL $CLASS_DISK0
+		log_must zpool iostat -H $TESTPOOL $CLASS_DISK1
+		log_must zpool iostat -H $TESTPOOL $CLASS_DISK2
+	elif [ "$type" = "raidz" ]; then
+		log_must zpool add $TESTPOOL special mirror \
+		    $CLASS_DISK0 $CLASS_DISK1
+		log_must zpool iostat -H $TESTPOOL $CLASS_DISK0
+		log_must zpool iostat -H $TESTPOOL $CLASS_DISK1
+	else
+		log_must zpool add $TESTPOOL special $CLASS_DISK0
+		log_must zpool iostat -H $TESTPOOL $CLASS_DISK0
+	fi
+
+	log_must zpool destroy -f $TESTPOOL
+done
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_004_pos.ksh

@@ -0,0 +1,67 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Checking if allocation_classes feature flag status is active after
+#	creating a pool with a special device.
+#
+claim="Checking active allocation classes feature flag status successful."
+
+verify_runnable "global"
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+typeset ac_value
+typeset stype=""
+typeset sdisks=""
+
+for type in "" "mirror" "raidz"
+do
+	if [ "$type" = "mirror" ]; then
+		stype="mirror"
+		sdisks="${CLASS_DISK0} ${CLASS_DISK1} ${CLASS_DISK2}"
+	elif [ "$type" = "raidz" ]; then
+		stype="mirror"
+		sdisks="${CLASS_DISK0} ${CLASS_DISK1}"
+	else
+		stype=""
+		sdisks="${CLASS_DISK0}"
+	fi
+
+	log_must zpool create $TESTPOOL $type $ZPOOL_DISKS \
+	    special $stype $sdisks
+
+	ac_value="$(zpool get all -H -o property,value | \
+	    egrep allocation_classes | nawk '{print $2}')"
+	if [ "$ac_value" = "active" ]; then
+		log_note "feature@allocation_classes is active"
+	else
+		log_fail "feature@allocation_classes not active, \
+		    status = $ac_value"
+	fi
+
+	log_must zpool destroy -f $TESTPOOL
+done
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_005_pos.ksh

@@ -0,0 +1,71 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Checking allocation_classes feature flag value after pool is created
+#	(should be enabled) and also after a special device added to existing
+#	pool (should be active).
+#
+
+verify_runnable "global"
+
+log_assert "Values of allocation_classes feature flag correct."
+log_onexit cleanup
+
+log_must disk_setup
+
+typeset ac_value
+
+for type in "" "mirror" "raidz"
+do
+	if [ "$type" = "mirror" ]; then
+		log_must zpool create $TESTPOOL $type $ZPOOL_DISK0 $ZPOOL_DISK1
+	else
+		log_must zpool create $TESTPOOL $type $ZPOOL_DISKS
+	fi
+	ac_value="$(zpool get all -H -o property,value | \
+	    egrep allocation_classes  | awk '{print $2}')"
+	if [ "$ac_value" = "enabled" ]; then
+		log_note "feature@allocation_classes is enabled"
+	else
+		log_fail "feature@allocation_classes not enabled, \
+		    status = $ac_value"
+	fi
+
+	if [ "$type" = "" ]; then
+		log_must zpool add $TESTPOOL special $CLASS_DISK0
+	else
+		log_must zpool add $TESTPOOL special mirror \
+		    $CLASS_DISK0 $CLASS_DISK1
+	fi
+	ac_value="$(zpool get all -H -o property,value | \
+	    egrep allocation_classes | awk '{print $2}')"
+	if [ "$ac_value" = "active" ]; then
+		log_note "feature@allocation_classes is active"
+	else
+		log_fail "feature@allocation_classes not active, \
+		    status = $ac_value"
+	fi
+
+	log_must zpool destroy -f $TESTPOOL
+done
+
+log_pass "Values of allocation_classes feature flag correct."

tests/zfs-tests/tests/functional/alloc_class/alloc_class_006_pos.ksh

@@ -0,0 +1,41 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Using zpool split command to detach disks from mirrored special pool
+#	to create a new pool with the detached disks.
+#
+claim="zpool split command succeeds with special devices present."
+
+verify_runnable "global"
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+log_must zpool create $TESTPOOL \
+    mirror $ZPOOL_DISK0 $ZPOOL_DISK1 \
+    special mirror $CLASS_DISK0 $CLASS_DISK1
+log_must zpool split $TESTPOOL split_pool
+log_must zpool destroy -f $TESTPOOL
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_007_pos.ksh

@@ -0,0 +1,41 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Replacing a special device succeeds
+#
+claim="Replacing a special device is successful."
+
+verify_runnable "global"
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+log_must zpool create $TESTPOOL raidz $ZPOOL_DISKS \
+   special mirror $CLASS_DISK0 $CLASS_DISK1
+log_must zpool replace $TESTPOOL $CLASS_DISK1 $CLASS_DISK2
+log_must sleep 10
+log_must zpool iostat -H $TESTPOOL $CLASS_DISK2
+log_must zpool destroy -f $TESTPOOL
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_008_pos.ksh

@@ -0,0 +1,56 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Adding an additional special device to a pool with special succeeds.
+#
+claim="Adding an additional special device succeeds."
+
+verify_runnable "global"
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+typeset special_type=""
+typeset create_disks=""
+typeset added_disks=""
+
+for type in "" "raidz"
+do
+	if [ "$type" = "raidz" ]; then
+		special_type="mirror"
+		create_disks="${CLASS_DISK0} ${CLASS_DISK1}"
+		added_disks="${CLASS_DISK2} ${CLASS_DISK3}"
+	else
+		special_type=""
+		create_disks="${CLASS_DISK0}"
+		added_disks="${CLASS_DISK1}"
+	fi
+	log_must zpool create $TESTPOOL $type $ZPOOL_DISKS \
+	    special $special_type $create_disks
+	log_must zpool add $TESTPOOL special $special_type $added_disks
+	log_must zpool iostat $TESTPOOL $added_disks
+	log_must zpool destroy -f $TESTPOOL
+done
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_009_pos.ksh

@@ -0,0 +1,58 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Importing and exporting pool with special device succeeds.
+#
+claim="Import/export of pool with special device mirror succeeds."
+
+verify_runnable "global"
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+typeset stype=""
+typeset sdisks=""
+
+for type in "" "mirror" "raidz"
+do
+	if [ "$type" = "mirror" ]; then
+		stype="mirror"
+		sdisks="${CLASS_DISK0} ${CLASS_DISK1} ${CLASS_DISK2}"
+	elif [ "$type" = "raidz" ]; then
+		stype="mirror"
+		sdisks="${CLASS_DISK0} ${CLASS_DISK1}"
+	else
+		stype=""
+		special_args="${CLASS_DISK0}"
+	fi
+
+	log_must zpool create $TESTPOOL $type $ZPOOL_DISKS \
+	    special $stype $sdisks
+	log_must zpool export $TESTPOOL
+	log_must zpool import -d $TEST_BASE_DIR -s $TESTPOOL
+	log_must display_status $TESTPOOL
+	log_must zpool destroy -f $TESTPOOL
+done
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_010_pos.ksh

@@ -0,0 +1,50 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Setting the special_small_blocks property to a valid value succeeds.
+#
+
+verify_runnable "global"
+
+claim="Setting the special_small_blocks property to a valid value succeeds."
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+
+log_must zpool create $TESTPOOL raidz $ZPOOL_DISKS special mirror \
+	$CLASS_DISK0 $CLASS_DISK1
+
+for value in 0 512 1024 2048 4096 8192 16384 32768 65536 131072
+do
+	log_must zfs set special_small_blocks=$value $TESTPOOL
+	ACTUAL=$(zfs get -p special_small_blocks $TESTPOOL | \
+		grep special_small_blocks | awk '{print $3}')
+	if [ "$ACTUAL" != "$value" ]
+	then
+		log_fail "v. $ACTUAL set for $TESTPOOL, expected v. $value!"
+	fi
+done
+
+log_must zpool destroy -f "$TESTPOOL"
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_011_neg.ksh

@@ -0,0 +1,43 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Setting the special_small_blocks property to invalid values fails.
+#
+
+verify_runnable "global"
+
+claim="Setting the special_small_blocks property to invalid values fails."
+
+log_assert $claim
+log_onexit cleanup
+
+log_must disk_setup
+log_must zpool create $TESTPOOL raidz $ZPOOL_DISKS special mirror \
+	$CLASS_DISK0 $CLASS_DISK1
+
+for value in 256 1025 262144
+do
+	log_mustnot zfs set special_small_blocks=$value $TESTPOOL
+done
+
+log_must zpool destroy -f "$TESTPOOL"
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_012_pos.ksh

@@ -0,0 +1,69 @@
+#!/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) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Removing a special device from a pool succeeds.
+#
+
+verify_runnable "global"
+
+claim= "Removing a special device from a pool succeeds."
+
+log_assert $claim
+log_onexit cleanup
+
+#
+# Create a non-raidz pool so we can remove top-level vdevs
+#
+log_must disk_setup
+log_must zpool create $TESTPOOL $ZPOOL_DISK0 $ZPOOL_DISK1 $ZPOOL_DISK2 \
+  special $CLASS_DISK0 special $CLASS_DISK1
+log_must display_status "$TESTPOOL"
+
+#
+# Generate some metadata and small blocks in the special class before removal
+#
+typeset -l i=1
+typeset -l blocks=25
+
+log_must zfs create -o special_small_blocks=32K -o recordsize=32K \
+	$TESTPOOL/$TESTFS
+for i in 1 2 3 4; do
+	log_must dd if=/dev/urandom of=/$TESTPOOL/$TESTFS/testfile.$i bs=1M \
+	    count=$blocks
+	((blocks = blocks + 25))
+done
+log_must sync_pool $TESTPOOL
+log_must zpool list -v $TESTPOOL
+
+#
+# remove a special allocation vdev and force a remapping
+#
+log_must zpool remove $TESTPOOL $CLASS_DISK0
+log_must zfs remap $TESTPOOL/$TESTFS
+
+sleep 5
+log_must sync_pool $TESTPOOL
+sleep 1
+
+log_must zdb -bbcc $TESTPOOL
+log_must zpool destroy -f "$TESTPOOL"
+
+log_pass $claim

tests/zfs-tests/tests/functional/alloc_class/alloc_class_013_pos.ksh

@@ -0,0 +1,63 @@
+#!/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) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+#
+# DESCRIPTION:
+#	Removing a dedup device from a pool succeeds.
+#
+
+verify_runnable "global"
+
+claim= "Removing a dedup device from a pool succeeds."
+
+log_assert $claim
+log_onexit cleanup
+
+#
+# Create a non-raidz pool so we can remove top-level vdevs
+#
+log_must disk_setup
+log_must zpool create $TESTPOOL $ZPOOL_DISKS dedup $CLASS_DISK0
+log_must display_status "$TESTPOOL"
+
+#
+# Generate some dedup data in the dedup class before removal
+#
+
+log_must zfs create -o dedup=on -V 2G $TESTPOOL/$TESTVOL
+
+log_must echo y | newfs $ZVOL_DEVDIR/$TESTPOOL/$TESTVOL >/dev/null 2>&1
+
+sync_pool
+log_must zpool list -v $TESTPOOL
+
+#
+# remove a dedup allocation vdev
+#
+log_must zpool remove $TESTPOOL $CLASS_DISK0
+
+sleep 5
+log_must sync_pool $TESTPOOL
+sleep 1
+
+log_must zdb -bbcc $TESTPOOL
+
+log_must zpool destroy -f "$TESTPOOL"
+
+log_pass $claim

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

@@ -0,0 +1,27 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018, Delphix
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+verify_runnable "global"
+
+pool_cleanup
+disk_cleanup
+
+log_pass

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

@@ -0,0 +1,26 @@
+#!/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) 2017, Intel Corporation.
+# Copyright (c) 2018 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/alloc_class/alloc_class.kshlib
+
+verify_runnable "global"
+
+disk_cleanup
+
+log_pass

tests/zfs-tests/tests/functional/cli_root/zpool_get/zpool_get.cfg

@@ -86,5 +86,6 @@ if is_linux; then
 	    "feature@userobj_accounting"
 	    "feature@encryption"
 	    "feature@project_quota"
+	    "feature@allocation_classes"
 	)
 fi

tests/zfs-tests/tests/functional/cli_root/zpool_split/zpool_split_vdevs.ksh

@@ -50,6 +50,7 @@ function cleanup
 #  "l" - log
 # "ll" - mirrored log
 #  "c" - cache
+#  "sc" - special class
 #
 function pool_config # <vdev-type>
 {
@@ -66,6 +67,7 @@ function pool_config # <vdev-type>
 	disks[l]="l1"
 	disks[ll]="l1 l2"
 	disks[c]="c1"
+	disks[sc]="sc1 sc2"
 	case $1 in
 	d|t) # single disk or stripe
 		vdev='' ;;
@@ -85,6 +87,8 @@ function pool_config # <vdev-type>
 		vdev='log mirror';;
 	c) # cache
 		vdev='cache';;
+	sc) # mirrored special class
+		vdev='special mirror';;
 	*)
 		log_fail "setup_pool: unsupported vdev type '$1'"
 	esac
@@ -108,7 +112,7 @@ log_onexit cleanup
 # "good" and "bad" pool layouts
 # first token is always used with "zpool create"
 # second to last tokens, if any, are used with "zpool add"
-typeset -a goodconfs=("m" "m l" "m s" "m c" "m m" "m3" "m3 m3" "m m3 l s c")
+typeset -a goodconfs=("m" "m l" "m s" "m c" "m m" "m3" "m3 m3" "m m3 l s c" "m m sc")
 typeset -a badconfs=("d" "z1" "z2" "z3" "m d" "m3 d" "m z1" "m z2" "m z3")
 typeset FILEDEV_PREFIX="$TEST_BASE_DIR/filedev"
 typeset altroot="$TESTDIR/altroot-$TESTPOOL2"