Illumos #3464

3464 zfs synctask code needs restructuring
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>

References:
  https://www.illumos.org/issues/3464
  illumos/illumos-gate@3b2aab1880

Ported-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #1495

cmd/zdb/zdb.c

@@ -1725,7 +1725,9 @@ dump_dir(objset_t *os)
 	int print_header = 1;
 	int i, error;
 
+	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
 	dmu_objset_fast_stat(os, &dds);
+	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
 
 	if (dds.dds_type < DMU_OST_NUMTYPES)
 		type = objset_types[dds.dds_type];
@@ -2171,7 +2173,6 @@ zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 
 		zio_nowait(zio_read(NULL, spa, bp, data, size,
 		    zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
-
 	}
 
 	zcb->zcb_readfails = 0;
@@ -2365,8 +2366,10 @@ dump_block_stats(spa_t *spa)
 	 */
 	(void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
 	    count_block_cb, &zcb, NULL);
-	(void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
+	if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
-	    count_block_cb, &zcb, NULL);
+		(void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
+		    count_block_cb, &zcb, NULL);
+	}
 	if (spa_feature_is_active(spa,
 	    &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
 		VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,

cmd/zfs/zfs_main.c

@@ -892,6 +892,7 @@ typedef struct destroy_cbdata {
 	boolean_t	cb_parsable;
 	boolean_t	cb_dryrun;
 	nvlist_t	*cb_nvl;
+	nvlist_t	*cb_batchedsnaps;
 
 	/* first snap in contiguous run */
 	char		*cb_firstsnap;
@@ -988,9 +989,27 @@ destroy_callback(zfs_handle_t *zhp, void *data)
 		zfs_close(zhp);
 		return (0);
 	}
+	if (cb->cb_dryrun) {
+		zfs_close(zhp);
+		return (0);
+	}
 
-	if (!cb->cb_dryrun) {
+	/*
-		if (zfs_unmount(zhp, NULL, cb->cb_force ? MS_FORCE : 0) != 0 ||
+	 * We batch up all contiguous snapshots (even of different
+	 * filesystems) and destroy them with one ioctl.  We can't
+	 * simply do all snap deletions and then all fs deletions,
+	 * because we must delete a clone before its origin.
+	 */
+	if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT) {
+		fnvlist_add_boolean(cb->cb_batchedsnaps, name);
+	} else {
+		int error = zfs_destroy_snaps_nvl(g_zfs,
+		    cb->cb_batchedsnaps, B_FALSE);
+		fnvlist_free(cb->cb_batchedsnaps);
+		cb->cb_batchedsnaps = fnvlist_alloc();
+
+		if (error != 0 ||
+		    zfs_unmount(zhp, NULL, cb->cb_force ? MS_FORCE : 0) != 0 ||
 		    zfs_destroy(zhp, cb->cb_defer_destroy) != 0) {
 			zfs_close(zhp);
 			return (-1);
@@ -1146,8 +1165,10 @@ static int
 zfs_do_destroy(int argc, char **argv)
 {
 	destroy_cbdata_t cb = { 0 };
+	int rv = 0;
+	int err = 0;
 	int c;
-	zfs_handle_t *zhp;
+	zfs_handle_t *zhp = NULL;
 	char *at;
 	zfs_type_t type = ZFS_TYPE_DATASET;
 
@@ -1201,11 +1222,9 @@ zfs_do_destroy(int argc, char **argv)
 
 	at = strchr(argv[0], '@');
 	if (at != NULL) {
-		int err = 0;
 
 		/* Build the list of snaps to destroy in cb_nvl. */
-		if (nvlist_alloc(&cb.cb_nvl, NV_UNIQUE_NAME, 0) != 0)
+		cb.cb_nvl = fnvlist_alloc();
-			nomem();
 
 		*at = '\0';
 		zhp = zfs_open(g_zfs, argv[0],
@@ -1216,17 +1235,15 @@ zfs_do_destroy(int argc, char **argv)
 		cb.cb_snapspec = at + 1;
 		if (gather_snapshots(zfs_handle_dup(zhp), &cb) != 0 ||
 		    cb.cb_error) {
-			zfs_close(zhp);
+			rv = 1;
-			nvlist_free(cb.cb_nvl);
+			goto out;
-			return (1);
 		}
 
 		if (nvlist_empty(cb.cb_nvl)) {
 			(void) fprintf(stderr, gettext("could not find any "
 			    "snapshots to destroy; check snapshot names.\n"));
-			zfs_close(zhp);
+			rv = 1;
-			nvlist_free(cb.cb_nvl);
+			goto out;
-			return (1);
 		}
 
 		if (cb.cb_verbose) {
@@ -1245,18 +1262,26 @@ zfs_do_destroy(int argc, char **argv)
 		}
 
 		if (!cb.cb_dryrun) {
-			if (cb.cb_doclones)
+			if (cb.cb_doclones) {
+				cb.cb_batchedsnaps = fnvlist_alloc();
 				err = destroy_clones(&cb);
+				if (err == 0) {
+					err = zfs_destroy_snaps_nvl(g_zfs,
+					    cb.cb_batchedsnaps, B_FALSE);
+				}
+				if (err != 0) {
+					rv = 1;
+					goto out;
+				}
+			}
 			if (err == 0) {
-				err = zfs_destroy_snaps_nvl(zhp, cb.cb_nvl,
+				err = zfs_destroy_snaps_nvl(g_zfs, cb.cb_nvl,
 				    cb.cb_defer_destroy);
 			}
 		}
 
-		zfs_close(zhp);
-		nvlist_free(cb.cb_nvl);
 		if (err != 0)
-			return (1);
+			rv = 1;
 	} else {
 		/* Open the given dataset */
 		if ((zhp = zfs_open(g_zfs, argv[0], type)) == NULL)
@@ -1277,8 +1302,8 @@ zfs_do_destroy(int argc, char **argv)
 			    zfs_get_name(zhp));
 			(void) fprintf(stderr, gettext("use 'zpool destroy %s' "
 			    "to destroy the pool itself\n"), zfs_get_name(zhp));
-			zfs_close(zhp);
+			rv = 1;
-			return (1);
+			goto out;
 		}
 
 		/*
@@ -1288,30 +1313,42 @@ zfs_do_destroy(int argc, char **argv)
 		if (!cb.cb_doclones &&
 		    zfs_iter_dependents(zhp, B_TRUE, destroy_check_dependent,
 		    &cb) != 0) {
-			zfs_close(zhp);
+			rv = 1;
-			return (1);
+			goto out;
 		}
 
 		if (cb.cb_error) {
-			zfs_close(zhp);
+			rv = 1;
-			return (1);
+			goto out;
 		}
 
+		cb.cb_batchedsnaps = fnvlist_alloc();
 		if (zfs_iter_dependents(zhp, B_FALSE, destroy_callback,
 		    &cb) != 0) {
-			zfs_close(zhp);
+			rv = 1;
-			return (1);
+			goto out;
 		}
 
 		/*
 		 * Do the real thing.  The callback will close the
 		 * handle regardless of whether it succeeds or not.
 		 */
-		if (destroy_callback(zhp, &cb) != 0)
+		err = destroy_callback(zhp, &cb);
-			return (1);
+		zhp = NULL;
+		if (err == 0) {
+			err = zfs_destroy_snaps_nvl(g_zfs,
+			    cb.cb_batchedsnaps, cb.cb_defer_destroy);
+		}
+		if (err != 0)
+			rv = 1;
 	}
 
-	return (0);
+out:
+	fnvlist_free(cb.cb_batchedsnaps);
+	fnvlist_free(cb.cb_nvl);
+	if (zhp != NULL)
+		zfs_close(zhp);
+	return (rv);
 }
 
 static boolean_t
@@ -5081,28 +5118,12 @@ cleanup2:
 	return (error);
 }
 
-/*
- * zfs allow [-r] [-t] <tag> <snap> ...
- *
- *	-r	Recursively hold
- *	-t	Temporary hold (hidden option)
- *
- * Apply a user-hold with the given tag to the list of snapshots.
- */
 static int
 zfs_do_allow(int argc, char **argv)
 {
 	return (zfs_do_allow_unallow_impl(argc, argv, B_FALSE));
 }
 
-/*
- * zfs unallow [-r] [-t] <tag> <snap> ...
- *
- *	-r	Recursively hold
- *	-t	Temporary hold (hidden option)
- *
- * Apply a user-hold with the given tag to the list of snapshots.
- */
 static int
 zfs_do_unallow(int argc, char **argv)
 {
@@ -5116,7 +5137,6 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding)
 	int i;
 	const char *tag;
 	boolean_t recursive = B_FALSE;
-	boolean_t temphold = B_FALSE;
 	const char *opts = holding ? "rt" : "r";
 	int c;
 
@@ -5126,9 +5146,6 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding)
 		case 'r':
 			recursive = B_TRUE;
 			break;
-		case 't':
-			temphold = B_TRUE;
-			break;
 		case '?':
 			(void) fprintf(stderr, gettext("invalid option '%c'\n"),
 			    optopt);
@@ -5177,7 +5194,7 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding)
 		}
 		if (holding) {
 			if (zfs_hold(zhp, delim+1, tag, recursive,
-			    temphold, B_FALSE, -1, 0, 0) != 0)
+			    B_FALSE, -1) != 0)
 				++errors;
 		} else {
 			if (zfs_release(zhp, delim+1, tag, recursive) != 0)
@@ -5193,7 +5210,6 @@ zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding)
  * zfs hold [-r] [-t] <tag> <snap> ...
  *
  *	-r	Recursively hold
- *	-t	Temporary hold (hidden option)
  *
  * Apply a user-hold with the given tag to the list of snapshots.
  */

cmd/zhack/zhack.c

@@ -46,6 +46,7 @@
 #include <sys/zio_checksum.h>
 #include <sys/zio_compress.h>
 #include <sys/zfeature.h>
+#include <sys/dmu_tx.h>
 #undef ZFS_MAXNAMELEN
 #include <libzfs.h>
 
@@ -123,7 +124,7 @@ import_pool(const char *target, boolean_t readonly)
 	spa_t *spa;
 	nvpair_t *elem;
 	nvlist_t *props;
-	const char *name;
+	char *name;
 
 	kernel_init(readonly ? FREAD : (FREAD | FWRITE));
 	g_zfs = libzfs_init();
@@ -273,10 +274,10 @@ zhack_do_feature_stat(int argc, char **argv)
 }
 
 static void
-feature_enable_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+feature_enable_sync(void *arg, dmu_tx_t *tx)
 {
-	spa_t *spa = arg1;
+	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
-	zfeature_info_t *feature = arg2;
+	zfeature_info_t *feature = arg;
 
 	spa_feature_enable(spa, feature, tx);
 	spa_history_log_internal(spa, "zhack enable feature", tx,
@@ -344,8 +345,8 @@ zhack_do_feature_enable(int argc, char **argv)
 	if (0 == zap_contains(mos, spa->spa_feat_desc_obj, feature.fi_guid))
 		fatal("feature already enabled: %s", feature.fi_guid);
 
-	VERIFY3U(0, ==, dsl_sync_task_do(spa->spa_dsl_pool, NULL,
+	VERIFY0(dsl_sync_task(spa_name(spa), NULL,
-	    feature_enable_sync, spa, &feature, 5));
+	    feature_enable_sync, &feature, 5));
 
 	spa_close(spa, FTAG);
 
@@ -353,10 +354,10 @@ zhack_do_feature_enable(int argc, char **argv)
 }
 
 static void
-feature_incr_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+feature_incr_sync(void *arg, dmu_tx_t *tx)
 {
-	spa_t *spa = arg1;
+	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
-	zfeature_info_t *feature = arg2;
+	zfeature_info_t *feature = arg;
 
 	spa_feature_incr(spa, feature, tx);
 	spa_history_log_internal(spa, "zhack feature incr", tx,
@@ -364,10 +365,10 @@ feature_incr_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 }
 
 static void
-feature_decr_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+feature_decr_sync(void *arg, dmu_tx_t *tx)
 {
-	spa_t *spa = arg1;
+	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
-	zfeature_info_t *feature = arg2;
+	zfeature_info_t *feature = arg;
 
 	spa_feature_decr(spa, feature, tx);
 	spa_history_log_internal(spa, "zhack feature decr", tx,
@@ -442,8 +443,8 @@ zhack_do_feature_ref(int argc, char **argv)
 	if (decr && !spa_feature_is_active(spa, &feature))
 		fatal("feature refcount already 0: %s", feature.fi_guid);
 
-	VERIFY3U(0, ==, dsl_sync_task_do(spa->spa_dsl_pool, NULL,
+	VERIFY0(dsl_sync_task(spa_name(spa), NULL,
-	    decr ? feature_decr_sync : feature_incr_sync, spa, &feature, 5));
+	    decr ? feature_decr_sync : feature_incr_sync, &feature, 5));
 
 	spa_close(spa, FTAG);
 }

cmd/ztest/ztest.c

@@ -106,10 +106,12 @@
 #include <sys/metaslab_impl.h>
 #include <sys/dsl_prop.h>
 #include <sys/dsl_dataset.h>
+#include <sys/dsl_destroy.h>
 #include <sys/dsl_scan.h>
 #include <sys/zio_checksum.h>
 #include <sys/refcount.h>
 #include <sys/zfeature.h>
+#include <sys/dsl_userhold.h>
 #include <stdio.h>
 #include <stdio_ext.h>
 #include <stdlib.h>
@@ -367,7 +369,7 @@ ztest_info_t ztest_info[] = {
 	{ ztest_scrub,				1,	&zopt_rarely	},
 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
-	{ ztest_vdev_attach_detach,		1,	&zopt_rarely	},
+	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
 	{ ztest_vdev_add_remove,		1,
 	    &ztest_opts.zo_vdevtime				},
@@ -1031,9 +1033,8 @@ ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
 	uint64_t curval;
 	int error;
 
-	error = dsl_prop_set(osname, propname,
+	error = dsl_prop_set_int(osname, propname,
-	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
+	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
-	    sizeof (value), 1, &value);
 
 	if (error == ENOSPC) {
 		ztest_record_enospc(FTAG);
@@ -1042,8 +1043,7 @@ ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
 	ASSERT0(error);
 
 	setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
-	VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
+	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
-	    1, &curval, setpoint), ==, 0);
 
 	if (ztest_opts.zo_verbose >= 6) {
 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
@@ -2484,8 +2484,7 @@ ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
 	int error;
 
 	mutex_enter(&ztest_vdev_lock);
-	leaves =
+	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
-		MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
 
 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
 
@@ -2507,7 +2506,7 @@ ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
 		 * prevent a race between removing a slog (dmu_objset_find)
 		 * and destroying a dataset. Removing the slog will
 		 * grab a reference on the dataset which may cause
-		 * dmu_objset_destroy() to fail with EBUSY thus
+		 * dsl_destroy_head() to fail with EBUSY thus
 		 * leaving the dataset in an inconsistent state.
 		 */
 		rw_enter(&ztest_name_lock, RW_WRITER);
@@ -3196,7 +3195,7 @@ ztest_objset_destroy_cb(const char *name, void *arg)
 	/*
 	 * Verify that the dataset contains a directory object.
 	 */
-	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
+	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
 	if (error != ENOENT) {
 		/* We could have crashed in the middle of destroying it */
@@ -3204,12 +3203,16 @@ ztest_objset_destroy_cb(const char *name, void *arg)
 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
 	}
-	dmu_objset_rele(os, FTAG);
+	dmu_objset_disown(os, FTAG);
 
 	/*
 	 * Destroy the dataset.
 	 */
-	VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
+	if (strchr(name, '@') != NULL) {
+		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
+	} else {
+		VERIFY0(dsl_destroy_head(name));
+	}
 	return (0);
 }
 
@@ -3219,16 +3222,17 @@ ztest_snapshot_create(char *osname, uint64_t id)
 	char snapname[MAXNAMELEN];
 	int error;
 
-	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
+	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
-	    (u_longlong_t)id);
 
-	error = dmu_objset_snapshot_one(osname, strchr(snapname, '@') + 1);
+	error = dmu_objset_snapshot_one(osname, snapname);
 	if (error == ENOSPC) {
 		ztest_record_enospc(FTAG);
 		return (B_FALSE);
 	}
-	if (error != 0 && error != EEXIST)
+	if (error != 0 && error != EEXIST) {
-		fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
+		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
+		    snapname, error);
+	}
 	return (B_TRUE);
 }
 
@@ -3241,7 +3245,7 @@ ztest_snapshot_destroy(char *osname, uint64_t id)
 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
 	    (u_longlong_t)id);
 
-	error = dmu_objset_destroy(snapname, B_FALSE);
+	error = dsl_destroy_snapshot(snapname, B_FALSE);
 	if (error != 0 && error != ENOENT)
 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
 	return (B_TRUE);
@@ -3269,7 +3273,7 @@ ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
 
 	/*
 	 * If this dataset exists from a previous run, process its replay log
-	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
+	 * half of the time.  If we don't replay it, then dsl_destroy_head()
 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
 	 */
 	if (ztest_random(2) == 0 &&
@@ -3291,7 +3295,8 @@ ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
 	/*
 	 * Verify that the destroyed dataset is no longer in the namespace.
 	 */
-	VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
+	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
+	    FTAG, &os));
 
 	/*
 	 * Verify that we can create a new dataset.
@@ -3305,8 +3310,7 @@ ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
 	}
 
-	VERIFY3U(0, ==,
+	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
-	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
 
 	ztest_zd_init(zdtmp, NULL, os);
 
@@ -3396,21 +3400,21 @@ ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu",
 	    clone1name, (u_longlong_t)id);
 
-	error = dmu_objset_destroy(clone2name, B_FALSE);
+	error = dsl_destroy_head(clone2name);
 	if (error && error != ENOENT)
-		fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
+		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
-	error = dmu_objset_destroy(snap3name, B_FALSE);
+	error = dsl_destroy_snapshot(snap3name, B_FALSE);
 	if (error && error != ENOENT)
-		fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
+		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
-	error = dmu_objset_destroy(snap2name, B_FALSE);
+	error = dsl_destroy_snapshot(snap2name, B_FALSE);
 	if (error && error != ENOENT)
-		fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
+		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
-	error = dmu_objset_destroy(clone1name, B_FALSE);
+	error = dsl_destroy_head(clone1name);
 	if (error && error != ENOENT)
-		fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
+		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
-	error = dmu_objset_destroy(snap1name, B_FALSE);
+	error = dsl_destroy_snapshot(snap1name, B_FALSE);
 	if (error && error != ENOENT)
-		fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
+		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
 
 	umem_free(snap1name, MAXNAMELEN);
 	umem_free(clone1name, MAXNAMELEN);
@@ -3425,8 +3429,7 @@ ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
 void
 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
 {
-	objset_t *clone;
+	objset_t *os;
-	dsl_dataset_t *ds;
 	char *snap1name;
 	char *clone1name;
 	char *snap2name;
@@ -3465,12 +3468,7 @@ ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
 	}
 
-	error = dmu_objset_hold(snap1name, FTAG, &clone);
+	error = dmu_objset_clone(clone1name, snap1name);
-	if (error)
-		fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
-
-	error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
-	dmu_objset_rele(clone, FTAG);
 	if (error) {
 		if (error == ENOSPC) {
 			ztest_record_enospc(FTAG);
@@ -3497,12 +3495,7 @@ ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
 	}
 
-	error = dmu_objset_hold(snap3name, FTAG, &clone);
+	error = dmu_objset_clone(clone2name, snap3name);
-	if (error)
-		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
-
-	error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
-	dmu_objset_rele(clone, FTAG);
 	if (error) {
 		if (error == ENOSPC) {
 			ztest_record_enospc(FTAG);
@@ -3511,14 +3504,14 @@ ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
 	}
 
-	error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
+	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
 	if (error)
-		fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
+		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
 	error = dsl_dataset_promote(clone2name, NULL);
 	if (error != EBUSY)
 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
 		    error);
-	dsl_dataset_disown(ds, FTAG);
+	dmu_objset_disown(os, FTAG);
 
 out:
 	ztest_dsl_dataset_cleanup(osname, id);
@@ -4392,7 +4385,7 @@ ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
 	}
 
 	count = -1ULL;
-	VERIFY(zap_count(os, object, &count) == 0);
+	VERIFY0(zap_count(os, object, &count));
 	ASSERT(count != -1ULL);
 
 	/*
@@ -4710,6 +4703,22 @@ ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
 	(void) rw_exit(&ztest_name_lock);
 }
 
+static int
+user_release_one(const char *snapname, const char *holdname)
+{
+	nvlist_t *snaps, *holds;
+	int error;
+
+	snaps = fnvlist_alloc();
+	holds = fnvlist_alloc();
+	fnvlist_add_boolean(holds, holdname);
+	fnvlist_add_nvlist(snaps, snapname, holds);
+	fnvlist_free(holds);
+	error = dsl_dataset_user_release(snaps, NULL);
+	fnvlist_free(snaps);
+	return (error);
+}
+
 /*
  * Test snapshot hold/release and deferred destroy.
  */
@@ -4724,22 +4733,30 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
 	char clonename[100];
 	char tag[100];
 	char osname[MAXNAMELEN];
+	nvlist_t *holds;
 
 	(void) rw_enter(&ztest_name_lock, RW_READER);
 
 	dmu_objset_name(os, osname);
 
-	(void) snprintf(snapname, 100, "sh1_%llu", (u_longlong_t)id);
+	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", (long long unsigned int)id);
-	(void) snprintf(fullname, 100, "%s@%s", osname, snapname);
+	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
-	(void) snprintf(clonename, 100, "%s/ch1_%llu",osname,(u_longlong_t)id);
+	(void) snprintf(clonename, sizeof (clonename),
-	(void) snprintf(tag, 100, "tag_%llu", (u_longlong_t)id);
+	    "%s/ch1_%llu", osname, (long long unsigned int)id);
+	(void) snprintf(tag, sizeof (tag), "tag_%llu", (long long unsigned int)id);
 
 	/*
 	 * Clean up from any previous run.
 	 */
-	(void) dmu_objset_destroy(clonename, B_FALSE);
+	error = dsl_destroy_head(clonename);
-	(void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
+	if (error != ENOENT)
-	(void) dmu_objset_destroy(fullname, B_FALSE);
+		ASSERT0(error);
+	error = user_release_one(fullname, tag);
+	if (error != ESRCH && error != ENOENT)
+		ASSERT0(error);
+	error = dsl_destroy_snapshot(fullname, B_FALSE);
+	if (error != ENOENT)
+		ASSERT0(error);
 
 	/*
 	 * Create snapshot, clone it, mark snap for deferred destroy,
@@ -4754,12 +4771,7 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
 	}
 
-	error = dmu_objset_hold(fullname, FTAG, &origin);
+	error = dmu_objset_clone(clonename, fullname);
-	if (error)
-		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
-
-	error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
-	dmu_objset_rele(origin, FTAG);
 	if (error) {
 		if (error == ENOSPC) {
 			ztest_record_enospc("dmu_objset_clone");
@@ -4768,15 +4780,15 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
 	}
 
-	error = dmu_objset_destroy(fullname, B_TRUE);
+	error = dsl_destroy_snapshot(fullname, B_TRUE);
 	if (error) {
-		fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
+		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
 		    fullname, error);
 	}
 
-	error = dmu_objset_destroy(clonename, B_FALSE);
+	error = dsl_destroy_head(clonename);
 	if (error)
-		fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
+		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
 
 	error = dmu_objset_hold(fullname, FTAG, &origin);
 	if (error != ENOENT)
@@ -4796,28 +4808,31 @@ ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
 	}
 
-	error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE,
+	holds = fnvlist_alloc();
-	    B_TRUE, -1);
+	fnvlist_add_string(holds, fullname, tag);
+	error = dsl_dataset_user_hold(holds, 0, NULL);
+	fnvlist_free(holds);
+
 	if (error)
 		fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
 
-	error = dmu_objset_destroy(fullname, B_FALSE);
+	error = dsl_destroy_snapshot(fullname, B_FALSE);
 	if (error != EBUSY) {
-		fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
+		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
 		    fullname, error);
 	}
 
-	error = dmu_objset_destroy(fullname, B_TRUE);
+	error = dsl_destroy_snapshot(fullname, B_TRUE);
 	if (error) {
-		fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
+		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
 		    fullname, error);
 	}
 
-	error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
+	error = user_release_one(fullname, tag);
 	if (error)
-		fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
+		fatal(0, "user_release_one(%s)", fullname, tag);
 
-	VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
+	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
 
 out:
 	(void) rw_exit(&ztest_name_lock);
@@ -4947,7 +4962,7 @@ ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
 			 * prevent a race between offlining a slog and
 			 * destroying a dataset. Offlining the slog will
 			 * grab a reference on the dataset which may cause
-			 * dmu_objset_destroy() to fail with EBUSY thus
+			 * dsl_destroy_head() to fail with EBUSY thus
 			 * leaving the dataset in an inconsistent state.
 			 */
 			if (islog)
@@ -5084,8 +5099,12 @@ ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
 	 */
 	for (i = 0; i < copies; i++) {
 		uint64_t offset = i * blocksize;
-		VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db,
+		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
-		    DMU_READ_NO_PREFETCH) == 0);
+		    DMU_READ_NO_PREFETCH);
+		if (error != 0) {
+			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
+			    os, (long long)object, (long long) offset, error);
+		}
 		ASSERT(db->db_offset == offset);
 		ASSERT(db->db_size == blocksize);
 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
@@ -5300,6 +5319,7 @@ ztest_spa_import_export(char *oldname, char *newname)
 	nvlist_t *config, *newconfig;
 	uint64_t pool_guid;
 	spa_t *spa;
+	int error;
 
 	if (ztest_opts.zo_verbose >= 4) {
 		(void) printf("import/export: old = %s, new = %s\n",
@@ -5344,7 +5364,12 @@ ztest_spa_import_export(char *oldname, char *newname)
 	/*
 	 * Import it under the new name.
 	 */
-	VERIFY3U(0, ==, spa_import(newname, config, NULL, 0));
+	error = spa_import(newname, config, NULL, 0);
+	if (error != 0) {
+		dump_nvlist(config, 0);
+		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
+		    oldname, newname, error);
+	}
 
 	ztest_walk_pool_directory("pools after import");
 
@@ -5551,7 +5576,7 @@ ztest_dataset_open(int d)
 	}
 	ASSERT(error == 0 || error == EEXIST);
 
-	VERIFY0(dmu_objset_hold(name, zd, &os));
+	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
 	(void) rw_exit(&ztest_name_lock);
 
 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
@@ -5592,7 +5617,7 @@ ztest_dataset_close(int d)
 	ztest_ds_t *zd = &ztest_ds[d];
 
 	zil_close(zd->zd_zilog);
-	dmu_objset_rele(zd->zd_os, zd);
+	dmu_objset_disown(zd->zd_os, zd);
 
 	ztest_zd_fini(zd);
 }
@@ -5638,13 +5663,14 @@ ztest_run(ztest_shared_t *zs)
 	 * Open our pool.
 	 */
 	kernel_init(FREAD | FWRITE);
-	VERIFY(spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0);
+	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
 	spa->spa_debug = B_TRUE;
 	ztest_spa = spa;
 
-	VERIFY3U(0, ==, dmu_objset_hold(ztest_opts.zo_pool, FTAG, &os));
+	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
+	    DMU_OST_ANY, B_TRUE, FTAG, &os));
 	zs->zs_guid = dmu_objset_fsid_guid(os);
-	dmu_objset_rele(os, FTAG);
+	dmu_objset_disown(os, FTAG);
 
 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
 

include/libzfs.h

@@ -563,7 +563,7 @@ extern int zfs_create(libzfs_handle_t *, const char *, zfs_type_t,
 extern int zfs_create_ancestors(libzfs_handle_t *, const char *);
 extern int zfs_destroy(zfs_handle_t *, boolean_t);
 extern int zfs_destroy_snaps(zfs_handle_t *, char *, boolean_t);
-extern int zfs_destroy_snaps_nvl(zfs_handle_t *, nvlist_t *, boolean_t);
+extern int zfs_destroy_snaps_nvl(libzfs_handle_t *, nvlist_t *, boolean_t);
 extern int zfs_clone(zfs_handle_t *, const char *, nvlist_t *);
 extern int zfs_snapshot(libzfs_handle_t *, const char *, boolean_t, nvlist_t *);
 extern int zfs_snapshot_nvl(libzfs_handle_t *hdl, nvlist_t *snaps,
@@ -606,8 +606,8 @@ extern int zfs_send(zfs_handle_t *, const char *, const char *,
     sendflags_t *, int, snapfilter_cb_t, void *, nvlist_t **);
 
 extern int zfs_promote(zfs_handle_t *);
-extern int zfs_hold(zfs_handle_t *, const char *, const char *, boolean_t,
+extern int zfs_hold(zfs_handle_t *, const char *, const char *,
-    boolean_t, boolean_t, int, uint64_t, uint64_t);
+    boolean_t, boolean_t, int);
 extern int zfs_release(zfs_handle_t *, const char *, const char *, boolean_t);
 extern int zfs_get_holds(zfs_handle_t *, nvlist_t **);
 extern uint64_t zvol_volsize_to_reservation(uint64_t, nvlist_t *);

include/libzfs_core.h

@@ -46,6 +46,10 @@ int lzc_destroy_snaps(nvlist_t *snaps, boolean_t defer, nvlist_t **errlist);
 int lzc_snaprange_space(const char *firstsnap, const char *lastsnap,
     uint64_t *usedp);
 
+int lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist);
+int lzc_release(nvlist_t *holds, nvlist_t **errlist);
+int lzc_get_holds(const char *snapname, nvlist_t **holdsp);
+
 int lzc_send(const char *snapname, const char *fromsnap, int fd);
 int lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
     boolean_t force, int fd);

include/sys/Makefile.am

@@ -12,6 +12,7 @@ COMMON_H = \
 	$(top_srcdir)/include/sys/dmu.h \
 	$(top_srcdir)/include/sys/dmu_impl.h \
 	$(top_srcdir)/include/sys/dmu_objset.h \
+	$(top_srcdir)/include/sys/dmu_send.h \
 	$(top_srcdir)/include/sys/dmu_traverse.h \
 	$(top_srcdir)/include/sys/dmu_tx.h \
 	$(top_srcdir)/include/sys/dmu_zfetch.h \
@@ -19,11 +20,13 @@ COMMON_H = \
 	$(top_srcdir)/include/sys/dsl_dataset.h \
 	$(top_srcdir)/include/sys/dsl_deadlist.h \
 	$(top_srcdir)/include/sys/dsl_deleg.h \
+	$(top_srcdir)/include/sys/dsl_destroy.h \
 	$(top_srcdir)/include/sys/dsl_dir.h \
 	$(top_srcdir)/include/sys/dsl_pool.h \
 	$(top_srcdir)/include/sys/dsl_prop.h \
 	$(top_srcdir)/include/sys/dsl_scan.h \
 	$(top_srcdir)/include/sys/dsl_synctask.h \
+	$(top_srcdir)/include/sys/dsl_userhold.h \
 	$(top_srcdir)/include/sys/efi_partition.h \
 	$(top_srcdir)/include/sys/metaslab.h \
 	$(top_srcdir)/include/sys/metaslab_impl.h \
@@ -65,8 +68,8 @@ COMMON_H = \
 	$(top_srcdir)/include/sys/zfs_sa.h \
 	$(top_srcdir)/include/sys/zfs_stat.h \
 	$(top_srcdir)/include/sys/zfs_vfsops.h \
-	$(top_srcdir)/include/sys/zfs_znode.h \
 	$(top_srcdir)/include/sys/zfs_vnops.h \
+	$(top_srcdir)/include/sys/zfs_znode.h \
 	$(top_srcdir)/include/sys/zil.h \
 	$(top_srcdir)/include/sys/zil_impl.h \
 	$(top_srcdir)/include/sys/zio_checksum.h \

include/sys/arc.h

@@ -100,7 +100,7 @@ arc_buf_t *arc_loan_buf(spa_t *spa, int size);
 void arc_return_buf(arc_buf_t *buf, void *tag);
 void arc_loan_inuse_buf(arc_buf_t *buf, void *tag);
 void arc_buf_add_ref(arc_buf_t *buf, void *tag);
-int arc_buf_remove_ref(arc_buf_t *buf, void *tag);
+boolean_t arc_buf_remove_ref(arc_buf_t *buf, void *tag);
 int arc_buf_size(arc_buf_t *buf);
 void arc_release(arc_buf_t *buf, void *tag);
 int arc_released(arc_buf_t *buf);

include/sys/dbuf.h

@@ -307,20 +307,17 @@ void dbuf_fini(void);
 
 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
 
-#define	DBUF_IS_METADATA(_db)	\
-	(dbuf_is_metadata(_db))
-
 #define	DBUF_GET_BUFC_TYPE(_db)	\
-	(DBUF_IS_METADATA(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
+	(dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
 
 #define	DBUF_IS_CACHEABLE(_db)						\
 	((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
-	(DBUF_IS_METADATA(_db) &&					\
+	(dbuf_is_metadata(_db) &&					\
 	((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
 
 #define	DBUF_IS_L2CACHEABLE(_db)					\
 	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||	\
-	(DBUF_IS_METADATA(_db) &&					\
+	(dbuf_is_metadata(_db) &&					\
 	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
 
 #define	DBUF_IS_L2COMPRESSIBLE(_db)					\

include/sys/dmu.h

@@ -214,6 +214,11 @@ typedef enum dmu_object_type {
 	DMU_OTN_ZAP_METADATA = DMU_OT(DMU_BSWAP_ZAP, B_TRUE),
 } dmu_object_type_t;
 
+typedef enum txg_how {
+	TXG_WAIT = 1,
+	TXG_NOWAIT,
+} txg_how_t;
+
 void byteswap_uint64_array(void *buf, size_t size);
 void byteswap_uint32_array(void *buf, size_t size);
 void byteswap_uint16_array(void *buf, size_t size);
@@ -252,22 +257,19 @@ void dmu_objset_rele(objset_t *os, void *tag);
 void dmu_objset_disown(objset_t *os, void *tag);
 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
 
-int dmu_objset_evict_dbufs(objset_t *os);
+void dmu_objset_evict_dbufs(objset_t *os);
 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
-int dmu_objset_clone(const char *name, struct dsl_dataset *clone_origin,
+int dmu_objset_clone(const char *name, const char *origin);
-    uint64_t flags);
+int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
-int dmu_objset_destroy(const char *name, boolean_t defer);
-int dmu_snapshots_destroy_nvl(struct nvlist *snaps, boolean_t defer,
     struct nvlist *errlist);
-int dmu_objset_snapshot(struct nvlist *snaps, struct nvlist *, struct nvlist *);
 int dmu_objset_snapshot_one(const char *fsname, const char *snapname);
 int dmu_objset_snapshot_tmp(const char *, const char *, int);
-int dmu_objset_rename(const char *name, const char *newname,
-    boolean_t recursive);
 int dmu_objset_find(char *name, int func(const char *, void *), void *arg,
     int flags);
 void dmu_objset_byteswap(void *buf, size_t size);
+int dsl_dataset_rename_snapshot(const char *fsname,
+    const char *oldsnapname, const char *newsnapname, boolean_t recursive);
 
 typedef struct dmu_buf {
 	uint64_t db_object;		/* object that this buffer is part of */
@@ -537,7 +539,7 @@ void dmu_tx_hold_spill(dmu_tx_t *tx, uint64_t object);
 void dmu_tx_hold_sa(dmu_tx_t *tx, struct sa_handle *hdl, boolean_t may_grow);
 void dmu_tx_hold_sa_create(dmu_tx_t *tx, int total_size);
 void dmu_tx_abort(dmu_tx_t *tx);
-int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
+int dmu_tx_assign(dmu_tx_t *tx, enum txg_how txg_how);
 void dmu_tx_wait(dmu_tx_t *tx);
 void dmu_tx_commit(dmu_tx_t *tx);
 
@@ -785,36 +787,8 @@ typedef void (*dmu_traverse_cb_t)(objset_t *os, void *arg, struct blkptr *bp,
 void dmu_traverse_objset(objset_t *os, uint64_t txg_start,
     dmu_traverse_cb_t cb, void *arg);
 
-int dmu_send(objset_t *tosnap, objset_t *fromsnap,
+int dmu_diff(const char *tosnap_name, const char *fromsnap_name,
-    int outfd, struct vnode *vp, offset_t *off);
+    struct vnode *vp, offset_t *offp);
-int dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep);
-
-typedef struct dmu_recv_cookie {
-	/*
-	 * This structure is opaque!
-	 *
-	 * If logical and real are different, we are recving the stream
-	 * into the "real" temporary clone, and then switching it with
-	 * the "logical" target.
-	 */
-	struct dsl_dataset *drc_logical_ds;
-	struct dsl_dataset *drc_real_ds;
-	struct drr_begin *drc_drrb;
-	char *drc_tosnap;
-	char *drc_top_ds;
-	boolean_t drc_newfs;
-	boolean_t drc_force;
-	struct avl_tree *drc_guid_to_ds_map;
-} dmu_recv_cookie_t;
-
-int dmu_recv_begin(char *tofs, char *tosnap, char *topds, struct drr_begin *,
-    boolean_t force, objset_t *origin, dmu_recv_cookie_t *);
-int dmu_recv_stream(dmu_recv_cookie_t *drc, struct vnode *vp, offset_t *voffp,
-    int cleanup_fd, uint64_t *action_handlep);
-int dmu_recv_end(dmu_recv_cookie_t *drc);
-
-int dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp,
-    offset_t *off);
 
 /* CRC64 table */
 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */

include/sys/dmu_objset.h

@@ -44,6 +44,7 @@ extern "C" {
 
 extern krwlock_t os_lock;
 
+struct dsl_pool;
 struct dsl_dataset;
 struct dmu_tx;
 
@@ -115,8 +116,6 @@ struct objset {
 	/* stuff we store for the user */
 	kmutex_t os_user_ptr_lock;
 	void *os_user_ptr;
-
-	/* SA layout/attribute registration */
 	sa_os_t *os_sa;
 };
 
@@ -146,10 +145,10 @@ void dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat);
 void dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
     uint64_t *usedobjsp, uint64_t *availobjsp);
 uint64_t dmu_objset_fsid_guid(objset_t *os);
-int dmu_objset_find_spa(spa_t *spa, const char *name,
+int dmu_objset_find_dp(struct dsl_pool *dp, uint64_t ddobj,
-    int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags);
+    int func(struct dsl_pool *, struct dsl_dataset *, void *),
-int dmu_objset_prefetch(const char *name, void *arg);
+    void *arg, int flags);
-int dmu_objset_evict_dbufs(objset_t *os);
+void dmu_objset_evict_dbufs(objset_t *os);
 timestruc_t dmu_objset_snap_cmtime(objset_t *os);
 
 /* called from dsl */
@@ -165,6 +164,7 @@ void dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx);
 boolean_t dmu_objset_userused_enabled(objset_t *os);
 int dmu_objset_userspace_upgrade(objset_t *os);
 boolean_t dmu_objset_userspace_present(objset_t *os);
+int dmu_fsname(const char *snapname, char *buf);
 
 void dmu_objset_init(void);
 void dmu_objset_fini(void);

include/sys/dmu_send.h

@@ -0,0 +1,66 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+#ifndef _DMU_SEND_H
+#define	_DMU_SEND_H
+
+#include <sys/inttypes.h>
+#include <sys/spa.h>
+
+struct vnode;
+struct dsl_dataset;
+struct drr_begin;
+struct avl_tree;
+
+int dmu_send(const char *tosnap, const char *fromsnap, int outfd,
+    struct vnode *vp, offset_t *off);
+int dmu_send_estimate(struct dsl_dataset *ds, struct dsl_dataset *fromds,
+    uint64_t *sizep);
+int dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
+    int outfd, struct vnode *vp, offset_t *off);
+
+typedef struct dmu_recv_cookie {
+	struct dsl_dataset *drc_ds;
+	struct drr_begin *drc_drrb;
+	const char *drc_tofs;
+	const char *drc_tosnap;
+	boolean_t drc_newfs;
+	boolean_t drc_byteswap;
+	boolean_t drc_force;
+	struct avl_tree *drc_guid_to_ds_map;
+	zio_cksum_t drc_cksum;
+	uint64_t drc_newsnapobj;
+} dmu_recv_cookie_t;
+
+int dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
+    boolean_t force, char *origin, dmu_recv_cookie_t *drc);
+int dmu_recv_stream(dmu_recv_cookie_t *drc, struct vnode *vp, offset_t *voffp,
+    int cleanup_fd, uint64_t *action_handlep);
+int dmu_recv_end(dmu_recv_cookie_t *drc);
+
+#endif /* _DMU_SEND_H */

include/sys/dmu_tx.h

@@ -22,6 +22,9 @@
  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
+/*
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ */
 
 #ifndef	_SYS_DMU_TX_H
 #define	_SYS_DMU_TX_H
@@ -133,10 +136,11 @@ extern dmu_tx_stats_t dmu_tx_stats;
  * These routines are defined in dmu.h, and are called by the user.
  */
 dmu_tx_t *dmu_tx_create(objset_t *dd);
-int dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how);
+int dmu_tx_assign(dmu_tx_t *tx, txg_how_t txg_how);
 void dmu_tx_commit(dmu_tx_t *tx);
 void dmu_tx_abort(dmu_tx_t *tx);
 uint64_t dmu_tx_get_txg(dmu_tx_t *tx);
+struct dsl_pool *dmu_tx_pool(dmu_tx_t *tx);
 void dmu_tx_wait(dmu_tx_t *tx);
 
 void dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *dcb_func,

include/sys/dsl_dataset.h

@@ -35,6 +35,7 @@
 #include <sys/dsl_synctask.h>
 #include <sys/zfs_context.h>
 #include <sys/dsl_deadlist.h>
+#include <sys/refcount.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -48,10 +49,8 @@ struct dsl_pool;
 #define	DS_IS_INCONSISTENT(ds)	\
 	((ds)->ds_phys->ds_flags & DS_FLAG_INCONSISTENT)
 /*
- * NB: nopromote can not yet be set, but we want support for it in this
+ * Note: nopromote can not yet be set, but we want support for it in this
- * on-disk version, so that we don't need to upgrade for it later.  It
+ * on-disk version, so that we don't need to upgrade for it later.
- * will be needed when we implement 'zfs split' (where the split off
- * clone should not be promoted).
  */
 #define	DS_FLAG_NOPROMOTE	(1ULL<<1)
 
@@ -76,6 +75,8 @@ struct dsl_pool;
  */
 #define	DS_FLAG_CI_DATASET	(1ULL<<16)
 
+#define	DS_CREATE_FLAG_NODIRTY	(1ULL<<24)
+
 typedef struct dsl_dataset_phys {
 	uint64_t ds_dir_obj;		/* DMU_OT_DSL_DIR */
 	uint64_t ds_prev_snap_obj;	/* DMU_OT_DSL_DATASET */
@@ -125,9 +126,6 @@ typedef struct dsl_dataset {
 	dsl_deadlist_t ds_deadlist;
 	bplist_t ds_pending_deadlist;
 
-	/* to protect against multiple concurrent incremental recv */
-	kmutex_t ds_recvlock;
-
 	/* protected by lock on pool's dp_dirty_datasets list */
 	txg_node_t ds_dirty_link;
 	list_node_t ds_synced_link;
@@ -139,13 +137,15 @@ typedef struct dsl_dataset {
 	kmutex_t ds_lock;
 	objset_t *ds_objset;
 	uint64_t ds_userrefs;
+	void *ds_owner;
 
 	/*
-	 * ds_owner is protected by the ds_rwlock and the ds_lock
+	 * Long holds prevent the ds from being destroyed; they allow the
+	 * ds to remain held even after dropping the dp_config_rwlock.
+	 * Owning counts as a long hold.  See the comments above
+	 * dsl_pool_hold() for details.
 	 */
-	krwlock_t ds_rwlock;
+	refcount_t ds_longholds;
-	kcondvar_t ds_exclusive_cv;
-	void *ds_owner;
 
 	/* no locking; only for making guesses */
 	uint64_t ds_trysnap_txg;
@@ -163,76 +163,42 @@ typedef struct dsl_dataset {
 	char ds_snapname[MAXNAMELEN];
 } dsl_dataset_t;
 
-struct dsl_ds_destroyarg {
-	dsl_dataset_t *ds;		/* ds to destroy */
-	dsl_dataset_t *rm_origin;	/* also remove our origin? */
-	boolean_t is_origin_rm;		/* set if removing origin snap */
-	boolean_t defer;		/* destroy -d requested? */
-	boolean_t releasing;		/* destroying due to release? */
-	boolean_t need_prep;		/* do we need to retry due to EBUSY? */
-};
-
 /*
  * The max length of a temporary tag prefix is the number of hex digits
  * required to express UINT64_MAX plus one for the hyphen.
  */
 #define	MAX_TAG_PREFIX_LEN	17
 
-struct dsl_ds_holdarg {
-	dsl_sync_task_group_t *dstg;
-	const char *htag;
-	char *snapname;
-	boolean_t recursive;
-	boolean_t gotone;
-	boolean_t temphold;
-	char failed[MAXPATHLEN];
-};
-
 #define	dsl_dataset_is_snapshot(ds) \
 	((ds)->ds_phys->ds_num_children != 0)
 
 #define	DS_UNIQUE_IS_ACCURATE(ds)	\
 	(((ds)->ds_phys->ds_flags & DS_FLAG_UNIQUE_ACCURATE) != 0)
 
-int dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp);
+int dsl_dataset_hold(struct dsl_pool *dp, const char *name, void *tag,
-int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj,
+    dsl_dataset_t **dsp);
-    void *tag, dsl_dataset_t **);
+int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj, void *tag,
-int dsl_dataset_own(const char *name, boolean_t inconsistentok,
+    dsl_dataset_t **);
+void dsl_dataset_rele(dsl_dataset_t *ds, void *tag);
+int dsl_dataset_own(struct dsl_pool *dp, const char *name,
     void *tag, dsl_dataset_t **dsp);
 int dsl_dataset_own_obj(struct dsl_pool *dp, uint64_t dsobj,
-    boolean_t inconsistentok, void *tag, dsl_dataset_t **dsp);
+    void *tag, dsl_dataset_t **dsp);
-void dsl_dataset_name(dsl_dataset_t *ds, char *name);
-void dsl_dataset_rele(dsl_dataset_t *ds, void *tag);
 void dsl_dataset_disown(dsl_dataset_t *ds, void *tag);
-void dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag);
+void dsl_dataset_name(dsl_dataset_t *ds, char *name);
-boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok,
+boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag);
-    void *tag);
-void dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *tag);
 void dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
     minor_t minor);
 uint64_t dsl_dataset_create_sync(dsl_dir_t *pds, const char *lastname,
     dsl_dataset_t *origin, uint64_t flags, cred_t *, dmu_tx_t *);
 uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
     uint64_t flags, dmu_tx_t *tx);
-int dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer);
+int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors);
-dsl_checkfunc_t dsl_dataset_destroy_check;
-dsl_syncfunc_t dsl_dataset_destroy_sync;
-dsl_syncfunc_t dsl_dataset_user_hold_sync;
-int dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *, dmu_tx_t *tx);
-void dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *, dmu_tx_t *tx);
-int dsl_dataset_rename(char *name, const char *newname, boolean_t recursive);
 int dsl_dataset_promote(const char *name, char *conflsnap);
-int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
+int dsl_dataset_rename_snapshot(const char *fsname,
-    boolean_t force);
+    const char *oldsnapname, const char *newsnapname, boolean_t recursive);
-int dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
+int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
-    boolean_t recursive, boolean_t temphold, int cleanup_fd);
+    minor_t cleanup_minor, const char *htag);
-int dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
-    boolean_t temphold);
-int dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
-    boolean_t recursive);
-int dsl_dataset_user_release_tmp(struct dsl_pool *dp, uint64_t dsobj,
-    char *htag, boolean_t retry);
-int dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp);
 
 blkptr_t *dsl_dataset_get_blkptr(dsl_dataset_t *ds);
 void dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx);
@@ -271,13 +237,35 @@ int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf);
 int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
     uint64_t asize, uint64_t inflight, uint64_t *used,
     uint64_t *ref_rsrv);
-int dsl_dataset_set_quota(const char *dsname, zprop_source_t source,
+int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
     uint64_t quota);
-dsl_syncfunc_t dsl_dataset_set_quota_sync;
+int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
-int dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
     uint64_t reservation);
 
-int dsl_destroy_inconsistent(const char *dsname, void *arg);
+boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier);
+void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag);
+void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag);
+boolean_t dsl_dataset_long_held(dsl_dataset_t *ds);
+
+int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
+    dsl_dataset_t *origin_head, boolean_t force);
+void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
+    dsl_dataset_t *origin_head, dmu_tx_t *tx);
+int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
+    dmu_tx_t *tx);
+void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
+    dmu_tx_t *tx);
+
+void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
+    dmu_tx_t *tx);
+void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds);
+int dsl_dataset_get_snapname(dsl_dataset_t *ds);
+int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name,
+    uint64_t *value);
+int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx);
+void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
+    zprop_source_t source, uint64_t value, dmu_tx_t *tx);
+int dsl_dataset_rollback(const char *fsname);
 
 #ifdef ZFS_DEBUG
 #define	dprintf_ds(ds, fmt, ...) do { \

include/sys/dsl_destroy.h

@@ -0,0 +1,52 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+#ifndef	_SYS_DSL_DESTROY_H
+#define	_SYS_DSL_DESTROY_H
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+struct nvlist;
+struct dsl_dataset;
+struct dmu_tx;
+
+int dsl_destroy_snapshots_nvl(struct nvlist *snaps, boolean_t defer,
+    struct nvlist *errlist);
+int dsl_destroy_snapshot(const char *name, boolean_t defer);
+int dsl_destroy_head(const char *name);
+int dsl_destroy_head_check_impl(struct dsl_dataset *ds, int expected_holds);
+void dsl_destroy_head_sync_impl(struct dsl_dataset *ds, struct dmu_tx *tx);
+int dsl_destroy_inconsistent(const char *dsname, void *arg);
+void dsl_destroy_snapshot_sync_impl(struct dsl_dataset *ds,
+    boolean_t defer, struct dmu_tx *tx);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* _SYS_DSL_DESTROY_H */

include/sys/dsl_dir.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 #ifndef	_SYS_DSL_DIR_H
@@ -101,18 +102,15 @@ struct dsl_dir {
 	char dd_myname[MAXNAMELEN];
 };
 
-void dsl_dir_close(dsl_dir_t *dd, void *tag);
+void dsl_dir_rele(dsl_dir_t *dd, void *tag);
-int dsl_dir_open(const char *name, void *tag, dsl_dir_t **, const char **tail);
+int dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
-int dsl_dir_open_spa(spa_t *spa, const char *name, void *tag, dsl_dir_t **,
+    dsl_dir_t **, const char **tail);
-    const char **tailp);
+int dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
-int dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
     const char *tail, void *tag, dsl_dir_t **);
 void dsl_dir_name(dsl_dir_t *dd, char *buf);
 int dsl_dir_namelen(dsl_dir_t *dd);
 uint64_t dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds,
     const char *name, dmu_tx_t *tx);
-dsl_checkfunc_t dsl_dir_destroy_check;
-dsl_syncfunc_t dsl_dir_destroy_sync;
 void dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv);
 uint64_t dsl_dir_space_available(dsl_dir_t *dd,
     dsl_dir_t *ancestor, int64_t delta, int ondiskonly);
@@ -131,14 +129,15 @@ int dsl_dir_set_quota(const char *ddname, zprop_source_t source,
     uint64_t quota);
 int dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
     uint64_t reservation);
-int dsl_dir_rename(dsl_dir_t *dd, const char *newname);
+int dsl_dir_rename(const char *oldname, const char *newname);
 int dsl_dir_transfer_possible(dsl_dir_t *sdd, dsl_dir_t *tdd, uint64_t space);
-int dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx);
 boolean_t dsl_dir_is_clone(dsl_dir_t *dd);
 void dsl_dir_new_refreservation(dsl_dir_t *dd, struct dsl_dataset *ds,
     uint64_t reservation, cred_t *cr, dmu_tx_t *tx);
 void dsl_dir_snap_cmtime_update(dsl_dir_t *dd);
 timestruc_t dsl_dir_snap_cmtime(dsl_dir_t *dd);
+void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value,
+    dmu_tx_t *tx);
 
 /* internal reserved dir name */
 #define	MOS_DIR_NAME "$MOS"

include/sys/dsl_pool.h

@@ -36,6 +36,7 @@
 #include <sys/arc.h>
 #include <sys/bpobj.h>
 #include <sys/bptree.h>
+#include <sys/rrwlock.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -129,7 +130,7 @@ typedef struct dsl_pool {
 	 * syncing context does not need to ever have it for read, since
 	 * nobody else could possibly have it for write.
 	 */
-	krwlock_t dp_config_rwlock;
+	rrwlock_t dp_config_rwlock;
 
 	zfs_all_blkstats_t *dp_blkstats;
 } dsl_pool_t;
@@ -155,15 +156,20 @@ void dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx);
 void dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx);
 void dsl_pool_mos_diduse_space(dsl_pool_t *dp,
     int64_t used, int64_t comp, int64_t uncomp);
+void dsl_pool_config_enter(dsl_pool_t *dp, void *tag);
+void dsl_pool_config_exit(dsl_pool_t *dp, void *tag);
+boolean_t dsl_pool_config_held(dsl_pool_t *dp);
 
 taskq_t *dsl_pool_iput_taskq(dsl_pool_t *dp);
 
-extern int dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj,
+int dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj,
-    const char *tag, uint64_t *now, dmu_tx_t *tx);
+    const char *tag, uint64_t now, dmu_tx_t *tx);
-extern int dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj,
+int dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj,
     const char *tag, dmu_tx_t *tx);
-extern void dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp);
+void dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp);
 int dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **);
+int dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp);
+void dsl_pool_rele(dsl_pool_t *dp, void *tag);
 
 void dsl_pool_tx_assign_add_usecs(dsl_pool_t *dp, uint64_t usecs);
 

include/sys/dsl_prop.h

@@ -54,58 +54,47 @@ typedef struct dsl_props_arg {
 	zprop_source_t pa_source;
 } dsl_props_arg_t;
 
-typedef struct dsl_prop_set_arg {
-	const char *psa_name;
-	zprop_source_t psa_source;
-	int psa_intsz;
-	int psa_numints;
-	const void *psa_value;
-
-	/*
-	 * Used to handle the special requirements of the quota and reservation
-	 * properties.
-	 */
-	uint64_t psa_effective_value;
-} dsl_prop_setarg_t;
-
 int dsl_prop_register(struct dsl_dataset *ds, const char *propname,
     dsl_prop_changed_cb_t *callback, void *cbarg);
 int dsl_prop_unregister(struct dsl_dataset *ds, const char *propname,
     dsl_prop_changed_cb_t *callback, void *cbarg);
-int dsl_prop_numcb(struct dsl_dataset *ds);
+void dsl_prop_notify_all(struct dsl_dir *dd);
+boolean_t dsl_prop_hascb(struct dsl_dataset *ds);
 
 int dsl_prop_get(const char *ddname, const char *propname,
     int intsz, int numints, void *buf, char *setpoint);
 int dsl_prop_get_integer(const char *ddname, const char *propname,
     uint64_t *valuep, char *setpoint);
 int dsl_prop_get_all(objset_t *os, nvlist_t **nvp);
-int dsl_prop_get_received(objset_t *os, nvlist_t **nvp);
+int dsl_prop_get_received(const char *dsname, nvlist_t **nvp);
 int dsl_prop_get_ds(struct dsl_dataset *ds, const char *propname,
     int intsz, int numints, void *buf, char *setpoint);
+int dsl_prop_get_int_ds(struct dsl_dataset *ds, const char *propname,
+    uint64_t *valuep);
 int dsl_prop_get_dd(struct dsl_dir *dd, const char *propname,
     int intsz, int numints, void *buf, char *setpoint,
     boolean_t snapshot);
 
-dsl_syncfunc_t dsl_props_set_sync;
+void dsl_props_set_sync_impl(struct dsl_dataset *ds, zprop_source_t source,
-int dsl_prop_set(const char *ddname, const char *propname,
+    nvlist_t *props, dmu_tx_t *tx);
-    zprop_source_t source, int intsz, int numints, const void *buf);
+void dsl_prop_set_sync_impl(struct dsl_dataset *ds, const char *propname,
+    zprop_source_t source, int intsz, int numints, const void *value,
+    dmu_tx_t *tx);
 int dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *nvl);
+int dsl_prop_set_int(const char *dsname, const char *propname,
+    zprop_source_t source, uint64_t value);
+int dsl_prop_set_string(const char *dsname, const char *propname,
+    zprop_source_t source, const char *value);
+int dsl_prop_inherit(const char *dsname, const char *propname,
+    zprop_source_t source);
 
-void dsl_prop_setarg_init_uint64(dsl_prop_setarg_t *psa, const char *propname,
+int dsl_prop_predict(dsl_dir_t *dd, const char *propname,
-    zprop_source_t source, uint64_t *value);
+    zprop_source_t source, uint64_t value, uint64_t *newvalp);
-int dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa);
-#ifdef	ZFS_DEBUG
-void dsl_prop_check_prediction(dsl_dir_t *dd, dsl_prop_setarg_t *psa);
-#define	DSL_PROP_CHECK_PREDICTION(dd, psa)	\
-	dsl_prop_check_prediction((dd), (psa))
-#else
-#define	DSL_PROP_CHECK_PREDICTION(dd, psa)	/* nothing */
-#endif
 
 /* flag first receive on or after SPA_VERSION_RECVD_PROPS */
-boolean_t dsl_prop_get_hasrecvd(objset_t *os);
+boolean_t dsl_prop_get_hasrecvd(const char *dsname);
-void dsl_prop_set_hasrecvd(objset_t *os);
+int dsl_prop_set_hasrecvd(const char *dsname);
-void dsl_prop_unset_hasrecvd(objset_t *os);
+void dsl_prop_unset_hasrecvd(const char *dsname);
 
 void dsl_prop_nvlist_add_uint64(nvlist_t *nv, zfs_prop_t prop, uint64_t value);
 void dsl_prop_nvlist_add_string(nvlist_t *nv,

include/sys/dsl_synctask.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 #ifndef	_SYS_DSL_SYNCTASK_H
@@ -34,43 +35,26 @@ extern "C" {
 
 struct dsl_pool;
 
-typedef int (dsl_checkfunc_t)(void *, void *, dmu_tx_t *);
+typedef int (dsl_checkfunc_t)(void *, dmu_tx_t *);
-typedef void (dsl_syncfunc_t)(void *, void *, dmu_tx_t *);
+typedef void (dsl_syncfunc_t)(void *, dmu_tx_t *);
 
 typedef struct dsl_sync_task {
-	list_node_t dst_node;
+	txg_node_t dst_node;
+	struct dsl_pool *dst_pool;
+	uint64_t dst_txg;
+	int dst_space;
 	dsl_checkfunc_t *dst_checkfunc;
 	dsl_syncfunc_t *dst_syncfunc;
-	void *dst_arg1;
+	void *dst_arg;
-	void *dst_arg2;
+	int dst_error;
-	int dst_err;
+	boolean_t dst_nowaiter;
 } dsl_sync_task_t;
 
-typedef struct dsl_sync_task_group {
+void dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx);
-	txg_node_t dstg_node;
+int dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
-	list_t dstg_tasks;
+    dsl_syncfunc_t *syncfunc, void *arg, int blocks_modified);
-	struct dsl_pool *dstg_pool;
+void dsl_sync_task_nowait(struct dsl_pool *dp, dsl_syncfunc_t *syncfunc,
-	uint64_t dstg_txg;
+    void *arg, int blocks_modified, dmu_tx_t *tx);
-	int dstg_err;
-	int dstg_space;
-	boolean_t dstg_nowaiter;
-} dsl_sync_task_group_t;
-
-dsl_sync_task_group_t *dsl_sync_task_group_create(struct dsl_pool *dp);
-void dsl_sync_task_create(dsl_sync_task_group_t *dstg,
-    dsl_checkfunc_t *, dsl_syncfunc_t *,
-    void *arg1, void *arg2, int blocks_modified);
-int dsl_sync_task_group_wait(dsl_sync_task_group_t *dstg);
-void dsl_sync_task_group_nowait(dsl_sync_task_group_t *dstg, dmu_tx_t *tx);
-void dsl_sync_task_group_destroy(dsl_sync_task_group_t *dstg);
-void dsl_sync_task_group_sync(dsl_sync_task_group_t *dstg, dmu_tx_t *tx);
-
-int dsl_sync_task_do(struct dsl_pool *dp,
-    dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc,
-    void *arg1, void *arg2, int blocks_modified);
-void dsl_sync_task_do_nowait(struct dsl_pool *dp,
-    dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc,
-    void *arg1, void *arg2, int blocks_modified, dmu_tx_t *tx);
 
 #ifdef	__cplusplus
 }

include/sys/dsl_userhold.h

@@ -0,0 +1,57 @@
+
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ * Copyright (c) 2012, Joyent, Inc. All rights reserved.
+ */
+
+#ifndef	_SYS_DSL_USERHOLD_H
+#define	_SYS_DSL_USERHOLD_H
+
+#include <sys/nvpair.h>
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+struct dsl_pool;
+struct dsl_dataset;
+struct dmu_tx;
+
+int dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor,
+    nvlist_t *errlist);
+int dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist);
+int dsl_dataset_get_holds(const char *dsname, nvlist_t *nvl);
+void dsl_dataset_user_release_tmp(struct dsl_pool *dp, uint64_t dsobj,
+    const char *htag);
+int dsl_dataset_user_hold_check_one(struct dsl_dataset *ds, const char *htag,
+    boolean_t temphold, struct dmu_tx *tx);
+void dsl_dataset_user_hold_sync_one(struct dsl_dataset *ds, const char *htag,
+    minor_t minor, uint64_t now, struct dmu_tx *tx);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* _SYS_DSL_USERHOLD_H */

include/sys/metaslab.h

@@ -20,7 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011 by Delphix. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 #ifndef _SYS_METASLAB_H
@@ -57,6 +57,7 @@ extern int metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
 extern void metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg,
     boolean_t now);
 extern int metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg);
+extern void metaslab_check_free(spa_t *spa, const blkptr_t *bp);
 extern void metaslab_fastwrite_mark(spa_t *spa, const blkptr_t *bp);
 extern void metaslab_fastwrite_unmark(spa_t *spa, const blkptr_t *bp);
 

include/sys/nvpair.h

@@ -285,6 +285,7 @@ void fnvlist_pack_free(char *, size_t);
 nvlist_t *fnvlist_unpack(char *, size_t);
 nvlist_t *fnvlist_dup(nvlist_t *);
 void fnvlist_merge(nvlist_t *, nvlist_t *);
+size_t fnvlist_num_pairs(nvlist_t *);
 
 void fnvlist_add_boolean(nvlist_t *, const char *);
 void fnvlist_add_boolean_value(nvlist_t *, const char *, boolean_t);

include/sys/refcount.h

@@ -50,15 +50,17 @@ typedef struct reference {
 
 typedef struct refcount {
 	kmutex_t rc_mtx;
+	boolean_t rc_tracked;
 	list_t rc_list;
 	list_t rc_removed;
 	int64_t rc_count;
 	int64_t rc_removed_count;
 } refcount_t;
 
-/* Note: refcount_t must be initialized with refcount_create() */
+/* Note: refcount_t must be initialized with refcount_create[_untracked]() */
 
 void refcount_create(refcount_t *rc);
+void refcount_create_untracked(refcount_t *rc);
 void refcount_destroy(refcount_t *rc);
 void refcount_destroy_many(refcount_t *rc, uint64_t number);
 int refcount_is_zero(refcount_t *rc);
@@ -79,6 +81,7 @@ typedef struct refcount {
 } refcount_t;
 
 #define	refcount_create(rc) ((rc)->rc_count = 0)
+#define	refcount_create_untracked(rc) ((rc)->rc_count = 0)
 #define	refcount_destroy(rc) ((rc)->rc_count = 0)
 #define	refcount_destroy_many(rc, number) ((rc)->rc_count = 0)
 #define	refcount_is_zero(rc) ((rc)->rc_count == 0)

include/sys/rrwlock.h

@@ -60,6 +60,7 @@ typedef struct rrwlock {
 	refcount_t	rr_anon_rcount;
 	refcount_t	rr_linked_rcount;
 	boolean_t	rr_writer_wanted;
+	boolean_t	rr_track_all;
 } rrwlock_t;
 
 /*
@@ -67,15 +68,19 @@ typedef struct rrwlock {
  * 'tag' must be the same in a rrw_enter() as in its
  * corresponding rrw_exit().
  */
-void rrw_init(rrwlock_t *rrl);
+void rrw_init(rrwlock_t *rrl, boolean_t track_all);
 void rrw_destroy(rrwlock_t *rrl);
 void rrw_enter(rrwlock_t *rrl, krw_t rw, void *tag);
+void rrw_enter_read(rrwlock_t *rrl, void *tag);
+void rrw_enter_write(rrwlock_t *rrl);
 void rrw_exit(rrwlock_t *rrl, void *tag);
 boolean_t rrw_held(rrwlock_t *rrl, krw_t rw);
 void rrw_tsd_destroy(void *arg);
 
 #define	RRW_READ_HELD(x)	rrw_held(x, RW_READER)
 #define	RRW_WRITE_HELD(x)	rrw_held(x, RW_WRITER)
+#define	RRW_LOCK_HELD(x) \
+	(rrw_held(x, RW_WRITER) || rrw_held(x, RW_READER))
 
 #ifdef	__cplusplus
 }

include/sys/spa.h

@@ -422,7 +422,7 @@ extern int spa_get_stats(const char *pool, nvlist_t **config, char *altroot,
 extern int spa_create(const char *pool, nvlist_t *config, nvlist_t *props,
     nvlist_t *zplprops);
 extern int spa_import_rootpool(char *devpath, char *devid);
-extern int spa_import(const char *pool, nvlist_t *config, nvlist_t *props,
+extern int spa_import(char *pool, nvlist_t *config, nvlist_t *props,
     uint64_t flags);
 extern nvlist_t *spa_tryimport(nvlist_t *tryconfig);
 extern int spa_destroy(char *pool);

include/sys/space_map.h

@@ -149,6 +149,8 @@ extern void space_map_add(space_map_t *sm, uint64_t start, uint64_t size);
 extern void space_map_remove(space_map_t *sm, uint64_t start, uint64_t size);
 extern boolean_t space_map_contains(space_map_t *sm,
     uint64_t start, uint64_t size);
+extern space_seg_t *space_map_find(space_map_t *sm, uint64_t start,
+    uint64_t size, avl_index_t *wherep);
 extern void space_map_swap(space_map_t **msrc, space_map_t **mdest);
 extern void space_map_vacate(space_map_t *sm,
     space_map_func_t *func, space_map_t *mdest);

include/sys/txg.h

@@ -45,9 +45,6 @@ extern "C" {
 /* Number of txgs worth of frees we defer adding to in-core spacemaps */
 #define	TXG_DEFER_SIZE		2
 
-#define	TXG_WAIT		1ULL
-#define	TXG_NOWAIT		2ULL
-
 typedef struct tx_cpu tx_cpu_t;
 
 typedef struct txg_handle {
@@ -125,11 +122,11 @@ extern void txg_wait_callbacks(struct dsl_pool *dp);
 extern void txg_list_create(txg_list_t *tl, size_t offset);
 extern void txg_list_destroy(txg_list_t *tl);
 extern boolean_t txg_list_empty(txg_list_t *tl, uint64_t txg);
-extern int txg_list_add(txg_list_t *tl, void *p, uint64_t txg);
+extern boolean_t txg_list_add(txg_list_t *tl, void *p, uint64_t txg);
-extern int txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg);
+extern boolean_t txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg);
 extern void *txg_list_remove(txg_list_t *tl, uint64_t txg);
 extern void *txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg);
-extern int txg_list_member(txg_list_t *tl, void *p, uint64_t txg);
+extern boolean_t txg_list_member(txg_list_t *tl, void *p, uint64_t txg);
 extern void *txg_list_head(txg_list_t *tl, uint64_t txg);
 extern void *txg_list_next(txg_list_t *tl, void *p, uint64_t txg);
 

include/sys/zfeature.h

@@ -26,7 +26,6 @@
 #ifndef _SYS_ZFEATURE_H
 #define	_SYS_ZFEATURE_H
 
-#include <sys/dmu.h>
 #include <sys/nvpair.h>
 #include "zfeature_common.h"
 
@@ -34,14 +33,18 @@
 extern "C" {
 #endif
 
-extern boolean_t feature_is_supported(objset_t *os, uint64_t obj,
+struct spa;
+struct dmu_tx;
+struct objset;
+
+extern boolean_t feature_is_supported(struct objset *os, uint64_t obj,
     uint64_t desc_obj, nvlist_t *unsup_feat, nvlist_t *enabled_feat);
 
-struct spa;
+extern void spa_feature_create_zap_objects(struct spa *, struct dmu_tx *);
-extern void spa_feature_create_zap_objects(struct spa *, dmu_tx_t *);
+extern void spa_feature_enable(struct spa *, zfeature_info_t *,
-extern void spa_feature_enable(struct spa *, zfeature_info_t *, dmu_tx_t *);
+    struct dmu_tx *);
-extern void spa_feature_incr(struct spa *, zfeature_info_t *, dmu_tx_t *);
+extern void spa_feature_incr(struct spa *, zfeature_info_t *, struct dmu_tx *);
-extern void spa_feature_decr(struct spa *, zfeature_info_t *, dmu_tx_t *);
+extern void spa_feature_decr(struct spa *, zfeature_info_t *, struct dmu_tx *);
 extern boolean_t spa_feature_is_enabled(struct spa *, zfeature_info_t *);
 extern boolean_t spa_feature_is_active(struct spa *, zfeature_info_t *);
 

include/sys/zfs_context.h

@@ -209,8 +209,6 @@ typedef struct kthread {
 	void *		t_arg;
 } kthread_t;
 
-#define	tsd_get(key)			pthread_getspecific(key)
-#define	tsd_set(key, val)		pthread_setspecific(key, val)
 #define	curthread			zk_thread_current()
 #define	thread_exit			zk_thread_exit
 #define	thread_create(stk, stksize, func, arg, len, pp, state, pri)	\
@@ -284,6 +282,12 @@ typedef int krw_t;
 #define	RW_WRITE_HELD(x)	((x)->rw_wr_owner == curthread)
 #define	RW_LOCK_HELD(x)		(RW_READ_HELD(x) || RW_WRITE_HELD(x))
 
+#undef RW_LOCK_HELD
+#define	RW_LOCK_HELD(x)		(RW_READ_HELD(x) || RW_WRITE_HELD(x))
+
+#undef RW_LOCK_HELD
+#define	RW_LOCK_HELD(x)		(RW_READ_HELD(x) || RW_WRITE_HELD(x))
+
 extern void rw_init(krwlock_t *rwlp, char *name, int type, void *arg);
 extern void rw_destroy(krwlock_t *rwlp);
 extern void rw_enter(krwlock_t *rwlp, krw_t rw);
@@ -320,6 +324,22 @@ extern void cv_broadcast(kcondvar_t *cv);
 #define cv_wait_interruptible(cv, mp)		cv_wait(cv, mp)
 #define cv_wait_io(cv, mp)			cv_wait(cv, mp)
 
+/*
+ * Thread-specific data
+ */
+#define	tsd_get(k) pthread_getspecific(k)
+#define	tsd_set(k, v) pthread_setspecific(k, v)
+#define	tsd_create(kp, d) pthread_key_create(kp, d)
+#define	tsd_destroy(kp) /* nothing */
+
+/*
+ * Thread-specific data
+ */
+#define	tsd_get(k) pthread_getspecific(k)
+#define	tsd_set(k, v) pthread_setspecific(k, v)
+#define	tsd_create(kp, d) pthread_key_create(kp, d)
+#define	tsd_destroy(kp) /* nothing */
+
 /*
  * kstat creation, installation and deletion
  */
@@ -592,7 +612,7 @@ typedef struct callb_cpr {
 
 extern char *kmem_vasprintf(const char *fmt, va_list adx);
 extern char *kmem_asprintf(const char *fmt, ...);
-#define	strfree(str) kmem_free((str), strlen(str)+1)
+#define	strfree(str) kmem_free((str), strlen(str) + 1)
 
 /*
  * Hostname information

include/sys/zfs_debug.h

@@ -43,11 +43,13 @@
 extern int zfs_flags;
 extern int zfs_recover;
 
-#define	ZFS_DEBUG_DPRINTF	0x0001
+#define	ZFS_DEBUG_DPRINTF	(1<<0)
-#define	ZFS_DEBUG_DBUF_VERIFY	0x0002
+#define	ZFS_DEBUG_DBUF_VERIFY	(1<<1)
-#define	ZFS_DEBUG_DNODE_VERIFY	0x0004
+#define	ZFS_DEBUG_DNODE_VERIFY	(1<<2)
-#define	ZFS_DEBUG_SNAPNAMES	0x0008
+#define	ZFS_DEBUG_SNAPNAMES	(1<<3)
-#define	ZFS_DEBUG_MODIFY	0x0010
+#define	ZFS_DEBUG_MODIFY	(1<<4)
+#define	ZFS_DEBUG_SPA		(1<<5)
+#define	ZFS_DEBUG_ZIO_FREE	(1<<6)
 
 /*
  * Always log zfs debug messages to the spl debug subsystem as SS_USER1.

include/sys/zfs_ioctl.h

@@ -302,7 +302,6 @@ typedef struct zfs_cmd {
 	uint64_t	zc_history;		/* really (char *) */
 	char		zc_value[MAXPATHLEN * 2];
 	char		zc_string[MAXNAMELEN];
-	char		zc_top_ds[MAXPATHLEN];
 	uint64_t	zc_guid;
 	uint64_t	zc_nvlist_conf;		/* really (char *) */
 	uint64_t	zc_nvlist_conf_size;
@@ -352,7 +351,8 @@ extern int zfs_secpolicy_snapshot_perms(const char *name, cred_t *cr);
 extern int zfs_secpolicy_rename_perms(const char *from,
     const char *to, cred_t *cr);
 extern int zfs_secpolicy_destroy_perms(const char *name, cred_t *cr);
-extern int zfs_unmount_snap(const char *, void *);
+extern void zfs_unmount_snap(const char *);
+extern void zfs_destroy_unmount_origin(const char *);
 
 enum zfsdev_state_type {
 	ZST_ONEXIT,

include/sys/zfs_znode.h

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 #ifndef	_SYS_FS_ZFS_ZNODE_H
@@ -254,7 +255,7 @@ typedef struct znode {
  */
 #define	ZFS_ENTER(zsb) \
 	{ \
-		rrw_enter(&(zsb)->z_teardown_lock, RW_READER, FTAG); \
+		rrw_enter_read(&(zsb)->z_teardown_lock, FTAG); \
 		if ((zsb)->z_unmounted) { \
 			ZFS_EXIT(zsb); \
 			return (EIO); \

include/sys/zil.h

@@ -470,8 +470,8 @@ extern int	zil_check_log_chain(const char *osname, void *txarg);
 extern void	zil_sync(zilog_t *zilog, dmu_tx_t *tx);
 extern void	zil_clean(zilog_t *zilog, uint64_t synced_txg);
 
-extern int	zil_suspend(zilog_t *zilog);
+extern int	zil_suspend(const char *osname, void **cookiep);
-extern void	zil_resume(zilog_t *zilog);
+extern void	zil_resume(void *cookie);
 
 extern void	zil_add_block(zilog_t *zilog, const blkptr_t *bp);
 extern int	zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp);

include/sys/zvol.h

@@ -39,7 +39,7 @@ extern int zvol_get_stats(objset_t *os, nvlist_t *nv);
 extern boolean_t zvol_is_zvol(const char *);
 extern void zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx);
 extern int zvol_create_minor(const char *);
-extern int zvol_create_minors(const char *);
+extern int zvol_create_minors(char *);
 extern int zvol_remove_minor(const char *);
 extern void zvol_remove_minors(const char *);
 extern int zvol_set_volsize(const char *, uint64_t);

lib/libzfs/libzfs_config.c

@@ -106,7 +106,7 @@ namespace_reload(libzfs_handle_t *hdl)
 	nvlist_t *config;
 	config_node_t *cn;
 	nvpair_t *elem;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	void *cookie;
 
 	if (hdl->libzfs_ns_gen == 0) {
@@ -261,7 +261,7 @@ zpool_get_features(zpool_handle_t *zhp)
 int
 zpool_refresh_stats(zpool_handle_t *zhp, boolean_t *missing)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int error;
 	nvlist_t *config;
 	libzfs_handle_t *hdl = zhp->zpool_hdl;

lib/libzfs/libzfs_dataset.c

@@ -313,7 +313,7 @@ get_recvd_props_ioctl(zfs_handle_t *zhp)
 {
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 	nvlist_t *recvdprops;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int err;
 
 	if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
@@ -376,7 +376,7 @@ static int
 get_stats(zfs_handle_t *zhp)
 {
 	int rc = 0;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	if (zcmd_alloc_dst_nvlist(zhp->zfs_hdl, &zc, 0) != 0)
 		return (-1);
@@ -439,7 +439,7 @@ make_dataset_handle_common(zfs_handle_t *zhp, zfs_cmd_t *zc)
 zfs_handle_t *
 make_dataset_handle(libzfs_handle_t *hdl, const char *path)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	zfs_handle_t *zhp = calloc(sizeof (zfs_handle_t), 1);
 
@@ -1427,7 +1427,7 @@ zfs_is_namespace_prop(zfs_prop_t prop)
 int
 zfs_prop_set(zfs_handle_t *zhp, const char *propname, const char *propval)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int ret = -1;
 	prop_changelist_t *cl = NULL;
 	char errbuf[1024];
@@ -1553,7 +1553,7 @@ error:
 int
 zfs_prop_inherit(zfs_handle_t *zhp, const char *propname, boolean_t received)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int ret;
 	prop_changelist_t *cl;
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
@@ -1728,7 +1728,7 @@ static int
 get_numeric_property(zfs_handle_t *zhp, zfs_prop_t prop, zprop_source_t *src,
     char **source, uint64_t *val)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	nvlist_t *zplprops = NULL;
 	struct mnttab mnt;
 	char *mntopt_on = NULL;
@@ -2002,10 +2002,7 @@ get_clones_cb(zfs_handle_t *zhp, void *arg)
 	    NULL, NULL, 0, B_TRUE) != 0)
 		goto out;
 	if (strcmp(gca->buf, gca->origin) == 0) {
-		if (nvlist_add_boolean(gca->value, zfs_get_name(zhp)) != 0) {
+		fnvlist_add_boolean(gca->value, zfs_get_name(zhp));
-			zfs_close(zhp);
-			return (no_memory(zhp->zfs_hdl));
-		}
 		gca->numclones--;
 	}
 
@@ -2580,7 +2577,7 @@ zfs_prop_get_userquota_common(zfs_handle_t *zhp, const char *propname,
     uint64_t *propvalue, zfs_userquota_prop_t *typep)
 {
 	int err;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 
@@ -2640,7 +2637,7 @@ zfs_prop_get_written_int(zfs_handle_t *zhp, const char *propname,
     uint64_t *propvalue)
 {
 	int err;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	const char *snapname;
 
 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
@@ -2760,7 +2757,7 @@ static int
 check_parents(libzfs_handle_t *hdl, const char *path, uint64_t *zoned,
     boolean_t accept_ancestor, int *prefixlen)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char parent[ZFS_MAXNAMELEN];
 	char *slash;
 	zfs_handle_t *zhp;
@@ -3120,7 +3117,7 @@ zfs_create(libzfs_handle_t *hdl, const char *path, zfs_type_t type,
 int
 zfs_destroy(zfs_handle_t *zhp, boolean_t defer)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 
@@ -3200,46 +3197,50 @@ zfs_destroy_snaps(zfs_handle_t *zhp, char *snapname, boolean_t defer)
 		    dgettext(TEXT_DOMAIN, "cannot destroy '%s@%s'"),
 		    zhp->zfs_name, snapname);
 	} else {
-		ret = zfs_destroy_snaps_nvl(zhp, dd.nvl, defer);
+		ret = zfs_destroy_snaps_nvl(zhp->zfs_hdl, dd.nvl, defer);
 	}
 	nvlist_free(dd.nvl);
 	return (ret);
 }
 
 /*
- * Destroys all the snapshots named in the nvlist.  They must be underneath
+ * Destroys all the snapshots named in the nvlist.
- * the zhp (either snapshots of it, or snapshots of its descendants).
  */
 int
-zfs_destroy_snaps_nvl(zfs_handle_t *zhp, nvlist_t *snaps, boolean_t defer)
+zfs_destroy_snaps_nvl(libzfs_handle_t *hdl, nvlist_t *snaps, boolean_t defer)
 {
 	int ret;
 	nvlist_t *errlist;
+	nvpair_t *pair;
 
 	ret = lzc_destroy_snaps(snaps, defer, &errlist);
 
-	if (ret != 0) {
+	if (ret == 0)
-		nvpair_t *pair;
+		return (0);
-		for (pair = nvlist_next_nvpair(errlist, NULL);
-		    pair != NULL; pair = nvlist_next_nvpair(errlist, pair)) {
-			char errbuf[1024];
-			(void) snprintf(errbuf, sizeof (errbuf),
-			    dgettext(TEXT_DOMAIN, "cannot destroy snapshot %s"),
-			    nvpair_name(pair));
 
-			switch (fnvpair_value_int32(pair)) {
+	if (nvlist_next_nvpair(errlist, NULL) == NULL) {
-			case EEXIST:
+		char errbuf[1024];
-				zfs_error_aux(zhp->zfs_hdl,
+		(void) snprintf(errbuf, sizeof (errbuf),
-				    dgettext(TEXT_DOMAIN,
+		    dgettext(TEXT_DOMAIN, "cannot destroy snapshots"));
-				    "snapshot is cloned"));
+
-				ret = zfs_error(zhp->zfs_hdl, EZFS_EXISTS,
+		ret = zfs_standard_error(hdl, ret, errbuf);
-				    errbuf);
+	}
-				break;
+	for (pair = nvlist_next_nvpair(errlist, NULL);
-			default:
+	    pair != NULL; pair = nvlist_next_nvpair(errlist, pair)) {
-				ret = zfs_standard_error(zhp->zfs_hdl, errno,
+		char errbuf[1024];
-				    errbuf);
+		(void) snprintf(errbuf, sizeof (errbuf),
-				break;
+		    dgettext(TEXT_DOMAIN, "cannot destroy snapshot %s"),
-			}
+		    nvpair_name(pair));
+
+		switch (fnvpair_value_int32(pair)) {
+		case EEXIST:
+			zfs_error_aux(hdl,
+			    dgettext(TEXT_DOMAIN, "snapshot is cloned"));
+			ret = zfs_error(hdl, EZFS_EXISTS, errbuf);
+			break;
+		default:
+			ret = zfs_standard_error(hdl, errno, errbuf);
+			break;
 		}
 	}
 
@@ -3388,7 +3389,7 @@ int
 zfs_promote(zfs_handle_t *zhp)
 {
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char parent[MAXPATHLEN];
 	char *cp;
 	int ret;
@@ -3726,7 +3727,7 @@ zfs_rollback(zfs_handle_t *zhp, zfs_handle_t *snap, boolean_t force)
 {
 	rollback_data_t cb = { 0 };
 	int err;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	boolean_t restore_resv = 0;
 	uint64_t old_volsize = 0, new_volsize;
 	zfs_prop_t resv_prop = { 0 };
@@ -3813,7 +3814,7 @@ zfs_rename(zfs_handle_t *zhp, const char *target, boolean_t recursive,
     boolean_t force_unmount)
 {
 	int ret;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char *delim;
 	prop_changelist_t *cl = NULL;
 	zfs_handle_t *zhrp = NULL;
@@ -4032,7 +4033,7 @@ zvol_create_link(libzfs_handle_t *hdl, const char *dataset)
 static int
 zvol_create_link_common(libzfs_handle_t *hdl, const char *dataset, int ifexists)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char path[MAXPATHLEN];
 	int error;
 
@@ -4096,7 +4097,7 @@ zvol_create_link_common(libzfs_handle_t *hdl, const char *dataset, int ifexists)
 int
 zvol_remove_link(libzfs_handle_t *hdl, const char *dataset)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int timeout = 3000; /* in milliseconds */
 	int error = 0;
 	int i;
@@ -4289,7 +4290,7 @@ static int
 zfs_smb_acl_mgmt(libzfs_handle_t *hdl, char *dataset, char *path,
     zfs_smb_acl_op_t cmd, char *resource1, char *resource2)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	nvlist_t *nvlist = NULL;
 	int error;
 
@@ -4371,7 +4372,7 @@ int
 zfs_userspace(zfs_handle_t *zhp, zfs_userquota_prop_t type,
     zfs_userspace_cb_t func, void *arg)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	zfs_useracct_t buf[100];
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 	int ret;
@@ -4408,37 +4409,83 @@ zfs_userspace(zfs_handle_t *zhp, zfs_userquota_prop_t type,
 	return (0);
 }
 
+struct holdarg {
+	nvlist_t *nvl;
+	const char *snapname;
+	const char *tag;
+	boolean_t recursive;
+};
+
+static int
+zfs_hold_one(zfs_handle_t *zhp, void *arg)
+{
+	struct holdarg *ha = arg;
+	zfs_handle_t *szhp;
+	char name[ZFS_MAXNAMELEN];
+	int rv = 0;
+
+	(void) snprintf(name, sizeof (name),
+	    "%s@%s", zhp->zfs_name, ha->snapname);
+
+	szhp = make_dataset_handle(zhp->zfs_hdl, name);
+	if (szhp) {
+		fnvlist_add_string(ha->nvl, name, ha->tag);
+		zfs_close(szhp);
+	}
+
+	if (ha->recursive)
+		rv = zfs_iter_filesystems(zhp, zfs_hold_one, ha);
+	zfs_close(zhp);
+	return (rv);
+}
+
 int
 zfs_hold(zfs_handle_t *zhp, const char *snapname, const char *tag,
-    boolean_t recursive, boolean_t temphold, boolean_t enoent_ok,
+    boolean_t recursive, boolean_t enoent_ok, int cleanup_fd)
-    int cleanup_fd, uint64_t dsobj, uint64_t createtxg)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	int ret;
+	struct holdarg ha;
+	nvlist_t *errors;
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
+	char errbuf[1024];
+	nvpair_t *elem;
 
-	ASSERT(!recursive || dsobj == 0);
+	ha.nvl = fnvlist_alloc();
+	ha.snapname = snapname;
+	ha.tag = tag;
+	ha.recursive = recursive;
+	(void) zfs_hold_one(zfs_handle_dup(zhp), &ha);
+	ret = lzc_hold(ha.nvl, cleanup_fd, &errors);
+	fnvlist_free(ha.nvl);
 
-	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
+	if (ret == 0)
-	(void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
+		return (0);
-	if (strlcpy(zc.zc_string, tag, sizeof (zc.zc_string))
-	    >= sizeof (zc.zc_string))
-		return (zfs_error(hdl, EZFS_TAGTOOLONG, tag));
-	zc.zc_cookie = recursive;
-	zc.zc_temphold = temphold;
-	zc.zc_cleanup_fd = cleanup_fd;
-	zc.zc_sendobj = dsobj;
-	zc.zc_createtxg = createtxg;
 
-	if (zfs_ioctl(hdl, ZFS_IOC_HOLD, &zc) != 0) {
+	if (nvlist_next_nvpair(errors, NULL) == NULL) {
-		char errbuf[ZFS_MAXNAMELEN+32];
+		/* no hold-specific errors */
+		(void) snprintf(errbuf, sizeof (errbuf),
+		    dgettext(TEXT_DOMAIN, "cannot hold"));
+		switch (ret) {
+		case ENOTSUP:
+			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
+			    "pool must be upgraded"));
+			(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
+			break;
+		case EINVAL:
+			(void) zfs_error(hdl, EZFS_BADTYPE, errbuf);
+			break;
+		default:
+			(void) zfs_standard_error(hdl, ret, errbuf);
+		}
+	}
 
-		/*
+	for (elem = nvlist_next_nvpair(errors, NULL);
-		 * if it was recursive, the one that actually failed will be in
+	    elem != NULL;
-		 * zc.zc_name.
+	    elem = nvlist_next_nvpair(errors, elem)) {
-		 */
+		(void) snprintf(errbuf, sizeof (errbuf),
-		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
+		    dgettext(TEXT_DOMAIN,
-		    "cannot hold '%s@%s'"), zc.zc_name, snapname);
+		    "cannot hold snapshot '%s'"), nvpair_name(elem));
-		switch (errno) {
+		switch (fnvpair_value_int32(elem)) {
 		case E2BIG:
 			/*
 			 * Temporary tags wind up having the ds object id
@@ -4446,77 +4493,133 @@ zfs_hold(zfs_handle_t *zhp, const char *snapname, const char *tag,
 			 * above, it's still possible for the tag to wind
 			 * up being slightly too long.
 			 */
-			return (zfs_error(hdl, EZFS_TAGTOOLONG, errbuf));
+			(void) zfs_error(hdl, EZFS_TAGTOOLONG, errbuf);
-		case ENOTSUP:
+			break;
-			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
-			    "pool must be upgraded"));
-			return (zfs_error(hdl, EZFS_BADVERSION, errbuf));
 		case EINVAL:
-			return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
+			(void) zfs_error(hdl, EZFS_BADTYPE, errbuf);
+			break;
 		case EEXIST:
-			return (zfs_error(hdl, EZFS_REFTAG_HOLD, errbuf));
+			(void) zfs_error(hdl, EZFS_REFTAG_HOLD, errbuf);
+			break;
 		case ENOENT:
 			if (enoent_ok)
 				return (ENOENT);
 			/* FALLTHROUGH */
 		default:
-			return (zfs_standard_error_fmt(hdl, errno, errbuf));
+			(void) zfs_standard_error(hdl,
+			    fnvpair_value_int32(elem), errbuf);
 		}
 	}
 
-	return (0);
+	fnvlist_free(errors);
+	return (ret);
+}
+
+struct releasearg {
+	nvlist_t *nvl;
+	const char *snapname;
+	const char *tag;
+	boolean_t recursive;
+};
+
+static int
+zfs_release_one(zfs_handle_t *zhp, void *arg)
+{
+	struct holdarg *ha = arg;
+	zfs_handle_t *szhp;
+	char name[ZFS_MAXNAMELEN];
+	int rv = 0;
+
+	(void) snprintf(name, sizeof (name),
+	    "%s@%s", zhp->zfs_name, ha->snapname);
+
+	szhp = make_dataset_handle(zhp->zfs_hdl, name);
+	if (szhp) {
+		nvlist_t *holds = fnvlist_alloc();
+		fnvlist_add_boolean(holds, ha->tag);
+		fnvlist_add_nvlist(ha->nvl, name, holds);
+		zfs_close(szhp);
+	}
+
+	if (ha->recursive)
+		rv = zfs_iter_filesystems(zhp, zfs_release_one, ha);
+	zfs_close(zhp);
+	return (rv);
 }
 
 int
 zfs_release(zfs_handle_t *zhp, const char *snapname, const char *tag,
     boolean_t recursive)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	int ret;
+	struct holdarg ha;
+	nvlist_t *errors;
+	nvpair_t *elem;
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 
-	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
+	ha.nvl = fnvlist_alloc();
-	(void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
+	ha.snapname = snapname;
-	if (strlcpy(zc.zc_string, tag, sizeof (zc.zc_string))
+	ha.tag = tag;
-	    >= sizeof (zc.zc_string))
+	ha.recursive = recursive;
-		return (zfs_error(hdl, EZFS_TAGTOOLONG, tag));
+	(void) zfs_release_one(zfs_handle_dup(zhp), &ha);
-	zc.zc_cookie = recursive;
+	ret = lzc_release(ha.nvl, &errors);
+	fnvlist_free(ha.nvl);
 
-	if (zfs_ioctl(hdl, ZFS_IOC_RELEASE, &zc) != 0) {
+	if (ret == 0)
-		char errbuf[ZFS_MAXNAMELEN+32];
+		return (0);
+
+	if (nvlist_next_nvpair(errors, NULL) == NULL) {
+		/* no hold-specific errors */
+		char errbuf[1024];
 
-		/*
-		 * if it was recursive, the one that actually failed will be in
-		 * zc.zc_name.
-		 */
 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
-		    "cannot release '%s' from '%s@%s'"), tag, zc.zc_name,
+		    "cannot release"));
-		    snapname);
 		switch (errno) {
-		case ESRCH:
-			return (zfs_error(hdl, EZFS_REFTAG_RELE, errbuf));
 		case ENOTSUP:
 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 			    "pool must be upgraded"));
-			return (zfs_error(hdl, EZFS_BADVERSION, errbuf));
+			(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
-		case EINVAL:
+			break;
-			return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
 		default:
-			return (zfs_standard_error_fmt(hdl, errno, errbuf));
+			(void) zfs_standard_error_fmt(hdl, errno, errbuf);
 		}
 	}
 
-	return (0);
+	for (elem = nvlist_next_nvpair(errors, NULL);
+	    elem != NULL;
+	    elem = nvlist_next_nvpair(errors, elem)) {
+		char errbuf[1024];
+
+		(void) snprintf(errbuf, sizeof (errbuf),
+		    dgettext(TEXT_DOMAIN,
+		    "cannot release hold from snapshot '%s'"),
+		    nvpair_name(elem));
+		switch (fnvpair_value_int32(elem)) {
+		case ESRCH:
+			(void) zfs_error(hdl, EZFS_REFTAG_RELE, errbuf);
+			break;
+		case EINVAL:
+			(void) zfs_error(hdl, EZFS_BADTYPE, errbuf);
+			break;
+		default:
+			(void) zfs_standard_error_fmt(hdl,
+			    fnvpair_value_int32(elem), errbuf);
+		}
+	}
+
+	fnvlist_free(errors);
+	return (ret);
 }
 
 int
 zfs_get_fsacl(zfs_handle_t *zhp, nvlist_t **nvl)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 	int nvsz = 2048;
 	void *nvbuf;
 	int err = 0;
-	char errbuf[ZFS_MAXNAMELEN+32];
+	char errbuf[1024];
 
 	assert(zhp->zfs_type == ZFS_TYPE_VOLUME ||
 	    zhp->zfs_type == ZFS_TYPE_FILESYSTEM);
@@ -4578,10 +4681,10 @@ out:
 int
 zfs_set_fsacl(zfs_handle_t *zhp, boolean_t un, nvlist_t *nvl)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 	char *nvbuf;
-	char errbuf[ZFS_MAXNAMELEN+32];
+	char errbuf[1024];
 	size_t nvsz;
 	int err;
 
@@ -4632,38 +4735,18 @@ zfs_set_fsacl(zfs_handle_t *zhp, boolean_t un, nvlist_t *nvl)
 int
 zfs_get_holds(zfs_handle_t *zhp, nvlist_t **nvl)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	int err;
-	libzfs_handle_t *hdl = zhp->zfs_hdl;
+	char errbuf[1024];
-	int nvsz = 2048;
-	void *nvbuf;
-	int err = 0;
-	char errbuf[ZFS_MAXNAMELEN+32];
 
-	assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
+	err = lzc_get_holds(zhp->zfs_name, nvl);
 
-tryagain:
+	if (err != 0) {
+		libzfs_handle_t *hdl = zhp->zfs_hdl;
 
-	nvbuf = malloc(nvsz);
-	if (nvbuf == NULL) {
-		err = (zfs_error(hdl, EZFS_NOMEM, strerror(errno)));
-		goto out;
-	}
-
-	zc.zc_nvlist_dst_size = nvsz;
-	zc.zc_nvlist_dst = (uintptr_t)nvbuf;
-
-	(void) strlcpy(zc.zc_name, zhp->zfs_name, ZFS_MAXNAMELEN);
-
-	if (zfs_ioctl(hdl, ZFS_IOC_GET_HOLDS, &zc) != 0) {
 		(void) snprintf(errbuf, sizeof (errbuf),
 		    dgettext(TEXT_DOMAIN, "cannot get holds for '%s'"),
-		    zc.zc_name);
+		    zhp->zfs_name);
-		switch (errno) {
+		switch (err) {
-		case ENOMEM:
-			free(nvbuf);
-			nvsz = zc.zc_nvlist_dst_size;
-			goto tryagain;
-
 		case ENOTSUP:
 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 			    "pool must be upgraded"));
@@ -4679,19 +4762,8 @@ tryagain:
 			err = zfs_standard_error_fmt(hdl, errno, errbuf);
 			break;
 		}
-	} else {
-		/* success */
-		int rc = nvlist_unpack(nvbuf, zc.zc_nvlist_dst_size, nvl, 0);
-		if (rc) {
-			(void) snprintf(errbuf, sizeof (errbuf),
-			    dgettext(TEXT_DOMAIN, "cannot get holds for '%s'"),
-			    zc.zc_name);
-			err = zfs_standard_error_fmt(hdl, rc, errbuf);
-		}
 	}
 
-	free(nvbuf);
-out:
 	return (err);
 }
 

lib/libzfs/libzfs_diff.c

@@ -90,7 +90,7 @@ static int
 get_stats_for_obj(differ_info_t *di, const char *dsname, uint64_t obj,
     char *pn, int maxlen, zfs_stat_t *sb)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int error;
 
 	(void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
@@ -379,7 +379,7 @@ describe_free(FILE *fp, differ_info_t *di, uint64_t object, char *namebuf,
 static int
 write_free_diffs(FILE *fp, differ_info_t *di, dmu_diff_record_t *dr)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *lhdl = di->zhp->zfs_hdl;
 	char fobjname[MAXPATHLEN];
 
@@ -507,7 +507,7 @@ static int
 make_temp_snapshot(differ_info_t *di)
 {
 	libzfs_handle_t *hdl = di->zhp->zfs_hdl;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) snprintf(zc.zc_value, sizeof (zc.zc_value),
 	    ZDIFF_PREFIX, getpid());
@@ -749,7 +749,7 @@ int
 zfs_show_diffs(zfs_handle_t *zhp, int outfd, const char *fromsnap,
     const char *tosnap, int flags)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char errbuf[1024];
 	differ_info_t di = { 0 };
 	pthread_t tid;

lib/libzfs/libzfs_fru.c

@@ -361,7 +361,7 @@ libzfs_fru_devpath(libzfs_handle_t *hdl, const char *fru)
 int
 zpool_fru_set(zpool_handle_t *zhp, uint64_t vdev_guid, const char *fru)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 	(void) strncpy(zc.zc_value, fru, sizeof (zc.zc_value));

lib/libzfs/libzfs_graph.c

@@ -379,7 +379,7 @@ zfs_graph_add(libzfs_handle_t *hdl, zfs_graph_t *zgp, const char *source,
 static int
 iterate_children(libzfs_handle_t *hdl, zfs_graph_t *zgp, const char *dataset)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	zfs_vertex_t *zvp;
 
 	/*
@@ -473,7 +473,7 @@ iterate_children(libzfs_handle_t *hdl, zfs_graph_t *zgp, const char *dataset)
 static boolean_t
 external_dependents(libzfs_handle_t *hdl, zfs_graph_t *zgp, const char *dataset)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	/*
 	 * Check whether this dataset is a clone or has clones since

lib/libzfs/libzfs_import.c

@@ -365,7 +365,7 @@ static nvlist_t *
 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
 {
 	nvlist_t *nvl;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int err;
 
 	if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)

lib/libzfs/libzfs_iter.c

@@ -103,7 +103,7 @@ top:
 int
 zfs_iter_filesystems(zfs_handle_t *zhp, zfs_iter_f func, void *data)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	zfs_handle_t *nzhp;
 	int ret;
 
@@ -140,7 +140,7 @@ int
 zfs_iter_snapshots(zfs_handle_t *zhp, boolean_t simple, zfs_iter_f func,
     void *data)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	zfs_handle_t *nzhp;
 	int ret;
 

lib/libzfs/libzfs_pool.c

@@ -64,7 +64,7 @@ typedef struct prop_flags {
 static int
 zpool_get_all_props(zpool_handle_t *zhp)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
@@ -692,7 +692,7 @@ error:
 int
 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int ret = -1;
 	char errbuf[1024];
 	nvlist_t *nvl = NULL;
@@ -1141,7 +1141,7 @@ int
 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
     nvlist_t *props, nvlist_t *fsprops)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	nvlist_t *zc_fsprops = NULL;
 	nvlist_t *zc_props = NULL;
 	char msg[1024];
@@ -1275,7 +1275,7 @@ create_failed:
 int
 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	zfs_handle_t *zfp = NULL;
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 	char msg[1024];
@@ -1319,7 +1319,7 @@ zpool_destroy(zpool_handle_t *zhp, const char *log_str)
 int
 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int ret;
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 	char msg[1024];
@@ -1446,7 +1446,7 @@ static int
 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
     const char *log_str)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 
 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
@@ -1721,7 +1721,7 @@ int
 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
     nvlist_t *props, int flags)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	zpool_rewind_policy_t policy;
 	nvlist_t *nv = NULL;
 	nvlist_t *nvinfo = NULL;
@@ -1913,7 +1913,7 @@ zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
 int
 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
@@ -2389,7 +2389,7 @@ int
 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
     vdev_state_t *newstate)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	nvlist_t *tgt;
 	boolean_t avail_spare, l2cache, islog;
@@ -2473,7 +2473,7 @@ zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
 int
 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	nvlist_t *tgt;
 	boolean_t avail_spare, l2cache;
@@ -2523,7 +2523,7 @@ zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
 int
 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
@@ -2558,7 +2558,7 @@ zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
 int
 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
@@ -2612,7 +2612,7 @@ int
 zpool_vdev_attach(zpool_handle_t *zhp,
     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	int ret;
 	nvlist_t *tgt;
@@ -2788,7 +2788,7 @@ zpool_vdev_attach(zpool_handle_t *zhp,
 int
 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	nvlist_t *tgt;
 	boolean_t avail_spare, l2cache;
@@ -2886,7 +2886,7 @@ int
 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
     nvlist_t *props, splitflags_t flags)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
 	nvlist_t **varray = NULL, *zc_props = NULL;
@@ -3097,7 +3097,7 @@ out:
 int
 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	nvlist_t *tgt;
 	boolean_t avail_spare, l2cache, islog;
@@ -3142,7 +3142,7 @@ zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
 int
 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	nvlist_t *tgt;
 	zpool_rewind_policy_t policy;
@@ -3218,7 +3218,7 @@ zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
 int
 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
@@ -3244,7 +3244,7 @@ zpool_reguid(zpool_handle_t *zhp)
 {
 	char msg[1024];
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) snprintf(msg, sizeof (msg),
 	    dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
@@ -3262,7 +3262,7 @@ zpool_reguid(zpool_handle_t *zhp)
 int
 zpool_reopen(zpool_handle_t *zhp)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
@@ -3342,7 +3342,7 @@ path_to_devid(const char *path)
 static void
 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 	(void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
@@ -3517,7 +3517,7 @@ zbookmark_compare(const void *a, const void *b)
 int
 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	uint64_t count;
 	zbookmark_t *zb = NULL;
 	int i;
@@ -3613,7 +3613,7 @@ nomem:
 int
 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
 	(void) strcpy(zc.zc_name, zhp->zpool_name);
@@ -3641,7 +3641,7 @@ zfs_save_arguments(int argc, char **argv, char *string, int len)
 int
 zpool_log_history(libzfs_handle_t *hdl, const char *message)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	nvlist_t *args;
 	int err;
 
@@ -3667,7 +3667,7 @@ zpool_log_history(libzfs_handle_t *hdl, const char *message)
 static int
 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *hdl = zhp->zpool_hdl;
 
 	(void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
@@ -3804,7 +3804,7 @@ int
 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
     int *dropped, int block, int cleanup_fd)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int error = 0;
 
 	*nvp = NULL;
@@ -3863,7 +3863,7 @@ out:
 int
 zpool_events_clear(libzfs_handle_t *hdl, int *count)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char msg[1024];
 
 	(void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
@@ -3882,7 +3882,7 @@ void
 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
     char *pathname, size_t len)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	boolean_t mounted = B_FALSE;
 	char *mntpnt = NULL;
 	char dsname[MAXNAMELEN];

lib/libzfs/libzfs_sendrecv.c

@@ -812,7 +812,7 @@ static int
 estimate_ioctl(zfs_handle_t *zhp, uint64_t fromsnap_obj,
     boolean_t fromorigin, uint64_t *sizep)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 
 	assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
@@ -876,7 +876,7 @@ static int
 dump_ioctl(zfs_handle_t *zhp, const char *fromsnap, uint64_t fromsnap_obj,
     boolean_t fromorigin, int outfd, nvlist_t *debugnv)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 	nvlist_t *thisdbg;
 
@@ -978,9 +978,7 @@ hold_for_send(zfs_handle_t *zhp, send_dump_data_t *sdd)
 	 */
 	if (pzhp) {
 		error = zfs_hold(pzhp, thissnap, sdd->holdtag,
-		    B_FALSE, B_TRUE, B_TRUE, sdd->cleanup_fd,
+		    B_FALSE, B_TRUE, sdd->cleanup_fd);
-		    zfs_prop_get_int(zhp, ZFS_PROP_OBJSETID),
-		    zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG));
 		zfs_close(pzhp);
 	}
 
@@ -992,7 +990,7 @@ send_progress_thread(void *arg)
 {
 	progress_arg_t *pa = arg;
 
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	zfs_handle_t *zhp = pa->pa_zhp;
 	libzfs_handle_t *hdl = zhp->zfs_hdl;
 	unsigned long long bytes;
@@ -1195,7 +1193,7 @@ dump_filesystem(zfs_handle_t *zhp, void *arg)
 	int rv = 0;
 	send_dump_data_t *sdd = arg;
 	boolean_t missingfrom = B_FALSE;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) snprintf(zc.zc_name, sizeof (zc.zc_name), "%s@%s",
 	    zhp->zfs_name, sdd->tosnap);
@@ -1683,7 +1681,7 @@ recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
     int baselen, char *newname, recvflags_t *flags)
 {
 	static int seq;
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int err;
 	prop_changelist_t *clp;
 	zfs_handle_t *zhp;
@@ -1719,12 +1717,11 @@ recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
 		err = ENOENT;
 	}
 
-	if (err != 0 && strncmp(name+baselen, "recv-", 5) != 0) {
+	if (err != 0 && strncmp(name + baselen, "recv-", 5) != 0) {
 		seq++;
 
-		(void) strncpy(newname, name, baselen);
+		(void) snprintf(newname, ZFS_MAXNAMELEN, "%.*srecv-%u-%u",
-		(void) snprintf(newname+baselen, ZFS_MAXNAMELEN-baselen,
+		    baselen, name, getpid(), seq);
-		    "recv-%ld-%u", (long) getpid(), seq);
 		(void) strlcpy(zc.zc_value, newname, sizeof (zc.zc_value));
 
 		if (flags->verbose) {
@@ -1756,7 +1753,7 @@ static int
 recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
     char *newname, recvflags_t *flags)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int err = 0;
 	prop_changelist_t *clp;
 	zfs_handle_t *zhp;
@@ -2015,7 +2012,7 @@ again:
 			    stream_originguid, originguid)) {
 			case 1: {
 				/* promote it! */
-				zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+				zfs_cmd_t zc = {"\0"};
 				nvlist_t *origin_nvfs;
 				char *origin_fsname;
 
@@ -2087,7 +2084,7 @@ again:
 			if (0 == nvlist_lookup_nvlist(stream_nvfs, "snapprops",
 			    &props) && 0 == nvlist_lookup_nvlist(props,
 			    stream_snapname, &props)) {
-				zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+				zfs_cmd_t zc = {"\0"};
 
 				zc.zc_cookie = B_TRUE; /* received */
 				(void) snprintf(zc.zc_name, sizeof (zc.zc_name),
@@ -2518,7 +2515,7 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
     nvlist_t *stream_nv, avl_tree_t *stream_avl, char **top_zfs, int cleanup_fd,
     uint64_t *action_handlep)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	time_t begin_time;
 	int ioctl_err, ioctl_errno, err;
 	char *cp;
@@ -2649,7 +2646,6 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
 	/*
 	 * Determine name of destination snapshot, store in zc_value.
 	 */
-	(void) strcpy(zc.zc_top_ds, tosnap);
 	(void) strcpy(zc.zc_value, tosnap);
 	(void) strlcat(zc.zc_value, chopprefix, sizeof (zc.zc_value));
 	free(cp);
@@ -2892,7 +2888,7 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
 	zcmd_free_nvlists(&zc);
 
 	if (err == 0 && snapprops_nvlist) {
-		zfs_cmd_t zc2 = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+		zfs_cmd_t zc2 = {"\0"};
 
 		(void) strcpy(zc2.zc_name, zc.zc_value);
 		zc2.zc_cookie = B_TRUE; /* received */

lib/libzfs_core/libzfs_core.c

@@ -118,7 +118,7 @@ static int
 lzc_ioctl(zfs_ioc_t ioc, const char *name,
     nvlist_t *source, nvlist_t **resultp)
 {
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	int error = 0;
 	char *packed;
 	size_t size;
@@ -132,6 +132,7 @@ lzc_ioctl(zfs_ioc_t ioc, const char *name,
 	zc.zc_nvlist_src_size = size;
 
 	if (resultp != NULL) {
+		*resultp = NULL;
 		zc.zc_nvlist_dst_size = MAX(size * 2, 128 * 1024);
 		zc.zc_nvlist_dst = (uint64_t)(uintptr_t)
 		    malloc(zc.zc_nvlist_dst_size);
@@ -159,8 +160,6 @@ lzc_ioctl(zfs_ioc_t ioc, const char *name,
 	if (zc.zc_nvlist_dst_filled) {
 		*resultp = fnvlist_unpack((void *)(uintptr_t)zc.zc_nvlist_dst,
 		    zc.zc_nvlist_dst_size);
-	} else if (resultp != NULL) {
-		*resultp = NULL;
 	}
 
 out:
@@ -209,7 +208,7 @@ lzc_clone(const char *fsname, const char *origin,
  * The value will be the (int32) error code.
  *
  * The return value will be 0 if all snapshots were created, otherwise it will
- * be the errno of a (undetermined) snapshot that failed.
+ * be the errno of a (unspecified) snapshot that failed.
  */
 int
 lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
@@ -258,7 +257,7 @@ lzc_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t **errlist)
  * The return value will be 0 if all snapshots were destroyed (or marked for
  * later destruction if 'defer' is set) or didn't exist to begin with.
  *
- * Otherwise the return value will be the errno of a (undetermined) snapshot
+ * Otherwise the return value will be the errno of a (unspecified) snapshot
  * that failed, no snapshots will be destroyed, and the errlist will have an
  * entry for each snapshot that failed.  The value in the errlist will be
  * the (int32) error code.
@@ -326,12 +325,107 @@ lzc_exists(const char *dataset)
 	 * The objset_stats ioctl is still legacy, so we need to construct our
 	 * own zfs_cmd_t rather than using zfsc_ioctl().
 	 */
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 
 	(void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
 	return (ioctl(g_fd, ZFS_IOC_OBJSET_STATS, &zc) == 0);
 }
 
+/*
+ * Create "user holds" on snapshots.  If there is a hold on a snapshot,
+ * the snapshot can not be destroyed.  (However, it can be marked for deletion
+ * by lzc_destroy_snaps(defer=B_TRUE).)
+ *
+ * The keys in the nvlist are snapshot names.
+ * The snapshots must all be in the same pool.
+ * The value is the name of the hold (string type).
+ *
+ * If cleanup_fd is not -1, it must be the result of open("/dev/zfs", O_EXCL).
+ * In this case, when the cleanup_fd is closed (including on process
+ * termination), the holds will be released.  If the system is shut down
+ * uncleanly, the holds will be released when the pool is next opened
+ * or imported.
+ *
+ * The return value will be 0 if all holds were created. Otherwise the return
+ * value will be the errno of a (unspecified) hold that failed, no holds will
+ * be created, and the errlist will have an entry for each hold that
+ * failed (name = snapshot).  The value in the errlist will be the error
+ * code (int32).
+ */
+int
+lzc_hold(nvlist_t *holds, int cleanup_fd, nvlist_t **errlist)
+{
+	char pool[MAXNAMELEN];
+	nvlist_t *args;
+	nvpair_t *elem;
+	int error;
+
+	/* determine the pool name */
+	elem = nvlist_next_nvpair(holds, NULL);
+	if (elem == NULL)
+		return (0);
+	(void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
+	pool[strcspn(pool, "/@")] = '\0';
+
+	args = fnvlist_alloc();
+	fnvlist_add_nvlist(args, "holds", holds);
+	if (cleanup_fd != -1)
+		fnvlist_add_int32(args, "cleanup_fd", cleanup_fd);
+
+	error = lzc_ioctl(ZFS_IOC_HOLD, pool, args, errlist);
+	nvlist_free(args);
+	return (error);
+}
+
+/*
+ * Release "user holds" on snapshots.  If the snapshot has been marked for
+ * deferred destroy (by lzc_destroy_snaps(defer=B_TRUE)), it does not have
+ * any clones, and all the user holds are removed, then the snapshot will be
+ * destroyed.
+ *
+ * The keys in the nvlist are snapshot names.
+ * The snapshots must all be in the same pool.
+ * The value is a nvlist whose keys are the holds to remove.
+ *
+ * The return value will be 0 if all holds were removed.
+ * Otherwise the return value will be the errno of a (unspecified) release
+ * that failed, no holds will be released, and the errlist will have an
+ * entry for each snapshot that has failed releases (name = snapshot).
+ * The value in the errlist will be the error code (int32) of a failed release.
+ */
+int
+lzc_release(nvlist_t *holds, nvlist_t **errlist)
+{
+	char pool[MAXNAMELEN];
+	nvpair_t *elem;
+
+	/* determine the pool name */
+	elem = nvlist_next_nvpair(holds, NULL);
+	if (elem == NULL)
+		return (0);
+	(void) strlcpy(pool, nvpair_name(elem), sizeof (pool));
+	pool[strcspn(pool, "/@")] = '\0';
+
+	return (lzc_ioctl(ZFS_IOC_RELEASE, pool, holds, errlist));
+}
+
+/*
+ * Retrieve list of user holds on the specified snapshot.
+ *
+ * On success, *holdsp will be set to a nvlist which the caller must free.
+ * The keys are the names of the holds, and the value is the creation time
+ * of the hold (uint64) in seconds since the epoch.
+ */
+int
+lzc_get_holds(const char *snapname, nvlist_t **holdsp)
+{
+	int error;
+	nvlist_t *innvl = fnvlist_alloc();
+	error = lzc_ioctl(ZFS_IOC_GET_HOLDS, snapname, innvl, holdsp);
+	fnvlist_free(innvl);
+	return (error);
+}
+
 /*
  * If fromsnap is NULL, a full (non-incremental) stream will be sent.
  */
@@ -411,7 +505,7 @@ lzc_receive(const char *snapname, nvlist_t *props, const char *origin,
 	 * The receive ioctl is still legacy, so we need to construct our own
 	 * zfs_cmd_t rather than using zfsc_ioctl().
 	 */
-	zfs_cmd_t zc = {"\0", 0, 0, 0, 0, 0, 0, 0, "\0", "\0", "\0"};
+	zfs_cmd_t zc = {"\0"};
 	char *atp;
 	char *packed = NULL;
 	size_t size;

lib/libzpool/Makefile.am

@@ -45,6 +45,8 @@ libzpool_la_SOURCES = \
 	$(top_srcdir)/module/zfs/dsl_prop.c \
 	$(top_srcdir)/module/zfs/dsl_scan.c \
 	$(top_srcdir)/module/zfs/dsl_synctask.c \
+	$(top_srcdir)/module/zfs/dsl_destroy.c \
+	$(top_srcdir)/module/zfs/dsl_userhold.c \
 	$(top_srcdir)/module/zfs/fm.c \
 	$(top_srcdir)/module/zfs/gzip.c \
 	$(top_srcdir)/module/zfs/lzjb.c \

lib/libzpool/kernel.c

@@ -34,6 +34,7 @@
 #include <sys/stat.h>
 #include <sys/processor.h>
 #include <sys/zfs_context.h>
+#include <sys/rrwlock.h>
 #include <sys/utsname.h>
 #include <sys/time.h>
 #include <sys/systeminfo.h>
@@ -1042,6 +1043,8 @@ umem_out_of_memory(void)
 void
 kernel_init(int mode)
 {
+	extern uint_t rrw_tsd_key;
+
 	umem_nofail_callback(umem_out_of_memory);
 
 	physmem = sysconf(_SC_PHYS_PAGES);
@@ -1059,6 +1062,8 @@ kernel_init(int mode)
 	system_taskq_init();
 
 	spa_init(mode);
+
+	tsd_create(&rrw_tsd_key, rrw_tsd_destroy);
 }
 
 void

man/man8/zfs.8

@@ -1511,7 +1511,9 @@ Destroy (or mark for deferred destruction) all snapshots with this name in desce
 .ad
 .sp .6
 .RS 4n
-Recursively destroy all dependents.
+Recursively destroy all clones of these snapshots, including the clones,
+snapshots, and children.  If this flag is specified, the \fB-d\fR flag will
+have no effect.
 .RE
 
 .sp
@@ -1547,7 +1549,7 @@ Print verbose information about the deleted data.
 .RE
 
 .sp
-Extreme care should be taken when applying either the \fB-r\fR or the \fB-f\fR
+Extreme care should be taken when applying either the \fB-r\fR or the \fB-R\fR
 options, as they can destroy large portions of a pool and cause unexpected
 behavior for mounted file systems in use.
 .RE

module/nvpair/fnvpair.c

@@ -26,6 +26,7 @@
 #include <sys/nvpair.h>
 #include <sys/kmem.h>
 #include <sys/debug.h>
+#include <sys/param.h>
 #ifndef _KERNEL
 #include <stdlib.h>
 #endif
@@ -114,6 +115,18 @@ fnvlist_merge(nvlist_t *dst, nvlist_t *src)
 	VERIFY0(nvlist_merge(dst, src, KM_SLEEP));
 }
 
+size_t
+fnvlist_num_pairs(nvlist_t *nvl)
+{
+	size_t count = 0;
+	nvpair_t *pair;
+
+	for (pair = nvlist_next_nvpair(nvl, 0); pair != NULL;
+	    pair = nvlist_next_nvpair(nvl, pair))
+		count++;
+	return (count);
+}
+
 void
 fnvlist_add_boolean(nvlist_t *nvl, const char *name)
 {
@@ -563,5 +576,6 @@ EXPORT_SYMBOL(fnvpair_value_int64);
 EXPORT_SYMBOL(fnvpair_value_uint64);
 EXPORT_SYMBOL(fnvpair_value_string);
 EXPORT_SYMBOL(fnvpair_value_nvlist);
+EXPORT_SYMBOL(fnvlist_num_pairs);
 
 #endif

module/zfs/Makefile.in

@@ -93,3 +93,5 @@ $(MODULE)-objs += @top_srcdir@/module/zfs/zpl_super.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/zpl_xattr.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/zrlock.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/zvol.o
+$(MODULE)-objs += @top_srcdir@/module/zfs/dsl_destroy.o
+$(MODULE)-objs += @top_srcdir@/module/zfs/dsl_userhold.o

module/zfs/arc.c

@@ -1643,12 +1643,12 @@ arc_buf_free(arc_buf_t *buf, void *tag)
 	}
 }
 
-int
+boolean_t
 arc_buf_remove_ref(arc_buf_t *buf, void* tag)
 {
 	arc_buf_hdr_t *hdr = buf->b_hdr;
 	kmutex_t *hash_lock = NULL;
-	int no_callback = (buf->b_efunc == NULL);
+	boolean_t no_callback = (buf->b_efunc == NULL);
 
 	if (hdr->b_state == arc_anon) {
 		ASSERT(hdr->b_datacnt == 1);
@@ -1854,7 +1854,7 @@ arc_evict(arc_state_t *state, uint64_t spa, int64_t bytes, boolean_t recycle,
 		ARCSTAT_INCR(arcstat_mutex_miss, missed);
 
 	/*
-	 * We have just evicted some date into the ghost state, make
+	 * We have just evicted some data into the ghost state, make
 	 * sure we also adjust the ghost state size if necessary.
 	 */
 	if (arc_no_grow &&
@@ -2772,7 +2772,7 @@ arc_bcopy_func(zio_t *zio, arc_buf_t *buf, void *arg)
 {
 	if (zio == NULL || zio->io_error == 0)
 		bcopy(buf->b_data, arg, buf->b_hdr->b_size);
-	VERIFY(arc_buf_remove_ref(buf, arg) == 1);
+	VERIFY(arc_buf_remove_ref(buf, arg));
 }
 
 /* a generic arc_done_func_t */
@@ -2781,7 +2781,7 @@ arc_getbuf_func(zio_t *zio, arc_buf_t *buf, void *arg)
 {
 	arc_buf_t **bufp = arg;
 	if (zio && zio->io_error) {
-		VERIFY(arc_buf_remove_ref(buf, arg) == 1);
+		VERIFY(arc_buf_remove_ref(buf, arg));
 		*bufp = NULL;
 	} else {
 		*bufp = buf;

module/zfs/bplist.c

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 #include <sys/bplist.h>
@@ -52,6 +53,12 @@ bplist_append(bplist_t *bpl, const blkptr_t *bp)
 	mutex_exit(&bpl->bpl_lock);
 }
 
+/*
+ * To aid debugging, we keep the most recently removed entry.  This way if
+ * we are in the callback, we can easily locate the entry.
+ */
+static bplist_entry_t *bplist_iterate_last_removed;
+
 void
 bplist_iterate(bplist_t *bpl, bplist_itor_t *func, void *arg, dmu_tx_t *tx)
 {
@@ -59,6 +66,7 @@ bplist_iterate(bplist_t *bpl, bplist_itor_t *func, void *arg, dmu_tx_t *tx)
 
 	mutex_enter(&bpl->bpl_lock);
 	while ((bpe = list_head(&bpl->bpl_list))) {
+		bplist_iterate_last_removed = bpe;
 		list_remove(&bpl->bpl_list, bpe);
 		mutex_exit(&bpl->bpl_lock);
 		func(arg, &bpe->bpe_blk, tx);

module/zfs/bpobj.c

@@ -366,6 +366,7 @@ bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
 {
 	bpobj_t subbpo;
 	uint64_t used, comp, uncomp, subsubobjs;
+	ASSERTV(dmu_object_info_t doi);
 
 	ASSERT(bpo->bpo_havesubobj);
 	ASSERT(bpo->bpo_havecomp);
@@ -392,6 +393,9 @@ bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
 		    DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx);
 	}
 
+	ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
+	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
+
 	mutex_enter(&bpo->bpo_lock);
 	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
 	    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),

module/zfs/dbuf.c

@@ -64,7 +64,7 @@ static void __dbuf_hold_impl_init(struct dbuf_hold_impl_data *dh,
 static int __dbuf_hold_impl(struct dbuf_hold_impl_data *dh);
 
 static void dbuf_destroy(dmu_buf_impl_t *db);
-static int dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
+static boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 static void dbuf_write(dbuf_dirty_record_t *dr, arc_buf_t *data, dmu_tx_t *tx);
 
 /*
@@ -546,7 +546,7 @@ dbuf_read_done(zio_t *zio, arc_buf_t *buf, void *vdb)
 	} else {
 		ASSERT(db->db_blkid != DMU_BONUS_BLKID);
 		ASSERT3P(db->db_buf, ==, NULL);
-		VERIFY(arc_buf_remove_ref(buf, db) == 1);
+		VERIFY(arc_buf_remove_ref(buf, db));
 		db->db_state = DB_UNCACHED;
 	}
 	cv_broadcast(&db->db_changed);
@@ -875,10 +875,12 @@ dbuf_free_range(dnode_t *dn, uint64_t start, uint64_t end, dmu_tx_t *tx)
 			continue;
 
 		/* found a level 0 buffer in the range */
-		if (dbuf_undirty(db, tx))
-			continue;
-
 		mutex_enter(&db->db_mtx);
+		if (dbuf_undirty(db, tx)) {
+			/* mutex has been dropped and dbuf destroyed */
+			continue;
+		}
+
 		if (db->db_state == DB_UNCACHED ||
 		    db->db_state == DB_NOFILL ||
 		    db->db_state == DB_EVICTING) {
@@ -1005,7 +1007,7 @@ dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx)
 
 	mutex_enter(&db->db_mtx);
 	dbuf_set_data(db, buf);
-	VERIFY(arc_buf_remove_ref(obuf, db) == 1);
+	VERIFY(arc_buf_remove_ref(obuf, db));
 	db->db.db_size = size;
 
 	if (db->db_level == 0) {
@@ -1306,7 +1308,10 @@ dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
 	return (dr);
 }
 
-static int
+/*
+ * Return TRUE if this evicted the dbuf.
+ */
+static boolean_t
 dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
 {
 	dnode_t *dn;
@@ -1315,18 +1320,17 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
 
 	ASSERT(txg != 0);
 	ASSERT(db->db_blkid != DMU_BONUS_BLKID);
+	ASSERT0(db->db_level);
+	ASSERT(MUTEX_HELD(&db->db_mtx));
 
-	mutex_enter(&db->db_mtx);
 	/*
 	 * If this buffer is not dirty, we're done.
 	 */
 	for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
 		if (dr->dr_txg <= txg)
 			break;
-	if (dr == NULL || dr->dr_txg < txg) {
+	if (dr == NULL || dr->dr_txg < txg)
-		mutex_exit(&db->db_mtx);
+		return (B_FALSE);
-		return (0);
-	}
 	ASSERT(dr->dr_txg == txg);
 	ASSERT(dr->dr_dbuf == db);
 
@@ -1334,24 +1338,12 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
 	dn = DB_DNODE(db);
 
 	/*
-	 * If this buffer is currently held, we cannot undirty
+	 * Note:  This code will probably work even if there are concurrent
-	 * it, since one of the current holders may be in the
+	 * holders, but it is untested in that scenerio, as the ZPL and
-	 * middle of an update.  Note that users of dbuf_undirty()
+	 * ztest have additional locking (the range locks) that prevents
-	 * should not place a hold on the dbuf before the call.
+	 * that type of concurrent access.
-	 * Also note: we can get here with a spill block, so
-	 * test for that similar to how dbuf_dirty does.
 	 */
-	if (refcount_count(&db->db_holds) > db->db_dirtycnt) {
+	ASSERT3U(refcount_count(&db->db_holds), ==, db->db_dirtycnt);
-		mutex_exit(&db->db_mtx);
-		/* Make sure we don't toss this buffer at sync phase */
-		if (db->db_blkid != DMU_SPILL_BLKID) {
-			mutex_enter(&dn->dn_mtx);
-			dnode_clear_range(dn, db->db_blkid, 1, tx);
-			mutex_exit(&dn->dn_mtx);
-		}
-		DB_DNODE_EXIT(db);
-		return (0);
-	}
 
 	dprintf_dbuf(db, "size=%llx\n", (u_longlong_t)db->db.db_size);
 
@@ -1380,21 +1372,13 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
 	}
 	DB_DNODE_EXIT(db);
 
-	if (db->db_level == 0) {
+	if (db->db_state != DB_NOFILL) {
-		if (db->db_state != DB_NOFILL) {
+		dbuf_unoverride(dr);
-			dbuf_unoverride(dr);
 
-			ASSERT(db->db_buf != NULL);
-			ASSERT(dr->dt.dl.dr_data != NULL);
-			if (dr->dt.dl.dr_data != db->db_buf)
-				VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data,
-				    db) == 1);
-		}
-	} else {
 		ASSERT(db->db_buf != NULL);
-		ASSERT(list_head(&dr->dt.di.dr_children) == NULL);
+		ASSERT(dr->dt.dl.dr_data != NULL);
-		mutex_destroy(&dr->dt.di.dr_mtx);
+		if (dr->dt.dl.dr_data != db->db_buf)
-		list_destroy(&dr->dt.di.dr_children);
+			VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data, db));
 	}
 	kmem_free(dr, sizeof (dbuf_dirty_record_t));
 
@@ -1406,13 +1390,12 @@ dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx)
 
 		ASSERT(db->db_state == DB_NOFILL || arc_released(buf));
 		dbuf_set_data(db, NULL);
-		VERIFY(arc_buf_remove_ref(buf, db) == 1);
+		VERIFY(arc_buf_remove_ref(buf, db));
 		dbuf_evict(db);
-		return (1);
+		return (B_TRUE);
 	}
 
-	mutex_exit(&db->db_mtx);
+	return (B_FALSE);
-	return (0);
 }
 
 #pragma weak dmu_buf_will_dirty = dbuf_will_dirty
@@ -1511,7 +1494,7 @@ dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx)
 		mutex_exit(&db->db_mtx);
 		(void) dbuf_dirty(db, tx);
 		bcopy(buf->b_data, db->db.db_data, db->db.db_size);
-		VERIFY(arc_buf_remove_ref(buf, db) == 1);
+		VERIFY(arc_buf_remove_ref(buf, db));
 		xuio_stat_wbuf_copied();
 		return;
 	}
@@ -1529,10 +1512,10 @@ dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx)
 				arc_release(db->db_buf, db);
 			}
 			dr->dt.dl.dr_data = buf;
-			VERIFY(arc_buf_remove_ref(db->db_buf, db) == 1);
+			VERIFY(arc_buf_remove_ref(db->db_buf, db));
 		} else if (dr == NULL || dr->dt.dl.dr_data != db->db_buf) {
 			arc_release(db->db_buf, db);
-			VERIFY(arc_buf_remove_ref(db->db_buf, db) == 1);
+			VERIFY(arc_buf_remove_ref(db->db_buf, db));
 		}
 		db->db_buf = NULL;
 	}
@@ -2168,10 +2151,10 @@ dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag)
 			 * This dbuf has anonymous data associated with it.
 			 */
 			dbuf_set_data(db, NULL);
-			VERIFY(arc_buf_remove_ref(buf, db) == 1);
+			VERIFY(arc_buf_remove_ref(buf, db));
 			dbuf_evict(db);
 		} else {
-			VERIFY(arc_buf_remove_ref(db->db_buf, db) == 0);
+			VERIFY(!arc_buf_remove_ref(db->db_buf, db));
 
 			/*
 			 * A dbuf will be eligible for eviction if either the
@@ -2669,7 +2652,7 @@ dbuf_write_done(zio_t *zio, arc_buf_t *buf, void *vdb)
 		if (db->db_state != DB_NOFILL) {
 			if (dr->dt.dl.dr_data != db->db_buf)
 				VERIFY(arc_buf_remove_ref(dr->dt.dl.dr_data,
-				    db) == 1);
+				    db));
 			else if (!arc_released(db->db_buf))
 				arc_set_callback(db->db_buf, dbuf_do_evict, db);
 		}

module/zfs/dmu.c

@@ -1382,7 +1382,7 @@ void
 dmu_return_arcbuf(arc_buf_t *buf)
 {
 	arc_return_buf(buf, FTAG);
-	VERIFY(arc_buf_remove_ref(buf, FTAG) == 1);
+	VERIFY(arc_buf_remove_ref(buf, FTAG));
 }
 
 /*

module/zfs/dmu_diff.c

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 #include <sys/dmu.h>
@@ -155,51 +156,49 @@ diff_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 }
 
 int
-dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp, offset_t *offp)
+dmu_diff(const char *tosnap_name, const char *fromsnap_name,
+    struct vnode *vp, offset_t *offp)
 {
 	struct diffarg da;
-	dsl_dataset_t *ds = tosnap->os_dsl_dataset;
+	dsl_dataset_t *fromsnap;
-	dsl_dataset_t *fromds = fromsnap->os_dsl_dataset;
+	dsl_dataset_t *tosnap;
-	dsl_dataset_t *findds;
+	dsl_pool_t *dp;
-	dsl_dataset_t *relds;
+	int error;
-	int err = 0;
+	uint64_t fromtxg;
 
-	/* make certain we are looking at snapshots */
+	if (strchr(tosnap_name, '@') == NULL ||
-	if (!dsl_dataset_is_snapshot(ds) || !dsl_dataset_is_snapshot(fromds))
+	    strchr(fromsnap_name, '@') == NULL)
 		return (EINVAL);
 
-	/* fromsnap must be earlier and from the same lineage as tosnap */
+	error = dsl_pool_hold(tosnap_name, FTAG, &dp);
-	if (fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg)
+	if (error != 0)
-		return (EXDEV);
+		return (error);
 
-	relds = NULL;
+	error = dsl_dataset_hold(dp, tosnap_name, FTAG, &tosnap);
-	findds = ds;
+	if (error != 0) {
-
+		dsl_pool_rele(dp, FTAG);
-	while (fromds->ds_dir != findds->ds_dir) {
+		return (error);
-		dsl_pool_t *dp = ds->ds_dir->dd_pool;
-
-		if (!dsl_dir_is_clone(findds->ds_dir)) {
-			if (relds)
-				dsl_dataset_rele(relds, FTAG);
-			return (EXDEV);
-		}
-
-		rw_enter(&dp->dp_config_rwlock, RW_READER);
-		err = dsl_dataset_hold_obj(dp,
-		    findds->ds_dir->dd_phys->dd_origin_obj, FTAG, &findds);
-		rw_exit(&dp->dp_config_rwlock);
-
-		if (relds)
-			dsl_dataset_rele(relds, FTAG);
-
-		if (err)
-			return (EXDEV);
-
-		relds = findds;
 	}
 
-	if (relds)
+	error = dsl_dataset_hold(dp, fromsnap_name, FTAG, &fromsnap);
-		dsl_dataset_rele(relds, FTAG);
+	if (error != 0) {
+		dsl_dataset_rele(tosnap, FTAG);
+		dsl_pool_rele(dp, FTAG);
+		return (error);
+	}
+
+	if (!dsl_dataset_is_before(tosnap, fromsnap)) {
+		dsl_dataset_rele(fromsnap, FTAG);
+		dsl_dataset_rele(tosnap, FTAG);
+		dsl_pool_rele(dp, FTAG);
+		return (EXDEV);
+	}
+
+	fromtxg = fromsnap->ds_phys->ds_creation_txg;
+	dsl_dataset_rele(fromsnap, FTAG);
+
+	dsl_dataset_long_hold(tosnap, FTAG);
+	dsl_pool_rele(dp, FTAG);
 
 	da.da_vp = vp;
 	da.da_offp = offp;
@@ -207,15 +206,18 @@ dmu_diff(objset_t *tosnap, objset_t *fromsnap, struct vnode *vp, offset_t *offp)
 	da.da_ddr.ddr_first = da.da_ddr.ddr_last = 0;
 	da.da_err = 0;
 
-	err = traverse_dataset(ds, fromds->ds_phys->ds_creation_txg,
+	error = traverse_dataset(tosnap, fromtxg,
 	    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, diff_cb, &da);
 
-	if (err) {
+	if (error != 0) {
-		da.da_err = err;
+		da.da_err = error;
 	} else {
 		/* we set the da.da_err we return as side-effect */
 		(void) write_record(&da);
 	}
 
+	dsl_dataset_long_rele(tosnap, FTAG);
+	dsl_dataset_rele(tosnap, FTAG);
+
 	return (da.da_err);
 }

module/zfs/dmu_traverse.c

@@ -265,7 +265,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 
 		err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
-		if (err)
+		if (err != 0)
 			return (err);
 		cbp = buf->b_data;
 
@@ -282,7 +282,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 			    zb->zb_level - 1,
 			    zb->zb_blkid * epb + i);
 			err = traverse_visitbp(td, dnp, &cbp[i], &czb);
-			if (err) {
+			if (err != 0) {
 				if (!hard)
 					break;
 				lasterr = err;
@@ -295,7 +295,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 
 		err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
-		if (err)
+		if (err != 0)
 			return (err);
 		dnp = buf->b_data;
 
@@ -308,7 +308,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 		for (i = 0; i < epb; i++) {
 			err = traverse_dnode(td, &dnp[i], zb->zb_objset,
 			    zb->zb_blkid * epb + i);
-			if (err) {
+			if (err != 0) {
 				if (!hard)
 					break;
 				lasterr = err;
@@ -321,7 +321,7 @@ traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
 
 		err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
-		if (err)
+		if (err != 0)
 			return (err);
 
 		osp = buf->b_data;
@@ -405,7 +405,7 @@ traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
 	for (j = 0; j < dnp->dn_nblkptr; j++) {
 		SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
 		err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb);
-		if (err) {
+		if (err != 0) {
 			if (!hard)
 				break;
 			lasterr = err;
@@ -415,7 +415,7 @@ traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
 	if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
 		SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
 		err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb);
-		if (err) {
+		if (err != 0) {
 			if (!hard)
 				return (err);
 			lasterr = err;
@@ -518,14 +518,20 @@ traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
 	cv_init(&pd->pd_cv, NULL, CV_DEFAULT, NULL);
 
 	/* See comment on ZIL traversal in dsl_scan_visitds. */
-	if (ds != NULL && !dsl_dataset_is_snapshot(ds)) {
+	if (ds != NULL && !dsl_dataset_is_snapshot(ds) && !BP_IS_HOLE(rootbp)) {
-		objset_t *os;
+		uint32_t flags = ARC_WAIT;
+		objset_phys_t *osp;
+		arc_buf_t *buf;
 
-		err = dmu_objset_from_ds(ds, &os);
+		err = arc_read(NULL, td->td_spa, rootbp,
-		if (err)
+		    arc_getbuf_func, &buf,
+		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL);
+		if (err != 0)
 			return (err);
 
-		traverse_zil(td, &os->os_zil_header);
+		osp = buf->b_data;
+		traverse_zil(td, &osp->os_zil_header);
+		(void) arc_buf_remove_ref(buf, &buf);
 	}
 
 	if (!(flags & TRAVERSE_PREFETCH_DATA) ||
@@ -591,7 +597,7 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
 	/* visit the MOS */
 	err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
 	    txg_start, NULL, flags, func, arg);
-	if (err)
+	if (err != 0)
 		return (err);
 
 	/* visit each dataset */
@@ -600,7 +606,7 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
 		dmu_object_info_t doi;
 
 		err = dmu_object_info(mos, obj, &doi);
-		if (err) {
+		if (err != 0) {
 			if (!hard)
 				return (err);
 			lasterr = err;
@@ -611,10 +617,10 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
 			dsl_dataset_t *ds;
 			uint64_t txg = txg_start;
 
-			rw_enter(&dp->dp_config_rwlock, RW_READER);
+			dsl_pool_config_enter(dp, FTAG);
 			err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
-			rw_exit(&dp->dp_config_rwlock);
+			dsl_pool_config_exit(dp, FTAG);
-			if (err) {
+			if (err != 0) {
 				if (!hard)
 					return (err);
 				lasterr = err;
@@ -624,7 +630,7 @@ traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
 				txg = ds->ds_phys->ds_prev_snap_txg;
 			err = traverse_dataset(ds, txg, flags, func, arg);
 			dsl_dataset_rele(ds, FTAG);
-			if (err) {
+			if (err != 0) {
 				if (!hard)
 					return (err);
 				lasterr = err;

module/zfs/dmu_tx.c

@@ -917,7 +917,7 @@ dmu_tx_dirty_buf(dmu_tx_t *tx, dmu_buf_impl_t *db)
 #endif
 
 static int
-dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how)
+dmu_tx_try_assign(dmu_tx_t *tx, txg_how_t txg_how)
 {
 	dmu_tx_hold_t *txh;
 	spa_t *spa = tx->tx_pool->dp_spa;
@@ -985,15 +985,6 @@ dmu_tx_try_assign(dmu_tx_t *tx, uint64_t txg_how)
 		fudge += txh->txh_fudge;
 	}
 
-	/*
-	 * NB: This check must be after we've held the dnodes, so that
-	 * the dmu_tx_unassign() logic will work properly
-	 */
-	if (txg_how >= TXG_INITIAL && txg_how != tx->tx_txg) {
-		DMU_TX_STAT_BUMP(dmu_tx_how);
-		return (ERESTART);
-	}
-
 	/*
 	 * If a snapshot has been taken since we made our estimates,
 	 * assume that we won't be able to free or overwrite anything.
@@ -1076,29 +1067,28 @@ dmu_tx_unassign(dmu_tx_t *tx)
  *
  * (1)	TXG_WAIT.  If the current open txg is full, waits until there's
  *	a new one.  This should be used when you're not holding locks.
- *	If will only fail if we're truly out of space (or over quota).
+ *	It will only fail if we're truly out of space (or over quota).
  *
  * (2)	TXG_NOWAIT.  If we can't assign into the current open txg without
  *	blocking, returns immediately with ERESTART.  This should be used
  *	whenever you're holding locks.  On an ERESTART error, the caller
  *	should drop locks, do a dmu_tx_wait(tx), and try again.
- *
- * (3)	A specific txg.  Use this if you need to ensure that multiple
- *	transactions all sync in the same txg.  Like TXG_NOWAIT, it
- *	returns ERESTART if it can't assign you into the requested txg.
  */
 int
-dmu_tx_assign(dmu_tx_t *tx, uint64_t txg_how)
+dmu_tx_assign(dmu_tx_t *tx, txg_how_t txg_how)
 {
 	hrtime_t before, after;
 	int err;
 
 	ASSERT(tx->tx_txg == 0);
-	ASSERT(txg_how != 0);
+	ASSERT(txg_how == TXG_WAIT || txg_how == TXG_NOWAIT);
 	ASSERT(!dsl_pool_sync_context(tx->tx_pool));
 
 	before = gethrtime();
 
+	/* If we might wait, we must not hold the config lock. */
+	ASSERT(txg_how != TXG_WAIT || !dsl_pool_config_held(tx->tx_pool));
+
 	while ((err = dmu_tx_try_assign(tx, txg_how)) != 0) {
 		dmu_tx_unassign(tx);
 
@@ -1124,6 +1114,7 @@ dmu_tx_wait(dmu_tx_t *tx)
 	spa_t *spa = tx->tx_pool->dp_spa;
 
 	ASSERT(tx->tx_txg == 0);
+	ASSERT(!dsl_pool_config_held(tx->tx_pool));
 
 	/*
 	 * It's possible that the pool has become active after this thread
@@ -1250,6 +1241,14 @@ dmu_tx_get_txg(dmu_tx_t *tx)
 	return (tx->tx_txg);
 }
 
+dsl_pool_t *
+dmu_tx_pool(dmu_tx_t *tx)
+{
+	ASSERT(tx->tx_pool != NULL);
+	return (tx->tx_pool);
+}
+
+
 void
 dmu_tx_callback_register(dmu_tx_t *tx, dmu_tx_callback_func_t *func, void *data)
 {

module/zfs/dnode.c

@@ -74,7 +74,11 @@ dnode_cons(void *arg, void *unused, int kmflag)
 	mutex_init(&dn->dn_dbufs_mtx, NULL, MUTEX_DEFAULT, NULL);
 	cv_init(&dn->dn_notxholds, NULL, CV_DEFAULT, NULL);
 
-	refcount_create(&dn->dn_holds);
+	/*
+	 * Every dbuf has a reference, and dropping a tracked reference is
+	 * O(number of references), so don't track dn_holds.
+	 */
+	refcount_create_untracked(&dn->dn_holds);
 	refcount_create(&dn->dn_tx_holds);
 	list_link_init(&dn->dn_link);
 

module/zfs/dnode_sync.c

@@ -481,6 +481,7 @@ dnode_sync_free(dnode_t *dn, dmu_tx_t *tx)
 	dnode_undirty_dbufs(&dn->dn_dirty_records[txgoff]);
 	dnode_evict_dbufs(dn);
 	ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL);
+	ASSERT3P(dn->dn_bonus, ==, NULL);
 
 	/*
 	 * XXX - It would be nice to assert this, but we may still

module/zfs/dsl_deleg.c

@@ -147,28 +147,37 @@ dsl_deleg_can_unallow(char *ddname, nvlist_t *nvp, cred_t *cr)
 	return (0);
 }
 
-static void
+typedef struct dsl_deleg_arg {
-dsl_deleg_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+	const char *dda_name;
-{
+	nvlist_t *dda_nvlist;
-	dsl_dir_t *dd = arg1;
+} dsl_deleg_arg_t;
-	nvlist_t *nvp = arg2;
-	objset_t *mos = dd->dd_pool->dp_meta_objset;
-	nvpair_t *whopair = NULL;
-	uint64_t zapobj = dd->dd_phys->dd_deleg_zapobj;
 
+static void
+dsl_deleg_set_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_deleg_arg_t *dda = arg;
+	dsl_dir_t *dd;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	objset_t *mos = dp->dp_meta_objset;
+	nvpair_t *whopair = NULL;
+	uint64_t zapobj;
+
+	VERIFY0(dsl_dir_hold(dp, dda->dda_name, FTAG, &dd, NULL));
+
+	zapobj = dd->dd_phys->dd_deleg_zapobj;
 	if (zapobj == 0) {
 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
 		zapobj = dd->dd_phys->dd_deleg_zapobj = zap_create(mos,
 		    DMU_OT_DSL_PERMS, DMU_OT_NONE, 0, tx);
 	}
 
-	while ((whopair = nvlist_next_nvpair(nvp, whopair))) {
+	while ((whopair = nvlist_next_nvpair(dda->dda_nvlist, whopair))) {
 		const char *whokey = nvpair_name(whopair);
 		nvlist_t *perms;
 		nvpair_t *permpair = NULL;
 		uint64_t jumpobj;
 
-		VERIFY(nvpair_value_nvlist(whopair, &perms) == 0);
+		perms = fnvpair_value_nvlist(whopair);
 
 		if (zap_lookup(mos, zapobj, whokey, 8, 1, &jumpobj) != 0) {
 			jumpobj = zap_create_link(mos, DMU_OT_DSL_PERMS,
@@ -185,21 +194,27 @@ dsl_deleg_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 			    "%s %s", whokey, perm);
 		}
 	}
+	dsl_dir_rele(dd, FTAG);
 }
 
 static void
-dsl_deleg_unset_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_deleg_unset_sync(void *arg, dmu_tx_t *tx)
 {
-	dsl_dir_t *dd = arg1;
+	dsl_deleg_arg_t *dda = arg;
-	nvlist_t *nvp = arg2;
+	dsl_dir_t *dd;
-	objset_t *mos = dd->dd_pool->dp_meta_objset;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	objset_t *mos = dp->dp_meta_objset;
 	nvpair_t *whopair = NULL;
-	uint64_t zapobj = dd->dd_phys->dd_deleg_zapobj;
+	uint64_t zapobj;
 
-	if (zapobj == 0)
+	VERIFY0(dsl_dir_hold(dp, dda->dda_name, FTAG, &dd, NULL));
+	zapobj = dd->dd_phys->dd_deleg_zapobj;
+	if (zapobj == 0) {
+		dsl_dir_rele(dd, FTAG);
 		return;
+	}
 
-	while ((whopair = nvlist_next_nvpair(nvp, whopair))) {
+	while ((whopair = nvlist_next_nvpair(dda->dda_nvlist, whopair))) {
 		const char *whokey = nvpair_name(whopair);
 		nvlist_t *perms;
 		nvpair_t *permpair = NULL;
@@ -234,35 +249,40 @@ dsl_deleg_unset_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 			    "%s %s", whokey, perm);
 		}
 	}
+	dsl_dir_rele(dd, FTAG);
+}
+
+static int
+dsl_deleg_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_deleg_arg_t *dda = arg;
+	dsl_dir_t *dd;
+	int error;
+
+	if (spa_version(dmu_tx_pool(tx)->dp_spa) <
+	    SPA_VERSION_DELEGATED_PERMS) {
+		return (ENOTSUP);
+	}
+
+	error = dsl_dir_hold(dmu_tx_pool(tx), dda->dda_name, FTAG, &dd, NULL);
+	if (error == 0)
+		dsl_dir_rele(dd, FTAG);
+	return (error);
 }
 
 int
 dsl_deleg_set(const char *ddname, nvlist_t *nvp, boolean_t unset)
 {
-	dsl_dir_t *dd;
+	dsl_deleg_arg_t dda;
-	int error;
-	nvpair_t *whopair = NULL;
-	int blocks_modified = 0;
 
-	error = dsl_dir_open(ddname, FTAG, &dd, NULL);
+	/* nvp must already have been verified to be valid */
-	if (error)
-		return (error);
 
-	if (spa_version(dmu_objset_spa(dd->dd_pool->dp_meta_objset)) <
+	dda.dda_name = ddname;
-	    SPA_VERSION_DELEGATED_PERMS) {
+	dda.dda_nvlist = nvp;
-		dsl_dir_close(dd, FTAG);
-		return (ENOTSUP);
-	}
 
-	while ((whopair = nvlist_next_nvpair(nvp, whopair)))
+	return (dsl_sync_task(ddname, dsl_deleg_check,
-		blocks_modified++;
-
-	error = dsl_sync_task_do(dd->dd_pool, NULL,
 	    unset ? dsl_deleg_unset_sync : dsl_deleg_set_sync,
-	    dd, nvp, blocks_modified);
+	    &dda, fnvlist_num_pairs(nvp)));
-	dsl_dir_close(dd, FTAG);
-
-	return (error);
 }
 
 /*
@@ -293,10 +313,16 @@ dsl_deleg_get(const char *ddname, nvlist_t **nvp)
 	zap_attribute_t *baseza, *za;
 	char *source;
 
-	error = dsl_dir_open(ddname, FTAG, &startdd, NULL);
+	error = dsl_pool_hold(ddname, FTAG, &dp);
-	if (error)
+	if (error != 0)
 		return (error);
 
+	error = dsl_dir_hold(dp, ddname, FTAG, &startdd, NULL);
+	if (error != 0) {
+		dsl_pool_rele(dp, FTAG);
+		return (error);
+	}
+
 	dp = startdd->dd_pool;
 	mos = dp->dp_meta_objset;
 
@@ -307,20 +333,16 @@ dsl_deleg_get(const char *ddname, nvlist_t **nvp)
 	source = kmem_alloc(MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, KM_SLEEP);
 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 
-	rw_enter(&dp->dp_config_rwlock, RW_READER);
 	for (dd = startdd; dd != NULL; dd = dd->dd_parent) {
 		nvlist_t *sp_nvp;
 		uint64_t n;
 
-		if (dd->dd_phys->dd_deleg_zapobj &&
+		if (dd->dd_phys->dd_deleg_zapobj == 0 ||
-		    (zap_count(mos, dd->dd_phys->dd_deleg_zapobj,
+		    zap_count(mos, dd->dd_phys->dd_deleg_zapobj, &n) != 0 ||
-		    &n) == 0) && n) {
+		    n == 0)
-			VERIFY(nvlist_alloc(&sp_nvp,
-			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
-		} else {
 			continue;
-		}
 
+		sp_nvp = fnvlist_alloc();
 		for (zap_cursor_init(basezc, mos,
 		    dd->dd_phys->dd_deleg_zapobj);
 		    zap_cursor_retrieve(basezc, baseza) == 0;
@@ -330,27 +352,23 @@ dsl_deleg_get(const char *ddname, nvlist_t **nvp)
 			ASSERT(baseza->za_integer_length == 8);
 			ASSERT(baseza->za_num_integers == 1);
 
-			VERIFY(nvlist_alloc(&perms_nvp,
+			perms_nvp = fnvlist_alloc();
-			    NV_UNIQUE_NAME, KM_SLEEP) == 0);
 			for (zap_cursor_init(zc, mos, baseza->za_first_integer);
 			    zap_cursor_retrieve(zc, za) == 0;
 			    zap_cursor_advance(zc)) {
-				VERIFY(nvlist_add_boolean(perms_nvp,
+				fnvlist_add_boolean(perms_nvp, za->za_name);
-				    za->za_name) == 0);
 			}
 			zap_cursor_fini(zc);
-			VERIFY(nvlist_add_nvlist(sp_nvp, baseza->za_name,
+			fnvlist_add_nvlist(sp_nvp, baseza->za_name, perms_nvp);
-			    perms_nvp) == 0);
+			fnvlist_free(perms_nvp);
-			nvlist_free(perms_nvp);
 		}
 
 		zap_cursor_fini(basezc);
 
 		dsl_dir_name(dd, source);
-		VERIFY(nvlist_add_nvlist(*nvp, source, sp_nvp) == 0);
+		fnvlist_add_nvlist(*nvp, source, sp_nvp);
 		nvlist_free(sp_nvp);
 	}
-	rw_exit(&dp->dp_config_rwlock);
 
 	kmem_free(source, MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
 	kmem_free(baseza, sizeof(zap_attribute_t));
@@ -358,7 +376,8 @@ dsl_deleg_get(const char *ddname, nvlist_t **nvp)
 	kmem_free(za, sizeof(zap_attribute_t));
 	kmem_free(zc, sizeof(zap_cursor_t));
 
-	dsl_dir_close(startdd, FTAG);
+	dsl_dir_rele(startdd, FTAG);
+	dsl_pool_rele(dp, FTAG);
 	return (0);
 }
 
@@ -564,7 +583,7 @@ dsl_deleg_access_impl(dsl_dataset_t *ds, const char *perm, cred_t *cr)
 	avl_create(&permsets, perm_set_compare, sizeof (perm_set_t),
 	    offsetof(perm_set_t, p_node));
 
-	rw_enter(&dp->dp_config_rwlock, RW_READER);
+	ASSERT(dsl_pool_config_held(dp));
 	for (dd = ds->ds_dir; dd != NULL; dd = dd->dd_parent,
 	    checkflag = ZFS_DELEG_DESCENDENT) {
 		uint64_t zapobj;
@@ -625,7 +644,6 @@ again:
 	}
 	error = EPERM;
 success:
-	rw_exit(&dp->dp_config_rwlock);
 
 	cookie = NULL;
 	while ((setnode = avl_destroy_nodes(&permsets, &cookie)) != NULL)
@@ -637,15 +655,19 @@ success:
 int
 dsl_deleg_access(const char *dsname, const char *perm, cred_t *cr)
 {
+	dsl_pool_t *dp;
 	dsl_dataset_t *ds;
 	int error;
 
-	error = dsl_dataset_hold(dsname, FTAG, &ds);
+	error = dsl_pool_hold(dsname, FTAG, &dp);
-	if (error)
+	if (error != 0)
 		return (error);
-
+	error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
-	error = dsl_deleg_access_impl(ds, perm, cr);
+	if (error == 0) {
-	dsl_dataset_rele(ds, FTAG);
+		error = dsl_deleg_access_impl(ds, perm, cr);
+		dsl_dataset_rele(ds, FTAG);
+	}
+	dsl_pool_rele(dp, FTAG);
 
 	return (error);
 }

module/zfs/dsl_destroy.c

@@ -0,0 +1,940 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/dsl_userhold.h>
+#include <sys/dsl_dataset.h>
+#include <sys/dsl_synctask.h>
+#include <sys/dmu_tx.h>
+#include <sys/dsl_pool.h>
+#include <sys/dsl_dir.h>
+#include <sys/dmu_traverse.h>
+#include <sys/dsl_scan.h>
+#include <sys/dmu_objset.h>
+#include <sys/zap.h>
+#include <sys/zfeature.h>
+#include <sys/zfs_ioctl.h>
+#include <sys/dsl_deleg.h>
+
+typedef struct dmu_snapshots_destroy_arg {
+	nvlist_t *dsda_snaps;
+	nvlist_t *dsda_successful_snaps;
+	boolean_t dsda_defer;
+	nvlist_t *dsda_errlist;
+} dmu_snapshots_destroy_arg_t;
+
+/*
+ * ds must be owned.
+ */
+static int
+dsl_destroy_snapshot_check_impl(dsl_dataset_t *ds, boolean_t defer)
+{
+	if (!dsl_dataset_is_snapshot(ds))
+		return (EINVAL);
+
+	if (dsl_dataset_long_held(ds))
+		return (EBUSY);
+
+	/*
+	 * Only allow deferred destroy on pools that support it.
+	 * NOTE: deferred destroy is only supported on snapshots.
+	 */
+	if (defer) {
+		if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
+		    SPA_VERSION_USERREFS)
+			return (ENOTSUP);
+		return (0);
+	}
+
+	/*
+	 * If this snapshot has an elevated user reference count,
+	 * we can't destroy it yet.
+	 */
+	if (ds->ds_userrefs > 0)
+		return (EBUSY);
+
+	/*
+	 * Can't delete a branch point.
+	 */
+	if (ds->ds_phys->ds_num_children > 1)
+		return (EEXIST);
+
+	return (0);
+}
+
+static int
+dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
+{
+	dmu_snapshots_destroy_arg_t *dsda = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	nvpair_t *pair;
+	int error = 0;
+
+	if (!dmu_tx_is_syncing(tx))
+		return (0);
+
+	for (pair = nvlist_next_nvpair(dsda->dsda_snaps, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(dsda->dsda_snaps, pair)) {
+		dsl_dataset_t *ds;
+
+		error = dsl_dataset_hold(dp, nvpair_name(pair),
+		    FTAG, &ds);
+
+		/*
+		 * If the snapshot does not exist, silently ignore it
+		 * (it's "already destroyed").
+		 */
+		if (error == ENOENT)
+			continue;
+
+		if (error == 0) {
+			error = dsl_destroy_snapshot_check_impl(ds,
+			    dsda->dsda_defer);
+			dsl_dataset_rele(ds, FTAG);
+		}
+
+		if (error == 0) {
+			fnvlist_add_boolean(dsda->dsda_successful_snaps,
+			    nvpair_name(pair));
+		} else {
+			fnvlist_add_int32(dsda->dsda_errlist,
+			    nvpair_name(pair), error);
+		}
+	}
+
+	pair = nvlist_next_nvpair(dsda->dsda_errlist, NULL);
+	if (pair != NULL)
+		return (fnvpair_value_int32(pair));
+	return (0);
+}
+
+struct process_old_arg {
+	dsl_dataset_t *ds;
+	dsl_dataset_t *ds_prev;
+	boolean_t after_branch_point;
+	zio_t *pio;
+	uint64_t used, comp, uncomp;
+};
+
+static int
+process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
+{
+	struct process_old_arg *poa = arg;
+	dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
+
+	if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
+		dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
+		if (poa->ds_prev && !poa->after_branch_point &&
+		    bp->blk_birth >
+		    poa->ds_prev->ds_phys->ds_prev_snap_txg) {
+			poa->ds_prev->ds_phys->ds_unique_bytes +=
+			    bp_get_dsize_sync(dp->dp_spa, bp);
+		}
+	} else {
+		poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
+		poa->comp += BP_GET_PSIZE(bp);
+		poa->uncomp += BP_GET_UCSIZE(bp);
+		dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
+	}
+	return (0);
+}
+
+static void
+process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
+    dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
+{
+	struct process_old_arg poa = { 0 };
+	dsl_pool_t *dp = ds->ds_dir->dd_pool;
+	objset_t *mos = dp->dp_meta_objset;
+	uint64_t deadlist_obj;
+
+	ASSERT(ds->ds_deadlist.dl_oldfmt);
+	ASSERT(ds_next->ds_deadlist.dl_oldfmt);
+
+	poa.ds = ds;
+	poa.ds_prev = ds_prev;
+	poa.after_branch_point = after_branch_point;
+	poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
+	VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
+	    process_old_cb, &poa, tx));
+	VERIFY0(zio_wait(poa.pio));
+	ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
+
+	/* change snapused */
+	dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
+	    -poa.used, -poa.comp, -poa.uncomp, tx);
+
+	/* swap next's deadlist to our deadlist */
+	dsl_deadlist_close(&ds->ds_deadlist);
+	dsl_deadlist_close(&ds_next->ds_deadlist);
+	deadlist_obj = ds->ds_phys->ds_deadlist_obj;
+	ds->ds_phys->ds_deadlist_obj = ds_next->ds_phys->ds_deadlist_obj;
+	ds_next->ds_phys->ds_deadlist_obj = deadlist_obj;
+	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
+	dsl_deadlist_open(&ds_next->ds_deadlist, mos,
+	    ds_next->ds_phys->ds_deadlist_obj);
+}
+
+static void
+dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
+{
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	zap_cursor_t zc;
+	zap_attribute_t za;
+
+	/*
+	 * If it is the old version, dd_clones doesn't exist so we can't
+	 * find the clones, but dsl_deadlist_remove_key() is a no-op so it
+	 * doesn't matter.
+	 */
+	if (ds->ds_dir->dd_phys->dd_clones == 0)
+		return;
+
+	for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
+	    zap_cursor_retrieve(&zc, &za) == 0;
+	    zap_cursor_advance(&zc)) {
+		dsl_dataset_t *clone;
+
+		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
+		    za.za_first_integer, FTAG, &clone));
+		if (clone->ds_dir->dd_origin_txg > mintxg) {
+			dsl_deadlist_remove_key(&clone->ds_deadlist,
+			    mintxg, tx);
+			dsl_dataset_remove_clones_key(clone, mintxg, tx);
+		}
+		dsl_dataset_rele(clone, FTAG);
+	}
+	zap_cursor_fini(&zc);
+}
+
+void
+dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
+{
+#ifdef ZFS_DEBUG
+	int err;
+#endif
+	int after_branch_point = FALSE;
+	dsl_pool_t *dp = ds->ds_dir->dd_pool;
+	objset_t *mos = dp->dp_meta_objset;
+	dsl_dataset_t *ds_prev = NULL;
+	uint64_t obj, old_unique, used = 0, comp = 0, uncomp = 0;
+	dsl_dataset_t *ds_next, *ds_head, *hds;
+
+
+	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
+	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
+	ASSERT(refcount_is_zero(&ds->ds_longholds));
+
+	if (defer &&
+	    (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1)) {
+		ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
+		dmu_buf_will_dirty(ds->ds_dbuf, tx);
+		ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
+		spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
+		return;
+	}
+
+	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
+
+	/* We need to log before removing it from the namespace. */
+	spa_history_log_internal_ds(ds, "destroy", tx, "");
+
+	dsl_scan_ds_destroyed(ds, tx);
+
+	obj = ds->ds_object;
+
+	if (ds->ds_phys->ds_prev_snap_obj != 0) {
+		ASSERT3P(ds->ds_prev, ==, NULL);
+		VERIFY0(dsl_dataset_hold_obj(dp,
+		    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
+		after_branch_point =
+		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
+
+		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
+		if (after_branch_point &&
+		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
+			dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
+			if (ds->ds_phys->ds_next_snap_obj != 0) {
+				VERIFY0(zap_add_int(mos,
+				    ds_prev->ds_phys->ds_next_clones_obj,
+				    ds->ds_phys->ds_next_snap_obj, tx));
+			}
+		}
+		if (!after_branch_point) {
+			ds_prev->ds_phys->ds_next_snap_obj =
+			    ds->ds_phys->ds_next_snap_obj;
+		}
+	}
+
+	VERIFY0(dsl_dataset_hold_obj(dp,
+	    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
+	ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
+
+	old_unique = ds_next->ds_phys->ds_unique_bytes;
+
+	dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
+	ds_next->ds_phys->ds_prev_snap_obj =
+	    ds->ds_phys->ds_prev_snap_obj;
+	ds_next->ds_phys->ds_prev_snap_txg =
+	    ds->ds_phys->ds_prev_snap_txg;
+	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
+	    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
+
+	if (ds_next->ds_deadlist.dl_oldfmt) {
+		process_old_deadlist(ds, ds_prev, ds_next,
+		    after_branch_point, tx);
+	} else {
+		/* Adjust prev's unique space. */
+		if (ds_prev && !after_branch_point) {
+			dsl_deadlist_space_range(&ds_next->ds_deadlist,
+			    ds_prev->ds_phys->ds_prev_snap_txg,
+			    ds->ds_phys->ds_prev_snap_txg,
+			    &used, &comp, &uncomp);
+			ds_prev->ds_phys->ds_unique_bytes += used;
+		}
+
+		/* Adjust snapused. */
+		dsl_deadlist_space_range(&ds_next->ds_deadlist,
+		    ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
+		    &used, &comp, &uncomp);
+		dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
+		    -used, -comp, -uncomp, tx);
+
+		/* Move blocks to be freed to pool's free list. */
+		dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
+		    &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
+		    tx);
+		dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
+		    DD_USED_HEAD, used, comp, uncomp, tx);
+
+		/* Merge our deadlist into next's and free it. */
+		dsl_deadlist_merge(&ds_next->ds_deadlist,
+		    ds->ds_phys->ds_deadlist_obj, tx);
+	}
+	dsl_deadlist_close(&ds->ds_deadlist);
+	dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
+	dmu_buf_will_dirty(ds->ds_dbuf, tx);
+	ds->ds_phys->ds_deadlist_obj = 0;
+
+	/* Collapse range in clone heads */
+	dsl_dataset_remove_clones_key(ds,
+	    ds->ds_phys->ds_creation_txg, tx);
+
+	if (dsl_dataset_is_snapshot(ds_next)) {
+		dsl_dataset_t *ds_nextnext;
+
+		/*
+		 * Update next's unique to include blocks which
+		 * were previously shared by only this snapshot
+		 * and it.  Those blocks will be born after the
+		 * prev snap and before this snap, and will have
+		 * died after the next snap and before the one
+		 * after that (ie. be on the snap after next's
+		 * deadlist).
+		 */
+		VERIFY0(dsl_dataset_hold_obj(dp,
+		    ds_next->ds_phys->ds_next_snap_obj, FTAG, &ds_nextnext));
+		dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
+		    ds->ds_phys->ds_prev_snap_txg,
+		    ds->ds_phys->ds_creation_txg,
+		    &used, &comp, &uncomp);
+		ds_next->ds_phys->ds_unique_bytes += used;
+		dsl_dataset_rele(ds_nextnext, FTAG);
+		ASSERT3P(ds_next->ds_prev, ==, NULL);
+
+		/* Collapse range in this head. */
+		VERIFY0(dsl_dataset_hold_obj(dp,
+		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &hds));
+		dsl_deadlist_remove_key(&hds->ds_deadlist,
+		    ds->ds_phys->ds_creation_txg, tx);
+		dsl_dataset_rele(hds, FTAG);
+
+	} else {
+		ASSERT3P(ds_next->ds_prev, ==, ds);
+		dsl_dataset_rele(ds_next->ds_prev, ds_next);
+		ds_next->ds_prev = NULL;
+		if (ds_prev) {
+			VERIFY0(dsl_dataset_hold_obj(dp,
+			    ds->ds_phys->ds_prev_snap_obj,
+			    ds_next, &ds_next->ds_prev));
+		}
+
+		dsl_dataset_recalc_head_uniq(ds_next);
+
+		/*
+		 * Reduce the amount of our unconsumed refreservation
+		 * being charged to our parent by the amount of
+		 * new unique data we have gained.
+		 */
+		if (old_unique < ds_next->ds_reserved) {
+			int64_t mrsdelta;
+			uint64_t new_unique =
+			    ds_next->ds_phys->ds_unique_bytes;
+
+			ASSERT(old_unique <= new_unique);
+			mrsdelta = MIN(new_unique - old_unique,
+			    ds_next->ds_reserved - old_unique);
+			dsl_dir_diduse_space(ds->ds_dir,
+			    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
+		}
+	}
+	dsl_dataset_rele(ds_next, FTAG);
+
+	/*
+	 * This must be done after the dsl_traverse(), because it will
+	 * re-open the objset.
+	 */
+	if (ds->ds_objset) {
+		dmu_objset_evict(ds->ds_objset);
+		ds->ds_objset = NULL;
+	}
+
+	/* remove from snapshot namespace */
+	ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
+	VERIFY0(dsl_dataset_hold_obj(dp,
+	    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
+	VERIFY0(dsl_dataset_get_snapname(ds));
+#ifdef ZFS_DEBUG
+	{
+		uint64_t val;
+
+		err = dsl_dataset_snap_lookup(ds_head,
+		    ds->ds_snapname, &val);
+		ASSERT0(err);
+		ASSERT3U(val, ==, obj);
+	}
+#endif
+	VERIFY0(dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx));
+	dsl_dataset_rele(ds_head, FTAG);
+
+	if (ds_prev != NULL)
+		dsl_dataset_rele(ds_prev, FTAG);
+
+	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
+
+	if (ds->ds_phys->ds_next_clones_obj != 0) {
+		ASSERTV(uint64_t count);
+		ASSERT0(zap_count(mos,
+		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
+		VERIFY0(dmu_object_free(mos,
+		    ds->ds_phys->ds_next_clones_obj, tx));
+	}
+	if (ds->ds_phys->ds_props_obj != 0)
+		VERIFY0(zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
+	if (ds->ds_phys->ds_userrefs_obj != 0)
+		VERIFY0(zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
+	dsl_dir_rele(ds->ds_dir, ds);
+	ds->ds_dir = NULL;
+	VERIFY0(dmu_object_free(mos, obj, tx));
+}
+
+static void
+dsl_destroy_snapshot_sync(void *arg, dmu_tx_t *tx)
+{
+	dmu_snapshots_destroy_arg_t *dsda = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	nvpair_t *pair;
+
+	for (pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, NULL);
+	    pair != NULL;
+	    pair = nvlist_next_nvpair(dsda->dsda_successful_snaps, pair)) {
+		dsl_dataset_t *ds;
+
+		VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
+
+		dsl_destroy_snapshot_sync_impl(ds, dsda->dsda_defer, tx);
+		dsl_dataset_rele(ds, FTAG);
+	}
+}
+
+/*
+ * The semantics of this function are described in the comment above
+ * lzc_destroy_snaps().  To summarize:
+ *
+ * The snapshots must all be in the same pool.
+ *
+ * Snapshots that don't exist will be silently ignored (considered to be
+ * "already deleted").
+ *
+ * On success, all snaps will be destroyed and this will return 0.
+ * On failure, no snaps will be destroyed, the errlist will be filled in,
+ * and this will return an errno.
+ */
+int
+dsl_destroy_snapshots_nvl(nvlist_t *snaps, boolean_t defer,
+    nvlist_t *errlist)
+{
+	dmu_snapshots_destroy_arg_t dsda;
+	int error;
+	nvpair_t *pair;
+
+	pair = nvlist_next_nvpair(snaps, NULL);
+	if (pair == NULL)
+		return (0);
+
+	dsda.dsda_snaps = snaps;
+        VERIFY0(nvlist_alloc(&dsda.dsda_successful_snaps, NV_UNIQUE_NAME, KM_PUSHPAGE));
+	dsda.dsda_defer = defer;
+	dsda.dsda_errlist = errlist;
+
+	error = dsl_sync_task(nvpair_name(pair),
+	    dsl_destroy_snapshot_check, dsl_destroy_snapshot_sync,
+	    &dsda, 0);
+	fnvlist_free(dsda.dsda_successful_snaps);
+
+	return (error);
+}
+
+int
+dsl_destroy_snapshot(const char *name, boolean_t defer)
+{
+	int error;
+	nvlist_t *nvl;
+	nvlist_t *errlist;
+
+        VERIFY0(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_PUSHPAGE));
+        VERIFY0(nvlist_alloc(&errlist, NV_UNIQUE_NAME, KM_PUSHPAGE));
+
+	fnvlist_add_boolean(nvl, name);
+	error = dsl_destroy_snapshots_nvl(nvl, defer, errlist);
+	fnvlist_free(errlist);
+	fnvlist_free(nvl);
+	return (error);
+}
+
+struct killarg {
+	dsl_dataset_t *ds;
+	dmu_tx_t *tx;
+};
+
+/* ARGSUSED */
+static int
+kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
+    const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
+{
+	struct killarg *ka = arg;
+	dmu_tx_t *tx = ka->tx;
+
+	if (bp == NULL)
+		return (0);
+
+	if (zb->zb_level == ZB_ZIL_LEVEL) {
+		ASSERT(zilog != NULL);
+		/*
+		 * It's a block in the intent log.  It has no
+		 * accounting, so just free it.
+		 */
+		dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
+	} else {
+		ASSERT(zilog == NULL);
+		ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
+		(void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
+	}
+
+	return (0);
+}
+
+static void
+old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
+{
+	struct killarg ka;
+
+	/*
+	 * Free everything that we point to (that's born after
+	 * the previous snapshot, if we are a clone)
+	 *
+	 * NB: this should be very quick, because we already
+	 * freed all the objects in open context.
+	 */
+	ka.ds = ds;
+	ka.tx = tx;
+	VERIFY0(traverse_dataset(ds,
+	    ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
+	    kill_blkptr, &ka));
+	ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
+}
+
+typedef struct dsl_destroy_head_arg {
+	const char *ddha_name;
+} dsl_destroy_head_arg_t;
+
+int
+dsl_destroy_head_check_impl(dsl_dataset_t *ds, int expected_holds)
+{
+	int error;
+	uint64_t count;
+	objset_t *mos;
+
+	if (dsl_dataset_is_snapshot(ds))
+		return (EINVAL);
+
+	if (refcount_count(&ds->ds_longholds) != expected_holds)
+		return (EBUSY);
+
+	mos = ds->ds_dir->dd_pool->dp_meta_objset;
+
+	/*
+	 * Can't delete a head dataset if there are snapshots of it.
+	 * (Except if the only snapshots are from the branch we cloned
+	 * from.)
+	 */
+	if (ds->ds_prev != NULL &&
+	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
+		return (EBUSY);
+
+	/*
+	 * Can't delete if there are children of this fs.
+	 */
+	error = zap_count(mos,
+	    ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
+	if (error != 0)
+		return (error);
+	if (count != 0)
+		return (EEXIST);
+
+	if (dsl_dir_is_clone(ds->ds_dir) && DS_IS_DEFER_DESTROY(ds->ds_prev) &&
+	    ds->ds_prev->ds_phys->ds_num_children == 2 &&
+	    ds->ds_prev->ds_userrefs == 0) {
+		/* We need to remove the origin snapshot as well. */
+		if (!refcount_is_zero(&ds->ds_prev->ds_longholds))
+			return (EBUSY);
+	}
+	return (0);
+}
+
+static int
+dsl_destroy_head_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_destroy_head_arg_t *ddha = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	dsl_dataset_t *ds;
+	int error;
+
+	error = dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds);
+	if (error != 0)
+		return (error);
+
+	error = dsl_destroy_head_check_impl(ds, 0);
+	dsl_dataset_rele(ds, FTAG);
+	return (error);
+}
+
+static void
+dsl_dir_destroy_sync(uint64_t ddobj, dmu_tx_t *tx)
+{
+	dsl_dir_t *dd;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	objset_t *mos = dp->dp_meta_objset;
+	dd_used_t t;
+
+	ASSERT(RRW_WRITE_HELD(&dmu_tx_pool(tx)->dp_config_rwlock));
+
+	VERIFY0(dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd));
+
+	ASSERT0(dd->dd_phys->dd_head_dataset_obj);
+
+	/*
+	 * Remove our reservation. The impl() routine avoids setting the
+	 * actual property, which would require the (already destroyed) ds.
+	 */
+	dsl_dir_set_reservation_sync_impl(dd, 0, tx);
+
+	ASSERT0(dd->dd_phys->dd_used_bytes);
+	ASSERT0(dd->dd_phys->dd_reserved);
+	for (t = 0; t < DD_USED_NUM; t++)
+		ASSERT0(dd->dd_phys->dd_used_breakdown[t]);
+
+	VERIFY0(zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
+	VERIFY0(zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
+	VERIFY0(dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
+	VERIFY0(zap_remove(mos,
+	    dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
+
+	dsl_dir_rele(dd, FTAG);
+	VERIFY0(dmu_object_free(mos, ddobj, tx));
+}
+
+void
+dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
+{
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	objset_t *mos = dp->dp_meta_objset;
+	uint64_t obj, ddobj, prevobj = 0;
+	boolean_t rmorigin;
+	zfeature_info_t *async_destroy;
+	objset_t *os;
+
+	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
+	ASSERT(ds->ds_prev == NULL ||
+	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
+	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
+	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
+
+	/* We need to log before removing it from the namespace. */
+	spa_history_log_internal_ds(ds, "destroy", tx, "");
+
+	rmorigin = (dsl_dir_is_clone(ds->ds_dir) &&
+	    DS_IS_DEFER_DESTROY(ds->ds_prev) &&
+	    ds->ds_prev->ds_phys->ds_num_children == 2 &&
+	    ds->ds_prev->ds_userrefs == 0);
+
+	/* Remove our reservation */
+	if (ds->ds_reserved != 0) {
+		dsl_dataset_set_refreservation_sync_impl(ds,
+		    (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
+		    0, tx);
+		ASSERT0(ds->ds_reserved);
+	}
+
+	dsl_scan_ds_destroyed(ds, tx);
+
+	obj = ds->ds_object;
+
+	if (ds->ds_phys->ds_prev_snap_obj != 0) {
+		/* This is a clone */
+		ASSERT(ds->ds_prev != NULL);
+		ASSERT3U(ds->ds_prev->ds_phys->ds_next_snap_obj, !=, obj);
+		ASSERT0(ds->ds_phys->ds_next_snap_obj);
+
+		dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
+		if (ds->ds_prev->ds_phys->ds_next_clones_obj != 0) {
+			dsl_dataset_remove_from_next_clones(ds->ds_prev,
+			    obj, tx);
+		}
+
+		ASSERT3U(ds->ds_prev->ds_phys->ds_num_children, >, 1);
+		ds->ds_prev->ds_phys->ds_num_children--;
+	}
+
+	async_destroy =
+	    &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
+
+	/*
+	 * Destroy the deadlist.  Unless it's a clone, the
+	 * deadlist should be empty.  (If it's a clone, it's
+	 * safe to ignore the deadlist contents.)
+	 */
+	dsl_deadlist_close(&ds->ds_deadlist);
+	dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
+	dmu_buf_will_dirty(ds->ds_dbuf, tx);
+	ds->ds_phys->ds_deadlist_obj = 0;
+
+	VERIFY0(dmu_objset_from_ds(ds, &os));
+
+	if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
+		old_synchronous_dataset_destroy(ds, tx);
+	} else {
+		/*
+		 * Move the bptree into the pool's list of trees to
+		 * clean up and update space accounting information.
+		 */
+		uint64_t used, comp, uncomp;
+
+		zil_destroy_sync(dmu_objset_zil(os), tx);
+
+		if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
+			spa_feature_incr(dp->dp_spa, async_destroy, tx);
+			dp->dp_bptree_obj = bptree_alloc(mos, tx);
+			VERIFY0(zap_add(mos,
+			    DMU_POOL_DIRECTORY_OBJECT,
+			    DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
+			    &dp->dp_bptree_obj, tx));
+		}
+
+		used = ds->ds_dir->dd_phys->dd_used_bytes;
+		comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
+		uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
+
+		ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
+		    ds->ds_phys->ds_unique_bytes == used);
+
+		bptree_add(mos, dp->dp_bptree_obj,
+		    &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
+		    used, comp, uncomp, tx);
+		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
+		    -used, -comp, -uncomp, tx);
+		dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
+		    used, comp, uncomp, tx);
+	}
+
+	if (ds->ds_prev != NULL) {
+		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
+			VERIFY0(zap_remove_int(mos,
+			    ds->ds_prev->ds_dir->dd_phys->dd_clones,
+			    ds->ds_object, tx));
+		}
+		prevobj = ds->ds_prev->ds_object;
+		dsl_dataset_rele(ds->ds_prev, ds);
+		ds->ds_prev = NULL;
+	}
+
+	/*
+	 * This must be done after the dsl_traverse(), because it will
+	 * re-open the objset.
+	 */
+	if (ds->ds_objset) {
+		dmu_objset_evict(ds->ds_objset);
+		ds->ds_objset = NULL;
+	}
+
+	/* Erase the link in the dir */
+	dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
+	ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
+	ddobj = ds->ds_dir->dd_object;
+	ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
+	VERIFY0(zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx));
+
+	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
+
+	ASSERT0(ds->ds_phys->ds_next_clones_obj);
+	ASSERT0(ds->ds_phys->ds_props_obj);
+	ASSERT0(ds->ds_phys->ds_userrefs_obj);
+	dsl_dir_rele(ds->ds_dir, ds);
+	ds->ds_dir = NULL;
+	VERIFY0(dmu_object_free(mos, obj, tx));
+
+	dsl_dir_destroy_sync(ddobj, tx);
+
+	if (rmorigin) {
+		dsl_dataset_t *prev;
+		VERIFY0(dsl_dataset_hold_obj(dp, prevobj, FTAG, &prev));
+		dsl_destroy_snapshot_sync_impl(prev, B_FALSE, tx);
+		dsl_dataset_rele(prev, FTAG);
+	}
+}
+
+static void
+dsl_destroy_head_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_destroy_head_arg_t *ddha = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	dsl_dataset_t *ds;
+
+	VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
+	dsl_destroy_head_sync_impl(ds, tx);
+	dsl_dataset_rele(ds, FTAG);
+}
+
+static void
+dsl_destroy_head_begin_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_destroy_head_arg_t *ddha = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	dsl_dataset_t *ds;
+
+	VERIFY0(dsl_dataset_hold(dp, ddha->ddha_name, FTAG, &ds));
+
+	/* Mark it as inconsistent on-disk, in case we crash */
+	dmu_buf_will_dirty(ds->ds_dbuf, tx);
+	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
+
+	spa_history_log_internal_ds(ds, "destroy begin", tx, "");
+	dsl_dataset_rele(ds, FTAG);
+}
+
+int
+dsl_destroy_head(const char *name)
+{
+	dsl_destroy_head_arg_t ddha;
+	int error;
+	spa_t *spa;
+	boolean_t isenabled;
+
+#ifdef _KERNEL
+	zfs_destroy_unmount_origin(name);
+#endif
+
+	error = spa_open(name, &spa, FTAG);
+	if (error != 0)
+		return (error);
+	isenabled = spa_feature_is_enabled(spa,
+	    &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]);
+	spa_close(spa, FTAG);
+
+	ddha.ddha_name = name;
+
+	if (!isenabled) {
+		objset_t *os;
+
+		error = dsl_sync_task(name, dsl_destroy_head_check,
+		    dsl_destroy_head_begin_sync, &ddha, 0);
+		if (error != 0)
+			return (error);
+
+		/*
+		 * Head deletion is processed in one txg on old pools;
+		 * remove the objects from open context so that the txg sync
+		 * is not too long.
+		 */
+		error = dmu_objset_own(name, DMU_OST_ANY, B_FALSE, FTAG, &os);
+		if (error == 0) {
+			uint64_t obj;
+			uint64_t prev_snap_txg =
+			    dmu_objset_ds(os)->ds_phys->ds_prev_snap_txg;
+			for (obj = 0; error == 0;
+			    error = dmu_object_next(os, &obj, FALSE,
+			    prev_snap_txg))
+				(void) dmu_free_object(os, obj);
+			/* sync out all frees */
+			txg_wait_synced(dmu_objset_pool(os), 0);
+			dmu_objset_disown(os, FTAG);
+		}
+	}
+
+	return (dsl_sync_task(name, dsl_destroy_head_check,
+	    dsl_destroy_head_sync, &ddha, 0));
+}
+
+/*
+ * Note, this function is used as the callback for dmu_objset_find().  We
+ * always return 0 so that we will continue to find and process
+ * inconsistent datasets, even if we encounter an error trying to
+ * process one of them.
+ */
+/* ARGSUSED */
+int
+dsl_destroy_inconsistent(const char *dsname, void *arg)
+{
+	objset_t *os;
+
+	if (dmu_objset_hold(dsname, FTAG, &os) == 0) {
+		boolean_t inconsistent = DS_IS_INCONSISTENT(dmu_objset_ds(os));
+		dmu_objset_rele(os, FTAG);
+		if (inconsistent)
+			(void) dsl_destroy_head(dsname);
+	}
+	return (0);
+}
+
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+EXPORT_SYMBOL(dsl_destroy_head);
+EXPORT_SYMBOL(dsl_destroy_head_sync_impl);
+EXPORT_SYMBOL(dsl_dataset_user_hold_check_one);
+EXPORT_SYMBOL(dsl_destroy_snapshot_sync_impl);
+EXPORT_SYMBOL(dsl_destroy_inconsistent);
+EXPORT_SYMBOL(dsl_dataset_user_release_tmp);
+EXPORT_SYMBOL(dsl_destroy_head_check_impl);
+#endif

module/zfs/dsl_dir.c

@@ -40,8 +40,6 @@
 #include "zfs_namecheck.h"
 
 static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd);
-static void dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd,
-    uint64_t value, dmu_tx_t *tx);
 
 /* ARGSUSED */
 static void
@@ -58,7 +56,7 @@ dsl_dir_evict(dmu_buf_t *db, void *arg)
 	}
 
 	if (dd->dd_parent)
-		dsl_dir_close(dd->dd_parent, dd);
+		dsl_dir_rele(dd->dd_parent, dd);
 
 	spa_close(dd->dd_pool->dp_spa, dd);
 
@@ -72,18 +70,17 @@ dsl_dir_evict(dmu_buf_t *db, void *arg)
 }
 
 int
-dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
+dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
     const char *tail, void *tag, dsl_dir_t **ddp)
 {
 	dmu_buf_t *dbuf;
 	dsl_dir_t *dd;
 	int err;
 
-	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
+	ASSERT(dsl_pool_config_held(dp));
-	    dsl_pool_sync_context(dp));
 
 	err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
-	if (err)
+	if (err != 0)
 		return (err);
 	dd = dmu_buf_get_user(dbuf);
 #ifdef ZFS_DEBUG
@@ -110,9 +107,9 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 		dsl_dir_snap_cmtime_update(dd);
 
 		if (dd->dd_phys->dd_parent_obj) {
-			err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj,
+			err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
 			    NULL, dd, &dd->dd_parent);
-			if (err)
+			if (err != 0)
 				goto errout;
 			if (tail) {
 #ifdef ZFS_DEBUG
@@ -129,7 +126,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 				    dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 				    ddobj, 0, dd->dd_myname);
 			}
-			if (err)
+			if (err != 0)
 				goto errout;
 		} else {
 			(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
@@ -146,7 +143,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 			 */
 			err = dmu_bonus_hold(dp->dp_meta_objset,
 			    dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
-			if (err)
+			if (err != 0)
 				goto errout;
 			origin_phys = origin_bonus->db_data;
 			dd->dd_origin_txg =
@@ -158,7 +155,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 		    dsl_dir_evict);
 		if (winner) {
 			if (dd->dd_parent)
-				dsl_dir_close(dd->dd_parent, dd);
+				dsl_dir_rele(dd->dd_parent, dd);
 			mutex_destroy(&dd->dd_lock);
 			kmem_free(dd, sizeof (dsl_dir_t));
 			dd = winner;
@@ -185,7 +182,7 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 
 errout:
 	if (dd->dd_parent)
-		dsl_dir_close(dd->dd_parent, dd);
+		dsl_dir_rele(dd->dd_parent, dd);
 	mutex_destroy(&dd->dd_lock);
 	kmem_free(dd, sizeof (dsl_dir_t));
 	dmu_buf_rele(dbuf, tag);
@@ -193,7 +190,7 @@ errout:
 }
 
 void
-dsl_dir_close(dsl_dir_t *dd, void *tag)
+dsl_dir_rele(dsl_dir_t *dd, void *tag)
 {
 	dprintf_dd(dd, "%s\n", "");
 	spa_close(dd->dd_pool->dp_spa, tag);
@@ -250,6 +247,7 @@ static int
 getcomponent(const char *path, char *component, const char **nextp)
 {
 	char *p;
+
 	if ((path == NULL) || (path[0] == '\0'))
 		return (ENOENT);
 	/* This would be a good place to reserve some namespace... */
@@ -272,10 +270,10 @@ getcomponent(const char *path, char *component, const char **nextp)
 		(void) strcpy(component, path);
 		p = NULL;
 	} else if (p[0] == '/') {
-		if (p-path >= MAXNAMELEN)
+		if (p - path >= MAXNAMELEN)
 			return (ENAMETOOLONG);
 		(void) strncpy(component, path, p - path);
-		component[p-path] = '\0';
+		component[p - path] = '\0';
 		p++;
 	} else if (p[0] == '@') {
 		/*
@@ -284,66 +282,57 @@ getcomponent(const char *path, char *component, const char **nextp)
 		 */
 		if (strchr(path, '/'))
 			return (EINVAL);
-		if (p-path >= MAXNAMELEN)
+		if (p - path >= MAXNAMELEN)
 			return (ENAMETOOLONG);
 		(void) strncpy(component, path, p - path);
-		component[p-path] = '\0';
+		component[p - path] = '\0';
 	} else {
-		ASSERT(!"invalid p");
+		panic("invalid p=%p", (void *)p);
 	}
 	*nextp = p;
 	return (0);
 }
 
 /*
- * same as dsl_dir_open, ignore the first component of name and use the
+ * Return the dsl_dir_t, and possibly the last component which couldn't
- * spa instead
+ * be found in *tail.  The name must be in the specified dsl_pool_t.  This
+ * thread must hold the dp_config_rwlock for the pool.  Returns NULL if the
+ * path is bogus, or if tail==NULL and we couldn't parse the whole name.
+ * (*tail)[0] == '@' means that the last component is a snapshot.
  */
 int
-dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
+dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
     dsl_dir_t **ddp, const char **tailp)
 {
 	char *buf;
-	const char *next, *nextnext = NULL;
+	const char *spaname, *next, *nextnext = NULL;
 	int err;
 	dsl_dir_t *dd;
-	dsl_pool_t *dp;
 	uint64_t ddobj;
-	int openedspa = FALSE;
-
-	dprintf("%s\n", name);
 
 	buf = kmem_alloc(MAXNAMELEN, KM_PUSHPAGE);
 	err = getcomponent(name, buf, &next);
-	if (err)
+	if (err != 0)
 		goto error;
-	if (spa == NULL) {
-		err = spa_open(buf, &spa, FTAG);
-		if (err) {
-			dprintf("spa_open(%s) failed\n", buf);
-			goto error;
-		}
-		openedspa = TRUE;
 
-		/* XXX this assertion belongs in spa_open */
+	/* Make sure the name is in the specified pool. */
-		ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
+	spaname = spa_name(dp->dp_spa);
+	if (strcmp(buf, spaname) != 0) {
+		err = EINVAL;
+		goto error;
 	}
 
-	dp = spa_get_dsl(spa);
+	ASSERT(dsl_pool_config_held(dp));
 
-	rw_enter(&dp->dp_config_rwlock, RW_READER);
+	err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
-	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
+	if (err != 0) {
-	if (err) {
-		rw_exit(&dp->dp_config_rwlock);
-		if (openedspa)
-			spa_close(spa, FTAG);
 		goto error;
 	}
 
 	while (next != NULL) {
 		dsl_dir_t *child_ds;
 		err = getcomponent(next, buf, &nextnext);
-		if (err)
+		if (err != 0)
 			break;
 		ASSERT(next[0] != '\0');
 		if (next[0] == '@')
@@ -354,25 +343,22 @@ dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
 		err = zap_lookup(dp->dp_meta_objset,
 		    dd->dd_phys->dd_child_dir_zapobj,
 		    buf, sizeof (ddobj), 1, &ddobj);
-		if (err) {
+		if (err != 0) {
 			if (err == ENOENT)
 				err = 0;
 			break;
 		}
 
-		err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
+		err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
-		if (err)
+		if (err != 0)
 			break;
-		dsl_dir_close(dd, tag);
+		dsl_dir_rele(dd, tag);
 		dd = child_ds;
 		next = nextnext;
 	}
-	rw_exit(&dp->dp_config_rwlock);
 
-	if (err) {
+	if (err != 0) {
-		dsl_dir_close(dd, tag);
+		dsl_dir_rele(dd, tag);
-		if (openedspa)
-			spa_close(spa, FTAG);
 		goto error;
 	}
 
@@ -383,32 +369,18 @@ dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
 	if (next != NULL &&
 	    (tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
 		/* bad path name */
-		dsl_dir_close(dd, tag);
+		dsl_dir_rele(dd, tag);
 		dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
 		err = ENOENT;
 	}
-	if (tailp)
+	if (tailp != NULL)
 		*tailp = next;
-	if (openedspa)
-		spa_close(spa, FTAG);
 	*ddp = dd;
 error:
 	kmem_free(buf, MAXNAMELEN);
 	return (err);
 }
 
-/*
- * Return the dsl_dir_t, and possibly the last component which couldn't
- * be found in *tail.  Return NULL if the path is bogus, or if
- * tail==NULL and we couldn't parse the whole name.  (*tail)[0] == '@'
- * means that the last component is a snapshot.
- */
-int
-dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp)
-{
-	return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp));
-}
-
 uint64_t
 dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
     dmu_tx_t *tx)
@@ -446,71 +418,6 @@ dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name,
 	return (ddobj);
 }
 
-/* ARGSUSED */
-int
-dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
-{
-	dsl_dir_t *dd = arg1;
-	dsl_pool_t *dp = dd->dd_pool;
-	objset_t *mos = dp->dp_meta_objset;
-	int err;
-	uint64_t count;
-
-	/*
-	 * There should be exactly two holds, both from
-	 * dsl_dataset_destroy: one on the dd directory, and one on its
-	 * head ds.  If there are more holds, then a concurrent thread is
-	 * performing a lookup inside this dir while we're trying to destroy
-	 * it.  To minimize this possibility, we perform this check only
-	 * in syncing context and fail the operation if we encounter
-	 * additional holds.  The dp_config_rwlock ensures that nobody else
-	 * opens it after we check.
-	 */
-	if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 2)
-		return (EBUSY);
-
-	err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
-	if (err)
-		return (err);
-	if (count != 0)
-		return (EEXIST);
-
-	return (0);
-}
-
-void
-dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
-{
-	dsl_dir_t *dd = arg1;
-	objset_t *mos = dd->dd_pool->dp_meta_objset;
-	uint64_t obj;
-	dd_used_t t;
-
-	ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock));
-	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
-
-	/*
-	 * Remove our reservation. The impl() routine avoids setting the
-	 * actual property, which would require the (already destroyed) ds.
-	 */
-	dsl_dir_set_reservation_sync_impl(dd, 0, tx);
-
-	ASSERT0(dd->dd_phys->dd_used_bytes);
-	ASSERT0(dd->dd_phys->dd_reserved);
-	for (t = 0; t < DD_USED_NUM; t++)
-		ASSERT0(dd->dd_phys->dd_used_breakdown[t]);
-
-	VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx));
-	VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx));
-	VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx));
-	VERIFY(0 == zap_remove(mos,
-	    dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx));
-
-	obj = dd->dd_object;
-	dsl_dir_close(dd, tag);
-	VERIFY(0 == dmu_object_free(mos, obj, tx));
-}
-
 boolean_t
 dsl_dir_is_clone(dsl_dir_t *dd)
 {
@@ -546,18 +453,16 @@ dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv)
 	}
 	mutex_exit(&dd->dd_lock);
 
-	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
 	if (dsl_dir_is_clone(dd)) {
 		dsl_dataset_t *ds;
 		char buf[MAXNAMELEN];
 
-		VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
+		VERIFY0(dsl_dataset_hold_obj(dd->dd_pool,
 		    dd->dd_phys->dd_origin_obj, FTAG, &ds));
 		dsl_dataset_name(ds, buf);
 		dsl_dataset_rele(ds, FTAG);
 		dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf);
 	}
-	rw_exit(&dd->dd_pool->dp_config_rwlock);
 }
 
 void
@@ -567,7 +472,7 @@ dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx)
 
 	ASSERT(dd->dd_phys);
 
-	if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg) == 0) {
+	if (txg_list_add(&dp->dp_dirty_dirs, dd, tx->tx_txg)) {
 		/* up the hold count until we can be written out */
 		dmu_buf_add_ref(dd->dd_dbuf, dd);
 	}
@@ -854,7 +759,7 @@ dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 		    FALSE, asize > usize, tr_list, tx, TRUE);
 	}
 
-	if (err)
+	if (err != 0)
 		dsl_dir_tempreserve_clear(tr_list, tx);
 	else
 		*tr_cookiep = tr_list;
@@ -1007,118 +912,123 @@ dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta,
 		mutex_exit(&dd->dd_lock);
 }
 
+typedef struct dsl_dir_set_qr_arg {
+	const char *ddsqra_name;
+	zprop_source_t ddsqra_source;
+	uint64_t ddsqra_value;
+} dsl_dir_set_qr_arg_t;
+
 static int
-dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_dir_set_quota_check(void *arg, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	dsl_dir_set_qr_arg_t *ddsqra = arg;
-	dsl_dir_t *dd = ds->ds_dir;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
-	dsl_prop_setarg_t *psa = arg2;
+	dsl_dataset_t *ds;
-	int err;
+	int error;
-	uint64_t towrite;
+	uint64_t towrite, newval;
 
-	if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
+	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
-		return (err);
+	if (error != 0)
+		return (error);
 
-	if (psa->psa_effective_value == 0)
+	error = dsl_prop_predict(ds->ds_dir, "quota",
+	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
+	if (error != 0) {
+		dsl_dataset_rele(ds, FTAG);
+		return (error);
+	}
+
+	if (newval == 0) {
+		dsl_dataset_rele(ds, FTAG);
 		return (0);
+	}
 
-	mutex_enter(&dd->dd_lock);
+	mutex_enter(&ds->ds_dir->dd_lock);
 	/*
 	 * If we are doing the preliminary check in open context, and
 	 * there are pending changes, then don't fail it, since the
 	 * pending changes could under-estimate the amount of space to be
 	 * freed up.
 	 */
-	towrite = dsl_dir_space_towrite(dd);
+	towrite = dsl_dir_space_towrite(ds->ds_dir);
 	if ((dmu_tx_is_syncing(tx) || towrite == 0) &&
-	    (psa->psa_effective_value < dd->dd_phys->dd_reserved ||
+	    (newval < ds->ds_dir->dd_phys->dd_reserved ||
-	    psa->psa_effective_value < dd->dd_phys->dd_used_bytes + towrite)) {
+	    newval < ds->ds_dir->dd_phys->dd_used_bytes + towrite)) {
-		err = ENOSPC;
+		error = ENOSPC;
 	}
-	mutex_exit(&dd->dd_lock);
+	mutex_exit(&ds->ds_dir->dd_lock);
-	return (err);
+	dsl_dataset_rele(ds, FTAG);
+	return (error);
 }
 
-extern dsl_syncfunc_t dsl_prop_set_sync;
-
 static void
-dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_dir_set_quota_sync(void *arg, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	dsl_dir_set_qr_arg_t *ddsqra = arg;
-	dsl_dir_t *dd = ds->ds_dir;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
-	dsl_prop_setarg_t *psa = arg2;
+	dsl_dataset_t *ds;
-	uint64_t effective_value = psa->psa_effective_value;
+	uint64_t newval;
 
-	dsl_prop_set_sync(ds, psa, tx);
+	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
-	DSL_PROP_CHECK_PREDICTION(dd, psa);
 
-	dmu_buf_will_dirty(dd->dd_dbuf, tx);
+	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_QUOTA),
+	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
+	    &ddsqra->ddsqra_value, tx);
 
-	mutex_enter(&dd->dd_lock);
+	VERIFY0(dsl_prop_get_int_ds(ds,
-	dd->dd_phys->dd_quota = effective_value;
+	    zfs_prop_to_name(ZFS_PROP_QUOTA), &newval));
-	mutex_exit(&dd->dd_lock);
 
-	spa_history_log_internal_dd(dd, "set quota", tx,
+	dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
-	    "quota=%lld", (longlong_t)effective_value);
+	mutex_enter(&ds->ds_dir->dd_lock);
+	ds->ds_dir->dd_phys->dd_quota = newval;
+	mutex_exit(&ds->ds_dir->dd_lock);
+	dsl_dataset_rele(ds, FTAG);
 }
 
 int
 dsl_dir_set_quota(const char *ddname, zprop_source_t source, uint64_t quota)
 {
-	dsl_dir_t *dd;
+	dsl_dir_set_qr_arg_t ddsqra;
-	dsl_dataset_t *ds;
-	dsl_prop_setarg_t psa;
-	int err;
 
-	dsl_prop_setarg_init_uint64(&psa, "quota", source, &quota);
+	ddsqra.ddsqra_name = ddname;
+	ddsqra.ddsqra_source = source;
+	ddsqra.ddsqra_value = quota;
 
-	err = dsl_dataset_hold(ddname, FTAG, &ds);
+	return (dsl_sync_task(ddname, dsl_dir_set_quota_check,
-	if (err)
+	    dsl_dir_set_quota_sync, &ddsqra, 0));
-		return (err);
-
-	err = dsl_dir_open(ddname, FTAG, &dd, NULL);
-	if (err) {
-		dsl_dataset_rele(ds, FTAG);
-		return (err);
-	}
-
-	ASSERT(ds->ds_dir == dd);
-
-	/*
-	 * If someone removes a file, then tries to set the quota, we want to
-	 * make sure the file freeing takes effect.
-	 */
-	txg_wait_open(dd->dd_pool, 0);
-
-	err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check,
-	    dsl_dir_set_quota_sync, ds, &psa, 0);
-
-	dsl_dir_close(dd, FTAG);
-	dsl_dataset_rele(ds, FTAG);
-	return (err);
 }
 
 int
-dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_dir_set_reservation_check(void *arg, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	dsl_dir_set_qr_arg_t *ddsqra = arg;
-	dsl_dir_t *dd = ds->ds_dir;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
-	dsl_prop_setarg_t *psa = arg2;
+	dsl_dataset_t *ds;
-	uint64_t effective_value;
+	dsl_dir_t *dd;
-	uint64_t used, avail;
+	uint64_t newval, used, avail;
-	int err;
+	int error;
 
-	if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
+	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
-		return (err);
+	if (error != 0)
-
+		return (error);
-	effective_value = psa->psa_effective_value;
+	dd = ds->ds_dir;
 
 	/*
 	 * If we are doing the preliminary check in open context, the
 	 * space estimates may be inaccurate.
 	 */
-	if (!dmu_tx_is_syncing(tx))
+	if (!dmu_tx_is_syncing(tx)) {
+		dsl_dataset_rele(ds, FTAG);
 		return (0);
+	}
+
+	error = dsl_prop_predict(ds->ds_dir,
+	    zfs_prop_to_name(ZFS_PROP_RESERVATION),
+	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
+	if (error != 0) {
+		dsl_dataset_rele(ds, FTAG);
+		return (error);
+	}
 
 	mutex_enter(&dd->dd_lock);
 	used = dd->dd_phys->dd_used_bytes;
@@ -1131,21 +1041,21 @@ dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
 		avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used;
 	}
 
-	if (MAX(used, effective_value) > MAX(used, dd->dd_phys->dd_reserved)) {
+	if (MAX(used, newval) > MAX(used, dd->dd_phys->dd_reserved)) {
-		uint64_t delta = MAX(used, effective_value) -
+		uint64_t delta = MAX(used, newval) -
 		    MAX(used, dd->dd_phys->dd_reserved);
 
-		if (delta > avail)
+		if (delta > avail ||
-			return (ENOSPC);
+		    (dd->dd_phys->dd_quota > 0 &&
-		if (dd->dd_phys->dd_quota > 0 &&
+		    newval > dd->dd_phys->dd_quota))
-		    effective_value > dd->dd_phys->dd_quota)
+			error = ENOSPC;
-			return (ENOSPC);
 	}
 
-	return (0);
+	dsl_dataset_rele(ds, FTAG);
+	return (error);
 }
 
-static void
+void
 dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
 {
 	uint64_t used;
@@ -1167,51 +1077,38 @@ dsl_dir_set_reservation_sync_impl(dsl_dir_t *dd, uint64_t value, dmu_tx_t *tx)
 }
 
 static void
-dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_dir_set_reservation_sync(void *arg, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	dsl_dir_set_qr_arg_t *ddsqra = arg;
-	dsl_dir_t *dd = ds->ds_dir;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
-	dsl_prop_setarg_t *psa = arg2;
+	dsl_dataset_t *ds;
-	uint64_t value = psa->psa_effective_value;
+	uint64_t newval;
 
-	dsl_prop_set_sync(ds, psa, tx);
+	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
-	DSL_PROP_CHECK_PREDICTION(dd, psa);
 
-	dsl_dir_set_reservation_sync_impl(dd, value, tx);
+	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_RESERVATION),
+	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
+	    &ddsqra->ddsqra_value, tx);
 
-	spa_history_log_internal_dd(dd, "set reservation", tx,
+	VERIFY0(dsl_prop_get_int_ds(ds,
-	    "reservation=%lld", (longlong_t)value);
+	    zfs_prop_to_name(ZFS_PROP_RESERVATION), &newval));
+ 
+	dsl_dir_set_reservation_sync_impl(ds->ds_dir, newval, tx);
+	dsl_dataset_rele(ds, FTAG);
  }
 
 int
 dsl_dir_set_reservation(const char *ddname, zprop_source_t source,
     uint64_t reservation)
 {
-	dsl_dir_t *dd;
+	dsl_dir_set_qr_arg_t ddsqra;
-	dsl_dataset_t *ds;
-	dsl_prop_setarg_t psa;
-	int err;
 
-	dsl_prop_setarg_init_uint64(&psa, "reservation", source, &reservation);
+	ddsqra.ddsqra_name = ddname;
+	ddsqra.ddsqra_source = source;
+	ddsqra.ddsqra_value = reservation;
 
-	err = dsl_dataset_hold(ddname, FTAG, &ds);
+	return (dsl_sync_task(ddname, dsl_dir_set_reservation_check,
-	if (err)
+	    dsl_dir_set_reservation_sync, &ddsqra, 0));
-		return (err);
-
-	err = dsl_dir_open(ddname, FTAG, &dd, NULL);
-	if (err) {
-		dsl_dataset_rele(ds, FTAG);
-		return (err);
-	}
-
-	ASSERT(ds->ds_dir == dd);
-
-	err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_reservation_check,
-	    dsl_dir_set_reservation_sync, ds, &psa, 0);
-
-	dsl_dir_close(dd, FTAG);
-	dsl_dataset_rele(ds, FTAG);
-	return (err);
 }
 
 static dsl_dir_t *
@@ -1243,79 +1140,123 @@ would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor)
 	return (would_change(dd->dd_parent, delta, ancestor));
 }
 
-struct renamearg {
+typedef struct dsl_dir_rename_arg {
-	dsl_dir_t *newparent;
+	const char *ddra_oldname;
-	const char *mynewname;
+	const char *ddra_newname;
-};
+} dsl_dir_rename_arg_t;
+
+/* ARGSUSED */
+static int
+dsl_valid_rename(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
+{
+	int *deltap = arg;
+	char namebuf[MAXNAMELEN];
+
+	dsl_dataset_name(ds, namebuf);
+
+	if (strlen(namebuf) + *deltap >= MAXNAMELEN)
+		return (ENAMETOOLONG);
+	return (0);
+}
 
 static int
-dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_dir_rename_check(void *arg, dmu_tx_t *tx)
 {
-	dsl_dir_t *dd = arg1;
+	dsl_dir_rename_arg_t *ddra = arg;
-	struct renamearg *ra = arg2;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
-	dsl_pool_t *dp = dd->dd_pool;
+	dsl_dir_t *dd, *newparent;
-	objset_t *mos = dp->dp_meta_objset;
+	const char *mynewname;
-	int err;
+	int error;
-	uint64_t val;
+	int delta = strlen(ddra->ddra_newname) - strlen(ddra->ddra_oldname);
 
-	/*
+	/* target dir should exist */
-	 * There should only be one reference, from dmu_objset_rename().
+	error = dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL);
-	 * Fleeting holds are also possible (eg, from "zfs list" getting
+	if (error != 0)
-	 * stats), but any that are present in open context will likely
+		return (error);
-	 * be gone by syncing context, so only fail from syncing
-	 * context.
-	 */
-	if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 1)
-		return (EBUSY);
 
-	/* check for existing name */
+	/* new parent should exist */
-	err = zap_lookup(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
+	error = dsl_dir_hold(dp, ddra->ddra_newname, FTAG,
-	    ra->mynewname, 8, 1, &val);
+	    &newparent, &mynewname);
-	if (err == 0)
+	if (error != 0) {
+		dsl_dir_rele(dd, FTAG);
+		return (error);
+	}
+
+	/* can't rename to different pool */
+	if (dd->dd_pool != newparent->dd_pool) {
+		dsl_dir_rele(newparent, FTAG);
+		dsl_dir_rele(dd, FTAG);
+		return (ENXIO);
+	}
+
+	/* new name should not already exist */
+	if (mynewname == NULL) {
+		dsl_dir_rele(newparent, FTAG);
+		dsl_dir_rele(dd, FTAG);
 		return (EEXIST);
-	if (err != ENOENT)
+	}
-		return (err);
 
-	if (ra->newparent != dd->dd_parent) {
+	/* if the name length is growing, validate child name lengths */
+	if (delta > 0) {
+		error = dmu_objset_find_dp(dp, dd->dd_object, dsl_valid_rename,
+		    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
+		if (error != 0) {
+			dsl_dir_rele(newparent, FTAG);
+			dsl_dir_rele(dd, FTAG);
+			return (error);
+		}
+	}
+
+	if (newparent != dd->dd_parent) {
 		/* is there enough space? */
 		uint64_t myspace =
 		    MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved);
 
 		/* no rename into our descendant */
-		if (closest_common_ancestor(dd, ra->newparent) == dd)
+		if (closest_common_ancestor(dd, newparent) == dd) {
+			dsl_dir_rele(newparent, FTAG);
+			dsl_dir_rele(dd, FTAG);
 			return (EINVAL);
+		}
 
-		if ((err = dsl_dir_transfer_possible(dd->dd_parent,
+		error = dsl_dir_transfer_possible(dd->dd_parent,
-		    ra->newparent, myspace)))
+		    newparent, myspace);
-			return (err);
+		if (error != 0) {
+			dsl_dir_rele(newparent, FTAG);
+			dsl_dir_rele(dd, FTAG);
+			return (error);
+		}
 	}
 
+	dsl_dir_rele(newparent, FTAG);
+	dsl_dir_rele(dd, FTAG);
 	return (0);
 }
 
 static void
-dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_dir_rename_sync(void *arg, dmu_tx_t *tx)
 {
-	dsl_dir_t *dd = arg1;
+	dsl_dir_rename_arg_t *ddra = arg;
-	struct renamearg *ra = arg2;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
-	dsl_pool_t *dp = dd->dd_pool;
+	dsl_dir_t *dd, *newparent;
+	const char *mynewname;
+	int error;
 	objset_t *mos = dp->dp_meta_objset;
-	int err;
-	char namebuf[MAXNAMELEN];
 
-	ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2);
+	VERIFY0(dsl_dir_hold(dp, ddra->ddra_oldname, FTAG, &dd, NULL));
+	VERIFY0(dsl_dir_hold(dp, ddra->ddra_newname, FTAG, &newparent,
+	    &mynewname));
 
 	/* Log this before we change the name. */
-	dsl_dir_name(ra->newparent, namebuf);
 	spa_history_log_internal_dd(dd, "rename", tx,
-	    "-> %s/%s", namebuf, ra->mynewname);
+	    "-> %s", ddra->ddra_newname);
 
-	if (ra->newparent != dd->dd_parent) {
+	if (newparent != dd->dd_parent) {
 		dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD,
 		    -dd->dd_phys->dd_used_bytes,
 		    -dd->dd_phys->dd_compressed_bytes,
 		    -dd->dd_phys->dd_uncompressed_bytes, tx);
-		dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD,
+		dsl_dir_diduse_space(newparent, DD_USED_CHILD,
 		    dd->dd_phys->dd_used_bytes,
 		    dd->dd_phys->dd_compressed_bytes,
 		    dd->dd_phys->dd_uncompressed_bytes, tx);
@@ -1326,7 +1267,7 @@ dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 
 			dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV,
 			    -unused_rsrv, 0, 0, tx);
-			dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV,
+			dsl_dir_diduse_space(newparent, DD_USED_CHILD_RSRV,
 			    unused_rsrv, 0, 0, tx);
 		}
 	}
@@ -1334,52 +1275,36 @@ dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
 
 	/* remove from old parent zapobj */
-	err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
+	error = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj,
 	    dd->dd_myname, tx);
-	ASSERT0(err);
+	ASSERT0(error);
 
-	(void) strcpy(dd->dd_myname, ra->mynewname);
+	(void) strcpy(dd->dd_myname, mynewname);
-	dsl_dir_close(dd->dd_parent, dd);
+	dsl_dir_rele(dd->dd_parent, dd);
-	dd->dd_phys->dd_parent_obj = ra->newparent->dd_object;
+	dd->dd_phys->dd_parent_obj = newparent->dd_object;
-	VERIFY(0 == dsl_dir_open_obj(dd->dd_pool,
+	VERIFY0(dsl_dir_hold_obj(dp,
-	    ra->newparent->dd_object, NULL, dd, &dd->dd_parent));
+	    newparent->dd_object, NULL, dd, &dd->dd_parent));
 
 	/* add to new parent zapobj */
-	err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj,
+	VERIFY0(zap_add(mos, newparent->dd_phys->dd_child_dir_zapobj,
-	    dd->dd_myname, 8, 1, &dd->dd_object, tx);
+	    dd->dd_myname, 8, 1, &dd->dd_object, tx));
-	ASSERT0(err);
 
+	dsl_prop_notify_all(dd);
+
+	dsl_dir_rele(newparent, FTAG);
+	dsl_dir_rele(dd, FTAG);
 }
 
 int
-dsl_dir_rename(dsl_dir_t *dd, const char *newname)
+dsl_dir_rename(const char *oldname, const char *newname)
 {
-	struct renamearg ra;
+	dsl_dir_rename_arg_t ddra;
-	int err;
 
-	/* new parent should exist */
+	ddra.ddra_oldname = oldname;
-	err = dsl_dir_open(newname, FTAG, &ra.newparent, &ra.mynewname);
+	ddra.ddra_newname = newname;
-	if (err)
-		return (err);
 
-	/* can't rename to different pool */
+	return (dsl_sync_task(oldname,
-	if (dd->dd_pool != ra.newparent->dd_pool) {
+	    dsl_dir_rename_check, dsl_dir_rename_sync, &ddra, 3));
-		err = ENXIO;
-		goto out;
-	}
-
-	/* new name should not already exist */
-	if (ra.mynewname == NULL) {
-		err = EEXIST;
-		goto out;
-	}
-
-	err = dsl_sync_task_do(dd->dd_pool,
-	    dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3);
-
-out:
-	dsl_dir_close(ra.newparent, FTAG);
-	return (err);
 }
 
 int
@@ -1424,6 +1349,4 @@ dsl_dir_snap_cmtime_update(dsl_dir_t *dd)
 #if defined(_KERNEL) && defined(HAVE_SPL)
 EXPORT_SYMBOL(dsl_dir_set_quota);
 EXPORT_SYMBOL(dsl_dir_set_reservation);
-EXPORT_SYMBOL(dsl_dir_open);
-EXPORT_SYMBOL(dsl_dir_close);
 #endif

module/zfs/dsl_pool.c

@@ -43,6 +43,7 @@
 #include <sys/bptree.h>
 #include <sys/zfeature.h>
 #include <sys/zil_impl.h>
+#include <sys/dsl_userhold.h>
 
 int zfs_no_write_throttle = 0;
 int zfs_write_limit_shift = 3;			/* 1/8th of physical memory */
@@ -264,7 +265,7 @@ dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
 	if (err)
 		return (err);
 
-	return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
+	return (dsl_dir_hold_obj(dp, obj, name, dp, ddp));
 }
 
 static dsl_pool_t *
@@ -276,7 +277,7 @@ dsl_pool_open_impl(spa_t *spa, uint64_t txg)
 	dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
 	dp->dp_spa = spa;
 	dp->dp_meta_rootbp = *bp;
-	rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
+	rrw_init(&dp->dp_config_rwlock, B_TRUE);
 	dp->dp_write_limit = zfs_write_limit_min;
 	txg_init(dp, txg);
 
@@ -287,7 +288,7 @@ dsl_pool_open_impl(spa_t *spa, uint64_t txg)
 	txg_list_create(&dp->dp_dirty_dirs,
 	    offsetof(dsl_dir_t, dd_dirty_link));
 	txg_list_create(&dp->dp_sync_tasks,
-	    offsetof(dsl_sync_task_group_t, dstg_node));
+	    offsetof(dsl_sync_task_t, dst_node));
 
 	mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
 
@@ -324,14 +325,14 @@ dsl_pool_open(dsl_pool_t *dp)
 	dsl_dataset_t *ds;
 	uint64_t obj;
 
-	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
+	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
 	    &dp->dp_root_dir_obj);
 	if (err)
 		goto out;
 
-	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
+	err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
 	    NULL, dp, &dp->dp_root_dir);
 	if (err)
 		goto out;
@@ -352,7 +353,7 @@ dsl_pool_open(dsl_pool_t *dp)
 			    &dp->dp_origin_snap);
 			dsl_dataset_rele(ds, FTAG);
 		}
-		dsl_dir_close(dd, dp);
+		dsl_dir_rele(dd, dp);
 		if (err)
 			goto out;
 	}
@@ -367,7 +368,7 @@ dsl_pool_open(dsl_pool_t *dp)
 		    DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
 		if (err)
 			goto out;
-		VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
+		VERIFY0(bpobj_open(&dp->dp_free_bpobj,
 		    dp->dp_meta_objset, obj));
 	}
 
@@ -400,7 +401,7 @@ dsl_pool_open(dsl_pool_t *dp)
 	err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg);
 
 out:
-	rw_exit(&dp->dp_config_rwlock);
+	rrw_exit(&dp->dp_config_rwlock, FTAG);
 	return (err);
 }
 
@@ -415,13 +416,13 @@ dsl_pool_close(dsl_pool_t *dp)
 	 * and not a hold, so just drop that here.
 	 */
 	if (dp->dp_origin_snap)
-		dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
+		dsl_dataset_rele(dp->dp_origin_snap, dp);
 	if (dp->dp_mos_dir)
-		dsl_dir_close(dp->dp_mos_dir, dp);
+		dsl_dir_rele(dp->dp_mos_dir, dp);
 	if (dp->dp_free_dir)
-		dsl_dir_close(dp->dp_free_dir, dp);
+		dsl_dir_rele(dp->dp_free_dir, dp);
 	if (dp->dp_root_dir)
-		dsl_dir_close(dp->dp_root_dir, dp);
+		dsl_dir_rele(dp->dp_root_dir, dp);
 
 	bpobj_close(&dp->dp_free_bpobj);
 
@@ -439,7 +440,7 @@ dsl_pool_close(dsl_pool_t *dp)
 	dsl_scan_fini(dp);
 	dsl_pool_tx_assign_destroy(dp);
 	dsl_pool_txg_history_destroy(dp);
-	rw_destroy(&dp->dp_config_rwlock);
+	rrw_destroy(&dp->dp_config_rwlock);
 	mutex_destroy(&dp->dp_lock);
 	taskq_destroy(dp->dp_iput_taskq);
 	if (dp->dp_blkstats)
@@ -457,6 +458,8 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
 	dsl_dataset_t *ds;
 	uint64_t obj;
 
+	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
+
 	/* create and open the MOS (meta-objset) */
 	dp->dp_meta_objset = dmu_objset_create_impl(spa,
 	    NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
@@ -467,30 +470,30 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
 	ASSERT0(err);
 
 	/* Initialize scan structures */
-	VERIFY3U(0, ==, dsl_scan_init(dp, txg));
+	VERIFY0(dsl_scan_init(dp, txg));
 
 	/* create and open the root dir */
 	dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
-	VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
+	VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
 	    NULL, dp, &dp->dp_root_dir));
 
 	/* create and open the meta-objset dir */
 	(void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
-	VERIFY(0 == dsl_pool_open_special_dir(dp,
+	VERIFY0(dsl_pool_open_special_dir(dp,
 	    MOS_DIR_NAME, &dp->dp_mos_dir));
 
 	if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
 		/* create and open the free dir */
 		(void) dsl_dir_create_sync(dp, dp->dp_root_dir,
 		    FREE_DIR_NAME, tx);
-		VERIFY(0 == dsl_pool_open_special_dir(dp,
+		VERIFY0(dsl_pool_open_special_dir(dp,
 		    FREE_DIR_NAME, &dp->dp_free_dir));
 
 		/* create and open the free_bplist */
 		obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
 		VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
 		    DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
-		VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
+		VERIFY0(bpobj_open(&dp->dp_free_bpobj,
 		    dp->dp_meta_objset, obj));
 	}
 
@@ -501,7 +504,7 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
 	obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
 
 	/* create the root objset */
-	VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
+	VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
 	VERIFY(NULL != (os = dmu_objset_create_impl(dp->dp_spa, ds,
 	    dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx)));
 #ifdef _KERNEL
@@ -511,6 +514,8 @@ dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
 
 	dmu_tx_commit(tx);
 
+	rrw_exit(&dp->dp_config_rwlock, FTAG);
+
 	return (dp);
 }
 
@@ -533,10 +538,7 @@ static int
 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
 {
 	dsl_deadlist_t *dl = arg;
-	dsl_pool_t *dp = dmu_objset_pool(dl->dl_os);
-	rw_enter(&dp->dp_config_rwlock, RW_READER);
 	dsl_deadlist_insert(dl, bp, tx);
-	rw_exit(&dp->dp_config_rwlock);
 	return (0);
 }
 
@@ -558,7 +560,7 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
 
 	/*
 	 * We need to copy dp_space_towrite() before doing
-	 * dsl_sync_task_group_sync(), because
+	 * dsl_sync_task_sync(), because
 	 * dsl_dataset_snapshot_reserve_space() will increase
 	 * dp_space_towrite but not actually write anything.
 	 */
@@ -673,14 +675,14 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
 	 */
 	DTRACE_PROBE(pool_sync__3task);
 	if (!txg_list_empty(&dp->dp_sync_tasks, txg)) {
-		dsl_sync_task_group_t *dstg;
+		dsl_sync_task_t *dst;
 		/*
 		 * No more sync tasks should have been added while we
 		 * were syncing.
 		 */
 		ASSERT(spa_sync_pass(dp->dp_spa) == 1);
-		while ((dstg = txg_list_remove(&dp->dp_sync_tasks, txg)))
+		while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)))
-			dsl_sync_task_group_sync(dstg, tx);
+			dsl_sync_task_sync(dst, tx);
 	}
 
 	dmu_tx_commit(tx);
@@ -857,14 +859,13 @@ dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
 
 /* ARGSUSED */
 static int
-upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
+upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
 {
 	dmu_tx_t *tx = arg;
 	dsl_dataset_t *ds, *prev = NULL;
 	int err;
-	dsl_pool_t *dp = spa_get_dsl(spa);
 
-	err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
+	err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
 	if (err)
 		return (err);
 
@@ -890,7 +891,7 @@ upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
 		 * The $ORIGIN can't have any data, or the accounting
 		 * will be wrong.
 		 */
-		ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
+		ASSERT0(prev->ds_phys->ds_bp.blk_birth);
 
 		/* The origin doesn't get attached to itself */
 		if (ds->ds_object == prev->ds_object) {
@@ -910,13 +911,13 @@ upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
 
 		if (ds->ds_phys->ds_next_snap_obj == 0) {
 			ASSERT(ds->ds_prev == NULL);
-			VERIFY(0 == dsl_dataset_hold_obj(dp,
+			VERIFY0(dsl_dataset_hold_obj(dp,
 			    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
 		}
 	}
 
-	ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
+	ASSERT3U(ds->ds_dir->dd_phys->dd_origin_obj, ==, prev->ds_object);
-	ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
+	ASSERT3U(ds->ds_phys->ds_prev_snap_obj, ==, prev->ds_object);
 
 	if (prev->ds_phys->ds_next_clones_obj == 0) {
 		dmu_buf_will_dirty(prev->ds_dbuf, tx);
@@ -924,7 +925,7 @@ upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
 		    zap_create(dp->dp_meta_objset,
 		    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
 	}
-	VERIFY(0 == zap_add_int(dp->dp_meta_objset,
+	VERIFY0(zap_add_int(dp->dp_meta_objset,
 	    prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
 
 	dsl_dataset_rele(ds, FTAG);
@@ -939,25 +940,21 @@ dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
 	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT(dp->dp_origin_snap != NULL);
 
-	VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
+	VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb,
 	    tx, DS_FIND_CHILDREN));
 }
 
 /* ARGSUSED */
 static int
-upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
+upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
 {
 	dmu_tx_t *tx = arg;
-	dsl_dataset_t *ds;
-	dsl_pool_t *dp = spa_get_dsl(spa);
 	objset_t *mos = dp->dp_meta_objset;
 
-	VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
+	if (ds->ds_dir->dd_phys->dd_origin_obj != 0) {
-
-	if (ds->ds_dir->dd_phys->dd_origin_obj) {
 		dsl_dataset_t *origin;
 
-		VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
+		VERIFY0(dsl_dataset_hold_obj(dp,
 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin));
 
 		if (origin->ds_dir->dd_phys->dd_clones == 0) {
@@ -966,13 +963,11 @@ upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
 		}
 
-		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
+		VERIFY0(zap_add_int(dp->dp_meta_objset,
-		    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
+		    origin->ds_dir->dd_phys->dd_clones, ds->ds_object, tx));
 
 		dsl_dataset_rele(origin, FTAG);
 	}
-
-	dsl_dataset_rele(ds, FTAG);
 	return (0);
 }
 
@@ -984,7 +979,7 @@ dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
 	ASSERT(dmu_tx_is_syncing(tx));
 
 	(void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
-	VERIFY(0 == dsl_pool_open_special_dir(dp,
+	VERIFY0(dsl_pool_open_special_dir(dp,
 	    FREE_DIR_NAME, &dp->dp_free_dir));
 
 	/*
@@ -994,12 +989,11 @@ dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
 	 */
 	obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
 	    SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
-	VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
+	VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
 	    DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
-	VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
+	VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj));
-	    dp->dp_meta_objset, obj));
 
-	VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL,
+	VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
 	    upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
 }
 
@@ -1011,17 +1005,16 @@ dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
 
 	ASSERT(dmu_tx_is_syncing(tx));
 	ASSERT(dp->dp_origin_snap == NULL);
+	ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER));
 
 	/* create the origin dir, ds, & snap-ds */
-	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
 	dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
 	    NULL, 0, kcred, tx);
-	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
+	VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
-	dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx);
+	dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx);
-	VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
+	VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
 	    dp, &dp->dp_origin_snap));
 	dsl_dataset_rele(ds, FTAG);
-	rw_exit(&dp->dp_config_rwlock);
 }
 
 taskq_t *
@@ -1056,7 +1049,7 @@ dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
 		*htag = '\0';
 		++htag;
 		dsobj = strtonum(za.za_name, NULL);
-		(void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE);
+		dsl_dataset_user_release_tmp(dp, dsobj, htag);
 	}
 	zap_cursor_fini(&zc);
 }
@@ -1078,7 +1071,7 @@ dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
 
 static int
 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
-    const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding)
+    const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding)
 {
 	objset_t *mos = dp->dp_meta_objset;
 	uint64_t zapobj = dp->dp_tmp_userrefs_obj;
@@ -1103,7 +1096,7 @@ dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
 
 	name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
 	if (holding)
-		error = zap_add(mos, zapobj, name, 8, 1, now, tx);
+		error = zap_add(mos, zapobj, name, 8, 1, &now, tx);
 	else
 		error = zap_remove(mos, zapobj, name, tx);
 	strfree(name);
@@ -1116,7 +1109,7 @@ dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
  */
 int
 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
-    uint64_t *now, dmu_tx_t *tx)
+    uint64_t now, dmu_tx_t *tx)
 {
 	return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
 }
@@ -1128,10 +1121,113 @@ int
 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
     dmu_tx_t *tx)
 {
-	return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,
+	return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0,
 	    tx, B_FALSE));
 }
 
+/*
+ * DSL Pool Configuration Lock
+ *
+ * The dp_config_rwlock protects against changes to DSL state (e.g. dataset
+ * creation / destruction / rename / property setting).  It must be held for
+ * read to hold a dataset or dsl_dir.  I.e. you must call
+ * dsl_pool_config_enter() or dsl_pool_hold() before calling
+ * dsl_{dataset,dir}_hold{_obj}.  In most circumstances, the dp_config_rwlock
+ * must be held continuously until all datasets and dsl_dirs are released.
+ *
+ * The only exception to this rule is that if a "long hold" is placed on
+ * a dataset, then the dp_config_rwlock may be dropped while the dataset
+ * is still held.  The long hold will prevent the dataset from being
+ * destroyed -- the destroy will fail with EBUSY.  A long hold can be
+ * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset
+ * (by calling dsl_{dataset,objset}_{try}own{_obj}).
+ *
+ * Legitimate long-holders (including owners) should be long-running, cancelable
+ * tasks that should cause "zfs destroy" to fail.  This includes DMU
+ * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open),
+ * "zfs send", and "zfs diff".  There are several other long-holders whose
+ * uses are suboptimal (e.g. "zfs promote", and zil_suspend()).
+ *
+ * The usual formula for long-holding would be:
+ * dsl_pool_hold()
+ * dsl_dataset_hold()
+ * ... perform checks ...
+ * dsl_dataset_long_hold()
+ * dsl_pool_rele()
+ * ... perform long-running task ...
+ * dsl_dataset_long_rele()
+ * dsl_dataset_rele()
+ *
+ * Note that when the long hold is released, the dataset is still held but
+ * the pool is not held.  The dataset may change arbitrarily during this time
+ * (e.g. it could be destroyed).  Therefore you shouldn't do anything to the
+ * dataset except release it.
+ *
+ * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only
+ * or modifying operations.
+ *
+ * Modifying operations should generally use dsl_sync_task().  The synctask
+ * infrastructure enforces proper locking strategy with respect to the
+ * dp_config_rwlock.  See the comment above dsl_sync_task() for details.
+ *
+ * Read-only operations will manually hold the pool, then the dataset, obtain
+ * information from the dataset, then release the pool and dataset.
+ * dmu_objset_{hold,rele}() are convenience routines that also do the pool
+ * hold/rele.
+ */
+
+int
+dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp)
+{
+	spa_t *spa;
+	int error;
+
+	error = spa_open(name, &spa, tag);
+	if (error == 0) {
+		*dp = spa_get_dsl(spa);
+		dsl_pool_config_enter(*dp, tag);
+	}
+	return (error);
+}
+
+void
+dsl_pool_rele(dsl_pool_t *dp, void *tag)
+{
+	dsl_pool_config_exit(dp, tag);
+	spa_close(dp->dp_spa, tag);
+}
+
+void
+dsl_pool_config_enter(dsl_pool_t *dp, void *tag)
+{
+	/*
+	 * We use a "reentrant" reader-writer lock, but not reentrantly.
+	 *
+	 * The rrwlock can (with the track_all flag) track all reading threads,
+	 * which is very useful for debugging which code path failed to release
+	 * the lock, and for verifying that the *current* thread does hold
+	 * the lock.
+	 *
+	 * (Unlike a rwlock, which knows that N threads hold it for
+	 * read, but not *which* threads, so rw_held(RW_READER) returns TRUE
+	 * if any thread holds it for read, even if this thread doesn't).
+	 */
+	ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER));
+	rrw_enter(&dp->dp_config_rwlock, RW_READER, tag);
+}
+
+void
+dsl_pool_config_exit(dsl_pool_t *dp, void *tag)
+{
+	rrw_exit(&dp->dp_config_rwlock, tag);
+}
+
+boolean_t
+dsl_pool_config_held(dsl_pool_t *dp)
+{
+	return (RRW_LOCK_HELD(&dp->dp_config_rwlock));
+}
+
 #if defined(_KERNEL) && defined(HAVE_SPL)
 module_param(zfs_no_write_throttle, int, 0644);
 MODULE_PARM_DESC(zfs_no_write_throttle, "Disable write throttling");

module/zfs/dsl_prop.c

@@ -82,7 +82,7 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname,
 	char *inheritstr;
 	char *recvdstr;
 
-	ASSERT(RW_LOCK_HELD(&dd->dd_pool->dp_config_rwlock));
+	ASSERT(dsl_pool_config_held(dd->dd_pool));
 
 	if (setpoint)
 		setpoint[0] = '\0';
@@ -97,8 +97,6 @@ dsl_prop_get_dd(dsl_dir_t *dd, const char *propname,
 	 * after this loop.
 	 */
 	for (; dd != NULL; dd = dd->dd_parent) {
-		ASSERT(RW_LOCK_HELD(&dd->dd_pool->dp_config_rwlock));
-
 		if (dd != target || snapshot) {
 			if (!inheritable)
 				break;
@@ -167,7 +165,7 @@ dsl_prop_get_ds(dsl_dataset_t *ds, const char *propname,
 	boolean_t snapshot;
 	uint64_t zapobj;
 
-	ASSERT(RW_LOCK_HELD(&ds->ds_dir->dd_pool->dp_config_rwlock));
+	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
 	inheritable = (prop == ZPROP_INVAL || zfs_prop_inheritable(prop));
 	snapshot = (ds->ds_phys != NULL && dsl_dataset_is_snapshot(ds));
 	zapobj = (ds->ds_phys == NULL ? 0 : ds->ds_phys->ds_props_obj);
@@ -231,22 +229,16 @@ dsl_prop_register(dsl_dataset_t *ds, const char *propname,
     dsl_prop_changed_cb_t *callback, void *cbarg)
 {
 	dsl_dir_t *dd = ds->ds_dir;
-	dsl_pool_t *dp = dd->dd_pool;
 	uint64_t value;
 	dsl_prop_cb_record_t *cbr;
 	int err;
-	int need_rwlock;
+	ASSERTV(dsl_pool_t *dp = dd->dd_pool);
 
-	need_rwlock = !RW_WRITE_HELD(&dp->dp_config_rwlock);
+	ASSERT(dsl_pool_config_held(dp));
-	if (need_rwlock)
-		rw_enter(&dp->dp_config_rwlock, RW_READER);
 
-	err = dsl_prop_get_ds(ds, propname, 8, 1, &value, NULL);
+	err = dsl_prop_get_int_ds(ds, propname, &value);
-	if (err != 0) {
+	if (err != 0)
-		if (need_rwlock)
-			rw_exit(&dp->dp_config_rwlock);
 		return (err);
-	}
 
 	cbr = kmem_alloc(sizeof (dsl_prop_cb_record_t), KM_PUSHPAGE);
 	cbr->cbr_ds = ds;
@@ -259,9 +251,6 @@ dsl_prop_register(dsl_dataset_t *ds, const char *propname,
 	mutex_exit(&dd->dd_lock);
 
 	cbr->cbr_func(cbr->cbr_arg, value);
-
-	if (need_rwlock)
-		rw_exit(&dp->dp_config_rwlock);
 	return (0);
 }
 
@@ -269,19 +258,18 @@ int
 dsl_prop_get(const char *dsname, const char *propname,
     int intsz, int numints, void *buf, char *setpoint)
 {
-	dsl_dataset_t *ds;
+	objset_t *os;
-	int err;
+	int error;
 
-	err = dsl_dataset_hold(dsname, FTAG, &ds);
+	error = dmu_objset_hold(dsname, FTAG, &os);
-	if (err)
+	if (error != 0)
-		return (err);
+		return (error);
 
-	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
+	error = dsl_prop_get_ds(dmu_objset_ds(os), propname,
-	err = dsl_prop_get_ds(ds, propname, intsz, numints, buf, setpoint);
+	    intsz, numints, buf, setpoint);
-	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
 
-	dsl_dataset_rele(ds, FTAG);
+	dmu_objset_rele(os, FTAG);
-	return (err);
+	return (error);
 }
 
 /*
@@ -299,17 +287,11 @@ dsl_prop_get_integer(const char *ddname, const char *propname,
 	return (dsl_prop_get(ddname, propname, 8, 1, valuep, setpoint));
 }
 
-void
+int
-dsl_prop_setarg_init_uint64(dsl_prop_setarg_t *psa, const char *propname,
+dsl_prop_get_int_ds(dsl_dataset_t *ds, const char *propname,
-    zprop_source_t source, uint64_t *value)
+    uint64_t *valuep)
 {
-	psa->psa_name = propname;
+	return (dsl_prop_get_ds(ds, propname, 8, 1, valuep, NULL));
-	psa->psa_source = source;
-	psa->psa_intsz = 8;
-	psa->psa_numints = 1;
-	psa->psa_value = value;
-
-	psa->psa_effective_value = -1ULL;
 }
 
 /*
@@ -323,11 +305,10 @@ dsl_prop_setarg_init_uint64(dsl_prop_setarg_t *psa, const char *propname,
  * a property not handled by this function.
  */
 int
-dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa)
+dsl_prop_predict(dsl_dir_t *dd, const char *propname,
+    zprop_source_t source, uint64_t value, uint64_t *newvalp)
 {
-	const char *propname = psa->psa_name;
 	zfs_prop_t prop = zfs_name_to_prop(propname);
-	zprop_source_t source = psa->psa_source;
 	objset_t *mos;
 	uint64_t zapobj;
 	uint64_t version;
@@ -359,36 +340,33 @@ dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa)
 	switch ((int)source) {
 	case ZPROP_SRC_NONE:
 		/* Revert to the received value, if any. */
-		err = zap_lookup(mos, zapobj, recvdstr, 8, 1,
+		err = zap_lookup(mos, zapobj, recvdstr, 8, 1, newvalp);
-		    &psa->psa_effective_value);
 		if (err == ENOENT)
-			psa->psa_effective_value = 0;
+			*newvalp = 0;
 		break;
 	case ZPROP_SRC_LOCAL:
-		psa->psa_effective_value = *(uint64_t *)psa->psa_value;
+		*newvalp = value;
 		break;
 	case ZPROP_SRC_RECEIVED:
 		/*
 		 * If there's no local setting, then the new received value will
 		 * be the effective value.
 		 */
-		err = zap_lookup(mos, zapobj, propname, 8, 1,
+		err = zap_lookup(mos, zapobj, propname, 8, 1, newvalp);
-		    &psa->psa_effective_value);
 		if (err == ENOENT)
-			psa->psa_effective_value = *(uint64_t *)psa->psa_value;
+			*newvalp = value;
 		break;
 	case (ZPROP_SRC_NONE | ZPROP_SRC_RECEIVED):
 		/*
 		 * We're clearing the received value, so the local setting (if
 		 * it exists) remains the effective value.
 		 */
-		err = zap_lookup(mos, zapobj, propname, 8, 1,
+		err = zap_lookup(mos, zapobj, propname, 8, 1, newvalp);
-		    &psa->psa_effective_value);
 		if (err == ENOENT)
-			psa->psa_effective_value = 0;
+			*newvalp = 0;
 		break;
 	default:
-		cmn_err(CE_PANIC, "unexpected property source: %d", source);
+		panic("unexpected property source: %d", source);
 	}
 
 	strfree(recvdstr);
@@ -399,39 +377,6 @@ dsl_prop_predict_sync(dsl_dir_t *dd, dsl_prop_setarg_t *psa)
 	return (err);
 }
 
-#ifdef	ZFS_DEBUG
-void
-dsl_prop_check_prediction(dsl_dir_t *dd, dsl_prop_setarg_t *psa)
-{
-	zfs_prop_t prop = zfs_name_to_prop(psa->psa_name);
-	uint64_t intval;
-	char setpoint[MAXNAMELEN];
-	uint64_t version = spa_version(dd->dd_pool->dp_spa);
-	int err;
-
-	if (version < SPA_VERSION_RECVD_PROPS) {
-		switch (prop) {
-		case ZFS_PROP_QUOTA:
-		case ZFS_PROP_RESERVATION:
-			return;
-		default:
-			break;
-		}
-	}
-
-	err = dsl_prop_get_dd(dd, psa->psa_name, 8, 1, &intval,
-	    setpoint, B_FALSE);
-	if (err == 0 && intval != psa->psa_effective_value) {
-		cmn_err(CE_PANIC, "%s property, source: %x, "
-		    "predicted effective value: %llu, "
-		    "actual effective value: %llu (setpoint: %s)",
-		    psa->psa_name, psa->psa_source,
-		    (unsigned long long)psa->psa_effective_value,
-		    (unsigned long long)intval, setpoint);
-	}
-}
-#endif
-
 /*
  * Unregister this callback.  Return 0 on success, ENOENT if ddname is
  * invalid, ENOMSG if no matching callback registered.
@@ -466,25 +411,57 @@ dsl_prop_unregister(dsl_dataset_t *ds, const char *propname,
 	return (0);
 }
 
-/*
+boolean_t
- * Return the number of callbacks that are registered for this dataset.
+dsl_prop_hascb(dsl_dataset_t *ds)
- */
+{
-int
+	dsl_dir_t *dd = ds->ds_dir;
-dsl_prop_numcb(dsl_dataset_t *ds)
+	boolean_t rv = B_FALSE;
+	dsl_prop_cb_record_t *cbr;
+
+	mutex_enter(&dd->dd_lock);
+	for (cbr = list_head(&dd->dd_prop_cbs); cbr;
+	    cbr = list_next(&dd->dd_prop_cbs, cbr)) {
+		if (cbr->cbr_ds == ds) {
+			rv = B_TRUE;
+			break;
+		}
+	}
+	mutex_exit(&dd->dd_lock);
+	return (rv);
+}
+
+/* ARGSUSED */
+static int
+dsl_prop_notify_all_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
 {
 	dsl_dir_t *dd = ds->ds_dir;
 	dsl_prop_cb_record_t *cbr;
-	int num = 0;
 
 	mutex_enter(&dd->dd_lock);
-	for (cbr = list_head(&dd->dd_prop_cbs);
+	for (cbr = list_head(&dd->dd_prop_cbs); cbr;
-	    cbr; cbr = list_next(&dd->dd_prop_cbs, cbr)) {
+	    cbr = list_next(&dd->dd_prop_cbs, cbr)) {
-		if (cbr->cbr_ds == ds)
+		uint64_t value;
-			num++;
+
+		if (dsl_prop_get_ds(cbr->cbr_ds, cbr->cbr_propname,
+		    sizeof (value), 1, &value, NULL) == 0)
+			cbr->cbr_func(cbr->cbr_arg, value);
 	}
 	mutex_exit(&dd->dd_lock);
 
-	return (num);
+	return (0);
+}
+
+/*
+ * Update all property values for ddobj & its descendants.  This is used
+ * when renaming the dir.
+ */
+void
+dsl_prop_notify_all(dsl_dir_t *dd)
+{
+	dsl_pool_t *dp = dd->dd_pool;
+	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
+	(void) dmu_objset_find_dp(dp, dd->dd_object, dsl_prop_notify_all_cb,
+	    NULL, DS_FIND_CHILDREN);
 }
 
 static void
@@ -498,8 +475,8 @@ dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj,
 	zap_attribute_t *za;
 	int err;
 
-	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
+	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
-	err = dsl_dir_open_obj(dp, ddobj, NULL, FTAG, &dd);
+	err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
 	if (err)
 		return;
 
@@ -510,7 +487,7 @@ dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj,
 		 */
 		err = zap_contains(mos, dd->dd_phys->dd_props_zapobj, propname);
 		if (err == 0) {
-			dsl_dir_close(dd, FTAG);
+			dsl_dir_rele(dd, FTAG);
 			return;
 		}
 		ASSERT3U(err, ==, ENOENT);
@@ -545,26 +522,24 @@ dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj,
 	}
 	kmem_free(za, sizeof (zap_attribute_t));
 	zap_cursor_fini(&zc);
-	dsl_dir_close(dd, FTAG);
+	dsl_dir_rele(dd, FTAG);
 }
 
 void
-dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_prop_set_sync_impl(dsl_dataset_t *ds, const char *propname,
+    zprop_source_t source, int intsz, int numints, const void *value,
+    dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
-	dsl_prop_setarg_t *psa = arg2;
 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
 	uint64_t zapobj, intval, dummy;
 	int isint;
 	char valbuf[32];
-	char *valstr = NULL;
+	const char *valstr = NULL;
 	char *inheritstr;
 	char *recvdstr;
 	char *tbuf = NULL;
 	int err;
 	uint64_t version = spa_version(ds->ds_dir->dd_pool->dp_spa);
-	const char *propname = psa->psa_name;
-	zprop_source_t source = psa->psa_source;
 
 	isint = (dodefault(propname, 8, 1, &intval) == 0);
 
@@ -614,8 +589,8 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 		 */
 		err = zap_remove(mos, zapobj, inheritstr, tx);
 		ASSERT(err == 0 || err == ENOENT);
-		VERIFY(0 == zap_update(mos, zapobj, propname,
+		VERIFY0(zap_update(mos, zapobj, propname,
-		    psa->psa_intsz, psa->psa_numints, psa->psa_value, tx));
+		    intsz, numints, value, tx));
 		break;
 	case ZPROP_SRC_INHERITED:
 		/*
@@ -626,12 +601,10 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 		err = zap_remove(mos, zapobj, propname, tx);
 		ASSERT(err == 0 || err == ENOENT);
 		if (version >= SPA_VERSION_RECVD_PROPS &&
-		    dsl_prop_get_ds(ds, ZPROP_HAS_RECVD, 8, 1, &dummy,
+		    dsl_prop_get_int_ds(ds, ZPROP_HAS_RECVD, &dummy) == 0) {
-		    NULL) == 0) {
 			dummy = 0;
-			err = zap_update(mos, zapobj, inheritstr,
+			VERIFY0(zap_update(mos, zapobj, inheritstr,
-			    8, 1, &dummy, tx);
+			    8, 1, &dummy, tx));
-			ASSERT(err == 0);
 		}
 		break;
 	case ZPROP_SRC_RECEIVED:
@@ -639,7 +612,7 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 		 * set propname$recvd -> value
 		 */
 		err = zap_update(mos, zapobj, recvdstr,
-		    psa->psa_intsz, psa->psa_numints, psa->psa_value, tx);
+		    intsz, numints, value, tx);
 		ASSERT(err == 0);
 		break;
 	case (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED):
@@ -669,7 +642,7 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 	strfree(recvdstr);
 
 	if (isint) {
-		VERIFY(0 == dsl_prop_get_ds(ds, propname, 8, 1, &intval, NULL));
+		VERIFY0(dsl_prop_get_int_ds(ds, propname, &intval));
 
 		if (ds->ds_phys != NULL && dsl_dataset_is_snapshot(ds)) {
 			dsl_prop_cb_record_t *cbr;
@@ -696,7 +669,7 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 		valstr = valbuf;
 	} else {
 		if (source == ZPROP_SRC_LOCAL) {
-			valstr = (char *)psa->psa_value;
+			valstr = value;
 		} else {
 			tbuf = kmem_alloc(ZAP_MAXVALUELEN, KM_PUSHPAGE);
 			if (dsl_prop_get_ds(ds, propname, 1,
@@ -713,118 +686,73 @@ dsl_prop_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 		kmem_free(tbuf, ZAP_MAXVALUELEN);
 }
 
-void
+int
-dsl_props_set_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_prop_set_int(const char *dsname, const char *propname,
+    zprop_source_t source, uint64_t value)
 {
-	dsl_dataset_t *ds = arg1;
+	nvlist_t *nvl = fnvlist_alloc();
-	dsl_props_arg_t *pa = arg2;
+	int error;
-	nvlist_t *props = pa->pa_props;
-	dsl_prop_setarg_t psa;
-	nvpair_t *elem = NULL;
 
-	psa.psa_source = pa->pa_source;
+	fnvlist_add_uint64(nvl, propname, value);
-
+	error = dsl_props_set(dsname, source, nvl);
-	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
+	fnvlist_free(nvl);
-		nvpair_t *pair = elem;
+	return (error);
-
-		psa.psa_name = nvpair_name(pair);
-
-		if (nvpair_type(pair) == DATA_TYPE_NVLIST) {
-			/*
-			 * dsl_prop_get_all_impl() returns properties in this
-			 * format.
-			 */
-			nvlist_t *attrs;
-			VERIFY(nvpair_value_nvlist(pair, &attrs) == 0);
-			VERIFY(nvlist_lookup_nvpair(attrs, ZPROP_VALUE,
-			    &pair) == 0);
-		}
-
-		if (nvpair_type(pair) == DATA_TYPE_STRING) {
-			VERIFY(nvpair_value_string(pair,
-			    (char **)&psa.psa_value) == 0);
-			psa.psa_intsz = 1;
-			psa.psa_numints = strlen(psa.psa_value) + 1;
-		} else {
-			uint64_t intval;
-			VERIFY(nvpair_value_uint64(pair, &intval) == 0);
-			psa.psa_intsz = sizeof (intval);
-			psa.psa_numints = 1;
-			psa.psa_value = &intval;
-		}
-		dsl_prop_set_sync(ds, &psa, tx);
-	}
 }
 
 int
-dsl_prop_set(const char *dsname, const char *propname, zprop_source_t source,
+dsl_prop_set_string(const char *dsname, const char *propname,
-    int intsz, int numints, const void *buf)
+    zprop_source_t source, const char *value)
 {
-	dsl_dataset_t *ds;
+	nvlist_t *nvl = fnvlist_alloc();
-	uint64_t version;
+	int error;
-	int err;
-	dsl_prop_setarg_t psa;
 
-	/*
+	fnvlist_add_string(nvl, propname, value);
-	 * We must do these checks before we get to the syncfunc, since
+	error = dsl_props_set(dsname, source, nvl);
-	 * it can't fail.
+	fnvlist_free(nvl);
-	 */
+	return (error);
-	if (strlen(propname) >= ZAP_MAXNAMELEN)
-		return (ENAMETOOLONG);
-
-	err = dsl_dataset_hold(dsname, FTAG, &ds);
-	if (err)
-		return (err);
-
-	version = spa_version(ds->ds_dir->dd_pool->dp_spa);
-	if (intsz * numints >= (version < SPA_VERSION_STMF_PROP ?
-	    ZAP_OLDMAXVALUELEN : ZAP_MAXVALUELEN)) {
-		dsl_dataset_rele(ds, FTAG);
-		return (E2BIG);
-	}
-	if (dsl_dataset_is_snapshot(ds) &&
-	    version < SPA_VERSION_SNAP_PROPS) {
-		dsl_dataset_rele(ds, FTAG);
-		return (ENOTSUP);
-	}
-
-	psa.psa_name = propname;
-	psa.psa_source = source;
-	psa.psa_intsz = intsz;
-	psa.psa_numints = numints;
-	psa.psa_value = buf;
-	psa.psa_effective_value = -1ULL;
-
-	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
-	    NULL, dsl_prop_set_sync, ds, &psa, 2);
-
-	dsl_dataset_rele(ds, FTAG);
-	return (err);
 }
 
 int
-dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *props)
+dsl_prop_inherit(const char *dsname, const char *propname,
+    zprop_source_t source)
 {
+	nvlist_t *nvl = fnvlist_alloc();
+	int error;
+
+	fnvlist_add_boolean(nvl, propname);
+	error = dsl_props_set(dsname, source, nvl);
+	fnvlist_free(nvl);
+	return (error);
+}
+
+typedef struct dsl_props_set_arg {
+	const char *dpsa_dsname;
+	zprop_source_t dpsa_source;
+	nvlist_t *dpsa_props;
+} dsl_props_set_arg_t;
+
+static int
+dsl_props_set_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_props_set_arg_t *dpsa = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
 	dsl_dataset_t *ds;
 	uint64_t version;
 	nvpair_t *elem = NULL;
-	dsl_props_arg_t pa;
 	int err;
 
-	if ((err = dsl_dataset_hold(dsname, FTAG, &ds)))
+	err = dsl_dataset_hold(dp, dpsa->dpsa_dsname, FTAG, &ds);
+	if (err != 0)
 		return (err);
-	/*
+
-	 * Do these checks before the syncfunc, since it can't fail.
-	 */
 	version = spa_version(ds->ds_dir->dd_pool->dp_spa);
-	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
+	while ((elem = nvlist_next_nvpair(dpsa->dpsa_props, elem)) != NULL) {
 		if (strlen(nvpair_name(elem)) >= ZAP_MAXNAMELEN) {
 			dsl_dataset_rele(ds, FTAG);
 			return (ENAMETOOLONG);
 		}
 		if (nvpair_type(elem) == DATA_TYPE_STRING) {
-			char *valstr;
+			char *valstr = fnvpair_value_string(elem);
-			VERIFY(nvpair_value_string(elem, &valstr) == 0);
 			if (strlen(valstr) >= (version <
 			    SPA_VERSION_STMF_PROP ?
 			    ZAP_OLDMAXVALUELEN : ZAP_MAXVALUELEN)) {
@@ -834,20 +762,83 @@ dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *props)
 		}
 	}
 
-	if (dsl_dataset_is_snapshot(ds) &&
+	if (dsl_dataset_is_snapshot(ds) && version < SPA_VERSION_SNAP_PROPS) {
-	    version < SPA_VERSION_SNAP_PROPS) {
 		dsl_dataset_rele(ds, FTAG);
 		return (ENOTSUP);
 	}
-
-	pa.pa_props = props;
-	pa.pa_source = source;
-
-	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
-	    NULL, dsl_props_set_sync, ds, &pa, 2);
-
 	dsl_dataset_rele(ds, FTAG);
-	return (err);
+	return (0);
+}
+
+void
+dsl_props_set_sync_impl(dsl_dataset_t *ds, zprop_source_t source,
+    nvlist_t *props, dmu_tx_t *tx)
+{
+	nvpair_t *elem = NULL;
+
+	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
+		nvpair_t *pair = elem;
+
+		if (nvpair_type(pair) == DATA_TYPE_NVLIST) {
+			/*
+			 * dsl_prop_get_all_impl() returns properties in this
+			 * format.
+			 */
+			nvlist_t *attrs = fnvpair_value_nvlist(pair);
+			pair = fnvlist_lookup_nvpair(attrs, ZPROP_VALUE);
+		}
+
+		if (nvpair_type(pair) == DATA_TYPE_STRING) {
+			const char *value = fnvpair_value_string(pair);
+			dsl_prop_set_sync_impl(ds, nvpair_name(pair),
+			    source, 1, strlen(value) + 1, value, tx);
+		} else if (nvpair_type(pair) == DATA_TYPE_UINT64) {
+			uint64_t intval = fnvpair_value_uint64(pair);
+			dsl_prop_set_sync_impl(ds, nvpair_name(pair),
+			    source, sizeof (intval), 1, &intval, tx);
+		} else if (nvpair_type(pair) == DATA_TYPE_BOOLEAN) {
+			dsl_prop_set_sync_impl(ds, nvpair_name(pair),
+			    source, 0, 0, NULL, tx);
+		} else {
+			panic("invalid nvpair type");
+		}
+	}
+}
+
+static void
+dsl_props_set_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_props_set_arg_t *dpsa = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	dsl_dataset_t *ds;
+
+	VERIFY0(dsl_dataset_hold(dp, dpsa->dpsa_dsname, FTAG, &ds));
+	dsl_props_set_sync_impl(ds, dpsa->dpsa_source, dpsa->dpsa_props, tx);
+	dsl_dataset_rele(ds, FTAG);
+}
+
+/*
+ * All-or-nothing; if any prop can't be set, nothing will be modified.
+ */
+int
+dsl_props_set(const char *dsname, zprop_source_t source, nvlist_t *props)
+{
+	dsl_props_set_arg_t dpsa;
+	int nblks = 0;
+
+	dpsa.dpsa_dsname = dsname;
+	dpsa.dpsa_source = source;
+	dpsa.dpsa_props = props;
+
+	/*
+	 * If the source includes NONE, then we will only be removing entries
+	 * from the ZAP object.  In that case don't check for ENOSPC.
+	 */
+	if ((source & ZPROP_SRC_NONE) == 0)
+		nblks = 2 * fnvlist_num_pairs(props);
+
+	return (dsl_sync_task(dsname, dsl_props_set_check, dsl_props_set_sync,
+	    &dpsa, nblks));
 }
 
 typedef enum dsl_prop_getflags {
@@ -997,7 +988,7 @@ dsl_prop_get_all_ds(dsl_dataset_t *ds, nvlist_t **nvp,
 	if (dsl_dataset_is_snapshot(ds))
 		flags |= DSL_PROP_GET_SNAPSHOT;
 
-	rw_enter(&dp->dp_config_rwlock, RW_READER);
+	ASSERT(dsl_pool_config_held(dp));
 
 	if (ds->ds_phys->ds_props_obj != 0) {
 		ASSERT(flags & DSL_PROP_GET_SNAPSHOT);
@@ -1022,58 +1013,51 @@ dsl_prop_get_all_ds(dsl_dataset_t *ds, nvlist_t **nvp,
 			break;
 	}
 out:
-	rw_exit(&dp->dp_config_rwlock);
 	return (err);
 }
 
 boolean_t
-dsl_prop_get_hasrecvd(objset_t *os)
+dsl_prop_get_hasrecvd(const char *dsname)
 {
-	dsl_dataset_t *ds = os->os_dsl_dataset;
-	int rc;
 	uint64_t dummy;
 
-	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
+	return (0 ==
-	rc = dsl_prop_get_ds(ds, ZPROP_HAS_RECVD, 8, 1, &dummy, NULL);
+	    dsl_prop_get_integer(dsname, ZPROP_HAS_RECVD, &dummy, NULL));
-	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
-	ASSERT(rc != 0 || spa_version(os->os_spa) >= SPA_VERSION_RECVD_PROPS);
-	return (rc == 0);
 }
 
-static void
+static int
-dsl_prop_set_hasrecvd_impl(objset_t *os, zprop_source_t source)
+dsl_prop_set_hasrecvd_impl(const char *dsname, zprop_source_t source)
 {
-	dsl_dataset_t *ds = os->os_dsl_dataset;
+	uint64_t version;
-	uint64_t dummy = 0;
+	spa_t *spa;
-	dsl_prop_setarg_t psa;
+	int error = 0;
 
-	if (spa_version(os->os_spa) < SPA_VERSION_RECVD_PROPS)
+	VERIFY0(spa_open(dsname, &spa, FTAG));
-		return;
+	version = spa_version(spa);
+	spa_close(spa, FTAG);
 
-	dsl_prop_setarg_init_uint64(&psa, ZPROP_HAS_RECVD, source, &dummy);
+	if (version >= SPA_VERSION_RECVD_PROPS)
-
+		error = dsl_prop_set_int(dsname, ZPROP_HAS_RECVD, source, 0);
-	(void) dsl_sync_task_do(ds->ds_dir->dd_pool, NULL,
+	return (error);
-	    dsl_prop_set_sync, ds, &psa, 2);
 }
 
 /*
  * Call after successfully receiving properties to ensure that only the first
  * receive on or after SPA_VERSION_RECVD_PROPS blows away local properties.
  */
-void
+int
-dsl_prop_set_hasrecvd(objset_t *os)
+dsl_prop_set_hasrecvd(const char *dsname)
 {
-	if (dsl_prop_get_hasrecvd(os)) {
+	int error = 0;
-		ASSERT(spa_version(os->os_spa) >= SPA_VERSION_RECVD_PROPS);
+	if (!dsl_prop_get_hasrecvd(dsname))
-		return;
+		error = dsl_prop_set_hasrecvd_impl(dsname, ZPROP_SRC_LOCAL);
-	}
+	return (error);
-	dsl_prop_set_hasrecvd_impl(os, ZPROP_SRC_LOCAL);
 }
 
 void
-dsl_prop_unset_hasrecvd(objset_t *os)
+dsl_prop_unset_hasrecvd(const char *dsname)
 {
-	dsl_prop_set_hasrecvd_impl(os, ZPROP_SRC_NONE);
+	VERIFY0(dsl_prop_set_hasrecvd_impl(dsname, ZPROP_SRC_NONE));
 }
 
 int
@@ -1083,16 +1067,25 @@ dsl_prop_get_all(objset_t *os, nvlist_t **nvp)
 }
 
 int
-dsl_prop_get_received(objset_t *os, nvlist_t **nvp)
+dsl_prop_get_received(const char *dsname, nvlist_t **nvp)
 {
+	objset_t *os;
+	int error;
+
 	/*
 	 * Received properties are not distinguishable from local properties
 	 * until the dataset has received properties on or after
 	 * SPA_VERSION_RECVD_PROPS.
 	 */
-	dsl_prop_getflags_t flags = (dsl_prop_get_hasrecvd(os) ?
+	dsl_prop_getflags_t flags = (dsl_prop_get_hasrecvd(dsname) ?
 	    DSL_PROP_GET_RECEIVED : DSL_PROP_GET_LOCAL);
-	return (dsl_prop_get_all_ds(os->os_dsl_dataset, nvp, flags));
+
+	error = dmu_objset_hold(dsname, FTAG, &os);
+	if (error != 0)
+		return (error);
+	error = dsl_prop_get_all_ds(os->os_dsl_dataset, nvp, flags);
+	dmu_objset_rele(os, FTAG);
+	return (error);
 }
 
 void
@@ -1138,8 +1131,6 @@ dsl_prop_nvlist_add_string(nvlist_t *nv, zfs_prop_t prop, const char *value)
 #if defined(_KERNEL) && defined(HAVE_SPL)
 EXPORT_SYMBOL(dsl_prop_register);
 EXPORT_SYMBOL(dsl_prop_unregister);
-EXPORT_SYMBOL(dsl_prop_numcb);
-EXPORT_SYMBOL(dsl_prop_set);
 EXPORT_SYMBOL(dsl_prop_get);
 EXPORT_SYMBOL(dsl_prop_get_integer);
 EXPORT_SYMBOL(dsl_prop_get_all);

module/zfs/dsl_scan.c

@@ -53,7 +53,7 @@
 typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_t *);
 
 static scan_cb_t dsl_scan_scrub_cb;
-static dsl_syncfunc_t dsl_scan_cancel_sync;
+static void dsl_scan_cancel_sync(void *, dmu_tx_t *);
 static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *tx);
 
 int zfs_top_maxinflight = 32;		/* maximum I/Os per top-level */
@@ -150,9 +150,9 @@ dsl_scan_fini(dsl_pool_t *dp)
 
 /* ARGSUSED */
 static int
-dsl_scan_setup_check(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
 {
-	dsl_scan_t *scn = arg1;
+	dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
 
 	if (scn->scn_phys.scn_state == DSS_SCANNING)
 		return (EBUSY);
@@ -160,12 +160,11 @@ dsl_scan_setup_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	return (0);
 }
 
-/* ARGSUSED */
 static void
-dsl_scan_setup_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
 {
-	dsl_scan_t *scn = arg1;
+	dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
-	pool_scan_func_t *funcp = arg2;
+	pool_scan_func_t *funcp = arg;
 	dmu_object_type_t ot = 0;
 	dsl_pool_t *dp = scn->scn_dp;
 	spa_t *spa = dp->dp_spa;
@@ -312,9 +311,9 @@ dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
 
 /* ARGSUSED */
 static int
-dsl_scan_cancel_check(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
 {
-	dsl_scan_t *scn = arg1;
+	dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
 
 	if (scn->scn_phys.scn_state != DSS_SCANNING)
 		return (ENOENT);
@@ -323,9 +322,9 @@ dsl_scan_cancel_check(void *arg1, void *arg2, dmu_tx_t *tx)
 
 /* ARGSUSED */
 static void
-dsl_scan_cancel_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx)
 {
-	dsl_scan_t *scn = arg1;
+	dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
 
 	dsl_scan_done(scn, B_FALSE, tx);
 	dsl_scan_sync_state(scn, tx);
@@ -334,12 +333,8 @@ dsl_scan_cancel_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 int
 dsl_scan_cancel(dsl_pool_t *dp)
 {
-	boolean_t complete = B_FALSE;
+	return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
-	int err;
+	    dsl_scan_cancel_sync, NULL, 3));
-
-	err = dsl_sync_task_do(dp, dsl_scan_cancel_check,
-	    dsl_scan_cancel_sync, dp->dp_scan, &complete, 3);
-	return (err);
 }
 
 static void dsl_scan_visitbp(blkptr_t *bp,
@@ -375,7 +370,7 @@ dsl_scan_ds_maxtxg(dsl_dataset_t *ds)
 static void
 dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx)
 {
-	VERIFY(0 == zap_update(scn->scn_dp->dp_meta_objset,
+	VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
 	    DMU_POOL_DIRECTORY_OBJECT,
 	    DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
 	    &scn->scn_phys, tx));
@@ -959,33 +954,33 @@ struct enqueue_clones_arg {
 
 /* ARGSUSED */
 static int
-enqueue_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
+enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
 {
 	struct enqueue_clones_arg *eca = arg;
 	dsl_dataset_t *ds;
 	int err;
-	dsl_pool_t *dp = spa->spa_dsl_pool;
 	dsl_scan_t *scn = dp->dp_scan;
 
-	err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
+	if (hds->ds_dir->dd_phys->dd_origin_obj != eca->originobj)
+		return (0);
+
+	err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
 	if (err)
 		return (err);
 
-	if (ds->ds_dir->dd_phys->dd_origin_obj == eca->originobj) {
+	while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) {
-		while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) {
+		dsl_dataset_t *prev;
-			dsl_dataset_t *prev;
+		err = dsl_dataset_hold_obj(dp,
-			err = dsl_dataset_hold_obj(dp,
+		    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
-			    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
 
-			dsl_dataset_rele(ds, FTAG);
+		dsl_dataset_rele(ds, FTAG);
-			if (err)
+		if (err)
-				return (err);
+			return (err);
-			ds = prev;
+		ds = prev;
-		}
-		VERIFY(zap_add_int_key(dp->dp_meta_objset,
-		    scn->scn_phys.scn_queue_obj, ds->ds_object,
-		    ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0);
 	}
+	VERIFY(zap_add_int_key(dp->dp_meta_objset,
+	    scn->scn_phys.scn_queue_obj, ds->ds_object,
+	    ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0);
 	dsl_dataset_rele(ds, FTAG);
 	return (0);
 }
@@ -1075,17 +1070,17 @@ dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx)
 		}
 
 		if (usenext) {
-			VERIFY(zap_join_key(dp->dp_meta_objset,
+			VERIFY0(zap_join_key(dp->dp_meta_objset,
 			    ds->ds_phys->ds_next_clones_obj,
 			    scn->scn_phys.scn_queue_obj,
-			    ds->ds_phys->ds_creation_txg, tx) == 0);
+			    ds->ds_phys->ds_creation_txg, tx));
 		} else {
 			struct enqueue_clones_arg eca;
 			eca.tx = tx;
 			eca.originobj = ds->ds_object;
 
-			(void) dmu_objset_find_spa(ds->ds_dir->dd_pool->dp_spa,
+			VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
-			    NULL, enqueue_clones_cb, &eca, DS_FIND_CHILDREN);
+			    enqueue_clones_cb, &eca, DS_FIND_CHILDREN));
 		}
 	}
 
@@ -1095,15 +1090,14 @@ out:
 
 /* ARGSUSED */
 static int
-enqueue_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
+enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
 {
 	dmu_tx_t *tx = arg;
 	dsl_dataset_t *ds;
 	int err;
-	dsl_pool_t *dp = spa->spa_dsl_pool;
 	dsl_scan_t *scn = dp->dp_scan;
 
-	err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
+	err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
 	if (err)
 		return (err);
 
@@ -1261,8 +1255,8 @@ dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx)
 			return;
 
 		if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) {
-			VERIFY(0 == dmu_objset_find_spa(dp->dp_spa,
+			VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
-			    NULL, enqueue_cb, tx, DS_FIND_CHILDREN));
+			    enqueue_cb, tx, DS_FIND_CHILDREN));
 		} else {
 			dsl_scan_visitds(scn,
 			    dp->dp_origin_snap->ds_object, tx);
@@ -1402,7 +1396,7 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 			func = POOL_SCAN_RESILVER;
 		zfs_dbgmsg("restarting scan func=%u txg=%llu",
 		    func, tx->tx_txg);
-		dsl_scan_setup_sync(scn, &func, tx);
+		dsl_scan_setup_sync(&func, tx);
 	}
 
 	if (!dsl_scan_active(scn) ||
@@ -1436,21 +1430,21 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 			err = bptree_iterate(dp->dp_meta_objset,
 			    dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb,
 			    scn, tx);
-			VERIFY3U(0, ==, zio_wait(scn->scn_zio_root));
+			VERIFY0(zio_wait(scn->scn_zio_root));
-			if (err != 0)
-				return;
 
-			/* disable async destroy feature */
+			if (err == 0) {
-			spa_feature_decr(spa,
+				zfeature_info_t *feat = &spa_feature_table
-			    &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY], tx);
+				    [SPA_FEATURE_ASYNC_DESTROY];
-			ASSERT(!spa_feature_is_active(spa,
+				/* finished; deactivate async destroy feature */
-			    &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]));
+				spa_feature_decr(spa, feat, tx);
-			VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
+				ASSERT(!spa_feature_is_active(spa, feat));
-			    DMU_POOL_DIRECTORY_OBJECT,
+				VERIFY0(zap_remove(dp->dp_meta_objset,
-			    DMU_POOL_BPTREE_OBJ, tx));
+				    DMU_POOL_DIRECTORY_OBJECT,
-			VERIFY3U(0, ==, bptree_free(dp->dp_meta_objset,
+				    DMU_POOL_BPTREE_OBJ, tx));
-			    dp->dp_bptree_obj, tx));
+				VERIFY0(bptree_free(dp->dp_meta_objset,
-			dp->dp_bptree_obj = 0;
+				    dp->dp_bptree_obj, tx));
+				dp->dp_bptree_obj = 0;
+			}
 		}
 		if (scn->scn_visited_this_txg) {
 			zfs_dbgmsg("freed %llu blocks in %llums from "
@@ -1497,7 +1491,9 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 
 	scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
 	    NULL, ZIO_FLAG_CANFAIL);
+	dsl_pool_config_enter(dp, FTAG);
 	dsl_scan_visit(scn, tx);
+	dsl_pool_config_exit(dp, FTAG);
 	(void) zio_wait(scn->scn_zio_root);
 	scn->scn_zio_root = NULL;
 
@@ -1734,8 +1730,8 @@ dsl_scan(dsl_pool_t *dp, pool_scan_func_t func)
 	spa->spa_scrub_reopen = B_FALSE;
 	(void) spa_vdev_state_exit(spa, NULL, 0);
 
-	return (dsl_sync_task_do(dp, dsl_scan_setup_check,
+	return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
-	    dsl_scan_setup_sync, dp->dp_scan, &func, 0));
+	    dsl_scan_setup_sync, &func, 0));
 }
 
 #if defined(_KERNEL) && defined(HAVE_SPL)

module/zfs/dsl_synctask.c

@@ -34,138 +34,115 @@
 
 /* ARGSUSED */
 static int
-dsl_null_checkfunc(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_null_checkfunc(void *arg, dmu_tx_t *tx)
 {
 	return (0);
 }
 
-dsl_sync_task_group_t *
+/*
-dsl_sync_task_group_create(dsl_pool_t *dp)
+ * Called from open context to perform a callback in syncing context.  Waits
-{
+ * for the operation to complete.
-	dsl_sync_task_group_t *dstg;
+ *
-
+ * The checkfunc will be called from open context as a preliminary check
-	dstg = kmem_zalloc(sizeof (dsl_sync_task_group_t), KM_SLEEP);
+ * which can quickly fail.  If it succeeds, it will be called again from
-	list_create(&dstg->dstg_tasks, sizeof (dsl_sync_task_t),
+ * syncing context.  The checkfunc should generally be designed to work
-	    offsetof(dsl_sync_task_t, dst_node));
+ * properly in either context, but if necessary it can check
-	dstg->dstg_pool = dp;
+ * dmu_tx_is_syncing(tx).
-
+ *
-	return (dstg);
+ * The synctask infrastructure enforces proper locking strategy with respect
-}
+ * to the dp_config_rwlock -- the lock will always be held when the callbacks
-
+ * are called.  It will be held for read during the open-context (preliminary)
-void
+ * call to the checkfunc, and then held for write from syncing context during
-dsl_sync_task_create(dsl_sync_task_group_t *dstg,
+ * the calls to the check and sync funcs.
-    dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc,
+ *
-    void *arg1, void *arg2, int blocks_modified)
+ * A dataset or pool name can be passed as the first argument.  Typically,
-{
+ * the check func will hold, check the return value of the hold, and then
-	dsl_sync_task_t *dst;
+ * release the dataset.  The sync func will VERIFYO(hold()) the dataset.
-
+ * This is safe because no changes can be made between the check and sync funcs,
-	if (checkfunc == NULL)
+ * and the sync func will only be called if the check func successfully opened
-		checkfunc = dsl_null_checkfunc;
+ * the dataset.
-	dst = kmem_zalloc(sizeof (dsl_sync_task_t), KM_SLEEP);
+ */
-	dst->dst_checkfunc = checkfunc;
-	dst->dst_syncfunc = syncfunc;
-	dst->dst_arg1 = arg1;
-	dst->dst_arg2 = arg2;
-	list_insert_tail(&dstg->dstg_tasks, dst);
-
-	dstg->dstg_space += blocks_modified << DST_AVG_BLKSHIFT;
-}
-
 int
-dsl_sync_task_group_wait(dsl_sync_task_group_t *dstg)
+dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
+    dsl_syncfunc_t *syncfunc, void *arg, int blocks_modified)
 {
+	spa_t *spa;
 	dmu_tx_t *tx;
-	uint64_t txg;
+	int err;
-	dsl_sync_task_t *dst;
+	dsl_sync_task_t dst = { { { NULL } } };
+	dsl_pool_t *dp;
+
+	err = spa_open(pool, &spa, FTAG);
+	if (err != 0)
+		return (err);
+	dp = spa_get_dsl(spa);
 
 top:
-	tx = dmu_tx_create_dd(dstg->dstg_pool->dp_mos_dir);
+	tx = dmu_tx_create_dd(dp->dp_mos_dir);
-	VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT));
+	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
 
-	txg = dmu_tx_get_txg(tx);
+	dst.dst_pool = dp;
+	dst.dst_txg = dmu_tx_get_txg(tx);
+	dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
+	dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
+	dst.dst_syncfunc = syncfunc;
+	dst.dst_arg = arg;
+	dst.dst_error = 0;
+	dst.dst_nowaiter = B_FALSE;
 
-	/* Do a preliminary error check. */
+	dsl_pool_config_enter(dp, FTAG);
-	dstg->dstg_err = 0;
+	err = dst.dst_checkfunc(arg, tx);
-#ifdef ZFS_DEBUG
+	dsl_pool_config_exit(dp, FTAG);
-	/*
-	 * Only check half the time, otherwise, the sync-context
-	 * check will almost never fail.
-	 */
-	if (spa_get_random(2) == 0)
-		goto skip;
-#endif
-	rw_enter(&dstg->dstg_pool->dp_config_rwlock, RW_READER);
-	for (dst = list_head(&dstg->dstg_tasks); dst;
-	    dst = list_next(&dstg->dstg_tasks, dst)) {
-		dst->dst_err =
-		    dst->dst_checkfunc(dst->dst_arg1, dst->dst_arg2, tx);
-		if (dst->dst_err)
-			dstg->dstg_err = dst->dst_err;
-	}
-	rw_exit(&dstg->dstg_pool->dp_config_rwlock);
 
-	if (dstg->dstg_err) {
+	if (err != 0) {
 		dmu_tx_commit(tx);
-		return (dstg->dstg_err);
+		spa_close(spa, FTAG);
+		return (err);
 	}
-#ifdef ZFS_DEBUG
-skip:
-#endif
 
-	/*
+	VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, &dst, dst.dst_txg));
-	 * We don't generally have many sync tasks, so pay the price of
-	 * add_tail to get the tasks executed in the right order.
-	 */
-	VERIFY(0 == txg_list_add_tail(&dstg->dstg_pool->dp_sync_tasks,
-	    dstg, txg));
 
 	dmu_tx_commit(tx);
 
-	txg_wait_synced(dstg->dstg_pool, txg);
+	txg_wait_synced(dp, dst.dst_txg);
 
-	if (dstg->dstg_err == EAGAIN) {
+	if (dst.dst_error == EAGAIN) {
-		txg_wait_synced(dstg->dstg_pool, txg + TXG_DEFER_SIZE);
+		txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
 		goto top;
 	}
 
-	return (dstg->dstg_err);
+	spa_close(spa, FTAG);
+	return (dst.dst_error);
 }
 
 void
-dsl_sync_task_group_nowait(dsl_sync_task_group_t *dstg, dmu_tx_t *tx)
+dsl_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg,
+    int blocks_modified, dmu_tx_t *tx)
 {
-	uint64_t txg;
+	dsl_sync_task_t *dst = kmem_zalloc(sizeof (*dst), KM_SLEEP);
 
-	dstg->dstg_nowaiter = B_TRUE;
+	dst->dst_pool = dp;
-	txg = dmu_tx_get_txg(tx);
+	dst->dst_txg = dmu_tx_get_txg(tx);
-	/*
+	dst->dst_space = blocks_modified << DST_AVG_BLKSHIFT;
-	 * We don't generally have many sync tasks, so pay the price of
+	dst->dst_checkfunc = dsl_null_checkfunc;
-	 * add_tail to get the tasks executed in the right order.
+	dst->dst_syncfunc = syncfunc;
-	 */
+	dst->dst_arg = arg;
-	VERIFY(0 == txg_list_add_tail(&dstg->dstg_pool->dp_sync_tasks,
+	dst->dst_error = 0;
-	    dstg, txg));
+	dst->dst_nowaiter = B_TRUE;
+
+	VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, dst, dst->dst_txg));
 }
 
+/*
+ * Called in syncing context to execute the synctask.
+ */
 void
-dsl_sync_task_group_destroy(dsl_sync_task_group_t *dstg)
+dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx)
 {
-	dsl_sync_task_t *dst;
+	dsl_pool_t *dp = dst->dst_pool;
-
-	while ((dst = list_head(&dstg->dstg_tasks))) {
-		list_remove(&dstg->dstg_tasks, dst);
-		kmem_free(dst, sizeof (dsl_sync_task_t));
-	}
-	kmem_free(dstg, sizeof (dsl_sync_task_group_t));
-}
-
-void
-dsl_sync_task_group_sync(dsl_sync_task_group_t *dstg, dmu_tx_t *tx)
-{
-	dsl_sync_task_t *dst;
-	dsl_pool_t *dp = dstg->dstg_pool;
 	uint64_t quota, used;
 
-	ASSERT0(dstg->dstg_err);
+	ASSERT0(dst->dst_error);
 
 	/*
 	 * Check for sufficient space.  We just check against what's
@@ -177,70 +154,24 @@ dsl_sync_task_group_sync(dsl_sync_task_group_t *dstg, dmu_tx_t *tx)
 	    metaslab_class_get_deferred(spa_normal_class(dp->dp_spa));
 	used = dp->dp_root_dir->dd_phys->dd_used_bytes;
 	/* MOS space is triple-dittoed, so we multiply by 3. */
-	if (dstg->dstg_space > 0 && used + dstg->dstg_space * 3 > quota) {
+	if (dst->dst_space > 0 && used + dst->dst_space * 3 > quota) {
-		dstg->dstg_err = ENOSPC;
+		dst->dst_error = ENOSPC;
+		if (dst->dst_nowaiter)
+			kmem_free(dst, sizeof (*dst));
 		return;
 	}
 
 	/*
-	 * Check for errors by calling checkfuncs.
+	 * Check for errors by calling checkfunc.
 	 */
-	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
+	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
-	for (dst = list_head(&dstg->dstg_tasks); dst;
+	dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx);
-	    dst = list_next(&dstg->dstg_tasks, dst)) {
+	if (dst->dst_error == 0)
-		dst->dst_err =
+		dst->dst_syncfunc(dst->dst_arg, tx);
-		    dst->dst_checkfunc(dst->dst_arg1, dst->dst_arg2, tx);
+	rrw_exit(&dp->dp_config_rwlock, FTAG);
-		if (dst->dst_err)
+	if (dst->dst_nowaiter)
-			dstg->dstg_err = dst->dst_err;
+		kmem_free(dst, sizeof (*dst));
-	}
-
-	if (dstg->dstg_err == 0) {
-		/*
-		 * Execute sync tasks.
-		 */
-		for (dst = list_head(&dstg->dstg_tasks); dst;
-		    dst = list_next(&dstg->dstg_tasks, dst)) {
-			dst->dst_syncfunc(dst->dst_arg1, dst->dst_arg2, tx);
-		}
-	}
-	rw_exit(&dp->dp_config_rwlock);
-
-	if (dstg->dstg_nowaiter)
-		dsl_sync_task_group_destroy(dstg);
-}
-
-int
-dsl_sync_task_do(dsl_pool_t *dp,
-    dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc,
-    void *arg1, void *arg2, int blocks_modified)
-{
-	dsl_sync_task_group_t *dstg;
-	int err;
-
-	ASSERT(spa_writeable(dp->dp_spa));
-
-	dstg = dsl_sync_task_group_create(dp);
-	dsl_sync_task_create(dstg, checkfunc, syncfunc,
-	    arg1, arg2, blocks_modified);
-	err = dsl_sync_task_group_wait(dstg);
-	dsl_sync_task_group_destroy(dstg);
-	return (err);
-}
-
-void
-dsl_sync_task_do_nowait(dsl_pool_t *dp,
-    dsl_checkfunc_t *checkfunc, dsl_syncfunc_t *syncfunc,
-    void *arg1, void *arg2, int blocks_modified, dmu_tx_t *tx)
-{
-	dsl_sync_task_group_t *dstg;
-
-	dstg = dsl_sync_task_group_create(dp);
-	dsl_sync_task_create(dstg, checkfunc, syncfunc,
-	    arg1, arg2, blocks_modified);
-	dsl_sync_task_group_nowait(dstg, tx);
 }
 
 #if defined(_KERNEL) && defined(HAVE_SPL)
-EXPORT_SYMBOL(dsl_sync_task_do);
-EXPORT_SYMBOL(dsl_sync_task_do_nowait);
 #endif

module/zfs/dsl_userhold.c

@@ -0,0 +1,537 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/dsl_userhold.h>
+#include <sys/dsl_dataset.h>
+#include <sys/dsl_destroy.h>
+#include <sys/dsl_synctask.h>
+#include <sys/dmu_tx.h>
+#include <sys/zfs_onexit.h>
+#include <sys/dsl_pool.h>
+#include <sys/dsl_dir.h>
+#include <sys/zfs_ioctl.h>
+#include <sys/zap.h>
+
+typedef struct dsl_dataset_user_hold_arg {
+	nvlist_t *dduha_holds;
+	nvlist_t *dduha_errlist;
+	minor_t dduha_minor;
+} dsl_dataset_user_hold_arg_t;
+
+/*
+ * If you add new checks here, you may need to add additional checks to the
+ * "temporary" case in snapshot_check() in dmu_objset.c.
+ */
+int
+dsl_dataset_user_hold_check_one(dsl_dataset_t *ds, const char *htag,
+    boolean_t temphold, dmu_tx_t *tx)
+{
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	objset_t *mos = dp->dp_meta_objset;
+	int error = 0;
+
+	if (strlen(htag) > MAXNAMELEN)
+		return (E2BIG);
+	/* Tempholds have a more restricted length */
+	if (temphold && strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
+		return (E2BIG);
+
+	/* tags must be unique (if ds already exists) */
+	if (ds != NULL) {
+		mutex_enter(&ds->ds_lock);
+		if (ds->ds_phys->ds_userrefs_obj != 0) {
+			uint64_t value;
+			error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj,
+			    htag, 8, 1, &value);
+			if (error == 0)
+				error = EEXIST;
+			else if (error == ENOENT)
+				error = 0;
+		}
+		mutex_exit(&ds->ds_lock);
+	}
+
+	return (error);
+}
+
+static int
+dsl_dataset_user_hold_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_dataset_user_hold_arg_t *dduha = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	nvpair_t *pair;
+	int rv = 0;
+
+	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS)
+		return (ENOTSUP);
+
+	for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL;
+	    pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
+		int error = 0;
+		dsl_dataset_t *ds;
+		char *htag;
+
+		/* must be a snapshot */
+		if (strchr(nvpair_name(pair), '@') == NULL)
+			error = EINVAL;
+
+		if (error == 0)
+			error = nvpair_value_string(pair, &htag);
+		if (error == 0) {
+			error = dsl_dataset_hold(dp,
+			    nvpair_name(pair), FTAG, &ds);
+		}
+		if (error == 0) {
+			error = dsl_dataset_user_hold_check_one(ds, htag,
+			    dduha->dduha_minor != 0, tx);
+			dsl_dataset_rele(ds, FTAG);
+		}
+
+		if (error != 0) {
+			rv = error;
+			fnvlist_add_int32(dduha->dduha_errlist,
+			    nvpair_name(pair), error);
+		}
+	}
+	return (rv);
+}
+
+void
+dsl_dataset_user_hold_sync_one(dsl_dataset_t *ds, const char *htag,
+    minor_t minor, uint64_t now, dmu_tx_t *tx)
+{
+	dsl_pool_t *dp = ds->ds_dir->dd_pool;
+	objset_t *mos = dp->dp_meta_objset;
+	uint64_t zapobj;
+
+	mutex_enter(&ds->ds_lock);
+	if (ds->ds_phys->ds_userrefs_obj == 0) {
+		/*
+		 * This is the first user hold for this dataset.  Create
+		 * the userrefs zap object.
+		 */
+		dmu_buf_will_dirty(ds->ds_dbuf, tx);
+		zapobj = ds->ds_phys->ds_userrefs_obj =
+		    zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
+	} else {
+		zapobj = ds->ds_phys->ds_userrefs_obj;
+	}
+	ds->ds_userrefs++;
+	mutex_exit(&ds->ds_lock);
+
+	VERIFY0(zap_add(mos, zapobj, htag, 8, 1, &now, tx));
+
+	if (minor != 0) {
+		VERIFY0(dsl_pool_user_hold(dp, ds->ds_object,
+		    htag, now, tx));
+		dsl_register_onexit_hold_cleanup(ds, htag, minor);
+	}
+
+	spa_history_log_internal_ds(ds, "hold", tx,
+	    "tag=%s temp=%d refs=%llu",
+	    htag, minor != 0, ds->ds_userrefs);
+}
+
+static void
+dsl_dataset_user_hold_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_dataset_user_hold_arg_t *dduha = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	nvpair_t *pair;
+	uint64_t now = gethrestime_sec();
+
+	for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL); pair != NULL;
+	    pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
+		dsl_dataset_t *ds;
+		VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
+		dsl_dataset_user_hold_sync_one(ds, fnvpair_value_string(pair),
+		    dduha->dduha_minor, now, tx);
+		dsl_dataset_rele(ds, FTAG);
+	}
+}
+
+/*
+ * holds is nvl of snapname -> holdname
+ * errlist will be filled in with snapname -> error
+ * if cleanup_minor is not 0, the holds will be temporary, cleaned up
+ * when the process exits.
+ *
+ * if any fails, all will fail.
+ */
+int
+dsl_dataset_user_hold(nvlist_t *holds, minor_t cleanup_minor, nvlist_t *errlist)
+{
+	dsl_dataset_user_hold_arg_t dduha;
+	nvpair_t *pair;
+
+	pair = nvlist_next_nvpair(holds, NULL);
+	if (pair == NULL)
+		return (0);
+
+	dduha.dduha_holds = holds;
+	dduha.dduha_errlist = errlist;
+	dduha.dduha_minor = cleanup_minor;
+
+	return (dsl_sync_task(nvpair_name(pair), dsl_dataset_user_hold_check,
+	    dsl_dataset_user_hold_sync, &dduha, fnvlist_num_pairs(holds)));
+}
+
+typedef struct dsl_dataset_user_release_arg {
+	nvlist_t *ddura_holds;
+	nvlist_t *ddura_todelete;
+	nvlist_t *ddura_errlist;
+} dsl_dataset_user_release_arg_t;
+
+static int
+dsl_dataset_user_release_check_one(dsl_dataset_t *ds,
+    nvlist_t *holds, boolean_t *todelete)
+{
+	uint64_t zapobj;
+	nvpair_t *pair;
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	int error;
+	int numholds = 0;
+
+	*todelete = B_FALSE;
+
+	if (!dsl_dataset_is_snapshot(ds))
+		return (EINVAL);
+
+	zapobj = ds->ds_phys->ds_userrefs_obj;
+	if (zapobj == 0)
+		return (ESRCH);
+
+	for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
+	    pair = nvlist_next_nvpair(holds, pair)) {
+		/* Make sure the hold exists */
+		uint64_t tmp;
+		error = zap_lookup(mos, zapobj, nvpair_name(pair), 8, 1, &tmp);
+		if (error == ENOENT)
+			error = ESRCH;
+		if (error != 0)
+			return (error);
+		numholds++;
+	}
+
+	if (DS_IS_DEFER_DESTROY(ds) && ds->ds_phys->ds_num_children == 1 &&
+	    ds->ds_userrefs == numholds) {
+		/* we need to destroy the snapshot as well */
+
+		if (dsl_dataset_long_held(ds))
+			return (EBUSY);
+		*todelete = B_TRUE;
+	}
+	return (0);
+}
+
+static int
+dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_dataset_user_release_arg_t *ddura = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	nvpair_t *pair;
+	int rv = 0;
+
+	if (!dmu_tx_is_syncing(tx))
+		return (0);
+
+	for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL;
+	    pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
+		const char *name = nvpair_name(pair);
+		int error;
+		dsl_dataset_t *ds;
+		nvlist_t *holds;
+
+		error = nvpair_value_nvlist(pair, &holds);
+		if (error != 0)
+			return (EINVAL);
+
+		error = dsl_dataset_hold(dp, name, FTAG, &ds);
+		if (error == 0) {
+			boolean_t deleteme;
+			error = dsl_dataset_user_release_check_one(ds,
+			    holds, &deleteme);
+			if (error == 0 && deleteme) {
+				fnvlist_add_boolean(ddura->ddura_todelete,
+				    name);
+			}
+			dsl_dataset_rele(ds, FTAG);
+		}
+		if (error != 0) {
+			if (ddura->ddura_errlist != NULL) {
+				fnvlist_add_int32(ddura->ddura_errlist,
+				    name, error);
+			}
+			rv = error;
+		}
+	}
+	return (rv);
+}
+
+static void
+dsl_dataset_user_release_sync_one(dsl_dataset_t *ds, nvlist_t *holds,
+    dmu_tx_t *tx)
+{
+	dsl_pool_t *dp = ds->ds_dir->dd_pool;
+	objset_t *mos = dp->dp_meta_objset;
+	uint64_t zapobj;
+	int error;
+	nvpair_t *pair;
+
+	for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
+	    pair = nvlist_next_nvpair(holds, pair)) {
+		ds->ds_userrefs--;
+		error = dsl_pool_user_release(dp, ds->ds_object,
+		    nvpair_name(pair), tx);
+		VERIFY(error == 0 || error == ENOENT);
+		zapobj = ds->ds_phys->ds_userrefs_obj;
+		VERIFY0(zap_remove(mos, zapobj, nvpair_name(pair), tx));
+
+		spa_history_log_internal_ds(ds, "release", tx,
+		    "tag=%s refs=%lld", nvpair_name(pair),
+		    (longlong_t)ds->ds_userrefs);
+	}
+}
+
+static void
+dsl_dataset_user_release_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_dataset_user_release_arg_t *ddura = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	nvpair_t *pair;
+
+	for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL;
+	    pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
+		dsl_dataset_t *ds;
+
+		VERIFY0(dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds));
+		dsl_dataset_user_release_sync_one(ds,
+		    fnvpair_value_nvlist(pair), tx);
+		if (nvlist_exists(ddura->ddura_todelete,
+		    nvpair_name(pair))) {
+			ASSERT(ds->ds_userrefs == 0 &&
+			    ds->ds_phys->ds_num_children == 1 &&
+			    DS_IS_DEFER_DESTROY(ds));
+			dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
+		}
+		dsl_dataset_rele(ds, FTAG);
+	}
+}
+
+/*
+ * holds is nvl of snapname -> { holdname, ... }
+ * errlist will be filled in with snapname -> error
+ *
+ * if any fails, all will fail.
+ */
+int
+dsl_dataset_user_release(nvlist_t *holds, nvlist_t *errlist)
+{
+	dsl_dataset_user_release_arg_t ddura;
+	nvpair_t *pair;
+	int error;
+
+	pair = nvlist_next_nvpair(holds, NULL);
+	if (pair == NULL)
+		return (0);
+
+	ddura.ddura_holds = holds;
+	ddura.ddura_errlist = errlist;
+	ddura.ddura_todelete = fnvlist_alloc();
+
+	error = dsl_sync_task(nvpair_name(pair), dsl_dataset_user_release_check,
+	    dsl_dataset_user_release_sync, &ddura, fnvlist_num_pairs(holds));
+	fnvlist_free(ddura.ddura_todelete);
+	return (error);
+}
+
+typedef struct dsl_dataset_user_release_tmp_arg {
+	uint64_t ddurta_dsobj;
+	nvlist_t *ddurta_holds;
+	boolean_t ddurta_deleteme;
+} dsl_dataset_user_release_tmp_arg_t;
+
+static int
+dsl_dataset_user_release_tmp_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_dataset_user_release_tmp_arg_t *ddurta = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	dsl_dataset_t *ds;
+	int error;
+
+	if (!dmu_tx_is_syncing(tx))
+		return (0);
+
+	error = dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds);
+	if (error)
+		return (error);
+
+	error = dsl_dataset_user_release_check_one(ds,
+	    ddurta->ddurta_holds, &ddurta->ddurta_deleteme);
+	dsl_dataset_rele(ds, FTAG);
+	return (error);
+}
+
+static void
+dsl_dataset_user_release_tmp_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_dataset_user_release_tmp_arg_t *ddurta = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	dsl_dataset_t *ds;
+
+	VERIFY0(dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds));
+	dsl_dataset_user_release_sync_one(ds, ddurta->ddurta_holds, tx);
+	if (ddurta->ddurta_deleteme) {
+		ASSERT(ds->ds_userrefs == 0 &&
+		    ds->ds_phys->ds_num_children == 1 &&
+		    DS_IS_DEFER_DESTROY(ds));
+		dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
+	}
+	dsl_dataset_rele(ds, FTAG);
+}
+
+/*
+ * Called at spa_load time to release a stale temporary user hold.
+ * Also called by the onexit code.
+ */
+void
+dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, const char *htag)
+{
+	dsl_dataset_user_release_tmp_arg_t ddurta;
+
+#ifdef _KERNEL
+	dsl_dataset_t *ds;
+	int error;
+
+	/* Make sure it is not mounted. */
+	dsl_pool_config_enter(dp, FTAG);
+	error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
+	if (error == 0) {
+		char name[MAXNAMELEN];
+		dsl_dataset_name(ds, name);
+		dsl_dataset_rele(ds, FTAG);
+		dsl_pool_config_exit(dp, FTAG);
+		zfs_unmount_snap(name);
+	} else {
+		dsl_pool_config_exit(dp, FTAG);
+	}
+#endif
+
+	ddurta.ddurta_dsobj = dsobj;
+	ddurta.ddurta_holds = fnvlist_alloc();
+	fnvlist_add_boolean(ddurta.ddurta_holds, htag);
+
+	(void) dsl_sync_task(spa_name(dp->dp_spa),
+	    dsl_dataset_user_release_tmp_check,
+	    dsl_dataset_user_release_tmp_sync, &ddurta, 1);
+	fnvlist_free(ddurta.ddurta_holds);
+}
+
+typedef struct zfs_hold_cleanup_arg {
+	char zhca_spaname[MAXNAMELEN];
+	uint64_t zhca_spa_load_guid;
+	uint64_t zhca_dsobj;
+	char zhca_htag[MAXNAMELEN];
+} zfs_hold_cleanup_arg_t;
+
+static void
+dsl_dataset_user_release_onexit(void *arg)
+{
+	zfs_hold_cleanup_arg_t *ca = arg;
+	spa_t *spa;
+	int error;
+
+	error = spa_open(ca->zhca_spaname, &spa, FTAG);
+	if (error != 0) {
+		zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s "
+		    "because pool is no longer loaded",
+		    ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag);
+		return;
+	}
+	if (spa_load_guid(spa) != ca->zhca_spa_load_guid) {
+		zfs_dbgmsg("couldn't release hold on pool=%s ds=%llu tag=%s "
+		    "because pool is no longer loaded (guid doesn't match)",
+		    ca->zhca_spaname, ca->zhca_dsobj, ca->zhca_htag);
+		spa_close(spa, FTAG);
+		return;
+	}
+
+	dsl_dataset_user_release_tmp(spa_get_dsl(spa),
+	    ca->zhca_dsobj, ca->zhca_htag);
+	kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
+	spa_close(spa, FTAG);
+}
+
+void
+dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
+    minor_t minor)
+{
+	zfs_hold_cleanup_arg_t *ca = kmem_alloc(sizeof (*ca), KM_SLEEP);
+	spa_t *spa = dsl_dataset_get_spa(ds);
+	(void) strlcpy(ca->zhca_spaname, spa_name(spa),
+	    sizeof (ca->zhca_spaname));
+	ca->zhca_spa_load_guid = spa_load_guid(spa);
+	ca->zhca_dsobj = ds->ds_object;
+	(void) strlcpy(ca->zhca_htag, htag, sizeof (ca->zhca_htag));
+	VERIFY0(zfs_onexit_add_cb(minor,
+	    dsl_dataset_user_release_onexit, ca, NULL));
+}
+
+int
+dsl_dataset_get_holds(const char *dsname, nvlist_t *nvl)
+{
+	dsl_pool_t *dp;
+	dsl_dataset_t *ds;
+	int err;
+
+	err = dsl_pool_hold(dsname, FTAG, &dp);
+	if (err != 0)
+		return (err);
+	err = dsl_dataset_hold(dp, dsname, FTAG, &ds);
+	if (err != 0) {
+		dsl_pool_rele(dp, FTAG);
+		return (err);
+	}
+
+	if (ds->ds_phys->ds_userrefs_obj != 0) {
+		zap_attribute_t *za;
+		zap_cursor_t zc;
+
+		za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
+		for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
+		    ds->ds_phys->ds_userrefs_obj);
+		    zap_cursor_retrieve(&zc, za) == 0;
+		    zap_cursor_advance(&zc)) {
+			fnvlist_add_uint64(nvl, za->za_name,
+			    za->za_first_integer);
+		}
+		zap_cursor_fini(&zc);
+		kmem_free(za, sizeof (zap_attribute_t));
+	}
+	dsl_dataset_rele(ds, FTAG);
+	dsl_pool_rele(dp, FTAG);
+	return (0);
+}

module/zfs/metaslab.c

@@ -1928,6 +1928,46 @@ void metaslab_fastwrite_unmark(spa_t *spa, const blkptr_t *bp)
 	spa_config_exit(spa, SCL_VDEV, FTAG);
 }
 
+static void
+checkmap(space_map_t *sm, uint64_t off, uint64_t size)
+{
+	space_seg_t *ss;
+	avl_index_t where;
+
+	mutex_enter(sm->sm_lock);
+	ss = space_map_find(sm, off, size, &where);
+	if (ss != NULL)
+		panic("freeing free block; ss=%p", (void *)ss);
+	mutex_exit(sm->sm_lock);
+}
+
+void
+metaslab_check_free(spa_t *spa, const blkptr_t *bp)
+{
+	int i, j;
+
+	if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0)
+		return;
+
+	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
+	for (i = 0; i < BP_GET_NDVAS(bp); i++) {
+		uint64_t vdid = DVA_GET_VDEV(&bp->blk_dva[i]);
+		vdev_t *vd = vdev_lookup_top(spa, vdid);
+		uint64_t off = DVA_GET_OFFSET(&bp->blk_dva[i]);
+		uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]);
+		metaslab_t *ms = vd->vdev_ms[off >> vd->vdev_ms_shift];
+
+		if (ms->ms_map->sm_loaded)
+			checkmap(ms->ms_map, off, size);
+
+		for (j = 0; j < TXG_SIZE; j++)
+			checkmap(ms->ms_freemap[j], off, size);
+		for (j = 0; j < TXG_DEFER_SIZE; j++)
+			checkmap(ms->ms_defermap[j], off, size);
+	}
+	spa_config_exit(spa, SCL_VDEV, FTAG);
+}
+
 #if defined(_KERNEL) && defined(HAVE_SPL)
 module_param(metaslab_debug, int, 0644);
 MODULE_PARM_DESC(metaslab_debug, "keep space maps in core to verify frees");

module/zfs/refcount.c

@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012 by Delphix. All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -32,7 +33,7 @@ int reference_tracking_enable = FALSE; /* runs out of memory too easily */
 #else
 int reference_tracking_enable = TRUE;
 #endif
-int reference_history = 4; /* tunable */
+int reference_history = 3; /* tunable */
 
 static kmem_cache_t *reference_cache;
 static kmem_cache_t *reference_history_cache;
@@ -64,6 +65,14 @@ refcount_create(refcount_t *rc)
 	    offsetof(reference_t, ref_link));
 	rc->rc_count = 0;
 	rc->rc_removed_count = 0;
+	rc->rc_tracked = reference_tracking_enable;
+}
+
+void
+refcount_create_untracked(refcount_t *rc)
+{
+	refcount_create(rc);
+	rc->rc_tracked = B_FALSE;
 }
 
 void
@@ -96,14 +105,12 @@ refcount_destroy(refcount_t *rc)
 int
 refcount_is_zero(refcount_t *rc)
 {
-	ASSERT(rc->rc_count >= 0);
 	return (rc->rc_count == 0);
 }
 
 int64_t
 refcount_count(refcount_t *rc)
 {
-	ASSERT(rc->rc_count >= 0);
 	return (rc->rc_count);
 }
 
@@ -113,14 +120,14 @@ refcount_add_many(refcount_t *rc, uint64_t number, void *holder)
 	reference_t *ref = NULL;
 	int64_t count;
 
-	if (reference_tracking_enable) {
+	if (rc->rc_tracked) {
 		ref = kmem_cache_alloc(reference_cache, KM_PUSHPAGE);
 		ref->ref_holder = holder;
 		ref->ref_number = number;
 	}
 	mutex_enter(&rc->rc_mtx);
 	ASSERT(rc->rc_count >= 0);
-	if (reference_tracking_enable)
+	if (rc->rc_tracked)
 		list_insert_head(&rc->rc_list, ref);
 	rc->rc_count += number;
 	count = rc->rc_count;
@@ -144,7 +151,7 @@ refcount_remove_many(refcount_t *rc, uint64_t number, void *holder)
 	mutex_enter(&rc->rc_mtx);
 	ASSERT(rc->rc_count >= number);
 
-	if (!reference_tracking_enable) {
+	if (!rc->rc_tracked) {
 		rc->rc_count -= number;
 		count = rc->rc_count;
 		mutex_exit(&rc->rc_mtx);
@@ -161,7 +168,7 @@ refcount_remove_many(refcount_t *rc, uint64_t number, void *holder)
 				    KM_PUSHPAGE);
 				list_insert_head(&rc->rc_removed, ref);
 				rc->rc_removed_count++;
-				if (rc->rc_removed_count >= reference_history) {
+				if (rc->rc_removed_count > reference_history) {
 					ref = list_tail(&rc->rc_removed);
 					list_remove(&rc->rc_removed, ref);
 					kmem_cache_free(reference_history_cache,

module/zfs/rrwlock.c

@@ -75,8 +75,9 @@
 uint_t rrw_tsd_key;
 
 typedef struct rrw_node {
-	struct rrw_node	*rn_next;
+	struct rrw_node *rn_next;
-	rrwlock_t	*rn_rrl;
+	rrwlock_t *rn_rrl;
+	void *rn_tag;
 } rrw_node_t;
 
 static rrw_node_t *
@@ -98,13 +99,14 @@ rrn_find(rrwlock_t *rrl)
  * Add a node to the head of the singly linked list.
  */
 static void
-rrn_add(rrwlock_t *rrl)
+rrn_add(rrwlock_t *rrl, void *tag)
 {
 	rrw_node_t *rn;
 
 	rn = kmem_alloc(sizeof (*rn), KM_SLEEP);
 	rn->rn_rrl = rrl;
 	rn->rn_next = tsd_get(rrw_tsd_key);
+	rn->rn_tag = tag;
 	VERIFY(tsd_set(rrw_tsd_key, rn) == 0);
 }
 
@@ -113,7 +115,7 @@ rrn_add(rrwlock_t *rrl)
  * thread's list and return TRUE; otherwise return FALSE.
  */
 static boolean_t
-rrn_find_and_remove(rrwlock_t *rrl)
+rrn_find_and_remove(rrwlock_t *rrl, void *tag)
 {
 	rrw_node_t *rn;
 	rrw_node_t *prev = NULL;
@@ -122,7 +124,7 @@ rrn_find_and_remove(rrwlock_t *rrl)
 		return (B_FALSE);
 
 	for (rn = tsd_get(rrw_tsd_key); rn != NULL; rn = rn->rn_next) {
-		if (rn->rn_rrl == rrl) {
+		if (rn->rn_rrl == rrl && rn->rn_tag == tag) {
 			if (prev)
 				prev->rn_next = rn->rn_next;
 			else
@@ -136,7 +138,7 @@ rrn_find_and_remove(rrwlock_t *rrl)
 }
 
 void
-rrw_init(rrwlock_t *rrl)
+rrw_init(rrwlock_t *rrl, boolean_t track_all)
 {
 	mutex_init(&rrl->rr_lock, NULL, MUTEX_DEFAULT, NULL);
 	cv_init(&rrl->rr_cv, NULL, CV_DEFAULT, NULL);
@@ -144,6 +146,7 @@ rrw_init(rrwlock_t *rrl)
 	refcount_create(&rrl->rr_anon_rcount);
 	refcount_create(&rrl->rr_linked_rcount);
 	rrl->rr_writer_wanted = B_FALSE;
+	rrl->rr_track_all = track_all;
 }
 
 void
@@ -156,12 +159,13 @@ rrw_destroy(rrwlock_t *rrl)
 	refcount_destroy(&rrl->rr_linked_rcount);
 }
 
-static void
+void
 rrw_enter_read(rrwlock_t *rrl, void *tag)
 {
 	mutex_enter(&rrl->rr_lock);
 #if !defined(DEBUG) && defined(_KERNEL)
-	if (!rrl->rr_writer && !rrl->rr_writer_wanted) {
+	if (rrl->rr_writer == NULL && !rrl->rr_writer_wanted &&
+	    !rrl->rr_track_all) {
 		rrl->rr_anon_rcount.rc_count++;
 		mutex_exit(&rrl->rr_lock);
 		return;
@@ -171,14 +175,14 @@ rrw_enter_read(rrwlock_t *rrl, void *tag)
 	ASSERT(rrl->rr_writer != curthread);
 	ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0);
 
-	while (rrl->rr_writer || (rrl->rr_writer_wanted &&
+	while (rrl->rr_writer != NULL || (rrl->rr_writer_wanted &&
 	    refcount_is_zero(&rrl->rr_anon_rcount) &&
 	    rrn_find(rrl) == NULL))
 		cv_wait(&rrl->rr_cv, &rrl->rr_lock);
 
-	if (rrl->rr_writer_wanted) {
+	if (rrl->rr_writer_wanted || rrl->rr_track_all) {
 		/* may or may not be a re-entrant enter */
-		rrn_add(rrl);
+		rrn_add(rrl, tag);
 		(void) refcount_add(&rrl->rr_linked_rcount, tag);
 	} else {
 		(void) refcount_add(&rrl->rr_anon_rcount, tag);
@@ -187,7 +191,7 @@ rrw_enter_read(rrwlock_t *rrl, void *tag)
 	mutex_exit(&rrl->rr_lock);
 }
 
-static void
+void
 rrw_enter_write(rrwlock_t *rrl)
 {
 	mutex_enter(&rrl->rr_lock);
@@ -233,10 +237,12 @@ rrw_exit(rrwlock_t *rrl, void *tag)
 
 	if (rrl->rr_writer == NULL) {
 		int64_t count;
-		if (rrn_find_and_remove(rrl))
+		if (rrn_find_and_remove(rrl, tag)) {
 			count = refcount_remove(&rrl->rr_linked_rcount, tag);
-		else
+		} else {
+			ASSERT(!rrl->rr_track_all);
 			count = refcount_remove(&rrl->rr_anon_rcount, tag);
+		}
 		if (count == 0)
 			cv_broadcast(&rrl->rr_cv);
 	} else {
@@ -249,6 +255,11 @@ rrw_exit(rrwlock_t *rrl, void *tag)
 	mutex_exit(&rrl->rr_lock);
 }
 
+/*
+ * If the lock was created with track_all, rrw_held(RW_READER) will return
+ * B_TRUE iff the current thread has the lock for reader.  Otherwise it may
+ * return B_TRUE if any thread has the lock for reader.
+ */
 boolean_t
 rrw_held(rrwlock_t *rrl, krw_t rw)
 {
@@ -259,7 +270,7 @@ rrw_held(rrwlock_t *rrl, krw_t rw)
 		held = (rrl->rr_writer == curthread);
 	} else {
 		held = (!refcount_is_zero(&rrl->rr_anon_rcount) ||
-		    !refcount_is_zero(&rrl->rr_linked_rcount));
+		    rrn_find(rrl) != NULL);
 	}
 	mutex_exit(&rrl->rr_lock);
 

module/zfs/sa.c

@@ -1019,10 +1019,10 @@ sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count,
 	sa_attr_type_t *tb;
 	int error;
 
-	mutex_enter(&os->os_lock);
+	mutex_enter(&os->os_user_ptr_lock);
 	if (os->os_sa) {
 		mutex_enter(&os->os_sa->sa_lock);
-		mutex_exit(&os->os_lock);
+		mutex_exit(&os->os_user_ptr_lock);
 		tb = os->os_sa->sa_user_table;
 		mutex_exit(&os->os_sa->sa_lock);
 		*user_table = tb;
@@ -1035,7 +1035,7 @@ sa_setup(objset_t *os, uint64_t sa_obj, sa_attr_reg_t *reg_attrs, int count,
 
 	os->os_sa = sa;
 	mutex_enter(&sa->sa_lock);
-	mutex_exit(&os->os_lock);
+	mutex_exit(&os->os_user_ptr_lock);
 	avl_create(&sa->sa_layout_num_tree, layout_num_compare,
 	    sizeof (sa_lot_t), offsetof(sa_lot_t, lot_num_node));
 	avl_create(&sa->sa_layout_hash_tree, layout_hash_compare,

module/zfs/spa.c

@@ -64,6 +64,7 @@
 #include <sys/zfs_ioctl.h>
 #include <sys/dsl_scan.h>
 #include <sys/zfeature.h>
+#include <sys/dsl_destroy.h>
 #include <sys/zvol.h>
 
 #ifdef	_KERNEL
@@ -131,10 +132,8 @@ const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
 };
 
-static dsl_syncfunc_t spa_sync_version;
+static void spa_sync_version(void *arg, dmu_tx_t *tx);
-static dsl_syncfunc_t spa_sync_props;
+static void spa_sync_props(void *arg, dmu_tx_t *tx);
-static dsl_checkfunc_t spa_change_guid_check;
-static dsl_syncfunc_t spa_change_guid_sync;
 static boolean_t spa_has_active_shared_spare(spa_t *spa);
 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
@@ -329,10 +328,10 @@ spa_prop_get(spa_t *spa, nvlist_t **nvp)
 				dsl_dataset_t *ds = NULL;
 
 				dp = spa_get_dsl(spa);
-				rw_enter(&dp->dp_config_rwlock, RW_READER);
+				dsl_pool_config_enter(dp, FTAG);
 				if ((err = dsl_dataset_hold_obj(dp,
 				    za.za_first_integer, FTAG, &ds))) {
-					rw_exit(&dp->dp_config_rwlock);
+					dsl_pool_config_exit(dp, FTAG);
 					break;
 				}
 
@@ -341,7 +340,7 @@ spa_prop_get(spa_t *spa, nvlist_t **nvp)
 				    KM_PUSHPAGE);
 				dsl_dataset_name(ds, strval);
 				dsl_dataset_rele(ds, FTAG);
-				rw_exit(&dp->dp_config_rwlock);
+				dsl_pool_config_exit(dp, FTAG);
 			} else {
 				strval = NULL;
 				intval = za.za_first_integer;
@@ -495,9 +494,10 @@ spa_prop_validate(spa_t *spa, nvlist_t *props)
 
 				if (dmu_objset_type(os) != DMU_OST_ZFS) {
 					error = ENOTSUP;
-				} else if ((error = dsl_prop_get_integer(strval,
+				} else if ((error =
+				    dsl_prop_get_int_ds(dmu_objset_ds(os),
 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
-				    &compress, NULL)) == 0 &&
+				    &compress)) == 0 &&
 				    !BOOTFS_COMPRESS_VALID(compress)) {
 					error = ENOTSUP;
 				} else {
@@ -661,8 +661,8 @@ spa_prop_set(spa_t *spa, nvlist_t *nvp)
 			 * read object, the features for write object, or the
 			 * feature descriptions object.
 			 */
-			error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
+			error = dsl_sync_task(spa->spa_name, NULL,
-			    spa_sync_version, spa, &ver, 6);
+			    spa_sync_version, &ver, 6);
 			if (error)
 				return (error);
 			continue;
@@ -673,8 +673,8 @@ spa_prop_set(spa_t *spa, nvlist_t *nvp)
 	}
 
 	if (need_sync) {
-		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
+		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
-		    spa, nvp, 6));
+		    nvp, 6));
 	}
 
 	return (0);
@@ -696,12 +696,12 @@ spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
 
 /*ARGSUSED*/
 static int
-spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
+spa_change_guid_check(void *arg, dmu_tx_t *tx)
 {
-	spa_t *spa = arg1;
+	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 	vdev_t *rvd = spa->spa_root_vdev;
 	uint64_t vdev_state;
-	ASSERTV(uint64_t *newguid = arg2);
+	ASSERTV(uint64_t *newguid = arg);
 
 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
 	vdev_state = rvd->vdev_state;
@@ -716,10 +716,10 @@ spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx)
 }
 
 static void
-spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+spa_change_guid_sync(void *arg, dmu_tx_t *tx)
 {
-	spa_t *spa = arg1;
+	uint64_t *newguid = arg;
-	uint64_t *newguid = arg2;
+	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 	uint64_t oldguid;
 	vdev_t *rvd = spa->spa_root_vdev;
 
@@ -753,8 +753,8 @@ spa_change_guid(spa_t *spa)
 	mutex_enter(&spa_namespace_lock);
 	guid = spa_generate_guid(NULL);
 
-	error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check,
+	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
-	    spa_change_guid_sync, spa, &guid, 5);
+	    spa_change_guid_sync, &guid, 5);
 
 	if (error == 0) {
 		spa_config_sync(spa, B_FALSE, B_TRUE);
@@ -1729,23 +1729,24 @@ spa_config_valid(spa_t *spa, nvlist_t *config)
 /*
  * Check for missing log devices
  */
-static int
+static boolean_t
 spa_check_logs(spa_t *spa)
 {
+	boolean_t rv = B_FALSE;
+
 	switch (spa->spa_log_state) {
 	default:
 		break;
 	case SPA_LOG_MISSING:
 		/* need to recheck in case slog has been restored */
 	case SPA_LOG_UNKNOWN:
-		if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
+		rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
-		    DS_FIND_CHILDREN)) {
+		    NULL, DS_FIND_CHILDREN) != 0);
+		if (rv)
 			spa_set_log_state(spa, SPA_LOG_MISSING);
-			return (1);
-		}
 		break;
 	}
-	return (0);
+	return (rv);
 }
 
 static boolean_t
@@ -1793,11 +1794,11 @@ spa_activate_log(spa_t *spa)
 int
 spa_offline_log(spa_t *spa)
 {
-	int error = 0;
+	int error;
-
-	if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
-	    NULL, DS_FIND_CHILDREN)) == 0) {
 
+	error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
+	    NULL, DS_FIND_CHILDREN);
+	if (error == 0) {
 		/*
 		 * We successfully offlined the log device, sync out the
 		 * current txg so that the "stubby" block can be removed
@@ -3610,7 +3611,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
 	if (props != NULL) {
 		spa_configfile_set(spa, props, B_FALSE);
-		spa_sync_props(spa, props, tx);
+		spa_sync_props(props, tx);
 	}
 
 	dmu_tx_commit(tx);
@@ -3844,7 +3845,7 @@ out:
  * Import a non-root pool into the system.
  */
 int
-spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
+spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
 {
 	spa_t *spa;
 	char *altroot = NULL;
@@ -5878,10 +5879,11 @@ spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
 }
 
 static void
-spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
+spa_sync_version(void *arg, dmu_tx_t *tx)
 {
-	spa_t *spa = arg1;
+	uint64_t *versionp = arg;
-	uint64_t version = *(uint64_t *)arg2;
+	uint64_t version = *versionp;
+	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 
 	/*
 	 * Setting the version is special cased when first creating the pool.
@@ -5900,11 +5902,11 @@ spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
  * Set zpool properties.
  */
 static void
-spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
+spa_sync_props(void *arg, dmu_tx_t *tx)
 {
-	spa_t *spa = arg1;
+	nvlist_t *nvp = arg;
+	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 	objset_t *mos = spa->spa_meta_objset;
-	nvlist_t *nvp = arg2;
 	nvpair_t *elem = NULL;
 
 	mutex_enter(&spa->spa_props_lock);
@@ -6056,6 +6058,8 @@ spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
 
 	ASSERT(spa->spa_sync_pass == 1);
 
+	rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
+
 	if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
 	    spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
 		dsl_pool_create_origin(dp, tx);
@@ -6081,6 +6085,7 @@ spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
 	    spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
 		spa_feature_create_zap_objects(spa, tx);
 	}
+	rrw_exit(&dp->dp_config_rwlock, FTAG);
 }
 
 /*

module/zfs/spa_history.c

@@ -197,10 +197,10 @@ spa_history_zone(void)
  */
 /*ARGSUSED*/
 static void
-spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+spa_history_log_sync(void *arg, dmu_tx_t *tx)
 {
-	spa_t		*spa = arg1;
+	nvlist_t	*nvl = arg;
-	nvlist_t	*nvl = arg2;
+	spa_t		*spa = dmu_tx_pool(tx)->dp_spa;
 	objset_t	*mos = spa->spa_meta_objset;
 	dmu_buf_t	*dbp;
 	spa_history_phys_t *shpp;
@@ -222,7 +222,7 @@ spa_history_log_sync(void *arg1, void *arg2, dmu_tx_t *tx)
 	 * Get the offset of where we need to write via the bonus buffer.
 	 * Update the offset when the write completes.
 	 */
-	VERIFY(0 == dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
+	VERIFY0(dmu_bonus_hold(mos, spa->spa_history, FTAG, &dbp));
 	shpp = dbp->db_data;
 
 	dmu_buf_will_dirty(dbp, tx);
@@ -326,8 +326,8 @@ spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
 	fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
 
 	/* Kick this off asynchronously; errors are ignored. */
-	dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
+	dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
-	    spa_history_log_sync, spa, nvarg, 0, tx);
+	    nvarg, 0, tx);
 	dmu_tx_commit(tx);
 
 	/* spa_history_log_sync will free nvl */
@@ -465,10 +465,10 @@ log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
 	fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
 
 	if (dmu_tx_is_syncing(tx)) {
-		spa_history_log_sync(spa, nvl, tx);
+		spa_history_log_sync(nvl, tx);
 	} else {
-		dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
+		dsl_sync_task_nowait(spa_get_dsl(spa),
-		    spa_history_log_sync, spa, nvl, 0, tx);
+		    spa_history_log_sync, nvl, 0, tx);
 	}
 	/* spa_history_log_sync() will free nvl */
 }
@@ -544,17 +544,11 @@ spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
 void
 spa_history_log_version(spa_t *spa, const char *operation)
 {
-#ifdef _KERNEL
-	uint64_t current_vers = spa_version(spa);
-
 	spa_history_log_internal(spa, operation, NULL,
 	    "pool version %llu; software version %llu/%d; uts %s %s %s %s",
-	    (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
+	    (u_longlong_t)spa_version(spa), SPA_VERSION, ZPL_VERSION,
 	    utsname.nodename, utsname.release, utsname.version,
 	    utsname.machine);
-	cmn_err(CE_CONT, "!%s version %llu pool %s using %llu", operation,
-	    (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
-#endif
 }
 
 #if defined(_KERNEL) && defined(HAVE_SPL)

module/zfs/spa_misc.c

@@ -268,7 +268,7 @@ spa_config_lock_init(spa_t *spa)
 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
 		mutex_init(&scl->scl_lock, NULL, MUTEX_DEFAULT, NULL);
 		cv_init(&scl->scl_cv, NULL, CV_DEFAULT, NULL);
-		refcount_create(&scl->scl_count);
+		refcount_create_untracked(&scl->scl_count);
 		scl->scl_writer = NULL;
 		scl->scl_write_wanted = 0;
 	}
@@ -326,6 +326,8 @@ spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
 	int wlocks_held = 0;
 	int i;
 
+	ASSERT3U(SCL_LOCKS, <, sizeof (wlocks_held) * NBBY);
+
 	for (i = 0; i < SCL_LOCKS; i++) {
 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
 		if (scl->scl_writer == curthread)
@@ -406,27 +408,22 @@ spa_lookup(const char *name)
 	static spa_t search;	/* spa_t is large; don't allocate on stack */
 	spa_t *spa;
 	avl_index_t where;
-	char c = 0;
 	char *cp;
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
 
+	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
+
 	/*
 	 * If it's a full dataset name, figure out the pool name and
 	 * just use that.
 	 */
-	cp = strpbrk(name, "/@");
+	cp = strpbrk(search.spa_name, "/@");
-	if (cp) {
+	if (cp != NULL)
-		c = *cp;
 		*cp = '\0';
-	}
 
-	(void) strlcpy(search.spa_name, name, sizeof (search.spa_name));
 	spa = avl_find(&spa_namespace_avl, &search, &where);
 
-	if (cp)
-		*cp = c;
-
 	return (spa);
 }
 
@@ -539,6 +536,8 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 		    KM_SLEEP) == 0);
 	}
 
+	spa->spa_debug = ((zfs_flags & ZFS_DEBUG_SPA) != 0);
+
 	return (spa);
 }
 

module/zfs/space_map.c

@@ -102,7 +102,7 @@ void
 space_map_add(space_map_t *sm, uint64_t start, uint64_t size)
 {
 	avl_index_t where;
-	space_seg_t ssearch, *ss_before, *ss_after, *ss;
+	space_seg_t *ss_before, *ss_after, *ss;
 	uint64_t end = start + size;
 	int merge_before, merge_after;
 
@@ -115,11 +115,8 @@ space_map_add(space_map_t *sm, uint64_t start, uint64_t size)
 	VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
 	VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
 
-	ssearch.ss_start = start;
+	ss = space_map_find(sm, start, size, &where);
-	ssearch.ss_end = end;
+	if (ss != NULL) {
-	ss = avl_find(&sm->sm_root, &ssearch, &where);
-
-	if (ss != NULL && ss->ss_start <= start && ss->ss_end >= end) {
 		zfs_panic_recover("zfs: allocating allocated segment"
 		    "(offset=%llu size=%llu)\n",
 		    (longlong_t)start, (longlong_t)size);
@@ -171,19 +168,12 @@ void
 space_map_remove(space_map_t *sm, uint64_t start, uint64_t size)
 {
 	avl_index_t where;
-	space_seg_t ssearch, *ss, *newseg;
+	space_seg_t *ss, *newseg;
 	uint64_t end = start + size;
 	int left_over, right_over;
 
-	ASSERT(MUTEX_HELD(sm->sm_lock));
 	VERIFY(!sm->sm_condensing);
-	VERIFY(size != 0);
+	ss = space_map_find(sm, start, size, &where);
-	VERIFY(P2PHASE(start, 1ULL << sm->sm_shift) == 0);
-	VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
-
-	ssearch.ss_start = start;
-	ssearch.ss_end = end;
-	ss = avl_find(&sm->sm_root, &ssearch, &where);
 
 	/* Make sure we completely overlap with someone */
 	if (ss == NULL) {
@@ -226,12 +216,11 @@ space_map_remove(space_map_t *sm, uint64_t start, uint64_t size)
 	sm->sm_space -= size;
 }
 
-boolean_t
+space_seg_t *
-space_map_contains(space_map_t *sm, uint64_t start, uint64_t size)
+space_map_find(space_map_t *sm, uint64_t start, uint64_t size,
+    avl_index_t *wherep)
 {
-	avl_index_t where;
 	space_seg_t ssearch, *ss;
-	uint64_t end = start + size;
 
 	ASSERT(MUTEX_HELD(sm->sm_lock));
 	VERIFY(size != 0);
@@ -239,10 +228,20 @@ space_map_contains(space_map_t *sm, uint64_t start, uint64_t size)
 	VERIFY(P2PHASE(size, 1ULL << sm->sm_shift) == 0);
 
 	ssearch.ss_start = start;
-	ssearch.ss_end = end;
+	ssearch.ss_end = start + size;
-	ss = avl_find(&sm->sm_root, &ssearch, &where);
+	ss = avl_find(&sm->sm_root, &ssearch, wherep);
 
-	return (ss != NULL && ss->ss_start <= start && ss->ss_end >= end);
+	if (ss != NULL && ss->ss_start <= start && ss->ss_end >= start + size)
+		return (ss);
+	return (NULL);
+}
+
+boolean_t
+space_map_contains(space_map_t *sm, uint64_t start, uint64_t size)
+{
+	avl_index_t where;
+
+	return (space_map_find(sm, start, size, &where) != 0);
 }
 
 void

module/zfs/txg.c

@@ -659,6 +659,8 @@ txg_wait_synced(dsl_pool_t *dp, uint64_t txg)
 {
 	tx_state_t *tx = &dp->dp_tx;
 
+	ASSERT(!dsl_pool_config_held(dp));
+
 	mutex_enter(&tx->tx_sync_lock);
 	ASSERT(tx->tx_threads == 2);
 	if (txg == 0)
@@ -682,6 +684,8 @@ txg_wait_open(dsl_pool_t *dp, uint64_t txg)
 {
 	tx_state_t *tx = &dp->dp_tx;
 
+	ASSERT(!dsl_pool_config_held(dp));
+
 	mutex_enter(&tx->tx_sync_lock);
 	ASSERT(tx->tx_threads == 2);
 	if (txg == 0)
@@ -747,42 +751,43 @@ txg_list_empty(txg_list_t *tl, uint64_t txg)
 }
 
 /*
- * Add an entry to the list.
+ * Add an entry to the list (unless it's already on the list).
- * Returns 0 if it's a new entry, 1 if it's already there.
+ * Returns B_TRUE if it was actually added.
  */
-int
+boolean_t
 txg_list_add(txg_list_t *tl, void *p, uint64_t txg)
 {
 	int t = txg & TXG_MASK;
 	txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset);
-	int already_on_list;
+	boolean_t add;
 
 	mutex_enter(&tl->tl_lock);
-	already_on_list = tn->tn_member[t];
+	add = (tn->tn_member[t] == 0);
-	if (!already_on_list) {
+	if (add) {
 		tn->tn_member[t] = 1;
 		tn->tn_next[t] = tl->tl_head[t];
 		tl->tl_head[t] = tn;
 	}
 	mutex_exit(&tl->tl_lock);
 
-	return (already_on_list);
+	return (add);
 }
 
 /*
- * Add an entry to the end of the list (walks list to find end).
+ * Add an entry to the end of the list, unless it's already on the list.
- * Returns 0 if it's a new entry, 1 if it's already there.
+ * (walks list to find end)
+ * Returns B_TRUE if it was actually added.
  */
-int
+boolean_t
 txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg)
 {
 	int t = txg & TXG_MASK;
 	txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset);
-	int already_on_list;
+	boolean_t add;
 
 	mutex_enter(&tl->tl_lock);
-	already_on_list = tn->tn_member[t];
+	add = (tn->tn_member[t] == 0);
-	if (!already_on_list) {
+	if (add) {
 		txg_node_t **tp;
 
 		for (tp = &tl->tl_head[t]; *tp != NULL; tp = &(*tp)->tn_next[t])
@@ -794,7 +799,7 @@ txg_list_add_tail(txg_list_t *tl, void *p, uint64_t txg)
 	}
 	mutex_exit(&tl->tl_lock);
 
-	return (already_on_list);
+	return (add);
 }
 
 /*
@@ -845,13 +850,13 @@ txg_list_remove_this(txg_list_t *tl, void *p, uint64_t txg)
 	return (NULL);
 }
 
-int
+boolean_t
 txg_list_member(txg_list_t *tl, void *p, uint64_t txg)
 {
 	int t = txg & TXG_MASK;
 	txg_node_t *tn = (txg_node_t *)((char *)p + tl->tl_offset);
 
-	return (tn->tn_member[t]);
+	return (tn->tn_member[t] != 0);
 }
 
 /*

module/zfs/zfs_ctldir.c

@@ -80,6 +80,7 @@
 #include <sys/zfs_vnops.h>
 #include <sys/stat.h>
 #include <sys/dmu.h>
+#include <sys/dsl_destroy.h>
 #include <sys/dsl_deleg.h>
 #include <sys/mount.h>
 #include <sys/zpl.h>
@@ -488,13 +489,13 @@ zfsctl_rename_snap(zfs_sb_t *zsb, zfs_snapentry_t *sep, const char *name)
  */
 /*ARGSUSED*/
 int
-zfsctl_snapdir_rename(struct inode *sdip, char *sname,
+zfsctl_snapdir_rename(struct inode *sdip, char *snm,
-    struct inode *tdip, char *tname, cred_t *cr, int flags)
+    struct inode *tdip, char *tnm, cred_t *cr, int flags)
 {
 	zfs_sb_t *zsb = ITOZSB(sdip);
 	zfs_snapentry_t search, *sep;
 	avl_index_t where;
-	char *to, *from, *real;
+	char *to, *from, *real, *fsname;
 	int error;
 
 	ZFS_ENTER(zsb);
@@ -502,23 +503,26 @@ zfsctl_snapdir_rename(struct inode *sdip, char *sname,
 	to = kmem_alloc(MAXNAMELEN, KM_SLEEP);
 	from = kmem_alloc(MAXNAMELEN, KM_SLEEP);
 	real = kmem_alloc(MAXNAMELEN, KM_SLEEP);
+	fsname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
 
 	if (zsb->z_case == ZFS_CASE_INSENSITIVE) {
-		error = dmu_snapshot_realname(zsb->z_os, sname, real,
+		error = dmu_snapshot_realname(zsb->z_os, snm, real,
 		    MAXNAMELEN, NULL);
 		if (error == 0) {
-			sname = real;
+			snm = real;
 		} else if (error != ENOTSUP) {
 			goto out;
 		}
 	}
 
-	error = zfsctl_snapshot_zname(sdip, sname, MAXNAMELEN, from);
+	dmu_objset_name(zsb->z_os, fsname);
-	if (!error)
+
-		error = zfsctl_snapshot_zname(tdip, tname, MAXNAMELEN, to);
+	error = zfsctl_snapshot_zname(sdip, snm, MAXNAMELEN, from);
-	if (!error)
+	if (error == 0)
+		error = zfsctl_snapshot_zname(tdip, tnm, MAXNAMELEN, to);
+	if (error == 0)
 		error = zfs_secpolicy_rename_perms(from, to, cr);
-	if (error)
+	if (error != 0)
 		goto out;
 
 	/*
@@ -532,21 +536,21 @@ zfsctl_snapdir_rename(struct inode *sdip, char *sname,
 	/*
 	 * No-op when names are identical.
 	 */
-	if (strcmp(sname, tname) == 0) {
+	if (strcmp(snm, tnm) == 0) {
 		error = 0;
 		goto out;
 	}
 
 	mutex_enter(&zsb->z_ctldir_lock);
 
-	error = dmu_objset_rename(from, to, B_FALSE);
+	error = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE);
 	if (error)
 		goto out_unlock;
 
-	search.se_name = (char *)sname;
+	search.se_name = (char *)snm;
 	sep = avl_find(&zsb->z_ctldir_snaps, &search, &where);
 	if (sep)
-		zfsctl_rename_snap(zsb, sep, tname);
+		zfsctl_rename_snap(zsb, sep, tnm);
 
 out_unlock:
 	mutex_exit(&zsb->z_ctldir_lock);
@@ -554,6 +558,7 @@ out:
 	kmem_free(from, MAXNAMELEN);
 	kmem_free(to, MAXNAMELEN);
 	kmem_free(real, MAXNAMELEN);
+	kmem_free(fsname, MAXNAMELEN);
 
 	ZFS_EXIT(zsb);
 
@@ -588,14 +593,14 @@ zfsctl_snapdir_remove(struct inode *dip, char *name, cred_t *cr, int flags)
 	}
 
 	error = zfsctl_snapshot_zname(dip, name, MAXNAMELEN, snapname);
-	if (!error)
+	if (error == 0)
 		error = zfs_secpolicy_destroy_perms(snapname, cr);
-	if (error)
+	if (error != 0)
 		goto out;
 
 	error = zfsctl_unmount_snapshot(zsb, name, MNT_FORCE);
 	if ((error == 0) || (error == ENOENT))
-		error = dmu_objset_destroy(snapname, B_FALSE);
+		error = dsl_destroy_snapshot(snapname, B_FALSE);
 out:
 	kmem_free(snapname, MAXNAMELEN);
 	kmem_free(real, MAXNAMELEN);
@@ -628,12 +633,12 @@ zfsctl_snapdir_mkdir(struct inode *dip, char *dirname, vattr_t *vap,
 	dmu_objset_name(zsb->z_os, dsname);
 
 	error = zfs_secpolicy_snapshot_perms(dsname, cr);
-	if (error)
+	if (error != 0)
 		goto out;
 
 	if (error == 0) {
 		error = dmu_objset_snapshot_one(dsname, dirname);
-		if (error)
+		if (error != 0)
 			goto out;
 
 		error = zfsctl_snapdir_lookup(dip, dirname, ipp,

module/zfs/zfs_vfsops.c

@@ -248,28 +248,31 @@ zfs_register_callbacks(zfs_sb_t *zsb)
 	 * overboard...
 	 */
 	ds = dmu_objset_ds(os);
+	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
 	error = dsl_prop_register(ds,
-	    "atime", atime_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "xattr", xattr_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "recordsize", blksz_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "readonly", readonly_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "devices", devices_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "setuid", setuid_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "exec", exec_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "snapdir", snapdir_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "aclinherit", acl_inherit_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
+	    zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "vscan", vscan_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zsb);
 	error = error ? error : dsl_prop_register(ds,
-	    "nbmand", nbmand_changed_cb, zsb);
+	    zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zsb);
+	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
 	if (error)
 		goto unregister;
 
@@ -284,18 +287,28 @@ unregister:
 	 * registered, but this is OK; it will simply return ENOMSG,
 	 * which we will ignore.
 	 */
-	(void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
-	(void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zsb);
+	    atime_changed_cb, zsb);
-	(void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
-	(void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zsb);
+	    xattr_changed_cb, zsb);
-	(void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
-	(void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zsb);
+	    blksz_changed_cb, zsb);
-	(void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
-	(void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zsb);
+	    readonly_changed_cb, zsb);
-	(void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
-	    zsb);
+	    devices_changed_cb, zsb);
-	(void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
-	(void) dsl_prop_unregister(ds, "nbmand", nbmand_changed_cb, zsb);
+	    setuid_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
+	    exec_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
+	    snapdir_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
+	    acl_inherit_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
+	    vscan_changed_cb, zsb);
+	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_NBMAND),
+	    nbmand_changed_cb, zsb);
 
 	return (error);
 }
@@ -305,8 +318,6 @@ static int
 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
     uint64_t *userp, uint64_t *groupp)
 {
-	int error = 0;
-
 	/*
 	 * Is it a valid type of object to track?
 	 */
@@ -363,7 +374,7 @@ zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
 			*groupp = BSWAP_64(*groupp);
 		}
 	}
-	return (error);
+	return (0);
 }
 
 static void
@@ -726,7 +737,7 @@ zfs_sb_create(const char *osname, zfs_sb_t **zsbp)
 	mutex_init(&zsb->z_lock, NULL, MUTEX_DEFAULT, NULL);
 	list_create(&zsb->z_all_znodes, sizeof (znode_t),
 	    offsetof(znode_t, z_link_node));
-	rrw_init(&zsb->z_teardown_lock);
+	rrw_init(&zsb->z_teardown_lock, B_FALSE);
 	rw_init(&zsb->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
 	rw_init(&zsb->z_fuid_lock, NULL, RW_DEFAULT, NULL);
 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
@@ -1138,7 +1149,7 @@ zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting)
 	if (dsl_dataset_is_dirty(dmu_objset_ds(zsb->z_os)) &&
 	    !zfs_is_readonly(zsb))
 		txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
-	(void) dmu_objset_evict_dbufs(zsb->z_os);
+	dmu_objset_evict_dbufs(zsb->z_os);
 
 	return (0);
 }

module/zfs/zil.c

@@ -257,7 +257,7 @@ zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, blkptr_t *nbp, void *dst,
 			}
 		}
 
-		VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
+		VERIFY(arc_buf_remove_ref(abuf, &abuf));
 	}
 
 	return (error);
@@ -356,7 +356,7 @@ zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func,
 			break;
 
 		error = zil_read_log_block(zilog, &blk, &next_blk, lrbuf, &end);
-		if (error)
+		if (error != 0)
 			break;
 
 		for (lrp = lrbuf; lrp < end; lrp += reclen) {
@@ -492,7 +492,7 @@ zilog_dirty(zilog_t *zilog, uint64_t txg)
 	if (dsl_dataset_is_snapshot(ds))
 		panic("dirtying snapshot!");
 
-	if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg) == 0) {
+	if (txg_list_add(&dp->dp_dirty_zilogs, zilog, txg)) {
 		/* up the hold count until we can be written out */
 		dmu_buf_add_ref(ds->ds_dbuf, zilog);
 	}
@@ -658,8 +658,8 @@ zil_claim(const char *osname, void *txarg)
 	objset_t *os;
 	int error;
 
-	error = dmu_objset_hold(osname, FTAG, &os);
+	error = dmu_objset_own(osname, DMU_OST_ANY, B_FALSE, FTAG, &os);
-	if (error) {
+	if (error != 0) {
 		cmn_err(CE_WARN, "can't open objset for %s", osname);
 		return (0);
 	}
@@ -672,7 +672,7 @@ zil_claim(const char *osname, void *txarg)
 			zio_free_zil(zilog->zl_spa, first_txg, &zh->zh_log);
 		BP_ZERO(&zh->zh_log);
 		dsl_dataset_dirty(dmu_objset_ds(os), tx);
-		dmu_objset_rele(os, FTAG);
+		dmu_objset_disown(os, FTAG);
 		return (0);
 	}
 
@@ -697,7 +697,7 @@ zil_claim(const char *osname, void *txarg)
 	}
 
 	ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
-	dmu_objset_rele(os, FTAG);
+	dmu_objset_disown(os, FTAG);
 	return (0);
 }
 
@@ -717,7 +717,7 @@ zil_check_log_chain(const char *osname, void *tx)
 	ASSERT(tx == NULL);
 
 	error = dmu_objset_hold(osname, FTAG, &os);
-	if (error) {
+	if (error != 0) {
 		cmn_err(CE_WARN, "can't open objset for %s", osname);
 		return (0);
 	}
@@ -1014,7 +1014,8 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
 
 	BP_ZERO(bp);
 	use_slog = USE_SLOG(zilog);
-	error = zio_alloc_zil(spa, txg, bp, zil_blksz, USE_SLOG(zilog));
+	error = zio_alloc_zil(spa, txg, bp, zil_blksz,
+	    USE_SLOG(zilog));
 	if (use_slog)
 	{
 		ZIL_STAT_BUMP(zil_itx_metaslab_slog_count);
@@ -1025,7 +1026,7 @@ zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
 		ZIL_STAT_BUMP(zil_itx_metaslab_normal_count);
 		ZIL_STAT_INCR(zil_itx_metaslab_normal_bytes, lwb->lwb_nused);
 	}
-	if (!error) {
+	if (error == 0) {
 		ASSERT3U(bp->blk_birth, ==, txg);
 		bp->blk_cksum = lwb->lwb_blk.blk_cksum;
 		bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++;
@@ -1145,7 +1146,7 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 				txg_wait_synced(zilog->zl_dmu_pool, txg);
 				return (lwb);
 			}
-			if (error) {
+			if (error != 0) {
 				ASSERT(error == ENOENT || error == EEXIST ||
 				    error == EALREADY);
 				return (lwb);
@@ -1807,6 +1808,9 @@ zil_free(zilog_t *zilog)
 
 	zilog->zl_stop_sync = 1;
 
+	ASSERT0(zilog->zl_suspend);
+	ASSERT0(zilog->zl_suspending);
+
 	ASSERT(list_is_empty(&zilog->zl_lwb_list));
 	list_destroy(&zilog->zl_lwb_list);
 
@@ -1905,32 +1909,100 @@ zil_close(zilog_t *zilog)
 	mutex_exit(&zilog->zl_lock);
 }
 
+static char *suspend_tag = "zil suspending";
+
 /*
  * Suspend an intent log.  While in suspended mode, we still honor
  * synchronous semantics, but we rely on txg_wait_synced() to do it.
- * We suspend the log briefly when taking a snapshot so that the snapshot
+ * On old version pools, we suspend the log briefly when taking a
- * contains all the data it's supposed to, and has an empty intent log.
+ * snapshot so that it will have an empty intent log.
+ *
+ * Long holds are not really intended to be used the way we do here --
+ * held for such a short time.  A concurrent caller of dsl_dataset_long_held()
+ * could fail.  Therefore we take pains to only put a long hold if it is
+ * actually necessary.  Fortunately, it will only be necessary if the
+ * objset is currently mounted (or the ZVOL equivalent).  In that case it
+ * will already have a long hold, so we are not really making things any worse.
+ *
+ * Ideally, we would locate the existing long-holder (i.e. the zfsvfs_t or
+ * zvol_state_t), and use their mechanism to prevent their hold from being
+ * dropped (e.g. VFS_HOLD()).  However, that would be even more pain for
+ * very little gain.
+ *
+ * if cookiep == NULL, this does both the suspend & resume.
+ * Otherwise, it returns with the dataset "long held", and the cookie
+ * should be passed into zil_resume().
  */
 int
-zil_suspend(zilog_t *zilog)
+zil_suspend(const char *osname, void **cookiep)
 {
-	const zil_header_t *zh = zilog->zl_header;
+	objset_t *os;
+	zilog_t *zilog;
+	const zil_header_t *zh;
+	int error;
+
+	error = dmu_objset_hold(osname, suspend_tag, &os);
+	if (error != 0)
+		return (error);
+	zilog = dmu_objset_zil(os);
 
 	mutex_enter(&zilog->zl_lock);
+	zh = zilog->zl_header;
+
 	if (zh->zh_flags & ZIL_REPLAY_NEEDED) {		/* unplayed log */
 		mutex_exit(&zilog->zl_lock);
+		dmu_objset_rele(os, suspend_tag);
 		return (EBUSY);
 	}
-	if (zilog->zl_suspend++ != 0) {
+
+	/*
+	 * Don't put a long hold in the cases where we can avoid it.  This
+	 * is when there is no cookie so we are doing a suspend & resume
+	 * (i.e. called from zil_vdev_offline()), and there's nothing to do
+	 * for the suspend because it's already suspended, or there's no ZIL.
+	 */
+	if (cookiep == NULL && !zilog->zl_suspending &&
+	    (zilog->zl_suspend > 0 || BP_IS_HOLE(&zh->zh_log))) {
+		mutex_exit(&zilog->zl_lock);
+		dmu_objset_rele(os, suspend_tag);
+		return (0);
+	}
+
+	dsl_dataset_long_hold(dmu_objset_ds(os), suspend_tag);
+	dsl_pool_rele(dmu_objset_pool(os), suspend_tag);
+
+	zilog->zl_suspend++;
+
+	if (zilog->zl_suspend > 1) {
 		/*
-		 * Someone else already began a suspend.
+		 * Someone else is already suspending it.
 		 * Just wait for them to finish.
 		 */
+
 		while (zilog->zl_suspending)
 			cv_wait(&zilog->zl_cv_suspend, &zilog->zl_lock);
 		mutex_exit(&zilog->zl_lock);
+
+		if (cookiep == NULL)
+			zil_resume(os);
+		else
+			*cookiep = os;
 		return (0);
 	}
+
+	/*
+	 * If there is no pointer to an on-disk block, this ZIL must not
+	 * be active (e.g. filesystem not mounted), so there's nothing
+	 * to clean up.
+	 */
+	if (BP_IS_HOLE(&zh->zh_log)) {
+		ASSERT(cookiep != NULL); /* fast path already handled */
+
+		*cookiep = os;
+		mutex_exit(&zilog->zl_lock);
+		return (0);
+	}
+
 	zilog->zl_suspending = B_TRUE;
 	mutex_exit(&zilog->zl_lock);
 
@@ -1943,16 +2015,25 @@ zil_suspend(zilog_t *zilog)
 	cv_broadcast(&zilog->zl_cv_suspend);
 	mutex_exit(&zilog->zl_lock);
 
+	if (cookiep == NULL)
+		zil_resume(os);
+	else
+		*cookiep = os;
 	return (0);
 }
 
 void
-zil_resume(zilog_t *zilog)
+zil_resume(void *cookie)
 {
+	objset_t *os = cookie;
+	zilog_t *zilog = dmu_objset_zil(os);
+
 	mutex_enter(&zilog->zl_lock);
 	ASSERT(zilog->zl_suspend != 0);
 	zilog->zl_suspend--;
 	mutex_exit(&zilog->zl_lock);
+	dsl_dataset_long_rele(dmu_objset_ds(os), suspend_tag);
+	dsl_dataset_rele(dmu_objset_ds(os), suspend_tag);
 }
 
 typedef struct zil_replay_arg {
@@ -2025,7 +2106,7 @@ zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
 	if (txtype == TX_WRITE && reclen == sizeof (lr_write_t)) {
 		error = zil_read_log_data(zilog, (lr_write_t *)lr,
 		    zr->zr_lr + reclen);
-		if (error)
+		if (error != 0)
 			return (zil_replay_error(zilog, lr, error));
 	}
 
@@ -2046,7 +2127,7 @@ zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
 	 * is updated if we are in replay mode.
 	 */
 	error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, zr->zr_byteswap);
-	if (error) {
+	if (error != 0) {
 		/*
 		 * The DMU's dnode layer doesn't see removes until the txg
 		 * commits, so a subsequent claim can spuriously fail with
@@ -2056,7 +2137,7 @@ zil_replay_log_record(zilog_t *zilog, lr_t *lr, void *zra, uint64_t claim_txg)
 		 */
 		txg_wait_synced(spa_get_dsl(zilog->zl_spa), 0);
 		error = zr->zr_replay[txtype](zr->zr_arg, zr->zr_lr, B_FALSE);
-		if (error)
+		if (error != 0)
 			return (zil_replay_error(zilog, lr, error));
 	}
 	return (0);
@@ -2128,21 +2209,12 @@ zil_replaying(zilog_t *zilog, dmu_tx_t *tx)
 int
 zil_vdev_offline(const char *osname, void *arg)
 {
-	objset_t *os;
-	zilog_t *zilog;
 	int error;
 
-	error = dmu_objset_hold(osname, FTAG, &os);
+	error = zil_suspend(osname, NULL);
-	if (error)
+	if (error != 0)
-		return (error);
+		return (EEXIST);
-
+	return (0);
-	zilog = dmu_objset_zil(os);
-	if (zil_suspend(zilog) != 0)
-		error = EEXIST;
-	else
-		zil_resume(zilog);
-	dmu_objset_rele(os, FTAG);
-	return (error);
 }
 
 #if defined(_KERNEL) && defined(HAVE_SPL)

module/zfs/zio.c

@@ -767,6 +767,7 @@ zio_write_override(zio_t *zio, blkptr_t *bp, int copies)
 void
 zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp)
 {
+	metaslab_check_free(spa, bp);
 	bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp);
 }
 
@@ -785,6 +786,8 @@ zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
 
 	arc_freed(spa, bp);
 
+	metaslab_check_free(spa, bp);
+
 	zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp),
 	    NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags,
 	    NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE);
@@ -2060,7 +2063,7 @@ zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde)
 				    bcmp(abuf->b_data, zio->io_orig_data,
 				    zio->io_orig_size) != 0)
 					error = EEXIST;
-				VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
+				VERIFY(arc_buf_remove_ref(abuf, &abuf));
 			}
 
 			ddt_enter(ddt);
@@ -2656,8 +2659,9 @@ zio_vdev_io_assess(zio_t *zio)
 	 * set vdev_cant_write so that we stop trying to allocate from it.
 	 */
 	if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE &&
-	    vd != NULL && !vd->vdev_ops->vdev_op_leaf)
+	    vd != NULL && !vd->vdev_ops->vdev_op_leaf) {
 		vd->vdev_cant_write = B_TRUE;
+	}
 
 	if (zio->io_error)
 		zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;

module/zfs/zvol.c

@@ -315,6 +315,13 @@ zvol_set_volsize(const char *name, uint64_t volsize)
 	uint64_t readonly;
 	int error;
 
+	error = dsl_prop_get_integer(name,
+	    zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
+	if (error != 0)
+		return (error);
+	if (readonly)
+		return (EROFS);
+
 	mutex_enter(&zvol_state_lock);
 
 	zv = zvol_find_by_name(name);
@@ -1459,8 +1466,7 @@ zvol_remove_minor(const char *name)
 }
 
 static int
-zvol_create_minors_cb(spa_t *spa, uint64_t dsobj,
+zvol_create_minors_cb(const char *dsname, void *arg)
-		      const char *dsname, void *arg)
 {
 	if (strchr(dsname, '/') == NULL)
 		return 0;
@@ -1474,7 +1480,7 @@ zvol_create_minors_cb(spa_t *spa, uint64_t dsobj,
  * for all available pools.
  */
 int
-zvol_create_minors(const char *pool)
+zvol_create_minors(char *pool)
 {
 	spa_t *spa = NULL;
 	int error = 0;
@@ -1484,13 +1490,12 @@ zvol_create_minors(const char *pool)
 
 	mutex_enter(&zvol_state_lock);
 	if (pool) {
-		error = dmu_objset_find_spa(NULL, pool, zvol_create_minors_cb,
+		error = dmu_objset_find(pool, zvol_create_minors_cb,
 		    NULL, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
 	} else {
 		mutex_enter(&spa_namespace_lock);
 		while ((spa = spa_next(spa)) != NULL) {
-			error = dmu_objset_find_spa(NULL,
+			error = dmu_objset_find(spa_name(spa), zvol_create_minors_cb, NULL,
-			    spa_name(spa), zvol_create_minors_cb, NULL,
 			    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
 			if (error)
 				break;

module/zpios/pios.c

@@ -34,6 +34,7 @@
 #include <sys/zfs_context.h>
 #include <sys/dmu.h>
 #include <sys/txg.h>
+#include <sys/dsl_destroy.h>
 #include <linux/cdev.h>
 #include "zpios-internal.h"
 
@@ -224,9 +225,9 @@ zpios_dmu_setup(run_args_t *run_args)
 	run_args->os = os;
 out_destroy:
 	if (rc) {
-		rc2 = dmu_objset_destroy(name, B_FALSE);
+		rc2 = dsl_destroy_head(name);
 		if (rc2)
-			zpios_print(run_args->file, "Error dmu_objset_destroy"
+			zpios_print(run_args->file, "Error dsl_destroy_head"
 				    "(%s, ...) failed: %d\n", name, rc2);
 	}
 out:
@@ -395,9 +396,9 @@ zpios_remove_objset(run_args_t *run_args)
 	dmu_objset_disown(run_args->os, zpios_tag);
 
 	if (run_args->flags & DMU_REMOVE) {
-		rc = dmu_objset_destroy(name, B_FALSE);
+		rc = dsl_destroy_head(name);
 		if (rc)
-			zpios_print(run_args->file, "Error dmu_objset_destroy"
+			zpios_print(run_args->file, "Error dsl_destroy_head"
 				    "(%s, ...) failed: %d\n", name, rc);
 	}