Merge commit 'refs/top-bases/linux-kernel-mem' into linux-kernel-mem

ZFS.RELEASE

@@ -1 +1 @@
-http://dlc.sun.com/osol/on/downloads/b117/on-src.tar.bz2
+http://dlc.sun.com/osol/on/downloads/b121/on-src.tar.bz2

cmd/zdb/zdb.c

@@ -818,6 +818,8 @@ dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
 	    (u_longlong_t)ds->ds_snapnames_zapobj);
 	(void) printf("\t\tnum_children = %llu\n",
 	    (u_longlong_t)ds->ds_num_children);
+	(void) printf("\t\tuserrefs_obj = %llu\n",
+	    (u_longlong_t)ds->ds_userrefs_obj);
 	(void) printf("\t\tcreation_time = %s", ctime(&crtime));
 	(void) printf("\t\tcreation_txg = %llu\n",
 	    (u_longlong_t)ds->ds_creation_txg);
@@ -1050,6 +1052,7 @@ static object_viewer_t *object_viewer[DMU_OT_NUMTYPES] = {
 	dump_zap,		/* DSL scrub queue		*/
 	dump_zap,		/* ZFS user/group used		*/
 	dump_zap,		/* ZFS user/group quota		*/
+	dump_zap,		/* snapshot refcount tags	*/
 };
 
 static void

cmd/zfs/zfs_iter.c

@@ -362,7 +362,7 @@ zfs_for_each(int argc, char **argv, int flags, zfs_type_t types,
 	cb.cb_types = types;
 	cb.cb_depth_limit = limit;
 	/*
-	 * If cb_proplist is provided then in the zfs_handles created  we
+	 * If cb_proplist is provided then in the zfs_handles created we
 	 * retain only those properties listed in cb_proplist and sortcol.
 	 * The rest are pruned. So, the caller should make sure that no other
 	 * properties other than those listed in cb_proplist/sortcol are

cmd/zfs/zfs_main.c

@@ -80,6 +80,8 @@ static int zfs_do_receive(int argc, char **argv);
 static int zfs_do_promote(int argc, char **argv);
 static int zfs_do_userspace(int argc, char **argv);
 static int zfs_do_python(int argc, char **argv);
+static int zfs_do_hold(int argc, char **argv);
+static int zfs_do_release(int argc, char **argv);
 
 /*
  * Enable a reasonable set of defaults for libumem debugging on DEBUG builds.
@@ -121,7 +123,10 @@ typedef enum {
 	HELP_ALLOW,
 	HELP_UNALLOW,
 	HELP_USERSPACE,
-	HELP_GROUPSPACE
+	HELP_GROUPSPACE,
+	HELP_HOLD,
+	HELP_HOLDS,
+	HELP_RELEASE
 } zfs_help_t;
 
 typedef struct zfs_command {
@@ -169,6 +174,10 @@ static zfs_command_t command_table[] = {
 	{ "allow",	zfs_do_python,		HELP_ALLOW		},
 	{ NULL },
 	{ "unallow",	zfs_do_python,		HELP_UNALLOW		},
+	{ NULL },
+	{ "hold",	zfs_do_hold,		HELP_HOLD		},
+	{ "holds",	zfs_do_python,		HELP_HOLDS		},
+	{ "release",	zfs_do_release,		HELP_RELEASE		},
 };
 
 #define	NCOMMAND	(sizeof (command_table) / sizeof (command_table[0]))
@@ -189,7 +198,8 @@ get_usage(zfs_help_t idx)
 		    "-V <size> <volume>\n"));
 	case HELP_DESTROY:
 		return (gettext("\tdestroy [-rRf] "
-		    "<filesystem|volume|snapshot>\n"));
+		    "<filesystem|volume|snapshot>\n"
+		    "\tdestroy -d [-r] <filesystem|volume|snapshot>\n"));
 	case HELP_GET:
 		return (gettext("\tget [-rHp] [-d max] "
 		    "[-o field[,...]] [-s source[,...]]\n"
@@ -236,7 +246,7 @@ get_usage(zfs_help_t idx)
 		return (gettext("\tunmount [-f] "
 		    "<-a | filesystem|mountpoint>\n"));
 	case HELP_UNSHARE:
-		return (gettext("\tunshare [-f] "
+		return (gettext("\tunshare "
 		    "<-a | filesystem|mountpoint>\n"));
 	case HELP_ALLOW:
 		return (gettext("\tallow <filesystem|volume>\n"
@@ -266,6 +276,12 @@ get_usage(zfs_help_t idx)
 		return (gettext("\tgroupspace [-hniHpU] [-o field[,...]] "
 		    "[-sS field] ... [-t type[,...]]\n"
 		    "\t    <filesystem|snapshot>\n"));
+	case HELP_HOLD:
+		return (gettext("\thold [-r] <tag> <snapshot> ...\n"));
+	case HELP_HOLDS:
+		return (gettext("\tholds [-r] <snapshot> ...\n"));
+	case HELP_RELEASE:
+		return (gettext("\trelease [-r] <tag> <snapshot> ...\n"));
 	}
 
 	abort();
@@ -769,11 +785,13 @@ badusage:
 }
 
 /*
- * zfs destroy [-rf] <fs, snap, vol>
+ * zfs destroy [-rRf] <fs, snap, vol>
+ * zfs destroy -d [-r] <fs, snap, vol>
  *
  * 	-r	Recursively destroy all children
  * 	-R	Recursively destroy all dependents, including clones
  * 	-f	Force unmounting of any dependents
+ *	-d	If we can't destroy now, mark for deferred destruction
  *
  * Destroys the given dataset.  By default, it will unmount any filesystems,
  * and refuse to destroy a dataset that has any dependents.  A dependent can
@@ -789,6 +807,7 @@ typedef struct destroy_cbdata {
 	boolean_t	cb_closezhp;
 	zfs_handle_t	*cb_target;
 	char		*cb_snapname;
+	boolean_t	cb_defer_destroy;
 } destroy_cbdata_t;
 
 /*
@@ -869,7 +888,7 @@ destroy_callback(zfs_handle_t *zhp, void *data)
 	 * Bail out on the first error.
 	 */
 	if (zfs_unmount(zhp, NULL, cbp->cb_force ? MS_FORCE : 0) != 0 ||
-	    zfs_destroy(zhp) != 0) {
+	    zfs_destroy(zhp, cbp->cb_defer_destroy) != 0) {
 		zfs_close(zhp);
 		return (-1);
 	}
@@ -923,8 +942,11 @@ zfs_do_destroy(int argc, char **argv)
 	char *cp;
 
 	/* check options */
-	while ((c = getopt(argc, argv, "frR")) != -1) {
+	while ((c = getopt(argc, argv, "dfrR")) != -1) {
 		switch (c) {
+		case 'd':
+			cb.cb_defer_destroy = B_TRUE;
+			break;
 		case 'f':
 			cb.cb_force = 1;
 			break;
@@ -956,6 +978,9 @@ zfs_do_destroy(int argc, char **argv)
 		usage(B_FALSE);
 	}
 
+	if (cb.cb_defer_destroy && cb.cb_doclones)
+		usage(B_FALSE);
+
 	/*
 	 * If we are doing recursive destroy of a snapshot, then the
 	 * named snapshot may not exist.  Go straight to libzfs.
@@ -977,7 +1002,7 @@ zfs_do_destroy(int argc, char **argv)
 			}
 		}
 
-		ret = zfs_destroy_snaps(zhp, cp);
+		ret = zfs_destroy_snaps(zhp, cp, cb.cb_defer_destroy);
 		zfs_close(zhp);
 		if (ret) {
 			(void) fprintf(stderr,
@@ -986,7 +1011,6 @@ zfs_do_destroy(int argc, char **argv)
 		return (ret != 0);
 	}
 
-
 	/* Open the given dataset */
 	if ((zhp = zfs_open(g_zfs, argv[0], ZFS_TYPE_DATASET)) == NULL)
 		return (1);
@@ -1014,15 +1038,15 @@ zfs_do_destroy(int argc, char **argv)
 	 * Check for any dependents and/or clones.
 	 */
 	cb.cb_first = B_TRUE;
-	if (!cb.cb_doclones &&
+	if (!cb.cb_doclones && !cb.cb_defer_destroy &&
 	    zfs_iter_dependents(zhp, B_TRUE, destroy_check_dependent,
 	    &cb) != 0) {
 		zfs_close(zhp);
 		return (1);
 	}
 
-	if (cb.cb_error ||
+	if (cb.cb_error || (!cb.cb_defer_destroy &&
-	    zfs_iter_dependents(zhp, B_FALSE, destroy_callback, &cb) != 0) {
+	    (zfs_iter_dependents(zhp, B_FALSE, destroy_callback, &cb) != 0))) {
 		zfs_close(zhp);
 		return (1);
 	}
@@ -1035,7 +1059,6 @@ zfs_do_destroy(int argc, char **argv)
 	if (destroy_callback(zhp, &cb) != 0)
 		return (1);
 
-
 	return (0);
 }
 
@@ -1612,7 +1635,7 @@ zfs_do_upgrade(int argc, char **argv)
 		(void) printf(gettext(" 1   Initial ZFS filesystem version\n"));
 		(void) printf(gettext(" 2   Enhanced directory entries\n"));
 		(void) printf(gettext(" 3   Case insensitive and File system "
-		    "unique identifer (FUID)\n"));
+		    "unique identifier (FUID)\n"));
 		(void) printf(gettext(" 4   userquota, groupquota "
 		    "properties\n"));
 		(void) printf(gettext("\nFor more information on a particular "
@@ -2650,6 +2673,108 @@ zfs_do_receive(int argc, char **argv)
 	return (err != 0);
 }
 
+static int
+zfs_do_hold_rele_impl(int argc, char **argv, boolean_t holding)
+{
+	int errors = 0;
+	int i;
+	const char *tag;
+	boolean_t recursive = B_FALSE;
+	int c;
+	int (*func)(zfs_handle_t *, const char *, const char *, boolean_t);
+
+	/* check options */
+	while ((c = getopt(argc, argv, "r")) != -1) {
+		switch (c) {
+		case 'r':
+			recursive = B_TRUE;
+			break;
+		case '?':
+			(void) fprintf(stderr, gettext("invalid option '%c'\n"),
+			    optopt);
+			usage(B_FALSE);
+		}
+	}
+
+	argc -= optind;
+	argv += optind;
+
+	/* check number of arguments */
+	if (argc < 2)
+		usage(B_FALSE);
+
+	tag = argv[0];
+	--argc;
+	++argv;
+
+	if (holding) {
+		if (tag[0] == '.') {
+			/* tags starting with '.' are reserved for libzfs */
+			(void) fprintf(stderr,
+			    gettext("tag may not start with '.'\n"));
+			usage(B_FALSE);
+		}
+		func = zfs_hold;
+	} else {
+		func = zfs_release;
+	}
+
+	for (i = 0; i < argc; ++i) {
+		zfs_handle_t *zhp;
+		char parent[ZFS_MAXNAMELEN];
+		const char *delim;
+		char *path = argv[i];
+
+		delim = strchr(path, '@');
+		if (delim == NULL) {
+			(void) fprintf(stderr,
+			    gettext("'%s' is not a snapshot\n"), path);
+			++errors;
+			continue;
+		}
+		(void) strncpy(parent, path, delim - path);
+		parent[delim - path] = '\0';
+
+		zhp = zfs_open(g_zfs, parent,
+		    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME);
+		if (zhp == NULL) {
+			++errors;
+			continue;
+		}
+		if (func(zhp, delim+1, tag, recursive) != 0)
+			++errors;
+		zfs_close(zhp);
+	}
+
+	return (errors != 0);
+}
+
+/*
+ * zfs hold [-r] <tag> <snap> ...
+ *
+ * 	-r	Recursively hold
+ *
+ * Apply a user-hold with the given tag to the list of snapshots.
+ */
+static int
+zfs_do_hold(int argc, char **argv)
+{
+	return (zfs_do_hold_rele_impl(argc, argv, B_TRUE));
+}
+
+/*
+ * zfs release [-r] <tag> <snap> ...
+ *
+ * 	-r	Recursively release
+ *
+ * Release a user-hold with the given tag from the list of snapshots.
+ */
+static int
+zfs_do_release(int argc, char **argv)
+{
+	return (zfs_do_hold_rele_impl(argc, argv, B_FALSE));
+}
+
 typedef struct get_all_cbdata {
 	zfs_handle_t	**cb_handles;
 	size_t		cb_alloc;

cmd/zpool/zpool_main.c

@@ -1469,7 +1469,7 @@ show_import(nvlist_t *config)
  */
 static int
 do_import(nvlist_t *config, const char *newname, const char *mntopts,
-    int force, nvlist_t *props, boolean_t allowfaulted)
+    int force, nvlist_t *props, boolean_t do_verbatim)
 {
 	zpool_handle_t *zhp;
 	char *name;
@@ -1522,14 +1522,14 @@ do_import(nvlist_t *config, const char *newname, const char *mntopts,
 		}
 	}
 
-	if (zpool_import_props(g_zfs, config, newname, props,
+	if (zpool_import_props(g_zfs, config, newname, props, do_verbatim) != 0)
-	    allowfaulted) != 0)
 		return (1);
 
 	if (newname != NULL)
 		name = (char *)newname;
 
-	verify((zhp = zpool_open_canfail(g_zfs, name)) != NULL);
+	if ((zhp = zpool_open_canfail(g_zfs, name)) == NULL)
+		return (1);
 
 	if (zpool_get_state(zhp) != POOL_STATE_UNAVAIL &&
 	    zpool_enable_datasets(zhp, mntopts, 0) != 0) {
@@ -1566,7 +1566,8 @@ do_import(nvlist_t *config, const char *newname, const char *mntopts,
  *       -F	Import even in the presence of faulted vdevs.  This is an
  *       	intentionally undocumented option for testing purposes, and
  *       	treats the pool configuration as complete, leaving any bad
- *		vdevs in the FAULTED state.
+ *		vdevs in the FAULTED state. In other words, it does verbatim
+ *		import.
  *
  *       -a	Import all pools found.
  *
@@ -1595,7 +1596,7 @@ zpool_do_import(int argc, char **argv)
 	nvlist_t *found_config;
 	nvlist_t *props = NULL;
 	boolean_t first;
-	boolean_t allow_faulted = B_FALSE;
+	boolean_t do_verbatim = B_FALSE;
 	uint64_t pool_state;
 	char *cachefile = NULL;
 
@@ -1628,7 +1629,7 @@ zpool_do_import(int argc, char **argv)
 			do_force = B_TRUE;
 			break;
 		case 'F':
-			allow_faulted = B_TRUE;
+			do_verbatim = B_TRUE;
 			break;
 		case 'o':
 			if ((propval = strchr(optarg, '=')) != NULL) {
@@ -1778,7 +1779,7 @@ zpool_do_import(int argc, char **argv)
 
 			if (do_all)
 				err |= do_import(config, NULL, mntopts,
-				    do_force, props, allow_faulted);
+				    do_force, props, do_verbatim);
 			else
 				show_import(config);
 		} else if (searchname != NULL) {
@@ -1826,7 +1827,7 @@ zpool_do_import(int argc, char **argv)
 			err = B_TRUE;
 		} else {
 			err |= do_import(found_config, argc == 1 ? NULL :
-			    argv[1], mntopts, do_force, props, allow_faulted);
+			    argv[1], mntopts, do_force, props, do_verbatim);
 		}
 	}
 
@@ -3090,6 +3091,17 @@ status_callback(zpool_handle_t *zhp, void *data)
 		    "replace'.\n"));
 		break;
 
+	case ZPOOL_STATUS_REMOVED_DEV:
+		(void) printf(gettext("status: One or more devices has "
+		    "been removed by the administrator.\n\tSufficient "
+		    "replicas exist for the pool to continue functioning in "
+		    "a\n\tdegraded state.\n"));
+		(void) printf(gettext("action: Online the device using "
+		    "'zpool online' or replace the device with\n\t'zpool "
+		    "replace'.\n"));
+		break;
+
+
 	case ZPOOL_STATUS_RESILVERING:
 		(void) printf(gettext("status: One or more devices is "
 		    "currently being resilvered.  The pool will\n\tcontinue "
@@ -3512,6 +3524,8 @@ zpool_do_upgrade(int argc, char **argv)
 		(void) printf(gettext(" 14  passthrough-x aclinherit\n"));
 		(void) printf(gettext(" 15  user/group space accounting\n"));
 		(void) printf(gettext(" 16  stmf property support\n"));
+		(void) printf(gettext(" 17  Triple-parity RAID-Z\n"));
+		(void) printf(gettext(" 18  snapshot user holds\n"));
 		(void) printf(gettext("For more information on a particular "
 		    "version, including supported releases, see:\n\n"));
 		(void) printf("http://www.opensolaris.org/os/community/zfs/"
@@ -3597,6 +3611,8 @@ char *hist_event_table[LOG_END] = {
 	"refquota set",
 	"refreservation set",
 	"pool scrub done",
+	"user hold",
+	"user release",
 };
 
 /*

cmd/zpool/zpool_vdev.c

@@ -20,7 +20,7 @@
  */
 
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -67,6 +67,7 @@
 #include <libdiskmgt.h>
 #include <libintl.h>
 #include <libnvpair.h>
+#include <limits.h>
 #include <stdio.h>
 #include <string.h>
 #include <unistd.h>
@@ -1093,19 +1094,34 @@ check_in_use(nvlist_t *config, nvlist_t *nv, int force, int isreplacing,
 }
 
 static const char *
-is_grouping(const char *type, int *mindev)
+is_grouping(const char *type, int *mindev, int *maxdev)
 {
-	if (strcmp(type, "raidz") == 0 || strcmp(type, "raidz1") == 0) {
+	if (strncmp(type, "raidz", 5) == 0) {
+		const char *p = type + 5;
+		char *end;
+		long nparity;
+
+		if (*p == '\0') {
+			nparity = 1;
+		} else if (*p == '0') {
+			return (NULL); /* no zero prefixes allowed */
+		} else {
+			errno = 0;
+			nparity = strtol(p, &end, 10);
+			if (errno != 0 || nparity < 1 || nparity >= 255 ||
+			    *end != '\0')
+				return (NULL);
+		}
+
 		if (mindev != NULL)
-			*mindev = 2;
+			*mindev = nparity + 1;
+		if (maxdev != NULL)
+			*maxdev = 255;
 		return (VDEV_TYPE_RAIDZ);
 	}
 
-	if (strcmp(type, "raidz2") == 0) {
+	if (maxdev != NULL)
-		if (mindev != NULL)
+		*maxdev = INT_MAX;
-			*mindev = 3;
-		return (VDEV_TYPE_RAIDZ);
-	}
 
 	if (strcmp(type, "mirror") == 0) {
 		if (mindev != NULL)
@@ -1144,7 +1160,7 @@ nvlist_t *
 construct_spec(int argc, char **argv)
 {
 	nvlist_t *nvroot, *nv, **top, **spares, **l2cache;
-	int t, toplevels, mindev, nspares, nlogs, nl2cache;
+	int t, toplevels, mindev, maxdev, nspares, nlogs, nl2cache;
 	const char *type;
 	uint64_t is_log;
 	boolean_t seen_logs;
@@ -1166,7 +1182,7 @@ construct_spec(int argc, char **argv)
 		 * If it's a mirror or raidz, the subsequent arguments are
 		 * its leaves -- until we encounter the next mirror or raidz.
 		 */
-		if ((type = is_grouping(argv[0], &mindev)) != NULL) {
+		if ((type = is_grouping(argv[0], &mindev, &maxdev)) != NULL) {
 			nvlist_t **child = NULL;
 			int c, children = 0;
 
@@ -1223,7 +1239,7 @@ construct_spec(int argc, char **argv)
 			}
 
 			for (c = 1; c < argc; c++) {
-				if (is_grouping(argv[c], NULL) != NULL)
+				if (is_grouping(argv[c], NULL, NULL) != NULL)
 					break;
 				children++;
 				child = realloc(child,
@@ -1243,6 +1259,13 @@ construct_spec(int argc, char **argv)
 				return (NULL);
 			}
 
+			if (children > maxdev) {
+				(void) fprintf(stderr, gettext("invalid vdev "
+				    "specification: %s supports no more than "
+				    "%d devices\n"), argv[0], maxdev);
+				return (NULL);
+			}
+
 			argc -= c;
 			argv += c;
 

cmd/ztest/ztest.c

@@ -493,7 +493,7 @@ process_options(int argc, char **argv)
 			zopt_raidz = MAX(1, value);
 			break;
 		case 'R':
-			zopt_raidz_parity = MIN(MAX(value, 1), 2);
+			zopt_raidz_parity = MIN(MAX(value, 1), 3);
 			break;
 		case 'd':
 			zopt_datasets = MAX(1, value);
@@ -1404,7 +1404,7 @@ ztest_destroy_cb(char *name, void *arg)
 	/*
 	 * Destroy the dataset.
 	 */
-	error = dmu_objset_destroy(name);
+	error = dmu_objset_destroy(name, B_FALSE);
 	if (error) {
 		(void) dmu_objset_open(name, DMU_OST_OTHER,
 		    DS_MODE_USER | DS_MODE_READONLY, &os);
@@ -1577,7 +1577,7 @@ ztest_dmu_objset_create_destroy(ztest_args_t *za)
 	zil_close(zilog);
 	dmu_objset_close(os);
 
-	error = dmu_objset_destroy(name);
+	error = dmu_objset_destroy(name, B_FALSE);
 	if (error)
 		fatal(0, "dmu_objset_destroy(%s) = %d", name, error);
 
@@ -1600,7 +1600,7 @@ ztest_dmu_snapshot_create_destroy(ztest_args_t *za)
 	(void) snprintf(snapname, 100, "%s@%llu", osname,
 	    (u_longlong_t)za->za_instance);
 
-	error = dmu_objset_destroy(snapname);
+	error = dmu_objset_destroy(snapname, B_FALSE);
 	if (error != 0 && error != ENOENT)
 		fatal(0, "dmu_objset_destroy() = %d", error);
 	error = dmu_objset_snapshot(osname, strchr(snapname, '@')+1,
@@ -1636,19 +1636,19 @@ ztest_dsl_dataset_cleanup(char *osname, uint64_t curval)
 	(void) snprintf(snap3name, 100, "%s@s3_%llu",
 			clone1name, (u_longlong_t)curval);
 
-	error = dmu_objset_destroy(clone2name);
+	error = dmu_objset_destroy(clone2name, B_FALSE);
 	if (error && error != ENOENT)
 		fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
-	error = dmu_objset_destroy(snap3name);
+	error = dmu_objset_destroy(snap3name, B_FALSE);
 	if (error && error != ENOENT)
 		fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
-	error = dmu_objset_destroy(snap2name);
+	error = dmu_objset_destroy(snap2name, B_FALSE);
 	if (error && error != ENOENT)
 		fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
-	error = dmu_objset_destroy(clone1name);
+	error = dmu_objset_destroy(clone1name, B_FALSE);
 	if (error && error != ENOENT)
 		fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
-	error = dmu_objset_destroy(snap1name);
+	error = dmu_objset_destroy(snap1name, B_FALSE);
 	if (error && error != ENOENT)
 		fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
 }

lib/libzfs/include/libzfs.h

@@ -117,6 +117,8 @@ enum {
 	EZFS_NOTSUP,		/* ops not supported on this dataset */
 	EZFS_ACTIVE_SPARE,	/* pool has active shared spare devices */
 	EZFS_UNPLAYED_LOGS,	/* log device has unplayed logs */
+	EZFS_REFTAG_RELE,	/* snapshot release: tag not found */
+	EZFS_REFTAG_HOLD,	/* snapshot hold: tag already exists */
 	EZFS_UNKNOWN
 };
 
@@ -286,6 +288,7 @@ typedef enum {
 	ZPOOL_STATUS_VERSION_OLDER,	/* older on-disk version */
 	ZPOOL_STATUS_RESILVERING,	/* device being resilvered */
 	ZPOOL_STATUS_OFFLINE_DEV,	/* device online */
+	ZPOOL_STATUS_REMOVED_DEV,	/* removed device */
 
 	/*
 	 * Finally, the following indicates a healthy pool.
@@ -454,8 +457,8 @@ extern int zfs_iter_snapshots(zfs_handle_t *, zfs_iter_f, void *);
 extern int zfs_create(libzfs_handle_t *, const char *, zfs_type_t,
     nvlist_t *);
 extern int zfs_create_ancestors(libzfs_handle_t *, const char *);
-extern int zfs_destroy(zfs_handle_t *);
+extern int zfs_destroy(zfs_handle_t *, boolean_t);
-extern int zfs_destroy_snaps(zfs_handle_t *, char *);
+extern int zfs_destroy_snaps(zfs_handle_t *, char *, 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_rollback(zfs_handle_t *, zfs_handle_t *, boolean_t);
@@ -463,6 +466,8 @@ extern int zfs_rename(zfs_handle_t *, const char *, boolean_t);
 extern int zfs_send(zfs_handle_t *, const char *, const char *,
     boolean_t, boolean_t, boolean_t, boolean_t, int);
 extern int zfs_promote(zfs_handle_t *);
+extern int zfs_hold(zfs_handle_t *, const char *, const char *, boolean_t);
+extern int zfs_release(zfs_handle_t *, const char *, const char *, boolean_t);
 
 typedef int (*zfs_userspace_cb_t)(void *arg, const char *domain,
     uid_t rid, uint64_t space);

lib/libzfs/libzfs_changelist.c

@@ -512,6 +512,14 @@ change_one(zfs_handle_t *zhp, void *data)
 			    &idx);
 			uu_list_insert(clp->cl_list, cn, idx);
 		} else {
+			/*
+			 * Add this child to beginning of the list. Children
+			 * below this one in the hierarchy will get added above
+			 * this one in the list. This produces a list in
+			 * reverse dataset name order.
+			 * This is necessary when the original mountpoint
+			 * is legacy or none.
+			 */
 			ASSERT(!clp->cl_alldependents);
 			verify(uu_list_insert_before(clp->cl_list,
 			    uu_list_first(clp->cl_list), cn) == 0);
@@ -578,6 +586,7 @@ changelist_gather(zfs_handle_t *zhp, zfs_prop_t prop, int gather_flags,
 	zfs_handle_t *temp;
 	char property[ZFS_MAXPROPLEN];
 	uu_compare_fn_t *compare = NULL;
+	boolean_t legacy = B_FALSE;
 
 	if ((clp = zfs_alloc(zhp->zfs_hdl, sizeof (prop_changelist_t))) == NULL)
 		return (NULL);
@@ -590,8 +599,19 @@ changelist_gather(zfs_handle_t *zhp, zfs_prop_t prop, int gather_flags,
 	if (prop == ZFS_PROP_NAME || prop == ZFS_PROP_ZONED ||
 	    prop == ZFS_PROP_MOUNTPOINT || prop == ZFS_PROP_SHARENFS ||
 	    prop == ZFS_PROP_SHARESMB) {
-		compare = compare_mountpoints;
+
-		clp->cl_sorted = B_TRUE;
+		if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT,
+		    property, sizeof (property),
+		    NULL, NULL, 0, B_FALSE) == 0 &&
+		    (strcmp(property, "legacy") == 0 ||
+		    strcmp(property, "none") == 0)) {
+
+			legacy = B_TRUE;
+		}
+		if (!legacy) {
+			compare = compare_mountpoints;
+			clp->cl_sorted = B_TRUE;
+		}
 	}
 
 	clp->cl_pool = uu_list_pool_create("changelist_pool",
@@ -699,6 +719,12 @@ changelist_gather(zfs_handle_t *zhp, zfs_prop_t prop, int gather_flags,
 		(void) uu_list_find(clp->cl_list, cn, NULL, &idx);
 		uu_list_insert(clp->cl_list, cn, idx);
 	} else {
+		/*
+		 * Add the target dataset to the end of the list.
+		 * The list is not really unsorted. The list will be
+		 * in reverse dataset name order. This is necessary
+		 * when the original mountpoint is legacy or none.
+		 */
 		verify(uu_list_insert_after(clp->cl_list,
 		    uu_list_last(clp->cl_list), cn) == 0);
 	}
@@ -707,11 +733,7 @@ changelist_gather(zfs_handle_t *zhp, zfs_prop_t prop, int gather_flags,
 	 * If the mountpoint property was previously 'legacy', or 'none',
 	 * record it as the behavior of changelist_postfix() will be different.
 	 */
-	if ((clp->cl_prop == ZFS_PROP_MOUNTPOINT) &&
+	if ((clp->cl_prop == ZFS_PROP_MOUNTPOINT) && legacy) {
-	    (zfs_prop_get(zhp, prop, property, sizeof (property),
-	    NULL, NULL, 0, B_FALSE) == 0 &&
-	    (strcmp(property, "legacy") == 0 ||
-	    strcmp(property, "none") == 0))) {
 		/*
 		 * do not automatically mount ex-legacy datasets if
 		 * we specifically set canmount to noauto

lib/libzfs/libzfs_dataset.c

@@ -47,6 +47,7 @@
 #include <ucred.h>
 #include <idmap.h>
 #include <aclutils.h>
+#include <directory.h>
 
 #include <sys/spa.h>
 #include <sys/zap.h>
@@ -1642,21 +1643,13 @@ get_numeric_property(zfs_handle_t *zhp, zfs_prop_t prop, zprop_source_t *src,
 		(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 		if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_OBJSET_ZPLPROPS, &zc)) {
 			zcmd_free_nvlists(&zc);
-			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
+			return (-1);
-			    "unable to get %s property"),
-			    zfs_prop_to_name(prop));
-			return (zfs_error(zhp->zfs_hdl, EZFS_BADVERSION,
-			    dgettext(TEXT_DOMAIN, "internal error")));
 		}
 		if (zcmd_read_dst_nvlist(zhp->zfs_hdl, &zc, &zplprops) != 0 ||
 		    nvlist_lookup_uint64(zplprops, zfs_prop_to_name(prop),
 		    val) != 0) {
 			zcmd_free_nvlists(&zc);
-			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
+			return (-1);
-			    "unable to get %s property"),
-			    zfs_prop_to_name(prop));
-			return (zfs_error(zhp->zfs_hdl, EZFS_NOMEM,
-			    dgettext(TEXT_DOMAIN, "internal error")));
 		}
 		if (zplprops)
 			nvlist_free(zplprops);
@@ -2042,6 +2035,7 @@ userquota_propname_decode(const char *propname, boolean_t zoned,
 {
 	zfs_userquota_prop_t type;
 	char *cp, *end;
+	char *numericsid = NULL;
 	boolean_t isuser;
 
 	domain[0] = '\0';
@@ -2064,33 +2058,41 @@ userquota_propname_decode(const char *propname, boolean_t zoned,
 	if (strchr(cp, '@')) {
 		/*
 		 * It's a SID name (eg "user@domain") that needs to be
-		 * turned into S-1-domainID-RID.  There should be a
+		 * turned into S-1-domainID-RID.
-		 * better way to do this, but for now just translate it
-		 * to the (possibly ephemeral) uid and then back to the
-		 * SID.  This is like getsidname(noresolve=TRUE).
 		 */
-		uid_t id;
+		directory_error_t e;
-		idmap_rid_t rid;
-		char *mapdomain;
-
 		if (zoned && getzoneid() == GLOBAL_ZONEID)
 			return (ENOENT);
-		if (sid_to_id(cp, isuser, &id) != 0)
+		if (isuser) {
+			e = directory_sid_from_user_name(NULL,
+			    cp, &numericsid);
+		} else {
+			e = directory_sid_from_group_name(NULL,
+			    cp, &numericsid);
+		}
+		if (e != NULL) {
+			directory_error_free(e);
 			return (ENOENT);
-		if (idmap_id_to_numeric_domain_rid(id, isuser,
+		}
-		    &mapdomain, &rid) != 0)
+		if (numericsid == NULL)
 			return (ENOENT);
-		(void) strlcpy(domain, mapdomain, domainlen);
+		cp = numericsid;
-		*ridp = rid;
+		/* will be further decoded below */
-	} else if (strncmp(cp, "S-1-", 4) == 0) {
+	}
+
+	if (strncmp(cp, "S-1-", 4) == 0) {
 		/* It's a numeric SID (eg "S-1-234-567-89") */
-		(void) strcpy(domain, cp);
+		(void) strlcpy(domain, cp, domainlen);
 		cp = strrchr(domain, '-');
 		*cp = '\0';
 		cp++;
 
 		errno = 0;
 		*ridp = strtoull(cp, &end, 10);
+		if (numericsid) {
+			free(numericsid);
+			numericsid = NULL;
+		}
 		if (errno != 0 || *end != '\0')
 			return (EINVAL);
 	} else if (!isdigit(*cp)) {
@@ -2126,13 +2128,14 @@ userquota_propname_decode(const char *propname, boolean_t zoned,
 			if (idmap_id_to_numeric_domain_rid(id, isuser,
 			    &mapdomain, &rid) != 0)
 				return (ENOENT);
-			(void) strcpy(domain, mapdomain);
+			(void) strlcpy(domain, mapdomain, domainlen);
 			*ridp = rid;
 		} else {
 			*ridp = id;
 		}
 	}
 
+	ASSERT3P(numericsid, ==, NULL);
 	return (0);
 }
 
@@ -2732,7 +2735,7 @@ zfs_create(libzfs_handle_t *hdl, const char *path, zfs_type_t type,
  * isn't mounted, and that there are no active dependents.
  */
 int
-zfs_destroy(zfs_handle_t *zhp)
+zfs_destroy(zfs_handle_t *zhp, boolean_t defer)
 {
 	zfs_cmd_t zc = { "\0", "\0", "\0", 0 };
 
@@ -2756,6 +2759,7 @@ zfs_destroy(zfs_handle_t *zhp)
 		zc.zc_objset_type = DMU_OST_ZFS;
 	}
 
+	zc.zc_defer_destroy = defer;
 	if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_DESTROY, &zc) != 0) {
 		return (zfs_standard_error_fmt(zhp->zfs_hdl, errno,
 		    dgettext(TEXT_DOMAIN, "cannot destroy '%s'"),
@@ -2812,7 +2816,7 @@ zfs_remove_link_cb(zfs_handle_t *zhp, void *arg)
  * Destroys all snapshots with the given name in zhp & descendants.
  */
 int
-zfs_destroy_snaps(zfs_handle_t *zhp, char *snapname)
+zfs_destroy_snaps(zfs_handle_t *zhp, char *snapname, boolean_t defer)
 {
 	zfs_cmd_t zc = { "\0", "\0", "\0", 0 };
 	int ret;
@@ -2829,6 +2833,7 @@ zfs_destroy_snaps(zfs_handle_t *zhp, char *snapname)
 
 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
 	(void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
+	zc.zc_defer_destroy = defer;
 
 	ret = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_DESTROY_SNAPS, &zc);
 	if (ret != 0) {
@@ -3244,7 +3249,7 @@ rollback_destroy(zfs_handle_t *zhp, void *data)
 
 			logstr = zhp->zfs_hdl->libzfs_log_str;
 			zhp->zfs_hdl->libzfs_log_str = NULL;
-			cbp->cb_error |= zfs_destroy(zhp);
+			cbp->cb_error |= zfs_destroy(zhp, B_FALSE);
 			zhp->zfs_hdl->libzfs_log_str = logstr;
 		}
 	} else {
@@ -3258,7 +3263,7 @@ rollback_destroy(zfs_handle_t *zhp, void *data)
 			zfs_close(zhp);
 			return (0);
 		}
-		if (zfs_destroy(zhp) != 0)
+		if (zfs_destroy(zhp, B_FALSE) != 0)
 			cbp->cb_error = B_TRUE;
 		else
 			changelist_remove(clp, zhp->zfs_name);
@@ -4059,3 +4064,79 @@ zfs_userspace(zfs_handle_t *zhp, zfs_userquota_prop_t type,
 
 	return (error);
 }
+
+int
+zfs_hold(zfs_handle_t *zhp, const char *snapname, const char *tag,
+    boolean_t recursive)
+{
+	zfs_cmd_t zc = { 0 };
+	libzfs_handle_t *hdl = zhp->zfs_hdl;
+
+	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
+	(void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
+	(void) strlcpy(zc.zc_string, tag, sizeof (zc.zc_string));
+	zc.zc_cookie = recursive;
+
+	if (zfs_ioctl(hdl, ZFS_IOC_HOLD, &zc) != 0) {
+		char errbuf[ZFS_MAXNAMELEN+32];
+
+		/*
+		 * if it was recursive, the one that actually failed will be in
+		 * zc.zc_name.
+		 */
+		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
+		    "cannot hold '%s@%s'"), zc.zc_name, snapname);
+		switch (errno) {
+		case ENOTSUP:
+			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));
+		case EEXIST:
+			return (zfs_error(hdl, EZFS_REFTAG_HOLD, errbuf));
+		default:
+			return (zfs_standard_error_fmt(hdl, errno, errbuf));
+		}
+	}
+
+	return (0);
+}
+
+int
+zfs_release(zfs_handle_t *zhp, const char *snapname, const char *tag,
+    boolean_t recursive)
+{
+	zfs_cmd_t zc = { 0 };
+	libzfs_handle_t *hdl = zhp->zfs_hdl;
+
+	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
+	(void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
+	(void) strlcpy(zc.zc_string, tag, sizeof (zc.zc_string));
+	zc.zc_cookie = recursive;
+
+	if (zfs_ioctl(hdl, ZFS_IOC_RELEASE, &zc) != 0) {
+		char errbuf[ZFS_MAXNAMELEN+32];
+
+		/*
+		 * 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@%s'"), zc.zc_name, 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));
+		case EINVAL:
+			return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
+		default:
+			return (zfs_standard_error_fmt(hdl, errno, errbuf));
+		}
+	}
+
+	return (0);
+}

lib/libzfs/libzfs_sendrecv.c

@@ -113,6 +113,9 @@ fsavl_destroy(avl_tree_t *avl)
 	free(avl);
 }
 
+/*
+ * Given an nvlist, produce an avl tree of snapshots, ordered by guid
+ */
 static avl_tree_t *
 fsavl_create(nvlist_t *fss)
 {
@@ -243,7 +246,9 @@ send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 			continue;
 
 		verify(nvpair_value_nvlist(elem, &propnv) == 0);
-		if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION) {
+		if (prop == ZFS_PROP_QUOTA || prop == ZFS_PROP_RESERVATION ||
+		    prop == ZFS_PROP_REFQUOTA ||
+		    prop == ZFS_PROP_REFRESERVATION) {
 			/* these guys are modifyable, but have no source */
 			uint64_t value;
 			verify(nvlist_lookup_uint64(propnv,
@@ -274,6 +279,11 @@ send_iterate_prop(zfs_handle_t *zhp, nvlist_t *nv)
 	}
 }
 
+/*
+ * recursively generate nvlists describing datasets.  See comment
+ * for the data structure send_data_t above for description of contents
+ * of the nvlist.
+ */
 static int
 send_iterate_fs(zfs_handle_t *zhp, void *arg)
 {
@@ -689,9 +699,20 @@ again:
 }
 
 /*
- * Dumps a backup of tosnap, incremental from fromsnap if it isn't NULL.
+ * Generate a send stream for the dataset identified by the argument zhp.
- * If 'doall', dump all intermediate snaps.
+ *
- * If 'replicate', dump special header and do recursively.
+ * The content of the send stream is the snapshot identified by
+ * 'tosnap'.  Incremental streams are requested in two ways:
+ *     - from the snapshot identified by "fromsnap" (if non-null) or
+ *     - from the origin of the dataset identified by zhp, which must
+ *	 be a clone.  In this case, "fromsnap" is null and "fromorigin"
+ *	 is TRUE.
+ *
+ * The send stream is recursive (i.e. dumps a hierarchy of snapshots) and
+ * uses a special header (with a version field of DMU_BACKUP_HEADER_VERSION)
+ * if "replicate" is set.  If "doall" is set, dump all the intermediate
+ * snapshots. The DMU_BACKUP_HEADER_VERSION header is used in the "doall"
+ * case too.
  */
 int
 zfs_send(zfs_handle_t *zhp, const char *fromsnap, const char *tosnap,
@@ -900,11 +921,12 @@ recv_rename(libzfs_handle_t *hdl, const char *name, const char *tryname,
 	if (err)
 		return (err);
 
+	zc.zc_objset_type = DMU_OST_ZFS;
+	(void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
+
 	if (tryname) {
 		(void) strcpy(newname, tryname);
 
-		zc.zc_objset_type = DMU_OST_ZFS;
-		(void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
 		(void) strlcpy(zc.zc_value, tryname, sizeof (zc.zc_value));
 
 		if (flags.verbose) {
@@ -959,12 +981,18 @@ recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
 	int err = 0;
 	prop_changelist_t *clp;
 	zfs_handle_t *zhp;
+	boolean_t defer = B_FALSE;
+	int spa_version;
 
 	zhp = zfs_open(hdl, name, ZFS_TYPE_DATASET);
 	if (zhp == NULL)
 		return (-1);
 	clp = changelist_gather(zhp, ZFS_PROP_NAME, 0,
 	    flags.force ? MS_FORCE : 0);
+	if (zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
+	    zfs_spa_version(zhp, &spa_version) == 0 &&
+	    spa_version >= SPA_VERSION_USERREFS)
+		defer = B_TRUE;
 	zfs_close(zhp);
 	if (clp == NULL)
 		return (-1);
@@ -973,12 +1001,12 @@ recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
 		return (err);
 
 	zc.zc_objset_type = DMU_OST_ZFS;
+	zc.zc_defer_destroy = defer;
 	(void) strlcpy(zc.zc_name, name, sizeof (zc.zc_name));
 
 	if (flags.verbose)
 		(void) printf("attempting destroy %s\n", zc.zc_name);
 	err = ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc);
-
 	if (err == 0) {
 		if (flags.verbose)
 			(void) printf("success\n");
@@ -988,7 +1016,12 @@ recv_destroy(libzfs_handle_t *hdl, const char *name, int baselen,
 	(void) changelist_postfix(clp);
 	changelist_free(clp);
 
-	if (err != 0)
+	/*
+	 * Deferred destroy should always succeed. Since we can't tell
+	 * if it destroyed the dataset or just marked it for deferred
+	 * destroy, always do the rename just in case.
+	 */
+	if (err != 0 || defer)
 		err = recv_rename(hdl, name, NULL, baselen, newname, flags);
 
 	return (err);
@@ -1775,11 +1808,13 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
 			/* We can't do online recv in this case */
 			clp = changelist_gather(zhp, ZFS_PROP_NAME, 0, 0);
 			if (clp == NULL) {
+				zfs_close(zhp);
 				zcmd_free_nvlists(&zc);
 				return (-1);
 			}
 			if (changelist_prefix(clp) != 0) {
 				changelist_free(clp);
+				zfs_close(zhp);
 				zcmd_free_nvlists(&zc);
 				return (-1);
 			}
@@ -1936,7 +1971,8 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
 	 * (if created, or if we tore them down to do an incremental
 	 * restore), and the /dev links for the new snapshot (if
 	 * created). Also mount any children of the target filesystem
-	 * if we did an incremental receive.
+	 * if we did a replication receive (indicated by stream_avl
+	 * being non-NULL).
 	 */
 	cp = strchr(zc.zc_value, '@');
 	if (cp && (ioctl_err == 0 || !newfs)) {
@@ -1952,7 +1988,7 @@ zfs_receive_one(libzfs_handle_t *hdl, int infd, const char *tosnap,
 				if (err == 0 && ioctl_err == 0)
 					err = zvol_create_link(hdl,
 					    zc.zc_value);
-			} else if (newfs) {
+			} else if (newfs || stream_avl) {
 				/*
 				 * Track the first/top of hierarchy fs,
 				 * for mounting and sharing later.

lib/libzfs/libzfs_status.c

@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -104,6 +104,13 @@ vdev_offlined(uint64_t state, uint64_t aux, uint64_t errs)
 	return (state == VDEV_STATE_OFFLINE);
 }
 
+/* ARGSUSED */
+static int
+vdev_removed(uint64_t state, uint64_t aux, uint64_t errs)
+{
+	return (state == VDEV_STATE_REMOVED);
+}
+
 /*
  * Detect if any leaf devices that have seen errors or could not be opened.
  */
@@ -275,6 +282,12 @@ check_status(nvlist_t *config, boolean_t isimport)
 	if (find_vdev_problem(nvroot, vdev_offlined))
 		return (ZPOOL_STATUS_OFFLINE_DEV);
 
+	/*
+	 * Removed device
+	 */
+	if (find_vdev_problem(nvroot, vdev_removed))
+		return (ZPOOL_STATUS_REMOVED_DEV);
+
 	/*
 	 * Currently resilvering
 	 */

lib/libzfs/libzfs_util.c

@@ -213,6 +213,11 @@ libzfs_error_description(libzfs_handle_t *hdl)
 	case EZFS_UNPLAYED_LOGS:
 		return (dgettext(TEXT_DOMAIN, "log device has unplayed intent "
 		    "logs"));
+	case EZFS_REFTAG_RELE:
+		return (dgettext(TEXT_DOMAIN, "no such tag on this dataset"));
+	case EZFS_REFTAG_HOLD:
+		return (dgettext(TEXT_DOMAIN, "tag already exists on this "
+		    "dataset"));
 	case EZFS_UNKNOWN:
 		return (dgettext(TEXT_DOMAIN, "unknown error"));
 	default:

module/zcommon/include/sys/fs/zfs.h

@@ -18,6 +18,7 @@
  *
  * CDDL HEADER END
  */
+
 /*
  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
@@ -26,6 +27,8 @@
 #ifndef	_SYS_FS_ZFS_H
 #define	_SYS_FS_ZFS_H
 
+#include <sys/time.h>
+
 #ifdef	__cplusplus
 extern "C" {
 #endif
@@ -111,6 +114,8 @@ typedef enum {
 	ZFS_PROP_USEDREFRESERV,
 	ZFS_PROP_USERACCOUNTING,	/* not exposed to the user */
 	ZFS_PROP_STMF_SHAREINFO,	/* not exposed to the user */
+	ZFS_PROP_DEFER_DESTROY,
+	ZFS_PROP_USERREFS,
 	ZFS_NUM_PROPS
 } zfs_prop_t;
 
@@ -280,14 +285,16 @@ typedef enum zfs_cache_type {
 #define	SPA_VERSION_14			14ULL
 #define	SPA_VERSION_15			15ULL
 #define	SPA_VERSION_16			16ULL
+#define	SPA_VERSION_17			17ULL
+#define	SPA_VERSION_18			18ULL
 /*
  * When bumping up SPA_VERSION, make sure GRUB ZFS understands the on-disk
  * format change. Go to usr/src/grub/grub-0.97/stage2/{zfs-include/, fsys_zfs*},
  * and do the appropriate changes.  Also bump the version number in
  * usr/src/grub/capability.
  */
-#define	SPA_VERSION			SPA_VERSION_16
+#define	SPA_VERSION			SPA_VERSION_18
-#define	SPA_VERSION_STRING		"16"
+#define	SPA_VERSION_STRING		"18"
 
 /*
  * Symbolic names for the changes that caused a SPA_VERSION switch.
@@ -303,7 +310,7 @@ typedef enum zfs_cache_type {
 #define	SPA_VERSION_INITIAL		SPA_VERSION_1
 #define	SPA_VERSION_DITTO_BLOCKS	SPA_VERSION_2
 #define	SPA_VERSION_SPARES		SPA_VERSION_3
-#define	SPA_VERSION_RAID6		SPA_VERSION_3
+#define	SPA_VERSION_RAIDZ2		SPA_VERSION_3
 #define	SPA_VERSION_BPLIST_ACCOUNT	SPA_VERSION_3
 #define	SPA_VERSION_RAIDZ_DEFLATE	SPA_VERSION_3
 #define	SPA_VERSION_DNODE_BYTES		SPA_VERSION_3
@@ -325,6 +332,8 @@ typedef enum zfs_cache_type {
 #define	SPA_VERSION_PASSTHROUGH_X	SPA_VERSION_14
 #define	SPA_VERSION_USERSPACE		SPA_VERSION_15
 #define	SPA_VERSION_STMF_PROP		SPA_VERSION_16
+#define	SPA_VERSION_RAIDZ3		SPA_VERSION_17
+#define	SPA_VERSION_USERREFS		SPA_VERSION_18
 
 /*
  * ZPL version - rev'd whenever an incompatible on-disk format change
@@ -601,7 +610,10 @@ typedef enum zfs_ioc {
 	ZFS_IOC_SMB_ACL,
 	ZFS_IOC_USERSPACE_ONE,
 	ZFS_IOC_USERSPACE_MANY,
-	ZFS_IOC_USERSPACE_UPGRADE
+	ZFS_IOC_USERSPACE_UPGRADE,
+	ZFS_IOC_HOLD,
+	ZFS_IOC_RELEASE,
+	ZFS_IOC_GET_HOLDS
 } zfs_ioc_t;
 
 /*
@@ -715,6 +727,8 @@ typedef enum history_internal_events {
 	LOG_DS_REFQUOTA,
 	LOG_DS_REFRESERV,
 	LOG_POOL_SCRUB_DONE,
+	LOG_DS_USER_HOLD,
+	LOG_DS_USER_RELEASE,
 	LOG_END
 } history_internal_events_t;
 

module/zcommon/include/zfs_deleg.h

@@ -61,6 +61,8 @@ typedef enum {
 	ZFS_DELEG_NOTE_GROUPQUOTA,
 	ZFS_DELEG_NOTE_USERUSED,
 	ZFS_DELEG_NOTE_GROUPUSED,
+	ZFS_DELEG_NOTE_HOLD,
+	ZFS_DELEG_NOTE_RELEASE,
 	ZFS_DELEG_NOTE_NONE
 } zfs_deleg_note_t;
 

module/zcommon/zfs_deleg.c

@@ -67,6 +67,8 @@ zfs_deleg_perm_tab_t zfs_deleg_perm_tab[] = {
 	{ZFS_DELEG_PERM_GROUPQUOTA, ZFS_DELEG_NOTE_GROUPQUOTA },
 	{ZFS_DELEG_PERM_USERUSED, ZFS_DELEG_NOTE_USERUSED },
 	{ZFS_DELEG_PERM_GROUPUSED, ZFS_DELEG_NOTE_GROUPUSED },
+	{ZFS_DELEG_PERM_HOLD, ZFS_DELEG_NOTE_HOLD },
+	{ZFS_DELEG_PERM_RELEASE, ZFS_DELEG_NOTE_RELEASE },
 	{NULL, ZFS_DELEG_NOTE_NONE }
 };
 

module/zcommon/zfs_namecheck.c

@@ -59,7 +59,7 @@ valid_char(char c)
  * Snapshot names must be made up of alphanumeric characters plus the following
  * characters:
  *
- * 	[-_.:]
+ * 	[-_.: ]
  */
 int
 snapshot_namecheck(const char *path, namecheck_err_t *why, char *what)

module/zcommon/zfs_prop.c

@@ -235,6 +235,9 @@ zfs_prop_init(void)
 	/* readonly index (boolean) properties */
 	register_index(ZFS_PROP_MOUNTED, "mounted", 0, PROP_READONLY,
 	    ZFS_TYPE_FILESYSTEM, "yes | no", "MOUNTED", boolean_table);
+	register_index(ZFS_PROP_DEFER_DESTROY, "defer_destroy", 0,
+	    PROP_READONLY, ZFS_TYPE_SNAPSHOT, "yes | no", "DEFER_DESTROY",
+	    boolean_table);
 
 	/* set once index properties */
 	register_index(ZFS_PROP_NORMALIZE, "normalization", 0,
@@ -286,6 +289,8 @@ zfs_prop_init(void)
 	register_number(ZFS_PROP_USEDREFRESERV, "usedbyrefreservation", 0,
 	    PROP_READONLY,
 	    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME, "<size>", "USEDREFRESERV");
+	register_number(ZFS_PROP_USERREFS, "userrefs", 0, PROP_READONLY,
+	    ZFS_TYPE_SNAPSHOT, "<count>", "USERREFS");
 
 	/* default number properties */
 	register_number(ZFS_PROP_QUOTA, "quota", 0, PROP_DEFAULT,

module/zfs/dmu.c

@@ -87,6 +87,7 @@ const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
 	{	zap_byteswap,		TRUE,	"scrub work queue"	},
 	{	zap_byteswap,		TRUE,	"ZFS user/group used"	},
 	{	zap_byteswap,		TRUE,	"ZFS user/group quota"	},
+	{	zap_byteswap,		TRUE,	"snapshot refcount tags"},
 };
 
 int
@@ -195,7 +196,7 @@ dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
 
 	ASSERT(length <= DMU_MAX_ACCESS);
 
-	dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT;
+	dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
 	if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
 		dbuf_flags |= DB_RF_NOPREFETCH;
 
@@ -212,6 +213,7 @@ dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
 			    os_dsl_dataset->ds_object,
 			    (longlong_t)dn->dn_object, dn->dn_datablksz,
 			    (longlong_t)offset, (longlong_t)length);
+			rw_exit(&dn->dn_struct_rwlock);
 			return (EIO);
 		}
 		nblks = 1;
@@ -234,9 +236,7 @@ dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
 		}
 		/* initiate async i/o */
 		if (rd) {
-			rw_exit(&dn->dn_struct_rwlock);
 			(void) dbuf_read(db, zio, dbuf_flags);
-			rw_enter(&dn->dn_struct_rwlock, RW_READER);
 		}
 		dbp[i] = &db->db;
 	}
@@ -376,56 +376,51 @@ dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
 	dnode_rele(dn, FTAG);
 }
 
+/*
+ * Get the next "chunk" of file data to free.  We traverse the file from
+ * the end so that the file gets shorter over time (if we crashes in the
+ * middle, this will leave us in a better state).  We find allocated file
+ * data by simply searching the allocated level 1 indirects.
+ */
 static int
-get_next_chunk(dnode_t *dn, uint64_t *offset, uint64_t limit)
+get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
 {
-	uint64_t len = *offset - limit;
+	uint64_t len = *start - limit;
-	uint64_t chunk_len = dn->dn_datablksz * DMU_MAX_DELETEBLKCNT;
+	uint64_t blkcnt = 0;
-	uint64_t subchunk =
+	uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
+	uint64_t iblkrange =
 	    dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
 
-	ASSERT(limit <= *offset);
+	ASSERT(limit <= *start);
 
-	if (len <= chunk_len) {
+	if (len <= iblkrange * maxblks) {
-		*offset = limit;
+		*start = limit;
 		return (0);
 	}
+	ASSERT(ISP2(iblkrange));
 
-	ASSERT(ISP2(subchunk));
+	while (*start > limit && blkcnt < maxblks) {
-
-	while (*offset > limit) {
-		uint64_t initial_offset = P2ROUNDUP(*offset, subchunk);
-		uint64_t delta;
 		int err;
 
-		/* skip over allocated data */
+		/* find next allocated L1 indirect */
 		err = dnode_next_offset(dn,
-		    DNODE_FIND_HOLE|DNODE_FIND_BACKWARDS, offset, 1, 1, 0);
+		    DNODE_FIND_BACKWARDS, start, 2, 1, 0);
-		if (err == ESRCH)
-			*offset = limit;
-		else if (err)
-			return (err);
 
-		ASSERT3U(*offset, <=, initial_offset);
+		/* if there are no more, then we are done */
-		*offset = P2ALIGN(*offset, subchunk);
+		if (err == ESRCH) {
-		delta = initial_offset - *offset;
+			*start = limit;
-		if (delta >= chunk_len) {
-			*offset += delta - chunk_len;
 			return (0);
-		}
+		} else if (err) {
-		chunk_len -= delta;
-
-		/* skip over unallocated data */
-		err = dnode_next_offset(dn,
-		    DNODE_FIND_BACKWARDS, offset, 1, 1, 0);
-		if (err == ESRCH)
-			*offset = limit;
-		else if (err)
 			return (err);
+		}
+		blkcnt += 1;
 
-		if (*offset < limit)
+		/* reset offset to end of "next" block back */
-			*offset = limit;
+		*start = P2ALIGN(*start, iblkrange);
-		ASSERT3U(*offset, <, initial_offset);
+		if (*start <= limit)
+			*start = limit;
+		else
+			*start -= 1;
 	}
 	return (0);
 }
@@ -548,7 +543,7 @@ dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 {
 	dnode_t *dn;
 	dmu_buf_t **dbp;
-	int numbufs, i, err;
+	int numbufs, err;
 
 	err = dnode_hold(os->os, object, FTAG, &dn);
 	if (err)
@@ -559,7 +554,7 @@ dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 	 * block.  If we ever do the tail block optimization, we will need to
 	 * handle that here as well.
 	 */
-	if (dn->dn_datablkshift == 0) {
+	if (dn->dn_maxblkid == 0) {
 		int newsz = offset > dn->dn_datablksz ? 0 :
 		    MIN(size, dn->dn_datablksz - offset);
 		bzero((char *)buf + newsz, size - newsz);
@@ -568,6 +563,7 @@ dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 
 	while (size > 0) {
 		uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
+		int i;
 
 		/*
 		 * NB: we could do this block-at-a-time, but it's nice
@@ -803,9 +799,6 @@ dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
 		if (tocpy == db->db_size)
 			dmu_buf_fill_done(db, tx);
 
-		if (err)
-			break;
-
 		offset += tocpy;
 		size -= tocpy;
 	}

module/zfs/dmu_objset.c

@@ -679,7 +679,7 @@ dmu_objset_create(const char *name, dmu_objset_type_t type,
 }
 
 int
-dmu_objset_destroy(const char *name)
+dmu_objset_destroy(const char *name, boolean_t defer)
 {
 	objset_t *os;
 	int error;
@@ -696,7 +696,7 @@ dmu_objset_destroy(const char *name)
 		dsl_dataset_t *ds = os->os->os_dsl_dataset;
 		zil_destroy(dmu_objset_zil(os), B_FALSE);
 
-		error = dsl_dataset_destroy(ds, os);
+		error = dsl_dataset_destroy(ds, os, defer);
 		/*
 		 * dsl_dataset_destroy() closes the ds.
 		 */
@@ -1136,7 +1136,7 @@ dmu_objset_userspace_upgrade(objset_t *os)
 	 */
 
 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
-		dmu_tx_t *tx = dmu_tx_create(os);
+		dmu_tx_t *tx;
 		dmu_buf_t *db;
 		int objerr;
 
@@ -1146,6 +1146,7 @@ dmu_objset_userspace_upgrade(objset_t *os)
 		objerr = dmu_bonus_hold(os, obj, FTAG, &db);
 		if (objerr)
 			continue;
+		tx = dmu_tx_create(os);
 		dmu_tx_hold_bonus(tx, obj);
 		objerr = dmu_tx_assign(tx, TXG_WAIT);
 		if (objerr) {

module/zfs/dmu_send.c

@@ -393,6 +393,7 @@ recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	dsl_dataset_t *ds = arg1;
 	struct recvbeginsyncarg *rbsa = arg2;
 	int err;
+	struct dsl_ds_destroyarg dsda = {0};
 
 	/* must be a head ds */
 	if (ds->ds_phys->ds_next_snap_obj != 0)
@@ -402,7 +403,8 @@ recv_full_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	if (dsl_dir_is_clone(ds->ds_dir))
 		return (EINVAL);
 
-	err = dsl_dataset_destroy_check(ds, rbsa->tag, tx);
+	dsda.ds = ds;
+	err = dsl_dataset_destroy_check(&dsda, rbsa->tag, tx);
 	if (err)
 		return (err);
 
@@ -427,13 +429,16 @@ recv_full_existing_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	dsl_dir_t *dd = ds->ds_dir;
 	uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
 	uint64_t dsobj;
+	struct dsl_ds_destroyarg dsda = {0};
 
 	/*
 	 * NB: caller must provide an extra hold on the dsl_dir_t, so it
 	 * won't go away when dsl_dataset_destroy_sync() closes the
 	 * dataset.
 	 */
-	dsl_dataset_destroy_sync(ds, rbsa->tag, cr, tx);
+	dsda.ds = ds;
+	dsl_dataset_destroy_sync(&dsda, rbsa->tag, cr, tx);
+	ASSERT3P(dsda.rm_origin, ==, NULL);
 
 	dsobj = dsl_dataset_create_sync_dd(dd, rbsa->origin, flags, tx);
 
@@ -483,7 +488,7 @@ recv_incremental_check(void *arg1, void *arg2, dmu_tx_t *tx)
 
 /* ARGSUSED */
 static void
-recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
+recv_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 {
 	dsl_dataset_t *ohds = arg1;
 	struct recvbeginsyncarg *rbsa = arg2;
@@ -513,27 +518,13 @@ recv_online_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	    dp->dp_spa, tx, cr, "dataset = %lld", dsobj);
 }
 
-/* ARGSUSED */
-static void
-recv_offline_incremental_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
-{
-	dsl_dataset_t *ds = arg1;
-
-	dmu_buf_will_dirty(ds->ds_dbuf, tx);
-	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
-
-	spa_history_internal_log(LOG_DS_REPLAY_INC_SYNC,
-	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "dataset = %lld",
-	    ds->ds_object);
-}
-
 /*
  * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
  * succeeds; otherwise we will leak the holds on the datasets.
  */
 int
 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
-    boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *drc)
+    boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
 {
 	int err = 0;
 	boolean_t byteswap;
@@ -582,36 +573,8 @@ dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 	/*
 	 * Process the begin in syncing context.
 	 */
-	if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE) && !online) {
+	if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
-		/* offline incremental receive */
+		/* incremental receive */
-		err = dsl_dataset_own(tofs, 0, dmu_recv_tag, &ds);
-		if (err)
-			return (err);
-
-		/*
-		 * Only do the rollback if the most recent snapshot
-		 * matches the incremental source
-		 */
-		if (force) {
-			if (ds->ds_prev == NULL ||
-			    ds->ds_prev->ds_phys->ds_guid !=
-			    rbsa.fromguid) {
-				dsl_dataset_disown(ds, dmu_recv_tag);
-				return (ENODEV);
-			}
-			(void) dsl_dataset_rollback(ds, DMU_OST_NONE);
-		}
-		rbsa.force = B_FALSE;
-		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
-		    recv_incremental_check,
-		    recv_offline_incremental_sync, ds, &rbsa, 1);
-		if (err) {
-			dsl_dataset_disown(ds, dmu_recv_tag);
-			return (err);
-		}
-		drc->drc_logical_ds = drc->drc_real_ds = ds;
-	} else if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE)) {
-		/* online incremental receive */
 
 		/* tmp clone name is: tofs/%tosnap" */
 		(void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
@@ -622,11 +585,18 @@ dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
 		if (err)
 			return (err);
 
+		/* must not have an incremental recv already in progress */
+		if (!mutex_tryenter(&ds->ds_recvlock)) {
+			dsl_dataset_rele(ds, dmu_recv_tag);
+			return (EBUSY);
+		}
+
 		rbsa.force = force;
 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
 		    recv_incremental_check,
-		    recv_online_incremental_sync, ds, &rbsa, 5);
+		    recv_incremental_sync, ds, &rbsa, 5);
 		if (err) {
+			mutex_exit(&ds->ds_recvlock);
 			dsl_dataset_rele(ds, dmu_recv_tag);
 			return (err);
 		}
@@ -931,26 +901,6 @@ restore_free(struct restorearg *ra, objset_t *os,
 	return (err);
 }
 
-void
-dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc)
-{
-	if (drc->drc_newfs || drc->drc_real_ds != drc->drc_logical_ds) {
-		/*
-		 * online incremental or new fs: destroy the fs (which
-		 * may be a clone) that we created
-		 */
-		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
-		if (drc->drc_real_ds != drc->drc_logical_ds)
-			dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
-	} else {
-		/*
-		 * offline incremental: rollback to most recent snapshot.
-		 */
-		(void) dsl_dataset_rollback(drc->drc_real_ds, DMU_OST_NONE);
-		dsl_dataset_disown(drc->drc_real_ds, dmu_recv_tag);
-	}
-}
-
 /*
  * Compute checksum of drr_begin record
  */
@@ -1085,11 +1035,17 @@ out:
 
 	if (ra.err != 0) {
 		/*
-		 * rollback or destroy what we created, so we don't
+		 * destroy what we created, so we don't leave it in the
-		 * leave it in the restoring state.
+		 * inconsistent restoring state.
 		 */
 		txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
-		dmu_recv_abort_cleanup(drc);
+
+		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
+		    B_FALSE);
+		if (drc->drc_real_ds != drc->drc_logical_ds) {
+			mutex_exit(&drc->drc_logical_ds->ds_recvlock);
+			dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
+		}
 	}
 
 	vmem_free(ra.buf, ra.bufsize);
@@ -1156,7 +1112,9 @@ dmu_recv_end(dmu_recv_cookie_t *drc)
 			dsl_dataset_rele(ds, dmu_recv_tag);
 		}
 		/* dsl_dataset_destroy() will disown the ds */
-		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag);
+		(void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
+		    B_FALSE);
+		mutex_exit(&drc->drc_logical_ds->ds_recvlock);
 		if (err)
 			return (err);
 	}
@@ -1170,7 +1128,8 @@ dmu_recv_end(dmu_recv_cookie_t *drc)
 	if (err) {
 		if (drc->drc_newfs) {
 			ASSERT(ds == drc->drc_real_ds);
-			(void) dsl_dataset_destroy(ds, dmu_recv_tag);
+			(void) dsl_dataset_destroy(ds, dmu_recv_tag,
+			    B_FALSE);
 			return (err);
 		} else {
 			(void) dsl_dataset_rollback(ds, DMU_OST_NONE);

module/zfs/dmu_tx.c

@@ -699,8 +699,7 @@ dmu_tx_hold_zap(dmu_tx_t *tx, uint64_t object, int add, const char *name)
 	}
 
 	err = zap_count_write(&dn->dn_objset->os, dn->dn_object, name, add,
-	    &txh->txh_space_towrite, &txh->txh_space_tooverwrite,
+	    &txh->txh_space_towrite, &txh->txh_space_tooverwrite);
-	    txh->txh_dnode->dn_datablkshift);
 
 	/*
 	 * If the modified blocks are scattered to the four winds,

module/zfs/dnode.c

@@ -1308,6 +1308,22 @@ dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx)
 	dmu_tx_willuse_space(tx, space);
 }
 
+/*
+ * This function scans a block at the indicated "level" looking for
+ * a hole or data (depending on 'flags').  If level > 0, then we are
+ * scanning an indirect block looking at its pointers.  If level == 0,
+ * then we are looking at a block of dnodes.  If we don't find what we
+ * are looking for in the block, we return ESRCH.  Otherwise, return
+ * with *offset pointing to the beginning (if searching forwards) or
+ * end (if searching backwards) of the range covered by the block
+ * pointer we matched on (or dnode).
+ *
+ * The basic search algorithm used below by dnode_next_offset() is to
+ * use this function to search up the block tree (widen the search) until
+ * we find something (i.e., we don't return ESRCH) and then search back
+ * down the tree (narrow the search) until we reach our original search
+ * level.
+ */
 static int
 dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 	int lvl, uint64_t blkfill, uint64_t txg)
@@ -1378,6 +1394,7 @@ dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 			error = ESRCH;
 	} else {
 		blkptr_t *bp = data;
+		uint64_t start = *offset;
 		span = (lvl - 1) * epbs + dn->dn_datablkshift;
 		minfill = 0;
 		maxfill = blkfill << ((lvl - 1) * epbs);
@@ -1387,18 +1404,25 @@ dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
 		else
 			minfill++;
 
-		for (i = (*offset >> span) & ((1ULL << epbs) - 1);
+		*offset = *offset >> span;
+		for (i = BF64_GET(*offset, 0, epbs);
 		    i >= 0 && i < epb; i += inc) {
 			if (bp[i].blk_fill >= minfill &&
 			    bp[i].blk_fill <= maxfill &&
 			    (hole || bp[i].blk_birth > txg))
 				break;
-			if (inc < 0 && *offset < (1ULL << span))
+			if (inc > 0 || *offset > 0)
-				*offset = 0;
+				*offset += inc;
-			else
-				*offset += (1ULL << span) * inc;
 		}
-		if (i < 0 || i == epb)
+		*offset = *offset << span;
+		if (inc < 0) {
+			/* traversing backwards; position offset at the end */
+			ASSERT3U(*offset, <=, start);
+			*offset = MIN(*offset + (1ULL << span) - 1, start);
+		} else if (*offset < start) {
+			*offset = start;
+		}
+		if (i < 0 || i >= epb)
 			error = ESRCH;
 	}
 

module/zfs/dsl_dataset.c

@@ -39,6 +39,7 @@
 #include <sys/spa.h>
 #include <sys/zfs_znode.h>
 #include <sys/sunddi.h>
+#include <sys/zvol.h>
 
 static char *dsl_reaper = "the grim reaper";
 
@@ -264,6 +265,7 @@ dsl_dataset_evict(dmu_buf_t *db, void *dsv)
 	ASSERT(!list_link_active(&ds->ds_synced_link));
 
 	mutex_destroy(&ds->ds_lock);
+	mutex_destroy(&ds->ds_recvlock);
 	mutex_destroy(&ds->ds_opening_lock);
 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
 	rw_destroy(&ds->ds_rwlock);
@@ -362,6 +364,7 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 		list_link_init(&ds->ds_synced_link);
 
 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
+		mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
 		    NULL);
@@ -380,6 +383,7 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 			 * just opened it.
 			 */
 			mutex_destroy(&ds->ds_lock);
+			mutex_destroy(&ds->ds_recvlock);
 			mutex_destroy(&ds->ds_opening_lock);
 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
 			rw_destroy(&ds->ds_rwlock);
@@ -409,8 +413,15 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 					dsl_dataset_rele(origin, FTAG);
 				}
 			}
-		} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
+		} else {
-			err = dsl_dataset_get_snapname(ds);
+			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
+				err = dsl_dataset_get_snapname(ds);
+			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
+				err = zap_count(
+				    ds->ds_dir->dd_pool->dp_meta_objset,
+				    ds->ds_phys->ds_userrefs_obj,
+				    &ds->ds_userrefs);
+			}
 		}
 
 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
@@ -451,6 +462,7 @@ dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 				dsl_dataset_drop_ref(ds->ds_prev, ds);
 			dsl_dir_close(ds->ds_dir, ds);
 			mutex_destroy(&ds->ds_lock);
+			mutex_destroy(&ds->ds_recvlock);
 			mutex_destroy(&ds->ds_opening_lock);
 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
 			rw_destroy(&ds->ds_rwlock);
@@ -848,6 +860,7 @@ struct destroyarg {
 	dsl_sync_task_group_t *dstg;
 	char *snapname;
 	char *failed;
+	boolean_t defer;
 };
 
 static int
@@ -855,23 +868,30 @@ dsl_snapshot_destroy_one(char *name, void *arg)
 {
 	struct destroyarg *da = arg;
 	dsl_dataset_t *ds;
-	char *cp;
 	int err;
+	char *dsname;
+	size_t buflen;
 
-	(void) strcat(name, "@");
+	/* alloc a buffer to hold name@snapname, plus the terminating NULL */
-	(void) strcat(name, da->snapname);
+	buflen = strlen(name) + strlen(da->snapname) + 2;
-	err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
+	dsname = kmem_alloc(buflen, KM_SLEEP);
+	(void) snprintf(dsname, buflen, "%s@%s", name, da->snapname);
+	err = dsl_dataset_own(dsname, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
 	    da->dstg, &ds);
-	cp = strchr(name, '@');
+	kmem_free(dsname, buflen);
-	*cp = '\0';
 	if (err == 0) {
+		struct dsl_ds_destroyarg *dsda;
+
 		dsl_dataset_make_exclusive(ds, da->dstg);
 		if (ds->ds_user_ptr) {
 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
 			ds->ds_user_ptr = NULL;
 		}
+		dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
+		dsda->ds = ds;
+		dsda->defer = da->defer;
 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
-		    dsl_dataset_destroy_sync, ds, da->dstg, 0);
+		    dsl_dataset_destroy_sync, dsda, da->dstg, 0);
 	} else if (err == ENOENT) {
 		err = 0;
 	} else {
@@ -885,7 +905,7 @@ dsl_snapshot_destroy_one(char *name, void *arg)
  */
 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
 int
-dsl_snapshots_destroy(char *fsname, char *snapname)
+dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
 {
 	int err;
 	struct destroyarg da;
@@ -898,6 +918,7 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
 	da.snapname = snapname;
 	da.failed = fsname;
+	da.defer = defer;
 
 	err = dmu_objset_find(fsname,
 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
@@ -907,7 +928,9 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 
 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
-		dsl_dataset_t *ds = dst->dst_arg1;
+		struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
+		dsl_dataset_t *ds = dsda->ds;
+
 		/*
 		 * Return the file system name that triggered the error
 		 */
@@ -915,7 +938,9 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 			dsl_dataset_name(ds, fsname);
 			*strchr(fsname, '@') = '\0';
 		}
+		ASSERT3P(dsda->rm_origin, ==, NULL);
 		dsl_dataset_disown(ds, da.dstg);
+		kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
 	}
 
 	dsl_sync_task_group_destroy(da.dstg);
@@ -923,18 +948,100 @@ dsl_snapshots_destroy(char *fsname, char *snapname)
 	return (err);
 }
 
+static boolean_t
+dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
+{
+	boolean_t might_destroy = B_FALSE;
+
+	mutex_enter(&ds->ds_lock);
+	if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
+	    DS_IS_DEFER_DESTROY(ds))
+		might_destroy = B_TRUE;
+	mutex_exit(&ds->ds_lock);
+
+	return (might_destroy);
+}
+
+#ifdef _KERNEL
+static int
+dsl_dataset_zvol_cleanup(dsl_dataset_t *ds, const char *name)
+{
+	int error;
+	objset_t *os;
+
+	error = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
+	if (error)
+		return (error);
+
+	if (dmu_objset_type(os) == DMU_OST_ZVOL)
+		error = zvol_remove_minor(name);
+	dmu_objset_close(os);
+
+	return (error);
+}
+#endif
+
+/*
+ * If we're removing a clone, and these three conditions are true:
+ *	1) the clone's origin has no other children
+ *	2) the clone's origin has no user references
+ *	3) the clone's origin has been marked for deferred destruction
+ * Then, prepare to remove the origin as part of this sync task group.
+ */
+static int
+dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
+{
+	dsl_dataset_t *ds = dsda->ds;
+	dsl_dataset_t *origin = ds->ds_prev;
+
+	if (dsl_dataset_might_destroy_origin(origin)) {
+		char *name;
+		int namelen;
+		int error;
+
+		namelen = dsl_dataset_namelen(origin) + 1;
+		name = kmem_alloc(namelen, KM_SLEEP);
+		dsl_dataset_name(origin, name);
+#ifdef _KERNEL
+		error = zfs_unmount_snap(name, NULL);
+		if (error) {
+			kmem_free(name, namelen);
+			return (error);
+		}
+		error = dsl_dataset_zvol_cleanup(origin, name);
+		if (error) {
+			kmem_free(name, namelen);
+			return (error);
+		}
+#endif
+		error = dsl_dataset_own(name,
+		    DS_MODE_READONLY | DS_MODE_INCONSISTENT,
+		    tag, &origin);
+		kmem_free(name, namelen);
+		if (error)
+			return (error);
+		dsda->rm_origin = origin;
+		dsl_dataset_make_exclusive(origin, tag);
+	}
+
+	return (0);
+}
+
 /*
  * ds must be opened as OWNER.  On return (whether successful or not),
  * ds will be closed and caller can no longer dereference it.
  */
 int
-dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
+dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
 {
 	int err;
 	dsl_sync_task_group_t *dstg;
 	objset_t *os;
 	dsl_dir_t *dd;
 	uint64_t obj;
+	struct dsl_ds_destroyarg dsda = {0};
+
+	dsda.ds = ds;
 
 	if (dsl_dataset_is_snapshot(ds)) {
 		/* Destroying a snapshot is simpler */
@@ -944,9 +1051,12 @@ dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
 			ds->ds_user_ptr = NULL;
 		}
+		/* NOTE: defer is always B_FALSE for non-snapshots */
+		dsda.defer = defer;
 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
-		    ds, tag, 0);
+		    &dsda, tag, 0);
+		ASSERT3P(dsda.rm_origin, ==, NULL);
 		goto out;
 	}
 
@@ -1027,13 +1137,45 @@ dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
 		ds->ds_user_ptr = NULL;
 	}
-	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
+
-	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
+	/*
-	    dsl_dataset_destroy_sync, ds, tag, 0);
+	 * If we're removing a clone, we might also need to remove its
-	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
+	 * origin.
-	    dsl_dir_destroy_sync, dd, FTAG, 0);
+	 */
-	err = dsl_sync_task_group_wait(dstg);
+	do {
-	dsl_sync_task_group_destroy(dstg);
+		dsda.need_prep = B_FALSE;
+		if (dsl_dir_is_clone(dd)) {
+			err = dsl_dataset_origin_rm_prep(&dsda, tag);
+			if (err) {
+				dsl_dir_close(dd, FTAG);
+				goto out;
+			}
+		}
+
+		dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
+		dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
+		    dsl_dataset_destroy_sync, &dsda, tag, 0);
+		dsl_sync_task_create(dstg, dsl_dir_destroy_check,
+		    dsl_dir_destroy_sync, dd, FTAG, 0);
+		err = dsl_sync_task_group_wait(dstg);
+		dsl_sync_task_group_destroy(dstg);
+
+		/*
+		 * We could be racing against 'zfs release' or 'zfs destroy -d'
+		 * on the origin snap, in which case we can get EBUSY if we
+		 * needed to destroy the origin snap but were not ready to
+		 * do so.
+		 */
+		if (dsda.need_prep) {
+			ASSERT(err == EBUSY);
+			ASSERT(dsl_dir_is_clone(dd));
+			ASSERT(dsda.rm_origin == NULL);
+		}
+	} while (dsda.need_prep);
+
+	if (dsda.rm_origin != NULL)
+		dsl_dataset_disown(dsda.rm_origin, tag);
+
 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
 	if (err)
 		dsl_dir_close(dd, FTAG);
@@ -1214,7 +1356,8 @@ dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	/*
 	 * We can only roll back to emptyness if it is a ZPL objset.
 	 */
-	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
+	if (*ost != DMU_OST_ZFS &&
+	    ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL)
 		return (EINVAL);
 
 	/*
@@ -1319,6 +1462,7 @@ dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	} else {
 		objset_impl_t *osi;
 
+		ASSERT(*ost != DMU_OST_ZVOL);
 		ASSERT3U(ds->ds_phys->ds_used_bytes, ==, 0);
 		ASSERT3U(ds->ds_phys->ds_compressed_bytes, ==, 0);
 		ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, ==, 0);
@@ -1388,18 +1532,63 @@ dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 	    cr, "dataset = %llu", ds->ds_object);
 }
 
+static int
+dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
+    dmu_tx_t *tx)
+{
+	dsl_dataset_t *ds = dsda->ds;
+	dsl_dataset_t *ds_prev = ds->ds_prev;
+
+	if (dsl_dataset_might_destroy_origin(ds_prev)) {
+		struct dsl_ds_destroyarg ndsda = {0};
+
+		/*
+		 * If we're not prepared to remove the origin, don't remove
+		 * the clone either.
+		 */
+		if (dsda->rm_origin == NULL) {
+			dsda->need_prep = B_TRUE;
+			return (EBUSY);
+		}
+
+		ndsda.ds = ds_prev;
+		ndsda.is_origin_rm = B_TRUE;
+		return (dsl_dataset_destroy_check(&ndsda, tag, tx));
+	}
+
+	/*
+	 * If we're not going to remove the origin after all,
+	 * undo the open context setup.
+	 */
+	if (dsda->rm_origin != NULL) {
+		dsl_dataset_disown(dsda->rm_origin, tag);
+		dsda->rm_origin = NULL;
+	}
+
+	return (0);
+}
+
 /* ARGSUSED */
 int
 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	struct dsl_ds_destroyarg *dsda = arg1;
+	dsl_dataset_t *ds = dsda->ds;
 
 	/* we have an owner hold, so noone else can destroy us */
 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
 
-	/* Can't delete a branch point. */
+	/*
-	if (ds->ds_phys->ds_num_children > 1)
+	 * Only allow deferred destroy on pools that support it.
-		return (EEXIST);
+	 * NOTE: deferred destroy is only supported on snapshots.
+	 */
+	if (dsda->defer) {
+		if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
+		    SPA_VERSION_USERREFS)
+			return (ENOTSUP);
+		ASSERT(dsl_dataset_is_snapshot(ds));
+		return (0);
+	}
 
 	/*
 	 * Can't delete a head dataset if there are snapshots of it.
@@ -1417,6 +1606,31 @@ dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
 		return (EAGAIN);
 
+	if (dsl_dataset_is_snapshot(ds)) {
+		/*
+		 * If this snapshot has an elevated user reference count,
+		 * we can't destroy it yet.
+		 */
+		if (ds->ds_userrefs > 0 && !dsda->releasing)
+			return (EBUSY);
+
+		mutex_enter(&ds->ds_lock);
+		/*
+		 * Can't delete a branch point. However, if we're destroying
+		 * a clone and removing its origin due to it having a user
+		 * hold count of 0 and having been marked for deferred destroy,
+		 * it's OK for the origin to have a single clone.
+		 */
+		if (ds->ds_phys->ds_num_children >
+		    (dsda->is_origin_rm ? 2 : 1)) {
+			mutex_exit(&ds->ds_lock);
+			return (EEXIST);
+		}
+		mutex_exit(&ds->ds_lock);
+	} else if (dsl_dir_is_clone(ds->ds_dir)) {
+		return (dsl_dataset_origin_check(dsda, arg2, tx));
+	}
+
 	/* XXX we should do some i/o error checking... */
 	return (0);
 }
@@ -1464,7 +1678,8 @@ dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
 void
 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 {
-	dsl_dataset_t *ds = arg1;
+	struct dsl_ds_destroyarg *dsda = arg1;
+	dsl_dataset_t *ds = dsda->ds;
 	zio_t *zio;
 	int err;
 	int after_branch_point = FALSE;
@@ -1474,11 +1689,20 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 	uint64_t obj;
 
 	ASSERT(ds->ds_owner);
-	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
+	ASSERT(dsda->defer || 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);
 
+	if (dsda->defer) {
+		ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
+		if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) {
+			dmu_buf_will_dirty(ds->ds_dbuf, tx);
+			ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
+			return;
+		}
+	}
+
 	/* signal any waiters that this dataset is going away */
 	mutex_enter(&ds->ds_lock);
 	ds->ds_owner = dsl_reaper;
@@ -1524,6 +1748,20 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 			/* This clone is toast. */
 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
 			ds_prev->ds_phys->ds_num_children--;
+
+			/*
+			 * If the clone's origin has no other clones, no
+			 * user holds, and has been marked for deferred
+			 * deletion, then we should have done the necessary
+			 * destroy setup for it.
+			 */
+			if (ds_prev->ds_phys->ds_num_children == 1 &&
+			    ds_prev->ds_userrefs == 0 &&
+			    DS_IS_DEFER_DESTROY(ds_prev)) {
+				ASSERT3P(dsda->rm_origin, !=, NULL);
+			} else {
+				ASSERT3P(dsda->rm_origin, ==, NULL);
+			}
 		} else if (!after_branch_point) {
 			ds_prev->ds_phys->ds_next_snap_obj =
 			    ds->ds_phys->ds_next_snap_obj;
@@ -1736,10 +1974,32 @@ dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
 	}
 	if (ds->ds_phys->ds_props_obj != 0)
 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
+	if (ds->ds_phys->ds_userrefs_obj != 0)
+		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
 	dsl_dir_close(ds->ds_dir, ds);
 	ds->ds_dir = NULL;
 	dsl_dataset_drain_refs(ds, tag);
 	VERIFY(0 == dmu_object_free(mos, obj, tx));
+
+	if (dsda->rm_origin) {
+		/*
+		 * Remove the origin of the clone we just destroyed.
+		 */
+		dsl_dataset_t *origin = ds->ds_prev;
+		struct dsl_ds_destroyarg ndsda = {0};
+
+		ASSERT3P(origin, ==, dsda->rm_origin);
+		if (origin->ds_user_ptr) {
+			origin->ds_user_evict_func(origin, origin->ds_user_ptr);
+			origin->ds_user_ptr = NULL;
+		}
+
+		dsl_dataset_rele(origin, tag);
+		ds->ds_prev = NULL;
+
+		ndsda.ds = origin;
+		dsl_dataset_destroy_sync(&ndsda, tag, cr, tx);
+	}
 }
 
 static int
@@ -1954,6 +2214,9 @@ dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
 	    ds->ds_reserved);
 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
 	    ds->ds_phys->ds_guid);
+	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, ds->ds_userrefs);
+	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
+	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
 
 	if (ds->ds_phys->ds_next_snap_obj) {
 		/*
@@ -3023,7 +3286,7 @@ dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 
 	ds->ds_quota = new_quota;
 
-	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
+	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
 
 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
 	    tx, cr, "%lld dataset = %llu ",
@@ -3118,7 +3381,7 @@ dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
 
 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
 	mutex_exit(&ds->ds_dir->dd_lock);
-	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
+	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refreservation",
 	    new_reservation, cr, tx);
 
 	spa_history_internal_log(LOG_DS_REFRESERV,
@@ -3142,3 +3405,421 @@ dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
 	dsl_dataset_rele(ds, FTAG);
 	return (err);
 }
+
+static int
+dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
+{
+	dsl_dataset_t *ds = arg1;
+	char *htag = arg2;
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	int error = 0;
+
+	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
+		return (ENOTSUP);
+
+	if (!dsl_dataset_is_snapshot(ds))
+		return (EINVAL);
+
+	if (strlen(htag) >= ZAP_MAXNAMELEN)
+		return (ENAMETOOLONG);
+
+	/* tags must be unique */
+	mutex_enter(&ds->ds_lock);
+	if (ds->ds_phys->ds_userrefs_obj) {
+		error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
+		    8, 1, tx);
+		if (error == 0)
+			error = EEXIST;
+		else if (error == ENOENT)
+			error = 0;
+	}
+	mutex_exit(&ds->ds_lock);
+
+	return (error);
+}
+
+static void
+dsl_dataset_user_hold_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
+{
+	dsl_dataset_t *ds = arg1;
+	char *htag = arg2;
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	time_t now = gethrestime_sec();
+	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);
+
+	VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
+
+	spa_history_internal_log(LOG_DS_USER_HOLD,
+	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "<%s> dataset = %llu",
+	    htag, ds->ds_object);
+}
+
+struct dsl_ds_holdarg {
+	dsl_sync_task_group_t *dstg;
+	char *htag;
+	char *snapname;
+	boolean_t recursive;
+	char failed[MAXPATHLEN];
+};
+
+static int
+dsl_dataset_user_hold_one(char *dsname, void *arg)
+{
+	struct dsl_ds_holdarg *ha = arg;
+	dsl_dataset_t *ds;
+	int error;
+	char *name;
+	size_t buflen;
+
+	/* alloc a buffer to hold dsname@snapname plus terminating NULL */
+	buflen = strlen(dsname) + strlen(ha->snapname) + 2;
+	name = kmem_alloc(buflen, KM_SLEEP);
+	(void) snprintf(name, buflen, "%s@%s", dsname, ha->snapname);
+	error = dsl_dataset_hold(name, ha->dstg, &ds);
+	kmem_free(name, buflen);
+	if (error == 0) {
+		dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
+		    dsl_dataset_user_hold_sync, ds, ha->htag, 0);
+	} else if (error == ENOENT && ha->recursive) {
+		error = 0;
+	} else {
+		(void) strcpy(ha->failed, dsname);
+	}
+	return (error);
+}
+
+int
+dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
+    boolean_t recursive)
+{
+	struct dsl_ds_holdarg *ha;
+	dsl_sync_task_t *dst;
+	spa_t *spa;
+	int error;
+
+	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
+
+	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
+
+	error = spa_open(dsname, &spa, FTAG);
+	if (error) {
+		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+		return (error);
+	}
+
+	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
+	ha->htag = htag;
+	ha->snapname = snapname;
+	ha->recursive = recursive;
+	if (recursive) {
+		error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
+		    ha, DS_FIND_CHILDREN);
+	} else {
+		error = dsl_dataset_user_hold_one(dsname, ha);
+	}
+	if (error == 0)
+		error = dsl_sync_task_group_wait(ha->dstg);
+
+	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
+	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
+		dsl_dataset_t *ds = dst->dst_arg1;
+
+		if (dst->dst_err) {
+			dsl_dataset_name(ds, ha->failed);
+			*strchr(ha->failed, '@') = '\0';
+		}
+		dsl_dataset_rele(ds, ha->dstg);
+	}
+
+	if (error)
+		(void) strcpy(dsname, ha->failed);
+
+	dsl_sync_task_group_destroy(ha->dstg);
+	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+	spa_close(spa, FTAG);
+	return (error);
+}
+
+struct dsl_ds_releasearg {
+	dsl_dataset_t *ds;
+	const char *htag;
+	boolean_t own;		/* do we own or just hold ds? */
+};
+
+static int
+dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
+    boolean_t *might_destroy)
+{
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	uint64_t zapobj;
+	uint64_t tmp;
+	int error;
+
+	*might_destroy = B_FALSE;
+
+	mutex_enter(&ds->ds_lock);
+	zapobj = ds->ds_phys->ds_userrefs_obj;
+	if (zapobj == 0) {
+		/* The tag can't possibly exist */
+		mutex_exit(&ds->ds_lock);
+		return (ESRCH);
+	}
+
+	/* Make sure the tag exists */
+	error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
+	if (error) {
+		mutex_exit(&ds->ds_lock);
+		if (error == ENOENT)
+			error = ESRCH;
+		return (error);
+	}
+
+	if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
+	    DS_IS_DEFER_DESTROY(ds))
+		*might_destroy = B_TRUE;
+
+	mutex_exit(&ds->ds_lock);
+	return (0);
+}
+
+static int
+dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
+{
+	struct dsl_ds_releasearg *ra = arg1;
+	dsl_dataset_t *ds = ra->ds;
+	boolean_t might_destroy;
+	int error;
+
+	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
+		return (ENOTSUP);
+
+	error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
+	if (error)
+		return (error);
+
+	if (might_destroy) {
+		struct dsl_ds_destroyarg dsda = {0};
+
+		if (dmu_tx_is_syncing(tx)) {
+			/*
+			 * If we're not prepared to remove the snapshot,
+			 * we can't allow the release to happen right now.
+			 */
+			if (!ra->own)
+				return (EBUSY);
+			if (ds->ds_user_ptr) {
+				ds->ds_user_evict_func(ds, ds->ds_user_ptr);
+				ds->ds_user_ptr = NULL;
+			}
+		}
+		dsda.ds = ds;
+		dsda.releasing = B_TRUE;
+		return (dsl_dataset_destroy_check(&dsda, tag, tx));
+	}
+
+	return (0);
+}
+
+static void
+dsl_dataset_user_release_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
+{
+	struct dsl_ds_releasearg *ra = arg1;
+	dsl_dataset_t *ds = ra->ds;
+	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	uint64_t zapobj;
+	uint64_t dsobj = ds->ds_object;
+	uint64_t refs;
+
+	mutex_enter(&ds->ds_lock);
+	ds->ds_userrefs--;
+	refs = ds->ds_userrefs;
+	mutex_exit(&ds->ds_lock);
+	zapobj = ds->ds_phys->ds_userrefs_obj;
+	VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
+	if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
+	    DS_IS_DEFER_DESTROY(ds)) {
+		struct dsl_ds_destroyarg dsda = {0};
+
+		ASSERT(ra->own);
+		dsda.ds = ds;
+		dsda.releasing = B_TRUE;
+		/* We already did the destroy_check */
+		dsl_dataset_destroy_sync(&dsda, tag, cr, tx);
+	}
+
+	spa_history_internal_log(LOG_DS_USER_RELEASE,
+	    spa, tx, cr, "<%s> %lld dataset = %llu",
+	    ra->htag, (longlong_t)refs, dsobj);
+}
+
+static int
+dsl_dataset_user_release_one(char *dsname, void *arg)
+{
+	struct dsl_ds_holdarg *ha = arg;
+	struct dsl_ds_releasearg *ra;
+	dsl_dataset_t *ds;
+	int error;
+	void *dtag = ha->dstg;
+	char *name;
+	size_t buflen;
+	boolean_t own = B_FALSE;
+	boolean_t might_destroy;
+
+	if (strlen(ha->htag) >= ZAP_MAXNAMELEN)
+		return (ENAMETOOLONG);
+
+	/* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
+	buflen = strlen(dsname) + strlen(ha->snapname) + 2;
+	name = kmem_alloc(buflen, KM_SLEEP);
+	(void) snprintf(name, buflen, "%s@%s", dsname, ha->snapname);
+	error = dsl_dataset_hold(name, dtag, &ds);
+	kmem_free(name, buflen);
+	if (error == ENOENT && ha->recursive)
+		return (0);
+	(void) strcpy(ha->failed, dsname);
+	if (error)
+		return (error);
+
+	ASSERT(dsl_dataset_is_snapshot(ds));
+
+	error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
+	if (error) {
+		dsl_dataset_rele(ds, dtag);
+		return (error);
+	}
+
+	if (might_destroy) {
+#ifdef _KERNEL
+		error = zfs_unmount_snap(name, NULL);
+		if (error) {
+			dsl_dataset_rele(ds, dtag);
+			return (error);
+		}
+		error = dsl_dataset_zvol_cleanup(ds, name);
+		if (error) {
+			dsl_dataset_rele(ds, dtag);
+			return (error);
+		}
+#endif
+		if (!dsl_dataset_tryown(ds,
+		    DS_MODE_READONLY | DS_MODE_INCONSISTENT, dtag)) {
+			dsl_dataset_rele(ds, dtag);
+			return (EBUSY);
+		} else {
+			own = B_TRUE;
+			dsl_dataset_make_exclusive(ds, dtag);
+		}
+	}
+
+	ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
+	ra->ds = ds;
+	ra->htag = ha->htag;
+	ra->own = own;
+	dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
+	    dsl_dataset_user_release_sync, ra, dtag, 0);
+
+	return (0);
+}
+
+int
+dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
+    boolean_t recursive)
+{
+	struct dsl_ds_holdarg *ha;
+	dsl_sync_task_t *dst;
+	spa_t *spa;
+	int error;
+
+	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
+
+	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
+
+	error = spa_open(dsname, &spa, FTAG);
+	if (error) {
+		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+		return (error);
+	}
+
+	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
+	ha->htag = htag;
+	ha->snapname = snapname;
+	ha->recursive = recursive;
+	if (recursive) {
+		error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
+		    ha, DS_FIND_CHILDREN);
+	} else {
+		error = dsl_dataset_user_release_one(dsname, ha);
+	}
+	if (error == 0)
+		error = dsl_sync_task_group_wait(ha->dstg);
+
+	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
+	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
+		struct dsl_ds_releasearg *ra = dst->dst_arg1;
+		dsl_dataset_t *ds = ra->ds;
+
+		if (dst->dst_err)
+			dsl_dataset_name(ds, ha->failed);
+
+		if (ra->own)
+			dsl_dataset_disown(ds, ha->dstg);
+		else
+			dsl_dataset_rele(ds, ha->dstg);
+
+		kmem_free(ra, sizeof (struct dsl_ds_releasearg));
+	}
+
+	if (error)
+		(void) strcpy(dsname, ha->failed);
+
+	dsl_sync_task_group_destroy(ha->dstg);
+	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
+	spa_close(spa, FTAG);
+	return (error);
+}
+
+int
+dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
+{
+	dsl_dataset_t *ds;
+	int err;
+
+	err = dsl_dataset_hold(dsname, FTAG, &ds);
+	if (err)
+		return (err);
+
+	VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
+	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)) {
+			VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
+			    za->za_first_integer));
+		}
+		zap_cursor_fini(&zc);
+		kmem_free(za, sizeof (zap_attribute_t));
+	}
+	dsl_dataset_rele(ds, FTAG);
+	return (0);
+}

module/zfs/dsl_deleg.c

@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -66,8 +66,6 @@
  * The ZAP OBJ is referred to as the jump object.
  */
 
-
-
 #include <sys/dmu.h>
 #include <sys/dmu_objset.h>
 #include <sys/dmu_tx.h>
@@ -540,7 +538,7 @@ dsl_deleg_access(const char *dsname, const char *perm, cred_t *cr)
 	dsl_pool_t *dp;
 	void *cookie;
 	int	error;
-	char	checkflag = ZFS_DELEG_LOCAL;
+	char	checkflag;
 	objset_t *mos;
 	avl_tree_t permsets;
 	perm_set_t *setnode;
@@ -563,6 +561,16 @@ dsl_deleg_access(const char *dsname, const char *perm, cred_t *cr)
 		return (EPERM);
 	}
 
+	if (dsl_dataset_is_snapshot(ds)) {
+		/*
+		 * Snapshots are treated as descendents only,
+		 * local permissions do not apply.
+		 */
+		checkflag = ZFS_DELEG_DESCENDENT;
+	} else {
+		checkflag = ZFS_DELEG_LOCAL;
+	}
+
 	avl_create(&permsets, perm_set_compare, sizeof (perm_set_t),
 	    offsetof(perm_set_t, p_node));
 

module/zfs/dsl_dir.c

@@ -96,7 +96,6 @@ dsl_dir_open_obj(dsl_pool_t *dp, uint64_t ddobj,
 #endif
 	if (dd == NULL) {
 		dsl_dir_t *winner;
-		int err;
 
 		dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
 		dd->dd_object = ddobj;

module/zfs/dsl_prop.c

@@ -442,7 +442,7 @@ dsl_props_set_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
 }
 
 void
-dsl_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
+dsl_dir_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
     cred_t *cr, dmu_tx_t *tx)
 {
 	objset_t *mos = dd->dd_pool->dp_meta_objset;

module/zfs/dsl_scrub.c

@@ -1025,6 +1025,8 @@ dsl_pool_scrub_clean_cb(dsl_pool_t *dp,
 int
 dsl_pool_scrub_clean(dsl_pool_t *dp)
 {
+	spa_t *spa = dp->dp_spa;
+
 	/*
 	 * Purge all vdev caches.  We do this here rather than in sync
 	 * context because this requires a writer lock on the spa_config
@@ -1032,11 +1034,11 @@ dsl_pool_scrub_clean(dsl_pool_t *dp)
 	 * spa_scrub_reopen flag indicates that vdev_open() should not
 	 * attempt to start another scrub.
 	 */
-	spa_config_enter(dp->dp_spa, SCL_ALL, FTAG, RW_WRITER);
+	spa_vdev_state_enter(spa);
-	dp->dp_spa->spa_scrub_reopen = B_TRUE;
+	spa->spa_scrub_reopen = B_TRUE;
-	vdev_reopen(dp->dp_spa->spa_root_vdev);
+	vdev_reopen(spa->spa_root_vdev);
-	dp->dp_spa->spa_scrub_reopen = B_FALSE;
+	spa->spa_scrub_reopen = B_FALSE;
-	spa_config_exit(dp->dp_spa, SCL_ALL, FTAG);
+	(void) spa_vdev_state_exit(spa, NULL, 0);
 
 	return (dsl_pool_scrub_setup(dp, SCRUB_FUNC_CLEAN));
 }

module/zfs/include/sys/dmu.h

@@ -117,6 +117,7 @@ typedef enum dmu_object_type {
 	DMU_OT_SCRUB_QUEUE,		/* ZAP */
 	DMU_OT_USERGROUP_USED,		/* ZAP */
 	DMU_OT_USERGROUP_QUOTA,		/* ZAP */
+	DMU_OT_USERREFS,		/* ZAP */
 	DMU_OT_NUMTYPES
 } dmu_object_type_t;
 
@@ -174,8 +175,8 @@ int dmu_objset_evict_dbufs(objset_t *os);
 int dmu_objset_create(const char *name, dmu_objset_type_t type,
     objset_t *clone_parent, uint64_t flags,
     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
-int dmu_objset_destroy(const char *name);
+int dmu_objset_destroy(const char *name, boolean_t defer);
-int dmu_snapshots_destroy(char *fsname, char *snapname);
+int dmu_snapshots_destroy(char *fsname, char *snapname, boolean_t defer);
 int dmu_objset_rollback(objset_t *os);
 int dmu_objset_snapshot(char *fsname, char *snapname, struct nvlist *props,
     boolean_t recursive);
@@ -666,10 +667,9 @@ typedef struct dmu_recv_cookie {
 } dmu_recv_cookie_t;
 
 int dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *,
-    boolean_t force, objset_t *origin, boolean_t online, dmu_recv_cookie_t *);
+    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 dmu_recv_end(dmu_recv_cookie_t *drc);
-void dmu_recv_abort_cleanup(dmu_recv_cookie_t *drc);
 
 /* CRC64 table */
 #define	ZFS_CRC64_POLY	0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */

module/zfs/include/sys/dmu_impl.h

@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -211,10 +211,11 @@ extern "C" {
  * ds_lock
  *    protects:
  *    	ds_user_ptr
- *    	ds_user_evice_func
+ *    	ds_user_evict_func
  *    	ds_open_refcount
  *    	ds_snapname
  *    	ds_phys accounting
+ *	ds_phys userrefs zapobj
  *	ds_reserved
  *    held from:
  *    	dsl_dataset_*

module/zfs/include/sys/dmu_objset.h

@@ -117,7 +117,7 @@ void dmu_objset_close(objset_t *os);
 int dmu_objset_create(const char *name, dmu_objset_type_t type,
     objset_t *clone_parent, uint64_t flags,
     void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg);
-int dmu_objset_destroy(const char *name);
+int dmu_objset_destroy(const char *name, boolean_t defer);
 int dmu_objset_rollback(objset_t *os);
 int dmu_objset_snapshot(char *fsname, char *snapname, nvlist_t *props,
     boolean_t recursive);

module/zfs/include/sys/dsl_dataset.h

@@ -62,6 +62,14 @@ typedef void dsl_dataset_evict_func_t(struct dsl_dataset *, void *);
  */
 #define	DS_FLAG_UNIQUE_ACCURATE	(1ULL<<2)
 
+/*
+ * DS_FLAG_DEFER_DESTROY is set after 'zfs destroy -d' has been called
+ * on a dataset. This allows the dataset to be destroyed using 'zfs release'.
+ */
+#define	DS_FLAG_DEFER_DESTROY	(1ULL<<3)
+#define	DS_IS_DEFER_DESTROY(ds)	\
+	((ds)->ds_phys->ds_flags & DS_FLAG_DEFER_DESTROY)
+
 /*
  * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose
  * name lookups should be performed case-insensitively.
@@ -93,7 +101,8 @@ typedef struct dsl_dataset_phys {
 	blkptr_t ds_bp;
 	uint64_t ds_next_clones_obj;	/* DMU_OT_DSL_CLONES */
 	uint64_t ds_props_obj;		/* DMU_OT_DSL_PROPS for snaps */
-	uint64_t ds_pad[6]; /* pad out to 320 bytes for good measure */
+	uint64_t ds_userrefs_obj;	/* DMU_OT_USERREFS */
+	uint64_t ds_pad[5]; /* pad out to 320 bytes for good measure */
 } dsl_dataset_phys_t;
 
 typedef struct dsl_dataset {
@@ -111,6 +120,9 @@ typedef struct dsl_dataset {
 	/* has internal locking: */
 	bplist_t ds_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;
@@ -122,6 +134,7 @@ typedef struct dsl_dataset {
 	kmutex_t ds_lock;
 	void *ds_user_ptr;
 	dsl_dataset_evict_func_t *ds_user_evict_func;
+	uint64_t ds_userrefs;
 
 	/*
 	 * ds_owner is protected by the ds_rwlock and the ds_lock
@@ -143,6 +156,15 @@ 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? */
+};
+
 #define	dsl_dataset_is_snapshot(ds)	\
 	((ds)->ds_phys->ds_num_children != 0)
 
@@ -167,8 +189,8 @@ 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);
+int dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer);
-int dsl_snapshots_destroy(char *fsname, char *snapname);
+int dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer);
 dsl_checkfunc_t dsl_dataset_destroy_check;
 dsl_syncfunc_t dsl_dataset_destroy_sync;
 dsl_checkfunc_t dsl_dataset_snapshot_check;
@@ -178,6 +200,11 @@ int dsl_dataset_rename(char *name, const char *newname, boolean_t recursive);
 int dsl_dataset_promote(const char *name);
 int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
     boolean_t force);
+int dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
+    boolean_t recursive);
+int dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
+    boolean_t recursive);
+int dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp);
 
 void *dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
     void *p, dsl_dataset_evict_func_t func);

module/zfs/include/sys/dsl_deleg.h

@@ -53,6 +53,8 @@ extern "C" {
 #define	ZFS_DELEG_PERM_GROUPQUOTA	"groupquota"
 #define	ZFS_DELEG_PERM_USERUSED		"userused"
 #define	ZFS_DELEG_PERM_GROUPUSED	"groupused"
+#define	ZFS_DELEG_PERM_HOLD		"hold"
+#define	ZFS_DELEG_PERM_RELEASE		"release"
 
 /*
  * Note: the names of properties that are marked delegatable are also

module/zfs/include/sys/dsl_prop.h

@@ -69,7 +69,7 @@ dsl_syncfunc_t dsl_props_set_sync;
 int dsl_prop_set(const char *ddname, const char *propname,
     int intsz, int numints, const void *buf);
 int dsl_props_set(const char *dsname, nvlist_t *nvl);
-void dsl_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
+void dsl_dir_prop_set_uint64_sync(dsl_dir_t *dd, const char *name, uint64_t val,
     cred_t *cr, dmu_tx_t *tx);
 
 void dsl_prop_nvlist_add_uint64(nvlist_t *nv, zfs_prop_t prop, uint64_t value);

module/zfs/include/sys/spa.h

@@ -500,8 +500,9 @@ extern int spa_history_get(spa_t *spa, uint64_t *offset, uint64_t *len_read,
     char *his_buf);
 extern int spa_history_log(spa_t *spa, const char *his_buf,
     history_log_type_t what);
-void spa_history_internal_log(history_internal_events_t event, spa_t *spa,
+extern void spa_history_internal_log(history_internal_events_t event,
-    dmu_tx_t *tx, cred_t *cr, const char *fmt, ...);
+    spa_t *spa, dmu_tx_t *tx, cred_t *cr, const char *fmt, ...);
+extern void spa_history_log_version(spa_t *spa, history_internal_events_t evt);
 
 /* error handling */
 struct zbookmark;

module/zfs/include/sys/spa_impl.h

@@ -105,6 +105,7 @@ struct spa {
 	int		spa_inject_ref;		/* injection references */
 	uint8_t		spa_sync_on;		/* sync threads are running */
 	spa_load_state_t spa_load_state;	/* current load operation */
+	boolean_t	spa_load_verbatim;	/* load the given config? */
 	taskq_t		*spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES];
 	dsl_pool_t	*spa_dsl_pool;
 	metaslab_class_t *spa_normal_class;	/* normal data class */

module/zfs/include/sys/vdev.h

@@ -47,6 +47,7 @@ typedef enum vdev_dtl_type {
 extern boolean_t zfs_nocacheflush;
 
 extern int vdev_open(vdev_t *);
+extern void vdev_open_children(vdev_t *vd);
 extern int vdev_validate(vdev_t *);
 extern void vdev_close(vdev_t *);
 extern int vdev_create(vdev_t *, uint64_t txg, boolean_t isreplace);

module/zfs/include/sys/vdev_impl.h

@@ -127,6 +127,8 @@ struct vdev {
 	space_map_t	vdev_dtl[DTL_TYPES]; /* in-core dirty time logs	*/
 	vdev_stat_t	vdev_stat;	/* virtual device statistics	*/
 	boolean_t	vdev_expanding;	/* expand the vdev?		*/
+	int		vdev_open_error; /* error on last open		*/
+	kthread_t	*vdev_open_thread; /* thread opening children	*/
 
 	/*
 	 * Top-level vdev state.

module/zfs/include/sys/zap.h

@@ -182,8 +182,7 @@ int zap_lookup_norm(objset_t *ds, uint64_t zapobj, const char *name,
     boolean_t *normalization_conflictp);
 
 int zap_count_write(objset_t *os, uint64_t zapobj, const char *name,
-    int add, uint64_t *towrite, uint64_t *tooverwrite,
+    int add, uint64_t *towrite, uint64_t *tooverwrite);
-    uint64_t dn_datablkshift);
 
 /*
  * Create an attribute with the given name and value.

module/zfs/include/sys/zfs_acl.h

@@ -203,6 +203,7 @@ void zfs_oldace_byteswap(ace_t *, int);
 void zfs_ace_byteswap(void *, size_t, boolean_t);
 extern boolean_t zfs_has_access(struct znode *zp, cred_t *cr);
 extern int zfs_zaccess(struct znode *, int, int, boolean_t, cred_t *);
+int zfs_fastaccesschk_execute(struct znode *, cred_t *);
 extern int zfs_zaccess_rwx(struct znode *, mode_t, int, cred_t *);
 extern int zfs_zaccess_unix(struct znode *, mode_t, cred_t *);
 extern int zfs_acl_access(struct znode *, int, cred_t *);

module/zfs/include/sys/zfs_ioctl.h

@@ -165,6 +165,7 @@ typedef struct zfs_cmd {
 	dmu_objset_stats_t zc_objset_stats;
 	struct drr_begin zc_begin_record;
 	zinject_record_t zc_inject_record;
+	boolean_t	zc_defer_destroy;
 } zfs_cmd_t;
 
 typedef struct zfs_useracct {

module/zfs/include/sys/zfs_vfsops.h

@@ -73,7 +73,6 @@ struct zfsvfs {
 	boolean_t	z_vscan;	/* virus scan on/off */
 	boolean_t	z_use_fuids;	/* version allows fuids */
 	boolean_t	z_replay;	/* set during ZIL replay */
-	kmutex_t	z_online_recv_lock; /* held while recv in progress */
 	uint64_t	z_version;	/* ZPL version */
 	uint64_t	z_shares_dir;	/* hidden shares dir */
 	kmutex_t	z_lock;

module/zfs/include/sys/zfs_znode.h

@@ -77,6 +77,7 @@ extern "C" {
 #define	ZFS_ACL_DEFAULTED	0x20		/* ACL should be defaulted */
 #define	ZFS_ACL_AUTO_INHERIT	0x40		/* ACL should be inherited */
 #define	ZFS_BONUS_SCANSTAMP	0x80		/* Scanstamp in bonus area */
+#define	ZFS_NO_EXECS_DENIED	0x100		/* exec was given to everyone */
 
 /*
  * Is ID ephemeral?
@@ -200,6 +201,7 @@ typedef struct znode {
 	uint64_t	z_gen;		/* generation (same as zp_gen) */
 	uint32_t	z_sync_cnt;	/* synchronous open count */
 	kmutex_t	z_acl_lock;	/* acl data lock */
+	zfs_acl_t	*z_acl_cached;	/* cached acl */
 	list_node_t	z_link_node;	/* all znodes in fs link */
 	/*
 	 * These are dmu managed fields.

module/zfs/include/sys/zio.h

@@ -143,6 +143,8 @@ enum zio_compress {
 #define	ZIO_FLAG_GODFATHER		0x080000
 
 #define	ZIO_FLAG_TRYHARD		0x100000
+#define	ZIO_FLAG_NODATA			0x200000
+#define	ZIO_FLAG_OPTIONAL		0x400000
 
 #define	ZIO_FLAG_GANG_INHERIT		\
 	(ZIO_FLAG_CANFAIL |		\
@@ -161,7 +163,9 @@ enum zio_compress {
 	ZIO_FLAG_IO_REPAIR |		\
 	ZIO_FLAG_IO_RETRY |		\
 	ZIO_FLAG_PROBE |		\
-	ZIO_FLAG_TRYHARD)
+	ZIO_FLAG_TRYHARD |		\
+	ZIO_FLAG_NODATA |		\
+	ZIO_FLAG_OPTIONAL)
 
 #define	ZIO_FLAG_AGG_INHERIT		\
 	(ZIO_FLAG_DONT_AGGREGATE |	\

module/zfs/rrwlock.c

@@ -19,12 +19,10 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
-
-
 #include <sys/refcount.h>
 #include <sys/rrwlock.h>
 
@@ -159,6 +157,14 @@ static 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) {
+		rrl->rr_anon_rcount.rc_count++;
+		mutex_exit(&rrl->rr_lock);
+		return;
+	}
+	DTRACE_PROBE(zfs__rrwfastpath__rdmiss);
+#endif
 	ASSERT(rrl->rr_writer != curthread);
 	ASSERT(refcount_count(&rrl->rr_anon_rcount) >= 0);
 
@@ -208,19 +214,28 @@ void
 rrw_exit(rrwlock_t *rrl, void *tag)
 {
 	mutex_enter(&rrl->rr_lock);
+#if !defined(DEBUG) && defined(_KERNEL)
+	if (!rrl->rr_writer && rrl->rr_linked_rcount.rc_count == 0) {
+		rrl->rr_anon_rcount.rc_count--;
+		if (rrl->rr_anon_rcount.rc_count == 0)
+			cv_broadcast(&rrl->rr_cv);
+		mutex_exit(&rrl->rr_lock);
+		return;
+	}
+	DTRACE_PROBE(zfs__rrwfastpath__exitmiss);
+#endif
 	ASSERT(!refcount_is_zero(&rrl->rr_anon_rcount) ||
 	    !refcount_is_zero(&rrl->rr_linked_rcount) ||
 	    rrl->rr_writer != NULL);
 
 	if (rrl->rr_writer == NULL) {
-		if (rrn_find_and_remove(rrl)) {
+		int64_t count;
-			if (refcount_remove(&rrl->rr_linked_rcount, tag) == 0)
+		if (rrn_find_and_remove(rrl))
-				cv_broadcast(&rrl->rr_cv);
+			count = refcount_remove(&rrl->rr_linked_rcount, tag);
-
+		else
-		} else {
+			count = refcount_remove(&rrl->rr_anon_rcount, tag);
-			if (refcount_remove(&rrl->rr_anon_rcount, tag) == 0)
+		if (count == 0)
-				cv_broadcast(&rrl->rr_cv);
+			cv_broadcast(&rrl->rr_cv);
-		}
 	} else {
 		ASSERT(rrl->rr_writer == curthread);
 		ASSERT(refcount_is_zero(&rrl->rr_anon_rcount) &&

module/zfs/spa.c

@@ -1587,9 +1587,12 @@ spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig)
 		/*
 		 * If the config cache is stale, or we have uninitialized
 		 * metaslabs (see spa_vdev_add()), then update the config.
+		 *
+		 * If spa_load_verbatim is true, trust the current
+		 * in-core spa_config and update the disk labels.
 		 */
 		if (config_cache_txg != spa->spa_config_txg ||
-		    state == SPA_LOAD_IMPORT)
+		    state == SPA_LOAD_IMPORT || spa->spa_load_verbatim)
 			need_update = B_TRUE;
 
 		for (c = 0; c < rvd->vdev_children; c++)
@@ -2285,6 +2288,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
 	if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
 		(void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
+	spa_history_log_version(spa, LOG_POOL_CREATE);
 
 	spa->spa_minref = refcount_count(&spa->spa_refcount);
 
@@ -2420,6 +2424,7 @@ spa_import_rootpool(char *devpath, char *devid)
 
 	spa = spa_add(pname, NULL);
 	spa->spa_is_root = B_TRUE;
+	spa->spa_load_verbatim = B_TRUE;
 
 	/*
 	 * Build up a vdev tree based on the boot device's label config.
@@ -2475,6 +2480,7 @@ spa_import_rootpool(char *devpath, char *devid)
 
 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
 	error = 0;
+	spa_history_log_version(spa, LOG_POOL_IMPORT);
 out:
 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 	vdev_free(rvd);
@@ -2507,6 +2513,8 @@ spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
 	    zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
 	spa = spa_add(pool, altroot);
 
+	spa->spa_load_verbatim = B_TRUE;
+
 	VERIFY(nvlist_dup(config, &spa->spa_config, 0) == 0);
 
 	if (props != NULL)
@@ -2515,6 +2523,7 @@ spa_import_verbatim(const char *pool, nvlist_t *config, nvlist_t *props)
 	spa_config_sync(spa, B_FALSE, B_TRUE);
 
 	mutex_exit(&spa_namespace_lock);
+	spa_history_log_version(spa, LOG_POOL_IMPORT);
 
 	return (0);
 }
@@ -2640,7 +2649,7 @@ spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
 		/*
 		 * Update the config cache to include the newly-imported pool.
 		 */
-		spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, B_FALSE);
+		spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
 	}
 
 	/*
@@ -2650,6 +2659,7 @@ spa_import(const char *pool, nvlist_t *config, nvlist_t *props)
 	spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
 
 	mutex_exit(&spa_namespace_lock);
+	spa_history_log_version(spa, LOG_POOL_IMPORT);
 
 	return (0);
 }
@@ -3007,7 +3017,6 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
 	ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
 	vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
 	vdev_ops_t *pvops;
-	dmu_tx_t *tx;
 	char *oldvdpath, *newvdpath;
 	int newvd_isspare;
 	int error;
@@ -3163,17 +3172,11 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
 
 	(void) spa_vdev_exit(spa, newrootvd, open_txg, 0);
 
-	tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
+	spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, NULL,
-	if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
+	    CRED(),  "%s vdev=%s %s vdev=%s",
-		spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx,
+	    replacing && newvd_isspare ? "spare in" :
-		    CRED(),  "%s vdev=%s %s vdev=%s",
+	    replacing ? "replace" : "attach", newvdpath,
-		    replacing && newvd_isspare ? "spare in" :
+	    replacing ? "for" : "to", oldvdpath);
-		    replacing ? "replace" : "attach", newvdpath,
-		    replacing ? "for" : "to", oldvdpath);
-		dmu_tx_commit(tx);
-	} else {
-		dmu_tx_abort(tx);
-	}
 
 	spa_strfree(oldvdpath);
 	spa_strfree(newvdpath);
@@ -3773,19 +3776,11 @@ spa_async_thread(spa_t *spa)
 		 * then log an internal history event.
 		 */
 		if (space_update) {
-			dmu_tx_t *tx;
+			spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
-
+			    spa, NULL, CRED(),
-			tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
+			    "pool '%s' size: %llu(+%llu)",
-			if (dmu_tx_assign(tx, TXG_WAIT) == 0) {
+			    spa_name(spa), spa_get_space(spa),
-				spa_history_internal_log(LOG_POOL_VDEV_ONLINE,
+			    space_update);
-				    spa, tx, CRED(),
-				    "pool '%s' size: %llu(+%llu)",
-				    spa_name(spa), spa_get_space(spa),
-				    space_update);
-				dmu_tx_commit(tx);
-			} else {
-				dmu_tx_abort(tx);
-			}
 		}
 	}
 

module/zfs/spa_config.c

@@ -209,7 +209,7 @@ spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
 
-	if (rootdir == NULL)
+	if (rootdir == NULL || !(spa_mode_global & FWRITE))
 		return;
 
 	/*
@@ -393,24 +393,13 @@ spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 	return (config);
 }
 
-/*
- * For a pool that's not currently a booting rootpool, update all disk labels,
- * generate a fresh config based on the current in-core state, and sync the
- * global config cache.
- */
-void
-spa_config_update(spa_t *spa, int what)
-{
-	spa_config_update_common(spa, what, FALSE);
-}
-
 /*
  * Update all disk labels, generate a fresh config based on the current
  * in-core state, and sync the global config cache (do not sync the config
  * cache if this is a booting rootpool).
  */
 void
-spa_config_update_common(spa_t *spa, int what, boolean_t isroot)
+spa_config_update(spa_t *spa, int what)
 {
 	vdev_t *rvd = spa->spa_root_vdev;
 	uint64_t txg;
@@ -447,9 +436,9 @@ spa_config_update_common(spa_t *spa, int what, boolean_t isroot)
 	/*
 	 * Update the global config cache to reflect the new mosconfig.
 	 */
-	if (!isroot)
+	if (!spa->spa_is_root)
 		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
 
 	if (what == SPA_CONFIG_UPDATE_POOL)
-		spa_config_update_common(spa, SPA_CONFIG_UPDATE_VDEVS, isroot);
+		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
 }

module/zfs/spa_history.c

@@ -394,13 +394,12 @@ spa_history_get(spa_t *spa, uint64_t *offp, uint64_t *len, char *buf)
 	return (err);
 }
 
-void
+static void
-spa_history_internal_log(history_internal_events_t event, spa_t *spa,
+log_internal(history_internal_events_t event, spa_t *spa,
-    dmu_tx_t *tx, cred_t *cr, const char *fmt, ...)
+    dmu_tx_t *tx, cred_t *cr, const char *fmt, va_list adx)
 {
 	history_arg_t *hap;
 	char *str;
-	va_list adx;
 
 	/*
 	 * If this is part of creating a pool, not everything is
@@ -412,9 +411,7 @@ spa_history_internal_log(history_internal_events_t event, spa_t *spa,
 	hap = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
 	str = vmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);
 
-	va_start(adx, fmt);
 	(void) vsnprintf(str, HIS_MAX_RECORD_LEN, fmt, adx);
-	va_end(adx);
 
 	hap->ha_log_type = LOG_INTERNAL;
 	hap->ha_history_str = str;
@@ -429,3 +426,48 @@ spa_history_internal_log(history_internal_events_t event, spa_t *spa,
 	}
 	/* spa_history_log_sync() will free hap and str */
 }
+
+void
+spa_history_internal_log(history_internal_events_t event, spa_t *spa,
+    dmu_tx_t *tx, cred_t *cr, const char *fmt, ...)
+{
+	dmu_tx_t *htx = tx;
+	va_list adx;
+
+	/* create a tx if we didn't get one */
+	if (tx == NULL) {
+		htx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
+		if (dmu_tx_assign(htx, TXG_WAIT) != 0) {
+			dmu_tx_abort(htx);
+			return;
+		}
+	}
+
+	va_start(adx, fmt);
+	log_internal(event, spa, htx, cr, fmt, adx);
+	va_end(adx);
+
+	/* if we didn't get a tx from the caller, commit the one we made */
+	if (tx == NULL)
+		dmu_tx_commit(htx);
+}
+
+void
+spa_history_log_version(spa_t *spa, history_internal_events_t event)
+{
+#ifdef _KERNEL
+	uint64_t current_vers = spa_version(spa);
+
+	if (current_vers >= SPA_VERSION_ZPOOL_HISTORY) {
+		spa_history_internal_log(event, spa, NULL, CRED(),
+		    "pool spa %llu; zfs spa %llu; zpl %d; uts %s %s %s %s",
+		    (u_longlong_t)current_vers, SPA_VERSION, ZPL_VERSION,
+		    utsname.nodename, utsname.release, utsname.version,
+		    utsname.machine);
+	}
+	cmn_err(CE_CONT, "!%s version %llu pool %s using %llu",
+	    event == LOG_POOL_IMPORT ? "imported" :
+	    event == LOG_POOL_CREATE ? "created" : "accessed",
+	    (u_longlong_t)current_vers, spa_name(spa), SPA_VERSION);
+#endif
+}

module/zfs/spa_misc.c

@@ -316,10 +316,13 @@ spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw)
 void
 spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
 {
+	int wlocks_held = 0;
 	int i;
 
 	for (i = 0; i < SCL_LOCKS; i++) {
 		spa_config_lock_t *scl = &spa->spa_config_lock[i];
+		if (scl->scl_writer == curthread)
+			wlocks_held |= (1 << i);
 		if (!(locks & (1 << i)))
 			continue;
 		mutex_enter(&scl->scl_lock);
@@ -339,6 +342,7 @@ spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw)
 		(void) refcount_add(&scl->scl_count, tag);
 		mutex_exit(&scl->scl_lock);
 	}
+	ASSERT(wlocks_held <= locks);
 }
 
 void

module/zfs/vdev.c

@@ -412,22 +412,26 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 		if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
 		    &nparity) == 0) {
 			/*
-			 * Currently, we can only support 2 parity devices.
+			 * Currently, we can only support 3 parity devices.
 			 */
-			if (nparity == 0 || nparity > 2)
+			if (nparity == 0 || nparity > 3)
 				return (EINVAL);
 			/*
-			 * Older versions can only support 1 parity device.
+			 * Previous versions could only support 1 or 2 parity
+			 * device.
 			 */
-			if (nparity == 2 &&
+			if (nparity > 1 &&
-			    spa_version(spa) < SPA_VERSION_RAID6)
+			    spa_version(spa) < SPA_VERSION_RAIDZ2)
+				return (ENOTSUP);
+			if (nparity > 2 &&
+			    spa_version(spa) < SPA_VERSION_RAIDZ3)
 				return (ENOTSUP);
 		} else {
 			/*
 			 * We require the parity to be specified for SPAs that
 			 * support multiple parity levels.
 			 */
-			if (spa_version(spa) >= SPA_VERSION_RAID6)
+			if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
 				return (EINVAL);
 			/*
 			 * Otherwise, we default to 1 parity device for RAID-Z.
@@ -1004,6 +1008,32 @@ vdev_probe(vdev_t *vd, zio_t *zio)
 	return (NULL);
 }
 
+static void
+vdev_open_child(void *arg)
+{
+	vdev_t *vd = arg;
+
+	vd->vdev_open_thread = curthread;
+	vd->vdev_open_error = vdev_open(vd);
+	vd->vdev_open_thread = NULL;
+}
+
+void
+vdev_open_children(vdev_t *vd)
+{
+	taskq_t *tq;
+	int children = vd->vdev_children;
+
+	tq = taskq_create("vdev_open", children, minclsyspri,
+	    children, children, TASKQ_PREPOPULATE);
+
+	for (int c = 0; c < children; c++)
+		VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
+		    TQ_SLEEP) != NULL);
+
+	taskq_destroy(tq);
+}
+
 /*
  * Prepare a virtual device for access.
  */
@@ -1017,8 +1047,8 @@ vdev_open(vdev_t *vd)
 	uint64_t ashift = 0;
 	int c;
 
-	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
+	ASSERT(vd->vdev_open_thread == curthread ||
-
+	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
 	ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
 	    vd->vdev_state == VDEV_STATE_CANT_OPEN ||
 	    vd->vdev_state == VDEV_STATE_OFFLINE);
@@ -1230,7 +1260,12 @@ vdev_validate(vdev_t *vd)
 
 		nvlist_free(label);
 
-		if (spa->spa_load_state == SPA_LOAD_OPEN &&
+		/*
+		 * If spa->spa_load_verbatim is true, no need to check the
+		 * state of the pool.
+		 */
+		if (!spa->spa_load_verbatim &&
+		    spa->spa_load_state == SPA_LOAD_OPEN &&
 		    state != POOL_STATE_ACTIVE)
 			return (EBADF);
 

module/zfs/vdev_label.c

@@ -246,8 +246,10 @@ vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
 		 * into a crufty old storage pool.
 		 */
 		ASSERT(vd->vdev_nparity == 1 ||
-		    (vd->vdev_nparity == 2 &&
+		    (vd->vdev_nparity <= 2 &&
-		    spa_version(spa) >= SPA_VERSION_RAID6));
+		    spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
+		    (vd->vdev_nparity <= 3 &&
+		    spa_version(spa) >= SPA_VERSION_RAIDZ3));
 
 		/*
 		 * Note that we'll add the nparity tag even on storage pools
@@ -509,7 +511,7 @@ vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
 	int error;
 	uint64_t spare_guid = 0, l2cache_guid;
 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
-	int c, l, n;
+	int c, l;
 	vdev_t *pvd;
 
 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
@@ -647,8 +649,8 @@ vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
 	/*
 	 * Initialize uberblock template.
 	 */
-	ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
+	ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
-	bzero(ub, VDEV_UBERBLOCK_SIZE(vd));
+	bzero(ub, VDEV_UBERBLOCK_RING);
 	*ub = spa->spa_uberblock;
 	ub->ub_txg = 0;
 
@@ -677,11 +679,9 @@ retry:
 		    offsetof(vdev_label_t, vl_pad2),
 		    VDEV_PAD_SIZE, NULL, NULL, flags);
 
-		for (n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
+		vdev_label_write(zio, vd, l, ub,
-			vdev_label_write(zio, vd, l, ub,
+		    offsetof(vdev_label_t, vl_uberblock),
-			    VDEV_UBERBLOCK_OFFSET(vd, n),
+		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
-			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags);
-		}
 	}
 
 	error = zio_wait(zio);
@@ -693,7 +693,7 @@ retry:
 
 	nvlist_free(label);
 	zio_buf_free(pad2, VDEV_PAD_SIZE);
-	zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd));
+	zio_buf_free(ub, VDEV_UBERBLOCK_RING);
 	zio_buf_free(vp, sizeof (vdev_phys_t));
 
 	/*

module/zfs/vdev_mirror.c

@@ -124,21 +124,21 @@ vdev_mirror_map_alloc(zio_t *zio)
 static int
 vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 {
-	vdev_t *cvd;
-	uint64_t c;
 	int numerrors = 0;
-	int ret, lasterror = 0;
+	int lasterror = 0;
 
 	if (vd->vdev_children == 0) {
 		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
 		return (EINVAL);
 	}
 
-	for (c = 0; c < vd->vdev_children; c++) {
+	vdev_open_children(vd);
-		cvd = vd->vdev_child[c];
 
-		if ((ret = vdev_open(cvd)) != 0) {
+	for (int c = 0; c < vd->vdev_children; c++) {
-			lasterror = ret;
+		vdev_t *cvd = vd->vdev_child[c];
+
+		if (cvd->vdev_open_error) {
+			lasterror = cvd->vdev_open_error;
 			numerrors++;
 			continue;
 		}
@@ -158,9 +158,7 @@ vdev_mirror_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 static void
 vdev_mirror_close(vdev_t *vd)
 {
-	uint64_t c;
+	for (int c = 0; c < vd->vdev_children; c++)
-
-	for (c = 0; c < vd->vdev_children; c++)
 		vdev_close(vd->vdev_child[c]);
 }
 

module/zfs/vdev_queue.c

@@ -24,7 +24,7 @@
  */
 
 #include <sys/zfs_context.h>
-#include <sys/spa.h>
+#include <sys/spa_impl.h>
 #include <sys/vdev_impl.h>
 #include <sys/zio.h>
 #include <sys/avl.h>
@@ -48,11 +48,14 @@ int zfs_vdev_time_shift = 6;
 int zfs_vdev_ramp_rate = 2;
 
 /*
- * To reduce IOPs, we aggregate small adjacent i/os into one large i/o.
+ * To reduce IOPs, we aggregate small adjacent I/Os into one large I/O.
- * For read i/os, we also aggregate across small adjacency gaps.
+ * For read I/Os, we also aggregate across small adjacency gaps; for writes
+ * we include spans of optional I/Os to aid aggregation at the disk even when
+ * they aren't able to help us aggregate at this level.
  */
 int zfs_vdev_aggregation_limit = SPA_MAXBLOCKSIZE;
 int zfs_vdev_read_gap_limit = 32 << 10;
+int zfs_vdev_write_gap_limit = 4 << 10;
 
 /*
  * Virtual device vector for disk I/O scheduling.
@@ -172,12 +175,14 @@ vdev_queue_agg_io_done(zio_t *aio)
 static zio_t *
 vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 {
-	zio_t *fio, *lio, *aio, *dio, *nio;
+	zio_t *fio, *lio, *aio, *dio, *nio, *mio;
 	avl_tree_t *t;
 	int flags;
 	uint64_t maxspan = zfs_vdev_aggregation_limit;
 	uint64_t maxgap;
+	int stretch;
 
+again:
 	ASSERT(MUTEX_HELD(&vq->vq_lock));
 
 	if (avl_numnodes(&vq->vq_pending_tree) >= pending_limit ||
@@ -192,21 +197,88 @@ vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 
 	if (!(flags & ZIO_FLAG_DONT_AGGREGATE)) {
 		/*
-		 * We can aggregate I/Os that are adjacent and of the
+		 * We can aggregate I/Os that are sufficiently adjacent and of
-		 * same flavor, as expressed by the AGG_INHERIT flags.
+		 * the same flavor, as expressed by the AGG_INHERIT flags.
-		 * The latter is necessary so that certain attributes
+		 * The latter requirement is necessary so that certain
-		 * of the I/O, such as whether it's a normal I/O or a
+		 * attributes of the I/O, such as whether it's a normal I/O
-		 * scrub/resilver, can be preserved in the aggregate.
+		 * or a scrub/resilver, can be preserved in the aggregate.
+		 * We can include optional I/Os, but don't allow them
+		 * to begin a range as they add no benefit in that situation.
+		 */
+
+		/*
+		 * We keep track of the last non-optional I/O.
+		 */
+		mio = (fio->io_flags & ZIO_FLAG_OPTIONAL) ? NULL : fio;
+
+		/*
+		 * Walk backwards through sufficiently contiguous I/Os
+		 * recording the last non-option I/O.
 		 */
 		while ((dio = AVL_PREV(t, fio)) != NULL &&
 		    (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags &&
-		    IO_SPAN(dio, lio) <= maxspan && IO_GAP(dio, fio) <= maxgap)
+		    IO_SPAN(dio, lio) <= maxspan &&
+		    IO_GAP(dio, fio) <= maxgap) {
 			fio = dio;
+			if (mio == NULL && !(fio->io_flags & ZIO_FLAG_OPTIONAL))
+				mio = fio;
+		}
 
+		/*
+		 * Skip any initial optional I/Os.
+		 */
+		while ((fio->io_flags & ZIO_FLAG_OPTIONAL) && fio != lio) {
+			fio = AVL_NEXT(t, fio);
+			ASSERT(fio != NULL);
+		}
+
+		/*
+		 * Walk forward through sufficiently contiguous I/Os.
+		 */
 		while ((dio = AVL_NEXT(t, lio)) != NULL &&
 		    (dio->io_flags & ZIO_FLAG_AGG_INHERIT) == flags &&
-		    IO_SPAN(fio, dio) <= maxspan && IO_GAP(lio, dio) <= maxgap)
+		    IO_SPAN(fio, dio) <= maxspan &&
+		    IO_GAP(lio, dio) <= maxgap) {
 			lio = dio;
+			if (!(lio->io_flags & ZIO_FLAG_OPTIONAL))
+				mio = lio;
+		}
+
+		/*
+		 * Now that we've established the range of the I/O aggregation
+		 * we must decide what to do with trailing optional I/Os.
+		 * For reads, there's nothing to do. While we are unable to
+		 * aggregate further, it's possible that a trailing optional
+		 * I/O would allow the underlying device to aggregate with
+		 * subsequent I/Os. We must therefore determine if the next
+		 * non-optional I/O is close enough to make aggregation
+		 * worthwhile.
+		 */
+		stretch = B_FALSE;
+		if (t != &vq->vq_read_tree && mio != NULL) {
+			nio = lio;
+			while ((dio = AVL_NEXT(t, nio)) != NULL &&
+			    IO_GAP(nio, dio) == 0 &&
+			    IO_GAP(mio, dio) <= zfs_vdev_write_gap_limit) {
+				nio = dio;
+				if (!(nio->io_flags & ZIO_FLAG_OPTIONAL)) {
+					stretch = B_TRUE;
+					break;
+				}
+			}
+		}
+
+		if (stretch) {
+			/* This may be a no-op. */
+			VERIFY((dio = AVL_NEXT(t, lio)) != NULL);
+			dio->io_flags &= ~ZIO_FLAG_OPTIONAL;
+		} else {
+			while (lio != mio && lio != fio) {
+				ASSERT(lio->io_flags & ZIO_FLAG_OPTIONAL);
+				lio = AVL_PREV(t, lio);
+				ASSERT(lio != NULL);
+			}
+		}
 	}
 
 	if (fio != lio) {
@@ -225,10 +297,15 @@ vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 			ASSERT(dio->io_type == aio->io_type);
 			ASSERT(dio->io_vdev_tree == t);
 
-			if (dio->io_type == ZIO_TYPE_WRITE)
+			if (dio->io_flags & ZIO_FLAG_NODATA) {
+				ASSERT(dio->io_type == ZIO_TYPE_WRITE);
+				bzero((char *)aio->io_data + (dio->io_offset -
+				    aio->io_offset), dio->io_size);
+			} else if (dio->io_type == ZIO_TYPE_WRITE) {
 				bcopy(dio->io_data, (char *)aio->io_data +
 				    (dio->io_offset - aio->io_offset),
 				    dio->io_size);
+			}
 
 			zio_add_child(dio, aio);
 			vdev_queue_io_remove(vq, dio);
@@ -244,6 +321,20 @@ vdev_queue_io_to_issue(vdev_queue_t *vq, uint64_t pending_limit)
 	ASSERT(fio->io_vdev_tree == t);
 	vdev_queue_io_remove(vq, fio);
 
+	/*
+	 * If the I/O is or was optional and therefore has no data, we need to
+	 * simply discard it. We need to drop the vdev queue's lock to avoid a
+	 * deadlock that we could encounter since this I/O will complete
+	 * immediately.
+	 */
+	if (fio->io_flags & ZIO_FLAG_NODATA) {
+		mutex_exit(&vq->vq_lock);
+		zio_vdev_io_bypass(fio);
+		zio_execute(fio);
+		mutex_enter(&vq->vq_lock);
+		goto again;
+	}
+
 	avl_add(&vq->vq_pending_tree, fio);
 
 	return (fio);

module/zfs/vdev_root.c

@@ -19,7 +19,7 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
  */
 
@@ -52,7 +52,6 @@ too_many_errors(vdev_t *vd, int numerrors)
 static int
 vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 {
-	int c;
 	int lasterror = 0;
 	int numerrors = 0;
 
@@ -61,15 +60,14 @@ vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 		return (EINVAL);
 	}
 
-	for (c = 0; c < vd->vdev_children; c++) {
+	vdev_open_children(vd);
-		vdev_t *cvd = vd->vdev_child[c];
-		int error;
 
-		if ((error = vdev_open(cvd)) != 0 &&
+	for (int c = 0; c < vd->vdev_children; c++) {
-		    !cvd->vdev_islog) {
+		vdev_t *cvd = vd->vdev_child[c];
-			lasterror = error;
+
+		if (cvd->vdev_open_error && !cvd->vdev_islog) {
+			lasterror = cvd->vdev_open_error;
 			numerrors++;
-			continue;
 		}
 	}
 
@@ -87,9 +85,7 @@ vdev_root_open(vdev_t *vd, uint64_t *asize, uint64_t *ashift)
 static void
 vdev_root_close(vdev_t *vd)
 {
-	int c;
+	for (int c = 0; c < vd->vdev_children; c++)
-
-	for (c = 0; c < vd->vdev_children; c++)
 		vdev_close(vd->vdev_child[c]);
 }
 

module/zfs/zap_micro.c

@@ -1107,7 +1107,7 @@ zap_get_stats(objset_t *os, uint64_t zapobj, zap_stats_t *zs)
 
 int
 zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
-    uint64_t *towrite, uint64_t *tooverwrite, uint64_t dn_datablkshift)
+    uint64_t *towrite, uint64_t *tooverwrite)
 {
 	zap_t *zap;
 	int err = 0;
@@ -1152,28 +1152,28 @@ zap_count_write(objset_t *os, uint64_t zapobj, const char *name, int add,
 			*towrite += (3 + (add ? 4 : 0)) * SPA_MAXBLOCKSIZE;
 		}
 	} else {
-		if (!add) {
+		/*
-			if (dmu_buf_freeable(zap->zap_dbuf))
+		 * We are here if (name != NULL) and this is a micro-zap.
-				*tooverwrite += SPA_MAXBLOCKSIZE;
+		 * We account for the header block depending on whether it
-			else
+		 * is freeable.
-				*towrite += SPA_MAXBLOCKSIZE;
+		 *
-		} else {
+		 * Incase of an add-operation it is hard to find out
-			/*
+		 * if this add will promote this microzap to fatzap.
-			 * We are here if we are adding and (name != NULL).
+		 * Hence, we consider the worst case and account for the
-			 * It is hard to find out if this add will promote this
+		 * blocks assuming this microzap would be promoted to a
-			 * microzap to fatzap. Hence, we assume the worst case
+		 * fatzap.
-			 * and account for the blocks assuming this microzap
+		 *
-			 * would be promoted to a fatzap.
+		 * 1 block overwritten  : header block
-			 *
+		 * 4 new blocks written : 2 new split leaf, 2 grown
-			 * 1 block overwritten  : header block
+		 *			ptrtbl blocks
-			 * 4 new blocks written : 2 new split leaf, 2 grown
+		 */
-			 *			ptrtbl blocks
+		if (dmu_buf_freeable(zap->zap_dbuf))
-			 */
+			*tooverwrite += SPA_MAXBLOCKSIZE;
-			if (dmu_buf_freeable(zap->zap_dbuf))
+		else
-				*tooverwrite += 1 << dn_datablkshift;
+			*towrite += SPA_MAXBLOCKSIZE;
-			else
+
-				*towrite += 1 << dn_datablkshift;
+		if (add) {
-			*towrite += 4 << dn_datablkshift;
+			*towrite += 4 * SPA_MAXBLOCKSIZE;
 		}
 	}
 

module/zfs/zfs_acl.c

@@ -93,6 +93,8 @@
 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
     ZFS_ACL_OBJ_ACE)
 
+#define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
+
 static uint16_t
 zfs_ace_v0_get_type(void *acep)
 {
@@ -781,6 +783,7 @@ zfs_mode_compute(znode_t *zp, zfs_acl_t *aclp)
 	uint64_t	who;
 	uint16_t	iflags, type;
 	uint32_t	access_mask;
+	boolean_t	an_exec_denied = B_FALSE;
 
 	mode = (zp->z_phys->zp_mode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
 
@@ -905,8 +908,32 @@ zfs_mode_compute(znode_t *zp, zfs_acl_t *aclp)
 					}
 				}
 			}
+		} else {
+			/*
+			 * Only care if this IDENTIFIER_GROUP or
+			 * USER ACE denies execute access to someone,
+			 * mode is not affected
+			 */
+			if ((access_mask & ACE_EXECUTE) && type == DENY)
+				an_exec_denied = B_TRUE;
 		}
 	}
+
+	/*
+	 * Failure to allow is effectively a deny, so execute permission
+	 * is denied if it was never mentioned or if we explicitly
+	 * weren't allowed it.
+	 */
+	if (!an_exec_denied &&
+	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
+	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
+		an_exec_denied = B_TRUE;
+
+	if (an_exec_denied)
+		zp->z_phys->zp_flags &= ~ZFS_NO_EXECS_DENIED;
+	else
+		zp->z_phys->zp_flags |= ZFS_NO_EXECS_DENIED;
+
 	return (mode);
 }
 
@@ -946,7 +973,8 @@ zfs_acl_node_read_internal(znode_t *zp, boolean_t will_modify)
 }
 
 /*
- * Read an external acl object.
+ * Read an external acl object.  If the intent is to modify, always
+ * create a new acl and leave any cached acl in place.
  */
 static int
 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
@@ -960,8 +988,15 @@ zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
 
 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
 
+	if (zp->z_acl_cached && !will_modify) {
+		*aclpp = zp->z_acl_cached;
+		return (0);
+	}
+
 	if (zp->z_phys->zp_acl.z_acl_extern_obj == 0) {
 		*aclpp = zfs_acl_node_read_internal(zp, will_modify);
+		if (!will_modify)
+			zp->z_acl_cached = *aclpp;
 		return (0);
 	}
 
@@ -995,6 +1030,8 @@ zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
 	}
 
 	*aclpp = aclp;
+	if (!will_modify)
+		zp->z_acl_cached = aclp;
 	return (0);
 }
 
@@ -1019,11 +1056,16 @@ zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
 
 	dmu_buf_will_dirty(zp->z_dbuf, tx);
 
+	if (zp->z_acl_cached) {
+		zfs_acl_free(zp->z_acl_cached);
+		zp->z_acl_cached = NULL;
+	}
+
 	zphys->zp_mode = zfs_mode_compute(zp, aclp);
 
 	/*
-	 * Decide which opbject type to use.  If we are forced to
+	 * Decide which object type to use.  If we are forced to
-	 * use old ACL format than transform ACL into zfs_oldace_t
+	 * use old ACL format then transform ACL into zfs_oldace_t
 	 * layout.
 	 */
 	if (!zfsvfs->z_use_fuids) {
@@ -1869,7 +1911,6 @@ zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
 			mutex_exit(&dzp->z_acl_lock);
 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
-			zfs_acl_free(paclp);
 		} else {
 			acl_ids->z_aclp =
 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
@@ -1998,8 +2039,6 @@ zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
 
 	mutex_exit(&zp->z_acl_lock);
 
-	zfs_acl_free(aclp);
-
 	return (0);
 }
 
@@ -2095,11 +2134,6 @@ zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
 		aclp->z_hints |= (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
 	}
 top:
-	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)) {
-		zfs_acl_free(aclp);
-		return (error);
-	}
-
 	mutex_enter(&zp->z_lock);
 	mutex_enter(&zp->z_acl_lock);
 
@@ -2145,6 +2179,7 @@ top:
 
 	error = zfs_aclset_common(zp, aclp, cr, tx);
 	ASSERT(error == 0);
+	zp->z_acl_cached = aclp;
 
 	if (fuid_dirtied)
 		zfs_fuid_sync(zfsvfs, tx);
@@ -2154,7 +2189,6 @@ top:
 
 	if (fuidp)
 		zfs_fuid_info_free(fuidp);
-	zfs_acl_free(aclp);
 	dmu_tx_commit(tx);
 done:
 	mutex_exit(&zp->z_acl_lock);
@@ -2301,7 +2335,6 @@ zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
 					checkit = B_TRUE;
 				break;
 			} else {
-				zfs_acl_free(aclp);
 				mutex_exit(&zp->z_acl_lock);
 				return (EIO);
 			}
@@ -2321,7 +2354,6 @@ zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
 				    uint32_t, mask_matched);
 				if (anyaccess) {
 					mutex_exit(&zp->z_acl_lock);
-					zfs_acl_free(aclp);
 					return (0);
 				}
 			}
@@ -2334,7 +2366,6 @@ zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
 	}
 
 	mutex_exit(&zp->z_acl_lock);
-	zfs_acl_free(aclp);
 
 	/* Put the found 'denies' back on the working mode */
 	if (deny_mask) {
@@ -2366,8 +2397,7 @@ zfs_has_access(znode_t *zp, cred_t *cr)
 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VREAD) == 0 ||
 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VWRITE) == 0 ||
 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VEXEC) == 0 ||
-		    secpolicy_vnode_chown(cr, B_TRUE) == 0 ||
+		    secpolicy_vnode_chown(cr, owner) == 0 ||
-		    secpolicy_vnode_chown(cr, B_FALSE) == 0 ||
 		    secpolicy_vnode_setdac(cr, owner) == 0 ||
 		    secpolicy_vnode_remove(cr) == 0);
 	}
@@ -2421,6 +2451,78 @@ zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
 	    check_privs, B_FALSE, cr));
 }
 
+int
+zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
+{
+	boolean_t owner = B_FALSE;
+	boolean_t groupmbr = B_FALSE;
+	boolean_t is_attr;
+	uid_t fowner;
+	uid_t gowner;
+	uid_t uid = crgetuid(cr);
+	int error;
+
+	if (zdp->z_phys->zp_flags & ZFS_AV_QUARANTINED)
+		return (EACCES);
+
+	is_attr = ((zdp->z_phys->zp_flags & ZFS_XATTR) &&
+	    (ZTOV(zdp)->v_type == VDIR));
+	if (is_attr)
+		goto slow;
+
+	mutex_enter(&zdp->z_acl_lock);
+
+	if (zdp->z_phys->zp_flags & ZFS_NO_EXECS_DENIED) {
+		mutex_exit(&zdp->z_acl_lock);
+		return (0);
+	}
+
+	if (FUID_INDEX(zdp->z_phys->zp_uid) != 0 ||
+	    FUID_INDEX(zdp->z_phys->zp_gid) != 0) {
+		mutex_exit(&zdp->z_acl_lock);
+		goto slow;
+	}
+
+	fowner = (uid_t)zdp->z_phys->zp_uid;
+	gowner = (uid_t)zdp->z_phys->zp_gid;
+
+	if (uid == fowner) {
+		owner = B_TRUE;
+		if (zdp->z_phys->zp_mode & S_IXUSR) {
+			mutex_exit(&zdp->z_acl_lock);
+			return (0);
+		} else {
+			mutex_exit(&zdp->z_acl_lock);
+			goto slow;
+		}
+	}
+	if (groupmember(gowner, cr)) {
+		groupmbr = B_TRUE;
+		if (zdp->z_phys->zp_mode & S_IXGRP) {
+			mutex_exit(&zdp->z_acl_lock);
+			return (0);
+		} else {
+			mutex_exit(&zdp->z_acl_lock);
+			goto slow;
+		}
+	}
+	if (!owner && !groupmbr) {
+		if (zdp->z_phys->zp_mode & S_IXOTH) {
+			mutex_exit(&zdp->z_acl_lock);
+			return (0);
+		}
+	}
+
+	mutex_exit(&zdp->z_acl_lock);
+
+slow:
+	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
+	ZFS_ENTER(zdp->z_zfsvfs);
+	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
+	ZFS_EXIT(zdp->z_zfsvfs);
+	return (error);
+}
+
 /*
  * Determine whether Access should be granted/denied, invoking least
  * priv subsytem when a deny is determined.
@@ -2515,7 +2617,7 @@ zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
 			    owner, checkmode);
 
 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
-			error = secpolicy_vnode_chown(cr, B_TRUE);
+			error = secpolicy_vnode_chown(cr, owner);
 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
 			error = secpolicy_vnode_setdac(cr, owner);
 
@@ -2524,7 +2626,7 @@ zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
 			error = secpolicy_vnode_remove(cr);
 
 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
-			error = secpolicy_vnode_chown(cr, B_FALSE);
+			error = secpolicy_vnode_chown(cr, owner);
 		}
 		if (error == 0) {
 			/*

module/zfs/zfs_ctldir.c

@@ -700,7 +700,7 @@ zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
 		if (err)
 			avl_add(&sdp->sd_snaps, sep);
 		else
-			err = dmu_objset_destroy(snapname);
+			err = dmu_objset_destroy(snapname, B_FALSE);
 	} else {
 		err = ENOENT;
 	}

module/zfs/zfs_fuid.c

@@ -353,6 +353,7 @@ retry:
 		rw_exit(&zfsvfs->z_fuid_lock);
 		return (retidx);
 	} else {
+		rw_exit(&zfsvfs->z_fuid_lock);
 		return (-1);
 	}
 }

module/zfs/zfs_ioctl.c

@@ -763,6 +763,20 @@ zfs_secpolicy_userspace_upgrade(zfs_cmd_t *zc, cred_t *cr)
 	return (zfs_secpolicy_setprop(zc->zc_name, ZFS_PROP_VERSION, cr));
 }
 
+static int
+zfs_secpolicy_hold(zfs_cmd_t *zc, cred_t *cr)
+{
+	return (zfs_secpolicy_write_perms(zc->zc_name,
+	    ZFS_DELEG_PERM_HOLD, cr));
+}
+
+static int
+zfs_secpolicy_release(zfs_cmd_t *zc, cred_t *cr)
+{
+	return (zfs_secpolicy_write_perms(zc->zc_name,
+	    ZFS_DELEG_PERM_RELEASE, cr));
+}
+
 /*
  * Returns the nvlist as specified by the user in the zfs_cmd_t.
  */
@@ -2470,7 +2484,7 @@ zfs_ioc_create(zfs_cmd_t *zc)
 	 */
 	if (error == 0) {
 		if ((error = zfs_set_prop_nvlist(zc->zc_name, nvprops)) != 0)
-			(void) dmu_objset_destroy(zc->zc_name);
+			(void) dmu_objset_destroy(zc->zc_name, B_FALSE);
 	}
 	nvlist_free(nvprops);
 	return (error);
@@ -2557,8 +2571,9 @@ zfs_unmount_snap(char *name, void *arg)
 
 /*
  * inputs:
- * zc_name	name of filesystem
+ * zc_name		name of filesystem
- * zc_value	short name of snapshot
+ * zc_value		short name of snapshot
+ * zc_defer_destroy	mark for deferred destroy
  *
  * outputs:	none
  */
@@ -2573,13 +2588,15 @@ zfs_ioc_destroy_snaps(zfs_cmd_t *zc)
 	    zfs_unmount_snap, zc->zc_value, DS_FIND_CHILDREN);
 	if (err)
 		return (err);
-	return (dmu_snapshots_destroy(zc->zc_name, zc->zc_value));
+	return (dmu_snapshots_destroy(zc->zc_name, zc->zc_value,
+	    zc->zc_defer_destroy));
 }
 
 /*
  * inputs:
  * zc_name		name of dataset to destroy
  * zc_objset_type	type of objset
+ * zc_defer_destroy	mark for deferred destroy
  *
  * outputs:		none
  */
@@ -2592,7 +2609,7 @@ zfs_ioc_destroy(zfs_cmd_t *zc)
 			return (err);
 	}
 
-	return (dmu_objset_destroy(zc->zc_name));
+	return (dmu_objset_destroy(zc->zc_name, zc->zc_defer_destroy));
 }
 
 /*
@@ -2712,7 +2729,6 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 	file_t *fp;
 	objset_t *os;
 	dmu_recv_cookie_t drc;
-	zfsvfs_t *zfsvfs = NULL;
 	boolean_t force = (boolean_t)zc->zc_guid;
 	int error, fd;
 	offset_t off;
@@ -2744,25 +2760,12 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 		return (EBADF);
 	}
 
-	if (getzfsvfs(tofs, &zfsvfs) == 0) {
+	if (props && dmu_objset_open(tofs, DMU_OST_ANY,
-		if (!mutex_tryenter(&zfsvfs->z_online_recv_lock)) {
+	    DS_MODE_USER | DS_MODE_READONLY, &os) == 0) {
-			VFS_RELE(zfsvfs->z_vfs);
-			zfsvfs = NULL;
-			error = EBUSY;
-			goto out;
-		}
 		/*
 		 * If new properties are supplied, they are to completely
 		 * replace the existing ones, so stash away the existing ones.
 		 */
-		if (props)
-			(void) dsl_prop_get_all(zfsvfs->z_os, &origprops, TRUE);
-	} else if (props && dmu_objset_open(tofs, DMU_OST_ANY,
-	    DS_MODE_USER | DS_MODE_READONLY, &os) == 0) {
-		/*
-		 * Get the props even if there was no zfsvfs (zvol or
-		 * unmounted zpl).
-		 */
 		(void) dsl_prop_get_all(os, &origprops, TRUE);
 
 		dmu_objset_close(os);
@@ -2776,7 +2779,7 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 	}
 
 	error = dmu_recv_begin(tofs, tosnap, &zc->zc_begin_record,
-	    force, origin, zfsvfs != NULL, &drc);
+	    force, origin, &drc);
 	if (origin)
 		dmu_objset_close(origin);
 	if (error)
@@ -2797,25 +2800,33 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 	off = fp->f_offset;
 	error = dmu_recv_stream(&drc, fp->f_vnode, &off);
 
-	if (error == 0 && zfsvfs) {
+	if (error == 0) {
-		char *osname;
+		zfsvfs_t *zfsvfs = NULL;
-		int mode;
 
-		/* online recv */
+		if (getzfsvfs(tofs, &zfsvfs) == 0) {
-		osname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
+			/* online recv */
-		error = zfs_suspend_fs(zfsvfs, osname, &mode);
+			int end_err;
-		if (error == 0) {
+			char *osname;
-			int resume_err;
+			int mode;
 
-			error = dmu_recv_end(&drc);
+			osname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
-			resume_err = zfs_resume_fs(zfsvfs, osname, mode);
+			error = zfs_suspend_fs(zfsvfs, osname, &mode);
-			error = error ? error : resume_err;
+			/*
+			 * If the suspend fails, then the recv_end will
+			 * likely also fail, and clean up after itself.
+			 */
+			end_err = dmu_recv_end(&drc);
+			if (error == 0) {
+				int resume_err =
+				    zfs_resume_fs(zfsvfs, osname, mode);
+				error = error ? error : resume_err;
+			}
+			error = error ? error : end_err;
+			VFS_RELE(zfsvfs->z_vfs);
+			kmem_free(osname, MAXNAMELEN);
 		} else {
-			dmu_recv_abort_cleanup(&drc);
+			error = dmu_recv_end(&drc);
 		}
-		kmem_free(osname, MAXNAMELEN);
-	} else if (error == 0) {
-		error = dmu_recv_end(&drc);
 	}
 
 	zc->zc_cookie = off - fp->f_offset;
@@ -2830,10 +2841,6 @@ zfs_ioc_recv(zfs_cmd_t *zc)
 		(void) zfs_set_prop_nvlist(tofs, origprops);
 	}
 out:
-	if (zfsvfs) {
-		mutex_exit(&zfsvfs->z_online_recv_lock);
-		VFS_RELE(zfsvfs->z_vfs);
-	}
 	nvlist_free(props);
 	nvlist_free(origprops);
 	releasef(fd);
@@ -3435,6 +3442,69 @@ zfs_ioc_smb_acl(zfs_cmd_t *zc)
 	return (error);
 }
 
+/*
+ * inputs:
+ * zc_name	name of filesystem
+ * zc_value	short name of snap
+ * zc_string	user-supplied tag for this reference
+ * zc_cookie	recursive flag
+ *
+ * outputs:		none
+ */
+static int
+zfs_ioc_hold(zfs_cmd_t *zc)
+{
+	boolean_t recursive = zc->zc_cookie;
+
+	if (snapshot_namecheck(zc->zc_value, NULL, NULL) != 0)
+		return (EINVAL);
+
+	return (dsl_dataset_user_hold(zc->zc_name, zc->zc_value,
+	    zc->zc_string, recursive));
+}
+
+/*
+ * inputs:
+ * zc_name	name of dataset from which we're releasing a user reference
+ * zc_value	short name of snap
+ * zc_string	user-supplied tag for this reference
+ * zc_cookie	recursive flag
+ *
+ * outputs:		none
+ */
+static int
+zfs_ioc_release(zfs_cmd_t *zc)
+{
+	boolean_t recursive = zc->zc_cookie;
+
+	if (snapshot_namecheck(zc->zc_value, NULL, NULL) != 0)
+		return (EINVAL);
+
+	return (dsl_dataset_user_release(zc->zc_name, zc->zc_value,
+	    zc->zc_string, recursive));
+}
+
+/*
+ * inputs:
+ * zc_name		name of filesystem
+ *
+ * outputs:
+ * zc_nvlist_src{_size}	nvlist of snapshot holds
+ */
+static int
+zfs_ioc_get_holds(zfs_cmd_t *zc)
+{
+	nvlist_t *nvp;
+	int error;
+
+	if ((error = dsl_dataset_get_holds(zc->zc_name, &nvp)) == 0) {
+		error = put_nvlist(zc, nvp);
+		nvlist_free(nvp);
+	}
+
+	return (error);
+}
+
 /*
  * pool create, destroy, and export don't log the history as part of
  * zfsdev_ioctl, but rather zfs_ioc_pool_create, and zfs_ioc_pool_export
@@ -3515,8 +3585,8 @@ static zfs_ioc_vec_t zfs_ioc_vec[] = {
 	    B_TRUE },
 	{ zfs_ioc_dsobj_to_dsname, zfs_secpolicy_config, POOL_NAME, B_FALSE,
 	    B_FALSE },
-	{ zfs_ioc_obj_to_path, zfs_secpolicy_config, NO_NAME, B_FALSE,
+	{ zfs_ioc_obj_to_path, zfs_secpolicy_config, DATASET_NAME, B_FALSE,
-	    B_FALSE },
+	    B_TRUE },
 	{ zfs_ioc_pool_set_props, zfs_secpolicy_config,	POOL_NAME, B_TRUE,
 	    B_TRUE },
 	{ zfs_ioc_pool_get_props, zfs_secpolicy_read, POOL_NAME, B_FALSE,
@@ -3538,6 +3608,11 @@ static zfs_ioc_vec_t zfs_ioc_vec[] = {
 	    DATASET_NAME, B_FALSE, B_FALSE },
 	{ zfs_ioc_userspace_upgrade, zfs_secpolicy_userspace_upgrade,
 	    DATASET_NAME, B_FALSE, B_TRUE },
+	{ zfs_ioc_hold, zfs_secpolicy_hold, DATASET_NAME, B_TRUE, B_TRUE },
+	{ zfs_ioc_release, zfs_secpolicy_release, DATASET_NAME, B_TRUE,
+	    B_TRUE },
+	{ zfs_ioc_get_holds, zfs_secpolicy_read, DATASET_NAME, B_FALSE,
+	    B_TRUE }
 };
 
 int

module/zfs/zfs_vfsops.c

@@ -935,7 +935,6 @@ zfsvfs_create(const char *osname, int mode, zfsvfs_t **zvp)
 		goto out;
 
 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
-	mutex_init(&zfsvfs->z_online_recv_lock, NULL, MUTEX_DEFAULT, NULL);
 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
 	    offsetof(znode_t, z_link_node));
@@ -1051,7 +1050,6 @@ zfsvfs_free(zfsvfs_t *zfsvfs)
 	zfs_fuid_destroy(zfsvfs);
 
 	mutex_destroy(&zfsvfs->z_znodes_lock);
-	mutex_destroy(&zfsvfs->z_online_recv_lock);
 	mutex_destroy(&zfsvfs->z_lock);
 	list_destroy(&zfsvfs->z_all_znodes);
 	rrw_destroy(&zfsvfs->z_teardown_lock);

module/zfs/zfs_vnops.c

@@ -208,6 +208,12 @@ zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 	znode_t	*zp = VTOZ(vp);
 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 
+	/*
+	 * Clean up any locks held by this process on the vp.
+	 */
+	cleanlocks(vp, ddi_get_pid(), 0);
+	cleanshares(vp, ddi_get_pid());
+
 	ZFS_ENTER(zfsvfs);
 	ZFS_VERIFY_ZP(zp);
 
@@ -215,12 +221,6 @@ zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
 	if ((flag & (FSYNC | FDSYNC)) && (count == 1))
 		atomic_dec_32(&zp->z_sync_cnt);
 
-	/*
-	 * Clean up any locks held by this process on the vp.
-	 */
-	cleanlocks(vp, ddi_get_pid(), 0);
-	cleanshares(vp, ddi_get_pid());
-
 	if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan &&
 	    ZTOV(zp)->v_type == VREG &&
 	    !(zp->z_phys->zp_flags & ZFS_AV_QUARANTINED) &&
@@ -855,6 +855,10 @@ zfs_get_done(dmu_buf_t *db, void *vzgd)
 	kmem_free(zgd, sizeof (zgd_t));
 }
 
+#ifdef DEBUG
+static int zil_fault_io = 0;
+#endif
+
 /*
  * Get data to generate a TX_WRITE intent log record.
  */
@@ -936,7 +940,21 @@ zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
 		zgd->zgd_rl = rl;
 		zgd->zgd_zilog = zfsvfs->z_log;
 		zgd->zgd_bp = &lr->lr_blkptr;
-		VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db));
+#ifdef DEBUG
+		if (zil_fault_io) {
+			error = EIO;
+			zil_fault_io = 0;
+		} else {
+			error = dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db);
+		}
+#else
+		error = dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db);
+#endif
+		if (error != 0) {
+			kmem_free(zgd, sizeof (zgd_t));
+			goto out;
+		}
+
 		ASSERT(boff == db->db_offset);
 		lr->lr_blkoff = off - boff;
 		error = dmu_sync(zio, db, &lr->lr_blkptr,
@@ -987,6 +1005,27 @@ zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr,
 	return (error);
 }
 
+/*
+ * If vnode is for a device return a specfs vnode instead.
+ */
+static int
+specvp_check(vnode_t **vpp, cred_t *cr)
+{
+	int error = 0;
+
+	if (IS_DEVVP(*vpp)) {
+		struct vnode *svp;
+
+		svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
+		VN_RELE(*vpp);
+		if (svp == NULL)
+			error = ENOSYS;
+		*vpp = svp;
+	}
+	return (error);
+}
+
+
 /*
  * Lookup an entry in a directory, or an extended attribute directory.
  * If it exists, return a held vnode reference for it.
@@ -1017,7 +1056,46 @@ zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
 {
 	znode_t *zdp = VTOZ(dvp);
 	zfsvfs_t *zfsvfs = zdp->z_zfsvfs;
-	int	error;
+	int	error = 0;
+
+	/* fast path */
+	if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) {
+
+		if (dvp->v_type != VDIR) {
+			return (ENOTDIR);
+		} else if (zdp->z_dbuf == NULL) {
+			return (EIO);
+		}
+
+		if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) {
+			error = zfs_fastaccesschk_execute(zdp, cr);
+			if (!error) {
+				*vpp = dvp;
+				VN_HOLD(*vpp);
+				return (0);
+			}
+			return (error);
+		} else {
+			vnode_t *tvp = dnlc_lookup(dvp, nm);
+
+			if (tvp) {
+				error = zfs_fastaccesschk_execute(zdp, cr);
+				if (error) {
+					VN_RELE(tvp);
+					return (error);
+				}
+				if (tvp == DNLC_NO_VNODE) {
+					VN_RELE(tvp);
+					return (ENOENT);
+				} else {
+					*vpp = tvp;
+					return (specvp_check(vpp, cr));
+				}
+			}
+		}
+	}
+
+	DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm);
 
 	ZFS_ENTER(zfsvfs);
 	ZFS_VERIFY_ZP(zdp);
@@ -1082,21 +1160,8 @@ zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp,
 	}
 
 	error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp);
-	if (error == 0) {
+	if (error == 0)
-		/*
+		error = specvp_check(vpp, cr);
-		 * Convert device special files
-		 */
-		if (IS_DEVVP(*vpp)) {
-			vnode_t	*svp;
-
-			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
-			VN_RELE(*vpp);
-			if (svp == NULL)
-				error = ENOSYS;
-			else
-				*vpp = svp;
-		}
-	}
 
 	ZFS_EXIT(zfsvfs);
 	return (error);
@@ -1235,6 +1300,7 @@ top:
 		    &acl_ids)) != 0)
 			goto out;
 		if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
+			zfs_acl_ids_free(&acl_ids);
 			error = EDQUOT;
 			goto out;
 		}
@@ -1332,19 +1398,7 @@ out:
 			VN_RELE(ZTOV(zp));
 	} else {
 		*vpp = ZTOV(zp);
-		/*
+		error = specvp_check(vpp, cr);
-		 * If vnode is for a device return a specfs vnode instead.
-		 */
-		if (IS_DEVVP(*vpp)) {
-			struct vnode *svp;
-
-			svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
-			VN_RELE(*vpp);
-			if (svp == NULL) {
-				error = ENOSYS;
-			}
-			*vpp = svp;
-		}
 	}
 
 	ZFS_EXIT(zfsvfs);
@@ -1653,6 +1707,7 @@ top:
 		return (error);
 	}
 	if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
+		zfs_acl_ids_free(&acl_ids);
 		zfs_dirent_unlock(dl);
 		ZFS_EXIT(zfsvfs);
 		return (EDQUOT);
@@ -2456,6 +2511,7 @@ zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
 top:
 	attrzp = NULL;
 
+	/* Can this be moved to before the top label? */
 	if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
 		ZFS_EXIT(zfsvfs);
 		return (EROFS);
@@ -2765,6 +2821,8 @@ top:
 		zp->z_phys->zp_mode = new_mode;
 		err = zfs_aclset_common(zp, aclp, cr, tx);
 		ASSERT3U(err, ==, 0);
+		zp->z_acl_cached = aclp;
+		aclp = NULL;
 		mutex_exit(&zp->z_acl_lock);
 	}
 
@@ -2856,10 +2914,8 @@ out:
 	if (attrzp)
 		VN_RELE(ZTOV(attrzp));
 
-	if (aclp) {
+	if (aclp)
 		zfs_acl_free(aclp);
-		aclp = NULL;
-	}
 
 	if (fuidp) {
 		zfs_fuid_info_free(fuidp);
@@ -3724,8 +3780,8 @@ top:
 	if (err == 0) {
 		zfs_time_stamper(zp, CONTENT_MODIFIED, tx);
 		zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0);
-		dmu_tx_commit(tx);
 	}
+	dmu_tx_commit(tx);
 
 out:
 	pvn_write_done(pp, (err ? B_ERROR : 0) | flags);

module/zfs/zfs_znode.c

@@ -133,6 +133,7 @@ zfs_znode_cache_constructor(void *buf, void *arg, int kmflags)
 
 	zp->z_dbuf = NULL;
 	zp->z_dirlocks = NULL;
+	zp->z_acl_cached = NULL;
 	return (0);
 }
 
@@ -155,6 +156,7 @@ zfs_znode_cache_destructor(void *buf, void *arg)
 
 	ASSERT(zp->z_dbuf == NULL);
 	ASSERT(zp->z_dirlocks == NULL);
+	ASSERT(zp->z_acl_cached == NULL);
 }
 
 #ifdef	ZNODE_STATS
@@ -199,6 +201,18 @@ zfs_znode_move_impl(znode_t *ozp, znode_t *nzp)
 	nzp->z_phys = ozp->z_phys;
 	nzp->z_dbuf = ozp->z_dbuf;
 
+	/*
+	 * Release any cached ACL, since it *may* have
+	 * zfs_acl_node_t's that directly references an
+	 * embedded ACL in the zp_acl of the old znode_phys_t
+	 *
+	 * It will be recached the next time the ACL is needed.
+	 */
+	if (ozp->z_acl_cached) {
+		zfs_acl_free(ozp->z_acl_cached);
+		ozp->z_acl_cached = NULL;
+	}
+
 	/* Update back pointers. */
 	(void) dmu_buf_update_user(nzp->z_dbuf, ozp, nzp, &nzp->z_phys,
 	    znode_evict_error);
@@ -1081,6 +1095,11 @@ zfs_znode_free(znode_t *zp)
 	list_remove(&zfsvfs->z_all_znodes, zp);
 	mutex_exit(&zfsvfs->z_znodes_lock);
 
+	if (zp->z_acl_cached) {
+		zfs_acl_free(zp->z_acl_cached);
+		zp->z_acl_cached = NULL;
+	}
+
 	kmem_cache_free(znode_cache, zp);
 
 	VFS_RELE(zfsvfs->z_vfs);

module/zfs/zil.c

@@ -714,14 +714,15 @@ zil_lwb_write_done(zio_t *zio)
 	lwb->lwb_buf = NULL;
 	if (zio->io_error)
 		zilog->zl_log_error = B_TRUE;
-	mutex_exit(&zilog->zl_lock);
 
 	/*
 	 * Now that we've written this log block, we have a stable pointer
 	 * to the next block in the chain, so it's OK to let the txg in
-	 * which we allocated the next block sync.
+	 * which we allocated the next block sync. We still have the
+	 * zl_lock to ensure zil_sync doesn't kmem free the lwb.
 	 */
 	txg_rele_to_sync(&lwb->lwb_txgh);
+	mutex_exit(&zilog->zl_lock);
 }
 
 /*
@@ -925,6 +926,10 @@ zil_lwb_commit(zilog_t *zilog, itx_t *itx, lwb_t *lwb)
 			}
 			error = zilog->zl_get_data(
 			    itx->itx_private, lr, dbuf, lwb->lwb_zio);
+			if (error == EIO) {
+				txg_wait_synced(zilog->zl_dmu_pool, txg);
+				return (lwb);
+			}
 			if (error) {
 				ASSERT(error == ENOENT || error == EEXIST ||
 				    error == EALREADY);