Illumos 4368, 4369.

4369 implement zfs bookmarks
4368 zfs send filesystems from readonly pools
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Garrett D'Amore <garrett@damore.org>

References:
  https://www.illumos.org/issues/4369
  https://www.illumos.org/issues/4368
  https://github.com/illumos/illumos-gate/commit/78f1710

Ported by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #2530

module/zfs/Makefile.in

@@ -25,6 +25,7 @@ $(MODULE)-objs += @top_srcdir@/module/zfs/dnode_sync.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/dsl_dataset.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/dsl_deadlist.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/dsl_deleg.o
+$(MODULE)-objs += @top_srcdir@/module/zfs/dsl_bookmark.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/dsl_dir.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/dsl_pool.o
 $(MODULE)-objs += @top_srcdir@/module/zfs/dsl_prop.o

module/zfs/dmu_diff.c

@@ -187,7 +187,7 @@ dmu_diff(const char *tosnap_name, const char *fromsnap_name,
 		return (error);
 	}
 
-	if (!dsl_dataset_is_before(tosnap, fromsnap)) {
+	if (!dsl_dataset_is_before(tosnap, fromsnap, 0)) {
 		dsl_dataset_rele(fromsnap, FTAG);
 		dsl_dataset_rele(tosnap, FTAG);
 		dsl_pool_rele(dp, FTAG);

module/zfs/dmu_send.c

@@ -50,6 +50,7 @@
 #include <sys/zfs_onexit.h>
 #include <sys/dmu_send.h>
 #include <sys/dsl_destroy.h>
+#include <sys/dsl_bookmark.h>
 
 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
 int zfs_send_corrupt_data = B_FALSE;
@@ -385,6 +386,12 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 	if (zb->zb_object != DMU_META_DNODE_OBJECT &&
 	    DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
 		return (0);
+	} else if (zb->zb_level == ZB_ZIL_LEVEL) {
+		/*
+		 * If we are sending a non-snapshot (which is allowed on
+		 * read-only pools), it may have a ZIL, which must be ignored.
+		 */
+		return (0);
 	} else if (BP_IS_HOLE(bp) &&
 	    zb->zb_object == DMU_META_DNODE_OBJECT) {
 		uint64_t span = BP_SPAN(dnp, zb->zb_level);
@@ -433,6 +440,7 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 		arc_buf_t *abuf;
 		int blksz = BP_GET_LSIZE(bp);
 
+		ASSERT0(zb->zb_level);
 		if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
 		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
 		    &aflags, zb) != 0) {
@@ -460,11 +468,12 @@ backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
 }
 
 /*
- * Releases dp, ds, and fromds, using the specified tag.
+ * Releases dp using the specified tag.
  */
 static int
 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
-    dsl_dataset_t *fromds, int outfd, vnode_t *vp, offset_t *off)
+    zfs_bookmark_phys_t *fromzb, boolean_t is_clone, int outfd,
+    vnode_t *vp, offset_t *off)
 {
 	objset_t *os;
 	dmu_replay_record_t *drr;
@@ -472,18 +481,8 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 	int err;
 	uint64_t fromtxg = 0;
 
-	if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) {
-		dsl_dataset_rele(fromds, tag);
-		dsl_dataset_rele(ds, tag);
-		dsl_pool_rele(dp, tag);
-		return (SET_ERROR(EXDEV));
-	}
-
 	err = dmu_objset_from_ds(ds, &os);
 	if (err != 0) {
-		if (fromds != NULL)
-			dsl_dataset_rele(fromds, tag);
-		dsl_dataset_rele(ds, tag);
 		dsl_pool_rele(dp, tag);
 		return (err);
 	}
@@ -499,9 +498,6 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 		uint64_t version;
 		if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
 			kmem_free(drr, sizeof (dmu_replay_record_t));
-			if (fromds != NULL)
-				dsl_dataset_rele(fromds, tag);
-			dsl_dataset_rele(ds, tag);
 			dsl_pool_rele(dp, tag);
 			return (SET_ERROR(EINVAL));
 		}
@@ -516,20 +512,20 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 	drr->drr_u.drr_begin.drr_creation_time =
 	    ds->ds_phys->ds_creation_time;
 	drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
-	if (fromds != NULL && ds->ds_dir != fromds->ds_dir)
+	if (is_clone)
 		drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
 	drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
 		drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
 
-	if (fromds != NULL)
-		drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
+	if (fromzb != NULL) {
+		drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
+		fromtxg = fromzb->zbm_creation_txg;
+	}
 	dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
-
-	if (fromds != NULL) {
-		fromtxg = fromds->ds_phys->ds_creation_txg;
-		dsl_dataset_rele(fromds, tag);
-		fromds = NULL;
+	if (!dsl_dataset_is_snapshot(ds)) {
+		(void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
+		    sizeof (drr->drr_u.drr_begin.drr_toname));
 	}
 
 	dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
@@ -543,7 +539,7 @@ dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
 	dsp->dsa_toguid = ds->ds_phys->ds_guid;
 	ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
 	dsp->dsa_pending_op = PENDING_NONE;
-	dsp->dsa_incremental = (fromtxg != 0);
+	dsp->dsa_incremental = (fromzb != NULL);
 
 	mutex_enter(&ds->ds_sendstream_lock);
 	list_insert_head(&ds->ds_sendstreams, dsp);
@@ -589,7 +585,6 @@ out:
 	kmem_free(dsp, sizeof (dmu_sendarg_t));
 
 	dsl_dataset_long_rele(ds, FTAG);
-	dsl_dataset_rele(ds, tag);
 
 	return (err);
 }
@@ -614,15 +609,30 @@ dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
 	}
 
 	if (fromsnap != 0) {
+		zfs_bookmark_phys_t zb;
+		boolean_t is_clone;
+
 		err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
 		if (err != 0) {
 			dsl_dataset_rele(ds, FTAG);
 			dsl_pool_rele(dp, FTAG);
 			return (err);
 		}
+		if (!dsl_dataset_is_before(ds, fromds, 0))
+			err = SET_ERROR(EXDEV);
+		zb.zbm_creation_time = fromds->ds_phys->ds_creation_time;
+		zb.zbm_creation_txg = fromds->ds_phys->ds_creation_txg;
+		zb.zbm_guid = fromds->ds_phys->ds_guid;
+		is_clone = (fromds->ds_dir != ds->ds_dir);
+		dsl_dataset_rele(fromds, FTAG);
+		err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
+		    outfd, vp, off);
+	} else {
+		err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
+		    outfd, vp, off);
 	}
-
-	return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off));
+	dsl_dataset_rele(ds, FTAG);
+	return (err);
 }
 
 int
@@ -631,33 +641,79 @@ dmu_send(const char *tosnap, const char *fromsnap,
 {
 	dsl_pool_t *dp;
 	dsl_dataset_t *ds;
-	dsl_dataset_t *fromds = NULL;
 	int err;
+	boolean_t owned = B_FALSE;
 
-	if (strchr(tosnap, '@') == NULL)
-		return (SET_ERROR(EINVAL));
-	if (fromsnap != NULL && strchr(fromsnap, '@') == NULL)
+	if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
 		return (SET_ERROR(EINVAL));
 
 	err = dsl_pool_hold(tosnap, FTAG, &dp);
 	if (err != 0)
 		return (err);
 
-	err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
+	if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
+		/*
+		 * We are sending a filesystem or volume.  Ensure
+		 * that it doesn't change by owning the dataset.
+		 */
+		err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
+		owned = B_TRUE;
+	} else {
+		err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
+	}
 	if (err != 0) {
 		dsl_pool_rele(dp, FTAG);
 		return (err);
 	}
 
 	if (fromsnap != NULL) {
-		err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
+		zfs_bookmark_phys_t zb;
+		boolean_t is_clone = B_FALSE;
+		int fsnamelen = strchr(tosnap, '@') - tosnap;
+
+		/*
+		 * If the fromsnap is in a different filesystem, then
+		 * mark the send stream as a clone.
+		 */
+		if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
+		    (fromsnap[fsnamelen] != '@' &&
+		    fromsnap[fsnamelen] != '#')) {
+			is_clone = B_TRUE;
+		}
+
+		if (strchr(fromsnap, '@')) {
+			dsl_dataset_t *fromds;
+			err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
+			if (err == 0) {
+				if (!dsl_dataset_is_before(ds, fromds, 0))
+					err = SET_ERROR(EXDEV);
+				zb.zbm_creation_time =
+				    fromds->ds_phys->ds_creation_time;
+				zb.zbm_creation_txg =
+				    fromds->ds_phys->ds_creation_txg;
+				zb.zbm_guid = fromds->ds_phys->ds_guid;
+				is_clone = (ds->ds_dir != fromds->ds_dir);
+				dsl_dataset_rele(fromds, FTAG);
+			}
+		} else {
+			err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
+		}
 		if (err != 0) {
 			dsl_dataset_rele(ds, FTAG);
 			dsl_pool_rele(dp, FTAG);
 			return (err);
 		}
+		err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
+		    outfd, vp, off);
+	} else {
+		err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
+		    outfd, vp, off);
 	}
-	return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off));
+	if (owned)
+		dsl_dataset_disown(ds, FTAG);
+	else
+		dsl_dataset_rele(ds, FTAG);
+	return (err);
 }
 
 int
@@ -677,7 +733,7 @@ dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
 	 * fromsnap must be an earlier snapshot from the same fs as tosnap,
 	 * or the origin's fs.
 	 */
-	if (fromds != NULL && !dsl_dataset_is_before(ds, fromds))
+	if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
 		return (SET_ERROR(EXDEV));
 
 	/* Get uncompressed size estimate of changed data. */

module/zfs/dsl_bookmark.c

@@ -0,0 +1,459 @@
+/*
+ * CDDL HEADER START
+ *
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright (c) 2013 by Delphix. All rights reserved.
+ */
+
+#include <sys/zfs_context.h>
+#include <sys/dsl_dataset.h>
+#include <sys/dsl_dir.h>
+#include <sys/dsl_prop.h>
+#include <sys/dsl_synctask.h>
+#include <sys/dmu_impl.h>
+#include <sys/dmu_tx.h>
+#include <sys/arc.h>
+#include <sys/zap.h>
+#include <sys/zfeature.h>
+#include <sys/spa.h>
+#include <sys/dsl_bookmark.h>
+#include <zfs_namecheck.h>
+
+static int
+dsl_bookmark_hold_ds(dsl_pool_t *dp, const char *fullname,
+    dsl_dataset_t **dsp, void *tag, char **shortnamep)
+{
+	char buf[MAXNAMELEN];
+	char *hashp;
+
+	if (strlen(fullname) >= MAXNAMELEN)
+		return (SET_ERROR(ENAMETOOLONG));
+	hashp = strchr(fullname, '#');
+	if (hashp == NULL)
+		return (SET_ERROR(EINVAL));
+
+	*shortnamep = hashp + 1;
+	if (zfs_component_namecheck(*shortnamep, NULL, NULL))
+		return (SET_ERROR(EINVAL));
+	(void) strlcpy(buf, fullname, hashp - fullname + 1);
+	return (dsl_dataset_hold(dp, buf, tag, dsp));
+}
+
+/*
+ * Returns ESRCH if bookmark is not found.
+ */
+static int
+dsl_dataset_bmark_lookup(dsl_dataset_t *ds, const char *shortname,
+    zfs_bookmark_phys_t *bmark_phys)
+{
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	uint64_t bmark_zapobj = ds->ds_bookmarks;
+	matchtype_t mt;
+	int err;
+
+	if (bmark_zapobj == 0)
+		return (SET_ERROR(ESRCH));
+
+	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
+		mt = MT_FIRST;
+	else
+		mt = MT_EXACT;
+
+	err = zap_lookup_norm(mos, bmark_zapobj, shortname, sizeof (uint64_t),
+	    sizeof (*bmark_phys) / sizeof (uint64_t), bmark_phys, mt,
+	    NULL, 0, NULL);
+
+	return (err == ENOENT ? ESRCH : err);
+}
+
+/*
+ * If later_ds is non-NULL, this will return EXDEV if the the specified bookmark
+ * does not represents an earlier point in later_ds's timeline.
+ *
+ * Returns ENOENT if the dataset containing the bookmark does not exist.
+ * Returns ESRCH if the dataset exists but the bookmark was not found in it.
+ */
+int
+dsl_bookmark_lookup(dsl_pool_t *dp, const char *fullname,
+    dsl_dataset_t *later_ds, zfs_bookmark_phys_t *bmp)
+{
+	char *shortname;
+	dsl_dataset_t *ds;
+	int error;
+
+	error = dsl_bookmark_hold_ds(dp, fullname, &ds, FTAG, &shortname);
+	if (error != 0)
+		return (error);
+
+	error = dsl_dataset_bmark_lookup(ds, shortname, bmp);
+	if (error == 0 && later_ds != NULL) {
+		if (!dsl_dataset_is_before(later_ds, ds, bmp->zbm_creation_txg))
+			error = SET_ERROR(EXDEV);
+	}
+	dsl_dataset_rele(ds, FTAG);
+	return (error);
+}
+
+typedef struct dsl_bookmark_create_arg {
+	nvlist_t *dbca_bmarks;
+	nvlist_t *dbca_errors;
+} dsl_bookmark_create_arg_t;
+
+static int
+dsl_bookmark_create_check_impl(dsl_dataset_t *snapds, const char *bookmark_name,
+    dmu_tx_t *tx)
+{
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	dsl_dataset_t *bmark_fs;
+	char *shortname;
+	int error;
+	zfs_bookmark_phys_t bmark_phys;
+
+	if (!dsl_dataset_is_snapshot(snapds))
+		return (SET_ERROR(EINVAL));
+
+	error = dsl_bookmark_hold_ds(dp, bookmark_name,
+	    &bmark_fs, FTAG, &shortname);
+	if (error != 0)
+		return (error);
+
+	if (!dsl_dataset_is_before(bmark_fs, snapds, 0)) {
+		dsl_dataset_rele(bmark_fs, FTAG);
+		return (SET_ERROR(EINVAL));
+	}
+
+	error = dsl_dataset_bmark_lookup(bmark_fs, shortname,
+	    &bmark_phys);
+	dsl_dataset_rele(bmark_fs, FTAG);
+	if (error == 0)
+		return (SET_ERROR(EEXIST));
+	if (error == ESRCH)
+		return (0);
+	return (error);
+}
+
+static int
+dsl_bookmark_create_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_bookmark_create_arg_t *dbca = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	int rv = 0;
+	nvpair_t *pair;
+
+	if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARKS))
+		return (SET_ERROR(ENOTSUP));
+
+	for (pair = nvlist_next_nvpair(dbca->dbca_bmarks, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(dbca->dbca_bmarks, pair)) {
+		dsl_dataset_t *snapds;
+		int error;
+
+		/* note: validity of nvlist checked by ioctl layer */
+		error = dsl_dataset_hold(dp, fnvpair_value_string(pair),
+		    FTAG, &snapds);
+		if (error == 0) {
+			error = dsl_bookmark_create_check_impl(snapds,
+			    nvpair_name(pair), tx);
+			dsl_dataset_rele(snapds, FTAG);
+		}
+		if (error != 0) {
+			fnvlist_add_int32(dbca->dbca_errors,
+			    nvpair_name(pair), error);
+			rv = error;
+		}
+	}
+
+	return (rv);
+}
+
+static void
+dsl_bookmark_create_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_bookmark_create_arg_t *dbca = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	objset_t *mos = dp->dp_meta_objset;
+	nvpair_t *pair;
+
+	ASSERT(spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARKS));
+
+	for (pair = nvlist_next_nvpair(dbca->dbca_bmarks, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(dbca->dbca_bmarks, pair)) {
+		dsl_dataset_t *snapds, *bmark_fs;
+		zfs_bookmark_phys_t bmark_phys;
+		char *shortname;
+
+		VERIFY0(dsl_dataset_hold(dp, fnvpair_value_string(pair),
+		    FTAG, &snapds));
+		VERIFY0(dsl_bookmark_hold_ds(dp, nvpair_name(pair),
+		    &bmark_fs, FTAG, &shortname));
+		if (bmark_fs->ds_bookmarks == 0) {
+			bmark_fs->ds_bookmarks =
+			    zap_create_norm(mos, U8_TEXTPREP_TOUPPER,
+			    DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
+			spa_feature_incr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
+
+			dsl_dataset_zapify(bmark_fs, tx);
+			VERIFY0(zap_add(mos, bmark_fs->ds_object,
+			    DS_FIELD_BOOKMARK_NAMES,
+			    sizeof (bmark_fs->ds_bookmarks), 1,
+			    &bmark_fs->ds_bookmarks, tx));
+		}
+
+		bmark_phys.zbm_guid = snapds->ds_phys->ds_guid;
+		bmark_phys.zbm_creation_txg = snapds->ds_phys->ds_creation_txg;
+		bmark_phys.zbm_creation_time =
+		    snapds->ds_phys->ds_creation_time;
+
+		VERIFY0(zap_add(mos, bmark_fs->ds_bookmarks,
+		    shortname, sizeof (uint64_t),
+		    sizeof (zfs_bookmark_phys_t) / sizeof (uint64_t),
+		    &bmark_phys, tx));
+
+		spa_history_log_internal_ds(bmark_fs, "bookmark", tx,
+		    "name=%s creation_txg=%llu target_snap=%llu",
+		    shortname,
+		    (longlong_t)bmark_phys.zbm_creation_txg,
+		    (longlong_t)snapds->ds_object);
+
+		dsl_dataset_rele(bmark_fs, FTAG);
+		dsl_dataset_rele(snapds, FTAG);
+	}
+}
+
+/*
+ * The bookmarks must all be in the same pool.
+ */
+int
+dsl_bookmark_create(nvlist_t *bmarks, nvlist_t *errors)
+{
+	nvpair_t *pair;
+	dsl_bookmark_create_arg_t dbca;
+
+	pair = nvlist_next_nvpair(bmarks, NULL);
+	if (pair == NULL)
+		return (0);
+
+	dbca.dbca_bmarks = bmarks;
+	dbca.dbca_errors = errors;
+
+	return (dsl_sync_task(nvpair_name(pair), dsl_bookmark_create_check,
+	    dsl_bookmark_create_sync, &dbca, fnvlist_num_pairs(bmarks)));
+}
+
+int
+dsl_get_bookmarks_impl(dsl_dataset_t *ds, nvlist_t *props, nvlist_t *outnvl)
+{
+	int err = 0;
+	zap_cursor_t zc;
+	zap_attribute_t attr;
+	dsl_pool_t *dp = ds->ds_dir->dd_pool;
+
+	uint64_t bmark_zapobj = ds->ds_bookmarks;
+	if (bmark_zapobj == 0)
+		return (0);
+
+	for (zap_cursor_init(&zc, dp->dp_meta_objset, bmark_zapobj);
+	    zap_cursor_retrieve(&zc, &attr) == 0;
+	    zap_cursor_advance(&zc)) {
+		nvlist_t *out_props;
+		char *bmark_name = attr.za_name;
+		zfs_bookmark_phys_t bmark_phys;
+
+		err = dsl_dataset_bmark_lookup(ds, bmark_name, &bmark_phys);
+		ASSERT3U(err, !=, ENOENT);
+		if (err != 0)
+			break;
+
+		out_props = fnvlist_alloc();
+		if (nvlist_exists(props,
+		    zfs_prop_to_name(ZFS_PROP_GUID))) {
+			dsl_prop_nvlist_add_uint64(out_props,
+			    ZFS_PROP_GUID, bmark_phys.zbm_guid);
+		}
+		if (nvlist_exists(props,
+		    zfs_prop_to_name(ZFS_PROP_CREATETXG))) {
+			dsl_prop_nvlist_add_uint64(out_props,
+			    ZFS_PROP_CREATETXG, bmark_phys.zbm_creation_txg);
+		}
+		if (nvlist_exists(props,
+		    zfs_prop_to_name(ZFS_PROP_CREATION))) {
+			dsl_prop_nvlist_add_uint64(out_props,
+			    ZFS_PROP_CREATION, bmark_phys.zbm_creation_time);
+		}
+
+		fnvlist_add_nvlist(outnvl, bmark_name, out_props);
+		fnvlist_free(out_props);
+	}
+	zap_cursor_fini(&zc);
+	return (err);
+}
+
+/*
+ * Retrieve the bookmarks that exist in the specified dataset, and the
+ * requested properties of each bookmark.
+ *
+ * The "props" nvlist specifies which properties are requested.
+ * See lzc_get_bookmarks() for the list of valid properties.
+ */
+int
+dsl_get_bookmarks(const char *dsname, nvlist_t *props, nvlist_t *outnvl)
+{
+	dsl_pool_t *dp;
+	dsl_dataset_t *ds;
+	int err;
+
+	err = dsl_pool_hold(dsname, FTAG, &dp);
+	if (err != 0)
+		return (err);
+	err = dsl_dataset_hold(dp, dsname, FTAG, &ds);
+	if (err != 0) {
+		dsl_pool_rele(dp, FTAG);
+		return (err);
+	}
+
+	err = dsl_get_bookmarks_impl(ds, props, outnvl);
+
+	dsl_dataset_rele(ds, FTAG);
+	dsl_pool_rele(dp, FTAG);
+	return (err);
+}
+
+typedef struct dsl_bookmark_destroy_arg {
+	nvlist_t *dbda_bmarks;
+	nvlist_t *dbda_success;
+	nvlist_t *dbda_errors;
+} dsl_bookmark_destroy_arg_t;
+
+static int
+dsl_dataset_bookmark_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx)
+{
+	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
+	uint64_t bmark_zapobj = ds->ds_bookmarks;
+	matchtype_t mt;
+
+	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
+		mt = MT_FIRST;
+	else
+		mt = MT_EXACT;
+
+	return (zap_remove_norm(mos, bmark_zapobj, name, mt, tx));
+}
+
+static int
+dsl_bookmark_destroy_check(void *arg, dmu_tx_t *tx)
+{
+	dsl_bookmark_destroy_arg_t *dbda = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	int rv = 0;
+	nvpair_t *pair;
+
+	if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARKS))
+		return (0);
+
+	for (pair = nvlist_next_nvpair(dbda->dbda_bmarks, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(dbda->dbda_bmarks, pair)) {
+		const char *fullname = nvpair_name(pair);
+		dsl_dataset_t *ds;
+		zfs_bookmark_phys_t bm;
+		int error;
+		char *shortname;
+
+		error = dsl_bookmark_hold_ds(dp, fullname, &ds,
+		    FTAG, &shortname);
+		if (error == ENOENT) {
+			/* ignore it; the bookmark is "already destroyed" */
+			continue;
+		}
+		if (error == 0) {
+			error = dsl_dataset_bmark_lookup(ds, shortname, &bm);
+			dsl_dataset_rele(ds, FTAG);
+			if (error == ESRCH) {
+				/*
+				 * ignore it; the bookmark is
+				 * "already destroyed"
+				 */
+				continue;
+			}
+		}
+		if (error == 0) {
+			fnvlist_add_boolean(dbda->dbda_success, fullname);
+		} else {
+			fnvlist_add_int32(dbda->dbda_errors, fullname, error);
+			rv = error;
+		}
+	}
+	return (rv);
+}
+
+static void
+dsl_bookmark_destroy_sync(void *arg, dmu_tx_t *tx)
+{
+	dsl_bookmark_destroy_arg_t *dbda = arg;
+	dsl_pool_t *dp = dmu_tx_pool(tx);
+	objset_t *mos = dp->dp_meta_objset;
+	nvpair_t *pair;
+
+	for (pair = nvlist_next_nvpair(dbda->dbda_success, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(dbda->dbda_success, pair)) {
+		dsl_dataset_t *ds;
+		char *shortname;
+		uint64_t zap_cnt;
+
+		VERIFY0(dsl_bookmark_hold_ds(dp, nvpair_name(pair),
+		    &ds, FTAG, &shortname));
+		VERIFY0(dsl_dataset_bookmark_remove(ds, shortname, tx));
+
+		/*
+		 * If all of this dataset's bookmarks have been destroyed,
+		 * free the zap object and decrement the feature's use count.
+		 */
+		VERIFY0(zap_count(mos, ds->ds_bookmarks,
+		    &zap_cnt));
+		if (zap_cnt == 0) {
+			dmu_buf_will_dirty(ds->ds_dbuf, tx);
+			VERIFY0(zap_destroy(mos, ds->ds_bookmarks, tx));
+			ds->ds_bookmarks = 0;
+			spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
+			VERIFY0(zap_remove(mos, ds->ds_object,
+			    DS_FIELD_BOOKMARK_NAMES, tx));
+		}
+
+		spa_history_log_internal_ds(ds, "remove bookmark", tx,
+		    "name=%s", shortname);
+
+		dsl_dataset_rele(ds, FTAG);
+	}
+}
+
+/*
+ * The bookmarks must all be in the same pool.
+ */
+int
+dsl_bookmark_destroy(nvlist_t *bmarks, nvlist_t *errors)
+{
+	int rv;
+	dsl_bookmark_destroy_arg_t dbda;
+	nvpair_t *pair = nvlist_next_nvpair(bmarks, NULL);
+	if (pair == NULL)
+		return (0);
+
+	dbda.dbda_bmarks = bmarks;
+	dbda.dbda_errors = errors;
+	dbda.dbda_success = fnvlist_alloc();
+
+	rv = dsl_sync_task(nvpair_name(pair), dsl_bookmark_destroy_check,
+	    dsl_bookmark_destroy_sync, &dbda, fnvlist_num_pairs(bmarks));
+	fnvlist_free(dbda.dbda_success);
+	return (rv);
+}

module/zfs/dsl_dataset.c

@@ -48,6 +48,7 @@
 #include <sys/dsl_deadlist.h>
 #include <sys/dsl_destroy.h>
 #include <sys/dsl_userhold.h>
+#include <sys/dsl_bookmark.h>
 
 #define	SWITCH64(x, y) \
 	{ \
@@ -404,6 +405,14 @@ dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
 				    ds->ds_phys->ds_prev_snap_obj,
 				    ds, &ds->ds_prev);
 			}
+			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
+				int zaperr = zap_lookup(mos, ds->ds_object,
+				    DS_FIELD_BOOKMARK_NAMES,
+				    sizeof (ds->ds_bookmarks), 1,
+				    &ds->ds_bookmarks);
+				if (zaperr != ENOENT)
+					VERIFY0(zaperr);
+			}
 		} else {
 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
 				err = dsl_dataset_get_snapname(ds);
@@ -1734,6 +1743,8 @@ dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
 	dsl_dataset_t *ds;
 	int64_t unused_refres_delta;
 	int error;
+	nvpair_t *pair;
+	nvlist_t *proprequest, *bookmarks;
 
 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
 	if (error != 0)
@@ -1751,6 +1762,28 @@ dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
 		return (SET_ERROR(EINVAL));
 	}
 
+	/* must not have any bookmarks after the most recent snapshot */
+	proprequest = fnvlist_alloc();
+	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
+	bookmarks = fnvlist_alloc();
+	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
+	fnvlist_free(proprequest);
+	if (error != 0)
+		return (error);
+	for (pair = nvlist_next_nvpair(bookmarks, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
+		nvlist_t *valuenv =
+		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
+		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
+		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
+		if (createtxg > ds->ds_phys->ds_prev_snap_txg) {
+			fnvlist_free(bookmarks);
+			dsl_dataset_rele(ds, FTAG);
+			return (SET_ERROR(EEXIST));
+		}
+	}
+	fnvlist_free(bookmarks);
+
 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
 	if (error != 0) {
 		dsl_dataset_rele(ds, FTAG);
@@ -2972,9 +3005,12 @@ dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
  * filesystem.  Or 'earlier' could be the origin's origin.
+ *
+ * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
  */
 boolean_t
-dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
+dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
+	uint64_t earlier_txg)
 {
 	dsl_pool_t *dp = later->ds_dir->dd_pool;
 	int error;
@@ -2982,9 +3018,13 @@ dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
 	dsl_dataset_t *origin;
 
 	ASSERT(dsl_pool_config_held(dp));
+	ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
 
-	if (earlier->ds_phys->ds_creation_txg >=
-	    later->ds_phys->ds_creation_txg)
+	if (earlier_txg == 0)
+		earlier_txg = earlier->ds_phys->ds_creation_txg;
+
+	if (dsl_dataset_is_snapshot(later) &&
+	    earlier_txg >= later->ds_phys->ds_creation_txg)
 		return (B_FALSE);
 
 	if (later->ds_dir == earlier->ds_dir)
@@ -2998,7 +3038,7 @@ dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
 	    later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
 	if (error != 0)
 		return (B_FALSE);
-	ret = dsl_dataset_is_before(origin, earlier);
+	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
 	dsl_dataset_rele(origin, FTAG);
 	return (ret);
 }

module/zfs/dsl_destroy.c

@@ -816,6 +816,12 @@ dsl_destroy_head_sync_impl(dsl_dataset_t *ds, dmu_tx_t *tx)
 	ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
 	VERIFY0(zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx));
 
+	if (ds->ds_bookmarks != 0) {
+		VERIFY0(zap_destroy(mos,
+		    ds->ds_bookmarks, tx));
+		spa_feature_decr(dp->dp_spa, SPA_FEATURE_BOOKMARKS, tx);
+	}
+
 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
 
 	ASSERT0(ds->ds_phys->ds_next_clones_obj);

module/zfs/spa_misc.c

@@ -428,7 +428,7 @@ spa_lookup(const char *name)
 	 * If it's a full dataset name, figure out the pool name and
 	 * just use that.
 	 */
-	cp = strpbrk(search.spa_name, "/@");
+	cp = strpbrk(search.spa_name, "/@#");
 	if (cp != NULL)
 		*cp = '\0';
 

module/zfs/zfeature_common.c

@@ -187,8 +187,10 @@ zpool_feature_init(void)
 	    B_FALSE, NULL);
 
 	{
-	static spa_feature_t hole_birth_deps[] = { SPA_FEATURE_ENABLED_TXG,
-	    SPA_FEATURE_NONE };
+	static const spa_feature_t hole_birth_deps[] = {
+		SPA_FEATURE_ENABLED_TXG,
+		SPA_FEATURE_NONE
+	};
 	zfeature_register(SPA_FEATURE_HOLE_BIRTH,
 	    "com.delphix:hole_birth", "hole_birth",
 	    "Retain hole birth txg for more precise zfs send",
@@ -199,4 +201,16 @@ zpool_feature_init(void)
 	    "com.delphix:extensible_dataset", "extensible_dataset",
 	    "Enhanced dataset functionality, used by other features.",
 	    B_FALSE, B_FALSE, B_FALSE, NULL);
+
+	{
+	static const spa_feature_t bookmarks_deps[] = {
+		SPA_FEATURE_EXTENSIBLE_DATASET,
+		SPA_FEATURE_NONE
+	};
+
+	zfeature_register(SPA_FEATURE_BOOKMARKS,
+	    "com.delphix:bookmarks", "bookmarks",
+	    "\"zfs bookmark\" command",
+	    B_TRUE, B_FALSE, B_FALSE, bookmarks_deps);
+	}
 }

module/zfs/zfs_ctldir.c

@@ -354,7 +354,7 @@ zfsctl_snapshot_zname(struct inode *ip, const char *name, int len, char *zname)
 {
 	objset_t *os = ITOZSB(ip)->z_os;
 
-	if (snapshot_namecheck(name, NULL, NULL) != 0)
+	if (zfs_component_namecheck(name, NULL, NULL) != 0)
 		return (SET_ERROR(EILSEQ));
 
 	dmu_objset_name(os, zname);
@@ -630,7 +630,7 @@ zfsctl_snapdir_mkdir(struct inode *dip, char *dirname, vattr_t *vap,
 
 	dsname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
 
-	if (snapshot_namecheck(dirname, NULL, NULL) != 0) {
+	if (zfs_component_namecheck(dirname, NULL, NULL) != 0) {
 		error = SET_ERROR(EILSEQ);
 		goto out;
 	}

module/zfs/zfs_ioctl.c

@@ -180,6 +180,7 @@
 
 #include <sys/dmu_send.h>
 #include <sys/dsl_destroy.h>
+#include <sys/dsl_bookmark.h>
 #include <sys/dsl_userhold.h>
 #include <sys/zfeature.h>
 
@@ -812,22 +813,9 @@ zfs_secpolicy_destroy_snaps(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr)
 		return (SET_ERROR(EINVAL));
 	for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
 	    pair = nextpair) {
-		dsl_pool_t *dp;
-		dsl_dataset_t *ds;
-
-		error = dsl_pool_hold(nvpair_name(pair), FTAG, &dp);
-		if (error != 0)
-			break;
 		nextpair = nvlist_next_nvpair(snaps, pair);
-		error = dsl_dataset_hold(dp, nvpair_name(pair), FTAG, &ds);
-		if (error == 0)
-			dsl_dataset_rele(ds, FTAG);
-		dsl_pool_rele(dp, FTAG);
-
-		if (error == 0) {
-			error = zfs_secpolicy_destroy_perms(nvpair_name(pair),
-			    cr);
-		} else if (error == ENOENT) {
+		error = zfs_secpolicy_destroy_perms(nvpair_name(pair), cr);
+		if (error == ENOENT) {
 			/*
 			 * Ignore any snapshots that don't exist (we consider
 			 * them "already destroyed").  Remove the name from the
@@ -986,6 +974,76 @@ zfs_secpolicy_snapshot(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr)
 	return (error);
 }
 
+/*
+ * Check for permission to create each snapshot in the nvlist.
+ */
+/* ARGSUSED */
+static int
+zfs_secpolicy_bookmark(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr)
+{
+	int error = 0;
+	nvpair_t *pair;
+
+	for (pair = nvlist_next_nvpair(innvl, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(innvl, pair)) {
+		char *name = nvpair_name(pair);
+		char *hashp = strchr(name, '#');
+
+		if (hashp == NULL) {
+			error = SET_ERROR(EINVAL);
+			break;
+		}
+		*hashp = '\0';
+		error = zfs_secpolicy_write_perms(name,
+		    ZFS_DELEG_PERM_BOOKMARK, cr);
+		*hashp = '#';
+		if (error != 0)
+			break;
+	}
+	return (error);
+}
+
+/* ARGSUSED */
+static int
+zfs_secpolicy_destroy_bookmarks(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr)
+{
+	nvpair_t *pair, *nextpair;
+	int error = 0;
+
+	for (pair = nvlist_next_nvpair(innvl, NULL); pair != NULL;
+	    pair = nextpair) {
+		char *name = nvpair_name(pair);
+		char *hashp = strchr(name, '#');
+		nextpair = nvlist_next_nvpair(innvl, pair);
+
+		if (hashp == NULL) {
+			error = SET_ERROR(EINVAL);
+			break;
+		}
+
+		*hashp = '\0';
+		error = zfs_secpolicy_write_perms(name,
+		    ZFS_DELEG_PERM_DESTROY, cr);
+		*hashp = '#';
+		if (error == ENOENT) {
+			/*
+			 * Ignore any filesystems that don't exist (we consider
+			 * their bookmarks "already destroyed").  Remove
+			 * the name from the nvl here in case the filesystem
+			 * is created between now and when we try to destroy
+			 * the bookmark (in which case we don't want to
+			 * destroy it since we haven't checked for permission).
+			 */
+			fnvlist_remove_nvpair(innvl, pair);
+			error = 0;
+		}
+		if (error != 0)
+			break;
+	}
+
+	return (error);
+}
+
 /* ARGSUSED */
 static int
 zfs_secpolicy_log_history(zfs_cmd_t *zc, nvlist_t *innvl, cred_t *cr)
@@ -2551,7 +2609,6 @@ zfs_check_userprops(const char *fsname, nvlist_t *nvl)
 
 	while ((pair = nvlist_next_nvpair(nvl, pair)) != NULL) {
 		const char *propname = nvpair_name(pair);
-		char *valstr;
 
 		if (!zfs_prop_user(propname) ||
 		    nvpair_type(pair) != DATA_TYPE_STRING)
@@ -2564,8 +2621,7 @@ zfs_check_userprops(const char *fsname, nvlist_t *nvl)
 		if (strlen(propname) >= ZAP_MAXNAMELEN)
 			return (SET_ERROR(ENAMETOOLONG));
 
-		VERIFY(nvpair_value_string(pair, &valstr) == 0);
-		if (strlen(valstr) >= ZAP_MAXVALUELEN)
+		if (strlen(fnvpair_value_string(pair)) >= ZAP_MAXVALUELEN)
 			return (SET_ERROR(E2BIG));
 	}
 	return (0);
@@ -3242,7 +3298,8 @@ zfs_ioc_snapshot(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl)
 		 * The snap name must contain an @, and the part after it must
 		 * contain only valid characters.
 		 */
-		if (cp == NULL || snapshot_namecheck(cp + 1, NULL, NULL) != 0)
+		if (cp == NULL ||
+		    zfs_component_namecheck(cp + 1, NULL, NULL) != 0)
 			return (SET_ERROR(EINVAL));
 
 		/*
@@ -3396,10 +3453,10 @@ zfs_destroy_unmount_origin(const char *fsname)
  *
  * outnvl: snapshot -> error code (int32)
  */
+/* ARGSUSED */
 static int
 zfs_ioc_destroy_snaps(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl)
 {
-	int error, poollen;
 	nvlist_t *snaps;
 	nvpair_t *pair;
 	boolean_t defer;
@@ -3408,27 +3465,112 @@ zfs_ioc_destroy_snaps(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl)
 		return (SET_ERROR(EINVAL));
 	defer = nvlist_exists(innvl, "defer");
 
-	poollen = strlen(poolname);
 	for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
 	    pair = nvlist_next_nvpair(snaps, pair)) {
-		const char *name = nvpair_name(pair);
-
-		/*
-		 * The snap must be in the specified pool.
-		 */
-		if (strncmp(name, poolname, poollen) != 0 ||
-		    (name[poollen] != '/' && name[poollen] != '@'))
-			return (SET_ERROR(EXDEV));
-
-		error = zfs_unmount_snap(name);
-		if (error != 0)
-			return (error);
-		(void) zvol_remove_minor(name);
+		(void) zfs_unmount_snap(nvpair_name(pair));
+		(void) zvol_remove_minor(nvpair_name(pair));
 	}
 
 	return (dsl_destroy_snapshots_nvl(snaps, defer, outnvl));
 }
 
+/*
+ * Create bookmarks.  Bookmark names are of the form <fs>#<bmark>.
+ * All bookmarks must be in the same pool.
+ *
+ * innvl: {
+ *     bookmark1 -> snapshot1, bookmark2 -> snapshot2
+ * }
+ *
+ * outnvl: bookmark -> error code (int32)
+ *
+ */
+/* ARGSUSED */
+static int
+zfs_ioc_bookmark(const char *poolname, nvlist_t *innvl, nvlist_t *outnvl)
+{
+	nvpair_t *pair, *pair2;
+
+	for (pair = nvlist_next_nvpair(innvl, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(innvl, pair)) {
+		char *snap_name;
+
+		/*
+		 * Verify the snapshot argument.
+		 */
+		if (nvpair_value_string(pair, &snap_name) != 0)
+			return (SET_ERROR(EINVAL));
+
+
+		/* Verify that the keys (bookmarks) are unique */
+		for (pair2 = nvlist_next_nvpair(innvl, pair);
+		    pair2 != NULL; pair2 = nvlist_next_nvpair(innvl, pair2)) {
+			if (strcmp(nvpair_name(pair), nvpair_name(pair2)) == 0)
+				return (SET_ERROR(EINVAL));
+		}
+	}
+
+	return (dsl_bookmark_create(innvl, outnvl));
+}
+
+/*
+ * innvl: {
+ *     property 1, property 2, ...
+ * }
+ *
+ * outnvl: {
+ *     bookmark name 1 -> { property 1, property 2, ... },
+ *     bookmark name 2 -> { property 1, property 2, ... }
+ * }
+ *
+ */
+static int
+zfs_ioc_get_bookmarks(const char *fsname, nvlist_t *innvl, nvlist_t *outnvl)
+{
+	return (dsl_get_bookmarks(fsname, innvl, outnvl));
+}
+
+/*
+ * innvl: {
+ *     bookmark name 1, bookmark name 2
+ * }
+ *
+ * outnvl: bookmark -> error code (int32)
+ *
+ */
+static int
+zfs_ioc_destroy_bookmarks(const char *poolname, nvlist_t *innvl,
+    nvlist_t *outnvl)
+{
+	int error, poollen;
+	nvpair_t *pair;
+
+	poollen = strlen(poolname);
+	for (pair = nvlist_next_nvpair(innvl, NULL);
+	    pair != NULL; pair = nvlist_next_nvpair(innvl, pair)) {
+		const char *name = nvpair_name(pair);
+		const char *cp = strchr(name, '#');
+
+		/*
+		 * The bookmark name must contain an #, and the part after it
+		 * must contain only valid characters.
+		 */
+		if (cp == NULL ||
+		    zfs_component_namecheck(cp + 1, NULL, NULL) != 0)
+			return (SET_ERROR(EINVAL));
+
+		/*
+		 * The bookmark must be in the specified pool.
+		 */
+		if (strncmp(name, poolname, poollen) != 0 ||
+		    (name[poollen] != '/' && name[poollen] != '#'))
+			return (SET_ERROR(EXDEV));
+	}
+
+	error = dsl_bookmark_destroy(innvl, outnvl);
+	return (error);
+}
+
 /*
  * inputs:
  * zc_name		name of dataset to destroy
@@ -4097,7 +4239,8 @@ out:
  * zc_guid	if set, estimate size of stream only.  zc_cookie is ignored.
  *		output size in zc_objset_type.
  *
- * outputs: none
+ * outputs:
+ * zc_objset_type	estimated size, if zc_guid is set
  */
 static int
 zfs_ioc_send(zfs_cmd_t *zc)
@@ -5272,6 +5415,19 @@ zfs_ioctl_init(void)
 	    zfs_ioc_rollback, zfs_secpolicy_rollback, DATASET_NAME,
 	    POOL_CHECK_SUSPENDED | POOL_CHECK_READONLY, B_FALSE, B_TRUE);
 
+	zfs_ioctl_register("bookmark", ZFS_IOC_BOOKMARK,
+	    zfs_ioc_bookmark, zfs_secpolicy_bookmark, POOL_NAME,
+	    POOL_CHECK_SUSPENDED | POOL_CHECK_READONLY, B_TRUE, B_TRUE);
+
+	zfs_ioctl_register("get_bookmarks", ZFS_IOC_GET_BOOKMARKS,
+	    zfs_ioc_get_bookmarks, zfs_secpolicy_read, DATASET_NAME,
+	    POOL_CHECK_SUSPENDED, B_FALSE, B_FALSE);
+
+	zfs_ioctl_register("destroy_bookmarks", ZFS_IOC_DESTROY_BOOKMARKS,
+	    zfs_ioc_destroy_bookmarks, zfs_secpolicy_destroy_bookmarks,
+	    POOL_NAME,
+	    POOL_CHECK_SUSPENDED | POOL_CHECK_READONLY, B_TRUE, B_TRUE);
+
 	/* IOCTLS that use the legacy function signature */
 
 	zfs_ioctl_register_legacy(ZFS_IOC_POOL_FREEZE, zfs_ioc_pool_freeze,