xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 6a59053b922dcd35760db6b042abda7cb6ef1c80)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25  * Copyright (c) 2014 RackTop Systems.
26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  */
28 
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/arc.h>
38 #include <sys/zio.h>
39 #include <sys/zap.h>
40 #include <sys/zfeature.h>
41 #include <sys/unique.h>
42 #include <sys/zfs_context.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/spa.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_onexit.h>
47 #include <sys/zvol.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/dsl_deadlist.h>
50 #include <sys/dsl_destroy.h>
51 #include <sys/dsl_userhold.h>
52 #include <sys/dsl_bookmark.h>
53 
54 /*
55  * The SPA supports block sizes up to 16MB.  However, very large blocks
56  * can have an impact on i/o latency (e.g. tying up a spinning disk for
57  * ~300ms), and also potentially on the memory allocator.  Therefore,
58  * we do not allow the recordsize to be set larger than zfs_max_recordsize
59  * (default 1MB).  Larger blocks can be created by changing this tunable,
60  * and pools with larger blocks can always be imported and used, regardless
61  * of this setting.
62  */
63 int zfs_max_recordsize = 1 * 1024 * 1024;
64 
65 #define	SWITCH64(x, y) \
66 	{ \
67 		uint64_t __tmp = (x); \
68 		(x) = (y); \
69 		(y) = __tmp; \
70 	}
71 
72 #define	DS_REF_MAX	(1ULL << 62)
73 
74 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
75 
76 /*
77  * Figure out how much of this delta should be propogated to the dsl_dir
78  * layer.  If there's a refreservation, that space has already been
79  * partially accounted for in our ancestors.
80  */
81 static int64_t
82 parent_delta(dsl_dataset_t *ds, int64_t delta)
83 {
84 	dsl_dataset_phys_t *ds_phys;
85 	uint64_t old_bytes, new_bytes;
86 
87 	if (ds->ds_reserved == 0)
88 		return (delta);
89 
90 	ds_phys = dsl_dataset_phys(ds);
91 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
92 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
93 
94 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
95 	return (new_bytes - old_bytes);
96 }
97 
98 void
99 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
100 {
101 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
102 	int compressed = BP_GET_PSIZE(bp);
103 	int uncompressed = BP_GET_UCSIZE(bp);
104 	int64_t delta;
105 
106 	dprintf_bp(bp, "ds=%p", ds);
107 
108 	ASSERT(dmu_tx_is_syncing(tx));
109 	/* It could have been compressed away to nothing */
110 	if (BP_IS_HOLE(bp))
111 		return;
112 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
113 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
114 	if (ds == NULL) {
115 		dsl_pool_mos_diduse_space(tx->tx_pool,
116 		    used, compressed, uncompressed);
117 		return;
118 	}
119 
120 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
121 	mutex_enter(&ds->ds_lock);
122 	delta = parent_delta(ds, used);
123 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
124 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
125 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
126 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
127 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
128 		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
129 		    B_TRUE;
130 	}
131 	mutex_exit(&ds->ds_lock);
132 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
133 	    compressed, uncompressed, tx);
134 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
135 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
136 }
137 
138 int
139 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
140     boolean_t async)
141 {
142 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
143 	int compressed = BP_GET_PSIZE(bp);
144 	int uncompressed = BP_GET_UCSIZE(bp);
145 
146 	if (BP_IS_HOLE(bp))
147 		return (0);
148 
149 	ASSERT(dmu_tx_is_syncing(tx));
150 	ASSERT(bp->blk_birth <= tx->tx_txg);
151 
152 	if (ds == NULL) {
153 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
154 		dsl_pool_mos_diduse_space(tx->tx_pool,
155 		    -used, -compressed, -uncompressed);
156 		return (used);
157 	}
158 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
159 
160 	ASSERT(!ds->ds_is_snapshot);
161 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
162 
163 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
164 		int64_t delta;
165 
166 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
167 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
168 
169 		mutex_enter(&ds->ds_lock);
170 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
171 		    !DS_UNIQUE_IS_ACCURATE(ds));
172 		delta = parent_delta(ds, -used);
173 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
174 		mutex_exit(&ds->ds_lock);
175 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
176 		    delta, -compressed, -uncompressed, tx);
177 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
178 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
179 	} else {
180 		dprintf_bp(bp, "putting on dead list: %s", "");
181 		if (async) {
182 			/*
183 			 * We are here as part of zio's write done callback,
184 			 * which means we're a zio interrupt thread.  We can't
185 			 * call dsl_deadlist_insert() now because it may block
186 			 * waiting for I/O.  Instead, put bp on the deferred
187 			 * queue and let dsl_pool_sync() finish the job.
188 			 */
189 			bplist_append(&ds->ds_pending_deadlist, bp);
190 		} else {
191 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
192 		}
193 		ASSERT3U(ds->ds_prev->ds_object, ==,
194 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
195 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
196 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
197 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
198 		    ds->ds_object && bp->blk_birth >
199 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
200 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
201 			mutex_enter(&ds->ds_prev->ds_lock);
202 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
203 			mutex_exit(&ds->ds_prev->ds_lock);
204 		}
205 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
206 			dsl_dir_transfer_space(ds->ds_dir, used,
207 			    DD_USED_HEAD, DD_USED_SNAP, tx);
208 		}
209 	}
210 	mutex_enter(&ds->ds_lock);
211 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
212 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
213 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
214 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
215 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
216 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
217 	mutex_exit(&ds->ds_lock);
218 
219 	return (used);
220 }
221 
222 uint64_t
223 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
224 {
225 	uint64_t trysnap = 0;
226 
227 	if (ds == NULL)
228 		return (0);
229 	/*
230 	 * The snapshot creation could fail, but that would cause an
231 	 * incorrect FALSE return, which would only result in an
232 	 * overestimation of the amount of space that an operation would
233 	 * consume, which is OK.
234 	 *
235 	 * There's also a small window where we could miss a pending
236 	 * snapshot, because we could set the sync task in the quiescing
237 	 * phase.  So this should only be used as a guess.
238 	 */
239 	if (ds->ds_trysnap_txg >
240 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
241 		trysnap = ds->ds_trysnap_txg;
242 	return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
243 }
244 
245 boolean_t
246 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
247     uint64_t blk_birth)
248 {
249 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
250 	    (bp != NULL && BP_IS_HOLE(bp)))
251 		return (B_FALSE);
252 
253 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
254 
255 	return (B_TRUE);
256 }
257 
258 static void
259 dsl_dataset_evict(void *dbu)
260 {
261 	dsl_dataset_t *ds = dbu;
262 
263 	ASSERT(ds->ds_owner == NULL);
264 
265 	ds->ds_dbuf = NULL;
266 
267 	unique_remove(ds->ds_fsid_guid);
268 
269 	if (ds->ds_objset != NULL)
270 		dmu_objset_evict(ds->ds_objset);
271 
272 	if (ds->ds_prev) {
273 		dsl_dataset_rele(ds->ds_prev, ds);
274 		ds->ds_prev = NULL;
275 	}
276 
277 	bplist_destroy(&ds->ds_pending_deadlist);
278 	if (ds->ds_deadlist.dl_os != NULL)
279 		dsl_deadlist_close(&ds->ds_deadlist);
280 	if (ds->ds_dir)
281 		dsl_dir_async_rele(ds->ds_dir, ds);
282 
283 	ASSERT(!list_link_active(&ds->ds_synced_link));
284 
285 	mutex_destroy(&ds->ds_lock);
286 	mutex_destroy(&ds->ds_opening_lock);
287 	mutex_destroy(&ds->ds_sendstream_lock);
288 	refcount_destroy(&ds->ds_longholds);
289 
290 	kmem_free(ds, sizeof (dsl_dataset_t));
291 }
292 
293 int
294 dsl_dataset_get_snapname(dsl_dataset_t *ds)
295 {
296 	dsl_dataset_phys_t *headphys;
297 	int err;
298 	dmu_buf_t *headdbuf;
299 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
300 	objset_t *mos = dp->dp_meta_objset;
301 
302 	if (ds->ds_snapname[0])
303 		return (0);
304 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
305 		return (0);
306 
307 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
308 	    FTAG, &headdbuf);
309 	if (err != 0)
310 		return (err);
311 	headphys = headdbuf->db_data;
312 	err = zap_value_search(dp->dp_meta_objset,
313 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
314 	dmu_buf_rele(headdbuf, FTAG);
315 	return (err);
316 }
317 
318 int
319 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
320 {
321 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
322 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
323 	matchtype_t mt;
324 	int err;
325 
326 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
327 		mt = MT_FIRST;
328 	else
329 		mt = MT_EXACT;
330 
331 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
332 	    value, mt, NULL, 0, NULL);
333 	if (err == ENOTSUP && mt == MT_FIRST)
334 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
335 	return (err);
336 }
337 
338 int
339 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
340     boolean_t adj_cnt)
341 {
342 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
343 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
344 	matchtype_t mt;
345 	int err;
346 
347 	dsl_dir_snap_cmtime_update(ds->ds_dir);
348 
349 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
350 		mt = MT_FIRST;
351 	else
352 		mt = MT_EXACT;
353 
354 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
355 	if (err == ENOTSUP && mt == MT_FIRST)
356 		err = zap_remove(mos, snapobj, name, tx);
357 
358 	if (err == 0 && adj_cnt)
359 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
360 		    DD_FIELD_SNAPSHOT_COUNT, tx);
361 
362 	return (err);
363 }
364 
365 boolean_t
366 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
367 {
368 	dmu_buf_t *dbuf = ds->ds_dbuf;
369 	boolean_t result = B_FALSE;
370 
371 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
372 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
373 
374 		if (ds == dmu_buf_get_user(dbuf))
375 			result = B_TRUE;
376 		else
377 			dmu_buf_rele(dbuf, tag);
378 	}
379 
380 	return (result);
381 }
382 
383 int
384 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
385     dsl_dataset_t **dsp)
386 {
387 	objset_t *mos = dp->dp_meta_objset;
388 	dmu_buf_t *dbuf;
389 	dsl_dataset_t *ds;
390 	int err;
391 	dmu_object_info_t doi;
392 
393 	ASSERT(dsl_pool_config_held(dp));
394 
395 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
396 	if (err != 0)
397 		return (err);
398 
399 	/* Make sure dsobj has the correct object type. */
400 	dmu_object_info_from_db(dbuf, &doi);
401 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
402 		dmu_buf_rele(dbuf, tag);
403 		return (SET_ERROR(EINVAL));
404 	}
405 
406 	ds = dmu_buf_get_user(dbuf);
407 	if (ds == NULL) {
408 		dsl_dataset_t *winner = NULL;
409 
410 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
411 		ds->ds_dbuf = dbuf;
412 		ds->ds_object = dsobj;
413 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
414 
415 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
416 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
417 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
418 		refcount_create(&ds->ds_longholds);
419 
420 		bplist_create(&ds->ds_pending_deadlist);
421 		dsl_deadlist_open(&ds->ds_deadlist,
422 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
423 
424 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
425 		    offsetof(dmu_sendarg_t, dsa_link));
426 
427 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
428 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
429 				if (!(spa_feature_table[f].fi_flags &
430 				    ZFEATURE_FLAG_PER_DATASET))
431 					continue;
432 				err = zap_contains(mos, dsobj,
433 				    spa_feature_table[f].fi_guid);
434 				if (err == 0) {
435 					ds->ds_feature_inuse[f] = B_TRUE;
436 				} else {
437 					ASSERT3U(err, ==, ENOENT);
438 					err = 0;
439 				}
440 			}
441 		}
442 
443 		err = dsl_dir_hold_obj(dp,
444 		    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
445 		if (err != 0) {
446 			mutex_destroy(&ds->ds_lock);
447 			mutex_destroy(&ds->ds_opening_lock);
448 			mutex_destroy(&ds->ds_sendstream_lock);
449 			refcount_destroy(&ds->ds_longholds);
450 			bplist_destroy(&ds->ds_pending_deadlist);
451 			dsl_deadlist_close(&ds->ds_deadlist);
452 			kmem_free(ds, sizeof (dsl_dataset_t));
453 			dmu_buf_rele(dbuf, tag);
454 			return (err);
455 		}
456 
457 		if (!ds->ds_is_snapshot) {
458 			ds->ds_snapname[0] = '\0';
459 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
460 				err = dsl_dataset_hold_obj(dp,
461 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
462 				    ds, &ds->ds_prev);
463 			}
464 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
465 				int zaperr = zap_lookup(mos, ds->ds_object,
466 				    DS_FIELD_BOOKMARK_NAMES,
467 				    sizeof (ds->ds_bookmarks), 1,
468 				    &ds->ds_bookmarks);
469 				if (zaperr != ENOENT)
470 					VERIFY0(zaperr);
471 			}
472 		} else {
473 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
474 				err = dsl_dataset_get_snapname(ds);
475 			if (err == 0 &&
476 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
477 				err = zap_count(
478 				    ds->ds_dir->dd_pool->dp_meta_objset,
479 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
480 				    &ds->ds_userrefs);
481 			}
482 		}
483 
484 		if (err == 0 && !ds->ds_is_snapshot) {
485 			err = dsl_prop_get_int_ds(ds,
486 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
487 			    &ds->ds_reserved);
488 			if (err == 0) {
489 				err = dsl_prop_get_int_ds(ds,
490 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
491 				    &ds->ds_quota);
492 			}
493 		} else {
494 			ds->ds_reserved = ds->ds_quota = 0;
495 		}
496 
497 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
498 		if (err == 0)
499 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
500 
501 		if (err != 0 || winner != NULL) {
502 			bplist_destroy(&ds->ds_pending_deadlist);
503 			dsl_deadlist_close(&ds->ds_deadlist);
504 			if (ds->ds_prev)
505 				dsl_dataset_rele(ds->ds_prev, ds);
506 			dsl_dir_rele(ds->ds_dir, ds);
507 			mutex_destroy(&ds->ds_lock);
508 			mutex_destroy(&ds->ds_opening_lock);
509 			mutex_destroy(&ds->ds_sendstream_lock);
510 			refcount_destroy(&ds->ds_longholds);
511 			kmem_free(ds, sizeof (dsl_dataset_t));
512 			if (err != 0) {
513 				dmu_buf_rele(dbuf, tag);
514 				return (err);
515 			}
516 			ds = winner;
517 		} else {
518 			ds->ds_fsid_guid =
519 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
520 		}
521 	}
522 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
523 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
524 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
525 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
526 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
527 	*dsp = ds;
528 	return (0);
529 }
530 
531 int
532 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
533     void *tag, dsl_dataset_t **dsp)
534 {
535 	dsl_dir_t *dd;
536 	const char *snapname;
537 	uint64_t obj;
538 	int err = 0;
539 	dsl_dataset_t *ds;
540 
541 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
542 	if (err != 0)
543 		return (err);
544 
545 	ASSERT(dsl_pool_config_held(dp));
546 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
547 	if (obj != 0)
548 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
549 	else
550 		err = SET_ERROR(ENOENT);
551 
552 	/* we may be looking for a snapshot */
553 	if (err == 0 && snapname != NULL) {
554 		dsl_dataset_t *snap_ds;
555 
556 		if (*snapname++ != '@') {
557 			dsl_dataset_rele(ds, tag);
558 			dsl_dir_rele(dd, FTAG);
559 			return (SET_ERROR(ENOENT));
560 		}
561 
562 		dprintf("looking for snapshot '%s'\n", snapname);
563 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
564 		if (err == 0)
565 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
566 		dsl_dataset_rele(ds, tag);
567 
568 		if (err == 0) {
569 			mutex_enter(&snap_ds->ds_lock);
570 			if (snap_ds->ds_snapname[0] == 0)
571 				(void) strlcpy(snap_ds->ds_snapname, snapname,
572 				    sizeof (snap_ds->ds_snapname));
573 			mutex_exit(&snap_ds->ds_lock);
574 			ds = snap_ds;
575 		}
576 	}
577 	if (err == 0)
578 		*dsp = ds;
579 	dsl_dir_rele(dd, FTAG);
580 	return (err);
581 }
582 
583 int
584 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
585     void *tag, dsl_dataset_t **dsp)
586 {
587 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
588 	if (err != 0)
589 		return (err);
590 	if (!dsl_dataset_tryown(*dsp, tag)) {
591 		dsl_dataset_rele(*dsp, tag);
592 		*dsp = NULL;
593 		return (SET_ERROR(EBUSY));
594 	}
595 	return (0);
596 }
597 
598 int
599 dsl_dataset_own(dsl_pool_t *dp, const char *name,
600     void *tag, dsl_dataset_t **dsp)
601 {
602 	int err = dsl_dataset_hold(dp, name, tag, dsp);
603 	if (err != 0)
604 		return (err);
605 	if (!dsl_dataset_tryown(*dsp, tag)) {
606 		dsl_dataset_rele(*dsp, tag);
607 		return (SET_ERROR(EBUSY));
608 	}
609 	return (0);
610 }
611 
612 /*
613  * See the comment above dsl_pool_hold() for details.  In summary, a long
614  * hold is used to prevent destruction of a dataset while the pool hold
615  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
616  *
617  * The dataset and pool must be held when this function is called.  After it
618  * is called, the pool hold may be released while the dataset is still held
619  * and accessed.
620  */
621 void
622 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
623 {
624 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
625 	(void) refcount_add(&ds->ds_longholds, tag);
626 }
627 
628 void
629 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
630 {
631 	(void) refcount_remove(&ds->ds_longholds, tag);
632 }
633 
634 /* Return B_TRUE if there are any long holds on this dataset. */
635 boolean_t
636 dsl_dataset_long_held(dsl_dataset_t *ds)
637 {
638 	return (!refcount_is_zero(&ds->ds_longholds));
639 }
640 
641 void
642 dsl_dataset_name(dsl_dataset_t *ds, char *name)
643 {
644 	if (ds == NULL) {
645 		(void) strcpy(name, "mos");
646 	} else {
647 		dsl_dir_name(ds->ds_dir, name);
648 		VERIFY0(dsl_dataset_get_snapname(ds));
649 		if (ds->ds_snapname[0]) {
650 			(void) strcat(name, "@");
651 			/*
652 			 * We use a "recursive" mutex so that we
653 			 * can call dprintf_ds() with ds_lock held.
654 			 */
655 			if (!MUTEX_HELD(&ds->ds_lock)) {
656 				mutex_enter(&ds->ds_lock);
657 				(void) strcat(name, ds->ds_snapname);
658 				mutex_exit(&ds->ds_lock);
659 			} else {
660 				(void) strcat(name, ds->ds_snapname);
661 			}
662 		}
663 	}
664 }
665 
666 void
667 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
668 {
669 	dmu_buf_rele(ds->ds_dbuf, tag);
670 }
671 
672 void
673 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
674 {
675 	ASSERT3P(ds->ds_owner, ==, tag);
676 	ASSERT(ds->ds_dbuf != NULL);
677 
678 	mutex_enter(&ds->ds_lock);
679 	ds->ds_owner = NULL;
680 	mutex_exit(&ds->ds_lock);
681 	dsl_dataset_long_rele(ds, tag);
682 	dsl_dataset_rele(ds, tag);
683 }
684 
685 boolean_t
686 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
687 {
688 	boolean_t gotit = FALSE;
689 
690 	mutex_enter(&ds->ds_lock);
691 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
692 		ds->ds_owner = tag;
693 		dsl_dataset_long_hold(ds, tag);
694 		gotit = TRUE;
695 	}
696 	mutex_exit(&ds->ds_lock);
697 	return (gotit);
698 }
699 
700 static void
701 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
702 {
703 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
704 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
705 	uint64_t zero = 0;
706 
707 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
708 
709 	spa_feature_incr(spa, f, tx);
710 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
711 
712 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
713 	    sizeof (zero), 1, &zero, tx));
714 }
715 
716 void
717 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
718 {
719 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
720 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
721 
722 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
723 
724 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
725 	spa_feature_decr(spa, f, tx);
726 }
727 
728 uint64_t
729 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
730     uint64_t flags, dmu_tx_t *tx)
731 {
732 	dsl_pool_t *dp = dd->dd_pool;
733 	dmu_buf_t *dbuf;
734 	dsl_dataset_phys_t *dsphys;
735 	uint64_t dsobj;
736 	objset_t *mos = dp->dp_meta_objset;
737 
738 	if (origin == NULL)
739 		origin = dp->dp_origin_snap;
740 
741 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
742 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
743 	ASSERT(dmu_tx_is_syncing(tx));
744 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
745 
746 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
747 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
748 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
749 	dmu_buf_will_dirty(dbuf, tx);
750 	dsphys = dbuf->db_data;
751 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
752 	dsphys->ds_dir_obj = dd->dd_object;
753 	dsphys->ds_flags = flags;
754 	dsphys->ds_fsid_guid = unique_create();
755 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
756 	    sizeof (dsphys->ds_guid));
757 	dsphys->ds_snapnames_zapobj =
758 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
759 	    DMU_OT_NONE, 0, tx);
760 	dsphys->ds_creation_time = gethrestime_sec();
761 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
762 
763 	if (origin == NULL) {
764 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
765 	} else {
766 		dsl_dataset_t *ohds; /* head of the origin snapshot */
767 
768 		dsphys->ds_prev_snap_obj = origin->ds_object;
769 		dsphys->ds_prev_snap_txg =
770 		    dsl_dataset_phys(origin)->ds_creation_txg;
771 		dsphys->ds_referenced_bytes =
772 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
773 		dsphys->ds_compressed_bytes =
774 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
775 		dsphys->ds_uncompressed_bytes =
776 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
777 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
778 
779 		/*
780 		 * Inherit flags that describe the dataset's contents
781 		 * (INCONSISTENT) or properties (Case Insensitive).
782 		 */
783 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
784 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
785 
786 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
787 			if (origin->ds_feature_inuse[f])
788 				dsl_dataset_activate_feature(dsobj, f, tx);
789 		}
790 
791 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
792 		dsl_dataset_phys(origin)->ds_num_children++;
793 
794 		VERIFY0(dsl_dataset_hold_obj(dp,
795 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
796 		    FTAG, &ohds));
797 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
798 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
799 		dsl_dataset_rele(ohds, FTAG);
800 
801 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
802 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
803 				dsl_dataset_phys(origin)->ds_next_clones_obj =
804 				    zap_create(mos,
805 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
806 			}
807 			VERIFY0(zap_add_int(mos,
808 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
809 			    dsobj, tx));
810 		}
811 
812 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
813 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
814 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
815 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
816 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
817 				dsl_dir_phys(origin->ds_dir)->dd_clones =
818 				    zap_create(mos,
819 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
820 			}
821 			VERIFY0(zap_add_int(mos,
822 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
823 			    dsobj, tx));
824 		}
825 	}
826 
827 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
828 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
829 
830 	dmu_buf_rele(dbuf, FTAG);
831 
832 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
833 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
834 
835 	return (dsobj);
836 }
837 
838 static void
839 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
840 {
841 	objset_t *os;
842 
843 	VERIFY0(dmu_objset_from_ds(ds, &os));
844 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
845 	dsl_dataset_dirty(ds, tx);
846 }
847 
848 uint64_t
849 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
850     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
851 {
852 	dsl_pool_t *dp = pdd->dd_pool;
853 	uint64_t dsobj, ddobj;
854 	dsl_dir_t *dd;
855 
856 	ASSERT(dmu_tx_is_syncing(tx));
857 	ASSERT(lastname[0] != '@');
858 
859 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
860 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
861 
862 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
863 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
864 
865 	dsl_deleg_set_create_perms(dd, tx, cr);
866 
867 	/*
868 	 * Since we're creating a new node we know it's a leaf, so we can
869 	 * initialize the counts if the limit feature is active.
870 	 */
871 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
872 		uint64_t cnt = 0;
873 		objset_t *os = dd->dd_pool->dp_meta_objset;
874 
875 		dsl_dir_zapify(dd, tx);
876 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
877 		    sizeof (cnt), 1, &cnt, tx));
878 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
879 		    sizeof (cnt), 1, &cnt, tx));
880 	}
881 
882 	dsl_dir_rele(dd, FTAG);
883 
884 	/*
885 	 * If we are creating a clone, make sure we zero out any stale
886 	 * data from the origin snapshots zil header.
887 	 */
888 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
889 		dsl_dataset_t *ds;
890 
891 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
892 		dsl_dataset_zero_zil(ds, tx);
893 		dsl_dataset_rele(ds, FTAG);
894 	}
895 
896 	return (dsobj);
897 }
898 
899 /*
900  * The unique space in the head dataset can be calculated by subtracting
901  * the space used in the most recent snapshot, that is still being used
902  * in this file system, from the space currently in use.  To figure out
903  * the space in the most recent snapshot still in use, we need to take
904  * the total space used in the snapshot and subtract out the space that
905  * has been freed up since the snapshot was taken.
906  */
907 void
908 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
909 {
910 	uint64_t mrs_used;
911 	uint64_t dlused, dlcomp, dluncomp;
912 
913 	ASSERT(!ds->ds_is_snapshot);
914 
915 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
916 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
917 	else
918 		mrs_used = 0;
919 
920 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
921 
922 	ASSERT3U(dlused, <=, mrs_used);
923 	dsl_dataset_phys(ds)->ds_unique_bytes =
924 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
925 
926 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
927 	    SPA_VERSION_UNIQUE_ACCURATE)
928 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
929 }
930 
931 void
932 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
933     dmu_tx_t *tx)
934 {
935 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
936 	uint64_t count;
937 	int err;
938 
939 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
940 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
941 	    obj, tx);
942 	/*
943 	 * The err should not be ENOENT, but a bug in a previous version
944 	 * of the code could cause upgrade_clones_cb() to not set
945 	 * ds_next_snap_obj when it should, leading to a missing entry.
946 	 * If we knew that the pool was created after
947 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
948 	 * ENOENT.  However, at least we can check that we don't have
949 	 * too many entries in the next_clones_obj even after failing to
950 	 * remove this one.
951 	 */
952 	if (err != ENOENT)
953 		VERIFY0(err);
954 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
955 	    &count));
956 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
957 }
958 
959 
960 blkptr_t *
961 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
962 {
963 	return (&dsl_dataset_phys(ds)->ds_bp);
964 }
965 
966 void
967 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
968 {
969 	ASSERT(dmu_tx_is_syncing(tx));
970 	/* If it's the meta-objset, set dp_meta_rootbp */
971 	if (ds == NULL) {
972 		tx->tx_pool->dp_meta_rootbp = *bp;
973 	} else {
974 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
975 		dsl_dataset_phys(ds)->ds_bp = *bp;
976 	}
977 }
978 
979 spa_t *
980 dsl_dataset_get_spa(dsl_dataset_t *ds)
981 {
982 	return (ds->ds_dir->dd_pool->dp_spa);
983 }
984 
985 void
986 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
987 {
988 	dsl_pool_t *dp;
989 
990 	if (ds == NULL) /* this is the meta-objset */
991 		return;
992 
993 	ASSERT(ds->ds_objset != NULL);
994 
995 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
996 		panic("dirtying snapshot!");
997 
998 	dp = ds->ds_dir->dd_pool;
999 
1000 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1001 		/* up the hold count until we can be written out */
1002 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1003 	}
1004 }
1005 
1006 boolean_t
1007 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1008 {
1009 	for (int t = 0; t < TXG_SIZE; t++) {
1010 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1011 		    ds, t))
1012 			return (B_TRUE);
1013 	}
1014 	return (B_FALSE);
1015 }
1016 
1017 static int
1018 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1019 {
1020 	uint64_t asize;
1021 
1022 	if (!dmu_tx_is_syncing(tx))
1023 		return (0);
1024 
1025 	/*
1026 	 * If there's an fs-only reservation, any blocks that might become
1027 	 * owned by the snapshot dataset must be accommodated by space
1028 	 * outside of the reservation.
1029 	 */
1030 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1031 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1032 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1033 		return (SET_ERROR(ENOSPC));
1034 
1035 	/*
1036 	 * Propagate any reserved space for this snapshot to other
1037 	 * snapshot checks in this sync group.
1038 	 */
1039 	if (asize > 0)
1040 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1041 
1042 	return (0);
1043 }
1044 
1045 typedef struct dsl_dataset_snapshot_arg {
1046 	nvlist_t *ddsa_snaps;
1047 	nvlist_t *ddsa_props;
1048 	nvlist_t *ddsa_errors;
1049 	cred_t *ddsa_cr;
1050 } dsl_dataset_snapshot_arg_t;
1051 
1052 int
1053 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1054     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1055 {
1056 	int error;
1057 	uint64_t value;
1058 
1059 	ds->ds_trysnap_txg = tx->tx_txg;
1060 
1061 	if (!dmu_tx_is_syncing(tx))
1062 		return (0);
1063 
1064 	/*
1065 	 * We don't allow multiple snapshots of the same txg.  If there
1066 	 * is already one, try again.
1067 	 */
1068 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1069 		return (SET_ERROR(EAGAIN));
1070 
1071 	/*
1072 	 * Check for conflicting snapshot name.
1073 	 */
1074 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1075 	if (error == 0)
1076 		return (SET_ERROR(EEXIST));
1077 	if (error != ENOENT)
1078 		return (error);
1079 
1080 	/*
1081 	 * We don't allow taking snapshots of inconsistent datasets, such as
1082 	 * those into which we are currently receiving.  However, if we are
1083 	 * creating this snapshot as part of a receive, this check will be
1084 	 * executed atomically with respect to the completion of the receive
1085 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1086 	 * case we ignore this, knowing it will be fixed up for us shortly in
1087 	 * dmu_recv_end_sync().
1088 	 */
1089 	if (!recv && DS_IS_INCONSISTENT(ds))
1090 		return (SET_ERROR(EBUSY));
1091 
1092 	/*
1093 	 * Skip the check for temporary snapshots or if we have already checked
1094 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1095 	 * check the count here when we're receiving a stream.
1096 	 */
1097 	if (cnt != 0 && cr != NULL) {
1098 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1099 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1100 		if (error != 0)
1101 			return (error);
1102 	}
1103 
1104 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1105 	if (error != 0)
1106 		return (error);
1107 
1108 	return (0);
1109 }
1110 
1111 static int
1112 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1113 {
1114 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1115 	dsl_pool_t *dp = dmu_tx_pool(tx);
1116 	nvpair_t *pair;
1117 	int rv = 0;
1118 
1119 	/*
1120 	 * Pre-compute how many total new snapshots will be created for each
1121 	 * level in the tree and below. This is needed for validating the
1122 	 * snapshot limit when either taking a recursive snapshot or when
1123 	 * taking multiple snapshots.
1124 	 *
1125 	 * The problem is that the counts are not actually adjusted when
1126 	 * we are checking, only when we finally sync. For a single snapshot,
1127 	 * this is easy, the count will increase by 1 at each node up the tree,
1128 	 * but its more complicated for the recursive/multiple snapshot case.
1129 	 *
1130 	 * The dsl_fs_ss_limit_check function does recursively check the count
1131 	 * at each level up the tree but since it is validating each snapshot
1132 	 * independently we need to be sure that we are validating the complete
1133 	 * count for the entire set of snapshots. We do this by rolling up the
1134 	 * counts for each component of the name into an nvlist and then
1135 	 * checking each of those cases with the aggregated count.
1136 	 *
1137 	 * This approach properly handles not only the recursive snapshot
1138 	 * case (where we get all of those on the ddsa_snaps list) but also
1139 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1140 	 * validate the limit on 'a' using a count of 2).
1141 	 *
1142 	 * We validate the snapshot names in the third loop and only report
1143 	 * name errors once.
1144 	 */
1145 	if (dmu_tx_is_syncing(tx)) {
1146 		nvlist_t *cnt_track = NULL;
1147 		cnt_track = fnvlist_alloc();
1148 
1149 		/* Rollup aggregated counts into the cnt_track list */
1150 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1151 		    pair != NULL;
1152 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1153 			char *pdelim;
1154 			uint64_t val;
1155 			char nm[MAXPATHLEN];
1156 
1157 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1158 			pdelim = strchr(nm, '@');
1159 			if (pdelim == NULL)
1160 				continue;
1161 			*pdelim = '\0';
1162 
1163 			do {
1164 				if (nvlist_lookup_uint64(cnt_track, nm,
1165 				    &val) == 0) {
1166 					/* update existing entry */
1167 					fnvlist_add_uint64(cnt_track, nm,
1168 					    val + 1);
1169 				} else {
1170 					/* add to list */
1171 					fnvlist_add_uint64(cnt_track, nm, 1);
1172 				}
1173 
1174 				pdelim = strrchr(nm, '/');
1175 				if (pdelim != NULL)
1176 					*pdelim = '\0';
1177 			} while (pdelim != NULL);
1178 		}
1179 
1180 		/* Check aggregated counts at each level */
1181 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1182 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1183 			int error = 0;
1184 			char *name;
1185 			uint64_t cnt = 0;
1186 			dsl_dataset_t *ds;
1187 
1188 			name = nvpair_name(pair);
1189 			cnt = fnvpair_value_uint64(pair);
1190 			ASSERT(cnt > 0);
1191 
1192 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1193 			if (error == 0) {
1194 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1195 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1196 				    ddsa->ddsa_cr);
1197 				dsl_dataset_rele(ds, FTAG);
1198 			}
1199 
1200 			if (error != 0) {
1201 				if (ddsa->ddsa_errors != NULL)
1202 					fnvlist_add_int32(ddsa->ddsa_errors,
1203 					    name, error);
1204 				rv = error;
1205 				/* only report one error for this check */
1206 				break;
1207 			}
1208 		}
1209 		nvlist_free(cnt_track);
1210 	}
1211 
1212 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1213 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1214 		int error = 0;
1215 		dsl_dataset_t *ds;
1216 		char *name, *atp;
1217 		char dsname[MAXNAMELEN];
1218 
1219 		name = nvpair_name(pair);
1220 		if (strlen(name) >= MAXNAMELEN)
1221 			error = SET_ERROR(ENAMETOOLONG);
1222 		if (error == 0) {
1223 			atp = strchr(name, '@');
1224 			if (atp == NULL)
1225 				error = SET_ERROR(EINVAL);
1226 			if (error == 0)
1227 				(void) strlcpy(dsname, name, atp - name + 1);
1228 		}
1229 		if (error == 0)
1230 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1231 		if (error == 0) {
1232 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1233 			error = dsl_dataset_snapshot_check_impl(ds,
1234 			    atp + 1, tx, B_FALSE, 0, NULL);
1235 			dsl_dataset_rele(ds, FTAG);
1236 		}
1237 
1238 		if (error != 0) {
1239 			if (ddsa->ddsa_errors != NULL) {
1240 				fnvlist_add_int32(ddsa->ddsa_errors,
1241 				    name, error);
1242 			}
1243 			rv = error;
1244 		}
1245 	}
1246 
1247 	return (rv);
1248 }
1249 
1250 void
1251 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1252     dmu_tx_t *tx)
1253 {
1254 	static zil_header_t zero_zil;
1255 
1256 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1257 	dmu_buf_t *dbuf;
1258 	dsl_dataset_phys_t *dsphys;
1259 	uint64_t dsobj, crtxg;
1260 	objset_t *mos = dp->dp_meta_objset;
1261 	objset_t *os;
1262 
1263 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1264 
1265 	/*
1266 	 * If we are on an old pool, the zil must not be active, in which
1267 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1268 	 */
1269 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1270 	    dmu_objset_from_ds(ds, &os) != 0 ||
1271 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1272 	    sizeof (zero_zil)) == 0);
1273 
1274 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1275 
1276 	/*
1277 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1278 	 */
1279 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1280 		crtxg = 1;
1281 	else
1282 		crtxg = tx->tx_txg;
1283 
1284 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1285 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1286 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1287 	dmu_buf_will_dirty(dbuf, tx);
1288 	dsphys = dbuf->db_data;
1289 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1290 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1291 	dsphys->ds_fsid_guid = unique_create();
1292 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1293 	    sizeof (dsphys->ds_guid));
1294 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1295 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1296 	dsphys->ds_next_snap_obj = ds->ds_object;
1297 	dsphys->ds_num_children = 1;
1298 	dsphys->ds_creation_time = gethrestime_sec();
1299 	dsphys->ds_creation_txg = crtxg;
1300 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1301 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1302 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1303 	dsphys->ds_uncompressed_bytes =
1304 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1305 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1306 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1307 	dmu_buf_rele(dbuf, FTAG);
1308 
1309 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1310 		if (ds->ds_feature_inuse[f])
1311 			dsl_dataset_activate_feature(dsobj, f, tx);
1312 	}
1313 
1314 	ASSERT3U(ds->ds_prev != 0, ==,
1315 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1316 	if (ds->ds_prev) {
1317 		uint64_t next_clones_obj =
1318 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1319 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1320 		    ds->ds_object ||
1321 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1322 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1323 		    ds->ds_object) {
1324 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1325 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1326 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1327 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1328 		} else if (next_clones_obj != 0) {
1329 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1330 			    dsphys->ds_next_snap_obj, tx);
1331 			VERIFY0(zap_add_int(mos,
1332 			    next_clones_obj, dsobj, tx));
1333 		}
1334 	}
1335 
1336 	/*
1337 	 * If we have a reference-reservation on this dataset, we will
1338 	 * need to increase the amount of refreservation being charged
1339 	 * since our unique space is going to zero.
1340 	 */
1341 	if (ds->ds_reserved) {
1342 		int64_t delta;
1343 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1344 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1345 		    ds->ds_reserved);
1346 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1347 		    delta, 0, 0, tx);
1348 	}
1349 
1350 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1351 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1352 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1353 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1354 	dsl_deadlist_close(&ds->ds_deadlist);
1355 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1356 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1357 	dsl_deadlist_add_key(&ds->ds_deadlist,
1358 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1359 
1360 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1361 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1362 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1363 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1364 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1365 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1366 
1367 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1368 	    snapname, 8, 1, &dsobj, tx));
1369 
1370 	if (ds->ds_prev)
1371 		dsl_dataset_rele(ds->ds_prev, ds);
1372 	VERIFY0(dsl_dataset_hold_obj(dp,
1373 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1374 
1375 	dsl_scan_ds_snapshotted(ds, tx);
1376 
1377 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1378 
1379 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1380 }
1381 
1382 static void
1383 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1384 {
1385 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1386 	dsl_pool_t *dp = dmu_tx_pool(tx);
1387 	nvpair_t *pair;
1388 
1389 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1390 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1391 		dsl_dataset_t *ds;
1392 		char *name, *atp;
1393 		char dsname[MAXNAMELEN];
1394 
1395 		name = nvpair_name(pair);
1396 		atp = strchr(name, '@');
1397 		(void) strlcpy(dsname, name, atp - name + 1);
1398 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1399 
1400 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1401 		if (ddsa->ddsa_props != NULL) {
1402 			dsl_props_set_sync_impl(ds->ds_prev,
1403 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1404 		}
1405 		dsl_dataset_rele(ds, FTAG);
1406 	}
1407 }
1408 
1409 /*
1410  * The snapshots must all be in the same pool.
1411  * All-or-nothing: if there are any failures, nothing will be modified.
1412  */
1413 int
1414 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1415 {
1416 	dsl_dataset_snapshot_arg_t ddsa;
1417 	nvpair_t *pair;
1418 	boolean_t needsuspend;
1419 	int error;
1420 	spa_t *spa;
1421 	char *firstname;
1422 	nvlist_t *suspended = NULL;
1423 
1424 	pair = nvlist_next_nvpair(snaps, NULL);
1425 	if (pair == NULL)
1426 		return (0);
1427 	firstname = nvpair_name(pair);
1428 
1429 	error = spa_open(firstname, &spa, FTAG);
1430 	if (error != 0)
1431 		return (error);
1432 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1433 	spa_close(spa, FTAG);
1434 
1435 	if (needsuspend) {
1436 		suspended = fnvlist_alloc();
1437 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1438 		    pair = nvlist_next_nvpair(snaps, pair)) {
1439 			char fsname[MAXNAMELEN];
1440 			char *snapname = nvpair_name(pair);
1441 			char *atp;
1442 			void *cookie;
1443 
1444 			atp = strchr(snapname, '@');
1445 			if (atp == NULL) {
1446 				error = SET_ERROR(EINVAL);
1447 				break;
1448 			}
1449 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1450 
1451 			error = zil_suspend(fsname, &cookie);
1452 			if (error != 0)
1453 				break;
1454 			fnvlist_add_uint64(suspended, fsname,
1455 			    (uintptr_t)cookie);
1456 		}
1457 	}
1458 
1459 	ddsa.ddsa_snaps = snaps;
1460 	ddsa.ddsa_props = props;
1461 	ddsa.ddsa_errors = errors;
1462 	ddsa.ddsa_cr = CRED();
1463 
1464 	if (error == 0) {
1465 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1466 		    dsl_dataset_snapshot_sync, &ddsa,
1467 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1468 	}
1469 
1470 	if (suspended != NULL) {
1471 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1472 		    pair = nvlist_next_nvpair(suspended, pair)) {
1473 			zil_resume((void *)(uintptr_t)
1474 			    fnvpair_value_uint64(pair));
1475 		}
1476 		fnvlist_free(suspended);
1477 	}
1478 
1479 	return (error);
1480 }
1481 
1482 typedef struct dsl_dataset_snapshot_tmp_arg {
1483 	const char *ddsta_fsname;
1484 	const char *ddsta_snapname;
1485 	minor_t ddsta_cleanup_minor;
1486 	const char *ddsta_htag;
1487 } dsl_dataset_snapshot_tmp_arg_t;
1488 
1489 static int
1490 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1491 {
1492 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1493 	dsl_pool_t *dp = dmu_tx_pool(tx);
1494 	dsl_dataset_t *ds;
1495 	int error;
1496 
1497 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1498 	if (error != 0)
1499 		return (error);
1500 
1501 	/* NULL cred means no limit check for tmp snapshot */
1502 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1503 	    tx, B_FALSE, 0, NULL);
1504 	if (error != 0) {
1505 		dsl_dataset_rele(ds, FTAG);
1506 		return (error);
1507 	}
1508 
1509 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1510 		dsl_dataset_rele(ds, FTAG);
1511 		return (SET_ERROR(ENOTSUP));
1512 	}
1513 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1514 	    B_TRUE, tx);
1515 	if (error != 0) {
1516 		dsl_dataset_rele(ds, FTAG);
1517 		return (error);
1518 	}
1519 
1520 	dsl_dataset_rele(ds, FTAG);
1521 	return (0);
1522 }
1523 
1524 static void
1525 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1526 {
1527 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1528 	dsl_pool_t *dp = dmu_tx_pool(tx);
1529 	dsl_dataset_t *ds;
1530 
1531 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1532 
1533 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1534 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1535 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1536 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1537 
1538 	dsl_dataset_rele(ds, FTAG);
1539 }
1540 
1541 int
1542 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1543     minor_t cleanup_minor, const char *htag)
1544 {
1545 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1546 	int error;
1547 	spa_t *spa;
1548 	boolean_t needsuspend;
1549 	void *cookie;
1550 
1551 	ddsta.ddsta_fsname = fsname;
1552 	ddsta.ddsta_snapname = snapname;
1553 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1554 	ddsta.ddsta_htag = htag;
1555 
1556 	error = spa_open(fsname, &spa, FTAG);
1557 	if (error != 0)
1558 		return (error);
1559 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1560 	spa_close(spa, FTAG);
1561 
1562 	if (needsuspend) {
1563 		error = zil_suspend(fsname, &cookie);
1564 		if (error != 0)
1565 			return (error);
1566 	}
1567 
1568 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1569 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1570 
1571 	if (needsuspend)
1572 		zil_resume(cookie);
1573 	return (error);
1574 }
1575 
1576 
1577 void
1578 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1579 {
1580 	ASSERT(dmu_tx_is_syncing(tx));
1581 	ASSERT(ds->ds_objset != NULL);
1582 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1583 
1584 	/*
1585 	 * in case we had to change ds_fsid_guid when we opened it,
1586 	 * sync it out now.
1587 	 */
1588 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1589 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1590 
1591 	dmu_objset_sync(ds->ds_objset, zio, tx);
1592 
1593 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1594 		if (ds->ds_feature_activation_needed[f]) {
1595 			if (ds->ds_feature_inuse[f])
1596 				continue;
1597 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1598 			ds->ds_feature_inuse[f] = B_TRUE;
1599 		}
1600 	}
1601 }
1602 
1603 static void
1604 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1605 {
1606 	uint64_t count = 0;
1607 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1608 	zap_cursor_t zc;
1609 	zap_attribute_t za;
1610 	nvlist_t *propval = fnvlist_alloc();
1611 	nvlist_t *val = fnvlist_alloc();
1612 
1613 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1614 
1615 	/*
1616 	 * There may be missing entries in ds_next_clones_obj
1617 	 * due to a bug in a previous version of the code.
1618 	 * Only trust it if it has the right number of entries.
1619 	 */
1620 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1621 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1622 		    &count));
1623 	}
1624 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1625 		goto fail;
1626 	for (zap_cursor_init(&zc, mos,
1627 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1628 	    zap_cursor_retrieve(&zc, &za) == 0;
1629 	    zap_cursor_advance(&zc)) {
1630 		dsl_dataset_t *clone;
1631 		char buf[ZFS_MAXNAMELEN];
1632 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1633 		    za.za_first_integer, FTAG, &clone));
1634 		dsl_dir_name(clone->ds_dir, buf);
1635 		fnvlist_add_boolean(val, buf);
1636 		dsl_dataset_rele(clone, FTAG);
1637 	}
1638 	zap_cursor_fini(&zc);
1639 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1640 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1641 fail:
1642 	nvlist_free(val);
1643 	nvlist_free(propval);
1644 }
1645 
1646 void
1647 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1648 {
1649 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1650 	uint64_t refd, avail, uobjs, aobjs, ratio;
1651 
1652 	ASSERT(dsl_pool_config_held(dp));
1653 
1654 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1655 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1656 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1657 
1658 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1659 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1660 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1661 
1662 	if (ds->ds_is_snapshot) {
1663 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1664 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1665 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1666 		get_clones_stat(ds, nv);
1667 	} else {
1668 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1669 			char buf[MAXNAMELEN];
1670 			dsl_dataset_name(ds->ds_prev, buf);
1671 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1672 		}
1673 
1674 		dsl_dir_stats(ds->ds_dir, nv);
1675 	}
1676 
1677 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1678 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1679 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1680 
1681 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1682 	    dsl_dataset_phys(ds)->ds_creation_time);
1683 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1684 	    dsl_dataset_phys(ds)->ds_creation_txg);
1685 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1686 	    ds->ds_quota);
1687 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1688 	    ds->ds_reserved);
1689 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1690 	    dsl_dataset_phys(ds)->ds_guid);
1691 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1692 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1693 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1694 	    ds->ds_object);
1695 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1696 	    ds->ds_userrefs);
1697 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1698 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1699 
1700 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1701 		uint64_t written, comp, uncomp;
1702 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1703 		dsl_dataset_t *prev;
1704 
1705 		int err = dsl_dataset_hold_obj(dp,
1706 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1707 		if (err == 0) {
1708 			err = dsl_dataset_space_written(prev, ds, &written,
1709 			    &comp, &uncomp);
1710 			dsl_dataset_rele(prev, FTAG);
1711 			if (err == 0) {
1712 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1713 				    written);
1714 			}
1715 		}
1716 	}
1717 }
1718 
1719 void
1720 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1721 {
1722 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1723 	ASSERT(dsl_pool_config_held(dp));
1724 
1725 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1726 	stat->dds_inconsistent =
1727 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1728 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1729 	stat->dds_origin[0] = '\0';
1730 	if (ds->ds_is_snapshot) {
1731 		stat->dds_is_snapshot = B_TRUE;
1732 		stat->dds_num_clones =
1733 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1734 	} else {
1735 		stat->dds_is_snapshot = B_FALSE;
1736 		stat->dds_num_clones = 0;
1737 
1738 		if (dsl_dir_is_clone(ds->ds_dir)) {
1739 			dsl_dataset_t *ods;
1740 
1741 			VERIFY0(dsl_dataset_hold_obj(dp,
1742 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1743 			    FTAG, &ods));
1744 			dsl_dataset_name(ods, stat->dds_origin);
1745 			dsl_dataset_rele(ods, FTAG);
1746 		}
1747 	}
1748 }
1749 
1750 uint64_t
1751 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1752 {
1753 	return (ds->ds_fsid_guid);
1754 }
1755 
1756 void
1757 dsl_dataset_space(dsl_dataset_t *ds,
1758     uint64_t *refdbytesp, uint64_t *availbytesp,
1759     uint64_t *usedobjsp, uint64_t *availobjsp)
1760 {
1761 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1762 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1763 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1764 		*availbytesp +=
1765 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1766 	if (ds->ds_quota != 0) {
1767 		/*
1768 		 * Adjust available bytes according to refquota
1769 		 */
1770 		if (*refdbytesp < ds->ds_quota)
1771 			*availbytesp = MIN(*availbytesp,
1772 			    ds->ds_quota - *refdbytesp);
1773 		else
1774 			*availbytesp = 0;
1775 	}
1776 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1777 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1778 }
1779 
1780 boolean_t
1781 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1782 {
1783 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1784 
1785 	ASSERT(dsl_pool_config_held(dp));
1786 	if (snap == NULL)
1787 		return (B_FALSE);
1788 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1789 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1790 		objset_t *os, *os_snap;
1791 		/*
1792 		 * It may be that only the ZIL differs, because it was
1793 		 * reset in the head.  Don't count that as being
1794 		 * modified.
1795 		 */
1796 		if (dmu_objset_from_ds(ds, &os) != 0)
1797 			return (B_TRUE);
1798 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1799 			return (B_TRUE);
1800 		return (bcmp(&os->os_phys->os_meta_dnode,
1801 		    &os_snap->os_phys->os_meta_dnode,
1802 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1803 	}
1804 	return (B_FALSE);
1805 }
1806 
1807 typedef struct dsl_dataset_rename_snapshot_arg {
1808 	const char *ddrsa_fsname;
1809 	const char *ddrsa_oldsnapname;
1810 	const char *ddrsa_newsnapname;
1811 	boolean_t ddrsa_recursive;
1812 	dmu_tx_t *ddrsa_tx;
1813 } dsl_dataset_rename_snapshot_arg_t;
1814 
1815 /* ARGSUSED */
1816 static int
1817 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1818     dsl_dataset_t *hds, void *arg)
1819 {
1820 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1821 	int error;
1822 	uint64_t val;
1823 
1824 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1825 	if (error != 0) {
1826 		/* ignore nonexistent snapshots */
1827 		return (error == ENOENT ? 0 : error);
1828 	}
1829 
1830 	/* new name should not exist */
1831 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1832 	if (error == 0)
1833 		error = SET_ERROR(EEXIST);
1834 	else if (error == ENOENT)
1835 		error = 0;
1836 
1837 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1838 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1839 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1840 		error = SET_ERROR(ENAMETOOLONG);
1841 
1842 	return (error);
1843 }
1844 
1845 static int
1846 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1847 {
1848 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1849 	dsl_pool_t *dp = dmu_tx_pool(tx);
1850 	dsl_dataset_t *hds;
1851 	int error;
1852 
1853 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1854 	if (error != 0)
1855 		return (error);
1856 
1857 	if (ddrsa->ddrsa_recursive) {
1858 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1859 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1860 		    DS_FIND_CHILDREN);
1861 	} else {
1862 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1863 	}
1864 	dsl_dataset_rele(hds, FTAG);
1865 	return (error);
1866 }
1867 
1868 static int
1869 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1870     dsl_dataset_t *hds, void *arg)
1871 {
1872 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1873 	dsl_dataset_t *ds;
1874 	uint64_t val;
1875 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1876 	int error;
1877 
1878 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1879 	ASSERT(error == 0 || error == ENOENT);
1880 	if (error == ENOENT) {
1881 		/* ignore nonexistent snapshots */
1882 		return (0);
1883 	}
1884 
1885 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1886 
1887 	/* log before we change the name */
1888 	spa_history_log_internal_ds(ds, "rename", tx,
1889 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1890 
1891 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1892 	    B_FALSE));
1893 	mutex_enter(&ds->ds_lock);
1894 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1895 	mutex_exit(&ds->ds_lock);
1896 	VERIFY0(zap_add(dp->dp_meta_objset,
1897 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1898 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1899 
1900 	dsl_dataset_rele(ds, FTAG);
1901 	return (0);
1902 }
1903 
1904 static void
1905 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1906 {
1907 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1908 	dsl_pool_t *dp = dmu_tx_pool(tx);
1909 	dsl_dataset_t *hds;
1910 
1911 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1912 	ddrsa->ddrsa_tx = tx;
1913 	if (ddrsa->ddrsa_recursive) {
1914 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1915 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1916 		    DS_FIND_CHILDREN));
1917 	} else {
1918 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1919 	}
1920 	dsl_dataset_rele(hds, FTAG);
1921 }
1922 
1923 int
1924 dsl_dataset_rename_snapshot(const char *fsname,
1925     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1926 {
1927 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1928 
1929 	ddrsa.ddrsa_fsname = fsname;
1930 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1931 	ddrsa.ddrsa_newsnapname = newsnapname;
1932 	ddrsa.ddrsa_recursive = recursive;
1933 
1934 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1935 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
1936 	    1, ZFS_SPACE_CHECK_RESERVED));
1937 }
1938 
1939 /*
1940  * If we're doing an ownership handoff, we need to make sure that there is
1941  * only one long hold on the dataset.  We're not allowed to change anything here
1942  * so we don't permanently release the long hold or regular hold here.  We want
1943  * to do this only when syncing to avoid the dataset unexpectedly going away
1944  * when we release the long hold.
1945  */
1946 static int
1947 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1948 {
1949 	boolean_t held;
1950 
1951 	if (!dmu_tx_is_syncing(tx))
1952 		return (0);
1953 
1954 	if (owner != NULL) {
1955 		VERIFY3P(ds->ds_owner, ==, owner);
1956 		dsl_dataset_long_rele(ds, owner);
1957 	}
1958 
1959 	held = dsl_dataset_long_held(ds);
1960 
1961 	if (owner != NULL)
1962 		dsl_dataset_long_hold(ds, owner);
1963 
1964 	if (held)
1965 		return (SET_ERROR(EBUSY));
1966 
1967 	return (0);
1968 }
1969 
1970 typedef struct dsl_dataset_rollback_arg {
1971 	const char *ddra_fsname;
1972 	void *ddra_owner;
1973 	nvlist_t *ddra_result;
1974 } dsl_dataset_rollback_arg_t;
1975 
1976 static int
1977 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1978 {
1979 	dsl_dataset_rollback_arg_t *ddra = arg;
1980 	dsl_pool_t *dp = dmu_tx_pool(tx);
1981 	dsl_dataset_t *ds;
1982 	int64_t unused_refres_delta;
1983 	int error;
1984 
1985 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1986 	if (error != 0)
1987 		return (error);
1988 
1989 	/* must not be a snapshot */
1990 	if (ds->ds_is_snapshot) {
1991 		dsl_dataset_rele(ds, FTAG);
1992 		return (SET_ERROR(EINVAL));
1993 	}
1994 
1995 	/* must have a most recent snapshot */
1996 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
1997 		dsl_dataset_rele(ds, FTAG);
1998 		return (SET_ERROR(EINVAL));
1999 	}
2000 
2001 	/* must not have any bookmarks after the most recent snapshot */
2002 	nvlist_t *proprequest = fnvlist_alloc();
2003 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2004 	nvlist_t *bookmarks = fnvlist_alloc();
2005 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2006 	fnvlist_free(proprequest);
2007 	if (error != 0)
2008 		return (error);
2009 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2010 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2011 		nvlist_t *valuenv =
2012 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2013 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2014 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2015 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2016 			fnvlist_free(bookmarks);
2017 			dsl_dataset_rele(ds, FTAG);
2018 			return (SET_ERROR(EEXIST));
2019 		}
2020 	}
2021 	fnvlist_free(bookmarks);
2022 
2023 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2024 	if (error != 0) {
2025 		dsl_dataset_rele(ds, FTAG);
2026 		return (error);
2027 	}
2028 
2029 	/*
2030 	 * Check if the snap we are rolling back to uses more than
2031 	 * the refquota.
2032 	 */
2033 	if (ds->ds_quota != 0 &&
2034 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2035 		dsl_dataset_rele(ds, FTAG);
2036 		return (SET_ERROR(EDQUOT));
2037 	}
2038 
2039 	/*
2040 	 * When we do the clone swap, we will temporarily use more space
2041 	 * due to the refreservation (the head will no longer have any
2042 	 * unique space, so the entire amount of the refreservation will need
2043 	 * to be free).  We will immediately destroy the clone, freeing
2044 	 * this space, but the freeing happens over many txg's.
2045 	 */
2046 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2047 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2048 
2049 	if (unused_refres_delta > 0 &&
2050 	    unused_refres_delta >
2051 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2052 		dsl_dataset_rele(ds, FTAG);
2053 		return (SET_ERROR(ENOSPC));
2054 	}
2055 
2056 	dsl_dataset_rele(ds, FTAG);
2057 	return (0);
2058 }
2059 
2060 static void
2061 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2062 {
2063 	dsl_dataset_rollback_arg_t *ddra = arg;
2064 	dsl_pool_t *dp = dmu_tx_pool(tx);
2065 	dsl_dataset_t *ds, *clone;
2066 	uint64_t cloneobj;
2067 	char namebuf[ZFS_MAXNAMELEN];
2068 
2069 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2070 
2071 	dsl_dataset_name(ds->ds_prev, namebuf);
2072 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2073 
2074 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2075 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2076 
2077 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2078 
2079 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2080 	dsl_dataset_zero_zil(ds, tx);
2081 
2082 	dsl_destroy_head_sync_impl(clone, tx);
2083 
2084 	dsl_dataset_rele(clone, FTAG);
2085 	dsl_dataset_rele(ds, FTAG);
2086 }
2087 
2088 /*
2089  * Rolls back the given filesystem or volume to the most recent snapshot.
2090  * The name of the most recent snapshot will be returned under key "target"
2091  * in the result nvlist.
2092  *
2093  * If owner != NULL:
2094  * - The existing dataset MUST be owned by the specified owner at entry
2095  * - Upon return, dataset will still be held by the same owner, whether we
2096  *   succeed or not.
2097  *
2098  * This mode is required any time the existing filesystem is mounted.  See
2099  * notes above zfs_suspend_fs() for further details.
2100  */
2101 int
2102 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2103 {
2104 	dsl_dataset_rollback_arg_t ddra;
2105 
2106 	ddra.ddra_fsname = fsname;
2107 	ddra.ddra_owner = owner;
2108 	ddra.ddra_result = result;
2109 
2110 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2111 	    dsl_dataset_rollback_sync, &ddra,
2112 	    1, ZFS_SPACE_CHECK_RESERVED));
2113 }
2114 
2115 struct promotenode {
2116 	list_node_t link;
2117 	dsl_dataset_t *ds;
2118 };
2119 
2120 typedef struct dsl_dataset_promote_arg {
2121 	const char *ddpa_clonename;
2122 	dsl_dataset_t *ddpa_clone;
2123 	list_t shared_snaps, origin_snaps, clone_snaps;
2124 	dsl_dataset_t *origin_origin; /* origin of the origin */
2125 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2126 	char *err_ds;
2127 	cred_t *cr;
2128 } dsl_dataset_promote_arg_t;
2129 
2130 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2131 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2132     void *tag);
2133 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2134 
2135 static int
2136 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2137 {
2138 	dsl_dataset_promote_arg_t *ddpa = arg;
2139 	dsl_pool_t *dp = dmu_tx_pool(tx);
2140 	dsl_dataset_t *hds;
2141 	struct promotenode *snap;
2142 	dsl_dataset_t *origin_ds;
2143 	int err;
2144 	uint64_t unused;
2145 	uint64_t ss_mv_cnt;
2146 	size_t max_snap_len;
2147 
2148 	err = promote_hold(ddpa, dp, FTAG);
2149 	if (err != 0)
2150 		return (err);
2151 
2152 	hds = ddpa->ddpa_clone;
2153 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2154 
2155 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2156 		promote_rele(ddpa, FTAG);
2157 		return (SET_ERROR(EXDEV));
2158 	}
2159 
2160 	/*
2161 	 * Compute and check the amount of space to transfer.  Since this is
2162 	 * so expensive, don't do the preliminary check.
2163 	 */
2164 	if (!dmu_tx_is_syncing(tx)) {
2165 		promote_rele(ddpa, FTAG);
2166 		return (0);
2167 	}
2168 
2169 	snap = list_head(&ddpa->shared_snaps);
2170 	origin_ds = snap->ds;
2171 
2172 	/* compute origin's new unique space */
2173 	snap = list_tail(&ddpa->clone_snaps);
2174 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2175 	    origin_ds->ds_object);
2176 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2177 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2178 	    &ddpa->unique, &unused, &unused);
2179 
2180 	/*
2181 	 * Walk the snapshots that we are moving
2182 	 *
2183 	 * Compute space to transfer.  Consider the incremental changes
2184 	 * to used by each snapshot:
2185 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2186 	 * So each snapshot gave birth to:
2187 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2188 	 * So a sequence would look like:
2189 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2190 	 * Which simplifies to:
2191 	 * uN + kN + kN-1 + ... + k1 + k0
2192 	 * Note however, if we stop before we reach the ORIGIN we get:
2193 	 * uN + kN + kN-1 + ... + kM - uM-1
2194 	 */
2195 	ss_mv_cnt = 0;
2196 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2197 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2198 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2199 	for (snap = list_head(&ddpa->shared_snaps); snap;
2200 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2201 		uint64_t val, dlused, dlcomp, dluncomp;
2202 		dsl_dataset_t *ds = snap->ds;
2203 
2204 		ss_mv_cnt++;
2205 
2206 		/*
2207 		 * If there are long holds, we won't be able to evict
2208 		 * the objset.
2209 		 */
2210 		if (dsl_dataset_long_held(ds)) {
2211 			err = SET_ERROR(EBUSY);
2212 			goto out;
2213 		}
2214 
2215 		/* Check that the snapshot name does not conflict */
2216 		VERIFY0(dsl_dataset_get_snapname(ds));
2217 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2218 			err = SET_ERROR(ENAMETOOLONG);
2219 			goto out;
2220 		}
2221 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2222 		if (err == 0) {
2223 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2224 			err = SET_ERROR(EEXIST);
2225 			goto out;
2226 		}
2227 		if (err != ENOENT)
2228 			goto out;
2229 
2230 		/* The very first snapshot does not have a deadlist */
2231 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2232 			continue;
2233 
2234 		dsl_deadlist_space(&ds->ds_deadlist,
2235 		    &dlused, &dlcomp, &dluncomp);
2236 		ddpa->used += dlused;
2237 		ddpa->comp += dlcomp;
2238 		ddpa->uncomp += dluncomp;
2239 	}
2240 
2241 	/*
2242 	 * If we are a clone of a clone then we never reached ORIGIN,
2243 	 * so we need to subtract out the clone origin's used space.
2244 	 */
2245 	if (ddpa->origin_origin) {
2246 		ddpa->used -=
2247 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2248 		ddpa->comp -=
2249 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2250 		ddpa->uncomp -=
2251 		    dsl_dataset_phys(ddpa->origin_origin)->
2252 		    ds_uncompressed_bytes;
2253 	}
2254 
2255 	/* Check that there is enough space and limit headroom here */
2256 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2257 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2258 	if (err != 0)
2259 		goto out;
2260 
2261 	/*
2262 	 * Compute the amounts of space that will be used by snapshots
2263 	 * after the promotion (for both origin and clone).  For each,
2264 	 * it is the amount of space that will be on all of their
2265 	 * deadlists (that was not born before their new origin).
2266 	 */
2267 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2268 		uint64_t space;
2269 
2270 		/*
2271 		 * Note, typically this will not be a clone of a clone,
2272 		 * so dd_origin_txg will be < TXG_INITIAL, so
2273 		 * these snaplist_space() -> dsl_deadlist_space_range()
2274 		 * calls will be fast because they do not have to
2275 		 * iterate over all bps.
2276 		 */
2277 		snap = list_head(&ddpa->origin_snaps);
2278 		err = snaplist_space(&ddpa->shared_snaps,
2279 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2280 		if (err != 0)
2281 			goto out;
2282 
2283 		err = snaplist_space(&ddpa->clone_snaps,
2284 		    snap->ds->ds_dir->dd_origin_txg, &space);
2285 		if (err != 0)
2286 			goto out;
2287 		ddpa->cloneusedsnap += space;
2288 	}
2289 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2290 	    DD_FLAG_USED_BREAKDOWN) {
2291 		err = snaplist_space(&ddpa->origin_snaps,
2292 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2293 		    &ddpa->originusedsnap);
2294 		if (err != 0)
2295 			goto out;
2296 	}
2297 
2298 out:
2299 	promote_rele(ddpa, FTAG);
2300 	return (err);
2301 }
2302 
2303 static void
2304 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2305 {
2306 	dsl_dataset_promote_arg_t *ddpa = arg;
2307 	dsl_pool_t *dp = dmu_tx_pool(tx);
2308 	dsl_dataset_t *hds;
2309 	struct promotenode *snap;
2310 	dsl_dataset_t *origin_ds;
2311 	dsl_dataset_t *origin_head;
2312 	dsl_dir_t *dd;
2313 	dsl_dir_t *odd = NULL;
2314 	uint64_t oldnext_obj;
2315 	int64_t delta;
2316 
2317 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2318 	hds = ddpa->ddpa_clone;
2319 
2320 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2321 
2322 	snap = list_head(&ddpa->shared_snaps);
2323 	origin_ds = snap->ds;
2324 	dd = hds->ds_dir;
2325 
2326 	snap = list_head(&ddpa->origin_snaps);
2327 	origin_head = snap->ds;
2328 
2329 	/*
2330 	 * We need to explicitly open odd, since origin_ds's dd will be
2331 	 * changing.
2332 	 */
2333 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2334 	    NULL, FTAG, &odd));
2335 
2336 	/* change origin's next snap */
2337 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2338 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2339 	snap = list_tail(&ddpa->clone_snaps);
2340 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2341 	    origin_ds->ds_object);
2342 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2343 
2344 	/* change the origin's next clone */
2345 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2346 		dsl_dataset_remove_from_next_clones(origin_ds,
2347 		    snap->ds->ds_object, tx);
2348 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2349 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2350 		    oldnext_obj, tx));
2351 	}
2352 
2353 	/* change origin */
2354 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2355 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2356 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2357 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2358 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2359 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2360 	origin_head->ds_dir->dd_origin_txg =
2361 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2362 
2363 	/* change dd_clone entries */
2364 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2365 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2366 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2367 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2368 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2369 		    hds->ds_object, tx));
2370 
2371 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2372 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2373 		    origin_head->ds_object, tx));
2374 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2375 			dsl_dir_phys(dd)->dd_clones =
2376 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2377 			    DMU_OT_NONE, 0, tx);
2378 		}
2379 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2380 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2381 	}
2382 
2383 	/* move snapshots to this dir */
2384 	for (snap = list_head(&ddpa->shared_snaps); snap;
2385 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2386 		dsl_dataset_t *ds = snap->ds;
2387 
2388 		/*
2389 		 * Property callbacks are registered to a particular
2390 		 * dsl_dir.  Since ours is changing, evict the objset
2391 		 * so that they will be unregistered from the old dsl_dir.
2392 		 */
2393 		if (ds->ds_objset) {
2394 			dmu_objset_evict(ds->ds_objset);
2395 			ds->ds_objset = NULL;
2396 		}
2397 
2398 		/* move snap name entry */
2399 		VERIFY0(dsl_dataset_get_snapname(ds));
2400 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2401 		    ds->ds_snapname, tx, B_TRUE));
2402 		VERIFY0(zap_add(dp->dp_meta_objset,
2403 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2404 		    8, 1, &ds->ds_object, tx));
2405 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2406 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2407 
2408 		/* change containing dsl_dir */
2409 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2410 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2411 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2412 		ASSERT3P(ds->ds_dir, ==, odd);
2413 		dsl_dir_rele(ds->ds_dir, ds);
2414 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2415 		    NULL, ds, &ds->ds_dir));
2416 
2417 		/* move any clone references */
2418 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2419 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2420 			zap_cursor_t zc;
2421 			zap_attribute_t za;
2422 
2423 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2424 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2425 			    zap_cursor_retrieve(&zc, &za) == 0;
2426 			    zap_cursor_advance(&zc)) {
2427 				dsl_dataset_t *cnds;
2428 				uint64_t o;
2429 
2430 				if (za.za_first_integer == oldnext_obj) {
2431 					/*
2432 					 * We've already moved the
2433 					 * origin's reference.
2434 					 */
2435 					continue;
2436 				}
2437 
2438 				VERIFY0(dsl_dataset_hold_obj(dp,
2439 				    za.za_first_integer, FTAG, &cnds));
2440 				o = dsl_dir_phys(cnds->ds_dir)->
2441 				    dd_head_dataset_obj;
2442 
2443 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2444 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2445 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2446 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2447 				dsl_dataset_rele(cnds, FTAG);
2448 			}
2449 			zap_cursor_fini(&zc);
2450 		}
2451 
2452 		ASSERT(!dsl_prop_hascb(ds));
2453 	}
2454 
2455 	/*
2456 	 * Change space accounting.
2457 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2458 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2459 	 * is true for each of {clone,origin} independently.
2460 	 */
2461 
2462 	delta = ddpa->cloneusedsnap -
2463 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2464 	ASSERT3S(delta, >=, 0);
2465 	ASSERT3U(ddpa->used, >=, delta);
2466 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2467 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2468 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2469 
2470 	delta = ddpa->originusedsnap -
2471 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2472 	ASSERT3S(delta, <=, 0);
2473 	ASSERT3U(ddpa->used, >=, -delta);
2474 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2475 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2476 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2477 
2478 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2479 
2480 	/* log history record */
2481 	spa_history_log_internal_ds(hds, "promote", tx, "");
2482 
2483 	dsl_dir_rele(odd, FTAG);
2484 	promote_rele(ddpa, FTAG);
2485 }
2486 
2487 /*
2488  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2489  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2490  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2491  * snapshots back to this dataset's origin.
2492  */
2493 static int
2494 snaplist_make(dsl_pool_t *dp,
2495     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2496 {
2497 	uint64_t obj = last_obj;
2498 
2499 	list_create(l, sizeof (struct promotenode),
2500 	    offsetof(struct promotenode, link));
2501 
2502 	while (obj != first_obj) {
2503 		dsl_dataset_t *ds;
2504 		struct promotenode *snap;
2505 		int err;
2506 
2507 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2508 		ASSERT(err != ENOENT);
2509 		if (err != 0)
2510 			return (err);
2511 
2512 		if (first_obj == 0)
2513 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2514 
2515 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2516 		snap->ds = ds;
2517 		list_insert_tail(l, snap);
2518 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2519 	}
2520 
2521 	return (0);
2522 }
2523 
2524 static int
2525 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2526 {
2527 	struct promotenode *snap;
2528 
2529 	*spacep = 0;
2530 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2531 		uint64_t used, comp, uncomp;
2532 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2533 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2534 		*spacep += used;
2535 	}
2536 	return (0);
2537 }
2538 
2539 static void
2540 snaplist_destroy(list_t *l, void *tag)
2541 {
2542 	struct promotenode *snap;
2543 
2544 	if (l == NULL || !list_link_active(&l->list_head))
2545 		return;
2546 
2547 	while ((snap = list_tail(l)) != NULL) {
2548 		list_remove(l, snap);
2549 		dsl_dataset_rele(snap->ds, tag);
2550 		kmem_free(snap, sizeof (*snap));
2551 	}
2552 	list_destroy(l);
2553 }
2554 
2555 static int
2556 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2557 {
2558 	int error;
2559 	dsl_dir_t *dd;
2560 	struct promotenode *snap;
2561 
2562 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2563 	    &ddpa->ddpa_clone);
2564 	if (error != 0)
2565 		return (error);
2566 	dd = ddpa->ddpa_clone->ds_dir;
2567 
2568 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2569 	    !dsl_dir_is_clone(dd)) {
2570 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2571 		return (SET_ERROR(EINVAL));
2572 	}
2573 
2574 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2575 	    &ddpa->shared_snaps, tag);
2576 	if (error != 0)
2577 		goto out;
2578 
2579 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2580 	    &ddpa->clone_snaps, tag);
2581 	if (error != 0)
2582 		goto out;
2583 
2584 	snap = list_head(&ddpa->shared_snaps);
2585 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2586 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2587 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2588 	    &ddpa->origin_snaps, tag);
2589 	if (error != 0)
2590 		goto out;
2591 
2592 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2593 		error = dsl_dataset_hold_obj(dp,
2594 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2595 		    tag, &ddpa->origin_origin);
2596 		if (error != 0)
2597 			goto out;
2598 	}
2599 out:
2600 	if (error != 0)
2601 		promote_rele(ddpa, tag);
2602 	return (error);
2603 }
2604 
2605 static void
2606 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2607 {
2608 	snaplist_destroy(&ddpa->shared_snaps, tag);
2609 	snaplist_destroy(&ddpa->clone_snaps, tag);
2610 	snaplist_destroy(&ddpa->origin_snaps, tag);
2611 	if (ddpa->origin_origin != NULL)
2612 		dsl_dataset_rele(ddpa->origin_origin, tag);
2613 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2614 }
2615 
2616 /*
2617  * Promote a clone.
2618  *
2619  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2620  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2621  */
2622 int
2623 dsl_dataset_promote(const char *name, char *conflsnap)
2624 {
2625 	dsl_dataset_promote_arg_t ddpa = { 0 };
2626 	uint64_t numsnaps;
2627 	int error;
2628 	objset_t *os;
2629 
2630 	/*
2631 	 * We will modify space proportional to the number of
2632 	 * snapshots.  Compute numsnaps.
2633 	 */
2634 	error = dmu_objset_hold(name, FTAG, &os);
2635 	if (error != 0)
2636 		return (error);
2637 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2638 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2639 	    &numsnaps);
2640 	dmu_objset_rele(os, FTAG);
2641 	if (error != 0)
2642 		return (error);
2643 
2644 	ddpa.ddpa_clonename = name;
2645 	ddpa.err_ds = conflsnap;
2646 	ddpa.cr = CRED();
2647 
2648 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2649 	    dsl_dataset_promote_sync, &ddpa,
2650 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2651 }
2652 
2653 int
2654 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2655     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2656 {
2657 	int64_t unused_refres_delta;
2658 
2659 	/* they should both be heads */
2660 	if (clone->ds_is_snapshot ||
2661 	    origin_head->ds_is_snapshot)
2662 		return (SET_ERROR(EINVAL));
2663 
2664 	/* if we are not forcing, the branch point should be just before them */
2665 	if (!force && clone->ds_prev != origin_head->ds_prev)
2666 		return (SET_ERROR(EINVAL));
2667 
2668 	/* clone should be the clone (unless they are unrelated) */
2669 	if (clone->ds_prev != NULL &&
2670 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2671 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2672 		return (SET_ERROR(EINVAL));
2673 
2674 	/* the clone should be a child of the origin */
2675 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2676 		return (SET_ERROR(EINVAL));
2677 
2678 	/* origin_head shouldn't be modified unless 'force' */
2679 	if (!force &&
2680 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2681 		return (SET_ERROR(ETXTBSY));
2682 
2683 	/* origin_head should have no long holds (e.g. is not mounted) */
2684 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2685 		return (SET_ERROR(EBUSY));
2686 
2687 	/* check amount of any unconsumed refreservation */
2688 	unused_refres_delta =
2689 	    (int64_t)MIN(origin_head->ds_reserved,
2690 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2691 	    (int64_t)MIN(origin_head->ds_reserved,
2692 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2693 
2694 	if (unused_refres_delta > 0 &&
2695 	    unused_refres_delta >
2696 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2697 		return (SET_ERROR(ENOSPC));
2698 
2699 	/* clone can't be over the head's refquota */
2700 	if (origin_head->ds_quota != 0 &&
2701 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2702 	    origin_head->ds_quota)
2703 		return (SET_ERROR(EDQUOT));
2704 
2705 	return (0);
2706 }
2707 
2708 void
2709 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2710     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2711 {
2712 	dsl_pool_t *dp = dmu_tx_pool(tx);
2713 	int64_t unused_refres_delta;
2714 
2715 	ASSERT(clone->ds_reserved == 0);
2716 	ASSERT(origin_head->ds_quota == 0 ||
2717 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2718 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2719 
2720 	/*
2721 	 * Swap per-dataset feature flags.
2722 	 */
2723 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2724 		if (!(spa_feature_table[f].fi_flags &
2725 		    ZFEATURE_FLAG_PER_DATASET)) {
2726 			ASSERT(!clone->ds_feature_inuse[f]);
2727 			ASSERT(!origin_head->ds_feature_inuse[f]);
2728 			continue;
2729 		}
2730 
2731 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
2732 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2733 
2734 		if (clone_inuse) {
2735 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2736 			clone->ds_feature_inuse[f] = B_FALSE;
2737 		}
2738 		if (origin_head_inuse) {
2739 			dsl_dataset_deactivate_feature(origin_head->ds_object,
2740 			    f, tx);
2741 			origin_head->ds_feature_inuse[f] = B_FALSE;
2742 		}
2743 		if (clone_inuse) {
2744 			dsl_dataset_activate_feature(origin_head->ds_object,
2745 			    f, tx);
2746 			origin_head->ds_feature_inuse[f] = B_TRUE;
2747 		}
2748 		if (origin_head_inuse) {
2749 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
2750 			clone->ds_feature_inuse[f] = B_TRUE;
2751 		}
2752 	}
2753 
2754 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2755 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2756 
2757 	if (clone->ds_objset != NULL) {
2758 		dmu_objset_evict(clone->ds_objset);
2759 		clone->ds_objset = NULL;
2760 	}
2761 
2762 	if (origin_head->ds_objset != NULL) {
2763 		dmu_objset_evict(origin_head->ds_objset);
2764 		origin_head->ds_objset = NULL;
2765 	}
2766 
2767 	unused_refres_delta =
2768 	    (int64_t)MIN(origin_head->ds_reserved,
2769 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2770 	    (int64_t)MIN(origin_head->ds_reserved,
2771 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2772 
2773 	/*
2774 	 * Reset origin's unique bytes, if it exists.
2775 	 */
2776 	if (clone->ds_prev) {
2777 		dsl_dataset_t *origin = clone->ds_prev;
2778 		uint64_t comp, uncomp;
2779 
2780 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2781 		dsl_deadlist_space_range(&clone->ds_deadlist,
2782 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2783 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2784 	}
2785 
2786 	/* swap blkptrs */
2787 	{
2788 		blkptr_t tmp;
2789 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2790 		dsl_dataset_phys(origin_head)->ds_bp =
2791 		    dsl_dataset_phys(clone)->ds_bp;
2792 		dsl_dataset_phys(clone)->ds_bp = tmp;
2793 	}
2794 
2795 	/* set dd_*_bytes */
2796 	{
2797 		int64_t dused, dcomp, duncomp;
2798 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2799 		uint64_t odl_used, odl_comp, odl_uncomp;
2800 
2801 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2802 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2803 
2804 		dsl_deadlist_space(&clone->ds_deadlist,
2805 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2806 		dsl_deadlist_space(&origin_head->ds_deadlist,
2807 		    &odl_used, &odl_comp, &odl_uncomp);
2808 
2809 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2810 		    cdl_used -
2811 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2812 		    odl_used);
2813 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2814 		    cdl_comp -
2815 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2816 		    odl_comp);
2817 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2818 		    cdl_uncomp -
2819 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2820 		    odl_uncomp);
2821 
2822 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2823 		    dused, dcomp, duncomp, tx);
2824 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2825 		    -dused, -dcomp, -duncomp, tx);
2826 
2827 		/*
2828 		 * The difference in the space used by snapshots is the
2829 		 * difference in snapshot space due to the head's
2830 		 * deadlist (since that's the only thing that's
2831 		 * changing that affects the snapused).
2832 		 */
2833 		dsl_deadlist_space_range(&clone->ds_deadlist,
2834 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2835 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2836 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2837 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2838 		    &odl_used, &odl_comp, &odl_uncomp);
2839 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2840 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2841 	}
2842 
2843 	/* swap ds_*_bytes */
2844 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2845 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
2846 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2847 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
2848 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2849 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2850 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2851 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2852 
2853 	/* apply any parent delta for change in unconsumed refreservation */
2854 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2855 	    unused_refres_delta, 0, 0, tx);
2856 
2857 	/*
2858 	 * Swap deadlists.
2859 	 */
2860 	dsl_deadlist_close(&clone->ds_deadlist);
2861 	dsl_deadlist_close(&origin_head->ds_deadlist);
2862 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2863 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2864 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2865 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2866 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2867 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2868 
2869 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2870 
2871 	spa_history_log_internal_ds(clone, "clone swap", tx,
2872 	    "parent=%s", origin_head->ds_dir->dd_myname);
2873 }
2874 
2875 /*
2876  * Given a pool name and a dataset object number in that pool,
2877  * return the name of that dataset.
2878  */
2879 int
2880 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2881 {
2882 	dsl_pool_t *dp;
2883 	dsl_dataset_t *ds;
2884 	int error;
2885 
2886 	error = dsl_pool_hold(pname, FTAG, &dp);
2887 	if (error != 0)
2888 		return (error);
2889 
2890 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2891 	if (error == 0) {
2892 		dsl_dataset_name(ds, buf);
2893 		dsl_dataset_rele(ds, FTAG);
2894 	}
2895 	dsl_pool_rele(dp, FTAG);
2896 
2897 	return (error);
2898 }
2899 
2900 int
2901 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2902     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2903 {
2904 	int error = 0;
2905 
2906 	ASSERT3S(asize, >, 0);
2907 
2908 	/*
2909 	 * *ref_rsrv is the portion of asize that will come from any
2910 	 * unconsumed refreservation space.
2911 	 */
2912 	*ref_rsrv = 0;
2913 
2914 	mutex_enter(&ds->ds_lock);
2915 	/*
2916 	 * Make a space adjustment for reserved bytes.
2917 	 */
2918 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2919 		ASSERT3U(*used, >=,
2920 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2921 		*used -=
2922 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2923 		*ref_rsrv =
2924 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2925 	}
2926 
2927 	if (!check_quota || ds->ds_quota == 0) {
2928 		mutex_exit(&ds->ds_lock);
2929 		return (0);
2930 	}
2931 	/*
2932 	 * If they are requesting more space, and our current estimate
2933 	 * is over quota, they get to try again unless the actual
2934 	 * on-disk is over quota and there are no pending changes (which
2935 	 * may free up space for us).
2936 	 */
2937 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2938 	    ds->ds_quota) {
2939 		if (inflight > 0 ||
2940 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2941 			error = SET_ERROR(ERESTART);
2942 		else
2943 			error = SET_ERROR(EDQUOT);
2944 	}
2945 	mutex_exit(&ds->ds_lock);
2946 
2947 	return (error);
2948 }
2949 
2950 typedef struct dsl_dataset_set_qr_arg {
2951 	const char *ddsqra_name;
2952 	zprop_source_t ddsqra_source;
2953 	uint64_t ddsqra_value;
2954 } dsl_dataset_set_qr_arg_t;
2955 
2956 
2957 /* ARGSUSED */
2958 static int
2959 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2960 {
2961 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2962 	dsl_pool_t *dp = dmu_tx_pool(tx);
2963 	dsl_dataset_t *ds;
2964 	int error;
2965 	uint64_t newval;
2966 
2967 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2968 		return (SET_ERROR(ENOTSUP));
2969 
2970 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2971 	if (error != 0)
2972 		return (error);
2973 
2974 	if (ds->ds_is_snapshot) {
2975 		dsl_dataset_rele(ds, FTAG);
2976 		return (SET_ERROR(EINVAL));
2977 	}
2978 
2979 	error = dsl_prop_predict(ds->ds_dir,
2980 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2981 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2982 	if (error != 0) {
2983 		dsl_dataset_rele(ds, FTAG);
2984 		return (error);
2985 	}
2986 
2987 	if (newval == 0) {
2988 		dsl_dataset_rele(ds, FTAG);
2989 		return (0);
2990 	}
2991 
2992 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
2993 	    newval < ds->ds_reserved) {
2994 		dsl_dataset_rele(ds, FTAG);
2995 		return (SET_ERROR(ENOSPC));
2996 	}
2997 
2998 	dsl_dataset_rele(ds, FTAG);
2999 	return (0);
3000 }
3001 
3002 static void
3003 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3004 {
3005 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3006 	dsl_pool_t *dp = dmu_tx_pool(tx);
3007 	dsl_dataset_t *ds;
3008 	uint64_t newval;
3009 
3010 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3011 
3012 	dsl_prop_set_sync_impl(ds,
3013 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3014 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3015 	    &ddsqra->ddsqra_value, tx);
3016 
3017 	VERIFY0(dsl_prop_get_int_ds(ds,
3018 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3019 
3020 	if (ds->ds_quota != newval) {
3021 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3022 		ds->ds_quota = newval;
3023 	}
3024 	dsl_dataset_rele(ds, FTAG);
3025 }
3026 
3027 int
3028 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3029     uint64_t refquota)
3030 {
3031 	dsl_dataset_set_qr_arg_t ddsqra;
3032 
3033 	ddsqra.ddsqra_name = dsname;
3034 	ddsqra.ddsqra_source = source;
3035 	ddsqra.ddsqra_value = refquota;
3036 
3037 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3038 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3039 }
3040 
3041 static int
3042 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3043 {
3044 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3045 	dsl_pool_t *dp = dmu_tx_pool(tx);
3046 	dsl_dataset_t *ds;
3047 	int error;
3048 	uint64_t newval, unique;
3049 
3050 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3051 		return (SET_ERROR(ENOTSUP));
3052 
3053 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3054 	if (error != 0)
3055 		return (error);
3056 
3057 	if (ds->ds_is_snapshot) {
3058 		dsl_dataset_rele(ds, FTAG);
3059 		return (SET_ERROR(EINVAL));
3060 	}
3061 
3062 	error = dsl_prop_predict(ds->ds_dir,
3063 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3064 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3065 	if (error != 0) {
3066 		dsl_dataset_rele(ds, FTAG);
3067 		return (error);
3068 	}
3069 
3070 	/*
3071 	 * If we are doing the preliminary check in open context, the
3072 	 * space estimates may be inaccurate.
3073 	 */
3074 	if (!dmu_tx_is_syncing(tx)) {
3075 		dsl_dataset_rele(ds, FTAG);
3076 		return (0);
3077 	}
3078 
3079 	mutex_enter(&ds->ds_lock);
3080 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3081 		dsl_dataset_recalc_head_uniq(ds);
3082 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3083 	mutex_exit(&ds->ds_lock);
3084 
3085 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3086 		uint64_t delta = MAX(unique, newval) -
3087 		    MAX(unique, ds->ds_reserved);
3088 
3089 		if (delta >
3090 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3091 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3092 			dsl_dataset_rele(ds, FTAG);
3093 			return (SET_ERROR(ENOSPC));
3094 		}
3095 	}
3096 
3097 	dsl_dataset_rele(ds, FTAG);
3098 	return (0);
3099 }
3100 
3101 void
3102 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3103     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3104 {
3105 	uint64_t newval;
3106 	uint64_t unique;
3107 	int64_t delta;
3108 
3109 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3110 	    source, sizeof (value), 1, &value, tx);
3111 
3112 	VERIFY0(dsl_prop_get_int_ds(ds,
3113 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3114 
3115 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3116 	mutex_enter(&ds->ds_dir->dd_lock);
3117 	mutex_enter(&ds->ds_lock);
3118 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3119 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3120 	delta = MAX(0, (int64_t)(newval - unique)) -
3121 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3122 	ds->ds_reserved = newval;
3123 	mutex_exit(&ds->ds_lock);
3124 
3125 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3126 	mutex_exit(&ds->ds_dir->dd_lock);
3127 }
3128 
3129 static void
3130 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3131 {
3132 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3133 	dsl_pool_t *dp = dmu_tx_pool(tx);
3134 	dsl_dataset_t *ds;
3135 
3136 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3137 	dsl_dataset_set_refreservation_sync_impl(ds,
3138 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3139 	dsl_dataset_rele(ds, FTAG);
3140 }
3141 
3142 int
3143 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3144     uint64_t refreservation)
3145 {
3146 	dsl_dataset_set_qr_arg_t ddsqra;
3147 
3148 	ddsqra.ddsqra_name = dsname;
3149 	ddsqra.ddsqra_source = source;
3150 	ddsqra.ddsqra_value = refreservation;
3151 
3152 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3153 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3154 	    0, ZFS_SPACE_CHECK_NONE));
3155 }
3156 
3157 /*
3158  * Return (in *usedp) the amount of space written in new that is not
3159  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3160  * a snapshot before new, in new's filesystem (or its origin).  If not then
3161  * fail and return EINVAL.
3162  *
3163  * The written space is calculated by considering two components:  First, we
3164  * ignore any freed space, and calculate the written as new's used space
3165  * minus old's used space.  Next, we add in the amount of space that was freed
3166  * between the two snapshots, thus reducing new's used space relative to old's.
3167  * Specifically, this is the space that was born before old->ds_creation_txg,
3168  * and freed before new (ie. on new's deadlist or a previous deadlist).
3169  *
3170  * space freed                         [---------------------]
3171  * snapshots                       ---O-------O--------O-------O------
3172  *                                         oldsnap            new
3173  */
3174 int
3175 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3176     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3177 {
3178 	int err = 0;
3179 	uint64_t snapobj;
3180 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3181 
3182 	ASSERT(dsl_pool_config_held(dp));
3183 
3184 	*usedp = 0;
3185 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3186 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3187 
3188 	*compp = 0;
3189 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3190 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3191 
3192 	*uncompp = 0;
3193 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3194 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3195 
3196 	snapobj = new->ds_object;
3197 	while (snapobj != oldsnap->ds_object) {
3198 		dsl_dataset_t *snap;
3199 		uint64_t used, comp, uncomp;
3200 
3201 		if (snapobj == new->ds_object) {
3202 			snap = new;
3203 		} else {
3204 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3205 			if (err != 0)
3206 				break;
3207 		}
3208 
3209 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3210 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3211 			/*
3212 			 * The blocks in the deadlist can not be born after
3213 			 * ds_prev_snap_txg, so get the whole deadlist space,
3214 			 * which is more efficient (especially for old-format
3215 			 * deadlists).  Unfortunately the deadlist code
3216 			 * doesn't have enough information to make this
3217 			 * optimization itself.
3218 			 */
3219 			dsl_deadlist_space(&snap->ds_deadlist,
3220 			    &used, &comp, &uncomp);
3221 		} else {
3222 			dsl_deadlist_space_range(&snap->ds_deadlist,
3223 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3224 			    &used, &comp, &uncomp);
3225 		}
3226 		*usedp += used;
3227 		*compp += comp;
3228 		*uncompp += uncomp;
3229 
3230 		/*
3231 		 * If we get to the beginning of the chain of snapshots
3232 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3233 		 * was not a snapshot of/before new.
3234 		 */
3235 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3236 		if (snap != new)
3237 			dsl_dataset_rele(snap, FTAG);
3238 		if (snapobj == 0) {
3239 			err = SET_ERROR(EINVAL);
3240 			break;
3241 		}
3242 
3243 	}
3244 	return (err);
3245 }
3246 
3247 /*
3248  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3249  * lastsnap, and all snapshots in between are deleted.
3250  *
3251  * blocks that would be freed            [---------------------------]
3252  * snapshots                       ---O-------O--------O-------O--------O
3253  *                                        firstsnap        lastsnap
3254  *
3255  * This is the set of blocks that were born after the snap before firstsnap,
3256  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3257  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3258  * We calculate this by iterating over the relevant deadlists (from the snap
3259  * after lastsnap, backward to the snap after firstsnap), summing up the
3260  * space on the deadlist that was born after the snap before firstsnap.
3261  */
3262 int
3263 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3264     dsl_dataset_t *lastsnap,
3265     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3266 {
3267 	int err = 0;
3268 	uint64_t snapobj;
3269 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3270 
3271 	ASSERT(firstsnap->ds_is_snapshot);
3272 	ASSERT(lastsnap->ds_is_snapshot);
3273 
3274 	/*
3275 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3276 	 * is before lastsnap.
3277 	 */
3278 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3279 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3280 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3281 		return (SET_ERROR(EINVAL));
3282 
3283 	*usedp = *compp = *uncompp = 0;
3284 
3285 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3286 	while (snapobj != firstsnap->ds_object) {
3287 		dsl_dataset_t *ds;
3288 		uint64_t used, comp, uncomp;
3289 
3290 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3291 		if (err != 0)
3292 			break;
3293 
3294 		dsl_deadlist_space_range(&ds->ds_deadlist,
3295 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3296 		    &used, &comp, &uncomp);
3297 		*usedp += used;
3298 		*compp += comp;
3299 		*uncompp += uncomp;
3300 
3301 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3302 		ASSERT3U(snapobj, !=, 0);
3303 		dsl_dataset_rele(ds, FTAG);
3304 	}
3305 	return (err);
3306 }
3307 
3308 /*
3309  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3310  * For example, they could both be snapshots of the same filesystem, and
3311  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3312  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3313  * filesystem.  Or 'earlier' could be the origin's origin.
3314  *
3315  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3316  */
3317 boolean_t
3318 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3319 	uint64_t earlier_txg)
3320 {
3321 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3322 	int error;
3323 	boolean_t ret;
3324 
3325 	ASSERT(dsl_pool_config_held(dp));
3326 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3327 
3328 	if (earlier_txg == 0)
3329 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3330 
3331 	if (later->ds_is_snapshot &&
3332 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3333 		return (B_FALSE);
3334 
3335 	if (later->ds_dir == earlier->ds_dir)
3336 		return (B_TRUE);
3337 	if (!dsl_dir_is_clone(later->ds_dir))
3338 		return (B_FALSE);
3339 
3340 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3341 		return (B_TRUE);
3342 	dsl_dataset_t *origin;
3343 	error = dsl_dataset_hold_obj(dp,
3344 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3345 	if (error != 0)
3346 		return (B_FALSE);
3347 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3348 	dsl_dataset_rele(origin, FTAG);
3349 	return (ret);
3350 }
3351 
3352 void
3353 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3354 {
3355 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3356 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3357 }
3358