xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 29e54759b5b8e21fb481d44ee504a764aa7b6ea2)
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 #include <sys/dmu_send.h>
54 #include <sys/zio_checksum.h>
55 #include <sys/zio_compress.h>
56 #include <zfs_fletcher.h>
57 
58 /*
59  * The SPA supports block sizes up to 16MB.  However, very large blocks
60  * can have an impact on i/o latency (e.g. tying up a spinning disk for
61  * ~300ms), and also potentially on the memory allocator.  Therefore,
62  * we do not allow the recordsize to be set larger than zfs_max_recordsize
63  * (default 1MB).  Larger blocks can be created by changing this tunable,
64  * and pools with larger blocks can always be imported and used, regardless
65  * of this setting.
66  */
67 int zfs_max_recordsize = 1 * 1024 * 1024;
68 
69 #define	SWITCH64(x, y) \
70 	{ \
71 		uint64_t __tmp = (x); \
72 		(x) = (y); \
73 		(y) = __tmp; \
74 	}
75 
76 #define	DS_REF_MAX	(1ULL << 62)
77 
78 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
79 
80 /*
81  * Figure out how much of this delta should be propogated to the dsl_dir
82  * layer.  If there's a refreservation, that space has already been
83  * partially accounted for in our ancestors.
84  */
85 static int64_t
86 parent_delta(dsl_dataset_t *ds, int64_t delta)
87 {
88 	dsl_dataset_phys_t *ds_phys;
89 	uint64_t old_bytes, new_bytes;
90 
91 	if (ds->ds_reserved == 0)
92 		return (delta);
93 
94 	ds_phys = dsl_dataset_phys(ds);
95 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
96 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
97 
98 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
99 	return (new_bytes - old_bytes);
100 }
101 
102 void
103 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
104 {
105 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
106 	int compressed = BP_GET_PSIZE(bp);
107 	int uncompressed = BP_GET_UCSIZE(bp);
108 	int64_t delta;
109 
110 	dprintf_bp(bp, "ds=%p", ds);
111 
112 	ASSERT(dmu_tx_is_syncing(tx));
113 	/* It could have been compressed away to nothing */
114 	if (BP_IS_HOLE(bp))
115 		return;
116 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
117 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
118 	if (ds == NULL) {
119 		dsl_pool_mos_diduse_space(tx->tx_pool,
120 		    used, compressed, uncompressed);
121 		return;
122 	}
123 
124 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
125 	mutex_enter(&ds->ds_lock);
126 	delta = parent_delta(ds, used);
127 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
128 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
129 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
130 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
131 
132 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
133 		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
134 		    B_TRUE;
135 	}
136 
137 	spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
138 	if (f != SPA_FEATURE_NONE)
139 		ds->ds_feature_activation_needed[f] = B_TRUE;
140 
141 	mutex_exit(&ds->ds_lock);
142 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
143 	    compressed, uncompressed, tx);
144 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
145 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
146 }
147 
148 int
149 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
150     boolean_t async)
151 {
152 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
153 	int compressed = BP_GET_PSIZE(bp);
154 	int uncompressed = BP_GET_UCSIZE(bp);
155 
156 	if (BP_IS_HOLE(bp))
157 		return (0);
158 
159 	ASSERT(dmu_tx_is_syncing(tx));
160 	ASSERT(bp->blk_birth <= tx->tx_txg);
161 
162 	if (ds == NULL) {
163 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
164 		dsl_pool_mos_diduse_space(tx->tx_pool,
165 		    -used, -compressed, -uncompressed);
166 		return (used);
167 	}
168 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
169 
170 	ASSERT(!ds->ds_is_snapshot);
171 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
172 
173 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
174 		int64_t delta;
175 
176 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
177 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
178 
179 		mutex_enter(&ds->ds_lock);
180 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
181 		    !DS_UNIQUE_IS_ACCURATE(ds));
182 		delta = parent_delta(ds, -used);
183 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
184 		mutex_exit(&ds->ds_lock);
185 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
186 		    delta, -compressed, -uncompressed, tx);
187 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
188 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
189 	} else {
190 		dprintf_bp(bp, "putting on dead list: %s", "");
191 		if (async) {
192 			/*
193 			 * We are here as part of zio's write done callback,
194 			 * which means we're a zio interrupt thread.  We can't
195 			 * call dsl_deadlist_insert() now because it may block
196 			 * waiting for I/O.  Instead, put bp on the deferred
197 			 * queue and let dsl_pool_sync() finish the job.
198 			 */
199 			bplist_append(&ds->ds_pending_deadlist, bp);
200 		} else {
201 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
202 		}
203 		ASSERT3U(ds->ds_prev->ds_object, ==,
204 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
205 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
206 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
207 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
208 		    ds->ds_object && bp->blk_birth >
209 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
210 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
211 			mutex_enter(&ds->ds_prev->ds_lock);
212 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
213 			mutex_exit(&ds->ds_prev->ds_lock);
214 		}
215 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
216 			dsl_dir_transfer_space(ds->ds_dir, used,
217 			    DD_USED_HEAD, DD_USED_SNAP, tx);
218 		}
219 	}
220 	mutex_enter(&ds->ds_lock);
221 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
222 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
223 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
224 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
225 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
226 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
227 	mutex_exit(&ds->ds_lock);
228 
229 	return (used);
230 }
231 
232 uint64_t
233 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
234 {
235 	uint64_t trysnap = 0;
236 
237 	if (ds == NULL)
238 		return (0);
239 	/*
240 	 * The snapshot creation could fail, but that would cause an
241 	 * incorrect FALSE return, which would only result in an
242 	 * overestimation of the amount of space that an operation would
243 	 * consume, which is OK.
244 	 *
245 	 * There's also a small window where we could miss a pending
246 	 * snapshot, because we could set the sync task in the quiescing
247 	 * phase.  So this should only be used as a guess.
248 	 */
249 	if (ds->ds_trysnap_txg >
250 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
251 		trysnap = ds->ds_trysnap_txg;
252 	return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
253 }
254 
255 boolean_t
256 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
257     uint64_t blk_birth)
258 {
259 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
260 	    (bp != NULL && BP_IS_HOLE(bp)))
261 		return (B_FALSE);
262 
263 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
264 
265 	return (B_TRUE);
266 }
267 
268 static void
269 dsl_dataset_evict(void *dbu)
270 {
271 	dsl_dataset_t *ds = dbu;
272 
273 	ASSERT(ds->ds_owner == NULL);
274 
275 	ds->ds_dbuf = NULL;
276 
277 	unique_remove(ds->ds_fsid_guid);
278 
279 	if (ds->ds_objset != NULL)
280 		dmu_objset_evict(ds->ds_objset);
281 
282 	if (ds->ds_prev) {
283 		dsl_dataset_rele(ds->ds_prev, ds);
284 		ds->ds_prev = NULL;
285 	}
286 
287 	bplist_destroy(&ds->ds_pending_deadlist);
288 	if (ds->ds_deadlist.dl_os != NULL)
289 		dsl_deadlist_close(&ds->ds_deadlist);
290 	if (ds->ds_dir)
291 		dsl_dir_async_rele(ds->ds_dir, ds);
292 
293 	ASSERT(!list_link_active(&ds->ds_synced_link));
294 
295 	list_destroy(&ds->ds_prop_cbs);
296 	mutex_destroy(&ds->ds_lock);
297 	mutex_destroy(&ds->ds_opening_lock);
298 	mutex_destroy(&ds->ds_sendstream_lock);
299 	refcount_destroy(&ds->ds_longholds);
300 
301 	kmem_free(ds, sizeof (dsl_dataset_t));
302 }
303 
304 int
305 dsl_dataset_get_snapname(dsl_dataset_t *ds)
306 {
307 	dsl_dataset_phys_t *headphys;
308 	int err;
309 	dmu_buf_t *headdbuf;
310 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
311 	objset_t *mos = dp->dp_meta_objset;
312 
313 	if (ds->ds_snapname[0])
314 		return (0);
315 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
316 		return (0);
317 
318 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
319 	    FTAG, &headdbuf);
320 	if (err != 0)
321 		return (err);
322 	headphys = headdbuf->db_data;
323 	err = zap_value_search(dp->dp_meta_objset,
324 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
325 	dmu_buf_rele(headdbuf, FTAG);
326 	return (err);
327 }
328 
329 int
330 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
331 {
332 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
333 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
334 	matchtype_t mt;
335 	int err;
336 
337 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
338 		mt = MT_FIRST;
339 	else
340 		mt = MT_EXACT;
341 
342 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
343 	    value, mt, NULL, 0, NULL);
344 	if (err == ENOTSUP && mt == MT_FIRST)
345 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
346 	return (err);
347 }
348 
349 int
350 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
351     boolean_t adj_cnt)
352 {
353 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
354 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
355 	matchtype_t mt;
356 	int err;
357 
358 	dsl_dir_snap_cmtime_update(ds->ds_dir);
359 
360 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
361 		mt = MT_FIRST;
362 	else
363 		mt = MT_EXACT;
364 
365 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
366 	if (err == ENOTSUP && mt == MT_FIRST)
367 		err = zap_remove(mos, snapobj, name, tx);
368 
369 	if (err == 0 && adj_cnt)
370 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
371 		    DD_FIELD_SNAPSHOT_COUNT, tx);
372 
373 	return (err);
374 }
375 
376 boolean_t
377 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
378 {
379 	dmu_buf_t *dbuf = ds->ds_dbuf;
380 	boolean_t result = B_FALSE;
381 
382 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
383 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
384 
385 		if (ds == dmu_buf_get_user(dbuf))
386 			result = B_TRUE;
387 		else
388 			dmu_buf_rele(dbuf, tag);
389 	}
390 
391 	return (result);
392 }
393 
394 int
395 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
396     dsl_dataset_t **dsp)
397 {
398 	objset_t *mos = dp->dp_meta_objset;
399 	dmu_buf_t *dbuf;
400 	dsl_dataset_t *ds;
401 	int err;
402 	dmu_object_info_t doi;
403 
404 	ASSERT(dsl_pool_config_held(dp));
405 
406 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
407 	if (err != 0)
408 		return (err);
409 
410 	/* Make sure dsobj has the correct object type. */
411 	dmu_object_info_from_db(dbuf, &doi);
412 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
413 		dmu_buf_rele(dbuf, tag);
414 		return (SET_ERROR(EINVAL));
415 	}
416 
417 	ds = dmu_buf_get_user(dbuf);
418 	if (ds == NULL) {
419 		dsl_dataset_t *winner = NULL;
420 
421 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
422 		ds->ds_dbuf = dbuf;
423 		ds->ds_object = dsobj;
424 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
425 
426 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
427 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
428 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
429 		refcount_create(&ds->ds_longholds);
430 
431 		bplist_create(&ds->ds_pending_deadlist);
432 		dsl_deadlist_open(&ds->ds_deadlist,
433 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
434 
435 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
436 		    offsetof(dmu_sendarg_t, dsa_link));
437 
438 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
439 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
440 
441 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
442 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
443 				if (!(spa_feature_table[f].fi_flags &
444 				    ZFEATURE_FLAG_PER_DATASET))
445 					continue;
446 				err = zap_contains(mos, dsobj,
447 				    spa_feature_table[f].fi_guid);
448 				if (err == 0) {
449 					ds->ds_feature_inuse[f] = B_TRUE;
450 				} else {
451 					ASSERT3U(err, ==, ENOENT);
452 					err = 0;
453 				}
454 			}
455 		}
456 
457 		err = dsl_dir_hold_obj(dp,
458 		    dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
459 		if (err != 0) {
460 			mutex_destroy(&ds->ds_lock);
461 			mutex_destroy(&ds->ds_opening_lock);
462 			mutex_destroy(&ds->ds_sendstream_lock);
463 			refcount_destroy(&ds->ds_longholds);
464 			bplist_destroy(&ds->ds_pending_deadlist);
465 			dsl_deadlist_close(&ds->ds_deadlist);
466 			kmem_free(ds, sizeof (dsl_dataset_t));
467 			dmu_buf_rele(dbuf, tag);
468 			return (err);
469 		}
470 
471 		if (!ds->ds_is_snapshot) {
472 			ds->ds_snapname[0] = '\0';
473 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
474 				err = dsl_dataset_hold_obj(dp,
475 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
476 				    ds, &ds->ds_prev);
477 			}
478 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
479 				int zaperr = zap_lookup(mos, ds->ds_object,
480 				    DS_FIELD_BOOKMARK_NAMES,
481 				    sizeof (ds->ds_bookmarks), 1,
482 				    &ds->ds_bookmarks);
483 				if (zaperr != ENOENT)
484 					VERIFY0(zaperr);
485 			}
486 		} else {
487 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
488 				err = dsl_dataset_get_snapname(ds);
489 			if (err == 0 &&
490 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
491 				err = zap_count(
492 				    ds->ds_dir->dd_pool->dp_meta_objset,
493 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
494 				    &ds->ds_userrefs);
495 			}
496 		}
497 
498 		if (err == 0 && !ds->ds_is_snapshot) {
499 			err = dsl_prop_get_int_ds(ds,
500 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
501 			    &ds->ds_reserved);
502 			if (err == 0) {
503 				err = dsl_prop_get_int_ds(ds,
504 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
505 				    &ds->ds_quota);
506 			}
507 		} else {
508 			ds->ds_reserved = ds->ds_quota = 0;
509 		}
510 
511 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
512 		if (err == 0)
513 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
514 
515 		if (err != 0 || winner != NULL) {
516 			bplist_destroy(&ds->ds_pending_deadlist);
517 			dsl_deadlist_close(&ds->ds_deadlist);
518 			if (ds->ds_prev)
519 				dsl_dataset_rele(ds->ds_prev, ds);
520 			dsl_dir_rele(ds->ds_dir, ds);
521 			mutex_destroy(&ds->ds_lock);
522 			mutex_destroy(&ds->ds_opening_lock);
523 			mutex_destroy(&ds->ds_sendstream_lock);
524 			refcount_destroy(&ds->ds_longholds);
525 			kmem_free(ds, sizeof (dsl_dataset_t));
526 			if (err != 0) {
527 				dmu_buf_rele(dbuf, tag);
528 				return (err);
529 			}
530 			ds = winner;
531 		} else {
532 			ds->ds_fsid_guid =
533 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
534 		}
535 	}
536 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
537 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
538 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
539 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
540 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
541 	*dsp = ds;
542 	return (0);
543 }
544 
545 int
546 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
547     void *tag, dsl_dataset_t **dsp)
548 {
549 	dsl_dir_t *dd;
550 	const char *snapname;
551 	uint64_t obj;
552 	int err = 0;
553 	dsl_dataset_t *ds;
554 
555 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
556 	if (err != 0)
557 		return (err);
558 
559 	ASSERT(dsl_pool_config_held(dp));
560 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
561 	if (obj != 0)
562 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
563 	else
564 		err = SET_ERROR(ENOENT);
565 
566 	/* we may be looking for a snapshot */
567 	if (err == 0 && snapname != NULL) {
568 		dsl_dataset_t *snap_ds;
569 
570 		if (*snapname++ != '@') {
571 			dsl_dataset_rele(ds, tag);
572 			dsl_dir_rele(dd, FTAG);
573 			return (SET_ERROR(ENOENT));
574 		}
575 
576 		dprintf("looking for snapshot '%s'\n", snapname);
577 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
578 		if (err == 0)
579 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
580 		dsl_dataset_rele(ds, tag);
581 
582 		if (err == 0) {
583 			mutex_enter(&snap_ds->ds_lock);
584 			if (snap_ds->ds_snapname[0] == 0)
585 				(void) strlcpy(snap_ds->ds_snapname, snapname,
586 				    sizeof (snap_ds->ds_snapname));
587 			mutex_exit(&snap_ds->ds_lock);
588 			ds = snap_ds;
589 		}
590 	}
591 	if (err == 0)
592 		*dsp = ds;
593 	dsl_dir_rele(dd, FTAG);
594 	return (err);
595 }
596 
597 int
598 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
599     void *tag, dsl_dataset_t **dsp)
600 {
601 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
602 	if (err != 0)
603 		return (err);
604 	if (!dsl_dataset_tryown(*dsp, tag)) {
605 		dsl_dataset_rele(*dsp, tag);
606 		*dsp = NULL;
607 		return (SET_ERROR(EBUSY));
608 	}
609 	return (0);
610 }
611 
612 int
613 dsl_dataset_own(dsl_pool_t *dp, const char *name,
614     void *tag, dsl_dataset_t **dsp)
615 {
616 	int err = dsl_dataset_hold(dp, name, tag, dsp);
617 	if (err != 0)
618 		return (err);
619 	if (!dsl_dataset_tryown(*dsp, tag)) {
620 		dsl_dataset_rele(*dsp, tag);
621 		return (SET_ERROR(EBUSY));
622 	}
623 	return (0);
624 }
625 
626 /*
627  * See the comment above dsl_pool_hold() for details.  In summary, a long
628  * hold is used to prevent destruction of a dataset while the pool hold
629  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
630  *
631  * The dataset and pool must be held when this function is called.  After it
632  * is called, the pool hold may be released while the dataset is still held
633  * and accessed.
634  */
635 void
636 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
637 {
638 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
639 	(void) refcount_add(&ds->ds_longholds, tag);
640 }
641 
642 void
643 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
644 {
645 	(void) refcount_remove(&ds->ds_longholds, tag);
646 }
647 
648 /* Return B_TRUE if there are any long holds on this dataset. */
649 boolean_t
650 dsl_dataset_long_held(dsl_dataset_t *ds)
651 {
652 	return (!refcount_is_zero(&ds->ds_longholds));
653 }
654 
655 void
656 dsl_dataset_name(dsl_dataset_t *ds, char *name)
657 {
658 	if (ds == NULL) {
659 		(void) strcpy(name, "mos");
660 	} else {
661 		dsl_dir_name(ds->ds_dir, name);
662 		VERIFY0(dsl_dataset_get_snapname(ds));
663 		if (ds->ds_snapname[0]) {
664 			(void) strcat(name, "@");
665 			/*
666 			 * We use a "recursive" mutex so that we
667 			 * can call dprintf_ds() with ds_lock held.
668 			 */
669 			if (!MUTEX_HELD(&ds->ds_lock)) {
670 				mutex_enter(&ds->ds_lock);
671 				(void) strcat(name, ds->ds_snapname);
672 				mutex_exit(&ds->ds_lock);
673 			} else {
674 				(void) strcat(name, ds->ds_snapname);
675 			}
676 		}
677 	}
678 }
679 
680 void
681 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
682 {
683 	dmu_buf_rele(ds->ds_dbuf, tag);
684 }
685 
686 void
687 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
688 {
689 	ASSERT3P(ds->ds_owner, ==, tag);
690 	ASSERT(ds->ds_dbuf != NULL);
691 
692 	mutex_enter(&ds->ds_lock);
693 	ds->ds_owner = NULL;
694 	mutex_exit(&ds->ds_lock);
695 	dsl_dataset_long_rele(ds, tag);
696 	dsl_dataset_rele(ds, tag);
697 }
698 
699 boolean_t
700 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
701 {
702 	boolean_t gotit = FALSE;
703 
704 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
705 	mutex_enter(&ds->ds_lock);
706 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
707 		ds->ds_owner = tag;
708 		dsl_dataset_long_hold(ds, tag);
709 		gotit = TRUE;
710 	}
711 	mutex_exit(&ds->ds_lock);
712 	return (gotit);
713 }
714 
715 boolean_t
716 dsl_dataset_has_owner(dsl_dataset_t *ds)
717 {
718 	boolean_t rv;
719 	mutex_enter(&ds->ds_lock);
720 	rv = (ds->ds_owner != NULL);
721 	mutex_exit(&ds->ds_lock);
722 	return (rv);
723 }
724 
725 static void
726 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
727 {
728 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
729 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
730 	uint64_t zero = 0;
731 
732 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
733 
734 	spa_feature_incr(spa, f, tx);
735 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
736 
737 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
738 	    sizeof (zero), 1, &zero, tx));
739 }
740 
741 void
742 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
743 {
744 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
745 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
746 
747 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
748 
749 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
750 	spa_feature_decr(spa, f, tx);
751 }
752 
753 uint64_t
754 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
755     uint64_t flags, dmu_tx_t *tx)
756 {
757 	dsl_pool_t *dp = dd->dd_pool;
758 	dmu_buf_t *dbuf;
759 	dsl_dataset_phys_t *dsphys;
760 	uint64_t dsobj;
761 	objset_t *mos = dp->dp_meta_objset;
762 
763 	if (origin == NULL)
764 		origin = dp->dp_origin_snap;
765 
766 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
767 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
768 	ASSERT(dmu_tx_is_syncing(tx));
769 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
770 
771 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
772 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
773 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
774 	dmu_buf_will_dirty(dbuf, tx);
775 	dsphys = dbuf->db_data;
776 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
777 	dsphys->ds_dir_obj = dd->dd_object;
778 	dsphys->ds_flags = flags;
779 	dsphys->ds_fsid_guid = unique_create();
780 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
781 	    sizeof (dsphys->ds_guid));
782 	dsphys->ds_snapnames_zapobj =
783 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
784 	    DMU_OT_NONE, 0, tx);
785 	dsphys->ds_creation_time = gethrestime_sec();
786 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
787 
788 	if (origin == NULL) {
789 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
790 	} else {
791 		dsl_dataset_t *ohds; /* head of the origin snapshot */
792 
793 		dsphys->ds_prev_snap_obj = origin->ds_object;
794 		dsphys->ds_prev_snap_txg =
795 		    dsl_dataset_phys(origin)->ds_creation_txg;
796 		dsphys->ds_referenced_bytes =
797 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
798 		dsphys->ds_compressed_bytes =
799 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
800 		dsphys->ds_uncompressed_bytes =
801 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
802 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
803 
804 		/*
805 		 * Inherit flags that describe the dataset's contents
806 		 * (INCONSISTENT) or properties (Case Insensitive).
807 		 */
808 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
809 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
810 
811 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
812 			if (origin->ds_feature_inuse[f])
813 				dsl_dataset_activate_feature(dsobj, f, tx);
814 		}
815 
816 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
817 		dsl_dataset_phys(origin)->ds_num_children++;
818 
819 		VERIFY0(dsl_dataset_hold_obj(dp,
820 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
821 		    FTAG, &ohds));
822 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
823 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
824 		dsl_dataset_rele(ohds, FTAG);
825 
826 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
827 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
828 				dsl_dataset_phys(origin)->ds_next_clones_obj =
829 				    zap_create(mos,
830 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
831 			}
832 			VERIFY0(zap_add_int(mos,
833 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
834 			    dsobj, tx));
835 		}
836 
837 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
838 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
839 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
840 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
841 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
842 				dsl_dir_phys(origin->ds_dir)->dd_clones =
843 				    zap_create(mos,
844 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
845 			}
846 			VERIFY0(zap_add_int(mos,
847 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
848 			    dsobj, tx));
849 		}
850 	}
851 
852 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
853 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
854 
855 	dmu_buf_rele(dbuf, FTAG);
856 
857 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
858 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
859 
860 	return (dsobj);
861 }
862 
863 static void
864 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
865 {
866 	objset_t *os;
867 
868 	VERIFY0(dmu_objset_from_ds(ds, &os));
869 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
870 	dsl_dataset_dirty(ds, tx);
871 }
872 
873 uint64_t
874 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
875     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
876 {
877 	dsl_pool_t *dp = pdd->dd_pool;
878 	uint64_t dsobj, ddobj;
879 	dsl_dir_t *dd;
880 
881 	ASSERT(dmu_tx_is_syncing(tx));
882 	ASSERT(lastname[0] != '@');
883 
884 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
885 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
886 
887 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
888 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
889 
890 	dsl_deleg_set_create_perms(dd, tx, cr);
891 
892 	/*
893 	 * Since we're creating a new node we know it's a leaf, so we can
894 	 * initialize the counts if the limit feature is active.
895 	 */
896 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
897 		uint64_t cnt = 0;
898 		objset_t *os = dd->dd_pool->dp_meta_objset;
899 
900 		dsl_dir_zapify(dd, tx);
901 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
902 		    sizeof (cnt), 1, &cnt, tx));
903 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
904 		    sizeof (cnt), 1, &cnt, tx));
905 	}
906 
907 	dsl_dir_rele(dd, FTAG);
908 
909 	/*
910 	 * If we are creating a clone, make sure we zero out any stale
911 	 * data from the origin snapshots zil header.
912 	 */
913 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
914 		dsl_dataset_t *ds;
915 
916 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
917 		dsl_dataset_zero_zil(ds, tx);
918 		dsl_dataset_rele(ds, FTAG);
919 	}
920 
921 	return (dsobj);
922 }
923 
924 /*
925  * The unique space in the head dataset can be calculated by subtracting
926  * the space used in the most recent snapshot, that is still being used
927  * in this file system, from the space currently in use.  To figure out
928  * the space in the most recent snapshot still in use, we need to take
929  * the total space used in the snapshot and subtract out the space that
930  * has been freed up since the snapshot was taken.
931  */
932 void
933 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
934 {
935 	uint64_t mrs_used;
936 	uint64_t dlused, dlcomp, dluncomp;
937 
938 	ASSERT(!ds->ds_is_snapshot);
939 
940 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
941 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
942 	else
943 		mrs_used = 0;
944 
945 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
946 
947 	ASSERT3U(dlused, <=, mrs_used);
948 	dsl_dataset_phys(ds)->ds_unique_bytes =
949 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
950 
951 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
952 	    SPA_VERSION_UNIQUE_ACCURATE)
953 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
954 }
955 
956 void
957 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
958     dmu_tx_t *tx)
959 {
960 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
961 	uint64_t count;
962 	int err;
963 
964 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
965 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
966 	    obj, tx);
967 	/*
968 	 * The err should not be ENOENT, but a bug in a previous version
969 	 * of the code could cause upgrade_clones_cb() to not set
970 	 * ds_next_snap_obj when it should, leading to a missing entry.
971 	 * If we knew that the pool was created after
972 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
973 	 * ENOENT.  However, at least we can check that we don't have
974 	 * too many entries in the next_clones_obj even after failing to
975 	 * remove this one.
976 	 */
977 	if (err != ENOENT)
978 		VERIFY0(err);
979 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
980 	    &count));
981 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
982 }
983 
984 
985 blkptr_t *
986 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
987 {
988 	return (&dsl_dataset_phys(ds)->ds_bp);
989 }
990 
991 void
992 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
993 {
994 	ASSERT(dmu_tx_is_syncing(tx));
995 	/* If it's the meta-objset, set dp_meta_rootbp */
996 	if (ds == NULL) {
997 		tx->tx_pool->dp_meta_rootbp = *bp;
998 	} else {
999 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1000 		dsl_dataset_phys(ds)->ds_bp = *bp;
1001 	}
1002 }
1003 
1004 spa_t *
1005 dsl_dataset_get_spa(dsl_dataset_t *ds)
1006 {
1007 	return (ds->ds_dir->dd_pool->dp_spa);
1008 }
1009 
1010 void
1011 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1012 {
1013 	dsl_pool_t *dp;
1014 
1015 	if (ds == NULL) /* this is the meta-objset */
1016 		return;
1017 
1018 	ASSERT(ds->ds_objset != NULL);
1019 
1020 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1021 		panic("dirtying snapshot!");
1022 
1023 	dp = ds->ds_dir->dd_pool;
1024 
1025 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1026 		/* up the hold count until we can be written out */
1027 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1028 	}
1029 }
1030 
1031 boolean_t
1032 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1033 {
1034 	for (int t = 0; t < TXG_SIZE; t++) {
1035 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1036 		    ds, t))
1037 			return (B_TRUE);
1038 	}
1039 	return (B_FALSE);
1040 }
1041 
1042 static int
1043 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1044 {
1045 	uint64_t asize;
1046 
1047 	if (!dmu_tx_is_syncing(tx))
1048 		return (0);
1049 
1050 	/*
1051 	 * If there's an fs-only reservation, any blocks that might become
1052 	 * owned by the snapshot dataset must be accommodated by space
1053 	 * outside of the reservation.
1054 	 */
1055 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1056 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1057 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1058 		return (SET_ERROR(ENOSPC));
1059 
1060 	/*
1061 	 * Propagate any reserved space for this snapshot to other
1062 	 * snapshot checks in this sync group.
1063 	 */
1064 	if (asize > 0)
1065 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1066 
1067 	return (0);
1068 }
1069 
1070 typedef struct dsl_dataset_snapshot_arg {
1071 	nvlist_t *ddsa_snaps;
1072 	nvlist_t *ddsa_props;
1073 	nvlist_t *ddsa_errors;
1074 	cred_t *ddsa_cr;
1075 } dsl_dataset_snapshot_arg_t;
1076 
1077 int
1078 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1079     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1080 {
1081 	int error;
1082 	uint64_t value;
1083 
1084 	ds->ds_trysnap_txg = tx->tx_txg;
1085 
1086 	if (!dmu_tx_is_syncing(tx))
1087 		return (0);
1088 
1089 	/*
1090 	 * We don't allow multiple snapshots of the same txg.  If there
1091 	 * is already one, try again.
1092 	 */
1093 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1094 		return (SET_ERROR(EAGAIN));
1095 
1096 	/*
1097 	 * Check for conflicting snapshot name.
1098 	 */
1099 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1100 	if (error == 0)
1101 		return (SET_ERROR(EEXIST));
1102 	if (error != ENOENT)
1103 		return (error);
1104 
1105 	/*
1106 	 * We don't allow taking snapshots of inconsistent datasets, such as
1107 	 * those into which we are currently receiving.  However, if we are
1108 	 * creating this snapshot as part of a receive, this check will be
1109 	 * executed atomically with respect to the completion of the receive
1110 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1111 	 * case we ignore this, knowing it will be fixed up for us shortly in
1112 	 * dmu_recv_end_sync().
1113 	 */
1114 	if (!recv && DS_IS_INCONSISTENT(ds))
1115 		return (SET_ERROR(EBUSY));
1116 
1117 	/*
1118 	 * Skip the check for temporary snapshots or if we have already checked
1119 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1120 	 * check the count here when we're receiving a stream.
1121 	 */
1122 	if (cnt != 0 && cr != NULL) {
1123 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1124 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1125 		if (error != 0)
1126 			return (error);
1127 	}
1128 
1129 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1130 	if (error != 0)
1131 		return (error);
1132 
1133 	return (0);
1134 }
1135 
1136 static int
1137 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1138 {
1139 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1140 	dsl_pool_t *dp = dmu_tx_pool(tx);
1141 	nvpair_t *pair;
1142 	int rv = 0;
1143 
1144 	/*
1145 	 * Pre-compute how many total new snapshots will be created for each
1146 	 * level in the tree and below. This is needed for validating the
1147 	 * snapshot limit when either taking a recursive snapshot or when
1148 	 * taking multiple snapshots.
1149 	 *
1150 	 * The problem is that the counts are not actually adjusted when
1151 	 * we are checking, only when we finally sync. For a single snapshot,
1152 	 * this is easy, the count will increase by 1 at each node up the tree,
1153 	 * but its more complicated for the recursive/multiple snapshot case.
1154 	 *
1155 	 * The dsl_fs_ss_limit_check function does recursively check the count
1156 	 * at each level up the tree but since it is validating each snapshot
1157 	 * independently we need to be sure that we are validating the complete
1158 	 * count for the entire set of snapshots. We do this by rolling up the
1159 	 * counts for each component of the name into an nvlist and then
1160 	 * checking each of those cases with the aggregated count.
1161 	 *
1162 	 * This approach properly handles not only the recursive snapshot
1163 	 * case (where we get all of those on the ddsa_snaps list) but also
1164 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1165 	 * validate the limit on 'a' using a count of 2).
1166 	 *
1167 	 * We validate the snapshot names in the third loop and only report
1168 	 * name errors once.
1169 	 */
1170 	if (dmu_tx_is_syncing(tx)) {
1171 		nvlist_t *cnt_track = NULL;
1172 		cnt_track = fnvlist_alloc();
1173 
1174 		/* Rollup aggregated counts into the cnt_track list */
1175 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1176 		    pair != NULL;
1177 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1178 			char *pdelim;
1179 			uint64_t val;
1180 			char nm[MAXPATHLEN];
1181 
1182 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1183 			pdelim = strchr(nm, '@');
1184 			if (pdelim == NULL)
1185 				continue;
1186 			*pdelim = '\0';
1187 
1188 			do {
1189 				if (nvlist_lookup_uint64(cnt_track, nm,
1190 				    &val) == 0) {
1191 					/* update existing entry */
1192 					fnvlist_add_uint64(cnt_track, nm,
1193 					    val + 1);
1194 				} else {
1195 					/* add to list */
1196 					fnvlist_add_uint64(cnt_track, nm, 1);
1197 				}
1198 
1199 				pdelim = strrchr(nm, '/');
1200 				if (pdelim != NULL)
1201 					*pdelim = '\0';
1202 			} while (pdelim != NULL);
1203 		}
1204 
1205 		/* Check aggregated counts at each level */
1206 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1207 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1208 			int error = 0;
1209 			char *name;
1210 			uint64_t cnt = 0;
1211 			dsl_dataset_t *ds;
1212 
1213 			name = nvpair_name(pair);
1214 			cnt = fnvpair_value_uint64(pair);
1215 			ASSERT(cnt > 0);
1216 
1217 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1218 			if (error == 0) {
1219 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1220 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1221 				    ddsa->ddsa_cr);
1222 				dsl_dataset_rele(ds, FTAG);
1223 			}
1224 
1225 			if (error != 0) {
1226 				if (ddsa->ddsa_errors != NULL)
1227 					fnvlist_add_int32(ddsa->ddsa_errors,
1228 					    name, error);
1229 				rv = error;
1230 				/* only report one error for this check */
1231 				break;
1232 			}
1233 		}
1234 		nvlist_free(cnt_track);
1235 	}
1236 
1237 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1238 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1239 		int error = 0;
1240 		dsl_dataset_t *ds;
1241 		char *name, *atp;
1242 		char dsname[MAXNAMELEN];
1243 
1244 		name = nvpair_name(pair);
1245 		if (strlen(name) >= MAXNAMELEN)
1246 			error = SET_ERROR(ENAMETOOLONG);
1247 		if (error == 0) {
1248 			atp = strchr(name, '@');
1249 			if (atp == NULL)
1250 				error = SET_ERROR(EINVAL);
1251 			if (error == 0)
1252 				(void) strlcpy(dsname, name, atp - name + 1);
1253 		}
1254 		if (error == 0)
1255 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1256 		if (error == 0) {
1257 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1258 			error = dsl_dataset_snapshot_check_impl(ds,
1259 			    atp + 1, tx, B_FALSE, 0, NULL);
1260 			dsl_dataset_rele(ds, FTAG);
1261 		}
1262 
1263 		if (error != 0) {
1264 			if (ddsa->ddsa_errors != NULL) {
1265 				fnvlist_add_int32(ddsa->ddsa_errors,
1266 				    name, error);
1267 			}
1268 			rv = error;
1269 		}
1270 	}
1271 
1272 	return (rv);
1273 }
1274 
1275 void
1276 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1277     dmu_tx_t *tx)
1278 {
1279 	static zil_header_t zero_zil;
1280 
1281 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1282 	dmu_buf_t *dbuf;
1283 	dsl_dataset_phys_t *dsphys;
1284 	uint64_t dsobj, crtxg;
1285 	objset_t *mos = dp->dp_meta_objset;
1286 	objset_t *os;
1287 
1288 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1289 
1290 	/*
1291 	 * If we are on an old pool, the zil must not be active, in which
1292 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1293 	 */
1294 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1295 	    dmu_objset_from_ds(ds, &os) != 0 ||
1296 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1297 	    sizeof (zero_zil)) == 0);
1298 
1299 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1300 
1301 	/*
1302 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1303 	 */
1304 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1305 		crtxg = 1;
1306 	else
1307 		crtxg = tx->tx_txg;
1308 
1309 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1310 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1311 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1312 	dmu_buf_will_dirty(dbuf, tx);
1313 	dsphys = dbuf->db_data;
1314 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1315 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1316 	dsphys->ds_fsid_guid = unique_create();
1317 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1318 	    sizeof (dsphys->ds_guid));
1319 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1320 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1321 	dsphys->ds_next_snap_obj = ds->ds_object;
1322 	dsphys->ds_num_children = 1;
1323 	dsphys->ds_creation_time = gethrestime_sec();
1324 	dsphys->ds_creation_txg = crtxg;
1325 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1326 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1327 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1328 	dsphys->ds_uncompressed_bytes =
1329 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1330 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1331 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1332 	dmu_buf_rele(dbuf, FTAG);
1333 
1334 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1335 		if (ds->ds_feature_inuse[f])
1336 			dsl_dataset_activate_feature(dsobj, f, tx);
1337 	}
1338 
1339 	ASSERT3U(ds->ds_prev != 0, ==,
1340 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1341 	if (ds->ds_prev) {
1342 		uint64_t next_clones_obj =
1343 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1344 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1345 		    ds->ds_object ||
1346 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1347 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1348 		    ds->ds_object) {
1349 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1350 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1351 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1352 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1353 		} else if (next_clones_obj != 0) {
1354 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1355 			    dsphys->ds_next_snap_obj, tx);
1356 			VERIFY0(zap_add_int(mos,
1357 			    next_clones_obj, dsobj, tx));
1358 		}
1359 	}
1360 
1361 	/*
1362 	 * If we have a reference-reservation on this dataset, we will
1363 	 * need to increase the amount of refreservation being charged
1364 	 * since our unique space is going to zero.
1365 	 */
1366 	if (ds->ds_reserved) {
1367 		int64_t delta;
1368 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1369 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1370 		    ds->ds_reserved);
1371 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1372 		    delta, 0, 0, tx);
1373 	}
1374 
1375 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1376 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1377 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1378 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1379 	dsl_deadlist_close(&ds->ds_deadlist);
1380 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1381 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1382 	dsl_deadlist_add_key(&ds->ds_deadlist,
1383 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1384 
1385 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1386 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1387 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1388 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1389 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1390 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1391 
1392 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1393 	    snapname, 8, 1, &dsobj, tx));
1394 
1395 	if (ds->ds_prev)
1396 		dsl_dataset_rele(ds->ds_prev, ds);
1397 	VERIFY0(dsl_dataset_hold_obj(dp,
1398 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1399 
1400 	dsl_scan_ds_snapshotted(ds, tx);
1401 
1402 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1403 
1404 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1405 }
1406 
1407 static void
1408 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1409 {
1410 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1411 	dsl_pool_t *dp = dmu_tx_pool(tx);
1412 	nvpair_t *pair;
1413 
1414 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1415 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1416 		dsl_dataset_t *ds;
1417 		char *name, *atp;
1418 		char dsname[MAXNAMELEN];
1419 
1420 		name = nvpair_name(pair);
1421 		atp = strchr(name, '@');
1422 		(void) strlcpy(dsname, name, atp - name + 1);
1423 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1424 
1425 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1426 		if (ddsa->ddsa_props != NULL) {
1427 			dsl_props_set_sync_impl(ds->ds_prev,
1428 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1429 		}
1430 		dsl_dataset_rele(ds, FTAG);
1431 	}
1432 }
1433 
1434 /*
1435  * The snapshots must all be in the same pool.
1436  * All-or-nothing: if there are any failures, nothing will be modified.
1437  */
1438 int
1439 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1440 {
1441 	dsl_dataset_snapshot_arg_t ddsa;
1442 	nvpair_t *pair;
1443 	boolean_t needsuspend;
1444 	int error;
1445 	spa_t *spa;
1446 	char *firstname;
1447 	nvlist_t *suspended = NULL;
1448 
1449 	pair = nvlist_next_nvpair(snaps, NULL);
1450 	if (pair == NULL)
1451 		return (0);
1452 	firstname = nvpair_name(pair);
1453 
1454 	error = spa_open(firstname, &spa, FTAG);
1455 	if (error != 0)
1456 		return (error);
1457 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1458 	spa_close(spa, FTAG);
1459 
1460 	if (needsuspend) {
1461 		suspended = fnvlist_alloc();
1462 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1463 		    pair = nvlist_next_nvpair(snaps, pair)) {
1464 			char fsname[MAXNAMELEN];
1465 			char *snapname = nvpair_name(pair);
1466 			char *atp;
1467 			void *cookie;
1468 
1469 			atp = strchr(snapname, '@');
1470 			if (atp == NULL) {
1471 				error = SET_ERROR(EINVAL);
1472 				break;
1473 			}
1474 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1475 
1476 			error = zil_suspend(fsname, &cookie);
1477 			if (error != 0)
1478 				break;
1479 			fnvlist_add_uint64(suspended, fsname,
1480 			    (uintptr_t)cookie);
1481 		}
1482 	}
1483 
1484 	ddsa.ddsa_snaps = snaps;
1485 	ddsa.ddsa_props = props;
1486 	ddsa.ddsa_errors = errors;
1487 	ddsa.ddsa_cr = CRED();
1488 
1489 	if (error == 0) {
1490 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1491 		    dsl_dataset_snapshot_sync, &ddsa,
1492 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1493 	}
1494 
1495 	if (suspended != NULL) {
1496 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1497 		    pair = nvlist_next_nvpair(suspended, pair)) {
1498 			zil_resume((void *)(uintptr_t)
1499 			    fnvpair_value_uint64(pair));
1500 		}
1501 		fnvlist_free(suspended);
1502 	}
1503 
1504 	return (error);
1505 }
1506 
1507 typedef struct dsl_dataset_snapshot_tmp_arg {
1508 	const char *ddsta_fsname;
1509 	const char *ddsta_snapname;
1510 	minor_t ddsta_cleanup_minor;
1511 	const char *ddsta_htag;
1512 } dsl_dataset_snapshot_tmp_arg_t;
1513 
1514 static int
1515 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1516 {
1517 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1518 	dsl_pool_t *dp = dmu_tx_pool(tx);
1519 	dsl_dataset_t *ds;
1520 	int error;
1521 
1522 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1523 	if (error != 0)
1524 		return (error);
1525 
1526 	/* NULL cred means no limit check for tmp snapshot */
1527 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1528 	    tx, B_FALSE, 0, NULL);
1529 	if (error != 0) {
1530 		dsl_dataset_rele(ds, FTAG);
1531 		return (error);
1532 	}
1533 
1534 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1535 		dsl_dataset_rele(ds, FTAG);
1536 		return (SET_ERROR(ENOTSUP));
1537 	}
1538 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1539 	    B_TRUE, tx);
1540 	if (error != 0) {
1541 		dsl_dataset_rele(ds, FTAG);
1542 		return (error);
1543 	}
1544 
1545 	dsl_dataset_rele(ds, FTAG);
1546 	return (0);
1547 }
1548 
1549 static void
1550 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1551 {
1552 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1553 	dsl_pool_t *dp = dmu_tx_pool(tx);
1554 	dsl_dataset_t *ds;
1555 
1556 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1557 
1558 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1559 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1560 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1561 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1562 
1563 	dsl_dataset_rele(ds, FTAG);
1564 }
1565 
1566 int
1567 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1568     minor_t cleanup_minor, const char *htag)
1569 {
1570 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1571 	int error;
1572 	spa_t *spa;
1573 	boolean_t needsuspend;
1574 	void *cookie;
1575 
1576 	ddsta.ddsta_fsname = fsname;
1577 	ddsta.ddsta_snapname = snapname;
1578 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1579 	ddsta.ddsta_htag = htag;
1580 
1581 	error = spa_open(fsname, &spa, FTAG);
1582 	if (error != 0)
1583 		return (error);
1584 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1585 	spa_close(spa, FTAG);
1586 
1587 	if (needsuspend) {
1588 		error = zil_suspend(fsname, &cookie);
1589 		if (error != 0)
1590 			return (error);
1591 	}
1592 
1593 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1594 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1595 
1596 	if (needsuspend)
1597 		zil_resume(cookie);
1598 	return (error);
1599 }
1600 
1601 
1602 void
1603 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1604 {
1605 	ASSERT(dmu_tx_is_syncing(tx));
1606 	ASSERT(ds->ds_objset != NULL);
1607 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1608 
1609 	/*
1610 	 * in case we had to change ds_fsid_guid when we opened it,
1611 	 * sync it out now.
1612 	 */
1613 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1614 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1615 
1616 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1617 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1618 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1619 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1620 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1621 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1622 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1623 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1624 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1625 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1626 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1627 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1628 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1629 	}
1630 
1631 	dmu_objset_sync(ds->ds_objset, zio, tx);
1632 
1633 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1634 		if (ds->ds_feature_activation_needed[f]) {
1635 			if (ds->ds_feature_inuse[f])
1636 				continue;
1637 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1638 			ds->ds_feature_inuse[f] = B_TRUE;
1639 		}
1640 	}
1641 }
1642 
1643 static void
1644 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1645 {
1646 	uint64_t count = 0;
1647 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1648 	zap_cursor_t zc;
1649 	zap_attribute_t za;
1650 	nvlist_t *propval = fnvlist_alloc();
1651 	nvlist_t *val = fnvlist_alloc();
1652 
1653 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1654 
1655 	/*
1656 	 * There may be missing entries in ds_next_clones_obj
1657 	 * due to a bug in a previous version of the code.
1658 	 * Only trust it if it has the right number of entries.
1659 	 */
1660 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1661 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1662 		    &count));
1663 	}
1664 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1665 		goto fail;
1666 	for (zap_cursor_init(&zc, mos,
1667 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1668 	    zap_cursor_retrieve(&zc, &za) == 0;
1669 	    zap_cursor_advance(&zc)) {
1670 		dsl_dataset_t *clone;
1671 		char buf[ZFS_MAXNAMELEN];
1672 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1673 		    za.za_first_integer, FTAG, &clone));
1674 		dsl_dir_name(clone->ds_dir, buf);
1675 		fnvlist_add_boolean(val, buf);
1676 		dsl_dataset_rele(clone, FTAG);
1677 	}
1678 	zap_cursor_fini(&zc);
1679 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1680 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1681 fail:
1682 	nvlist_free(val);
1683 	nvlist_free(propval);
1684 }
1685 
1686 static void
1687 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1688 {
1689 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1690 
1691 	if (dsl_dataset_has_resume_receive_state(ds)) {
1692 		char *str;
1693 		void *packed;
1694 		uint8_t *compressed;
1695 		uint64_t val;
1696 		nvlist_t *token_nv = fnvlist_alloc();
1697 		size_t packed_size, compressed_size;
1698 
1699 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1700 		    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1701 			fnvlist_add_uint64(token_nv, "fromguid", val);
1702 		}
1703 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1704 		    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1705 			fnvlist_add_uint64(token_nv, "object", val);
1706 		}
1707 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1708 		    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1709 			fnvlist_add_uint64(token_nv, "offset", val);
1710 		}
1711 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1712 		    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1713 			fnvlist_add_uint64(token_nv, "bytes", val);
1714 		}
1715 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1716 		    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1717 			fnvlist_add_uint64(token_nv, "toguid", val);
1718 		}
1719 		char buf[256];
1720 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1721 		    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1722 			fnvlist_add_string(token_nv, "toname", buf);
1723 		}
1724 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1725 		    DS_FIELD_RESUME_EMBEDOK) == 0) {
1726 			fnvlist_add_boolean(token_nv, "embedok");
1727 		}
1728 		packed = fnvlist_pack(token_nv, &packed_size);
1729 		fnvlist_free(token_nv);
1730 		compressed = kmem_alloc(packed_size, KM_SLEEP);
1731 
1732 		compressed_size = gzip_compress(packed, compressed,
1733 		    packed_size, packed_size, 6);
1734 
1735 		zio_cksum_t cksum;
1736 		fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1737 
1738 		str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1739 		for (int i = 0; i < compressed_size; i++) {
1740 			(void) sprintf(str + i * 2, "%02x", compressed[i]);
1741 		}
1742 		str[compressed_size * 2] = '\0';
1743 		char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1744 		    ZFS_SEND_RESUME_TOKEN_VERSION,
1745 		    (longlong_t)cksum.zc_word[0],
1746 		    (longlong_t)packed_size, str);
1747 		dsl_prop_nvlist_add_string(nv,
1748 		    ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1749 		kmem_free(packed, packed_size);
1750 		kmem_free(str, compressed_size * 2 + 1);
1751 		kmem_free(compressed, packed_size);
1752 		strfree(propval);
1753 	}
1754 }
1755 
1756 void
1757 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1758 {
1759 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1760 	uint64_t refd, avail, uobjs, aobjs, ratio;
1761 
1762 	ASSERT(dsl_pool_config_held(dp));
1763 
1764 	ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1765 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1766 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1767 
1768 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1769 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1770 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1771 
1772 	if (ds->ds_is_snapshot) {
1773 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1774 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1775 		    dsl_dataset_phys(ds)->ds_unique_bytes);
1776 		get_clones_stat(ds, nv);
1777 	} else {
1778 		if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1779 			char buf[MAXNAMELEN];
1780 			dsl_dataset_name(ds->ds_prev, buf);
1781 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1782 		}
1783 
1784 		dsl_dir_stats(ds->ds_dir, nv);
1785 	}
1786 
1787 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1788 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1789 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1790 
1791 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1792 	    dsl_dataset_phys(ds)->ds_creation_time);
1793 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1794 	    dsl_dataset_phys(ds)->ds_creation_txg);
1795 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1796 	    ds->ds_quota);
1797 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1798 	    ds->ds_reserved);
1799 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1800 	    dsl_dataset_phys(ds)->ds_guid);
1801 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1802 	    dsl_dataset_phys(ds)->ds_unique_bytes);
1803 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1804 	    ds->ds_object);
1805 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1806 	    ds->ds_userrefs);
1807 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1808 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1809 
1810 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1811 		uint64_t written, comp, uncomp;
1812 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1813 		dsl_dataset_t *prev;
1814 
1815 		int err = dsl_dataset_hold_obj(dp,
1816 		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1817 		if (err == 0) {
1818 			err = dsl_dataset_space_written(prev, ds, &written,
1819 			    &comp, &uncomp);
1820 			dsl_dataset_rele(prev, FTAG);
1821 			if (err == 0) {
1822 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1823 				    written);
1824 			}
1825 		}
1826 	}
1827 
1828 	if (!dsl_dataset_is_snapshot(ds)) {
1829 		/*
1830 		 * A failed "newfs" (e.g. full) resumable receive leaves
1831 		 * the stats set on this dataset.  Check here for the prop.
1832 		 */
1833 		get_receive_resume_stats(ds, nv);
1834 
1835 		/*
1836 		 * A failed incremental resumable receive leaves the
1837 		 * stats set on our child named "%recv".  Check the child
1838 		 * for the prop.
1839 		 */
1840 		char recvname[ZFS_MAXNAMELEN];
1841 		dsl_dataset_t *recv_ds;
1842 		dsl_dataset_name(ds, recvname);
1843 		(void) strcat(recvname, "/");
1844 		(void) strcat(recvname, recv_clone_name);
1845 		if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1846 			get_receive_resume_stats(recv_ds, nv);
1847 			dsl_dataset_rele(recv_ds, FTAG);
1848 		}
1849 	}
1850 }
1851 
1852 void
1853 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1854 {
1855 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1856 	ASSERT(dsl_pool_config_held(dp));
1857 
1858 	stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1859 	stat->dds_inconsistent =
1860 	    dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1861 	stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1862 	stat->dds_origin[0] = '\0';
1863 	if (ds->ds_is_snapshot) {
1864 		stat->dds_is_snapshot = B_TRUE;
1865 		stat->dds_num_clones =
1866 		    dsl_dataset_phys(ds)->ds_num_children - 1;
1867 	} else {
1868 		stat->dds_is_snapshot = B_FALSE;
1869 		stat->dds_num_clones = 0;
1870 
1871 		if (dsl_dir_is_clone(ds->ds_dir)) {
1872 			dsl_dataset_t *ods;
1873 
1874 			VERIFY0(dsl_dataset_hold_obj(dp,
1875 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1876 			    FTAG, &ods));
1877 			dsl_dataset_name(ods, stat->dds_origin);
1878 			dsl_dataset_rele(ods, FTAG);
1879 		}
1880 	}
1881 }
1882 
1883 uint64_t
1884 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1885 {
1886 	return (ds->ds_fsid_guid);
1887 }
1888 
1889 void
1890 dsl_dataset_space(dsl_dataset_t *ds,
1891     uint64_t *refdbytesp, uint64_t *availbytesp,
1892     uint64_t *usedobjsp, uint64_t *availobjsp)
1893 {
1894 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1895 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1896 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1897 		*availbytesp +=
1898 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1899 	if (ds->ds_quota != 0) {
1900 		/*
1901 		 * Adjust available bytes according to refquota
1902 		 */
1903 		if (*refdbytesp < ds->ds_quota)
1904 			*availbytesp = MIN(*availbytesp,
1905 			    ds->ds_quota - *refdbytesp);
1906 		else
1907 			*availbytesp = 0;
1908 	}
1909 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1910 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1911 }
1912 
1913 boolean_t
1914 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1915 {
1916 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1917 
1918 	ASSERT(dsl_pool_config_held(dp));
1919 	if (snap == NULL)
1920 		return (B_FALSE);
1921 	if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1922 	    dsl_dataset_phys(snap)->ds_creation_txg) {
1923 		objset_t *os, *os_snap;
1924 		/*
1925 		 * It may be that only the ZIL differs, because it was
1926 		 * reset in the head.  Don't count that as being
1927 		 * modified.
1928 		 */
1929 		if (dmu_objset_from_ds(ds, &os) != 0)
1930 			return (B_TRUE);
1931 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1932 			return (B_TRUE);
1933 		return (bcmp(&os->os_phys->os_meta_dnode,
1934 		    &os_snap->os_phys->os_meta_dnode,
1935 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1936 	}
1937 	return (B_FALSE);
1938 }
1939 
1940 typedef struct dsl_dataset_rename_snapshot_arg {
1941 	const char *ddrsa_fsname;
1942 	const char *ddrsa_oldsnapname;
1943 	const char *ddrsa_newsnapname;
1944 	boolean_t ddrsa_recursive;
1945 	dmu_tx_t *ddrsa_tx;
1946 } dsl_dataset_rename_snapshot_arg_t;
1947 
1948 /* ARGSUSED */
1949 static int
1950 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1951     dsl_dataset_t *hds, void *arg)
1952 {
1953 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1954 	int error;
1955 	uint64_t val;
1956 
1957 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1958 	if (error != 0) {
1959 		/* ignore nonexistent snapshots */
1960 		return (error == ENOENT ? 0 : error);
1961 	}
1962 
1963 	/* new name should not exist */
1964 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1965 	if (error == 0)
1966 		error = SET_ERROR(EEXIST);
1967 	else if (error == ENOENT)
1968 		error = 0;
1969 
1970 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1971 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1972 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1973 		error = SET_ERROR(ENAMETOOLONG);
1974 
1975 	return (error);
1976 }
1977 
1978 static int
1979 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1980 {
1981 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1982 	dsl_pool_t *dp = dmu_tx_pool(tx);
1983 	dsl_dataset_t *hds;
1984 	int error;
1985 
1986 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1987 	if (error != 0)
1988 		return (error);
1989 
1990 	if (ddrsa->ddrsa_recursive) {
1991 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1992 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1993 		    DS_FIND_CHILDREN);
1994 	} else {
1995 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1996 	}
1997 	dsl_dataset_rele(hds, FTAG);
1998 	return (error);
1999 }
2000 
2001 static int
2002 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2003     dsl_dataset_t *hds, void *arg)
2004 {
2005 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2006 	dsl_dataset_t *ds;
2007 	uint64_t val;
2008 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2009 	int error;
2010 
2011 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2012 	ASSERT(error == 0 || error == ENOENT);
2013 	if (error == ENOENT) {
2014 		/* ignore nonexistent snapshots */
2015 		return (0);
2016 	}
2017 
2018 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2019 
2020 	/* log before we change the name */
2021 	spa_history_log_internal_ds(ds, "rename", tx,
2022 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2023 
2024 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2025 	    B_FALSE));
2026 	mutex_enter(&ds->ds_lock);
2027 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2028 	mutex_exit(&ds->ds_lock);
2029 	VERIFY0(zap_add(dp->dp_meta_objset,
2030 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2031 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2032 
2033 	dsl_dataset_rele(ds, FTAG);
2034 	return (0);
2035 }
2036 
2037 static void
2038 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2039 {
2040 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2041 	dsl_pool_t *dp = dmu_tx_pool(tx);
2042 	dsl_dataset_t *hds;
2043 
2044 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2045 	ddrsa->ddrsa_tx = tx;
2046 	if (ddrsa->ddrsa_recursive) {
2047 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2048 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2049 		    DS_FIND_CHILDREN));
2050 	} else {
2051 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2052 	}
2053 	dsl_dataset_rele(hds, FTAG);
2054 }
2055 
2056 int
2057 dsl_dataset_rename_snapshot(const char *fsname,
2058     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2059 {
2060 	dsl_dataset_rename_snapshot_arg_t ddrsa;
2061 
2062 	ddrsa.ddrsa_fsname = fsname;
2063 	ddrsa.ddrsa_oldsnapname = oldsnapname;
2064 	ddrsa.ddrsa_newsnapname = newsnapname;
2065 	ddrsa.ddrsa_recursive = recursive;
2066 
2067 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2068 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2069 	    1, ZFS_SPACE_CHECK_RESERVED));
2070 }
2071 
2072 /*
2073  * If we're doing an ownership handoff, we need to make sure that there is
2074  * only one long hold on the dataset.  We're not allowed to change anything here
2075  * so we don't permanently release the long hold or regular hold here.  We want
2076  * to do this only when syncing to avoid the dataset unexpectedly going away
2077  * when we release the long hold.
2078  */
2079 static int
2080 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2081 {
2082 	boolean_t held;
2083 
2084 	if (!dmu_tx_is_syncing(tx))
2085 		return (0);
2086 
2087 	if (owner != NULL) {
2088 		VERIFY3P(ds->ds_owner, ==, owner);
2089 		dsl_dataset_long_rele(ds, owner);
2090 	}
2091 
2092 	held = dsl_dataset_long_held(ds);
2093 
2094 	if (owner != NULL)
2095 		dsl_dataset_long_hold(ds, owner);
2096 
2097 	if (held)
2098 		return (SET_ERROR(EBUSY));
2099 
2100 	return (0);
2101 }
2102 
2103 typedef struct dsl_dataset_rollback_arg {
2104 	const char *ddra_fsname;
2105 	void *ddra_owner;
2106 	nvlist_t *ddra_result;
2107 } dsl_dataset_rollback_arg_t;
2108 
2109 static int
2110 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2111 {
2112 	dsl_dataset_rollback_arg_t *ddra = arg;
2113 	dsl_pool_t *dp = dmu_tx_pool(tx);
2114 	dsl_dataset_t *ds;
2115 	int64_t unused_refres_delta;
2116 	int error;
2117 
2118 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2119 	if (error != 0)
2120 		return (error);
2121 
2122 	/* must not be a snapshot */
2123 	if (ds->ds_is_snapshot) {
2124 		dsl_dataset_rele(ds, FTAG);
2125 		return (SET_ERROR(EINVAL));
2126 	}
2127 
2128 	/* must have a most recent snapshot */
2129 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2130 		dsl_dataset_rele(ds, FTAG);
2131 		return (SET_ERROR(EINVAL));
2132 	}
2133 
2134 	/* must not have any bookmarks after the most recent snapshot */
2135 	nvlist_t *proprequest = fnvlist_alloc();
2136 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2137 	nvlist_t *bookmarks = fnvlist_alloc();
2138 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2139 	fnvlist_free(proprequest);
2140 	if (error != 0)
2141 		return (error);
2142 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2143 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2144 		nvlist_t *valuenv =
2145 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2146 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2147 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2148 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2149 			fnvlist_free(bookmarks);
2150 			dsl_dataset_rele(ds, FTAG);
2151 			return (SET_ERROR(EEXIST));
2152 		}
2153 	}
2154 	fnvlist_free(bookmarks);
2155 
2156 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2157 	if (error != 0) {
2158 		dsl_dataset_rele(ds, FTAG);
2159 		return (error);
2160 	}
2161 
2162 	/*
2163 	 * Check if the snap we are rolling back to uses more than
2164 	 * the refquota.
2165 	 */
2166 	if (ds->ds_quota != 0 &&
2167 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2168 		dsl_dataset_rele(ds, FTAG);
2169 		return (SET_ERROR(EDQUOT));
2170 	}
2171 
2172 	/*
2173 	 * When we do the clone swap, we will temporarily use more space
2174 	 * due to the refreservation (the head will no longer have any
2175 	 * unique space, so the entire amount of the refreservation will need
2176 	 * to be free).  We will immediately destroy the clone, freeing
2177 	 * this space, but the freeing happens over many txg's.
2178 	 */
2179 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2180 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2181 
2182 	if (unused_refres_delta > 0 &&
2183 	    unused_refres_delta >
2184 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2185 		dsl_dataset_rele(ds, FTAG);
2186 		return (SET_ERROR(ENOSPC));
2187 	}
2188 
2189 	dsl_dataset_rele(ds, FTAG);
2190 	return (0);
2191 }
2192 
2193 static void
2194 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2195 {
2196 	dsl_dataset_rollback_arg_t *ddra = arg;
2197 	dsl_pool_t *dp = dmu_tx_pool(tx);
2198 	dsl_dataset_t *ds, *clone;
2199 	uint64_t cloneobj;
2200 	char namebuf[ZFS_MAXNAMELEN];
2201 
2202 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2203 
2204 	dsl_dataset_name(ds->ds_prev, namebuf);
2205 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2206 
2207 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2208 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2209 
2210 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2211 
2212 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2213 	dsl_dataset_zero_zil(ds, tx);
2214 
2215 	dsl_destroy_head_sync_impl(clone, tx);
2216 
2217 	dsl_dataset_rele(clone, FTAG);
2218 	dsl_dataset_rele(ds, FTAG);
2219 }
2220 
2221 /*
2222  * Rolls back the given filesystem or volume to the most recent snapshot.
2223  * The name of the most recent snapshot will be returned under key "target"
2224  * in the result nvlist.
2225  *
2226  * If owner != NULL:
2227  * - The existing dataset MUST be owned by the specified owner at entry
2228  * - Upon return, dataset will still be held by the same owner, whether we
2229  *   succeed or not.
2230  *
2231  * This mode is required any time the existing filesystem is mounted.  See
2232  * notes above zfs_suspend_fs() for further details.
2233  */
2234 int
2235 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2236 {
2237 	dsl_dataset_rollback_arg_t ddra;
2238 
2239 	ddra.ddra_fsname = fsname;
2240 	ddra.ddra_owner = owner;
2241 	ddra.ddra_result = result;
2242 
2243 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2244 	    dsl_dataset_rollback_sync, &ddra,
2245 	    1, ZFS_SPACE_CHECK_RESERVED));
2246 }
2247 
2248 struct promotenode {
2249 	list_node_t link;
2250 	dsl_dataset_t *ds;
2251 };
2252 
2253 typedef struct dsl_dataset_promote_arg {
2254 	const char *ddpa_clonename;
2255 	dsl_dataset_t *ddpa_clone;
2256 	list_t shared_snaps, origin_snaps, clone_snaps;
2257 	dsl_dataset_t *origin_origin; /* origin of the origin */
2258 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2259 	char *err_ds;
2260 	cred_t *cr;
2261 } dsl_dataset_promote_arg_t;
2262 
2263 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2264 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2265     void *tag);
2266 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2267 
2268 static int
2269 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2270 {
2271 	dsl_dataset_promote_arg_t *ddpa = arg;
2272 	dsl_pool_t *dp = dmu_tx_pool(tx);
2273 	dsl_dataset_t *hds;
2274 	struct promotenode *snap;
2275 	dsl_dataset_t *origin_ds;
2276 	int err;
2277 	uint64_t unused;
2278 	uint64_t ss_mv_cnt;
2279 	size_t max_snap_len;
2280 
2281 	err = promote_hold(ddpa, dp, FTAG);
2282 	if (err != 0)
2283 		return (err);
2284 
2285 	hds = ddpa->ddpa_clone;
2286 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2287 
2288 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2289 		promote_rele(ddpa, FTAG);
2290 		return (SET_ERROR(EXDEV));
2291 	}
2292 
2293 	/*
2294 	 * Compute and check the amount of space to transfer.  Since this is
2295 	 * so expensive, don't do the preliminary check.
2296 	 */
2297 	if (!dmu_tx_is_syncing(tx)) {
2298 		promote_rele(ddpa, FTAG);
2299 		return (0);
2300 	}
2301 
2302 	snap = list_head(&ddpa->shared_snaps);
2303 	origin_ds = snap->ds;
2304 
2305 	/* compute origin's new unique space */
2306 	snap = list_tail(&ddpa->clone_snaps);
2307 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2308 	    origin_ds->ds_object);
2309 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2310 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2311 	    &ddpa->unique, &unused, &unused);
2312 
2313 	/*
2314 	 * Walk the snapshots that we are moving
2315 	 *
2316 	 * Compute space to transfer.  Consider the incremental changes
2317 	 * to used by each snapshot:
2318 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2319 	 * So each snapshot gave birth to:
2320 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2321 	 * So a sequence would look like:
2322 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2323 	 * Which simplifies to:
2324 	 * uN + kN + kN-1 + ... + k1 + k0
2325 	 * Note however, if we stop before we reach the ORIGIN we get:
2326 	 * uN + kN + kN-1 + ... + kM - uM-1
2327 	 */
2328 	ss_mv_cnt = 0;
2329 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2330 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2331 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2332 	for (snap = list_head(&ddpa->shared_snaps); snap;
2333 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2334 		uint64_t val, dlused, dlcomp, dluncomp;
2335 		dsl_dataset_t *ds = snap->ds;
2336 
2337 		ss_mv_cnt++;
2338 
2339 		/*
2340 		 * If there are long holds, we won't be able to evict
2341 		 * the objset.
2342 		 */
2343 		if (dsl_dataset_long_held(ds)) {
2344 			err = SET_ERROR(EBUSY);
2345 			goto out;
2346 		}
2347 
2348 		/* Check that the snapshot name does not conflict */
2349 		VERIFY0(dsl_dataset_get_snapname(ds));
2350 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2351 			err = SET_ERROR(ENAMETOOLONG);
2352 			goto out;
2353 		}
2354 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2355 		if (err == 0) {
2356 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2357 			err = SET_ERROR(EEXIST);
2358 			goto out;
2359 		}
2360 		if (err != ENOENT)
2361 			goto out;
2362 
2363 		/* The very first snapshot does not have a deadlist */
2364 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2365 			continue;
2366 
2367 		dsl_deadlist_space(&ds->ds_deadlist,
2368 		    &dlused, &dlcomp, &dluncomp);
2369 		ddpa->used += dlused;
2370 		ddpa->comp += dlcomp;
2371 		ddpa->uncomp += dluncomp;
2372 	}
2373 
2374 	/*
2375 	 * If we are a clone of a clone then we never reached ORIGIN,
2376 	 * so we need to subtract out the clone origin's used space.
2377 	 */
2378 	if (ddpa->origin_origin) {
2379 		ddpa->used -=
2380 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2381 		ddpa->comp -=
2382 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2383 		ddpa->uncomp -=
2384 		    dsl_dataset_phys(ddpa->origin_origin)->
2385 		    ds_uncompressed_bytes;
2386 	}
2387 
2388 	/* Check that there is enough space and limit headroom here */
2389 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2390 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2391 	if (err != 0)
2392 		goto out;
2393 
2394 	/*
2395 	 * Compute the amounts of space that will be used by snapshots
2396 	 * after the promotion (for both origin and clone).  For each,
2397 	 * it is the amount of space that will be on all of their
2398 	 * deadlists (that was not born before their new origin).
2399 	 */
2400 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2401 		uint64_t space;
2402 
2403 		/*
2404 		 * Note, typically this will not be a clone of a clone,
2405 		 * so dd_origin_txg will be < TXG_INITIAL, so
2406 		 * these snaplist_space() -> dsl_deadlist_space_range()
2407 		 * calls will be fast because they do not have to
2408 		 * iterate over all bps.
2409 		 */
2410 		snap = list_head(&ddpa->origin_snaps);
2411 		err = snaplist_space(&ddpa->shared_snaps,
2412 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2413 		if (err != 0)
2414 			goto out;
2415 
2416 		err = snaplist_space(&ddpa->clone_snaps,
2417 		    snap->ds->ds_dir->dd_origin_txg, &space);
2418 		if (err != 0)
2419 			goto out;
2420 		ddpa->cloneusedsnap += space;
2421 	}
2422 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2423 	    DD_FLAG_USED_BREAKDOWN) {
2424 		err = snaplist_space(&ddpa->origin_snaps,
2425 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2426 		    &ddpa->originusedsnap);
2427 		if (err != 0)
2428 			goto out;
2429 	}
2430 
2431 out:
2432 	promote_rele(ddpa, FTAG);
2433 	return (err);
2434 }
2435 
2436 static void
2437 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2438 {
2439 	dsl_dataset_promote_arg_t *ddpa = arg;
2440 	dsl_pool_t *dp = dmu_tx_pool(tx);
2441 	dsl_dataset_t *hds;
2442 	struct promotenode *snap;
2443 	dsl_dataset_t *origin_ds;
2444 	dsl_dataset_t *origin_head;
2445 	dsl_dir_t *dd;
2446 	dsl_dir_t *odd = NULL;
2447 	uint64_t oldnext_obj;
2448 	int64_t delta;
2449 
2450 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2451 	hds = ddpa->ddpa_clone;
2452 
2453 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2454 
2455 	snap = list_head(&ddpa->shared_snaps);
2456 	origin_ds = snap->ds;
2457 	dd = hds->ds_dir;
2458 
2459 	snap = list_head(&ddpa->origin_snaps);
2460 	origin_head = snap->ds;
2461 
2462 	/*
2463 	 * We need to explicitly open odd, since origin_ds's dd will be
2464 	 * changing.
2465 	 */
2466 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2467 	    NULL, FTAG, &odd));
2468 
2469 	/* change origin's next snap */
2470 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2471 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2472 	snap = list_tail(&ddpa->clone_snaps);
2473 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2474 	    origin_ds->ds_object);
2475 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2476 
2477 	/* change the origin's next clone */
2478 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2479 		dsl_dataset_remove_from_next_clones(origin_ds,
2480 		    snap->ds->ds_object, tx);
2481 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2482 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2483 		    oldnext_obj, tx));
2484 	}
2485 
2486 	/* change origin */
2487 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2488 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2489 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2490 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2491 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2492 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2493 	origin_head->ds_dir->dd_origin_txg =
2494 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2495 
2496 	/* change dd_clone entries */
2497 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2498 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2499 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2500 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2501 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2502 		    hds->ds_object, tx));
2503 
2504 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2505 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2506 		    origin_head->ds_object, tx));
2507 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2508 			dsl_dir_phys(dd)->dd_clones =
2509 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2510 			    DMU_OT_NONE, 0, tx);
2511 		}
2512 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2513 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2514 	}
2515 
2516 	/* move snapshots to this dir */
2517 	for (snap = list_head(&ddpa->shared_snaps); snap;
2518 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2519 		dsl_dataset_t *ds = snap->ds;
2520 
2521 		/*
2522 		 * Property callbacks are registered to a particular
2523 		 * dsl_dir.  Since ours is changing, evict the objset
2524 		 * so that they will be unregistered from the old dsl_dir.
2525 		 */
2526 		if (ds->ds_objset) {
2527 			dmu_objset_evict(ds->ds_objset);
2528 			ds->ds_objset = NULL;
2529 		}
2530 
2531 		/* move snap name entry */
2532 		VERIFY0(dsl_dataset_get_snapname(ds));
2533 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2534 		    ds->ds_snapname, tx, B_TRUE));
2535 		VERIFY0(zap_add(dp->dp_meta_objset,
2536 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2537 		    8, 1, &ds->ds_object, tx));
2538 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2539 		    DD_FIELD_SNAPSHOT_COUNT, tx);
2540 
2541 		/* change containing dsl_dir */
2542 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2543 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2544 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2545 		ASSERT3P(ds->ds_dir, ==, odd);
2546 		dsl_dir_rele(ds->ds_dir, ds);
2547 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2548 		    NULL, ds, &ds->ds_dir));
2549 
2550 		/* move any clone references */
2551 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2552 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2553 			zap_cursor_t zc;
2554 			zap_attribute_t za;
2555 
2556 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2557 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
2558 			    zap_cursor_retrieve(&zc, &za) == 0;
2559 			    zap_cursor_advance(&zc)) {
2560 				dsl_dataset_t *cnds;
2561 				uint64_t o;
2562 
2563 				if (za.za_first_integer == oldnext_obj) {
2564 					/*
2565 					 * We've already moved the
2566 					 * origin's reference.
2567 					 */
2568 					continue;
2569 				}
2570 
2571 				VERIFY0(dsl_dataset_hold_obj(dp,
2572 				    za.za_first_integer, FTAG, &cnds));
2573 				o = dsl_dir_phys(cnds->ds_dir)->
2574 				    dd_head_dataset_obj;
2575 
2576 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2577 				    dsl_dir_phys(odd)->dd_clones, o, tx));
2578 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2579 				    dsl_dir_phys(dd)->dd_clones, o, tx));
2580 				dsl_dataset_rele(cnds, FTAG);
2581 			}
2582 			zap_cursor_fini(&zc);
2583 		}
2584 
2585 		ASSERT(!dsl_prop_hascb(ds));
2586 	}
2587 
2588 	/*
2589 	 * Change space accounting.
2590 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2591 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2592 	 * is true for each of {clone,origin} independently.
2593 	 */
2594 
2595 	delta = ddpa->cloneusedsnap -
2596 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2597 	ASSERT3S(delta, >=, 0);
2598 	ASSERT3U(ddpa->used, >=, delta);
2599 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2600 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2601 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2602 
2603 	delta = ddpa->originusedsnap -
2604 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2605 	ASSERT3S(delta, <=, 0);
2606 	ASSERT3U(ddpa->used, >=, -delta);
2607 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2608 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2609 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2610 
2611 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2612 
2613 	/* log history record */
2614 	spa_history_log_internal_ds(hds, "promote", tx, "");
2615 
2616 	dsl_dir_rele(odd, FTAG);
2617 	promote_rele(ddpa, FTAG);
2618 }
2619 
2620 /*
2621  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2622  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2623  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2624  * snapshots back to this dataset's origin.
2625  */
2626 static int
2627 snaplist_make(dsl_pool_t *dp,
2628     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2629 {
2630 	uint64_t obj = last_obj;
2631 
2632 	list_create(l, sizeof (struct promotenode),
2633 	    offsetof(struct promotenode, link));
2634 
2635 	while (obj != first_obj) {
2636 		dsl_dataset_t *ds;
2637 		struct promotenode *snap;
2638 		int err;
2639 
2640 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2641 		ASSERT(err != ENOENT);
2642 		if (err != 0)
2643 			return (err);
2644 
2645 		if (first_obj == 0)
2646 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2647 
2648 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2649 		snap->ds = ds;
2650 		list_insert_tail(l, snap);
2651 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2652 	}
2653 
2654 	return (0);
2655 }
2656 
2657 static int
2658 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2659 {
2660 	struct promotenode *snap;
2661 
2662 	*spacep = 0;
2663 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2664 		uint64_t used, comp, uncomp;
2665 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2666 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2667 		*spacep += used;
2668 	}
2669 	return (0);
2670 }
2671 
2672 static void
2673 snaplist_destroy(list_t *l, void *tag)
2674 {
2675 	struct promotenode *snap;
2676 
2677 	if (l == NULL || !list_link_active(&l->list_head))
2678 		return;
2679 
2680 	while ((snap = list_tail(l)) != NULL) {
2681 		list_remove(l, snap);
2682 		dsl_dataset_rele(snap->ds, tag);
2683 		kmem_free(snap, sizeof (*snap));
2684 	}
2685 	list_destroy(l);
2686 }
2687 
2688 static int
2689 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2690 {
2691 	int error;
2692 	dsl_dir_t *dd;
2693 	struct promotenode *snap;
2694 
2695 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2696 	    &ddpa->ddpa_clone);
2697 	if (error != 0)
2698 		return (error);
2699 	dd = ddpa->ddpa_clone->ds_dir;
2700 
2701 	if (ddpa->ddpa_clone->ds_is_snapshot ||
2702 	    !dsl_dir_is_clone(dd)) {
2703 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2704 		return (SET_ERROR(EINVAL));
2705 	}
2706 
2707 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2708 	    &ddpa->shared_snaps, tag);
2709 	if (error != 0)
2710 		goto out;
2711 
2712 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2713 	    &ddpa->clone_snaps, tag);
2714 	if (error != 0)
2715 		goto out;
2716 
2717 	snap = list_head(&ddpa->shared_snaps);
2718 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2719 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2720 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2721 	    &ddpa->origin_snaps, tag);
2722 	if (error != 0)
2723 		goto out;
2724 
2725 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2726 		error = dsl_dataset_hold_obj(dp,
2727 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2728 		    tag, &ddpa->origin_origin);
2729 		if (error != 0)
2730 			goto out;
2731 	}
2732 out:
2733 	if (error != 0)
2734 		promote_rele(ddpa, tag);
2735 	return (error);
2736 }
2737 
2738 static void
2739 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2740 {
2741 	snaplist_destroy(&ddpa->shared_snaps, tag);
2742 	snaplist_destroy(&ddpa->clone_snaps, tag);
2743 	snaplist_destroy(&ddpa->origin_snaps, tag);
2744 	if (ddpa->origin_origin != NULL)
2745 		dsl_dataset_rele(ddpa->origin_origin, tag);
2746 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2747 }
2748 
2749 /*
2750  * Promote a clone.
2751  *
2752  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2753  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2754  */
2755 int
2756 dsl_dataset_promote(const char *name, char *conflsnap)
2757 {
2758 	dsl_dataset_promote_arg_t ddpa = { 0 };
2759 	uint64_t numsnaps;
2760 	int error;
2761 	objset_t *os;
2762 
2763 	/*
2764 	 * We will modify space proportional to the number of
2765 	 * snapshots.  Compute numsnaps.
2766 	 */
2767 	error = dmu_objset_hold(name, FTAG, &os);
2768 	if (error != 0)
2769 		return (error);
2770 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2771 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2772 	    &numsnaps);
2773 	dmu_objset_rele(os, FTAG);
2774 	if (error != 0)
2775 		return (error);
2776 
2777 	ddpa.ddpa_clonename = name;
2778 	ddpa.err_ds = conflsnap;
2779 	ddpa.cr = CRED();
2780 
2781 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2782 	    dsl_dataset_promote_sync, &ddpa,
2783 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2784 }
2785 
2786 int
2787 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2788     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2789 {
2790 	int64_t unused_refres_delta;
2791 
2792 	/* they should both be heads */
2793 	if (clone->ds_is_snapshot ||
2794 	    origin_head->ds_is_snapshot)
2795 		return (SET_ERROR(EINVAL));
2796 
2797 	/* if we are not forcing, the branch point should be just before them */
2798 	if (!force && clone->ds_prev != origin_head->ds_prev)
2799 		return (SET_ERROR(EINVAL));
2800 
2801 	/* clone should be the clone (unless they are unrelated) */
2802 	if (clone->ds_prev != NULL &&
2803 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2804 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2805 		return (SET_ERROR(EINVAL));
2806 
2807 	/* the clone should be a child of the origin */
2808 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2809 		return (SET_ERROR(EINVAL));
2810 
2811 	/* origin_head shouldn't be modified unless 'force' */
2812 	if (!force &&
2813 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2814 		return (SET_ERROR(ETXTBSY));
2815 
2816 	/* origin_head should have no long holds (e.g. is not mounted) */
2817 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2818 		return (SET_ERROR(EBUSY));
2819 
2820 	/* check amount of any unconsumed refreservation */
2821 	unused_refres_delta =
2822 	    (int64_t)MIN(origin_head->ds_reserved,
2823 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2824 	    (int64_t)MIN(origin_head->ds_reserved,
2825 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2826 
2827 	if (unused_refres_delta > 0 &&
2828 	    unused_refres_delta >
2829 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2830 		return (SET_ERROR(ENOSPC));
2831 
2832 	/* clone can't be over the head's refquota */
2833 	if (origin_head->ds_quota != 0 &&
2834 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
2835 	    origin_head->ds_quota)
2836 		return (SET_ERROR(EDQUOT));
2837 
2838 	return (0);
2839 }
2840 
2841 void
2842 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2843     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2844 {
2845 	dsl_pool_t *dp = dmu_tx_pool(tx);
2846 	int64_t unused_refres_delta;
2847 
2848 	ASSERT(clone->ds_reserved == 0);
2849 	ASSERT(origin_head->ds_quota == 0 ||
2850 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2851 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2852 
2853 	/*
2854 	 * Swap per-dataset feature flags.
2855 	 */
2856 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2857 		if (!(spa_feature_table[f].fi_flags &
2858 		    ZFEATURE_FLAG_PER_DATASET)) {
2859 			ASSERT(!clone->ds_feature_inuse[f]);
2860 			ASSERT(!origin_head->ds_feature_inuse[f]);
2861 			continue;
2862 		}
2863 
2864 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
2865 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2866 
2867 		if (clone_inuse) {
2868 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2869 			clone->ds_feature_inuse[f] = B_FALSE;
2870 		}
2871 		if (origin_head_inuse) {
2872 			dsl_dataset_deactivate_feature(origin_head->ds_object,
2873 			    f, tx);
2874 			origin_head->ds_feature_inuse[f] = B_FALSE;
2875 		}
2876 		if (clone_inuse) {
2877 			dsl_dataset_activate_feature(origin_head->ds_object,
2878 			    f, tx);
2879 			origin_head->ds_feature_inuse[f] = B_TRUE;
2880 		}
2881 		if (origin_head_inuse) {
2882 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
2883 			clone->ds_feature_inuse[f] = B_TRUE;
2884 		}
2885 	}
2886 
2887 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2888 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2889 
2890 	if (clone->ds_objset != NULL) {
2891 		dmu_objset_evict(clone->ds_objset);
2892 		clone->ds_objset = NULL;
2893 	}
2894 
2895 	if (origin_head->ds_objset != NULL) {
2896 		dmu_objset_evict(origin_head->ds_objset);
2897 		origin_head->ds_objset = NULL;
2898 	}
2899 
2900 	unused_refres_delta =
2901 	    (int64_t)MIN(origin_head->ds_reserved,
2902 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2903 	    (int64_t)MIN(origin_head->ds_reserved,
2904 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2905 
2906 	/*
2907 	 * Reset origin's unique bytes, if it exists.
2908 	 */
2909 	if (clone->ds_prev) {
2910 		dsl_dataset_t *origin = clone->ds_prev;
2911 		uint64_t comp, uncomp;
2912 
2913 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2914 		dsl_deadlist_space_range(&clone->ds_deadlist,
2915 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2916 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2917 	}
2918 
2919 	/* swap blkptrs */
2920 	{
2921 		blkptr_t tmp;
2922 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
2923 		dsl_dataset_phys(origin_head)->ds_bp =
2924 		    dsl_dataset_phys(clone)->ds_bp;
2925 		dsl_dataset_phys(clone)->ds_bp = tmp;
2926 	}
2927 
2928 	/* set dd_*_bytes */
2929 	{
2930 		int64_t dused, dcomp, duncomp;
2931 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2932 		uint64_t odl_used, odl_comp, odl_uncomp;
2933 
2934 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2935 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2936 
2937 		dsl_deadlist_space(&clone->ds_deadlist,
2938 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2939 		dsl_deadlist_space(&origin_head->ds_deadlist,
2940 		    &odl_used, &odl_comp, &odl_uncomp);
2941 
2942 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2943 		    cdl_used -
2944 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2945 		    odl_used);
2946 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2947 		    cdl_comp -
2948 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2949 		    odl_comp);
2950 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2951 		    cdl_uncomp -
2952 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2953 		    odl_uncomp);
2954 
2955 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2956 		    dused, dcomp, duncomp, tx);
2957 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2958 		    -dused, -dcomp, -duncomp, tx);
2959 
2960 		/*
2961 		 * The difference in the space used by snapshots is the
2962 		 * difference in snapshot space due to the head's
2963 		 * deadlist (since that's the only thing that's
2964 		 * changing that affects the snapused).
2965 		 */
2966 		dsl_deadlist_space_range(&clone->ds_deadlist,
2967 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2968 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2969 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2970 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2971 		    &odl_used, &odl_comp, &odl_uncomp);
2972 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2973 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2974 	}
2975 
2976 	/* swap ds_*_bytes */
2977 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2978 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
2979 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2980 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
2981 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2982 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2983 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2984 	    dsl_dataset_phys(clone)->ds_unique_bytes);
2985 
2986 	/* apply any parent delta for change in unconsumed refreservation */
2987 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2988 	    unused_refres_delta, 0, 0, tx);
2989 
2990 	/*
2991 	 * Swap deadlists.
2992 	 */
2993 	dsl_deadlist_close(&clone->ds_deadlist);
2994 	dsl_deadlist_close(&origin_head->ds_deadlist);
2995 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2996 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2997 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2998 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
2999 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3000 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3001 
3002 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3003 
3004 	spa_history_log_internal_ds(clone, "clone swap", tx,
3005 	    "parent=%s", origin_head->ds_dir->dd_myname);
3006 }
3007 
3008 /*
3009  * Given a pool name and a dataset object number in that pool,
3010  * return the name of that dataset.
3011  */
3012 int
3013 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3014 {
3015 	dsl_pool_t *dp;
3016 	dsl_dataset_t *ds;
3017 	int error;
3018 
3019 	error = dsl_pool_hold(pname, FTAG, &dp);
3020 	if (error != 0)
3021 		return (error);
3022 
3023 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3024 	if (error == 0) {
3025 		dsl_dataset_name(ds, buf);
3026 		dsl_dataset_rele(ds, FTAG);
3027 	}
3028 	dsl_pool_rele(dp, FTAG);
3029 
3030 	return (error);
3031 }
3032 
3033 int
3034 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3035     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3036 {
3037 	int error = 0;
3038 
3039 	ASSERT3S(asize, >, 0);
3040 
3041 	/*
3042 	 * *ref_rsrv is the portion of asize that will come from any
3043 	 * unconsumed refreservation space.
3044 	 */
3045 	*ref_rsrv = 0;
3046 
3047 	mutex_enter(&ds->ds_lock);
3048 	/*
3049 	 * Make a space adjustment for reserved bytes.
3050 	 */
3051 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3052 		ASSERT3U(*used, >=,
3053 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3054 		*used -=
3055 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3056 		*ref_rsrv =
3057 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3058 	}
3059 
3060 	if (!check_quota || ds->ds_quota == 0) {
3061 		mutex_exit(&ds->ds_lock);
3062 		return (0);
3063 	}
3064 	/*
3065 	 * If they are requesting more space, and our current estimate
3066 	 * is over quota, they get to try again unless the actual
3067 	 * on-disk is over quota and there are no pending changes (which
3068 	 * may free up space for us).
3069 	 */
3070 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3071 	    ds->ds_quota) {
3072 		if (inflight > 0 ||
3073 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3074 			error = SET_ERROR(ERESTART);
3075 		else
3076 			error = SET_ERROR(EDQUOT);
3077 	}
3078 	mutex_exit(&ds->ds_lock);
3079 
3080 	return (error);
3081 }
3082 
3083 typedef struct dsl_dataset_set_qr_arg {
3084 	const char *ddsqra_name;
3085 	zprop_source_t ddsqra_source;
3086 	uint64_t ddsqra_value;
3087 } dsl_dataset_set_qr_arg_t;
3088 
3089 
3090 /* ARGSUSED */
3091 static int
3092 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3093 {
3094 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3095 	dsl_pool_t *dp = dmu_tx_pool(tx);
3096 	dsl_dataset_t *ds;
3097 	int error;
3098 	uint64_t newval;
3099 
3100 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3101 		return (SET_ERROR(ENOTSUP));
3102 
3103 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3104 	if (error != 0)
3105 		return (error);
3106 
3107 	if (ds->ds_is_snapshot) {
3108 		dsl_dataset_rele(ds, FTAG);
3109 		return (SET_ERROR(EINVAL));
3110 	}
3111 
3112 	error = dsl_prop_predict(ds->ds_dir,
3113 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3114 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3115 	if (error != 0) {
3116 		dsl_dataset_rele(ds, FTAG);
3117 		return (error);
3118 	}
3119 
3120 	if (newval == 0) {
3121 		dsl_dataset_rele(ds, FTAG);
3122 		return (0);
3123 	}
3124 
3125 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3126 	    newval < ds->ds_reserved) {
3127 		dsl_dataset_rele(ds, FTAG);
3128 		return (SET_ERROR(ENOSPC));
3129 	}
3130 
3131 	dsl_dataset_rele(ds, FTAG);
3132 	return (0);
3133 }
3134 
3135 static void
3136 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3137 {
3138 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3139 	dsl_pool_t *dp = dmu_tx_pool(tx);
3140 	dsl_dataset_t *ds;
3141 	uint64_t newval;
3142 
3143 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3144 
3145 	dsl_prop_set_sync_impl(ds,
3146 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3147 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3148 	    &ddsqra->ddsqra_value, tx);
3149 
3150 	VERIFY0(dsl_prop_get_int_ds(ds,
3151 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3152 
3153 	if (ds->ds_quota != newval) {
3154 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3155 		ds->ds_quota = newval;
3156 	}
3157 	dsl_dataset_rele(ds, FTAG);
3158 }
3159 
3160 int
3161 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3162     uint64_t refquota)
3163 {
3164 	dsl_dataset_set_qr_arg_t ddsqra;
3165 
3166 	ddsqra.ddsqra_name = dsname;
3167 	ddsqra.ddsqra_source = source;
3168 	ddsqra.ddsqra_value = refquota;
3169 
3170 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3171 	    dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3172 }
3173 
3174 static int
3175 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3176 {
3177 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3178 	dsl_pool_t *dp = dmu_tx_pool(tx);
3179 	dsl_dataset_t *ds;
3180 	int error;
3181 	uint64_t newval, unique;
3182 
3183 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3184 		return (SET_ERROR(ENOTSUP));
3185 
3186 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3187 	if (error != 0)
3188 		return (error);
3189 
3190 	if (ds->ds_is_snapshot) {
3191 		dsl_dataset_rele(ds, FTAG);
3192 		return (SET_ERROR(EINVAL));
3193 	}
3194 
3195 	error = dsl_prop_predict(ds->ds_dir,
3196 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3197 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3198 	if (error != 0) {
3199 		dsl_dataset_rele(ds, FTAG);
3200 		return (error);
3201 	}
3202 
3203 	/*
3204 	 * If we are doing the preliminary check in open context, the
3205 	 * space estimates may be inaccurate.
3206 	 */
3207 	if (!dmu_tx_is_syncing(tx)) {
3208 		dsl_dataset_rele(ds, FTAG);
3209 		return (0);
3210 	}
3211 
3212 	mutex_enter(&ds->ds_lock);
3213 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3214 		dsl_dataset_recalc_head_uniq(ds);
3215 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3216 	mutex_exit(&ds->ds_lock);
3217 
3218 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3219 		uint64_t delta = MAX(unique, newval) -
3220 		    MAX(unique, ds->ds_reserved);
3221 
3222 		if (delta >
3223 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3224 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3225 			dsl_dataset_rele(ds, FTAG);
3226 			return (SET_ERROR(ENOSPC));
3227 		}
3228 	}
3229 
3230 	dsl_dataset_rele(ds, FTAG);
3231 	return (0);
3232 }
3233 
3234 void
3235 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3236     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3237 {
3238 	uint64_t newval;
3239 	uint64_t unique;
3240 	int64_t delta;
3241 
3242 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3243 	    source, sizeof (value), 1, &value, tx);
3244 
3245 	VERIFY0(dsl_prop_get_int_ds(ds,
3246 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3247 
3248 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3249 	mutex_enter(&ds->ds_dir->dd_lock);
3250 	mutex_enter(&ds->ds_lock);
3251 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3252 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3253 	delta = MAX(0, (int64_t)(newval - unique)) -
3254 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3255 	ds->ds_reserved = newval;
3256 	mutex_exit(&ds->ds_lock);
3257 
3258 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3259 	mutex_exit(&ds->ds_dir->dd_lock);
3260 }
3261 
3262 static void
3263 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3264 {
3265 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3266 	dsl_pool_t *dp = dmu_tx_pool(tx);
3267 	dsl_dataset_t *ds;
3268 
3269 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3270 	dsl_dataset_set_refreservation_sync_impl(ds,
3271 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3272 	dsl_dataset_rele(ds, FTAG);
3273 }
3274 
3275 int
3276 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3277     uint64_t refreservation)
3278 {
3279 	dsl_dataset_set_qr_arg_t ddsqra;
3280 
3281 	ddsqra.ddsqra_name = dsname;
3282 	ddsqra.ddsqra_source = source;
3283 	ddsqra.ddsqra_value = refreservation;
3284 
3285 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3286 	    dsl_dataset_set_refreservation_sync, &ddsqra,
3287 	    0, ZFS_SPACE_CHECK_NONE));
3288 }
3289 
3290 /*
3291  * Return (in *usedp) the amount of space written in new that is not
3292  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3293  * a snapshot before new, in new's filesystem (or its origin).  If not then
3294  * fail and return EINVAL.
3295  *
3296  * The written space is calculated by considering two components:  First, we
3297  * ignore any freed space, and calculate the written as new's used space
3298  * minus old's used space.  Next, we add in the amount of space that was freed
3299  * between the two snapshots, thus reducing new's used space relative to old's.
3300  * Specifically, this is the space that was born before old->ds_creation_txg,
3301  * and freed before new (ie. on new's deadlist or a previous deadlist).
3302  *
3303  * space freed                         [---------------------]
3304  * snapshots                       ---O-------O--------O-------O------
3305  *                                         oldsnap            new
3306  */
3307 int
3308 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3309     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3310 {
3311 	int err = 0;
3312 	uint64_t snapobj;
3313 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3314 
3315 	ASSERT(dsl_pool_config_held(dp));
3316 
3317 	*usedp = 0;
3318 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3319 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3320 
3321 	*compp = 0;
3322 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3323 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3324 
3325 	*uncompp = 0;
3326 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3327 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3328 
3329 	snapobj = new->ds_object;
3330 	while (snapobj != oldsnap->ds_object) {
3331 		dsl_dataset_t *snap;
3332 		uint64_t used, comp, uncomp;
3333 
3334 		if (snapobj == new->ds_object) {
3335 			snap = new;
3336 		} else {
3337 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3338 			if (err != 0)
3339 				break;
3340 		}
3341 
3342 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3343 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3344 			/*
3345 			 * The blocks in the deadlist can not be born after
3346 			 * ds_prev_snap_txg, so get the whole deadlist space,
3347 			 * which is more efficient (especially for old-format
3348 			 * deadlists).  Unfortunately the deadlist code
3349 			 * doesn't have enough information to make this
3350 			 * optimization itself.
3351 			 */
3352 			dsl_deadlist_space(&snap->ds_deadlist,
3353 			    &used, &comp, &uncomp);
3354 		} else {
3355 			dsl_deadlist_space_range(&snap->ds_deadlist,
3356 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3357 			    &used, &comp, &uncomp);
3358 		}
3359 		*usedp += used;
3360 		*compp += comp;
3361 		*uncompp += uncomp;
3362 
3363 		/*
3364 		 * If we get to the beginning of the chain of snapshots
3365 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3366 		 * was not a snapshot of/before new.
3367 		 */
3368 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3369 		if (snap != new)
3370 			dsl_dataset_rele(snap, FTAG);
3371 		if (snapobj == 0) {
3372 			err = SET_ERROR(EINVAL);
3373 			break;
3374 		}
3375 
3376 	}
3377 	return (err);
3378 }
3379 
3380 /*
3381  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3382  * lastsnap, and all snapshots in between are deleted.
3383  *
3384  * blocks that would be freed            [---------------------------]
3385  * snapshots                       ---O-------O--------O-------O--------O
3386  *                                        firstsnap        lastsnap
3387  *
3388  * This is the set of blocks that were born after the snap before firstsnap,
3389  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3390  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3391  * We calculate this by iterating over the relevant deadlists (from the snap
3392  * after lastsnap, backward to the snap after firstsnap), summing up the
3393  * space on the deadlist that was born after the snap before firstsnap.
3394  */
3395 int
3396 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3397     dsl_dataset_t *lastsnap,
3398     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3399 {
3400 	int err = 0;
3401 	uint64_t snapobj;
3402 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3403 
3404 	ASSERT(firstsnap->ds_is_snapshot);
3405 	ASSERT(lastsnap->ds_is_snapshot);
3406 
3407 	/*
3408 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3409 	 * is before lastsnap.
3410 	 */
3411 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3412 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3413 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3414 		return (SET_ERROR(EINVAL));
3415 
3416 	*usedp = *compp = *uncompp = 0;
3417 
3418 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3419 	while (snapobj != firstsnap->ds_object) {
3420 		dsl_dataset_t *ds;
3421 		uint64_t used, comp, uncomp;
3422 
3423 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3424 		if (err != 0)
3425 			break;
3426 
3427 		dsl_deadlist_space_range(&ds->ds_deadlist,
3428 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3429 		    &used, &comp, &uncomp);
3430 		*usedp += used;
3431 		*compp += comp;
3432 		*uncompp += uncomp;
3433 
3434 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3435 		ASSERT3U(snapobj, !=, 0);
3436 		dsl_dataset_rele(ds, FTAG);
3437 	}
3438 	return (err);
3439 }
3440 
3441 /*
3442  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3443  * For example, they could both be snapshots of the same filesystem, and
3444  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3445  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
3446  * filesystem.  Or 'earlier' could be the origin's origin.
3447  *
3448  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3449  */
3450 boolean_t
3451 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3452     uint64_t earlier_txg)
3453 {
3454 	dsl_pool_t *dp = later->ds_dir->dd_pool;
3455 	int error;
3456 	boolean_t ret;
3457 
3458 	ASSERT(dsl_pool_config_held(dp));
3459 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3460 
3461 	if (earlier_txg == 0)
3462 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3463 
3464 	if (later->ds_is_snapshot &&
3465 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3466 		return (B_FALSE);
3467 
3468 	if (later->ds_dir == earlier->ds_dir)
3469 		return (B_TRUE);
3470 	if (!dsl_dir_is_clone(later->ds_dir))
3471 		return (B_FALSE);
3472 
3473 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3474 		return (B_TRUE);
3475 	dsl_dataset_t *origin;
3476 	error = dsl_dataset_hold_obj(dp,
3477 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3478 	if (error != 0)
3479 		return (B_FALSE);
3480 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3481 	dsl_dataset_rele(origin, FTAG);
3482 	return (ret);
3483 }
3484 
3485 void
3486 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3487 {
3488 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3489 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3490 }
3491 
3492 boolean_t
3493 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3494 {
3495 	dmu_object_info_t doi;
3496 
3497 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
3498 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3499 }
3500 
3501 boolean_t
3502 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3503 {
3504 	return (dsl_dataset_is_zapified(ds) &&
3505 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3506 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3507 }
3508