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 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25  * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26  * Copyright (c) 2014 RackTop Systems.
27  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28  * Copyright (c) 2014 Integros [integros.com]
29  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
30  * Copyright 2017 Nexenta Systems, Inc.
31  */
32 
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dmu_traverse.h>
39 #include <sys/dmu_impl.h>
40 #include <sys/dmu_tx.h>
41 #include <sys/arc.h>
42 #include <sys/zio.h>
43 #include <sys/zap.h>
44 #include <sys/zfeature.h>
45 #include <sys/unique.h>
46 #include <sys/zfs_context.h>
47 #include <sys/zfs_ioctl.h>
48 #include <sys/spa.h>
49 #include <sys/spa_impl.h>
50 #include <sys/vdev.h>
51 #include <sys/zfs_znode.h>
52 #include <sys/zfs_onexit.h>
53 #include <sys/zvol.h>
54 #include <sys/dsl_scan.h>
55 #include <sys/dsl_deadlist.h>
56 #include <sys/dsl_destroy.h>
57 #include <sys/dsl_userhold.h>
58 #include <sys/dsl_bookmark.h>
59 #include <sys/dmu_send.h>
60 #include <sys/zio_checksum.h>
61 #include <sys/zio_compress.h>
62 #include <zfs_fletcher.h>
63 
64 /*
65  * The SPA supports block sizes up to 16MB.  However, very large blocks
66  * can have an impact on i/o latency (e.g. tying up a spinning disk for
67  * ~300ms), and also potentially on the memory allocator.  Therefore,
68  * we do not allow the recordsize to be set larger than zfs_max_recordsize
69  * (default 1MB).  Larger blocks can be created by changing this tunable,
70  * and pools with larger blocks can always be imported and used, regardless
71  * of this setting.
72  */
73 int zfs_max_recordsize = 1 * 1024 * 1024;
74 
75 #define	SWITCH64(x, y) \
76 	{ \
77 		uint64_t __tmp = (x); \
78 		(x) = (y); \
79 		(y) = __tmp; \
80 	}
81 
82 #define	DS_REF_MAX	(1ULL << 62)
83 
84 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
85 
86 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
87     uint64_t obj, dmu_tx_t *tx);
88 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
89     dmu_tx_t *tx);
90 
91 extern int spa_asize_inflation;
92 
93 static zil_header_t zero_zil;
94 
95 /*
96  * Figure out how much of this delta should be propogated to the dsl_dir
97  * layer.  If there's a refreservation, that space has already been
98  * partially accounted for in our ancestors.
99  */
100 static int64_t
101 parent_delta(dsl_dataset_t *ds, int64_t delta)
102 {
103 	dsl_dataset_phys_t *ds_phys;
104 	uint64_t old_bytes, new_bytes;
105 
106 	if (ds->ds_reserved == 0)
107 		return (delta);
108 
109 	ds_phys = dsl_dataset_phys(ds);
110 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
111 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
112 
113 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
114 	return (new_bytes - old_bytes);
115 }
116 
117 void
118 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
119 {
120 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
121 	int compressed = BP_GET_PSIZE(bp);
122 	int uncompressed = BP_GET_UCSIZE(bp);
123 	int64_t delta;
124 
125 	dprintf_bp(bp, "ds=%p", ds);
126 
127 	ASSERT(dmu_tx_is_syncing(tx));
128 	/* It could have been compressed away to nothing */
129 	if (BP_IS_HOLE(bp))
130 		return;
131 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
132 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
133 	if (ds == NULL) {
134 		dsl_pool_mos_diduse_space(tx->tx_pool,
135 		    used, compressed, uncompressed);
136 		return;
137 	}
138 
139 	ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
140 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
141 	mutex_enter(&ds->ds_lock);
142 	delta = parent_delta(ds, used);
143 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
144 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
145 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
146 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
147 
148 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
149 		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
150 		    B_TRUE;
151 	}
152 
153 	spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
154 	if (f != SPA_FEATURE_NONE)
155 		ds->ds_feature_activation_needed[f] = B_TRUE;
156 
157 	mutex_exit(&ds->ds_lock);
158 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
159 	    compressed, uncompressed, tx);
160 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
161 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
162 }
163 
164 /*
165  * Called when the specified segment has been remapped, and is thus no
166  * longer referenced in the head dataset.  The vdev must be indirect.
167  *
168  * If the segment is referenced by a snapshot, put it on the remap deadlist.
169  * Otherwise, add this segment to the obsolete spacemap.
170  */
171 void
172 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset,
173     uint64_t size, uint64_t birth, dmu_tx_t *tx)
174 {
175 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
176 
177 	ASSERT(dmu_tx_is_syncing(tx));
178 	ASSERT(birth <= tx->tx_txg);
179 	ASSERT(!ds->ds_is_snapshot);
180 
181 	if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
182 		spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
183 	} else {
184 		blkptr_t fakebp;
185 		dva_t *dva = &fakebp.blk_dva[0];
186 
187 		ASSERT(ds != NULL);
188 
189 		mutex_enter(&ds->ds_remap_deadlist_lock);
190 		if (!dsl_dataset_remap_deadlist_exists(ds)) {
191 			dsl_dataset_create_remap_deadlist(ds, tx);
192 		}
193 		mutex_exit(&ds->ds_remap_deadlist_lock);
194 
195 		BP_ZERO(&fakebp);
196 		fakebp.blk_birth = birth;
197 		DVA_SET_VDEV(dva, vdev);
198 		DVA_SET_OFFSET(dva, offset);
199 		DVA_SET_ASIZE(dva, size);
200 
201 		dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, tx);
202 	}
203 }
204 
205 int
206 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
207     boolean_t async)
208 {
209 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
210 
211 	int used = bp_get_dsize_sync(spa, bp);
212 	int compressed = BP_GET_PSIZE(bp);
213 	int uncompressed = BP_GET_UCSIZE(bp);
214 
215 	if (BP_IS_HOLE(bp))
216 		return (0);
217 
218 	ASSERT(dmu_tx_is_syncing(tx));
219 	ASSERT(bp->blk_birth <= tx->tx_txg);
220 
221 	if (ds == NULL) {
222 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
223 		dsl_pool_mos_diduse_space(tx->tx_pool,
224 		    -used, -compressed, -uncompressed);
225 		return (used);
226 	}
227 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
228 
229 	ASSERT(!ds->ds_is_snapshot);
230 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
231 
232 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
233 		int64_t delta;
234 
235 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
236 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
237 
238 		mutex_enter(&ds->ds_lock);
239 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
240 		    !DS_UNIQUE_IS_ACCURATE(ds));
241 		delta = parent_delta(ds, -used);
242 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
243 		mutex_exit(&ds->ds_lock);
244 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
245 		    delta, -compressed, -uncompressed, tx);
246 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
247 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
248 	} else {
249 		dprintf_bp(bp, "putting on dead list: %s", "");
250 		if (async) {
251 			/*
252 			 * We are here as part of zio's write done callback,
253 			 * which means we're a zio interrupt thread.  We can't
254 			 * call dsl_deadlist_insert() now because it may block
255 			 * waiting for I/O.  Instead, put bp on the deferred
256 			 * queue and let dsl_pool_sync() finish the job.
257 			 */
258 			bplist_append(&ds->ds_pending_deadlist, bp);
259 		} else {
260 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
261 		}
262 		ASSERT3U(ds->ds_prev->ds_object, ==,
263 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
264 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
265 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
266 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
267 		    ds->ds_object && bp->blk_birth >
268 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
269 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
270 			mutex_enter(&ds->ds_prev->ds_lock);
271 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
272 			mutex_exit(&ds->ds_prev->ds_lock);
273 		}
274 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
275 			dsl_dir_transfer_space(ds->ds_dir, used,
276 			    DD_USED_HEAD, DD_USED_SNAP, tx);
277 		}
278 	}
279 	mutex_enter(&ds->ds_lock);
280 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
281 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
282 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
283 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
284 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
285 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
286 	mutex_exit(&ds->ds_lock);
287 
288 	return (used);
289 }
290 
291 /*
292  * We have to release the fsid syncronously or we risk that a subsequent
293  * mount of the same dataset will fail to unique_insert the fsid.  This
294  * failure would manifest itself as the fsid of this dataset changing
295  * between mounts which makes NFS clients quite unhappy.
296  */
297 static void
298 dsl_dataset_evict_sync(void *dbu)
299 {
300 	dsl_dataset_t *ds = dbu;
301 
302 	ASSERT(ds->ds_owner == NULL);
303 
304 	unique_remove(ds->ds_fsid_guid);
305 }
306 
307 static void
308 dsl_dataset_evict_async(void *dbu)
309 {
310 	dsl_dataset_t *ds = dbu;
311 
312 	ASSERT(ds->ds_owner == NULL);
313 
314 	ds->ds_dbuf = NULL;
315 
316 	if (ds->ds_objset != NULL)
317 		dmu_objset_evict(ds->ds_objset);
318 
319 	if (ds->ds_prev) {
320 		dsl_dataset_rele(ds->ds_prev, ds);
321 		ds->ds_prev = NULL;
322 	}
323 
324 	bplist_destroy(&ds->ds_pending_deadlist);
325 	if (dsl_deadlist_is_open(&ds->ds_deadlist))
326 		dsl_deadlist_close(&ds->ds_deadlist);
327 	if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
328 		dsl_deadlist_close(&ds->ds_remap_deadlist);
329 	if (ds->ds_dir)
330 		dsl_dir_async_rele(ds->ds_dir, ds);
331 
332 	ASSERT(!list_link_active(&ds->ds_synced_link));
333 
334 	list_destroy(&ds->ds_prop_cbs);
335 	mutex_destroy(&ds->ds_lock);
336 	mutex_destroy(&ds->ds_opening_lock);
337 	mutex_destroy(&ds->ds_sendstream_lock);
338 	mutex_destroy(&ds->ds_remap_deadlist_lock);
339 	refcount_destroy(&ds->ds_longholds);
340 	rrw_destroy(&ds->ds_bp_rwlock);
341 
342 	kmem_free(ds, sizeof (dsl_dataset_t));
343 }
344 
345 int
346 dsl_dataset_get_snapname(dsl_dataset_t *ds)
347 {
348 	dsl_dataset_phys_t *headphys;
349 	int err;
350 	dmu_buf_t *headdbuf;
351 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
352 	objset_t *mos = dp->dp_meta_objset;
353 
354 	if (ds->ds_snapname[0])
355 		return (0);
356 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
357 		return (0);
358 
359 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
360 	    FTAG, &headdbuf);
361 	if (err != 0)
362 		return (err);
363 	headphys = headdbuf->db_data;
364 	err = zap_value_search(dp->dp_meta_objset,
365 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
366 	dmu_buf_rele(headdbuf, FTAG);
367 	return (err);
368 }
369 
370 int
371 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
372 {
373 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
374 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
375 	matchtype_t mt = 0;
376 	int err;
377 
378 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
379 		mt = MT_NORMALIZE;
380 
381 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
382 	    value, mt, NULL, 0, NULL);
383 	if (err == ENOTSUP && (mt & MT_NORMALIZE))
384 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
385 	return (err);
386 }
387 
388 int
389 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
390     boolean_t adj_cnt)
391 {
392 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
393 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
394 	matchtype_t mt = 0;
395 	int err;
396 
397 	dsl_dir_snap_cmtime_update(ds->ds_dir);
398 
399 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
400 		mt = MT_NORMALIZE;
401 
402 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
403 	if (err == ENOTSUP && (mt & MT_NORMALIZE))
404 		err = zap_remove(mos, snapobj, name, tx);
405 
406 	if (err == 0 && adj_cnt)
407 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
408 		    DD_FIELD_SNAPSHOT_COUNT, tx);
409 
410 	return (err);
411 }
412 
413 boolean_t
414 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
415 {
416 	dmu_buf_t *dbuf = ds->ds_dbuf;
417 	boolean_t result = B_FALSE;
418 
419 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
420 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
421 
422 		if (ds == dmu_buf_get_user(dbuf))
423 			result = B_TRUE;
424 		else
425 			dmu_buf_rele(dbuf, tag);
426 	}
427 
428 	return (result);
429 }
430 
431 int
432 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
433     dsl_dataset_t **dsp)
434 {
435 	objset_t *mos = dp->dp_meta_objset;
436 	dmu_buf_t *dbuf;
437 	dsl_dataset_t *ds;
438 	int err;
439 	dmu_object_info_t doi;
440 
441 	ASSERT(dsl_pool_config_held(dp));
442 
443 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
444 	if (err != 0)
445 		return (err);
446 
447 	/* Make sure dsobj has the correct object type. */
448 	dmu_object_info_from_db(dbuf, &doi);
449 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
450 		dmu_buf_rele(dbuf, tag);
451 		return (SET_ERROR(EINVAL));
452 	}
453 
454 	ds = dmu_buf_get_user(dbuf);
455 	if (ds == NULL) {
456 		dsl_dataset_t *winner = NULL;
457 
458 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
459 		ds->ds_dbuf = dbuf;
460 		ds->ds_object = dsobj;
461 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
462 
463 		err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj,
464 		    NULL, ds, &ds->ds_dir);
465 		if (err != 0) {
466 			kmem_free(ds, sizeof (dsl_dataset_t));
467 			dmu_buf_rele(dbuf, tag);
468 			return (err);
469 		}
470 
471 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
472 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
473 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
474 		mutex_init(&ds->ds_remap_deadlist_lock,
475 		    NULL, MUTEX_DEFAULT, NULL);
476 		rrw_init(&ds->ds_bp_rwlock, B_FALSE);
477 		refcount_create(&ds->ds_longholds);
478 
479 		bplist_create(&ds->ds_pending_deadlist);
480 
481 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
482 		    offsetof(dmu_sendarg_t, dsa_link));
483 
484 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
485 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
486 
487 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
488 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
489 				if (!(spa_feature_table[f].fi_flags &
490 				    ZFEATURE_FLAG_PER_DATASET))
491 					continue;
492 				err = zap_contains(mos, dsobj,
493 				    spa_feature_table[f].fi_guid);
494 				if (err == 0) {
495 					ds->ds_feature_inuse[f] = B_TRUE;
496 				} else {
497 					ASSERT3U(err, ==, ENOENT);
498 					err = 0;
499 				}
500 			}
501 		}
502 
503 		if (!ds->ds_is_snapshot) {
504 			ds->ds_snapname[0] = '\0';
505 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
506 				err = dsl_dataset_hold_obj(dp,
507 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
508 				    ds, &ds->ds_prev);
509 			}
510 			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
511 				int zaperr = zap_lookup(mos, ds->ds_object,
512 				    DS_FIELD_BOOKMARK_NAMES,
513 				    sizeof (ds->ds_bookmarks), 1,
514 				    &ds->ds_bookmarks);
515 				if (zaperr != ENOENT)
516 					VERIFY0(zaperr);
517 			}
518 		} else {
519 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
520 				err = dsl_dataset_get_snapname(ds);
521 			if (err == 0 &&
522 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
523 				err = zap_count(
524 				    ds->ds_dir->dd_pool->dp_meta_objset,
525 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
526 				    &ds->ds_userrefs);
527 			}
528 		}
529 
530 		if (err == 0 && !ds->ds_is_snapshot) {
531 			err = dsl_prop_get_int_ds(ds,
532 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
533 			    &ds->ds_reserved);
534 			if (err == 0) {
535 				err = dsl_prop_get_int_ds(ds,
536 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
537 				    &ds->ds_quota);
538 			}
539 		} else {
540 			ds->ds_reserved = ds->ds_quota = 0;
541 		}
542 
543 		dsl_deadlist_open(&ds->ds_deadlist,
544 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
545 		uint64_t remap_deadlist_obj =
546 		    dsl_dataset_get_remap_deadlist_object(ds);
547 		if (remap_deadlist_obj != 0) {
548 			dsl_deadlist_open(&ds->ds_remap_deadlist, mos,
549 			    remap_deadlist_obj);
550 		}
551 
552 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
553 		    dsl_dataset_evict_async, &ds->ds_dbuf);
554 		if (err == 0)
555 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
556 
557 		if (err != 0 || winner != NULL) {
558 			bplist_destroy(&ds->ds_pending_deadlist);
559 			dsl_deadlist_close(&ds->ds_deadlist);
560 			if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
561 				dsl_deadlist_close(&ds->ds_remap_deadlist);
562 			if (ds->ds_prev)
563 				dsl_dataset_rele(ds->ds_prev, ds);
564 			dsl_dir_rele(ds->ds_dir, ds);
565 			mutex_destroy(&ds->ds_lock);
566 			mutex_destroy(&ds->ds_opening_lock);
567 			mutex_destroy(&ds->ds_sendstream_lock);
568 			refcount_destroy(&ds->ds_longholds);
569 			kmem_free(ds, sizeof (dsl_dataset_t));
570 			if (err != 0) {
571 				dmu_buf_rele(dbuf, tag);
572 				return (err);
573 			}
574 			ds = winner;
575 		} else {
576 			ds->ds_fsid_guid =
577 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
578 			if (ds->ds_fsid_guid !=
579 			    dsl_dataset_phys(ds)->ds_fsid_guid) {
580 				zfs_dbgmsg("ds_fsid_guid changed from "
581 				    "%llx to %llx for pool %s dataset id %llu",
582 				    (long long)
583 				    dsl_dataset_phys(ds)->ds_fsid_guid,
584 				    (long long)ds->ds_fsid_guid,
585 				    spa_name(dp->dp_spa),
586 				    dsobj);
587 			}
588 		}
589 	}
590 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
591 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
592 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
593 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
594 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
595 	*dsp = ds;
596 	return (0);
597 }
598 
599 int
600 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
601     void *tag, dsl_dataset_t **dsp)
602 {
603 	dsl_dir_t *dd;
604 	const char *snapname;
605 	uint64_t obj;
606 	int err = 0;
607 	dsl_dataset_t *ds;
608 
609 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
610 	if (err != 0)
611 		return (err);
612 
613 	ASSERT(dsl_pool_config_held(dp));
614 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
615 	if (obj != 0)
616 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
617 	else
618 		err = SET_ERROR(ENOENT);
619 
620 	/* we may be looking for a snapshot */
621 	if (err == 0 && snapname != NULL) {
622 		dsl_dataset_t *snap_ds;
623 
624 		if (*snapname++ != '@') {
625 			dsl_dataset_rele(ds, tag);
626 			dsl_dir_rele(dd, FTAG);
627 			return (SET_ERROR(ENOENT));
628 		}
629 
630 		dprintf("looking for snapshot '%s'\n", snapname);
631 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
632 		if (err == 0)
633 			err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
634 		dsl_dataset_rele(ds, tag);
635 
636 		if (err == 0) {
637 			mutex_enter(&snap_ds->ds_lock);
638 			if (snap_ds->ds_snapname[0] == 0)
639 				(void) strlcpy(snap_ds->ds_snapname, snapname,
640 				    sizeof (snap_ds->ds_snapname));
641 			mutex_exit(&snap_ds->ds_lock);
642 			ds = snap_ds;
643 		}
644 	}
645 	if (err == 0)
646 		*dsp = ds;
647 	dsl_dir_rele(dd, FTAG);
648 	return (err);
649 }
650 
651 int
652 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
653     void *tag, dsl_dataset_t **dsp)
654 {
655 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
656 	if (err != 0)
657 		return (err);
658 	if (!dsl_dataset_tryown(*dsp, tag)) {
659 		dsl_dataset_rele(*dsp, tag);
660 		*dsp = NULL;
661 		return (SET_ERROR(EBUSY));
662 	}
663 	return (0);
664 }
665 
666 int
667 dsl_dataset_own(dsl_pool_t *dp, const char *name,
668     void *tag, dsl_dataset_t **dsp)
669 {
670 	int err = dsl_dataset_hold(dp, name, tag, dsp);
671 	if (err != 0)
672 		return (err);
673 	if (!dsl_dataset_tryown(*dsp, tag)) {
674 		dsl_dataset_rele(*dsp, tag);
675 		return (SET_ERROR(EBUSY));
676 	}
677 	return (0);
678 }
679 
680 /*
681  * See the comment above dsl_pool_hold() for details.  In summary, a long
682  * hold is used to prevent destruction of a dataset while the pool hold
683  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
684  *
685  * The dataset and pool must be held when this function is called.  After it
686  * is called, the pool hold may be released while the dataset is still held
687  * and accessed.
688  */
689 void
690 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
691 {
692 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
693 	(void) refcount_add(&ds->ds_longholds, tag);
694 }
695 
696 void
697 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
698 {
699 	(void) refcount_remove(&ds->ds_longholds, tag);
700 }
701 
702 /* Return B_TRUE if there are any long holds on this dataset. */
703 boolean_t
704 dsl_dataset_long_held(dsl_dataset_t *ds)
705 {
706 	return (!refcount_is_zero(&ds->ds_longholds));
707 }
708 
709 void
710 dsl_dataset_name(dsl_dataset_t *ds, char *name)
711 {
712 	if (ds == NULL) {
713 		(void) strcpy(name, "mos");
714 	} else {
715 		dsl_dir_name(ds->ds_dir, name);
716 		VERIFY0(dsl_dataset_get_snapname(ds));
717 		if (ds->ds_snapname[0]) {
718 			VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
719 			    <, ZFS_MAX_DATASET_NAME_LEN);
720 			/*
721 			 * We use a "recursive" mutex so that we
722 			 * can call dprintf_ds() with ds_lock held.
723 			 */
724 			if (!MUTEX_HELD(&ds->ds_lock)) {
725 				mutex_enter(&ds->ds_lock);
726 				VERIFY3U(strlcat(name, ds->ds_snapname,
727 				    ZFS_MAX_DATASET_NAME_LEN), <,
728 				    ZFS_MAX_DATASET_NAME_LEN);
729 				mutex_exit(&ds->ds_lock);
730 			} else {
731 				VERIFY3U(strlcat(name, ds->ds_snapname,
732 				    ZFS_MAX_DATASET_NAME_LEN), <,
733 				    ZFS_MAX_DATASET_NAME_LEN);
734 			}
735 		}
736 	}
737 }
738 
739 int
740 dsl_dataset_namelen(dsl_dataset_t *ds)
741 {
742 	VERIFY0(dsl_dataset_get_snapname(ds));
743 	mutex_enter(&ds->ds_lock);
744 	int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
745 	mutex_exit(&ds->ds_lock);
746 	return (len);
747 }
748 
749 void
750 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
751 {
752 	dmu_buf_rele(ds->ds_dbuf, tag);
753 }
754 
755 void
756 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
757 {
758 	ASSERT3P(ds->ds_owner, ==, tag);
759 	ASSERT(ds->ds_dbuf != NULL);
760 
761 	mutex_enter(&ds->ds_lock);
762 	ds->ds_owner = NULL;
763 	mutex_exit(&ds->ds_lock);
764 	dsl_dataset_long_rele(ds, tag);
765 	dsl_dataset_rele(ds, tag);
766 }
767 
768 boolean_t
769 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
770 {
771 	boolean_t gotit = FALSE;
772 
773 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
774 	mutex_enter(&ds->ds_lock);
775 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
776 		ds->ds_owner = tag;
777 		dsl_dataset_long_hold(ds, tag);
778 		gotit = TRUE;
779 	}
780 	mutex_exit(&ds->ds_lock);
781 	return (gotit);
782 }
783 
784 boolean_t
785 dsl_dataset_has_owner(dsl_dataset_t *ds)
786 {
787 	boolean_t rv;
788 	mutex_enter(&ds->ds_lock);
789 	rv = (ds->ds_owner != NULL);
790 	mutex_exit(&ds->ds_lock);
791 	return (rv);
792 }
793 
794 static void
795 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
796 {
797 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
798 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
799 	uint64_t zero = 0;
800 
801 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
802 
803 	spa_feature_incr(spa, f, tx);
804 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
805 
806 	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
807 	    sizeof (zero), 1, &zero, tx));
808 }
809 
810 void
811 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
812 {
813 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
814 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
815 
816 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
817 
818 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
819 	spa_feature_decr(spa, f, tx);
820 }
821 
822 uint64_t
823 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
824     uint64_t flags, dmu_tx_t *tx)
825 {
826 	dsl_pool_t *dp = dd->dd_pool;
827 	dmu_buf_t *dbuf;
828 	dsl_dataset_phys_t *dsphys;
829 	uint64_t dsobj;
830 	objset_t *mos = dp->dp_meta_objset;
831 
832 	if (origin == NULL)
833 		origin = dp->dp_origin_snap;
834 
835 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
836 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
837 	ASSERT(dmu_tx_is_syncing(tx));
838 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
839 
840 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
841 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
842 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
843 	dmu_buf_will_dirty(dbuf, tx);
844 	dsphys = dbuf->db_data;
845 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
846 	dsphys->ds_dir_obj = dd->dd_object;
847 	dsphys->ds_flags = flags;
848 	dsphys->ds_fsid_guid = unique_create();
849 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
850 	    sizeof (dsphys->ds_guid));
851 	dsphys->ds_snapnames_zapobj =
852 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
853 	    DMU_OT_NONE, 0, tx);
854 	dsphys->ds_creation_time = gethrestime_sec();
855 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
856 
857 	if (origin == NULL) {
858 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
859 	} else {
860 		dsl_dataset_t *ohds; /* head of the origin snapshot */
861 
862 		dsphys->ds_prev_snap_obj = origin->ds_object;
863 		dsphys->ds_prev_snap_txg =
864 		    dsl_dataset_phys(origin)->ds_creation_txg;
865 		dsphys->ds_referenced_bytes =
866 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
867 		dsphys->ds_compressed_bytes =
868 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
869 		dsphys->ds_uncompressed_bytes =
870 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
871 		rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
872 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
873 		rrw_exit(&origin->ds_bp_rwlock, FTAG);
874 
875 		/*
876 		 * Inherit flags that describe the dataset's contents
877 		 * (INCONSISTENT) or properties (Case Insensitive).
878 		 */
879 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
880 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
881 
882 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
883 			if (origin->ds_feature_inuse[f])
884 				dsl_dataset_activate_feature(dsobj, f, tx);
885 		}
886 
887 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
888 		dsl_dataset_phys(origin)->ds_num_children++;
889 
890 		VERIFY0(dsl_dataset_hold_obj(dp,
891 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
892 		    FTAG, &ohds));
893 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
894 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
895 		dsl_dataset_rele(ohds, FTAG);
896 
897 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
898 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
899 				dsl_dataset_phys(origin)->ds_next_clones_obj =
900 				    zap_create(mos,
901 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
902 			}
903 			VERIFY0(zap_add_int(mos,
904 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
905 			    dsobj, tx));
906 		}
907 
908 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
909 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
910 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
911 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
912 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
913 				dsl_dir_phys(origin->ds_dir)->dd_clones =
914 				    zap_create(mos,
915 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
916 			}
917 			VERIFY0(zap_add_int(mos,
918 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
919 			    dsobj, tx));
920 		}
921 	}
922 
923 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
924 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
925 
926 	dmu_buf_rele(dbuf, FTAG);
927 
928 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
929 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
930 
931 	return (dsobj);
932 }
933 
934 static void
935 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
936 {
937 	objset_t *os;
938 
939 	VERIFY0(dmu_objset_from_ds(ds, &os));
940 	if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
941 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
942 		zio_t *zio;
943 
944 		bzero(&os->os_zil_header, sizeof (os->os_zil_header));
945 
946 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
947 		dsl_dataset_sync(ds, zio, tx);
948 		VERIFY0(zio_wait(zio));
949 
950 		/* dsl_dataset_sync_done will drop this reference. */
951 		dmu_buf_add_ref(ds->ds_dbuf, ds);
952 		dsl_dataset_sync_done(ds, tx);
953 	}
954 }
955 
956 uint64_t
957 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
958     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
959 {
960 	dsl_pool_t *dp = pdd->dd_pool;
961 	uint64_t dsobj, ddobj;
962 	dsl_dir_t *dd;
963 
964 	ASSERT(dmu_tx_is_syncing(tx));
965 	ASSERT(lastname[0] != '@');
966 
967 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
968 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
969 
970 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
971 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
972 
973 	dsl_deleg_set_create_perms(dd, tx, cr);
974 
975 	/*
976 	 * Since we're creating a new node we know it's a leaf, so we can
977 	 * initialize the counts if the limit feature is active.
978 	 */
979 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
980 		uint64_t cnt = 0;
981 		objset_t *os = dd->dd_pool->dp_meta_objset;
982 
983 		dsl_dir_zapify(dd, tx);
984 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
985 		    sizeof (cnt), 1, &cnt, tx));
986 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
987 		    sizeof (cnt), 1, &cnt, tx));
988 	}
989 
990 	dsl_dir_rele(dd, FTAG);
991 
992 	/*
993 	 * If we are creating a clone, make sure we zero out any stale
994 	 * data from the origin snapshots zil header.
995 	 */
996 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
997 		dsl_dataset_t *ds;
998 
999 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1000 		dsl_dataset_zero_zil(ds, tx);
1001 		dsl_dataset_rele(ds, FTAG);
1002 	}
1003 
1004 	return (dsobj);
1005 }
1006 
1007 /*
1008  * The unique space in the head dataset can be calculated by subtracting
1009  * the space used in the most recent snapshot, that is still being used
1010  * in this file system, from the space currently in use.  To figure out
1011  * the space in the most recent snapshot still in use, we need to take
1012  * the total space used in the snapshot and subtract out the space that
1013  * has been freed up since the snapshot was taken.
1014  */
1015 void
1016 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1017 {
1018 	uint64_t mrs_used;
1019 	uint64_t dlused, dlcomp, dluncomp;
1020 
1021 	ASSERT(!ds->ds_is_snapshot);
1022 
1023 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1024 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1025 	else
1026 		mrs_used = 0;
1027 
1028 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1029 
1030 	ASSERT3U(dlused, <=, mrs_used);
1031 	dsl_dataset_phys(ds)->ds_unique_bytes =
1032 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1033 
1034 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1035 	    SPA_VERSION_UNIQUE_ACCURATE)
1036 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1037 }
1038 
1039 void
1040 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1041     dmu_tx_t *tx)
1042 {
1043 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1044 	uint64_t count;
1045 	int err;
1046 
1047 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1048 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1049 	    obj, tx);
1050 	/*
1051 	 * The err should not be ENOENT, but a bug in a previous version
1052 	 * of the code could cause upgrade_clones_cb() to not set
1053 	 * ds_next_snap_obj when it should, leading to a missing entry.
1054 	 * If we knew that the pool was created after
1055 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1056 	 * ENOENT.  However, at least we can check that we don't have
1057 	 * too many entries in the next_clones_obj even after failing to
1058 	 * remove this one.
1059 	 */
1060 	if (err != ENOENT)
1061 		VERIFY0(err);
1062 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1063 	    &count));
1064 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1065 }
1066 
1067 
1068 blkptr_t *
1069 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1070 {
1071 	return (&dsl_dataset_phys(ds)->ds_bp);
1072 }
1073 
1074 spa_t *
1075 dsl_dataset_get_spa(dsl_dataset_t *ds)
1076 {
1077 	return (ds->ds_dir->dd_pool->dp_spa);
1078 }
1079 
1080 void
1081 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1082 {
1083 	dsl_pool_t *dp;
1084 
1085 	if (ds == NULL) /* this is the meta-objset */
1086 		return;
1087 
1088 	ASSERT(ds->ds_objset != NULL);
1089 
1090 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1091 		panic("dirtying snapshot!");
1092 
1093 	/* Must not dirty a dataset in the same txg where it got snapshotted. */
1094 	ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1095 
1096 	dp = ds->ds_dir->dd_pool;
1097 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1098 		/* up the hold count until we can be written out */
1099 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1100 	}
1101 }
1102 
1103 boolean_t
1104 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1105 {
1106 	for (int t = 0; t < TXG_SIZE; t++) {
1107 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1108 		    ds, t))
1109 			return (B_TRUE);
1110 	}
1111 	return (B_FALSE);
1112 }
1113 
1114 static int
1115 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1116 {
1117 	uint64_t asize;
1118 
1119 	if (!dmu_tx_is_syncing(tx))
1120 		return (0);
1121 
1122 	/*
1123 	 * If there's an fs-only reservation, any blocks that might become
1124 	 * owned by the snapshot dataset must be accommodated by space
1125 	 * outside of the reservation.
1126 	 */
1127 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1128 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1129 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1130 		return (SET_ERROR(ENOSPC));
1131 
1132 	/*
1133 	 * Propagate any reserved space for this snapshot to other
1134 	 * snapshot checks in this sync group.
1135 	 */
1136 	if (asize > 0)
1137 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1138 
1139 	return (0);
1140 }
1141 
1142 int
1143 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1144     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1145 {
1146 	int error;
1147 	uint64_t value;
1148 
1149 	ds->ds_trysnap_txg = tx->tx_txg;
1150 
1151 	if (!dmu_tx_is_syncing(tx))
1152 		return (0);
1153 
1154 	/*
1155 	 * We don't allow multiple snapshots of the same txg.  If there
1156 	 * is already one, try again.
1157 	 */
1158 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1159 		return (SET_ERROR(EAGAIN));
1160 
1161 	/*
1162 	 * Check for conflicting snapshot name.
1163 	 */
1164 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1165 	if (error == 0)
1166 		return (SET_ERROR(EEXIST));
1167 	if (error != ENOENT)
1168 		return (error);
1169 
1170 	/*
1171 	 * We don't allow taking snapshots of inconsistent datasets, such as
1172 	 * those into which we are currently receiving.  However, if we are
1173 	 * creating this snapshot as part of a receive, this check will be
1174 	 * executed atomically with respect to the completion of the receive
1175 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1176 	 * case we ignore this, knowing it will be fixed up for us shortly in
1177 	 * dmu_recv_end_sync().
1178 	 */
1179 	if (!recv && DS_IS_INCONSISTENT(ds))
1180 		return (SET_ERROR(EBUSY));
1181 
1182 	/*
1183 	 * Skip the check for temporary snapshots or if we have already checked
1184 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1185 	 * check the count here when we're receiving a stream.
1186 	 */
1187 	if (cnt != 0 && cr != NULL) {
1188 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1189 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1190 		if (error != 0)
1191 			return (error);
1192 	}
1193 
1194 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1195 	if (error != 0)
1196 		return (error);
1197 
1198 	return (0);
1199 }
1200 
1201 int
1202 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1203 {
1204 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1205 	dsl_pool_t *dp = dmu_tx_pool(tx);
1206 	nvpair_t *pair;
1207 	int rv = 0;
1208 
1209 	/*
1210 	 * Pre-compute how many total new snapshots will be created for each
1211 	 * level in the tree and below. This is needed for validating the
1212 	 * snapshot limit when either taking a recursive snapshot or when
1213 	 * taking multiple snapshots.
1214 	 *
1215 	 * The problem is that the counts are not actually adjusted when
1216 	 * we are checking, only when we finally sync. For a single snapshot,
1217 	 * this is easy, the count will increase by 1 at each node up the tree,
1218 	 * but its more complicated for the recursive/multiple snapshot case.
1219 	 *
1220 	 * The dsl_fs_ss_limit_check function does recursively check the count
1221 	 * at each level up the tree but since it is validating each snapshot
1222 	 * independently we need to be sure that we are validating the complete
1223 	 * count for the entire set of snapshots. We do this by rolling up the
1224 	 * counts for each component of the name into an nvlist and then
1225 	 * checking each of those cases with the aggregated count.
1226 	 *
1227 	 * This approach properly handles not only the recursive snapshot
1228 	 * case (where we get all of those on the ddsa_snaps list) but also
1229 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1230 	 * validate the limit on 'a' using a count of 2).
1231 	 *
1232 	 * We validate the snapshot names in the third loop and only report
1233 	 * name errors once.
1234 	 */
1235 	if (dmu_tx_is_syncing(tx)) {
1236 		nvlist_t *cnt_track = NULL;
1237 		cnt_track = fnvlist_alloc();
1238 
1239 		/* Rollup aggregated counts into the cnt_track list */
1240 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1241 		    pair != NULL;
1242 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1243 			char *pdelim;
1244 			uint64_t val;
1245 			char nm[MAXPATHLEN];
1246 
1247 			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1248 			pdelim = strchr(nm, '@');
1249 			if (pdelim == NULL)
1250 				continue;
1251 			*pdelim = '\0';
1252 
1253 			do {
1254 				if (nvlist_lookup_uint64(cnt_track, nm,
1255 				    &val) == 0) {
1256 					/* update existing entry */
1257 					fnvlist_add_uint64(cnt_track, nm,
1258 					    val + 1);
1259 				} else {
1260 					/* add to list */
1261 					fnvlist_add_uint64(cnt_track, nm, 1);
1262 				}
1263 
1264 				pdelim = strrchr(nm, '/');
1265 				if (pdelim != NULL)
1266 					*pdelim = '\0';
1267 			} while (pdelim != NULL);
1268 		}
1269 
1270 		/* Check aggregated counts at each level */
1271 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1272 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1273 			int error = 0;
1274 			char *name;
1275 			uint64_t cnt = 0;
1276 			dsl_dataset_t *ds;
1277 
1278 			name = nvpair_name(pair);
1279 			cnt = fnvpair_value_uint64(pair);
1280 			ASSERT(cnt > 0);
1281 
1282 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1283 			if (error == 0) {
1284 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1285 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1286 				    ddsa->ddsa_cr);
1287 				dsl_dataset_rele(ds, FTAG);
1288 			}
1289 
1290 			if (error != 0) {
1291 				if (ddsa->ddsa_errors != NULL)
1292 					fnvlist_add_int32(ddsa->ddsa_errors,
1293 					    name, error);
1294 				rv = error;
1295 				/* only report one error for this check */
1296 				break;
1297 			}
1298 		}
1299 		nvlist_free(cnt_track);
1300 	}
1301 
1302 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1303 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1304 		int error = 0;
1305 		dsl_dataset_t *ds;
1306 		char *name, *atp;
1307 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1308 
1309 		name = nvpair_name(pair);
1310 		if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1311 			error = SET_ERROR(ENAMETOOLONG);
1312 		if (error == 0) {
1313 			atp = strchr(name, '@');
1314 			if (atp == NULL)
1315 				error = SET_ERROR(EINVAL);
1316 			if (error == 0)
1317 				(void) strlcpy(dsname, name, atp - name + 1);
1318 		}
1319 		if (error == 0)
1320 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1321 		if (error == 0) {
1322 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1323 			error = dsl_dataset_snapshot_check_impl(ds,
1324 			    atp + 1, tx, B_FALSE, 0, NULL);
1325 			dsl_dataset_rele(ds, FTAG);
1326 		}
1327 
1328 		if (error != 0) {
1329 			if (ddsa->ddsa_errors != NULL) {
1330 				fnvlist_add_int32(ddsa->ddsa_errors,
1331 				    name, error);
1332 			}
1333 			rv = error;
1334 		}
1335 	}
1336 
1337 	return (rv);
1338 }
1339 
1340 void
1341 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1342     dmu_tx_t *tx)
1343 {
1344 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1345 	dmu_buf_t *dbuf;
1346 	dsl_dataset_phys_t *dsphys;
1347 	uint64_t dsobj, crtxg;
1348 	objset_t *mos = dp->dp_meta_objset;
1349 	objset_t *os;
1350 
1351 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1352 
1353 	/*
1354 	 * If we are on an old pool, the zil must not be active, in which
1355 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1356 	 */
1357 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1358 	    dmu_objset_from_ds(ds, &os) != 0 ||
1359 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1360 	    sizeof (zero_zil)) == 0);
1361 
1362 	/* Should not snapshot a dirty dataset. */
1363 	ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1364 	    ds, tx->tx_txg));
1365 
1366 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1367 
1368 	/*
1369 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1370 	 */
1371 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1372 		crtxg = 1;
1373 	else
1374 		crtxg = tx->tx_txg;
1375 
1376 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1377 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1378 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1379 	dmu_buf_will_dirty(dbuf, tx);
1380 	dsphys = dbuf->db_data;
1381 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1382 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1383 	dsphys->ds_fsid_guid = unique_create();
1384 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1385 	    sizeof (dsphys->ds_guid));
1386 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1387 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1388 	dsphys->ds_next_snap_obj = ds->ds_object;
1389 	dsphys->ds_num_children = 1;
1390 	dsphys->ds_creation_time = gethrestime_sec();
1391 	dsphys->ds_creation_txg = crtxg;
1392 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1393 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1394 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1395 	dsphys->ds_uncompressed_bytes =
1396 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1397 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1398 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1399 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1400 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
1401 	dmu_buf_rele(dbuf, FTAG);
1402 
1403 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1404 		if (ds->ds_feature_inuse[f])
1405 			dsl_dataset_activate_feature(dsobj, f, tx);
1406 	}
1407 
1408 	ASSERT3U(ds->ds_prev != 0, ==,
1409 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1410 	if (ds->ds_prev) {
1411 		uint64_t next_clones_obj =
1412 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1413 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1414 		    ds->ds_object ||
1415 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1416 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1417 		    ds->ds_object) {
1418 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1419 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1420 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1421 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1422 		} else if (next_clones_obj != 0) {
1423 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1424 			    dsphys->ds_next_snap_obj, tx);
1425 			VERIFY0(zap_add_int(mos,
1426 			    next_clones_obj, dsobj, tx));
1427 		}
1428 	}
1429 
1430 	/*
1431 	 * If we have a reference-reservation on this dataset, we will
1432 	 * need to increase the amount of refreservation being charged
1433 	 * since our unique space is going to zero.
1434 	 */
1435 	if (ds->ds_reserved) {
1436 		int64_t delta;
1437 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1438 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1439 		    ds->ds_reserved);
1440 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1441 		    delta, 0, 0, tx);
1442 	}
1443 
1444 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1445 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1446 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1447 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1448 	dsl_deadlist_close(&ds->ds_deadlist);
1449 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1450 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1451 	dsl_deadlist_add_key(&ds->ds_deadlist,
1452 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1453 
1454 	if (dsl_dataset_remap_deadlist_exists(ds)) {
1455 		uint64_t remap_deadlist_obj =
1456 		    dsl_dataset_get_remap_deadlist_object(ds);
1457 		/*
1458 		 * Move the remap_deadlist to the snapshot.  The head
1459 		 * will create a new remap deadlist on demand, from
1460 		 * dsl_dataset_block_remapped().
1461 		 */
1462 		dsl_dataset_unset_remap_deadlist_object(ds, tx);
1463 		dsl_deadlist_close(&ds->ds_remap_deadlist);
1464 
1465 		dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1466 		VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST,
1467 		    sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx));
1468 	}
1469 
1470 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1471 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1472 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1473 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1474 
1475 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1476 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1477 
1478 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1479 	    snapname, 8, 1, &dsobj, tx));
1480 
1481 	if (ds->ds_prev)
1482 		dsl_dataset_rele(ds->ds_prev, ds);
1483 	VERIFY0(dsl_dataset_hold_obj(dp,
1484 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1485 
1486 	dsl_scan_ds_snapshotted(ds, tx);
1487 
1488 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1489 
1490 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1491 }
1492 
1493 void
1494 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1495 {
1496 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1497 	dsl_pool_t *dp = dmu_tx_pool(tx);
1498 	nvpair_t *pair;
1499 
1500 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1501 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1502 		dsl_dataset_t *ds;
1503 		char *name, *atp;
1504 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1505 
1506 		name = nvpair_name(pair);
1507 		atp = strchr(name, '@');
1508 		(void) strlcpy(dsname, name, atp - name + 1);
1509 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1510 
1511 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1512 		if (ddsa->ddsa_props != NULL) {
1513 			dsl_props_set_sync_impl(ds->ds_prev,
1514 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1515 		}
1516 		dsl_dataset_rele(ds, FTAG);
1517 	}
1518 }
1519 
1520 /*
1521  * The snapshots must all be in the same pool.
1522  * All-or-nothing: if there are any failures, nothing will be modified.
1523  */
1524 int
1525 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1526 {
1527 	dsl_dataset_snapshot_arg_t ddsa;
1528 	nvpair_t *pair;
1529 	boolean_t needsuspend;
1530 	int error;
1531 	spa_t *spa;
1532 	char *firstname;
1533 	nvlist_t *suspended = NULL;
1534 
1535 	pair = nvlist_next_nvpair(snaps, NULL);
1536 	if (pair == NULL)
1537 		return (0);
1538 	firstname = nvpair_name(pair);
1539 
1540 	error = spa_open(firstname, &spa, FTAG);
1541 	if (error != 0)
1542 		return (error);
1543 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1544 	spa_close(spa, FTAG);
1545 
1546 	if (needsuspend) {
1547 		suspended = fnvlist_alloc();
1548 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1549 		    pair = nvlist_next_nvpair(snaps, pair)) {
1550 			char fsname[ZFS_MAX_DATASET_NAME_LEN];
1551 			char *snapname = nvpair_name(pair);
1552 			char *atp;
1553 			void *cookie;
1554 
1555 			atp = strchr(snapname, '@');
1556 			if (atp == NULL) {
1557 				error = SET_ERROR(EINVAL);
1558 				break;
1559 			}
1560 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1561 
1562 			error = zil_suspend(fsname, &cookie);
1563 			if (error != 0)
1564 				break;
1565 			fnvlist_add_uint64(suspended, fsname,
1566 			    (uintptr_t)cookie);
1567 		}
1568 	}
1569 
1570 	ddsa.ddsa_snaps = snaps;
1571 	ddsa.ddsa_props = props;
1572 	ddsa.ddsa_errors = errors;
1573 	ddsa.ddsa_cr = CRED();
1574 
1575 	if (error == 0) {
1576 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1577 		    dsl_dataset_snapshot_sync, &ddsa,
1578 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1579 	}
1580 
1581 	if (suspended != NULL) {
1582 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1583 		    pair = nvlist_next_nvpair(suspended, pair)) {
1584 			zil_resume((void *)(uintptr_t)
1585 			    fnvpair_value_uint64(pair));
1586 		}
1587 		fnvlist_free(suspended);
1588 	}
1589 
1590 	return (error);
1591 }
1592 
1593 typedef struct dsl_dataset_snapshot_tmp_arg {
1594 	const char *ddsta_fsname;
1595 	const char *ddsta_snapname;
1596 	minor_t ddsta_cleanup_minor;
1597 	const char *ddsta_htag;
1598 } dsl_dataset_snapshot_tmp_arg_t;
1599 
1600 static int
1601 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1602 {
1603 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1604 	dsl_pool_t *dp = dmu_tx_pool(tx);
1605 	dsl_dataset_t *ds;
1606 	int error;
1607 
1608 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1609 	if (error != 0)
1610 		return (error);
1611 
1612 	/* NULL cred means no limit check for tmp snapshot */
1613 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1614 	    tx, B_FALSE, 0, NULL);
1615 	if (error != 0) {
1616 		dsl_dataset_rele(ds, FTAG);
1617 		return (error);
1618 	}
1619 
1620 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1621 		dsl_dataset_rele(ds, FTAG);
1622 		return (SET_ERROR(ENOTSUP));
1623 	}
1624 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1625 	    B_TRUE, tx);
1626 	if (error != 0) {
1627 		dsl_dataset_rele(ds, FTAG);
1628 		return (error);
1629 	}
1630 
1631 	dsl_dataset_rele(ds, FTAG);
1632 	return (0);
1633 }
1634 
1635 static void
1636 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1637 {
1638 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1639 	dsl_pool_t *dp = dmu_tx_pool(tx);
1640 	dsl_dataset_t *ds;
1641 
1642 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1643 
1644 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1645 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1646 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1647 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1648 
1649 	dsl_dataset_rele(ds, FTAG);
1650 }
1651 
1652 int
1653 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1654     minor_t cleanup_minor, const char *htag)
1655 {
1656 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1657 	int error;
1658 	spa_t *spa;
1659 	boolean_t needsuspend;
1660 	void *cookie;
1661 
1662 	ddsta.ddsta_fsname = fsname;
1663 	ddsta.ddsta_snapname = snapname;
1664 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1665 	ddsta.ddsta_htag = htag;
1666 
1667 	error = spa_open(fsname, &spa, FTAG);
1668 	if (error != 0)
1669 		return (error);
1670 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1671 	spa_close(spa, FTAG);
1672 
1673 	if (needsuspend) {
1674 		error = zil_suspend(fsname, &cookie);
1675 		if (error != 0)
1676 			return (error);
1677 	}
1678 
1679 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1680 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1681 
1682 	if (needsuspend)
1683 		zil_resume(cookie);
1684 	return (error);
1685 }
1686 
1687 void
1688 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1689 {
1690 	ASSERT(dmu_tx_is_syncing(tx));
1691 	ASSERT(ds->ds_objset != NULL);
1692 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1693 
1694 	/*
1695 	 * in case we had to change ds_fsid_guid when we opened it,
1696 	 * sync it out now.
1697 	 */
1698 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1699 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1700 
1701 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1702 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1703 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1704 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1705 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1706 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1707 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1708 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1709 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1710 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1711 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1712 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1713 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1714 	}
1715 
1716 	dmu_objset_sync(ds->ds_objset, zio, tx);
1717 
1718 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1719 		if (ds->ds_feature_activation_needed[f]) {
1720 			if (ds->ds_feature_inuse[f])
1721 				continue;
1722 			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1723 			ds->ds_feature_inuse[f] = B_TRUE;
1724 		}
1725 	}
1726 }
1727 
1728 static int
1729 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1730 {
1731 	dsl_deadlist_t *dl = arg;
1732 	dsl_deadlist_insert(dl, bp, tx);
1733 	return (0);
1734 }
1735 
1736 void
1737 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
1738 {
1739 	objset_t *os = ds->ds_objset;
1740 
1741 	bplist_iterate(&ds->ds_pending_deadlist,
1742 	    deadlist_enqueue_cb, &ds->ds_deadlist, tx);
1743 
1744 	if (os->os_synced_dnodes != NULL) {
1745 		multilist_destroy(os->os_synced_dnodes);
1746 		os->os_synced_dnodes = NULL;
1747 	}
1748 
1749 	ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
1750 
1751 	dmu_buf_rele(ds->ds_dbuf, ds);
1752 }
1753 
1754 int
1755 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val)
1756 {
1757 	uint64_t count = 0;
1758 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1759 	zap_cursor_t zc;
1760 	zap_attribute_t za;
1761 
1762 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1763 
1764 	/*
1765 	 * There may be missing entries in ds_next_clones_obj
1766 	 * due to a bug in a previous version of the code.
1767 	 * Only trust it if it has the right number of entries.
1768 	 */
1769 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1770 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1771 		    &count));
1772 	}
1773 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1) {
1774 		return (ENOENT);
1775 	}
1776 	for (zap_cursor_init(&zc, mos,
1777 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1778 	    zap_cursor_retrieve(&zc, &za) == 0;
1779 	    zap_cursor_advance(&zc)) {
1780 		dsl_dataset_t *clone;
1781 		char buf[ZFS_MAX_DATASET_NAME_LEN];
1782 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1783 		    za.za_first_integer, FTAG, &clone));
1784 		dsl_dir_name(clone->ds_dir, buf);
1785 		fnvlist_add_boolean(val, buf);
1786 		dsl_dataset_rele(clone, FTAG);
1787 	}
1788 	zap_cursor_fini(&zc);
1789 	return (0);
1790 }
1791 
1792 void
1793 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1794 {
1795 	nvlist_t *propval = fnvlist_alloc();
1796 	nvlist_t *val;
1797 
1798 	/*
1799 	 * We use nvlist_alloc() instead of fnvlist_alloc() because the
1800 	 * latter would allocate the list with NV_UNIQUE_NAME flag.
1801 	 * As a result, every time a clone name is appended to the list
1802 	 * it would be (linearly) searched for for a duplicate name.
1803 	 * We already know that all clone names must be unique and we
1804 	 * want avoid the quadratic complexity of double-checking that
1805 	 * because we can have a large number of clones.
1806 	 */
1807 	VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP));
1808 
1809 	if (get_clones_stat_impl(ds, val) == 0) {
1810 		fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1811 		fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
1812 		    propval);
1813 	}
1814 
1815 	nvlist_free(val);
1816 	nvlist_free(propval);
1817 }
1818 
1819 /*
1820  * Returns a string that represents the receive resume stats token. It should
1821  * be freed with strfree().
1822  */
1823 char *
1824 get_receive_resume_stats_impl(dsl_dataset_t *ds)
1825 {
1826 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1827 
1828 	if (dsl_dataset_has_resume_receive_state(ds)) {
1829 		char *str;
1830 		void *packed;
1831 		uint8_t *compressed;
1832 		uint64_t val;
1833 		nvlist_t *token_nv = fnvlist_alloc();
1834 		size_t packed_size, compressed_size;
1835 
1836 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1837 		    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1838 			fnvlist_add_uint64(token_nv, "fromguid", val);
1839 		}
1840 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1841 		    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1842 			fnvlist_add_uint64(token_nv, "object", val);
1843 		}
1844 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1845 		    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1846 			fnvlist_add_uint64(token_nv, "offset", val);
1847 		}
1848 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1849 		    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1850 			fnvlist_add_uint64(token_nv, "bytes", val);
1851 		}
1852 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1853 		    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1854 			fnvlist_add_uint64(token_nv, "toguid", val);
1855 		}
1856 		char buf[256];
1857 		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1858 		    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1859 			fnvlist_add_string(token_nv, "toname", buf);
1860 		}
1861 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1862 		    DS_FIELD_RESUME_LARGEBLOCK) == 0) {
1863 			fnvlist_add_boolean(token_nv, "largeblockok");
1864 		}
1865 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1866 		    DS_FIELD_RESUME_EMBEDOK) == 0) {
1867 			fnvlist_add_boolean(token_nv, "embedok");
1868 		}
1869 		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1870 		    DS_FIELD_RESUME_COMPRESSOK) == 0) {
1871 			fnvlist_add_boolean(token_nv, "compressok");
1872 		}
1873 		packed = fnvlist_pack(token_nv, &packed_size);
1874 		fnvlist_free(token_nv);
1875 		compressed = kmem_alloc(packed_size, KM_SLEEP);
1876 
1877 		compressed_size = gzip_compress(packed, compressed,
1878 		    packed_size, packed_size, 6);
1879 
1880 		zio_cksum_t cksum;
1881 		fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1882 
1883 		str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1884 		for (int i = 0; i < compressed_size; i++) {
1885 			(void) sprintf(str + i * 2, "%02x", compressed[i]);
1886 		}
1887 		str[compressed_size * 2] = '\0';
1888 		char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1889 		    ZFS_SEND_RESUME_TOKEN_VERSION,
1890 		    (longlong_t)cksum.zc_word[0],
1891 		    (longlong_t)packed_size, str);
1892 		kmem_free(packed, packed_size);
1893 		kmem_free(str, compressed_size * 2 + 1);
1894 		kmem_free(compressed, packed_size);
1895 		return (propval);
1896 	}
1897 	return (strdup(""));
1898 }
1899 
1900 /*
1901  * Returns a string that represents the receive resume stats token of the
1902  * dataset's child. It should be freed with strfree().
1903  */
1904 char *
1905 get_child_receive_stats(dsl_dataset_t *ds)
1906 {
1907 	char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1908 	dsl_dataset_t *recv_ds;
1909 	dsl_dataset_name(ds, recvname);
1910 	if (strlcat(recvname, "/", sizeof (recvname)) <
1911 	    sizeof (recvname) &&
1912 	    strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1913 	    sizeof (recvname) &&
1914 	    dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG,
1915 	    &recv_ds)  == 0) {
1916 		char *propval = get_receive_resume_stats_impl(recv_ds);
1917 		dsl_dataset_rele(recv_ds, FTAG);
1918 		return (propval);
1919 	}
1920 	return (strdup(""));
1921 }
1922 
1923 static void
1924 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1925 {
1926 	char *propval = get_receive_resume_stats_impl(ds);
1927 	if (strcmp(propval, "") != 0) {
1928 		dsl_prop_nvlist_add_string(nv,
1929 		    ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1930 	} else {
1931 		char *childval = get_child_receive_stats(ds);
1932 		if (strcmp(childval, "") != 0) {
1933 			dsl_prop_nvlist_add_string(nv,
1934 			    ZFS_PROP_RECEIVE_RESUME_TOKEN, childval);
1935 		}
1936 		strfree(childval);
1937 	}
1938 	strfree(propval);
1939 }
1940 
1941 uint64_t
1942 dsl_get_refratio(dsl_dataset_t *ds)
1943 {
1944 	uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1945 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1946 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
1947 	return (ratio);
1948 }
1949 
1950 uint64_t
1951 dsl_get_logicalreferenced(dsl_dataset_t *ds)
1952 {
1953 	return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1954 }
1955 
1956 uint64_t
1957 dsl_get_compressratio(dsl_dataset_t *ds)
1958 {
1959 	if (ds->ds_is_snapshot) {
1960 		return (dsl_get_refratio(ds));
1961 	} else {
1962 		dsl_dir_t *dd = ds->ds_dir;
1963 		mutex_enter(&dd->dd_lock);
1964 		uint64_t val = dsl_dir_get_compressratio(dd);
1965 		mutex_exit(&dd->dd_lock);
1966 		return (val);
1967 	}
1968 }
1969 
1970 uint64_t
1971 dsl_get_used(dsl_dataset_t *ds)
1972 {
1973 	if (ds->ds_is_snapshot) {
1974 		return (dsl_dataset_phys(ds)->ds_unique_bytes);
1975 	} else {
1976 		dsl_dir_t *dd = ds->ds_dir;
1977 		mutex_enter(&dd->dd_lock);
1978 		uint64_t val = dsl_dir_get_used(dd);
1979 		mutex_exit(&dd->dd_lock);
1980 		return (val);
1981 	}
1982 }
1983 
1984 uint64_t
1985 dsl_get_creation(dsl_dataset_t *ds)
1986 {
1987 	return (dsl_dataset_phys(ds)->ds_creation_time);
1988 }
1989 
1990 uint64_t
1991 dsl_get_creationtxg(dsl_dataset_t *ds)
1992 {
1993 	return (dsl_dataset_phys(ds)->ds_creation_txg);
1994 }
1995 
1996 uint64_t
1997 dsl_get_refquota(dsl_dataset_t *ds)
1998 {
1999 	return (ds->ds_quota);
2000 }
2001 
2002 uint64_t
2003 dsl_get_refreservation(dsl_dataset_t *ds)
2004 {
2005 	return (ds->ds_reserved);
2006 }
2007 
2008 uint64_t
2009 dsl_get_guid(dsl_dataset_t *ds)
2010 {
2011 	return (dsl_dataset_phys(ds)->ds_guid);
2012 }
2013 
2014 uint64_t
2015 dsl_get_unique(dsl_dataset_t *ds)
2016 {
2017 	return (dsl_dataset_phys(ds)->ds_unique_bytes);
2018 }
2019 
2020 uint64_t
2021 dsl_get_objsetid(dsl_dataset_t *ds)
2022 {
2023 	return (ds->ds_object);
2024 }
2025 
2026 uint64_t
2027 dsl_get_userrefs(dsl_dataset_t *ds)
2028 {
2029 	return (ds->ds_userrefs);
2030 }
2031 
2032 uint64_t
2033 dsl_get_defer_destroy(dsl_dataset_t *ds)
2034 {
2035 	return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2036 }
2037 
2038 uint64_t
2039 dsl_get_referenced(dsl_dataset_t *ds)
2040 {
2041 	return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2042 }
2043 
2044 uint64_t
2045 dsl_get_numclones(dsl_dataset_t *ds)
2046 {
2047 	ASSERT(ds->ds_is_snapshot);
2048 	return (dsl_dataset_phys(ds)->ds_num_children - 1);
2049 }
2050 
2051 uint64_t
2052 dsl_get_inconsistent(dsl_dataset_t *ds)
2053 {
2054 	return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2055 	    1 : 0);
2056 }
2057 
2058 uint64_t
2059 dsl_get_available(dsl_dataset_t *ds)
2060 {
2061 	uint64_t refdbytes = dsl_get_referenced(ds);
2062 	uint64_t availbytes = dsl_dir_space_available(ds->ds_dir,
2063 	    NULL, 0, TRUE);
2064 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2065 		availbytes +=
2066 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2067 	}
2068 	if (ds->ds_quota != 0) {
2069 		/*
2070 		 * Adjust available bytes according to refquota
2071 		 */
2072 		if (refdbytes < ds->ds_quota) {
2073 			availbytes = MIN(availbytes,
2074 			    ds->ds_quota - refdbytes);
2075 		} else {
2076 			availbytes = 0;
2077 		}
2078 	}
2079 	return (availbytes);
2080 }
2081 
2082 int
2083 dsl_get_written(dsl_dataset_t *ds, uint64_t *written)
2084 {
2085 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2086 	dsl_dataset_t *prev;
2087 	int err = dsl_dataset_hold_obj(dp,
2088 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2089 	if (err == 0) {
2090 		uint64_t comp, uncomp;
2091 		err = dsl_dataset_space_written(prev, ds, written,
2092 		    &comp, &uncomp);
2093 		dsl_dataset_rele(prev, FTAG);
2094 	}
2095 	return (err);
2096 }
2097 
2098 /*
2099  * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2100  */
2101 int
2102 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap)
2103 {
2104 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2105 	if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
2106 		dsl_dataset_name(ds->ds_prev, snap);
2107 		return (0);
2108 	} else {
2109 		return (ENOENT);
2110 	}
2111 }
2112 
2113 /*
2114  * Returns the mountpoint property and source for the given dataset in the value
2115  * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2116  * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2117  * Returns 0 on success and an error on failure.
2118  */
2119 int
2120 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value,
2121     char *source)
2122 {
2123 	int error;
2124 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2125 
2126 	/* Retrieve the mountpoint value stored in the zap opbject */
2127 	error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1,
2128 	    ZAP_MAXVALUELEN, value, source);
2129 	if (error != 0) {
2130 		return (error);
2131 	}
2132 
2133 	/*
2134 	 * Process the dsname and source to find the full mountpoint string.
2135 	 * Can be skipped for 'legacy' or 'none'.
2136 	 */
2137 	if (value[0] == '/') {
2138 		char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
2139 		char *root = buf;
2140 		const char *relpath;
2141 
2142 		/*
2143 		 * If we inherit the mountpoint, even from a dataset
2144 		 * with a received value, the source will be the path of
2145 		 * the dataset we inherit from. If source is
2146 		 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2147 		 * inherited.
2148 		 */
2149 		if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) {
2150 			relpath = "";
2151 		} else {
2152 			ASSERT0(strncmp(dsname, source, strlen(source)));
2153 			relpath = dsname + strlen(source);
2154 			if (relpath[0] == '/')
2155 				relpath++;
2156 		}
2157 
2158 		spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN);
2159 
2160 		/*
2161 		 * Special case an alternate root of '/'. This will
2162 		 * avoid having multiple leading slashes in the
2163 		 * mountpoint path.
2164 		 */
2165 		if (strcmp(root, "/") == 0)
2166 			root++;
2167 
2168 		/*
2169 		 * If the mountpoint is '/' then skip over this
2170 		 * if we are obtaining either an alternate root or
2171 		 * an inherited mountpoint.
2172 		 */
2173 		char *mnt = value;
2174 		if (value[1] == '\0' && (root[0] != '\0' ||
2175 		    relpath[0] != '\0'))
2176 			mnt = value + 1;
2177 
2178 		if (relpath[0] == '\0') {
2179 			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s",
2180 			    root, mnt);
2181 		} else {
2182 			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s",
2183 			    root, mnt, relpath[0] == '@' ? "" : "/",
2184 			    relpath);
2185 		}
2186 		kmem_free(buf, ZAP_MAXVALUELEN);
2187 	}
2188 
2189 	return (0);
2190 }
2191 
2192 void
2193 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2194 {
2195 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2196 
2197 	ASSERT(dsl_pool_config_held(dp));
2198 
2199 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO,
2200 	    dsl_get_refratio(ds));
2201 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
2202 	    dsl_get_logicalreferenced(ds));
2203 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2204 	    dsl_get_compressratio(ds));
2205 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2206 	    dsl_get_used(ds));
2207 
2208 	if (ds->ds_is_snapshot) {
2209 		get_clones_stat(ds, nv);
2210 	} else {
2211 		char buf[ZFS_MAX_DATASET_NAME_LEN];
2212 		if (dsl_get_prev_snap(ds, buf) == 0)
2213 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP,
2214 			    buf);
2215 		dsl_dir_stats(ds->ds_dir, nv);
2216 	}
2217 
2218 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE,
2219 	    dsl_get_available(ds));
2220 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
2221 	    dsl_get_referenced(ds));
2222 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2223 	    dsl_get_creation(ds));
2224 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2225 	    dsl_get_creationtxg(ds));
2226 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2227 	    dsl_get_refquota(ds));
2228 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2229 	    dsl_get_refreservation(ds));
2230 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2231 	    dsl_get_guid(ds));
2232 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2233 	    dsl_get_unique(ds));
2234 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2235 	    dsl_get_objsetid(ds));
2236 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2237 	    dsl_get_userrefs(ds));
2238 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2239 	    dsl_get_defer_destroy(ds));
2240 
2241 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
2242 		uint64_t written;
2243 		if (dsl_get_written(ds, &written) == 0) {
2244 			dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2245 			    written);
2246 		}
2247 	}
2248 
2249 	if (!dsl_dataset_is_snapshot(ds)) {
2250 		/*
2251 		 * A failed "newfs" (e.g. full) resumable receive leaves
2252 		 * the stats set on this dataset.  Check here for the prop.
2253 		 */
2254 		get_receive_resume_stats(ds, nv);
2255 
2256 		/*
2257 		 * A failed incremental resumable receive leaves the
2258 		 * stats set on our child named "%recv".  Check the child
2259 		 * for the prop.
2260 		 */
2261 		/* 6 extra bytes for /%recv */
2262 		char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
2263 		dsl_dataset_t *recv_ds;
2264 		dsl_dataset_name(ds, recvname);
2265 		if (strlcat(recvname, "/", sizeof (recvname)) <
2266 		    sizeof (recvname) &&
2267 		    strlcat(recvname, recv_clone_name, sizeof (recvname)) <
2268 		    sizeof (recvname) &&
2269 		    dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
2270 			get_receive_resume_stats(recv_ds, nv);
2271 			dsl_dataset_rele(recv_ds, FTAG);
2272 		}
2273 	}
2274 }
2275 
2276 void
2277 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2278 {
2279 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2280 	ASSERT(dsl_pool_config_held(dp));
2281 
2282 	stat->dds_creation_txg = dsl_get_creationtxg(ds);
2283 	stat->dds_inconsistent = dsl_get_inconsistent(ds);
2284 	stat->dds_guid = dsl_get_guid(ds);
2285 	stat->dds_origin[0] = '\0';
2286 	if (ds->ds_is_snapshot) {
2287 		stat->dds_is_snapshot = B_TRUE;
2288 		stat->dds_num_clones = dsl_get_numclones(ds);
2289 	} else {
2290 		stat->dds_is_snapshot = B_FALSE;
2291 		stat->dds_num_clones = 0;
2292 
2293 		if (dsl_dir_is_clone(ds->ds_dir)) {
2294 			dsl_dir_get_origin(ds->ds_dir, stat->dds_origin);
2295 		}
2296 	}
2297 }
2298 
2299 uint64_t
2300 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2301 {
2302 	return (ds->ds_fsid_guid);
2303 }
2304 
2305 void
2306 dsl_dataset_space(dsl_dataset_t *ds,
2307     uint64_t *refdbytesp, uint64_t *availbytesp,
2308     uint64_t *usedobjsp, uint64_t *availobjsp)
2309 {
2310 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2311 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2312 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2313 		*availbytesp +=
2314 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2315 	if (ds->ds_quota != 0) {
2316 		/*
2317 		 * Adjust available bytes according to refquota
2318 		 */
2319 		if (*refdbytesp < ds->ds_quota)
2320 			*availbytesp = MIN(*availbytesp,
2321 			    ds->ds_quota - *refdbytesp);
2322 		else
2323 			*availbytesp = 0;
2324 	}
2325 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2326 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2327 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2328 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2329 }
2330 
2331 boolean_t
2332 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2333 {
2334 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2335 	uint64_t birth;
2336 
2337 	ASSERT(dsl_pool_config_held(dp));
2338 	if (snap == NULL)
2339 		return (B_FALSE);
2340 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2341 	birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2342 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2343 	if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2344 		objset_t *os, *os_snap;
2345 		/*
2346 		 * It may be that only the ZIL differs, because it was
2347 		 * reset in the head.  Don't count that as being
2348 		 * modified.
2349 		 */
2350 		if (dmu_objset_from_ds(ds, &os) != 0)
2351 			return (B_TRUE);
2352 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
2353 			return (B_TRUE);
2354 		return (bcmp(&os->os_phys->os_meta_dnode,
2355 		    &os_snap->os_phys->os_meta_dnode,
2356 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
2357 	}
2358 	return (B_FALSE);
2359 }
2360 
2361 typedef struct dsl_dataset_rename_snapshot_arg {
2362 	const char *ddrsa_fsname;
2363 	const char *ddrsa_oldsnapname;
2364 	const char *ddrsa_newsnapname;
2365 	boolean_t ddrsa_recursive;
2366 	dmu_tx_t *ddrsa_tx;
2367 } dsl_dataset_rename_snapshot_arg_t;
2368 
2369 /* ARGSUSED */
2370 static int
2371 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2372     dsl_dataset_t *hds, void *arg)
2373 {
2374 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2375 	int error;
2376 	uint64_t val;
2377 
2378 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2379 	if (error != 0) {
2380 		/* ignore nonexistent snapshots */
2381 		return (error == ENOENT ? 0 : error);
2382 	}
2383 
2384 	/* new name should not exist */
2385 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2386 	if (error == 0)
2387 		error = SET_ERROR(EEXIST);
2388 	else if (error == ENOENT)
2389 		error = 0;
2390 
2391 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2392 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
2393 	    strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2394 		error = SET_ERROR(ENAMETOOLONG);
2395 
2396 	return (error);
2397 }
2398 
2399 static int
2400 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2401 {
2402 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2403 	dsl_pool_t *dp = dmu_tx_pool(tx);
2404 	dsl_dataset_t *hds;
2405 	int error;
2406 
2407 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2408 	if (error != 0)
2409 		return (error);
2410 
2411 	if (ddrsa->ddrsa_recursive) {
2412 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2413 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2414 		    DS_FIND_CHILDREN);
2415 	} else {
2416 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2417 	}
2418 	dsl_dataset_rele(hds, FTAG);
2419 	return (error);
2420 }
2421 
2422 static int
2423 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2424     dsl_dataset_t *hds, void *arg)
2425 {
2426 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2427 	dsl_dataset_t *ds;
2428 	uint64_t val;
2429 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2430 	int error;
2431 
2432 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2433 	ASSERT(error == 0 || error == ENOENT);
2434 	if (error == ENOENT) {
2435 		/* ignore nonexistent snapshots */
2436 		return (0);
2437 	}
2438 
2439 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2440 
2441 	/* log before we change the name */
2442 	spa_history_log_internal_ds(ds, "rename", tx,
2443 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2444 
2445 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2446 	    B_FALSE));
2447 	mutex_enter(&ds->ds_lock);
2448 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2449 	mutex_exit(&ds->ds_lock);
2450 	VERIFY0(zap_add(dp->dp_meta_objset,
2451 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2452 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2453 
2454 	dsl_dataset_rele(ds, FTAG);
2455 	return (0);
2456 }
2457 
2458 static void
2459 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2460 {
2461 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2462 	dsl_pool_t *dp = dmu_tx_pool(tx);
2463 	dsl_dataset_t *hds;
2464 
2465 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2466 	ddrsa->ddrsa_tx = tx;
2467 	if (ddrsa->ddrsa_recursive) {
2468 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2469 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2470 		    DS_FIND_CHILDREN));
2471 	} else {
2472 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2473 	}
2474 	dsl_dataset_rele(hds, FTAG);
2475 }
2476 
2477 int
2478 dsl_dataset_rename_snapshot(const char *fsname,
2479     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2480 {
2481 	dsl_dataset_rename_snapshot_arg_t ddrsa;
2482 
2483 	ddrsa.ddrsa_fsname = fsname;
2484 	ddrsa.ddrsa_oldsnapname = oldsnapname;
2485 	ddrsa.ddrsa_newsnapname = newsnapname;
2486 	ddrsa.ddrsa_recursive = recursive;
2487 
2488 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2489 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2490 	    1, ZFS_SPACE_CHECK_RESERVED));
2491 }
2492 
2493 /*
2494  * If we're doing an ownership handoff, we need to make sure that there is
2495  * only one long hold on the dataset.  We're not allowed to change anything here
2496  * so we don't permanently release the long hold or regular hold here.  We want
2497  * to do this only when syncing to avoid the dataset unexpectedly going away
2498  * when we release the long hold.
2499  */
2500 static int
2501 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2502 {
2503 	boolean_t held;
2504 
2505 	if (!dmu_tx_is_syncing(tx))
2506 		return (0);
2507 
2508 	if (owner != NULL) {
2509 		VERIFY3P(ds->ds_owner, ==, owner);
2510 		dsl_dataset_long_rele(ds, owner);
2511 	}
2512 
2513 	held = dsl_dataset_long_held(ds);
2514 
2515 	if (owner != NULL)
2516 		dsl_dataset_long_hold(ds, owner);
2517 
2518 	if (held)
2519 		return (SET_ERROR(EBUSY));
2520 
2521 	return (0);
2522 }
2523 
2524 int
2525 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2526 {
2527 	dsl_dataset_rollback_arg_t *ddra = arg;
2528 	dsl_pool_t *dp = dmu_tx_pool(tx);
2529 	dsl_dataset_t *ds;
2530 	int64_t unused_refres_delta;
2531 	int error;
2532 
2533 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2534 	if (error != 0)
2535 		return (error);
2536 
2537 	/* must not be a snapshot */
2538 	if (ds->ds_is_snapshot) {
2539 		dsl_dataset_rele(ds, FTAG);
2540 		return (SET_ERROR(EINVAL));
2541 	}
2542 
2543 	/* must have a most recent snapshot */
2544 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2545 		dsl_dataset_rele(ds, FTAG);
2546 		return (SET_ERROR(ESRCH));
2547 	}
2548 
2549 	/*
2550 	 * No rollback to a snapshot created in the current txg, because
2551 	 * the rollback may dirty the dataset and create blocks that are
2552 	 * not reachable from the rootbp while having a birth txg that
2553 	 * falls into the snapshot's range.
2554 	 */
2555 	if (dmu_tx_is_syncing(tx) &&
2556 	    dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2557 		dsl_dataset_rele(ds, FTAG);
2558 		return (SET_ERROR(EAGAIN));
2559 	}
2560 
2561 	/*
2562 	 * If the expected target snapshot is specified, then check that
2563 	 * the latest snapshot is it.
2564 	 */
2565 	if (ddra->ddra_tosnap != NULL) {
2566 		dsl_dataset_t *snapds;
2567 
2568 		/* Check if the target snapshot exists at all. */
2569 		error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
2570 		if (error != 0) {
2571 			/*
2572 			 * ESRCH is used to signal that the target snapshot does
2573 			 * not exist, while ENOENT is used to report that
2574 			 * the rolled back dataset does not exist.
2575 			 * ESRCH is also used to cover other cases where the
2576 			 * target snapshot is not related to the dataset being
2577 			 * rolled back such as being in a different pool.
2578 			 */
2579 			if (error == ENOENT || error == EXDEV)
2580 				error = SET_ERROR(ESRCH);
2581 			dsl_dataset_rele(ds, FTAG);
2582 			return (error);
2583 		}
2584 		ASSERT(snapds->ds_is_snapshot);
2585 
2586 		/* Check if the snapshot is the latest snapshot indeed. */
2587 		if (snapds != ds->ds_prev) {
2588 			/*
2589 			 * Distinguish between the case where the only problem
2590 			 * is intervening snapshots (EEXIST) vs the snapshot
2591 			 * not being a valid target for rollback (ESRCH).
2592 			 */
2593 			if (snapds->ds_dir == ds->ds_dir ||
2594 			    (dsl_dir_is_clone(ds->ds_dir) &&
2595 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
2596 			    snapds->ds_object)) {
2597 				error = SET_ERROR(EEXIST);
2598 			} else {
2599 				error = SET_ERROR(ESRCH);
2600 			}
2601 			dsl_dataset_rele(snapds, FTAG);
2602 			dsl_dataset_rele(ds, FTAG);
2603 			return (error);
2604 		}
2605 		dsl_dataset_rele(snapds, FTAG);
2606 	}
2607 
2608 	/* must not have any bookmarks after the most recent snapshot */
2609 	nvlist_t *proprequest = fnvlist_alloc();
2610 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2611 	nvlist_t *bookmarks = fnvlist_alloc();
2612 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2613 	fnvlist_free(proprequest);
2614 	if (error != 0) {
2615 		dsl_dataset_rele(ds, FTAG);
2616 		return (error);
2617 	}
2618 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2619 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2620 		nvlist_t *valuenv =
2621 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2622 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2623 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2624 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2625 			fnvlist_free(bookmarks);
2626 			dsl_dataset_rele(ds, FTAG);
2627 			return (SET_ERROR(EEXIST));
2628 		}
2629 	}
2630 	fnvlist_free(bookmarks);
2631 
2632 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2633 	if (error != 0) {
2634 		dsl_dataset_rele(ds, FTAG);
2635 		return (error);
2636 	}
2637 
2638 	/*
2639 	 * Check if the snap we are rolling back to uses more than
2640 	 * the refquota.
2641 	 */
2642 	if (ds->ds_quota != 0 &&
2643 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2644 		dsl_dataset_rele(ds, FTAG);
2645 		return (SET_ERROR(EDQUOT));
2646 	}
2647 
2648 	/*
2649 	 * When we do the clone swap, we will temporarily use more space
2650 	 * due to the refreservation (the head will no longer have any
2651 	 * unique space, so the entire amount of the refreservation will need
2652 	 * to be free).  We will immediately destroy the clone, freeing
2653 	 * this space, but the freeing happens over many txg's.
2654 	 */
2655 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2656 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2657 
2658 	if (unused_refres_delta > 0 &&
2659 	    unused_refres_delta >
2660 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2661 		dsl_dataset_rele(ds, FTAG);
2662 		return (SET_ERROR(ENOSPC));
2663 	}
2664 
2665 	dsl_dataset_rele(ds, FTAG);
2666 	return (0);
2667 }
2668 
2669 void
2670 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2671 {
2672 	dsl_dataset_rollback_arg_t *ddra = arg;
2673 	dsl_pool_t *dp = dmu_tx_pool(tx);
2674 	dsl_dataset_t *ds, *clone;
2675 	uint64_t cloneobj;
2676 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2677 
2678 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2679 
2680 	dsl_dataset_name(ds->ds_prev, namebuf);
2681 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2682 
2683 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2684 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2685 
2686 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2687 
2688 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2689 	dsl_dataset_zero_zil(ds, tx);
2690 
2691 	dsl_destroy_head_sync_impl(clone, tx);
2692 
2693 	dsl_dataset_rele(clone, FTAG);
2694 	dsl_dataset_rele(ds, FTAG);
2695 }
2696 
2697 /*
2698  * Rolls back the given filesystem or volume to the most recent snapshot.
2699  * The name of the most recent snapshot will be returned under key "target"
2700  * in the result nvlist.
2701  *
2702  * If owner != NULL:
2703  * - The existing dataset MUST be owned by the specified owner at entry
2704  * - Upon return, dataset will still be held by the same owner, whether we
2705  *   succeed or not.
2706  *
2707  * This mode is required any time the existing filesystem is mounted.  See
2708  * notes above zfs_suspend_fs() for further details.
2709  */
2710 int
2711 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
2712     nvlist_t *result)
2713 {
2714 	dsl_dataset_rollback_arg_t ddra;
2715 
2716 	ddra.ddra_fsname = fsname;
2717 	ddra.ddra_tosnap = tosnap;
2718 	ddra.ddra_owner = owner;
2719 	ddra.ddra_result = result;
2720 
2721 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2722 	    dsl_dataset_rollback_sync, &ddra,
2723 	    1, ZFS_SPACE_CHECK_RESERVED));
2724 }
2725 
2726 struct promotenode {
2727 	list_node_t link;
2728 	dsl_dataset_t *ds;
2729 };
2730 
2731 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2732 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2733     void *tag);
2734 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2735 
2736 int
2737 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2738 {
2739 	dsl_dataset_promote_arg_t *ddpa = arg;
2740 	dsl_pool_t *dp = dmu_tx_pool(tx);
2741 	dsl_dataset_t *hds;
2742 	struct promotenode *snap;
2743 	dsl_dataset_t *origin_ds;
2744 	int err;
2745 	uint64_t unused;
2746 	uint64_t ss_mv_cnt;
2747 	size_t max_snap_len;
2748 	boolean_t conflicting_snaps;
2749 
2750 	err = promote_hold(ddpa, dp, FTAG);
2751 	if (err != 0)
2752 		return (err);
2753 
2754 	hds = ddpa->ddpa_clone;
2755 	snap = list_head(&ddpa->shared_snaps);
2756 	origin_ds = snap->ds;
2757 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2758 
2759 	snap = list_head(&ddpa->origin_snaps);
2760 
2761 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2762 		promote_rele(ddpa, FTAG);
2763 		return (SET_ERROR(EXDEV));
2764 	}
2765 
2766 	/*
2767 	 * Compute and check the amount of space to transfer.  Since this is
2768 	 * so expensive, don't do the preliminary check.
2769 	 */
2770 	if (!dmu_tx_is_syncing(tx)) {
2771 		promote_rele(ddpa, FTAG);
2772 		return (0);
2773 	}
2774 
2775 	/* compute origin's new unique space */
2776 	snap = list_tail(&ddpa->clone_snaps);
2777 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2778 	    origin_ds->ds_object);
2779 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2780 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2781 	    &ddpa->unique, &unused, &unused);
2782 
2783 	/*
2784 	 * Walk the snapshots that we are moving
2785 	 *
2786 	 * Compute space to transfer.  Consider the incremental changes
2787 	 * to used by each snapshot:
2788 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2789 	 * So each snapshot gave birth to:
2790 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2791 	 * So a sequence would look like:
2792 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2793 	 * Which simplifies to:
2794 	 * uN + kN + kN-1 + ... + k1 + k0
2795 	 * Note however, if we stop before we reach the ORIGIN we get:
2796 	 * uN + kN + kN-1 + ... + kM - uM-1
2797 	 */
2798 	conflicting_snaps = B_FALSE;
2799 	ss_mv_cnt = 0;
2800 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2801 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2802 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2803 	for (snap = list_head(&ddpa->shared_snaps); snap;
2804 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2805 		uint64_t val, dlused, dlcomp, dluncomp;
2806 		dsl_dataset_t *ds = snap->ds;
2807 
2808 		ss_mv_cnt++;
2809 
2810 		/*
2811 		 * If there are long holds, we won't be able to evict
2812 		 * the objset.
2813 		 */
2814 		if (dsl_dataset_long_held(ds)) {
2815 			err = SET_ERROR(EBUSY);
2816 			goto out;
2817 		}
2818 
2819 		/* Check that the snapshot name does not conflict */
2820 		VERIFY0(dsl_dataset_get_snapname(ds));
2821 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2822 			err = SET_ERROR(ENAMETOOLONG);
2823 			goto out;
2824 		}
2825 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2826 		if (err == 0) {
2827 			fnvlist_add_boolean(ddpa->err_ds,
2828 			    snap->ds->ds_snapname);
2829 			conflicting_snaps = B_TRUE;
2830 		} else if (err != ENOENT) {
2831 			goto out;
2832 		}
2833 
2834 		/* The very first snapshot does not have a deadlist */
2835 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2836 			continue;
2837 
2838 		dsl_deadlist_space(&ds->ds_deadlist,
2839 		    &dlused, &dlcomp, &dluncomp);
2840 		ddpa->used += dlused;
2841 		ddpa->comp += dlcomp;
2842 		ddpa->uncomp += dluncomp;
2843 	}
2844 
2845 	/*
2846 	 * In order to return the full list of conflicting snapshots, we check
2847 	 * whether there was a conflict after traversing all of them.
2848 	 */
2849 	if (conflicting_snaps) {
2850 		err = SET_ERROR(EEXIST);
2851 		goto out;
2852 	}
2853 
2854 	/*
2855 	 * If we are a clone of a clone then we never reached ORIGIN,
2856 	 * so we need to subtract out the clone origin's used space.
2857 	 */
2858 	if (ddpa->origin_origin) {
2859 		ddpa->used -=
2860 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2861 		ddpa->comp -=
2862 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2863 		ddpa->uncomp -=
2864 		    dsl_dataset_phys(ddpa->origin_origin)->
2865 		    ds_uncompressed_bytes;
2866 	}
2867 
2868 	/* Check that there is enough space and limit headroom here */
2869 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2870 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
2871 	if (err != 0)
2872 		goto out;
2873 
2874 	/*
2875 	 * Compute the amounts of space that will be used by snapshots
2876 	 * after the promotion (for both origin and clone).  For each,
2877 	 * it is the amount of space that will be on all of their
2878 	 * deadlists (that was not born before their new origin).
2879 	 */
2880 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2881 		uint64_t space;
2882 
2883 		/*
2884 		 * Note, typically this will not be a clone of a clone,
2885 		 * so dd_origin_txg will be < TXG_INITIAL, so
2886 		 * these snaplist_space() -> dsl_deadlist_space_range()
2887 		 * calls will be fast because they do not have to
2888 		 * iterate over all bps.
2889 		 */
2890 		snap = list_head(&ddpa->origin_snaps);
2891 		err = snaplist_space(&ddpa->shared_snaps,
2892 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2893 		if (err != 0)
2894 			goto out;
2895 
2896 		err = snaplist_space(&ddpa->clone_snaps,
2897 		    snap->ds->ds_dir->dd_origin_txg, &space);
2898 		if (err != 0)
2899 			goto out;
2900 		ddpa->cloneusedsnap += space;
2901 	}
2902 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2903 	    DD_FLAG_USED_BREAKDOWN) {
2904 		err = snaplist_space(&ddpa->origin_snaps,
2905 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
2906 		    &ddpa->originusedsnap);
2907 		if (err != 0)
2908 			goto out;
2909 	}
2910 
2911 out:
2912 	promote_rele(ddpa, FTAG);
2913 	return (err);
2914 }
2915 
2916 void
2917 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2918 {
2919 	dsl_dataset_promote_arg_t *ddpa = arg;
2920 	dsl_pool_t *dp = dmu_tx_pool(tx);
2921 	dsl_dataset_t *hds;
2922 	struct promotenode *snap;
2923 	dsl_dataset_t *origin_ds;
2924 	dsl_dataset_t *origin_head;
2925 	dsl_dir_t *dd;
2926 	dsl_dir_t *odd = NULL;
2927 	uint64_t oldnext_obj;
2928 	int64_t delta;
2929 
2930 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2931 	hds = ddpa->ddpa_clone;
2932 
2933 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2934 
2935 	snap = list_head(&ddpa->shared_snaps);
2936 	origin_ds = snap->ds;
2937 	dd = hds->ds_dir;
2938 
2939 	snap = list_head(&ddpa->origin_snaps);
2940 	origin_head = snap->ds;
2941 
2942 	/*
2943 	 * We need to explicitly open odd, since origin_ds's dd will be
2944 	 * changing.
2945 	 */
2946 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2947 	    NULL, FTAG, &odd));
2948 
2949 	/* change origin's next snap */
2950 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2951 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2952 	snap = list_tail(&ddpa->clone_snaps);
2953 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2954 	    origin_ds->ds_object);
2955 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2956 
2957 	/* change the origin's next clone */
2958 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2959 		dsl_dataset_remove_from_next_clones(origin_ds,
2960 		    snap->ds->ds_object, tx);
2961 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2962 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2963 		    oldnext_obj, tx));
2964 	}
2965 
2966 	/* change origin */
2967 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2968 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2969 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2970 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2971 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2972 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2973 	origin_head->ds_dir->dd_origin_txg =
2974 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
2975 
2976 	/* change dd_clone entries */
2977 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2978 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2979 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2980 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2981 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2982 		    hds->ds_object, tx));
2983 
2984 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2985 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2986 		    origin_head->ds_object, tx));
2987 		if (dsl_dir_phys(dd)->dd_clones == 0) {
2988 			dsl_dir_phys(dd)->dd_clones =
2989 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2990 			    DMU_OT_NONE, 0, tx);
2991 		}
2992 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2993 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2994 	}
2995 
2996 	/* move snapshots to this dir */
2997 	for (snap = list_head(&ddpa->shared_snaps); snap;
2998 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2999 		dsl_dataset_t *ds = snap->ds;
3000 
3001 		/*
3002 		 * Property callbacks are registered to a particular
3003 		 * dsl_dir.  Since ours is changing, evict the objset
3004 		 * so that they will be unregistered from the old dsl_dir.
3005 		 */
3006 		if (ds->ds_objset) {
3007 			dmu_objset_evict(ds->ds_objset);
3008 			ds->ds_objset = NULL;
3009 		}
3010 
3011 		/* move snap name entry */
3012 		VERIFY0(dsl_dataset_get_snapname(ds));
3013 		VERIFY0(dsl_dataset_snap_remove(origin_head,
3014 		    ds->ds_snapname, tx, B_TRUE));
3015 		VERIFY0(zap_add(dp->dp_meta_objset,
3016 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3017 		    8, 1, &ds->ds_object, tx));
3018 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3019 		    DD_FIELD_SNAPSHOT_COUNT, tx);
3020 
3021 		/* change containing dsl_dir */
3022 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3023 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3024 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3025 		ASSERT3P(ds->ds_dir, ==, odd);
3026 		dsl_dir_rele(ds->ds_dir, ds);
3027 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3028 		    NULL, ds, &ds->ds_dir));
3029 
3030 		/* move any clone references */
3031 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3032 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3033 			zap_cursor_t zc;
3034 			zap_attribute_t za;
3035 
3036 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
3037 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
3038 			    zap_cursor_retrieve(&zc, &za) == 0;
3039 			    zap_cursor_advance(&zc)) {
3040 				dsl_dataset_t *cnds;
3041 				uint64_t o;
3042 
3043 				if (za.za_first_integer == oldnext_obj) {
3044 					/*
3045 					 * We've already moved the
3046 					 * origin's reference.
3047 					 */
3048 					continue;
3049 				}
3050 
3051 				VERIFY0(dsl_dataset_hold_obj(dp,
3052 				    za.za_first_integer, FTAG, &cnds));
3053 				o = dsl_dir_phys(cnds->ds_dir)->
3054 				    dd_head_dataset_obj;
3055 
3056 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
3057 				    dsl_dir_phys(odd)->dd_clones, o, tx));
3058 				VERIFY0(zap_add_int(dp->dp_meta_objset,
3059 				    dsl_dir_phys(dd)->dd_clones, o, tx));
3060 				dsl_dataset_rele(cnds, FTAG);
3061 			}
3062 			zap_cursor_fini(&zc);
3063 		}
3064 
3065 		ASSERT(!dsl_prop_hascb(ds));
3066 	}
3067 
3068 	/*
3069 	 * Change space accounting.
3070 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3071 	 * both be valid, or both be 0 (resulting in delta == 0).  This
3072 	 * is true for each of {clone,origin} independently.
3073 	 */
3074 
3075 	delta = ddpa->cloneusedsnap -
3076 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3077 	ASSERT3S(delta, >=, 0);
3078 	ASSERT3U(ddpa->used, >=, delta);
3079 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3080 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
3081 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3082 
3083 	delta = ddpa->originusedsnap -
3084 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3085 	ASSERT3S(delta, <=, 0);
3086 	ASSERT3U(ddpa->used, >=, -delta);
3087 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3088 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
3089 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3090 
3091 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3092 
3093 	/* log history record */
3094 	spa_history_log_internal_ds(hds, "promote", tx, "");
3095 
3096 	dsl_dir_rele(odd, FTAG);
3097 	promote_rele(ddpa, FTAG);
3098 }
3099 
3100 /*
3101  * Make a list of dsl_dataset_t's for the snapshots between first_obj
3102  * (exclusive) and last_obj (inclusive).  The list will be in reverse
3103  * order (last_obj will be the list_head()).  If first_obj == 0, do all
3104  * snapshots back to this dataset's origin.
3105  */
3106 static int
3107 snaplist_make(dsl_pool_t *dp,
3108     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
3109 {
3110 	uint64_t obj = last_obj;
3111 
3112 	list_create(l, sizeof (struct promotenode),
3113 	    offsetof(struct promotenode, link));
3114 
3115 	while (obj != first_obj) {
3116 		dsl_dataset_t *ds;
3117 		struct promotenode *snap;
3118 		int err;
3119 
3120 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3121 		ASSERT(err != ENOENT);
3122 		if (err != 0)
3123 			return (err);
3124 
3125 		if (first_obj == 0)
3126 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3127 
3128 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3129 		snap->ds = ds;
3130 		list_insert_tail(l, snap);
3131 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3132 	}
3133 
3134 	return (0);
3135 }
3136 
3137 static int
3138 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3139 {
3140 	struct promotenode *snap;
3141 
3142 	*spacep = 0;
3143 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3144 		uint64_t used, comp, uncomp;
3145 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3146 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
3147 		*spacep += used;
3148 	}
3149 	return (0);
3150 }
3151 
3152 static void
3153 snaplist_destroy(list_t *l, void *tag)
3154 {
3155 	struct promotenode *snap;
3156 
3157 	if (l == NULL || !list_link_active(&l->list_head))
3158 		return;
3159 
3160 	while ((snap = list_tail(l)) != NULL) {
3161 		list_remove(l, snap);
3162 		dsl_dataset_rele(snap->ds, tag);
3163 		kmem_free(snap, sizeof (*snap));
3164 	}
3165 	list_destroy(l);
3166 }
3167 
3168 static int
3169 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
3170 {
3171 	int error;
3172 	dsl_dir_t *dd;
3173 	struct promotenode *snap;
3174 
3175 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3176 	    &ddpa->ddpa_clone);
3177 	if (error != 0)
3178 		return (error);
3179 	dd = ddpa->ddpa_clone->ds_dir;
3180 
3181 	if (ddpa->ddpa_clone->ds_is_snapshot ||
3182 	    !dsl_dir_is_clone(dd)) {
3183 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
3184 		return (SET_ERROR(EINVAL));
3185 	}
3186 
3187 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3188 	    &ddpa->shared_snaps, tag);
3189 	if (error != 0)
3190 		goto out;
3191 
3192 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3193 	    &ddpa->clone_snaps, tag);
3194 	if (error != 0)
3195 		goto out;
3196 
3197 	snap = list_head(&ddpa->shared_snaps);
3198 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3199 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3200 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3201 	    &ddpa->origin_snaps, tag);
3202 	if (error != 0)
3203 		goto out;
3204 
3205 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3206 		error = dsl_dataset_hold_obj(dp,
3207 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3208 		    tag, &ddpa->origin_origin);
3209 		if (error != 0)
3210 			goto out;
3211 	}
3212 out:
3213 	if (error != 0)
3214 		promote_rele(ddpa, tag);
3215 	return (error);
3216 }
3217 
3218 static void
3219 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
3220 {
3221 	snaplist_destroy(&ddpa->shared_snaps, tag);
3222 	snaplist_destroy(&ddpa->clone_snaps, tag);
3223 	snaplist_destroy(&ddpa->origin_snaps, tag);
3224 	if (ddpa->origin_origin != NULL)
3225 		dsl_dataset_rele(ddpa->origin_origin, tag);
3226 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
3227 }
3228 
3229 /*
3230  * Promote a clone.
3231  *
3232  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3233  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3234  */
3235 int
3236 dsl_dataset_promote(const char *name, char *conflsnap)
3237 {
3238 	dsl_dataset_promote_arg_t ddpa = { 0 };
3239 	uint64_t numsnaps;
3240 	int error;
3241 	nvpair_t *snap_pair;
3242 	objset_t *os;
3243 
3244 	/*
3245 	 * We will modify space proportional to the number of
3246 	 * snapshots.  Compute numsnaps.
3247 	 */
3248 	error = dmu_objset_hold(name, FTAG, &os);
3249 	if (error != 0)
3250 		return (error);
3251 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3252 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3253 	    &numsnaps);
3254 	dmu_objset_rele(os, FTAG);
3255 	if (error != 0)
3256 		return (error);
3257 
3258 	ddpa.ddpa_clonename = name;
3259 	ddpa.err_ds = fnvlist_alloc();
3260 	ddpa.cr = CRED();
3261 
3262 	error = dsl_sync_task(name, dsl_dataset_promote_check,
3263 	    dsl_dataset_promote_sync, &ddpa,
3264 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3265 
3266 	/*
3267 	 * Return the first conflicting snapshot found.
3268 	 */
3269 	snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3270 	if (snap_pair != NULL && conflsnap != NULL)
3271 		(void) strcpy(conflsnap, nvpair_name(snap_pair));
3272 
3273 	fnvlist_free(ddpa.err_ds);
3274 	return (error);
3275 }
3276 
3277 int
3278 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3279     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3280 {
3281 	/*
3282 	 * "slack" factor for received datasets with refquota set on them.
3283 	 * See the bottom of this function for details on its use.
3284 	 */
3285 	uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
3286 	int64_t unused_refres_delta;
3287 
3288 	/* they should both be heads */
3289 	if (clone->ds_is_snapshot ||
3290 	    origin_head->ds_is_snapshot)
3291 		return (SET_ERROR(EINVAL));
3292 
3293 	/* if we are not forcing, the branch point should be just before them */
3294 	if (!force && clone->ds_prev != origin_head->ds_prev)
3295 		return (SET_ERROR(EINVAL));
3296 
3297 	/* clone should be the clone (unless they are unrelated) */
3298 	if (clone->ds_prev != NULL &&
3299 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3300 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
3301 		return (SET_ERROR(EINVAL));
3302 
3303 	/* the clone should be a child of the origin */
3304 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3305 		return (SET_ERROR(EINVAL));
3306 
3307 	/* origin_head shouldn't be modified unless 'force' */
3308 	if (!force &&
3309 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3310 		return (SET_ERROR(ETXTBSY));
3311 
3312 	/* origin_head should have no long holds (e.g. is not mounted) */
3313 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
3314 		return (SET_ERROR(EBUSY));
3315 
3316 	/* check amount of any unconsumed refreservation */
3317 	unused_refres_delta =
3318 	    (int64_t)MIN(origin_head->ds_reserved,
3319 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3320 	    (int64_t)MIN(origin_head->ds_reserved,
3321 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3322 
3323 	if (unused_refres_delta > 0 &&
3324 	    unused_refres_delta >
3325 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3326 		return (SET_ERROR(ENOSPC));
3327 
3328 	/*
3329 	 * The clone can't be too much over the head's refquota.
3330 	 *
3331 	 * To ensure that the entire refquota can be used, we allow one
3332 	 * transaction to exceed the the refquota.  Therefore, this check
3333 	 * needs to also allow for the space referenced to be more than the
3334 	 * refquota.  The maximum amount of space that one transaction can use
3335 	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
3336 	 * overage ensures that we are able to receive a filesystem that
3337 	 * exceeds the refquota on the source system.
3338 	 *
3339 	 * So that overage is the refquota_slack we use below.
3340 	 */
3341 	if (origin_head->ds_quota != 0 &&
3342 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
3343 	    origin_head->ds_quota + refquota_slack)
3344 		return (SET_ERROR(EDQUOT));
3345 
3346 	return (0);
3347 }
3348 
3349 static void
3350 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3351     dsl_dataset_t *origin, dmu_tx_t *tx)
3352 {
3353 	uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3354 	dsl_pool_t *dp = dmu_tx_pool(tx);
3355 
3356 	ASSERT(dsl_pool_sync_context(dp));
3357 
3358 	clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3359 	origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3360 
3361 	if (clone_remap_dl_obj != 0) {
3362 		dsl_deadlist_close(&clone->ds_remap_deadlist);
3363 		dsl_dataset_unset_remap_deadlist_object(clone, tx);
3364 	}
3365 	if (origin_remap_dl_obj != 0) {
3366 		dsl_deadlist_close(&origin->ds_remap_deadlist);
3367 		dsl_dataset_unset_remap_deadlist_object(origin, tx);
3368 	}
3369 
3370 	if (clone_remap_dl_obj != 0) {
3371 		dsl_dataset_set_remap_deadlist_object(origin,
3372 		    clone_remap_dl_obj, tx);
3373 		dsl_deadlist_open(&origin->ds_remap_deadlist,
3374 		    dp->dp_meta_objset, clone_remap_dl_obj);
3375 	}
3376 	if (origin_remap_dl_obj != 0) {
3377 		dsl_dataset_set_remap_deadlist_object(clone,
3378 		    origin_remap_dl_obj, tx);
3379 		dsl_deadlist_open(&clone->ds_remap_deadlist,
3380 		    dp->dp_meta_objset, origin_remap_dl_obj);
3381 	}
3382 }
3383 
3384 void
3385 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3386     dsl_dataset_t *origin_head, dmu_tx_t *tx)
3387 {
3388 	dsl_pool_t *dp = dmu_tx_pool(tx);
3389 	int64_t unused_refres_delta;
3390 
3391 	ASSERT(clone->ds_reserved == 0);
3392 	/*
3393 	 * NOTE: On DEBUG kernels there could be a race between this and
3394 	 * the check function if spa_asize_inflation is adjusted...
3395 	 */
3396 	ASSERT(origin_head->ds_quota == 0 ||
3397 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3398 	    DMU_MAX_ACCESS * spa_asize_inflation);
3399 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3400 
3401 	/*
3402 	 * Swap per-dataset feature flags.
3403 	 */
3404 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3405 		if (!(spa_feature_table[f].fi_flags &
3406 		    ZFEATURE_FLAG_PER_DATASET)) {
3407 			ASSERT(!clone->ds_feature_inuse[f]);
3408 			ASSERT(!origin_head->ds_feature_inuse[f]);
3409 			continue;
3410 		}
3411 
3412 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
3413 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3414 
3415 		if (clone_inuse) {
3416 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3417 			clone->ds_feature_inuse[f] = B_FALSE;
3418 		}
3419 		if (origin_head_inuse) {
3420 			dsl_dataset_deactivate_feature(origin_head->ds_object,
3421 			    f, tx);
3422 			origin_head->ds_feature_inuse[f] = B_FALSE;
3423 		}
3424 		if (clone_inuse) {
3425 			dsl_dataset_activate_feature(origin_head->ds_object,
3426 			    f, tx);
3427 			origin_head->ds_feature_inuse[f] = B_TRUE;
3428 		}
3429 		if (origin_head_inuse) {
3430 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
3431 			clone->ds_feature_inuse[f] = B_TRUE;
3432 		}
3433 	}
3434 
3435 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
3436 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3437 
3438 	if (clone->ds_objset != NULL) {
3439 		dmu_objset_evict(clone->ds_objset);
3440 		clone->ds_objset = NULL;
3441 	}
3442 
3443 	if (origin_head->ds_objset != NULL) {
3444 		dmu_objset_evict(origin_head->ds_objset);
3445 		origin_head->ds_objset = NULL;
3446 	}
3447 
3448 	unused_refres_delta =
3449 	    (int64_t)MIN(origin_head->ds_reserved,
3450 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3451 	    (int64_t)MIN(origin_head->ds_reserved,
3452 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3453 
3454 	/*
3455 	 * Reset origin's unique bytes, if it exists.
3456 	 */
3457 	if (clone->ds_prev) {
3458 		dsl_dataset_t *origin = clone->ds_prev;
3459 		uint64_t comp, uncomp;
3460 
3461 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
3462 		dsl_deadlist_space_range(&clone->ds_deadlist,
3463 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3464 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3465 	}
3466 
3467 	/* swap blkptrs */
3468 	{
3469 		rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3470 		rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3471 		blkptr_t tmp;
3472 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
3473 		dsl_dataset_phys(origin_head)->ds_bp =
3474 		    dsl_dataset_phys(clone)->ds_bp;
3475 		dsl_dataset_phys(clone)->ds_bp = tmp;
3476 		rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3477 		rrw_exit(&clone->ds_bp_rwlock, FTAG);
3478 	}
3479 
3480 	/* set dd_*_bytes */
3481 	{
3482 		int64_t dused, dcomp, duncomp;
3483 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
3484 		uint64_t odl_used, odl_comp, odl_uncomp;
3485 
3486 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3487 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
3488 
3489 		dsl_deadlist_space(&clone->ds_deadlist,
3490 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3491 		dsl_deadlist_space(&origin_head->ds_deadlist,
3492 		    &odl_used, &odl_comp, &odl_uncomp);
3493 
3494 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3495 		    cdl_used -
3496 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3497 		    odl_used);
3498 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3499 		    cdl_comp -
3500 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3501 		    odl_comp);
3502 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3503 		    cdl_uncomp -
3504 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3505 		    odl_uncomp);
3506 
3507 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3508 		    dused, dcomp, duncomp, tx);
3509 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3510 		    -dused, -dcomp, -duncomp, tx);
3511 
3512 		/*
3513 		 * The difference in the space used by snapshots is the
3514 		 * difference in snapshot space due to the head's
3515 		 * deadlist (since that's the only thing that's
3516 		 * changing that affects the snapused).
3517 		 */
3518 		dsl_deadlist_space_range(&clone->ds_deadlist,
3519 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3520 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3521 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
3522 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3523 		    &odl_used, &odl_comp, &odl_uncomp);
3524 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3525 		    DD_USED_HEAD, DD_USED_SNAP, tx);
3526 	}
3527 
3528 	/* swap ds_*_bytes */
3529 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3530 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
3531 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3532 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
3533 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3534 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3535 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3536 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3537 
3538 	/* apply any parent delta for change in unconsumed refreservation */
3539 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3540 	    unused_refres_delta, 0, 0, tx);
3541 
3542 	/*
3543 	 * Swap deadlists.
3544 	 */
3545 	dsl_deadlist_close(&clone->ds_deadlist);
3546 	dsl_deadlist_close(&origin_head->ds_deadlist);
3547 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3548 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3549 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3550 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3551 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3552 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3553 	dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
3554 
3555 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3556 
3557 	spa_history_log_internal_ds(clone, "clone swap", tx,
3558 	    "parent=%s", origin_head->ds_dir->dd_myname);
3559 }
3560 
3561 /*
3562  * Given a pool name and a dataset object number in that pool,
3563  * return the name of that dataset.
3564  */
3565 int
3566 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3567 {
3568 	dsl_pool_t *dp;
3569 	dsl_dataset_t *ds;
3570 	int error;
3571 
3572 	error = dsl_pool_hold(pname, FTAG, &dp);
3573 	if (error != 0)
3574 		return (error);
3575 
3576 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3577 	if (error == 0) {
3578 		dsl_dataset_name(ds, buf);
3579 		dsl_dataset_rele(ds, FTAG);
3580 	}
3581 	dsl_pool_rele(dp, FTAG);
3582 
3583 	return (error);
3584 }
3585 
3586 int
3587 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3588     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3589 {
3590 	int error = 0;
3591 
3592 	ASSERT3S(asize, >, 0);
3593 
3594 	/*
3595 	 * *ref_rsrv is the portion of asize that will come from any
3596 	 * unconsumed refreservation space.
3597 	 */
3598 	*ref_rsrv = 0;
3599 
3600 	mutex_enter(&ds->ds_lock);
3601 	/*
3602 	 * Make a space adjustment for reserved bytes.
3603 	 */
3604 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3605 		ASSERT3U(*used, >=,
3606 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3607 		*used -=
3608 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3609 		*ref_rsrv =
3610 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3611 	}
3612 
3613 	if (!check_quota || ds->ds_quota == 0) {
3614 		mutex_exit(&ds->ds_lock);
3615 		return (0);
3616 	}
3617 	/*
3618 	 * If they are requesting more space, and our current estimate
3619 	 * is over quota, they get to try again unless the actual
3620 	 * on-disk is over quota and there are no pending changes (which
3621 	 * may free up space for us).
3622 	 */
3623 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3624 	    ds->ds_quota) {
3625 		if (inflight > 0 ||
3626 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3627 			error = SET_ERROR(ERESTART);
3628 		else
3629 			error = SET_ERROR(EDQUOT);
3630 	}
3631 	mutex_exit(&ds->ds_lock);
3632 
3633 	return (error);
3634 }
3635 
3636 typedef struct dsl_dataset_set_qr_arg {
3637 	const char *ddsqra_name;
3638 	zprop_source_t ddsqra_source;
3639 	uint64_t ddsqra_value;
3640 } dsl_dataset_set_qr_arg_t;
3641 
3642 
3643 /* ARGSUSED */
3644 static int
3645 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3646 {
3647 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3648 	dsl_pool_t *dp = dmu_tx_pool(tx);
3649 	dsl_dataset_t *ds;
3650 	int error;
3651 	uint64_t newval;
3652 
3653 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3654 		return (SET_ERROR(ENOTSUP));
3655 
3656 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3657 	if (error != 0)
3658 		return (error);
3659 
3660 	if (ds->ds_is_snapshot) {
3661 		dsl_dataset_rele(ds, FTAG);
3662 		return (SET_ERROR(EINVAL));
3663 	}
3664 
3665 	error = dsl_prop_predict(ds->ds_dir,
3666 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3667 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3668 	if (error != 0) {
3669 		dsl_dataset_rele(ds, FTAG);
3670 		return (error);
3671 	}
3672 
3673 	if (newval == 0) {
3674 		dsl_dataset_rele(ds, FTAG);
3675 		return (0);
3676 	}
3677 
3678 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3679 	    newval < ds->ds_reserved) {
3680 		dsl_dataset_rele(ds, FTAG);
3681 		return (SET_ERROR(ENOSPC));
3682 	}
3683 
3684 	dsl_dataset_rele(ds, FTAG);
3685 	return (0);
3686 }
3687 
3688 static void
3689 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3690 {
3691 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3692 	dsl_pool_t *dp = dmu_tx_pool(tx);
3693 	dsl_dataset_t *ds;
3694 	uint64_t newval;
3695 
3696 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3697 
3698 	dsl_prop_set_sync_impl(ds,
3699 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3700 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3701 	    &ddsqra->ddsqra_value, tx);
3702 
3703 	VERIFY0(dsl_prop_get_int_ds(ds,
3704 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3705 
3706 	if (ds->ds_quota != newval) {
3707 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3708 		ds->ds_quota = newval;
3709 	}
3710 	dsl_dataset_rele(ds, FTAG);
3711 }
3712 
3713 int
3714 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3715     uint64_t refquota)
3716 {
3717 	dsl_dataset_set_qr_arg_t ddsqra;
3718 
3719 	ddsqra.ddsqra_name = dsname;
3720 	ddsqra.ddsqra_source = source;
3721 	ddsqra.ddsqra_value = refquota;
3722 
3723 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3724 	    dsl_dataset_set_refquota_sync, &ddsqra, 0,
3725 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
3726 }
3727 
3728 static int
3729 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3730 {
3731 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3732 	dsl_pool_t *dp = dmu_tx_pool(tx);
3733 	dsl_dataset_t *ds;
3734 	int error;
3735 	uint64_t newval, unique;
3736 
3737 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3738 		return (SET_ERROR(ENOTSUP));
3739 
3740 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3741 	if (error != 0)
3742 		return (error);
3743 
3744 	if (ds->ds_is_snapshot) {
3745 		dsl_dataset_rele(ds, FTAG);
3746 		return (SET_ERROR(EINVAL));
3747 	}
3748 
3749 	error = dsl_prop_predict(ds->ds_dir,
3750 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3751 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3752 	if (error != 0) {
3753 		dsl_dataset_rele(ds, FTAG);
3754 		return (error);
3755 	}
3756 
3757 	/*
3758 	 * If we are doing the preliminary check in open context, the
3759 	 * space estimates may be inaccurate.
3760 	 */
3761 	if (!dmu_tx_is_syncing(tx)) {
3762 		dsl_dataset_rele(ds, FTAG);
3763 		return (0);
3764 	}
3765 
3766 	mutex_enter(&ds->ds_lock);
3767 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3768 		dsl_dataset_recalc_head_uniq(ds);
3769 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3770 	mutex_exit(&ds->ds_lock);
3771 
3772 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3773 		uint64_t delta = MAX(unique, newval) -
3774 		    MAX(unique, ds->ds_reserved);
3775 
3776 		if (delta >
3777 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3778 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3779 			dsl_dataset_rele(ds, FTAG);
3780 			return (SET_ERROR(ENOSPC));
3781 		}
3782 	}
3783 
3784 	dsl_dataset_rele(ds, FTAG);
3785 	return (0);
3786 }
3787 
3788 void
3789 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3790     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3791 {
3792 	uint64_t newval;
3793 	uint64_t unique;
3794 	int64_t delta;
3795 
3796 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3797 	    source, sizeof (value), 1, &value, tx);
3798 
3799 	VERIFY0(dsl_prop_get_int_ds(ds,
3800 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3801 
3802 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3803 	mutex_enter(&ds->ds_dir->dd_lock);
3804 	mutex_enter(&ds->ds_lock);
3805 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3806 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3807 	delta = MAX(0, (int64_t)(newval - unique)) -
3808 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3809 	ds->ds_reserved = newval;
3810 	mutex_exit(&ds->ds_lock);
3811 
3812 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3813 	mutex_exit(&ds->ds_dir->dd_lock);
3814 }
3815 
3816 static void
3817 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3818 {
3819 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3820 	dsl_pool_t *dp = dmu_tx_pool(tx);
3821 	dsl_dataset_t *ds;
3822 
3823 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3824 	dsl_dataset_set_refreservation_sync_impl(ds,
3825 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3826 	dsl_dataset_rele(ds, FTAG);
3827 }
3828 
3829 int
3830 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3831     uint64_t refreservation)
3832 {
3833 	dsl_dataset_set_qr_arg_t ddsqra;
3834 
3835 	ddsqra.ddsqra_name = dsname;
3836 	ddsqra.ddsqra_source = source;
3837 	ddsqra.ddsqra_value = refreservation;
3838 
3839 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3840 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0,
3841 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
3842 }
3843 
3844 /*
3845  * Return (in *usedp) the amount of space written in new that is not
3846  * present in oldsnap.  New may be a snapshot or the head.  Old must be
3847  * a snapshot before new, in new's filesystem (or its origin).  If not then
3848  * fail and return EINVAL.
3849  *
3850  * The written space is calculated by considering two components:  First, we
3851  * ignore any freed space, and calculate the written as new's used space
3852  * minus old's used space.  Next, we add in the amount of space that was freed
3853  * between the two snapshots, thus reducing new's used space relative to old's.
3854  * Specifically, this is the space that was born before old->ds_creation_txg,
3855  * and freed before new (ie. on new's deadlist or a previous deadlist).
3856  *
3857  * space freed                         [---------------------]
3858  * snapshots                       ---O-------O--------O-------O------
3859  *                                         oldsnap            new
3860  */
3861 int
3862 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3863     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3864 {
3865 	int err = 0;
3866 	uint64_t snapobj;
3867 	dsl_pool_t *dp = new->ds_dir->dd_pool;
3868 
3869 	ASSERT(dsl_pool_config_held(dp));
3870 
3871 	*usedp = 0;
3872 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3873 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3874 
3875 	*compp = 0;
3876 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3877 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3878 
3879 	*uncompp = 0;
3880 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3881 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3882 
3883 	snapobj = new->ds_object;
3884 	while (snapobj != oldsnap->ds_object) {
3885 		dsl_dataset_t *snap;
3886 		uint64_t used, comp, uncomp;
3887 
3888 		if (snapobj == new->ds_object) {
3889 			snap = new;
3890 		} else {
3891 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3892 			if (err != 0)
3893 				break;
3894 		}
3895 
3896 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3897 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3898 			/*
3899 			 * The blocks in the deadlist can not be born after
3900 			 * ds_prev_snap_txg, so get the whole deadlist space,
3901 			 * which is more efficient (especially for old-format
3902 			 * deadlists).  Unfortunately the deadlist code
3903 			 * doesn't have enough information to make this
3904 			 * optimization itself.
3905 			 */
3906 			dsl_deadlist_space(&snap->ds_deadlist,
3907 			    &used, &comp, &uncomp);
3908 		} else {
3909 			dsl_deadlist_space_range(&snap->ds_deadlist,
3910 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3911 			    &used, &comp, &uncomp);
3912 		}
3913 		*usedp += used;
3914 		*compp += comp;
3915 		*uncompp += uncomp;
3916 
3917 		/*
3918 		 * If we get to the beginning of the chain of snapshots
3919 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3920 		 * was not a snapshot of/before new.
3921 		 */
3922 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3923 		if (snap != new)
3924 			dsl_dataset_rele(snap, FTAG);
3925 		if (snapobj == 0) {
3926 			err = SET_ERROR(EINVAL);
3927 			break;
3928 		}
3929 
3930 	}
3931 	return (err);
3932 }
3933 
3934 /*
3935  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3936  * lastsnap, and all snapshots in between are deleted.
3937  *
3938  * blocks that would be freed            [---------------------------]
3939  * snapshots                       ---O-------O--------O-------O--------O
3940  *                                        firstsnap        lastsnap
3941  *
3942  * This is the set of blocks that were born after the snap before firstsnap,
3943  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3944  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3945  * We calculate this by iterating over the relevant deadlists (from the snap
3946  * after lastsnap, backward to the snap after firstsnap), summing up the
3947  * space on the deadlist that was born after the snap before firstsnap.
3948  */
3949 int
3950 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3951     dsl_dataset_t *lastsnap,
3952     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3953 {
3954 	int err = 0;
3955 	uint64_t snapobj;
3956 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3957 
3958 	ASSERT(firstsnap->ds_is_snapshot);
3959 	ASSERT(lastsnap->ds_is_snapshot);
3960 
3961 	/*
3962 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
3963 	 * is before lastsnap.
3964 	 */
3965 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
3966 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
3967 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
3968 		return (SET_ERROR(EINVAL));
3969 
3970 	*usedp = *compp = *uncompp = 0;
3971 
3972 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3973 	while (snapobj != firstsnap->ds_object) {
3974 		dsl_dataset_t *ds;
3975 		uint64_t used, comp, uncomp;
3976 
3977 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3978 		if (err != 0)
3979 			break;
3980 
3981 		dsl_deadlist_space_range(&ds->ds_deadlist,
3982 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3983 		    &used, &comp, &uncomp);
3984 		*usedp += used;
3985 		*compp += comp;
3986 		*uncompp += uncomp;
3987 
3988 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3989 		ASSERT3U(snapobj, !=, 0);
3990 		dsl_dataset_rele(ds, FTAG);
3991 	}
3992 	return (err);
3993 }
3994 
3995 /*
3996  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3997  * For example, they could both be snapshots of the same filesystem, and
3998  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
3999  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
4000  * filesystem.  Or 'earlier' could be the origin's origin.
4001  *
4002  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4003  */
4004 boolean_t
4005 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4006     uint64_t earlier_txg)
4007 {
4008 	dsl_pool_t *dp = later->ds_dir->dd_pool;
4009 	int error;
4010 	boolean_t ret;
4011 
4012 	ASSERT(dsl_pool_config_held(dp));
4013 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4014 
4015 	if (earlier_txg == 0)
4016 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4017 
4018 	if (later->ds_is_snapshot &&
4019 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4020 		return (B_FALSE);
4021 
4022 	if (later->ds_dir == earlier->ds_dir)
4023 		return (B_TRUE);
4024 	if (!dsl_dir_is_clone(later->ds_dir))
4025 		return (B_FALSE);
4026 
4027 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
4028 		return (B_TRUE);
4029 	dsl_dataset_t *origin;
4030 	error = dsl_dataset_hold_obj(dp,
4031 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4032 	if (error != 0)
4033 		return (B_FALSE);
4034 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4035 	dsl_dataset_rele(origin, FTAG);
4036 	return (ret);
4037 }
4038 
4039 void
4040 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4041 {
4042 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4043 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4044 }
4045 
4046 boolean_t
4047 dsl_dataset_is_zapified(dsl_dataset_t *ds)
4048 {
4049 	dmu_object_info_t doi;
4050 
4051 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
4052 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4053 }
4054 
4055 boolean_t
4056 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4057 {
4058 	return (dsl_dataset_is_zapified(ds) &&
4059 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4060 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4061 }
4062 
4063 uint64_t
4064 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4065 {
4066 	uint64_t remap_deadlist_obj;
4067 	int err;
4068 
4069 	if (!dsl_dataset_is_zapified(ds))
4070 		return (0);
4071 
4072 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4073 	    DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4074 	    &remap_deadlist_obj);
4075 
4076 	if (err != 0) {
4077 		VERIFY3S(err, ==, ENOENT);
4078 		return (0);
4079 	}
4080 
4081 	ASSERT(remap_deadlist_obj != 0);
4082 	return (remap_deadlist_obj);
4083 }
4084 
4085 boolean_t
4086 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4087 {
4088 	EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4089 	    dsl_dataset_get_remap_deadlist_object(ds) != 0);
4090 	return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4091 }
4092 
4093 static void
4094 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4095     dmu_tx_t *tx)
4096 {
4097 	ASSERT(obj != 0);
4098 	dsl_dataset_zapify(ds, tx);
4099 	VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4100 	    DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4101 }
4102 
4103 static void
4104 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4105 {
4106 	VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4107 	    ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4108 }
4109 
4110 void
4111 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4112 {
4113 	uint64_t remap_deadlist_object;
4114 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4115 
4116 	ASSERT(dmu_tx_is_syncing(tx));
4117 	ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4118 
4119 	remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4120 	dsl_deadlist_close(&ds->ds_remap_deadlist);
4121 	dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4122 	dsl_dataset_unset_remap_deadlist_object(ds, tx);
4123 	spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4124 }
4125 
4126 void
4127 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4128 {
4129 	uint64_t remap_deadlist_obj;
4130 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4131 
4132 	ASSERT(dmu_tx_is_syncing(tx));
4133 	ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4134 	/*
4135 	 * Currently we only create remap deadlists when there are indirect
4136 	 * vdevs with referenced mappings.
4137 	 */
4138 	ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4139 
4140 	remap_deadlist_obj = dsl_deadlist_clone(
4141 	    &ds->ds_deadlist, UINT64_MAX,
4142 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4143 	dsl_dataset_set_remap_deadlist_object(ds,
4144 	    remap_deadlist_obj, tx);
4145 	dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4146 	    remap_deadlist_obj);
4147 	spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4148 }
4149