xref: /freebsd/sys/contrib/openzfs/module/zfs/dsl_dataset.c (revision 1db64f89363c97858961c4df0b7d02f3223723cf)
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 https://opensource.org/licenses/CDDL-1.0.
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, 2020 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) 2016 Actifio, Inc. All rights reserved.
29  * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
30  * Copyright 2017 Nexenta Systems, Inc.
31  * Copyright (c) 2019, Klara Inc.
32  * Copyright (c) 2019, Allan Jude
33  * Copyright (c) 2020 The FreeBSD Foundation [1]
34  *
35  * [1] Portions of this software were developed by Allan Jude
36  *     under sponsorship from the FreeBSD Foundation.
37  */
38 
39 #include <sys/dmu_objset.h>
40 #include <sys/dsl_dataset.h>
41 #include <sys/dsl_dir.h>
42 #include <sys/dsl_prop.h>
43 #include <sys/dsl_synctask.h>
44 #include <sys/dmu_traverse.h>
45 #include <sys/dmu_impl.h>
46 #include <sys/dmu_tx.h>
47 #include <sys/arc.h>
48 #include <sys/zio.h>
49 #include <sys/zap.h>
50 #include <sys/zfeature.h>
51 #include <sys/unique.h>
52 #include <sys/zfs_context.h>
53 #include <sys/zfs_ioctl.h>
54 #include <sys/spa.h>
55 #include <sys/spa_impl.h>
56 #include <sys/vdev.h>
57 #include <sys/zfs_znode.h>
58 #include <sys/zfs_onexit.h>
59 #include <sys/zvol.h>
60 #include <sys/dsl_scan.h>
61 #include <sys/dsl_deadlist.h>
62 #include <sys/dsl_destroy.h>
63 #include <sys/dsl_userhold.h>
64 #include <sys/dsl_bookmark.h>
65 #include <sys/policy.h>
66 #include <sys/dmu_send.h>
67 #include <sys/dmu_recv.h>
68 #include <sys/zio_compress.h>
69 #include <zfs_fletcher.h>
70 #include <sys/zio_checksum.h>
71 
72 /*
73  * The SPA supports block sizes up to 16MB.  However, very large blocks
74  * can have an impact on i/o latency (e.g. tying up a spinning disk for
75  * ~300ms), and also potentially on the memory allocator.  Therefore,
76  * we did not allow the recordsize to be set larger than zfs_max_recordsize
77  * (former default: 1MB).  Larger blocks could be created by changing this
78  * tunable, and pools with larger blocks could always be imported and used,
79  * regardless of this setting.
80  *
81  * We do, however, still limit it by default to 1M on x86_32, because Linux's
82  * 3/1 memory split doesn't leave much room for 16M chunks.
83  */
84 #ifdef _ILP32
85 uint_t zfs_max_recordsize =  1 * 1024 * 1024;
86 #else
87 uint_t zfs_max_recordsize = 16 * 1024 * 1024;
88 #endif
89 static int zfs_allow_redacted_dataset_mount = 0;
90 
91 int zfs_snapshot_history_enabled = 1;
92 
93 #define	SWITCH64(x, y) \
94 	{ \
95 		uint64_t __tmp = (x); \
96 		(x) = (y); \
97 		(y) = __tmp; \
98 	}
99 
100 #define	DS_REF_MAX	(1ULL << 62)
101 
102 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
103     uint64_t obj, dmu_tx_t *tx);
104 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
105     dmu_tx_t *tx);
106 
107 static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f);
108 
109 extern uint_t spa_asize_inflation;
110 
111 static zil_header_t zero_zil;
112 
113 /*
114  * Figure out how much of this delta should be propagated to the dsl_dir
115  * layer.  If there's a refreservation, that space has already been
116  * partially accounted for in our ancestors.
117  */
118 static int64_t
119 parent_delta(dsl_dataset_t *ds, int64_t delta)
120 {
121 	dsl_dataset_phys_t *ds_phys;
122 	uint64_t old_bytes, new_bytes;
123 
124 	if (ds->ds_reserved == 0)
125 		return (delta);
126 
127 	ds_phys = dsl_dataset_phys(ds);
128 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
129 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
130 
131 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
132 	return (new_bytes - old_bytes);
133 }
134 
135 void
136 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
137 {
138 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
139 	int used = bp_get_dsize_sync(spa, bp);
140 	int compressed = BP_GET_PSIZE(bp);
141 	int uncompressed = BP_GET_UCSIZE(bp);
142 	int64_t delta;
143 	spa_feature_t f;
144 
145 	dprintf_bp(bp, "ds=%p", ds);
146 
147 	ASSERT(dmu_tx_is_syncing(tx));
148 	/* It could have been compressed away to nothing */
149 	if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
150 		return;
151 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
152 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
153 	if (ds == NULL) {
154 		dsl_pool_mos_diduse_space(tx->tx_pool,
155 		    used, compressed, uncompressed);
156 		return;
157 	}
158 
159 	ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
160 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
161 	mutex_enter(&ds->ds_lock);
162 	delta = parent_delta(ds, used);
163 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
164 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
165 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
166 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
167 
168 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
169 		ds->ds_feature_activation[SPA_FEATURE_LARGE_BLOCKS] =
170 		    (void *)B_TRUE;
171 	}
172 
173 
174 	f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
175 	if (f != SPA_FEATURE_NONE) {
176 		ASSERT3S(spa_feature_table[f].fi_type, ==,
177 		    ZFEATURE_TYPE_BOOLEAN);
178 		ds->ds_feature_activation[f] = (void *)B_TRUE;
179 	}
180 
181 	f = zio_compress_to_feature(BP_GET_COMPRESS(bp));
182 	if (f != SPA_FEATURE_NONE) {
183 		ASSERT3S(spa_feature_table[f].fi_type, ==,
184 		    ZFEATURE_TYPE_BOOLEAN);
185 		ds->ds_feature_activation[f] = (void *)B_TRUE;
186 	}
187 
188 	/*
189 	 * Track block for livelist, but ignore embedded blocks because
190 	 * they do not need to be freed.
191 	 */
192 	if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
193 	    bp->blk_birth > ds->ds_dir->dd_origin_txg &&
194 	    !(BP_IS_EMBEDDED(bp))) {
195 		ASSERT(dsl_dir_is_clone(ds->ds_dir));
196 		ASSERT(spa_feature_is_enabled(spa,
197 		    SPA_FEATURE_LIVELIST));
198 		bplist_append(&ds->ds_dir->dd_pending_allocs, bp);
199 	}
200 
201 	mutex_exit(&ds->ds_lock);
202 	dsl_dir_diduse_transfer_space(ds->ds_dir, delta,
203 	    compressed, uncompressed, used,
204 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
205 }
206 
207 /*
208  * Called when the specified segment has been remapped, and is thus no
209  * longer referenced in the head dataset.  The vdev must be indirect.
210  *
211  * If the segment is referenced by a snapshot, put it on the remap deadlist.
212  * Otherwise, add this segment to the obsolete spacemap.
213  */
214 void
215 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset,
216     uint64_t size, uint64_t birth, dmu_tx_t *tx)
217 {
218 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
219 
220 	ASSERT(dmu_tx_is_syncing(tx));
221 	ASSERT(birth <= tx->tx_txg);
222 	ASSERT(!ds->ds_is_snapshot);
223 
224 	if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
225 		spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
226 	} else {
227 		blkptr_t fakebp;
228 		dva_t *dva = &fakebp.blk_dva[0];
229 
230 		ASSERT(ds != NULL);
231 
232 		mutex_enter(&ds->ds_remap_deadlist_lock);
233 		if (!dsl_dataset_remap_deadlist_exists(ds)) {
234 			dsl_dataset_create_remap_deadlist(ds, tx);
235 		}
236 		mutex_exit(&ds->ds_remap_deadlist_lock);
237 
238 		BP_ZERO(&fakebp);
239 		fakebp.blk_birth = birth;
240 		DVA_SET_VDEV(dva, vdev);
241 		DVA_SET_OFFSET(dva, offset);
242 		DVA_SET_ASIZE(dva, size);
243 		dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, B_FALSE,
244 		    tx);
245 	}
246 }
247 
248 int
249 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
250     boolean_t async)
251 {
252 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
253 
254 	int used = bp_get_dsize_sync(spa, bp);
255 	int compressed = BP_GET_PSIZE(bp);
256 	int uncompressed = BP_GET_UCSIZE(bp);
257 
258 	if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
259 		return (0);
260 
261 	ASSERT(dmu_tx_is_syncing(tx));
262 	ASSERT(bp->blk_birth <= tx->tx_txg);
263 
264 	if (ds == NULL) {
265 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
266 		dsl_pool_mos_diduse_space(tx->tx_pool,
267 		    -used, -compressed, -uncompressed);
268 		return (used);
269 	}
270 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
271 
272 	ASSERT(!ds->ds_is_snapshot);
273 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
274 
275 	/*
276 	 * Track block for livelist, but ignore embedded blocks because
277 	 * they do not need to be freed.
278 	 */
279 	if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
280 	    bp->blk_birth > ds->ds_dir->dd_origin_txg &&
281 	    !(BP_IS_EMBEDDED(bp))) {
282 		ASSERT(dsl_dir_is_clone(ds->ds_dir));
283 		ASSERT(spa_feature_is_enabled(spa,
284 		    SPA_FEATURE_LIVELIST));
285 		bplist_append(&ds->ds_dir->dd_pending_frees, bp);
286 	}
287 
288 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
289 		int64_t delta;
290 
291 		dprintf_bp(bp, "freeing ds=%llu", (u_longlong_t)ds->ds_object);
292 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
293 
294 		mutex_enter(&ds->ds_lock);
295 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
296 		    !DS_UNIQUE_IS_ACCURATE(ds));
297 		delta = parent_delta(ds, -used);
298 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
299 		mutex_exit(&ds->ds_lock);
300 		dsl_dir_diduse_transfer_space(ds->ds_dir,
301 		    delta, -compressed, -uncompressed, -used,
302 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
303 	} else {
304 		dprintf_bp(bp, "putting on dead list: %s", "");
305 		if (async) {
306 			/*
307 			 * We are here as part of zio's write done callback,
308 			 * which means we're a zio interrupt thread.  We can't
309 			 * call dsl_deadlist_insert() now because it may block
310 			 * waiting for I/O.  Instead, put bp on the deferred
311 			 * queue and let dsl_pool_sync() finish the job.
312 			 */
313 			bplist_append(&ds->ds_pending_deadlist, bp);
314 		} else {
315 			dsl_deadlist_insert(&ds->ds_deadlist, bp, B_FALSE, tx);
316 		}
317 		ASSERT3U(ds->ds_prev->ds_object, ==,
318 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
319 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
320 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
321 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
322 		    ds->ds_object && bp->blk_birth >
323 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
324 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
325 			mutex_enter(&ds->ds_prev->ds_lock);
326 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
327 			mutex_exit(&ds->ds_prev->ds_lock);
328 		}
329 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
330 			dsl_dir_transfer_space(ds->ds_dir, used,
331 			    DD_USED_HEAD, DD_USED_SNAP, tx);
332 		}
333 	}
334 
335 	dsl_bookmark_block_killed(ds, bp, tx);
336 
337 	mutex_enter(&ds->ds_lock);
338 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
339 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
340 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
341 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
342 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
343 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
344 	mutex_exit(&ds->ds_lock);
345 
346 	return (used);
347 }
348 
349 struct feature_type_uint64_array_arg {
350 	uint64_t length;
351 	uint64_t *array;
352 };
353 
354 static void
355 unload_zfeature(dsl_dataset_t *ds, spa_feature_t f)
356 {
357 	switch (spa_feature_table[f].fi_type) {
358 	case ZFEATURE_TYPE_BOOLEAN:
359 		break;
360 	case ZFEATURE_TYPE_UINT64_ARRAY:
361 	{
362 		struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f];
363 		kmem_free(ftuaa->array, ftuaa->length * sizeof (uint64_t));
364 		kmem_free(ftuaa, sizeof (*ftuaa));
365 		break;
366 	}
367 	default:
368 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
369 	}
370 }
371 
372 static int
373 load_zfeature(objset_t *mos, dsl_dataset_t *ds, spa_feature_t f)
374 {
375 	int err = 0;
376 	switch (spa_feature_table[f].fi_type) {
377 	case ZFEATURE_TYPE_BOOLEAN:
378 		err = zap_contains(mos, ds->ds_object,
379 		    spa_feature_table[f].fi_guid);
380 		if (err == 0) {
381 			ds->ds_feature[f] = (void *)B_TRUE;
382 		} else {
383 			ASSERT3U(err, ==, ENOENT);
384 			err = 0;
385 		}
386 		break;
387 	case ZFEATURE_TYPE_UINT64_ARRAY:
388 	{
389 		uint64_t int_size, num_int;
390 		uint64_t *data;
391 		err = zap_length(mos, ds->ds_object,
392 		    spa_feature_table[f].fi_guid, &int_size, &num_int);
393 		if (err != 0) {
394 			ASSERT3U(err, ==, ENOENT);
395 			err = 0;
396 			break;
397 		}
398 		ASSERT3U(int_size, ==, sizeof (uint64_t));
399 		data = kmem_alloc(int_size * num_int, KM_SLEEP);
400 		VERIFY0(zap_lookup(mos, ds->ds_object,
401 		    spa_feature_table[f].fi_guid, int_size, num_int, data));
402 		struct feature_type_uint64_array_arg *ftuaa =
403 		    kmem_alloc(sizeof (*ftuaa), KM_SLEEP);
404 		ftuaa->length = num_int;
405 		ftuaa->array = data;
406 		ds->ds_feature[f] = ftuaa;
407 		break;
408 	}
409 	default:
410 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
411 	}
412 	return (err);
413 }
414 
415 /*
416  * We have to release the fsid synchronously or we risk that a subsequent
417  * mount of the same dataset will fail to unique_insert the fsid.  This
418  * failure would manifest itself as the fsid of this dataset changing
419  * between mounts which makes NFS clients quite unhappy.
420  */
421 static void
422 dsl_dataset_evict_sync(void *dbu)
423 {
424 	dsl_dataset_t *ds = dbu;
425 
426 	ASSERT(ds->ds_owner == NULL);
427 
428 	unique_remove(ds->ds_fsid_guid);
429 }
430 
431 static void
432 dsl_dataset_evict_async(void *dbu)
433 {
434 	dsl_dataset_t *ds = dbu;
435 
436 	ASSERT(ds->ds_owner == NULL);
437 
438 	ds->ds_dbuf = NULL;
439 
440 	if (ds->ds_objset != NULL)
441 		dmu_objset_evict(ds->ds_objset);
442 
443 	if (ds->ds_prev) {
444 		dsl_dataset_rele(ds->ds_prev, ds);
445 		ds->ds_prev = NULL;
446 	}
447 
448 	dsl_bookmark_fini_ds(ds);
449 
450 	bplist_destroy(&ds->ds_pending_deadlist);
451 	if (dsl_deadlist_is_open(&ds->ds_deadlist))
452 		dsl_deadlist_close(&ds->ds_deadlist);
453 	if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
454 		dsl_deadlist_close(&ds->ds_remap_deadlist);
455 	if (ds->ds_dir)
456 		dsl_dir_async_rele(ds->ds_dir, ds);
457 
458 	ASSERT(!list_link_active(&ds->ds_synced_link));
459 
460 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
461 		if (dsl_dataset_feature_is_active(ds, f))
462 			unload_zfeature(ds, f);
463 	}
464 
465 	list_destroy(&ds->ds_prop_cbs);
466 	mutex_destroy(&ds->ds_lock);
467 	mutex_destroy(&ds->ds_opening_lock);
468 	mutex_destroy(&ds->ds_sendstream_lock);
469 	mutex_destroy(&ds->ds_remap_deadlist_lock);
470 	zfs_refcount_destroy(&ds->ds_longholds);
471 	rrw_destroy(&ds->ds_bp_rwlock);
472 
473 	kmem_free(ds, sizeof (dsl_dataset_t));
474 }
475 
476 int
477 dsl_dataset_get_snapname(dsl_dataset_t *ds)
478 {
479 	dsl_dataset_phys_t *headphys;
480 	int err;
481 	dmu_buf_t *headdbuf;
482 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
483 	objset_t *mos = dp->dp_meta_objset;
484 
485 	if (ds->ds_snapname[0])
486 		return (0);
487 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
488 		return (0);
489 
490 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
491 	    FTAG, &headdbuf);
492 	if (err != 0)
493 		return (err);
494 	headphys = headdbuf->db_data;
495 	err = zap_value_search(dp->dp_meta_objset,
496 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
497 	if (err != 0 && zfs_recover == B_TRUE) {
498 		err = 0;
499 		(void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname),
500 		    "SNAPOBJ=%llu-ERR=%d",
501 		    (unsigned long long)ds->ds_object, err);
502 	}
503 	dmu_buf_rele(headdbuf, FTAG);
504 	return (err);
505 }
506 
507 int
508 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
509 {
510 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
511 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
512 	matchtype_t mt = 0;
513 	int err;
514 
515 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
516 		mt = MT_NORMALIZE;
517 
518 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
519 	    value, mt, NULL, 0, NULL);
520 	if (err == ENOTSUP && (mt & MT_NORMALIZE))
521 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
522 	return (err);
523 }
524 
525 int
526 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
527     boolean_t adj_cnt)
528 {
529 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
530 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
531 	matchtype_t mt = 0;
532 	int err;
533 
534 	dsl_dir_snap_cmtime_update(ds->ds_dir, tx);
535 
536 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
537 		mt = MT_NORMALIZE;
538 
539 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
540 	if (err == ENOTSUP && (mt & MT_NORMALIZE))
541 		err = zap_remove(mos, snapobj, name, tx);
542 
543 	if (err == 0 && adj_cnt)
544 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
545 		    DD_FIELD_SNAPSHOT_COUNT, tx);
546 
547 	return (err);
548 }
549 
550 boolean_t
551 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, const void *tag)
552 {
553 	dmu_buf_t *dbuf = ds->ds_dbuf;
554 	boolean_t result = B_FALSE;
555 
556 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
557 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
558 
559 		if (ds == dmu_buf_get_user(dbuf))
560 			result = B_TRUE;
561 		else
562 			dmu_buf_rele(dbuf, tag);
563 	}
564 
565 	return (result);
566 }
567 
568 int
569 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, const void *tag,
570     dsl_dataset_t **dsp)
571 {
572 	objset_t *mos = dp->dp_meta_objset;
573 	dmu_buf_t *dbuf;
574 	dsl_dataset_t *ds;
575 	int err;
576 	dmu_object_info_t doi;
577 
578 	ASSERT(dsl_pool_config_held(dp));
579 
580 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
581 	if (err != 0)
582 		return (err);
583 
584 	/* Make sure dsobj has the correct object type. */
585 	dmu_object_info_from_db(dbuf, &doi);
586 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
587 		dmu_buf_rele(dbuf, tag);
588 		return (SET_ERROR(EINVAL));
589 	}
590 
591 	ds = dmu_buf_get_user(dbuf);
592 	if (ds == NULL) {
593 		dsl_dataset_t *winner = NULL;
594 
595 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
596 		ds->ds_dbuf = dbuf;
597 		ds->ds_object = dsobj;
598 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
599 		list_link_init(&ds->ds_synced_link);
600 
601 		err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj,
602 		    NULL, ds, &ds->ds_dir);
603 		if (err != 0) {
604 			kmem_free(ds, sizeof (dsl_dataset_t));
605 			dmu_buf_rele(dbuf, tag);
606 			return (err);
607 		}
608 
609 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
610 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
611 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
612 		mutex_init(&ds->ds_remap_deadlist_lock,
613 		    NULL, MUTEX_DEFAULT, NULL);
614 		rrw_init(&ds->ds_bp_rwlock, B_FALSE);
615 		zfs_refcount_create(&ds->ds_longholds);
616 
617 		bplist_create(&ds->ds_pending_deadlist);
618 
619 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendstatus_t),
620 		    offsetof(dmu_sendstatus_t, dss_link));
621 
622 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
623 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
624 
625 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
626 			spa_feature_t f;
627 
628 			for (f = 0; f < SPA_FEATURES; f++) {
629 				if (!(spa_feature_table[f].fi_flags &
630 				    ZFEATURE_FLAG_PER_DATASET))
631 					continue;
632 				err = load_zfeature(mos, ds, f);
633 			}
634 		}
635 
636 		if (!ds->ds_is_snapshot) {
637 			ds->ds_snapname[0] = '\0';
638 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
639 				err = dsl_dataset_hold_obj(dp,
640 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
641 				    ds, &ds->ds_prev);
642 			}
643 			if (err != 0)
644 				goto after_dsl_bookmark_fini;
645 			err = dsl_bookmark_init_ds(ds);
646 		} else {
647 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
648 				err = dsl_dataset_get_snapname(ds);
649 			if (err == 0 &&
650 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
651 				err = zap_count(
652 				    ds->ds_dir->dd_pool->dp_meta_objset,
653 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
654 				    &ds->ds_userrefs);
655 			}
656 		}
657 
658 		if (err == 0 && !ds->ds_is_snapshot) {
659 			err = dsl_prop_get_int_ds(ds,
660 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
661 			    &ds->ds_reserved);
662 			if (err == 0) {
663 				err = dsl_prop_get_int_ds(ds,
664 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
665 				    &ds->ds_quota);
666 			}
667 		} else {
668 			ds->ds_reserved = ds->ds_quota = 0;
669 		}
670 
671 		if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 &&
672 		    ds->ds_is_snapshot &&
673 		    zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) {
674 			dp->dp_spa->spa_errata =
675 			    ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
676 		}
677 
678 		dsl_deadlist_open(&ds->ds_deadlist,
679 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
680 		uint64_t remap_deadlist_obj =
681 		    dsl_dataset_get_remap_deadlist_object(ds);
682 		if (remap_deadlist_obj != 0) {
683 			dsl_deadlist_open(&ds->ds_remap_deadlist, mos,
684 			    remap_deadlist_obj);
685 		}
686 
687 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
688 		    dsl_dataset_evict_async, &ds->ds_dbuf);
689 		if (err == 0)
690 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
691 
692 		if (err != 0 || winner != NULL) {
693 			dsl_deadlist_close(&ds->ds_deadlist);
694 			if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
695 				dsl_deadlist_close(&ds->ds_remap_deadlist);
696 			dsl_bookmark_fini_ds(ds);
697 after_dsl_bookmark_fini:
698 			if (ds->ds_prev)
699 				dsl_dataset_rele(ds->ds_prev, ds);
700 			dsl_dir_rele(ds->ds_dir, ds);
701 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
702 				if (dsl_dataset_feature_is_active(ds, f))
703 					unload_zfeature(ds, f);
704 			}
705 
706 			list_destroy(&ds->ds_prop_cbs);
707 			list_destroy(&ds->ds_sendstreams);
708 			bplist_destroy(&ds->ds_pending_deadlist);
709 			mutex_destroy(&ds->ds_lock);
710 			mutex_destroy(&ds->ds_opening_lock);
711 			mutex_destroy(&ds->ds_sendstream_lock);
712 			mutex_destroy(&ds->ds_remap_deadlist_lock);
713 			zfs_refcount_destroy(&ds->ds_longholds);
714 			rrw_destroy(&ds->ds_bp_rwlock);
715 			kmem_free(ds, sizeof (dsl_dataset_t));
716 			if (err != 0) {
717 				dmu_buf_rele(dbuf, tag);
718 				return (err);
719 			}
720 			ds = winner;
721 		} else {
722 			ds->ds_fsid_guid =
723 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
724 			if (ds->ds_fsid_guid !=
725 			    dsl_dataset_phys(ds)->ds_fsid_guid) {
726 				zfs_dbgmsg("ds_fsid_guid changed from "
727 				    "%llx to %llx for pool %s dataset id %llu",
728 				    (long long)
729 				    dsl_dataset_phys(ds)->ds_fsid_guid,
730 				    (long long)ds->ds_fsid_guid,
731 				    spa_name(dp->dp_spa),
732 				    (u_longlong_t)dsobj);
733 			}
734 		}
735 	}
736 
737 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
738 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
739 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
740 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
741 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
742 	*dsp = ds;
743 
744 	return (0);
745 }
746 
747 int
748 dsl_dataset_create_key_mapping(dsl_dataset_t *ds)
749 {
750 	dsl_dir_t *dd = ds->ds_dir;
751 
752 	if (dd->dd_crypto_obj == 0)
753 		return (0);
754 
755 	return (spa_keystore_create_mapping(dd->dd_pool->dp_spa,
756 	    ds, ds, &ds->ds_key_mapping));
757 }
758 
759 int
760 dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj,
761     ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp)
762 {
763 	int err;
764 
765 	err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
766 	if (err != 0)
767 		return (err);
768 
769 	ASSERT3P(*dsp, !=, NULL);
770 
771 	if (flags & DS_HOLD_FLAG_DECRYPT) {
772 		err = dsl_dataset_create_key_mapping(*dsp);
773 		if (err != 0)
774 			dsl_dataset_rele(*dsp, tag);
775 	}
776 
777 	return (err);
778 }
779 
780 int
781 dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
782     const void *tag, dsl_dataset_t **dsp)
783 {
784 	dsl_dir_t *dd;
785 	const char *snapname;
786 	uint64_t obj;
787 	int err = 0;
788 	dsl_dataset_t *ds;
789 
790 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
791 	if (err != 0)
792 		return (err);
793 
794 	ASSERT(dsl_pool_config_held(dp));
795 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
796 	if (obj != 0)
797 		err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds);
798 	else
799 		err = SET_ERROR(ENOENT);
800 
801 	/* we may be looking for a snapshot */
802 	if (err == 0 && snapname != NULL) {
803 		dsl_dataset_t *snap_ds;
804 
805 		if (*snapname++ != '@') {
806 			dsl_dataset_rele_flags(ds, flags, tag);
807 			dsl_dir_rele(dd, FTAG);
808 			return (SET_ERROR(ENOENT));
809 		}
810 
811 		dprintf("looking for snapshot '%s'\n", snapname);
812 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
813 		if (err == 0) {
814 			err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag,
815 			    &snap_ds);
816 		}
817 		dsl_dataset_rele_flags(ds, flags, tag);
818 
819 		if (err == 0) {
820 			mutex_enter(&snap_ds->ds_lock);
821 			if (snap_ds->ds_snapname[0] == 0)
822 				(void) strlcpy(snap_ds->ds_snapname, snapname,
823 				    sizeof (snap_ds->ds_snapname));
824 			mutex_exit(&snap_ds->ds_lock);
825 			ds = snap_ds;
826 		}
827 	}
828 	if (err == 0)
829 		*dsp = ds;
830 	dsl_dir_rele(dd, FTAG);
831 	return (err);
832 }
833 
834 int
835 dsl_dataset_hold(dsl_pool_t *dp, const char *name, const void *tag,
836     dsl_dataset_t **dsp)
837 {
838 	return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp));
839 }
840 
841 static int
842 dsl_dataset_own_obj_impl(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
843     const void *tag, boolean_t override, dsl_dataset_t **dsp)
844 {
845 	int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp);
846 	if (err != 0)
847 		return (err);
848 	if (!dsl_dataset_tryown(*dsp, tag, override)) {
849 		dsl_dataset_rele_flags(*dsp, flags, tag);
850 		*dsp = NULL;
851 		return (SET_ERROR(EBUSY));
852 	}
853 	return (0);
854 }
855 
856 
857 int
858 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
859     const void *tag, dsl_dataset_t **dsp)
860 {
861 	return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_FALSE, dsp));
862 }
863 
864 int
865 dsl_dataset_own_obj_force(dsl_pool_t *dp, uint64_t dsobj,
866     ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp)
867 {
868 	return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_TRUE, dsp));
869 }
870 
871 static int
872 dsl_dataset_own_impl(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
873     const void *tag, boolean_t override, dsl_dataset_t **dsp)
874 {
875 	int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp);
876 	if (err != 0)
877 		return (err);
878 	if (!dsl_dataset_tryown(*dsp, tag, override)) {
879 		dsl_dataset_rele_flags(*dsp, flags, tag);
880 		return (SET_ERROR(EBUSY));
881 	}
882 	return (0);
883 }
884 
885 int
886 dsl_dataset_own_force(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
887     const void *tag, dsl_dataset_t **dsp)
888 {
889 	return (dsl_dataset_own_impl(dp, name, flags, tag, B_TRUE, dsp));
890 }
891 
892 int
893 dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
894     const void *tag, dsl_dataset_t **dsp)
895 {
896 	return (dsl_dataset_own_impl(dp, name, flags, tag, B_FALSE, dsp));
897 }
898 
899 /*
900  * See the comment above dsl_pool_hold() for details.  In summary, a long
901  * hold is used to prevent destruction of a dataset while the pool hold
902  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
903  *
904  * The dataset and pool must be held when this function is called.  After it
905  * is called, the pool hold may be released while the dataset is still held
906  * and accessed.
907  */
908 void
909 dsl_dataset_long_hold(dsl_dataset_t *ds, const void *tag)
910 {
911 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
912 	(void) zfs_refcount_add(&ds->ds_longholds, tag);
913 }
914 
915 void
916 dsl_dataset_long_rele(dsl_dataset_t *ds, const void *tag)
917 {
918 	(void) zfs_refcount_remove(&ds->ds_longholds, tag);
919 }
920 
921 /* Return B_TRUE if there are any long holds on this dataset. */
922 boolean_t
923 dsl_dataset_long_held(dsl_dataset_t *ds)
924 {
925 	return (!zfs_refcount_is_zero(&ds->ds_longholds));
926 }
927 
928 void
929 dsl_dataset_name(dsl_dataset_t *ds, char *name)
930 {
931 	if (ds == NULL) {
932 		(void) strlcpy(name, "mos", ZFS_MAX_DATASET_NAME_LEN);
933 	} else {
934 		dsl_dir_name(ds->ds_dir, name);
935 		VERIFY0(dsl_dataset_get_snapname(ds));
936 		if (ds->ds_snapname[0]) {
937 			VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
938 			    <, ZFS_MAX_DATASET_NAME_LEN);
939 			/*
940 			 * We use a "recursive" mutex so that we
941 			 * can call dprintf_ds() with ds_lock held.
942 			 */
943 			if (!MUTEX_HELD(&ds->ds_lock)) {
944 				mutex_enter(&ds->ds_lock);
945 				VERIFY3U(strlcat(name, ds->ds_snapname,
946 				    ZFS_MAX_DATASET_NAME_LEN), <,
947 				    ZFS_MAX_DATASET_NAME_LEN);
948 				mutex_exit(&ds->ds_lock);
949 			} else {
950 				VERIFY3U(strlcat(name, ds->ds_snapname,
951 				    ZFS_MAX_DATASET_NAME_LEN), <,
952 				    ZFS_MAX_DATASET_NAME_LEN);
953 			}
954 		}
955 	}
956 }
957 
958 int
959 dsl_dataset_namelen(dsl_dataset_t *ds)
960 {
961 	VERIFY0(dsl_dataset_get_snapname(ds));
962 	mutex_enter(&ds->ds_lock);
963 	int len = strlen(ds->ds_snapname);
964 	mutex_exit(&ds->ds_lock);
965 	/* add '@' if ds is a snap */
966 	if (len > 0)
967 		len++;
968 	len += dsl_dir_namelen(ds->ds_dir);
969 	return (len);
970 }
971 
972 void
973 dsl_dataset_rele(dsl_dataset_t *ds, const void *tag)
974 {
975 	dmu_buf_rele(ds->ds_dbuf, tag);
976 }
977 
978 void
979 dsl_dataset_remove_key_mapping(dsl_dataset_t *ds)
980 {
981 	dsl_dir_t *dd = ds->ds_dir;
982 
983 	if (dd == NULL || dd->dd_crypto_obj == 0)
984 		return;
985 
986 	(void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa,
987 	    ds->ds_object, ds);
988 }
989 
990 void
991 dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags,
992     const void *tag)
993 {
994 	if (flags & DS_HOLD_FLAG_DECRYPT)
995 		dsl_dataset_remove_key_mapping(ds);
996 
997 	dsl_dataset_rele(ds, tag);
998 }
999 
1000 void
1001 dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, const void *tag)
1002 {
1003 	ASSERT3P(ds->ds_owner, ==, tag);
1004 	ASSERT(ds->ds_dbuf != NULL);
1005 
1006 	mutex_enter(&ds->ds_lock);
1007 	ds->ds_owner = NULL;
1008 	mutex_exit(&ds->ds_lock);
1009 	dsl_dataset_long_rele(ds, tag);
1010 	dsl_dataset_rele_flags(ds, flags, tag);
1011 }
1012 
1013 boolean_t
1014 dsl_dataset_tryown(dsl_dataset_t *ds, const void *tag, boolean_t override)
1015 {
1016 	boolean_t gotit = FALSE;
1017 
1018 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1019 	mutex_enter(&ds->ds_lock);
1020 	if (ds->ds_owner == NULL && (override || !(DS_IS_INCONSISTENT(ds) ||
1021 	    (dsl_dataset_feature_is_active(ds,
1022 	    SPA_FEATURE_REDACTED_DATASETS) &&
1023 	    !zfs_allow_redacted_dataset_mount)))) {
1024 		ds->ds_owner = tag;
1025 		dsl_dataset_long_hold(ds, tag);
1026 		gotit = TRUE;
1027 	}
1028 	mutex_exit(&ds->ds_lock);
1029 	return (gotit);
1030 }
1031 
1032 boolean_t
1033 dsl_dataset_has_owner(dsl_dataset_t *ds)
1034 {
1035 	boolean_t rv;
1036 	mutex_enter(&ds->ds_lock);
1037 	rv = (ds->ds_owner != NULL);
1038 	mutex_exit(&ds->ds_lock);
1039 	return (rv);
1040 }
1041 
1042 static boolean_t
1043 zfeature_active(spa_feature_t f, void *arg)
1044 {
1045 	switch (spa_feature_table[f].fi_type) {
1046 	case ZFEATURE_TYPE_BOOLEAN: {
1047 		boolean_t val = (boolean_t)(uintptr_t)arg;
1048 		ASSERT(val == B_FALSE || val == B_TRUE);
1049 		return (val);
1050 	}
1051 	case ZFEATURE_TYPE_UINT64_ARRAY:
1052 		/*
1053 		 * In this case, arg is a uint64_t array.  The feature is active
1054 		 * if the array is non-null.
1055 		 */
1056 		return (arg != NULL);
1057 	default:
1058 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
1059 		return (B_FALSE);
1060 	}
1061 }
1062 
1063 boolean_t
1064 dsl_dataset_feature_is_active(dsl_dataset_t *ds, spa_feature_t f)
1065 {
1066 	return (zfeature_active(f, ds->ds_feature[f]));
1067 }
1068 
1069 /*
1070  * The buffers passed out by this function are references to internal buffers;
1071  * they should not be freed by callers of this function, and they should not be
1072  * used after the dataset has been released.
1073  */
1074 boolean_t
1075 dsl_dataset_get_uint64_array_feature(dsl_dataset_t *ds, spa_feature_t f,
1076     uint64_t *outlength, uint64_t **outp)
1077 {
1078 	VERIFY(spa_feature_table[f].fi_type & ZFEATURE_TYPE_UINT64_ARRAY);
1079 	if (!dsl_dataset_feature_is_active(ds, f)) {
1080 		return (B_FALSE);
1081 	}
1082 	struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f];
1083 	*outp = ftuaa->array;
1084 	*outlength = ftuaa->length;
1085 	return (B_TRUE);
1086 }
1087 
1088 void
1089 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, void *arg,
1090     dmu_tx_t *tx)
1091 {
1092 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1093 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
1094 	uint64_t zero = 0;
1095 
1096 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
1097 
1098 	spa_feature_incr(spa, f, tx);
1099 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1100 
1101 	switch (spa_feature_table[f].fi_type) {
1102 	case ZFEATURE_TYPE_BOOLEAN:
1103 		ASSERT3S((boolean_t)(uintptr_t)arg, ==, B_TRUE);
1104 		VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
1105 		    sizeof (zero), 1, &zero, tx));
1106 		break;
1107 	case ZFEATURE_TYPE_UINT64_ARRAY:
1108 	{
1109 		struct feature_type_uint64_array_arg *ftuaa = arg;
1110 		VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
1111 		    sizeof (uint64_t), ftuaa->length, ftuaa->array, tx));
1112 		break;
1113 	}
1114 	default:
1115 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
1116 	}
1117 }
1118 
1119 static void
1120 dsl_dataset_deactivate_feature_impl(dsl_dataset_t *ds, spa_feature_t f,
1121     dmu_tx_t *tx)
1122 {
1123 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1124 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
1125 	uint64_t dsobj = ds->ds_object;
1126 
1127 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
1128 
1129 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
1130 	spa_feature_decr(spa, f, tx);
1131 	ds->ds_feature[f] = NULL;
1132 }
1133 
1134 void
1135 dsl_dataset_deactivate_feature(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx)
1136 {
1137 	unload_zfeature(ds, f);
1138 	dsl_dataset_deactivate_feature_impl(ds, f, tx);
1139 }
1140 
1141 uint64_t
1142 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
1143     dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx)
1144 {
1145 	dsl_pool_t *dp = dd->dd_pool;
1146 	dmu_buf_t *dbuf;
1147 	dsl_dataset_phys_t *dsphys;
1148 	uint64_t dsobj;
1149 	objset_t *mos = dp->dp_meta_objset;
1150 
1151 	if (origin == NULL)
1152 		origin = dp->dp_origin_snap;
1153 
1154 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
1155 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
1156 	ASSERT(dmu_tx_is_syncing(tx));
1157 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1158 
1159 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1160 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1161 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1162 	dmu_buf_will_dirty(dbuf, tx);
1163 	dsphys = dbuf->db_data;
1164 	memset(dsphys, 0, sizeof (dsl_dataset_phys_t));
1165 	dsphys->ds_dir_obj = dd->dd_object;
1166 	dsphys->ds_flags = flags;
1167 	dsphys->ds_fsid_guid = unique_create();
1168 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1169 	    sizeof (dsphys->ds_guid));
1170 	dsphys->ds_snapnames_zapobj =
1171 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
1172 	    DMU_OT_NONE, 0, tx);
1173 	dsphys->ds_creation_time = gethrestime_sec();
1174 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
1175 
1176 	if (origin == NULL) {
1177 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
1178 	} else {
1179 		dsl_dataset_t *ohds; /* head of the origin snapshot */
1180 
1181 		dsphys->ds_prev_snap_obj = origin->ds_object;
1182 		dsphys->ds_prev_snap_txg =
1183 		    dsl_dataset_phys(origin)->ds_creation_txg;
1184 		dsphys->ds_referenced_bytes =
1185 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
1186 		dsphys->ds_compressed_bytes =
1187 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
1188 		dsphys->ds_uncompressed_bytes =
1189 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
1190 		rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
1191 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
1192 		rrw_exit(&origin->ds_bp_rwlock, FTAG);
1193 
1194 		/*
1195 		 * Inherit flags that describe the dataset's contents
1196 		 * (INCONSISTENT) or properties (Case Insensitive).
1197 		 */
1198 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
1199 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
1200 
1201 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1202 			if (zfeature_active(f, origin->ds_feature[f])) {
1203 				dsl_dataset_activate_feature(dsobj, f,
1204 				    origin->ds_feature[f], tx);
1205 			}
1206 		}
1207 
1208 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
1209 		dsl_dataset_phys(origin)->ds_num_children++;
1210 
1211 		VERIFY0(dsl_dataset_hold_obj(dp,
1212 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
1213 		    FTAG, &ohds));
1214 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
1215 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
1216 		dsl_dataset_rele(ohds, FTAG);
1217 
1218 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
1219 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
1220 				dsl_dataset_phys(origin)->ds_next_clones_obj =
1221 				    zap_create(mos,
1222 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
1223 			}
1224 			VERIFY0(zap_add_int(mos,
1225 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
1226 			    dsobj, tx));
1227 		}
1228 
1229 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
1230 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
1231 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1232 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
1233 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
1234 				dsl_dir_phys(origin->ds_dir)->dd_clones =
1235 				    zap_create(mos,
1236 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
1237 			}
1238 			VERIFY0(zap_add_int(mos,
1239 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
1240 			    dsobj, tx));
1241 		}
1242 	}
1243 
1244 	/* handle encryption */
1245 	dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx);
1246 
1247 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1248 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1249 
1250 	dmu_buf_rele(dbuf, FTAG);
1251 
1252 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1253 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
1254 
1255 	return (dsobj);
1256 }
1257 
1258 static void
1259 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
1260 {
1261 	objset_t *os;
1262 
1263 	VERIFY0(dmu_objset_from_ds(ds, &os));
1264 	if (memcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
1265 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1266 		zio_t *zio;
1267 
1268 		memset(&os->os_zil_header, 0, sizeof (os->os_zil_header));
1269 		if (os->os_encrypted)
1270 			os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE;
1271 
1272 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1273 		dsl_dataset_sync(ds, zio, tx);
1274 		VERIFY0(zio_wait(zio));
1275 		dsl_dataset_sync_done(ds, tx);
1276 	}
1277 }
1278 
1279 uint64_t
1280 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
1281     dsl_dataset_t *origin, uint64_t flags, cred_t *cr,
1282     dsl_crypto_params_t *dcp, dmu_tx_t *tx)
1283 {
1284 	dsl_pool_t *dp = pdd->dd_pool;
1285 	uint64_t dsobj, ddobj;
1286 	dsl_dir_t *dd;
1287 
1288 	ASSERT(dmu_tx_is_syncing(tx));
1289 	ASSERT(lastname[0] != '@');
1290 	/*
1291 	 * Filesystems will eventually have their origin set to dp_origin_snap,
1292 	 * but that's taken care of in dsl_dataset_create_sync_dd. When
1293 	 * creating a filesystem, this function is called with origin equal to
1294 	 * NULL.
1295 	 */
1296 	if (origin != NULL)
1297 		ASSERT3P(origin, !=, dp->dp_origin_snap);
1298 
1299 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
1300 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
1301 
1302 	dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp,
1303 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
1304 
1305 	dsl_deleg_set_create_perms(dd, tx, cr);
1306 
1307 	/*
1308 	 * If we are creating a clone and the livelist feature is enabled,
1309 	 * add the entry DD_FIELD_LIVELIST to ZAP.
1310 	 */
1311 	if (origin != NULL &&
1312 	    spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LIVELIST)) {
1313 		objset_t *mos = dd->dd_pool->dp_meta_objset;
1314 		dsl_dir_zapify(dd, tx);
1315 		uint64_t obj = dsl_deadlist_alloc(mos, tx);
1316 		VERIFY0(zap_add(mos, dd->dd_object, DD_FIELD_LIVELIST,
1317 		    sizeof (uint64_t), 1, &obj, tx));
1318 		spa_feature_incr(dp->dp_spa, SPA_FEATURE_LIVELIST, tx);
1319 	}
1320 
1321 	/*
1322 	 * Since we're creating a new node we know it's a leaf, so we can
1323 	 * initialize the counts if the limit feature is active.
1324 	 */
1325 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
1326 		uint64_t cnt = 0;
1327 		objset_t *os = dd->dd_pool->dp_meta_objset;
1328 
1329 		dsl_dir_zapify(dd, tx);
1330 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1331 		    sizeof (cnt), 1, &cnt, tx));
1332 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1333 		    sizeof (cnt), 1, &cnt, tx));
1334 	}
1335 
1336 	dsl_dir_rele(dd, FTAG);
1337 
1338 	/*
1339 	 * If we are creating a clone, make sure we zero out any stale
1340 	 * data from the origin snapshots zil header.
1341 	 */
1342 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
1343 		dsl_dataset_t *ds;
1344 
1345 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1346 		dsl_dataset_zero_zil(ds, tx);
1347 		dsl_dataset_rele(ds, FTAG);
1348 	}
1349 
1350 	return (dsobj);
1351 }
1352 
1353 /*
1354  * The unique space in the head dataset can be calculated by subtracting
1355  * the space used in the most recent snapshot, that is still being used
1356  * in this file system, from the space currently in use.  To figure out
1357  * the space in the most recent snapshot still in use, we need to take
1358  * the total space used in the snapshot and subtract out the space that
1359  * has been freed up since the snapshot was taken.
1360  */
1361 void
1362 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1363 {
1364 	uint64_t mrs_used;
1365 	uint64_t dlused, dlcomp, dluncomp;
1366 
1367 	ASSERT(!ds->ds_is_snapshot);
1368 
1369 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1370 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1371 	else
1372 		mrs_used = 0;
1373 
1374 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1375 
1376 	ASSERT3U(dlused, <=, mrs_used);
1377 	dsl_dataset_phys(ds)->ds_unique_bytes =
1378 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1379 
1380 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1381 	    SPA_VERSION_UNIQUE_ACCURATE)
1382 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1383 }
1384 
1385 void
1386 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1387     dmu_tx_t *tx)
1388 {
1389 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1390 	uint64_t count __maybe_unused;
1391 	int err;
1392 
1393 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1394 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1395 	    obj, tx);
1396 	/*
1397 	 * The err should not be ENOENT, but a bug in a previous version
1398 	 * of the code could cause upgrade_clones_cb() to not set
1399 	 * ds_next_snap_obj when it should, leading to a missing entry.
1400 	 * If we knew that the pool was created after
1401 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1402 	 * ENOENT.  However, at least we can check that we don't have
1403 	 * too many entries in the next_clones_obj even after failing to
1404 	 * remove this one.
1405 	 */
1406 	if (err != ENOENT)
1407 		VERIFY0(err);
1408 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1409 	    &count));
1410 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1411 }
1412 
1413 
1414 blkptr_t *
1415 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1416 {
1417 	return (&dsl_dataset_phys(ds)->ds_bp);
1418 }
1419 
1420 spa_t *
1421 dsl_dataset_get_spa(dsl_dataset_t *ds)
1422 {
1423 	return (ds->ds_dir->dd_pool->dp_spa);
1424 }
1425 
1426 void
1427 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1428 {
1429 	dsl_pool_t *dp;
1430 
1431 	if (ds == NULL) /* this is the meta-objset */
1432 		return;
1433 
1434 	ASSERT(ds->ds_objset != NULL);
1435 
1436 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1437 		panic("dirtying snapshot!");
1438 
1439 	/* Must not dirty a dataset in the same txg where it got snapshotted. */
1440 	ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1441 
1442 	dp = ds->ds_dir->dd_pool;
1443 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1444 		objset_t *os = ds->ds_objset;
1445 
1446 		/* up the hold count until we can be written out */
1447 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1448 
1449 		/* if this dataset is encrypted, grab a reference to the DCK */
1450 		if (ds->ds_dir->dd_crypto_obj != 0 &&
1451 		    !os->os_raw_receive &&
1452 		    !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
1453 			ASSERT3P(ds->ds_key_mapping, !=, NULL);
1454 			key_mapping_add_ref(ds->ds_key_mapping, ds);
1455 		}
1456 	}
1457 }
1458 
1459 static int
1460 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1461 {
1462 	uint64_t asize;
1463 
1464 	if (!dmu_tx_is_syncing(tx))
1465 		return (0);
1466 
1467 	/*
1468 	 * If there's an fs-only reservation, any blocks that might become
1469 	 * owned by the snapshot dataset must be accommodated by space
1470 	 * outside of the reservation.
1471 	 */
1472 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1473 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1474 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1475 		return (SET_ERROR(ENOSPC));
1476 
1477 	/*
1478 	 * Propagate any reserved space for this snapshot to other
1479 	 * snapshot checks in this sync group.
1480 	 */
1481 	if (asize > 0)
1482 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1483 
1484 	return (0);
1485 }
1486 
1487 int
1488 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1489     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr, proc_t *proc)
1490 {
1491 	int error;
1492 	uint64_t value;
1493 
1494 	ds->ds_trysnap_txg = tx->tx_txg;
1495 
1496 	if (!dmu_tx_is_syncing(tx))
1497 		return (0);
1498 
1499 	/*
1500 	 * We don't allow multiple snapshots of the same txg.  If there
1501 	 * is already one, try again.
1502 	 */
1503 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1504 		return (SET_ERROR(EAGAIN));
1505 
1506 	/*
1507 	 * Check for conflicting snapshot name.
1508 	 */
1509 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1510 	if (error == 0)
1511 		return (SET_ERROR(EEXIST));
1512 	if (error != ENOENT)
1513 		return (error);
1514 
1515 	/*
1516 	 * We don't allow taking snapshots of inconsistent datasets, such as
1517 	 * those into which we are currently receiving.  However, if we are
1518 	 * creating this snapshot as part of a receive, this check will be
1519 	 * executed atomically with respect to the completion of the receive
1520 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1521 	 * case we ignore this, knowing it will be fixed up for us shortly in
1522 	 * dmu_recv_end_sync().
1523 	 */
1524 	if (!recv && DS_IS_INCONSISTENT(ds))
1525 		return (SET_ERROR(EBUSY));
1526 
1527 	/*
1528 	 * Skip the check for temporary snapshots or if we have already checked
1529 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1530 	 * check the count here when we're receiving a stream.
1531 	 */
1532 	if (cnt != 0 && cr != NULL) {
1533 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1534 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr, proc);
1535 		if (error != 0)
1536 			return (error);
1537 	}
1538 
1539 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1540 	if (error != 0)
1541 		return (error);
1542 
1543 	return (0);
1544 }
1545 
1546 int
1547 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1548 {
1549 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1550 	dsl_pool_t *dp = dmu_tx_pool(tx);
1551 	nvpair_t *pair;
1552 	int rv = 0;
1553 
1554 	/*
1555 	 * Pre-compute how many total new snapshots will be created for each
1556 	 * level in the tree and below. This is needed for validating the
1557 	 * snapshot limit when either taking a recursive snapshot or when
1558 	 * taking multiple snapshots.
1559 	 *
1560 	 * The problem is that the counts are not actually adjusted when
1561 	 * we are checking, only when we finally sync. For a single snapshot,
1562 	 * this is easy, the count will increase by 1 at each node up the tree,
1563 	 * but its more complicated for the recursive/multiple snapshot case.
1564 	 *
1565 	 * The dsl_fs_ss_limit_check function does recursively check the count
1566 	 * at each level up the tree but since it is validating each snapshot
1567 	 * independently we need to be sure that we are validating the complete
1568 	 * count for the entire set of snapshots. We do this by rolling up the
1569 	 * counts for each component of the name into an nvlist and then
1570 	 * checking each of those cases with the aggregated count.
1571 	 *
1572 	 * This approach properly handles not only the recursive snapshot
1573 	 * case (where we get all of those on the ddsa_snaps list) but also
1574 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1575 	 * validate the limit on 'a' using a count of 2).
1576 	 *
1577 	 * We validate the snapshot names in the third loop and only report
1578 	 * name errors once.
1579 	 */
1580 	if (dmu_tx_is_syncing(tx)) {
1581 		char *nm;
1582 		nvlist_t *cnt_track = NULL;
1583 		cnt_track = fnvlist_alloc();
1584 
1585 		nm = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1586 
1587 		/* Rollup aggregated counts into the cnt_track list */
1588 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1589 		    pair != NULL;
1590 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1591 			char *pdelim;
1592 			uint64_t val;
1593 
1594 			(void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN);
1595 			pdelim = strchr(nm, '@');
1596 			if (pdelim == NULL)
1597 				continue;
1598 			*pdelim = '\0';
1599 
1600 			do {
1601 				if (nvlist_lookup_uint64(cnt_track, nm,
1602 				    &val) == 0) {
1603 					/* update existing entry */
1604 					fnvlist_add_uint64(cnt_track, nm,
1605 					    val + 1);
1606 				} else {
1607 					/* add to list */
1608 					fnvlist_add_uint64(cnt_track, nm, 1);
1609 				}
1610 
1611 				pdelim = strrchr(nm, '/');
1612 				if (pdelim != NULL)
1613 					*pdelim = '\0';
1614 			} while (pdelim != NULL);
1615 		}
1616 
1617 		kmem_free(nm, MAXPATHLEN);
1618 
1619 		/* Check aggregated counts at each level */
1620 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1621 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1622 			int error = 0;
1623 			const char *name;
1624 			uint64_t cnt = 0;
1625 			dsl_dataset_t *ds;
1626 
1627 			name = nvpair_name(pair);
1628 			cnt = fnvpair_value_uint64(pair);
1629 			ASSERT(cnt > 0);
1630 
1631 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1632 			if (error == 0) {
1633 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1634 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1635 				    ddsa->ddsa_cr, ddsa->ddsa_proc);
1636 				dsl_dataset_rele(ds, FTAG);
1637 			}
1638 
1639 			if (error != 0) {
1640 				if (ddsa->ddsa_errors != NULL)
1641 					fnvlist_add_int32(ddsa->ddsa_errors,
1642 					    name, error);
1643 				rv = error;
1644 				/* only report one error for this check */
1645 				break;
1646 			}
1647 		}
1648 		nvlist_free(cnt_track);
1649 	}
1650 
1651 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1652 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1653 		int error = 0;
1654 		dsl_dataset_t *ds;
1655 		const char *name, *atp = NULL;
1656 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1657 
1658 		name = nvpair_name(pair);
1659 		if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1660 			error = SET_ERROR(ENAMETOOLONG);
1661 		if (error == 0) {
1662 			atp = strchr(name, '@');
1663 			if (atp == NULL)
1664 				error = SET_ERROR(EINVAL);
1665 			if (error == 0)
1666 				(void) strlcpy(dsname, name, atp - name + 1);
1667 		}
1668 		if (error == 0)
1669 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1670 		if (error == 0) {
1671 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1672 			error = dsl_dataset_snapshot_check_impl(ds,
1673 			    atp + 1, tx, B_FALSE, 0, NULL, NULL);
1674 			dsl_dataset_rele(ds, FTAG);
1675 		}
1676 
1677 		if (error != 0) {
1678 			if (ddsa->ddsa_errors != NULL) {
1679 				fnvlist_add_int32(ddsa->ddsa_errors,
1680 				    name, error);
1681 			}
1682 			rv = error;
1683 		}
1684 	}
1685 
1686 	return (rv);
1687 }
1688 
1689 void
1690 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1691     dmu_tx_t *tx)
1692 {
1693 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1694 	dmu_buf_t *dbuf;
1695 	dsl_dataset_phys_t *dsphys;
1696 	uint64_t dsobj, crtxg;
1697 	objset_t *mos = dp->dp_meta_objset;
1698 	objset_t *os __maybe_unused;
1699 
1700 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1701 
1702 	/*
1703 	 * If we are on an old pool, the zil must not be active, in which
1704 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1705 	 */
1706 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1707 	    dmu_objset_from_ds(ds, &os) != 0 ||
1708 	    memcmp(&os->os_phys->os_zil_header, &zero_zil,
1709 	    sizeof (zero_zil)) == 0);
1710 
1711 	/* Should not snapshot a dirty dataset. */
1712 	ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1713 	    ds, tx->tx_txg));
1714 
1715 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1716 
1717 	/*
1718 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1719 	 */
1720 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1721 		crtxg = 1;
1722 	else
1723 		crtxg = tx->tx_txg;
1724 
1725 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1726 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1727 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1728 	dmu_buf_will_dirty(dbuf, tx);
1729 	dsphys = dbuf->db_data;
1730 	memset(dsphys, 0, sizeof (dsl_dataset_phys_t));
1731 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1732 	dsphys->ds_fsid_guid = unique_create();
1733 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1734 	    sizeof (dsphys->ds_guid));
1735 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1736 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1737 	dsphys->ds_next_snap_obj = ds->ds_object;
1738 	dsphys->ds_num_children = 1;
1739 	dsphys->ds_creation_time = gethrestime_sec();
1740 	dsphys->ds_creation_txg = crtxg;
1741 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1742 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1743 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1744 	dsphys->ds_uncompressed_bytes =
1745 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1746 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1747 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1748 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1749 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
1750 	dmu_buf_rele(dbuf, FTAG);
1751 
1752 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1753 		if (zfeature_active(f, ds->ds_feature[f])) {
1754 			dsl_dataset_activate_feature(dsobj, f,
1755 			    ds->ds_feature[f], tx);
1756 		}
1757 	}
1758 
1759 	ASSERT3U(ds->ds_prev != 0, ==,
1760 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1761 	if (ds->ds_prev) {
1762 		uint64_t next_clones_obj =
1763 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1764 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1765 		    ds->ds_object ||
1766 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1767 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1768 		    ds->ds_object) {
1769 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1770 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1771 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1772 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1773 		} else if (next_clones_obj != 0) {
1774 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1775 			    dsphys->ds_next_snap_obj, tx);
1776 			VERIFY0(zap_add_int(mos,
1777 			    next_clones_obj, dsobj, tx));
1778 		}
1779 	}
1780 
1781 	/*
1782 	 * If we have a reference-reservation on this dataset, we will
1783 	 * need to increase the amount of refreservation being charged
1784 	 * since our unique space is going to zero.
1785 	 */
1786 	if (ds->ds_reserved) {
1787 		int64_t delta;
1788 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1789 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1790 		    ds->ds_reserved);
1791 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1792 		    delta, 0, 0, tx);
1793 	}
1794 
1795 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1796 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1797 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1798 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1799 	dsl_deadlist_close(&ds->ds_deadlist);
1800 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1801 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1802 	dsl_deadlist_add_key(&ds->ds_deadlist,
1803 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1804 	dsl_bookmark_snapshotted(ds, tx);
1805 
1806 	if (dsl_dataset_remap_deadlist_exists(ds)) {
1807 		uint64_t remap_deadlist_obj =
1808 		    dsl_dataset_get_remap_deadlist_object(ds);
1809 		/*
1810 		 * Move the remap_deadlist to the snapshot.  The head
1811 		 * will create a new remap deadlist on demand, from
1812 		 * dsl_dataset_block_remapped().
1813 		 */
1814 		dsl_dataset_unset_remap_deadlist_object(ds, tx);
1815 		dsl_deadlist_close(&ds->ds_remap_deadlist);
1816 
1817 		dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1818 		VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST,
1819 		    sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx));
1820 	}
1821 
1822 	/*
1823 	 * Create a ivset guid for this snapshot if the dataset is
1824 	 * encrypted. This may be overridden by a raw receive. A
1825 	 * previous implementation of this code did not have this
1826 	 * field as part of the on-disk format for ZFS encryption
1827 	 * (see errata #4). As part of the remediation for this
1828 	 * issue, we ask the user to enable the bookmark_v2 feature
1829 	 * which is now a dependency of the encryption feature. We
1830 	 * use this as a heuristic to determine when the user has
1831 	 * elected to correct any datasets created with the old code.
1832 	 * As a result, we only do this step if the bookmark_v2
1833 	 * feature is enabled, which limits the number of states a
1834 	 * given pool / dataset can be in with regards to terms of
1835 	 * correcting the issue.
1836 	 */
1837 	if (ds->ds_dir->dd_crypto_obj != 0 &&
1838 	    spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) {
1839 		uint64_t ivset_guid = unique_create();
1840 
1841 		dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1842 		VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID,
1843 		    sizeof (ivset_guid), 1, &ivset_guid, tx));
1844 	}
1845 
1846 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1847 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1848 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1849 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1850 
1851 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1852 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1853 
1854 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1855 	    snapname, 8, 1, &dsobj, tx));
1856 
1857 	if (ds->ds_prev)
1858 		dsl_dataset_rele(ds->ds_prev, ds);
1859 	VERIFY0(dsl_dataset_hold_obj(dp,
1860 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1861 
1862 	dsl_scan_ds_snapshotted(ds, tx);
1863 
1864 	dsl_dir_snap_cmtime_update(ds->ds_dir, tx);
1865 
1866 	if (zfs_snapshot_history_enabled)
1867 		spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, " ");
1868 }
1869 
1870 void
1871 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1872 {
1873 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1874 	dsl_pool_t *dp = dmu_tx_pool(tx);
1875 	nvpair_t *pair;
1876 
1877 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1878 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1879 		dsl_dataset_t *ds;
1880 		const char *name, *atp;
1881 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1882 
1883 		name = nvpair_name(pair);
1884 		atp = strchr(name, '@');
1885 		(void) strlcpy(dsname, name, atp - name + 1);
1886 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1887 
1888 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1889 		if (ddsa->ddsa_props != NULL) {
1890 			dsl_props_set_sync_impl(ds->ds_prev,
1891 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1892 		}
1893 		dsl_dataset_rele(ds, FTAG);
1894 	}
1895 }
1896 
1897 /*
1898  * The snapshots must all be in the same pool.
1899  * All-or-nothing: if there are any failures, nothing will be modified.
1900  */
1901 int
1902 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1903 {
1904 	dsl_dataset_snapshot_arg_t ddsa;
1905 	nvpair_t *pair;
1906 	boolean_t needsuspend;
1907 	int error;
1908 	spa_t *spa;
1909 	const char *firstname;
1910 	nvlist_t *suspended = NULL;
1911 
1912 	pair = nvlist_next_nvpair(snaps, NULL);
1913 	if (pair == NULL)
1914 		return (0);
1915 	firstname = nvpair_name(pair);
1916 
1917 	error = spa_open(firstname, &spa, FTAG);
1918 	if (error != 0)
1919 		return (error);
1920 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1921 	spa_close(spa, FTAG);
1922 
1923 	if (needsuspend) {
1924 		suspended = fnvlist_alloc();
1925 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1926 		    pair = nvlist_next_nvpair(snaps, pair)) {
1927 			char fsname[ZFS_MAX_DATASET_NAME_LEN];
1928 			const char *snapname = nvpair_name(pair);
1929 			const char *atp;
1930 			void *cookie;
1931 
1932 			atp = strchr(snapname, '@');
1933 			if (atp == NULL) {
1934 				error = SET_ERROR(EINVAL);
1935 				break;
1936 			}
1937 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1938 
1939 			error = zil_suspend(fsname, &cookie);
1940 			if (error != 0)
1941 				break;
1942 			fnvlist_add_uint64(suspended, fsname,
1943 			    (uintptr_t)cookie);
1944 		}
1945 	}
1946 
1947 	ddsa.ddsa_snaps = snaps;
1948 	ddsa.ddsa_props = props;
1949 	ddsa.ddsa_errors = errors;
1950 	ddsa.ddsa_cr = CRED();
1951 	ddsa.ddsa_proc = curproc;
1952 
1953 	if (error == 0) {
1954 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1955 		    dsl_dataset_snapshot_sync, &ddsa,
1956 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1957 	}
1958 
1959 	if (suspended != NULL) {
1960 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1961 		    pair = nvlist_next_nvpair(suspended, pair)) {
1962 			zil_resume((void *)(uintptr_t)
1963 			    fnvpair_value_uint64(pair));
1964 		}
1965 		fnvlist_free(suspended);
1966 	}
1967 
1968 	if (error == 0) {
1969 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1970 		    pair = nvlist_next_nvpair(snaps, pair)) {
1971 			zvol_create_minor(nvpair_name(pair));
1972 		}
1973 	}
1974 
1975 	return (error);
1976 }
1977 
1978 typedef struct dsl_dataset_snapshot_tmp_arg {
1979 	const char *ddsta_fsname;
1980 	const char *ddsta_snapname;
1981 	minor_t ddsta_cleanup_minor;
1982 	const char *ddsta_htag;
1983 } dsl_dataset_snapshot_tmp_arg_t;
1984 
1985 static int
1986 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1987 {
1988 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1989 	dsl_pool_t *dp = dmu_tx_pool(tx);
1990 	dsl_dataset_t *ds;
1991 	int error;
1992 
1993 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1994 	if (error != 0)
1995 		return (error);
1996 
1997 	/* NULL cred means no limit check for tmp snapshot */
1998 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1999 	    tx, B_FALSE, 0, NULL, NULL);
2000 	if (error != 0) {
2001 		dsl_dataset_rele(ds, FTAG);
2002 		return (error);
2003 	}
2004 
2005 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
2006 		dsl_dataset_rele(ds, FTAG);
2007 		return (SET_ERROR(ENOTSUP));
2008 	}
2009 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
2010 	    B_TRUE, tx);
2011 	if (error != 0) {
2012 		dsl_dataset_rele(ds, FTAG);
2013 		return (error);
2014 	}
2015 
2016 	dsl_dataset_rele(ds, FTAG);
2017 	return (0);
2018 }
2019 
2020 static void
2021 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
2022 {
2023 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
2024 	dsl_pool_t *dp = dmu_tx_pool(tx);
2025 	dsl_dataset_t *ds = NULL;
2026 
2027 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
2028 
2029 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
2030 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
2031 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
2032 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
2033 
2034 	dsl_dataset_rele(ds, FTAG);
2035 }
2036 
2037 int
2038 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
2039     minor_t cleanup_minor, const char *htag)
2040 {
2041 	dsl_dataset_snapshot_tmp_arg_t ddsta;
2042 	int error;
2043 	spa_t *spa;
2044 	boolean_t needsuspend;
2045 	void *cookie;
2046 
2047 	ddsta.ddsta_fsname = fsname;
2048 	ddsta.ddsta_snapname = snapname;
2049 	ddsta.ddsta_cleanup_minor = cleanup_minor;
2050 	ddsta.ddsta_htag = htag;
2051 
2052 	error = spa_open(fsname, &spa, FTAG);
2053 	if (error != 0)
2054 		return (error);
2055 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
2056 	spa_close(spa, FTAG);
2057 
2058 	if (needsuspend) {
2059 		error = zil_suspend(fsname, &cookie);
2060 		if (error != 0)
2061 			return (error);
2062 	}
2063 
2064 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
2065 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
2066 
2067 	if (needsuspend)
2068 		zil_resume(cookie);
2069 	return (error);
2070 }
2071 
2072 /* Nonblocking dataset sync. Assumes dataset:objset is always 1:1 */
2073 void
2074 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *rio, dmu_tx_t *tx)
2075 {
2076 	ASSERT(dmu_tx_is_syncing(tx));
2077 	ASSERT(ds->ds_objset != NULL);
2078 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
2079 
2080 	/*
2081 	 * in case we had to change ds_fsid_guid when we opened it,
2082 	 * sync it out now.
2083 	 */
2084 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2085 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
2086 
2087 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
2088 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2089 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
2090 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
2091 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2092 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
2093 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
2094 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2095 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
2096 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
2097 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
2098 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
2099 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
2100 	}
2101 
2102 	dmu_objset_sync(ds->ds_objset, rio, tx);
2103 }
2104 
2105 /*
2106  * Check if the percentage of blocks shared between the clone and the
2107  * snapshot (as opposed to those that are clone only) is below a certain
2108  * threshold
2109  */
2110 static boolean_t
2111 dsl_livelist_should_disable(dsl_dataset_t *ds)
2112 {
2113 	uint64_t used, referenced;
2114 	int percent_shared;
2115 
2116 	used = dsl_dir_get_usedds(ds->ds_dir);
2117 	referenced = dsl_get_referenced(ds);
2118 	if (referenced == 0)
2119 		return (B_FALSE);
2120 	percent_shared = (100 * (referenced - used)) / referenced;
2121 	if (percent_shared <= zfs_livelist_min_percent_shared)
2122 		return (B_TRUE);
2123 	return (B_FALSE);
2124 }
2125 
2126 /*
2127  *  Check if it is possible to combine two livelist entries into one.
2128  *  This is the case if the combined number of 'live' blkptrs (ALLOCs that
2129  *  don't have a matching FREE) is under the maximum sublist size.
2130  *  We check this by subtracting twice the total number of frees from the total
2131  *  number of blkptrs. FREEs are counted twice because each FREE blkptr
2132  *  will cancel out an ALLOC blkptr when the livelist is processed.
2133  */
2134 static boolean_t
2135 dsl_livelist_should_condense(dsl_deadlist_entry_t *first,
2136     dsl_deadlist_entry_t *next)
2137 {
2138 	uint64_t total_free = first->dle_bpobj.bpo_phys->bpo_num_freed +
2139 	    next->dle_bpobj.bpo_phys->bpo_num_freed;
2140 	uint64_t total_entries = first->dle_bpobj.bpo_phys->bpo_num_blkptrs +
2141 	    next->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2142 	if ((total_entries - (2 * total_free)) < zfs_livelist_max_entries)
2143 		return (B_TRUE);
2144 	return (B_FALSE);
2145 }
2146 
2147 typedef struct try_condense_arg {
2148 	spa_t *spa;
2149 	dsl_dataset_t *ds;
2150 } try_condense_arg_t;
2151 
2152 /*
2153  * Iterate over the livelist entries, searching for a pair to condense.
2154  * A nonzero return value means stop, 0 means keep looking.
2155  */
2156 static int
2157 dsl_livelist_try_condense(void *arg, dsl_deadlist_entry_t *first)
2158 {
2159 	try_condense_arg_t *tca = arg;
2160 	spa_t *spa = tca->spa;
2161 	dsl_dataset_t *ds = tca->ds;
2162 	dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist;
2163 	dsl_deadlist_entry_t *next;
2164 
2165 	/* The condense thread has not yet been created at import */
2166 	if (spa->spa_livelist_condense_zthr == NULL)
2167 		return (1);
2168 
2169 	/* A condense is already in progress */
2170 	if (spa->spa_to_condense.ds != NULL)
2171 		return (1);
2172 
2173 	next = AVL_NEXT(&ll->dl_tree, &first->dle_node);
2174 	/* The livelist has only one entry - don't condense it */
2175 	if (next == NULL)
2176 		return (1);
2177 
2178 	/* Next is the newest entry - don't condense it */
2179 	if (AVL_NEXT(&ll->dl_tree, &next->dle_node) == NULL)
2180 		return (1);
2181 
2182 	/* This pair is not ready to condense but keep looking */
2183 	if (!dsl_livelist_should_condense(first, next))
2184 		return (0);
2185 
2186 	/*
2187 	 * Add a ref to prevent the dataset from being evicted while
2188 	 * the condense zthr or synctask are running. Ref will be
2189 	 * released at the end of the condense synctask
2190 	 */
2191 	dmu_buf_add_ref(ds->ds_dbuf, spa);
2192 
2193 	spa->spa_to_condense.ds = ds;
2194 	spa->spa_to_condense.first = first;
2195 	spa->spa_to_condense.next = next;
2196 	spa->spa_to_condense.syncing = B_FALSE;
2197 	spa->spa_to_condense.cancelled = B_FALSE;
2198 
2199 	zthr_wakeup(spa->spa_livelist_condense_zthr);
2200 	return (1);
2201 }
2202 
2203 static void
2204 dsl_flush_pending_livelist(dsl_dataset_t *ds, dmu_tx_t *tx)
2205 {
2206 	dsl_dir_t *dd = ds->ds_dir;
2207 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
2208 	dsl_deadlist_entry_t *last = dsl_deadlist_last(&dd->dd_livelist);
2209 
2210 	/* Check if we need to add a new sub-livelist */
2211 	if (last == NULL) {
2212 		/* The livelist is empty */
2213 		dsl_deadlist_add_key(&dd->dd_livelist,
2214 		    tx->tx_txg - 1, tx);
2215 	} else if (spa_sync_pass(spa) == 1) {
2216 		/*
2217 		 * Check if the newest entry is full. If it is, make a new one.
2218 		 * We only do this once per sync because we could overfill a
2219 		 * sublist in one sync pass and don't want to add another entry
2220 		 * for a txg that is already represented. This ensures that
2221 		 * blkptrs born in the same txg are stored in the same sublist.
2222 		 */
2223 		bpobj_t bpobj = last->dle_bpobj;
2224 		uint64_t all = bpobj.bpo_phys->bpo_num_blkptrs;
2225 		uint64_t free = bpobj.bpo_phys->bpo_num_freed;
2226 		uint64_t alloc = all - free;
2227 		if (alloc > zfs_livelist_max_entries) {
2228 			dsl_deadlist_add_key(&dd->dd_livelist,
2229 			    tx->tx_txg - 1, tx);
2230 		}
2231 	}
2232 
2233 	/* Insert each entry into the on-disk livelist */
2234 	bplist_iterate(&dd->dd_pending_allocs,
2235 	    dsl_deadlist_insert_alloc_cb, &dd->dd_livelist, tx);
2236 	bplist_iterate(&dd->dd_pending_frees,
2237 	    dsl_deadlist_insert_free_cb, &dd->dd_livelist, tx);
2238 
2239 	/* Attempt to condense every pair of adjacent entries */
2240 	try_condense_arg_t arg = {
2241 	    .spa = spa,
2242 	    .ds = ds
2243 	};
2244 	dsl_deadlist_iterate(&dd->dd_livelist, dsl_livelist_try_condense,
2245 	    &arg);
2246 }
2247 
2248 void
2249 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
2250 {
2251 	objset_t *os = ds->ds_objset;
2252 
2253 	bplist_iterate(&ds->ds_pending_deadlist,
2254 	    dsl_deadlist_insert_alloc_cb, &ds->ds_deadlist, tx);
2255 
2256 	if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) {
2257 		dsl_flush_pending_livelist(ds, tx);
2258 		if (dsl_livelist_should_disable(ds)) {
2259 			dsl_dir_remove_livelist(ds->ds_dir, tx, B_TRUE);
2260 		}
2261 	}
2262 
2263 	dsl_bookmark_sync_done(ds, tx);
2264 
2265 	multilist_destroy(&os->os_synced_dnodes);
2266 
2267 	if (os->os_encrypted)
2268 		os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE;
2269 	else
2270 		ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]);
2271 
2272 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2273 		if (zfeature_active(f,
2274 		    ds->ds_feature_activation[f])) {
2275 			if (zfeature_active(f, ds->ds_feature[f]))
2276 				continue;
2277 			dsl_dataset_activate_feature(ds->ds_object, f,
2278 			    ds->ds_feature_activation[f], tx);
2279 			ds->ds_feature[f] = ds->ds_feature_activation[f];
2280 		}
2281 	}
2282 
2283 	ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
2284 }
2285 
2286 int
2287 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val)
2288 {
2289 	uint64_t count = 0;
2290 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2291 	zap_cursor_t zc;
2292 	zap_attribute_t za;
2293 
2294 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
2295 
2296 	/*
2297 	 * There may be missing entries in ds_next_clones_obj
2298 	 * due to a bug in a previous version of the code.
2299 	 * Only trust it if it has the right number of entries.
2300 	 */
2301 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
2302 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
2303 		    &count));
2304 	}
2305 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1) {
2306 		return (SET_ERROR(ENOENT));
2307 	}
2308 	for (zap_cursor_init(&zc, mos,
2309 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
2310 	    zap_cursor_retrieve(&zc, &za) == 0;
2311 	    zap_cursor_advance(&zc)) {
2312 		dsl_dataset_t *clone;
2313 		char buf[ZFS_MAX_DATASET_NAME_LEN];
2314 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2315 		    za.za_first_integer, FTAG, &clone));
2316 		dsl_dir_name(clone->ds_dir, buf);
2317 		fnvlist_add_boolean(val, buf);
2318 		dsl_dataset_rele(clone, FTAG);
2319 	}
2320 	zap_cursor_fini(&zc);
2321 	return (0);
2322 }
2323 
2324 void
2325 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2326 {
2327 	nvlist_t *propval = fnvlist_alloc();
2328 	nvlist_t *val = fnvlist_alloc();
2329 
2330 	if (get_clones_stat_impl(ds, val) == 0) {
2331 		fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
2332 		fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2333 		    propval);
2334 	}
2335 
2336 	nvlist_free(val);
2337 	nvlist_free(propval);
2338 }
2339 
2340 static char *
2341 get_receive_resume_token_impl(dsl_dataset_t *ds)
2342 {
2343 	if (!dsl_dataset_has_resume_receive_state(ds))
2344 		return (NULL);
2345 
2346 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2347 	char *str;
2348 	void *packed;
2349 	uint8_t *compressed;
2350 	uint64_t val;
2351 	nvlist_t *token_nv = fnvlist_alloc();
2352 	size_t packed_size, compressed_size;
2353 
2354 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2355 	    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
2356 		fnvlist_add_uint64(token_nv, "fromguid", val);
2357 	}
2358 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2359 	    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
2360 		fnvlist_add_uint64(token_nv, "object", val);
2361 	}
2362 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2363 	    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
2364 		fnvlist_add_uint64(token_nv, "offset", val);
2365 	}
2366 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2367 	    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
2368 		fnvlist_add_uint64(token_nv, "bytes", val);
2369 	}
2370 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2371 	    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
2372 		fnvlist_add_uint64(token_nv, "toguid", val);
2373 	}
2374 	char buf[MAXNAMELEN];
2375 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2376 	    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
2377 		fnvlist_add_string(token_nv, "toname", buf);
2378 	}
2379 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2380 	    DS_FIELD_RESUME_LARGEBLOCK) == 0) {
2381 		fnvlist_add_boolean(token_nv, "largeblockok");
2382 	}
2383 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2384 	    DS_FIELD_RESUME_EMBEDOK) == 0) {
2385 		fnvlist_add_boolean(token_nv, "embedok");
2386 	}
2387 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2388 	    DS_FIELD_RESUME_COMPRESSOK) == 0) {
2389 		fnvlist_add_boolean(token_nv, "compressok");
2390 	}
2391 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2392 	    DS_FIELD_RESUME_RAWOK) == 0) {
2393 		fnvlist_add_boolean(token_nv, "rawok");
2394 	}
2395 	if (dsl_dataset_feature_is_active(ds,
2396 	    SPA_FEATURE_REDACTED_DATASETS)) {
2397 		uint64_t num_redact_snaps = 0;
2398 		uint64_t *redact_snaps = NULL;
2399 		VERIFY3B(dsl_dataset_get_uint64_array_feature(ds,
2400 		    SPA_FEATURE_REDACTED_DATASETS, &num_redact_snaps,
2401 		    &redact_snaps), ==, B_TRUE);
2402 		fnvlist_add_uint64_array(token_nv, "redact_snaps",
2403 		    redact_snaps, num_redact_snaps);
2404 	}
2405 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2406 	    DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS) == 0) {
2407 		uint64_t num_redact_snaps = 0, int_size = 0;
2408 		uint64_t *redact_snaps = NULL;
2409 		VERIFY0(zap_length(dp->dp_meta_objset, ds->ds_object,
2410 		    DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, &int_size,
2411 		    &num_redact_snaps));
2412 		ASSERT3U(int_size, ==, sizeof (uint64_t));
2413 
2414 		redact_snaps = kmem_alloc(int_size * num_redact_snaps,
2415 		    KM_SLEEP);
2416 		VERIFY0(zap_lookup(dp->dp_meta_objset, ds->ds_object,
2417 		    DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, int_size,
2418 		    num_redact_snaps, redact_snaps));
2419 		fnvlist_add_uint64_array(token_nv, "book_redact_snaps",
2420 		    redact_snaps, num_redact_snaps);
2421 		kmem_free(redact_snaps, int_size * num_redact_snaps);
2422 	}
2423 	packed = fnvlist_pack(token_nv, &packed_size);
2424 	fnvlist_free(token_nv);
2425 	compressed = kmem_alloc(packed_size, KM_SLEEP);
2426 
2427 	compressed_size = gzip_compress(packed, compressed,
2428 	    packed_size, packed_size, 6);
2429 
2430 	zio_cksum_t cksum;
2431 	fletcher_4_native_varsize(compressed, compressed_size, &cksum);
2432 
2433 	size_t alloc_size = compressed_size * 2 + 1;
2434 	str = kmem_alloc(alloc_size, KM_SLEEP);
2435 	for (int i = 0; i < compressed_size; i++) {
2436 		size_t offset = i * 2;
2437 		(void) snprintf(str + offset, alloc_size - offset,
2438 	    "%02x", compressed[i]);
2439 	}
2440 	str[compressed_size * 2] = '\0';
2441 	char *propval = kmem_asprintf("%u-%llx-%llx-%s",
2442 	    ZFS_SEND_RESUME_TOKEN_VERSION,
2443 	    (longlong_t)cksum.zc_word[0],
2444 	    (longlong_t)packed_size, str);
2445 	kmem_free(packed, packed_size);
2446 	kmem_free(str, alloc_size);
2447 	kmem_free(compressed, packed_size);
2448 	return (propval);
2449 }
2450 
2451 /*
2452  * Returns a string that represents the receive resume state token. It should
2453  * be freed with strfree(). NULL is returned if no resume state is present.
2454  */
2455 char *
2456 get_receive_resume_token(dsl_dataset_t *ds)
2457 {
2458 	/*
2459 	 * A failed "newfs" (e.g. full) resumable receive leaves
2460 	 * the stats set on this dataset.  Check here for the prop.
2461 	 */
2462 	char *token = get_receive_resume_token_impl(ds);
2463 	if (token != NULL)
2464 		return (token);
2465 	/*
2466 	 * A failed incremental resumable receive leaves the
2467 	 * stats set on our child named "%recv".  Check the child
2468 	 * for the prop.
2469 	 */
2470 	/* 6 extra bytes for /%recv */
2471 	char name[ZFS_MAX_DATASET_NAME_LEN + 6];
2472 	dsl_dataset_t *recv_ds;
2473 	dsl_dataset_name(ds, name);
2474 	if (strlcat(name, "/", sizeof (name)) < sizeof (name) &&
2475 	    strlcat(name, recv_clone_name, sizeof (name)) < sizeof (name) &&
2476 	    dsl_dataset_hold(ds->ds_dir->dd_pool, name, FTAG, &recv_ds) == 0) {
2477 		token = get_receive_resume_token_impl(recv_ds);
2478 		dsl_dataset_rele(recv_ds, FTAG);
2479 	}
2480 	return (token);
2481 }
2482 
2483 uint64_t
2484 dsl_get_refratio(dsl_dataset_t *ds)
2485 {
2486 	uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
2487 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
2488 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
2489 	return (ratio);
2490 }
2491 
2492 uint64_t
2493 dsl_get_logicalreferenced(dsl_dataset_t *ds)
2494 {
2495 	return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
2496 }
2497 
2498 uint64_t
2499 dsl_get_compressratio(dsl_dataset_t *ds)
2500 {
2501 	if (ds->ds_is_snapshot) {
2502 		return (dsl_get_refratio(ds));
2503 	} else {
2504 		dsl_dir_t *dd = ds->ds_dir;
2505 		mutex_enter(&dd->dd_lock);
2506 		uint64_t val = dsl_dir_get_compressratio(dd);
2507 		mutex_exit(&dd->dd_lock);
2508 		return (val);
2509 	}
2510 }
2511 
2512 uint64_t
2513 dsl_get_used(dsl_dataset_t *ds)
2514 {
2515 	if (ds->ds_is_snapshot) {
2516 		return (dsl_dataset_phys(ds)->ds_unique_bytes);
2517 	} else {
2518 		dsl_dir_t *dd = ds->ds_dir;
2519 		mutex_enter(&dd->dd_lock);
2520 		uint64_t val = dsl_dir_get_used(dd);
2521 		mutex_exit(&dd->dd_lock);
2522 		return (val);
2523 	}
2524 }
2525 
2526 uint64_t
2527 dsl_get_creation(dsl_dataset_t *ds)
2528 {
2529 	return (dsl_dataset_phys(ds)->ds_creation_time);
2530 }
2531 
2532 uint64_t
2533 dsl_get_creationtxg(dsl_dataset_t *ds)
2534 {
2535 	return (dsl_dataset_phys(ds)->ds_creation_txg);
2536 }
2537 
2538 uint64_t
2539 dsl_get_refquota(dsl_dataset_t *ds)
2540 {
2541 	return (ds->ds_quota);
2542 }
2543 
2544 uint64_t
2545 dsl_get_refreservation(dsl_dataset_t *ds)
2546 {
2547 	return (ds->ds_reserved);
2548 }
2549 
2550 uint64_t
2551 dsl_get_guid(dsl_dataset_t *ds)
2552 {
2553 	return (dsl_dataset_phys(ds)->ds_guid);
2554 }
2555 
2556 uint64_t
2557 dsl_get_unique(dsl_dataset_t *ds)
2558 {
2559 	return (dsl_dataset_phys(ds)->ds_unique_bytes);
2560 }
2561 
2562 uint64_t
2563 dsl_get_objsetid(dsl_dataset_t *ds)
2564 {
2565 	return (ds->ds_object);
2566 }
2567 
2568 uint64_t
2569 dsl_get_userrefs(dsl_dataset_t *ds)
2570 {
2571 	return (ds->ds_userrefs);
2572 }
2573 
2574 uint64_t
2575 dsl_get_defer_destroy(dsl_dataset_t *ds)
2576 {
2577 	return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2578 }
2579 
2580 uint64_t
2581 dsl_get_referenced(dsl_dataset_t *ds)
2582 {
2583 	return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2584 }
2585 
2586 uint64_t
2587 dsl_get_numclones(dsl_dataset_t *ds)
2588 {
2589 	ASSERT(ds->ds_is_snapshot);
2590 	return (dsl_dataset_phys(ds)->ds_num_children - 1);
2591 }
2592 
2593 uint64_t
2594 dsl_get_inconsistent(dsl_dataset_t *ds)
2595 {
2596 	return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2597 	    1 : 0);
2598 }
2599 
2600 uint64_t
2601 dsl_get_redacted(dsl_dataset_t *ds)
2602 {
2603 	return (dsl_dataset_feature_is_active(ds,
2604 	    SPA_FEATURE_REDACTED_DATASETS));
2605 }
2606 
2607 uint64_t
2608 dsl_get_available(dsl_dataset_t *ds)
2609 {
2610 	uint64_t refdbytes = dsl_get_referenced(ds);
2611 	uint64_t availbytes = dsl_dir_space_available(ds->ds_dir,
2612 	    NULL, 0, TRUE);
2613 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2614 		availbytes +=
2615 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2616 	}
2617 	if (ds->ds_quota != 0) {
2618 		/*
2619 		 * Adjust available bytes according to refquota
2620 		 */
2621 		if (refdbytes < ds->ds_quota) {
2622 			availbytes = MIN(availbytes,
2623 			    ds->ds_quota - refdbytes);
2624 		} else {
2625 			availbytes = 0;
2626 		}
2627 	}
2628 	return (availbytes);
2629 }
2630 
2631 int
2632 dsl_get_written(dsl_dataset_t *ds, uint64_t *written)
2633 {
2634 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2635 	dsl_dataset_t *prev;
2636 	int err = dsl_dataset_hold_obj(dp,
2637 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2638 	if (err == 0) {
2639 		uint64_t comp, uncomp;
2640 		err = dsl_dataset_space_written(prev, ds, written,
2641 		    &comp, &uncomp);
2642 		dsl_dataset_rele(prev, FTAG);
2643 	}
2644 	return (err);
2645 }
2646 
2647 /*
2648  * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2649  */
2650 int
2651 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap)
2652 {
2653 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2654 	if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
2655 		dsl_dataset_name(ds->ds_prev, snap);
2656 		return (0);
2657 	} else {
2658 		return (SET_ERROR(ENOENT));
2659 	}
2660 }
2661 
2662 void
2663 dsl_get_redact_snaps(dsl_dataset_t *ds, nvlist_t *propval)
2664 {
2665 	uint64_t nsnaps;
2666 	uint64_t *snaps;
2667 	if (dsl_dataset_get_uint64_array_feature(ds,
2668 	    SPA_FEATURE_REDACTED_DATASETS, &nsnaps, &snaps)) {
2669 		fnvlist_add_uint64_array(propval, ZPROP_VALUE, snaps,
2670 		    nsnaps);
2671 	}
2672 }
2673 
2674 /*
2675  * Returns the mountpoint property and source for the given dataset in the value
2676  * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2677  * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2678  * Returns 0 on success and an error on failure.
2679  */
2680 int
2681 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value,
2682     char *source)
2683 {
2684 	int error;
2685 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2686 
2687 	/* Retrieve the mountpoint value stored in the zap object */
2688 	error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1,
2689 	    ZAP_MAXVALUELEN, value, source);
2690 	if (error != 0) {
2691 		return (error);
2692 	}
2693 
2694 	/*
2695 	 * Process the dsname and source to find the full mountpoint string.
2696 	 * Can be skipped for 'legacy' or 'none'.
2697 	 */
2698 	if (value[0] == '/') {
2699 		char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
2700 		char *root = buf;
2701 		const char *relpath;
2702 
2703 		/*
2704 		 * If we inherit the mountpoint, even from a dataset
2705 		 * with a received value, the source will be the path of
2706 		 * the dataset we inherit from. If source is
2707 		 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2708 		 * inherited.
2709 		 */
2710 		if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) {
2711 			relpath = "";
2712 		} else {
2713 			ASSERT0(strncmp(dsname, source, strlen(source)));
2714 			relpath = dsname + strlen(source);
2715 			if (relpath[0] == '/')
2716 				relpath++;
2717 		}
2718 
2719 		spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN);
2720 
2721 		/*
2722 		 * Special case an alternate root of '/'. This will
2723 		 * avoid having multiple leading slashes in the
2724 		 * mountpoint path.
2725 		 */
2726 		if (strcmp(root, "/") == 0)
2727 			root++;
2728 
2729 		/*
2730 		 * If the mountpoint is '/' then skip over this
2731 		 * if we are obtaining either an alternate root or
2732 		 * an inherited mountpoint.
2733 		 */
2734 		char *mnt = value;
2735 		if (value[1] == '\0' && (root[0] != '\0' ||
2736 		    relpath[0] != '\0'))
2737 			mnt = value + 1;
2738 
2739 		mnt = kmem_strdup(mnt);
2740 
2741 		if (relpath[0] == '\0') {
2742 			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s",
2743 			    root, mnt);
2744 		} else {
2745 			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s",
2746 			    root, mnt, relpath[0] == '@' ? "" : "/",
2747 			    relpath);
2748 		}
2749 		kmem_free(buf, ZAP_MAXVALUELEN);
2750 		kmem_strfree(mnt);
2751 	}
2752 
2753 	return (0);
2754 }
2755 
2756 void
2757 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2758 {
2759 	dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2760 
2761 	ASSERT(dsl_pool_config_held(dp));
2762 
2763 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO,
2764 	    dsl_get_refratio(ds));
2765 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
2766 	    dsl_get_logicalreferenced(ds));
2767 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2768 	    dsl_get_compressratio(ds));
2769 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2770 	    dsl_get_used(ds));
2771 
2772 	if (ds->ds_is_snapshot) {
2773 		get_clones_stat(ds, nv);
2774 	} else {
2775 		char buf[ZFS_MAX_DATASET_NAME_LEN];
2776 		if (dsl_get_prev_snap(ds, buf) == 0)
2777 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP,
2778 			    buf);
2779 		dsl_dir_stats(ds->ds_dir, nv);
2780 	}
2781 
2782 	nvlist_t *propval = fnvlist_alloc();
2783 	dsl_get_redact_snaps(ds, propval);
2784 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS),
2785 	    propval);
2786 	nvlist_free(propval);
2787 
2788 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE,
2789 	    dsl_get_available(ds));
2790 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
2791 	    dsl_get_referenced(ds));
2792 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2793 	    dsl_get_creation(ds));
2794 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2795 	    dsl_get_creationtxg(ds));
2796 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2797 	    dsl_get_refquota(ds));
2798 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2799 	    dsl_get_refreservation(ds));
2800 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2801 	    dsl_get_guid(ds));
2802 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2803 	    dsl_get_unique(ds));
2804 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2805 	    dsl_get_objsetid(ds));
2806 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2807 	    dsl_get_userrefs(ds));
2808 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2809 	    dsl_get_defer_destroy(ds));
2810 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOTS_CHANGED,
2811 	    dsl_dir_snap_cmtime(ds->ds_dir).tv_sec);
2812 	dsl_dataset_crypt_stats(ds, nv);
2813 
2814 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
2815 		uint64_t written;
2816 		if (dsl_get_written(ds, &written) == 0) {
2817 			dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2818 			    written);
2819 		}
2820 	}
2821 
2822 	if (!dsl_dataset_is_snapshot(ds)) {
2823 		char *token = get_receive_resume_token(ds);
2824 		if (token != NULL) {
2825 			dsl_prop_nvlist_add_string(nv,
2826 			    ZFS_PROP_RECEIVE_RESUME_TOKEN, token);
2827 			kmem_strfree(token);
2828 		}
2829 	}
2830 }
2831 
2832 void
2833 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2834 {
2835 	dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2836 	ASSERT(dsl_pool_config_held(dp));
2837 
2838 	stat->dds_creation_txg = dsl_get_creationtxg(ds);
2839 	stat->dds_inconsistent = dsl_get_inconsistent(ds);
2840 	stat->dds_guid = dsl_get_guid(ds);
2841 	stat->dds_redacted = dsl_get_redacted(ds);
2842 	stat->dds_origin[0] = '\0';
2843 	if (ds->ds_is_snapshot) {
2844 		stat->dds_is_snapshot = B_TRUE;
2845 		stat->dds_num_clones = dsl_get_numclones(ds);
2846 	} else {
2847 		stat->dds_is_snapshot = B_FALSE;
2848 		stat->dds_num_clones = 0;
2849 
2850 		if (dsl_dir_is_clone(ds->ds_dir)) {
2851 			dsl_dir_get_origin(ds->ds_dir, stat->dds_origin);
2852 		}
2853 	}
2854 }
2855 
2856 uint64_t
2857 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2858 {
2859 	return (ds->ds_fsid_guid);
2860 }
2861 
2862 void
2863 dsl_dataset_space(dsl_dataset_t *ds,
2864     uint64_t *refdbytesp, uint64_t *availbytesp,
2865     uint64_t *usedobjsp, uint64_t *availobjsp)
2866 {
2867 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2868 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2869 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2870 		*availbytesp +=
2871 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2872 	if (ds->ds_quota != 0) {
2873 		/*
2874 		 * Adjust available bytes according to refquota
2875 		 */
2876 		if (*refdbytesp < ds->ds_quota)
2877 			*availbytesp = MIN(*availbytesp,
2878 			    ds->ds_quota - *refdbytesp);
2879 		else
2880 			*availbytesp = 0;
2881 	}
2882 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2883 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2884 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2885 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2886 }
2887 
2888 boolean_t
2889 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2890 {
2891 	dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2892 	uint64_t birth;
2893 
2894 	ASSERT(dsl_pool_config_held(dp));
2895 	if (snap == NULL)
2896 		return (B_FALSE);
2897 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2898 	birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2899 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2900 	if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2901 		objset_t *os, *os_snap;
2902 		/*
2903 		 * It may be that only the ZIL differs, because it was
2904 		 * reset in the head.  Don't count that as being
2905 		 * modified.
2906 		 */
2907 		if (dmu_objset_from_ds(ds, &os) != 0)
2908 			return (B_TRUE);
2909 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
2910 			return (B_TRUE);
2911 		return (memcmp(&os->os_phys->os_meta_dnode,
2912 		    &os_snap->os_phys->os_meta_dnode,
2913 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
2914 	}
2915 	return (B_FALSE);
2916 }
2917 
2918 static int
2919 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2920     dsl_dataset_t *hds, void *arg)
2921 {
2922 	(void) dp;
2923 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2924 	int error;
2925 	uint64_t val;
2926 
2927 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2928 	if (error != 0) {
2929 		/* ignore nonexistent snapshots */
2930 		return (error == ENOENT ? 0 : error);
2931 	}
2932 
2933 	/* new name should not exist */
2934 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2935 	if (error == 0)
2936 		error = SET_ERROR(EEXIST);
2937 	else if (error == ENOENT)
2938 		error = 0;
2939 
2940 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2941 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
2942 	    strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2943 		error = SET_ERROR(ENAMETOOLONG);
2944 
2945 	return (error);
2946 }
2947 
2948 int
2949 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2950 {
2951 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2952 	dsl_pool_t *dp = dmu_tx_pool(tx);
2953 	dsl_dataset_t *hds;
2954 	int error;
2955 
2956 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2957 	if (error != 0)
2958 		return (error);
2959 
2960 	if (ddrsa->ddrsa_recursive) {
2961 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2962 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2963 		    DS_FIND_CHILDREN);
2964 	} else {
2965 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2966 	}
2967 	dsl_dataset_rele(hds, FTAG);
2968 	return (error);
2969 }
2970 
2971 static int
2972 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2973     dsl_dataset_t *hds, void *arg)
2974 {
2975 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2976 	dsl_dataset_t *ds;
2977 	uint64_t val;
2978 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2979 	int error;
2980 
2981 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2982 	ASSERT(error == 0 || error == ENOENT);
2983 	if (error == ENOENT) {
2984 		/* ignore nonexistent snapshots */
2985 		return (0);
2986 	}
2987 
2988 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2989 
2990 	/* log before we change the name */
2991 	spa_history_log_internal_ds(ds, "rename", tx,
2992 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2993 
2994 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2995 	    B_FALSE));
2996 	mutex_enter(&ds->ds_lock);
2997 	(void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname,
2998 	    sizeof (ds->ds_snapname));
2999 	mutex_exit(&ds->ds_lock);
3000 	VERIFY0(zap_add(dp->dp_meta_objset,
3001 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
3002 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
3003 	zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname,
3004 	    ddrsa->ddrsa_newsnapname, B_TRUE);
3005 
3006 	dsl_dataset_rele(ds, FTAG);
3007 	return (0);
3008 }
3009 
3010 void
3011 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
3012 {
3013 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
3014 	dsl_pool_t *dp = dmu_tx_pool(tx);
3015 	dsl_dataset_t *hds = NULL;
3016 
3017 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
3018 	ddrsa->ddrsa_tx = tx;
3019 	if (ddrsa->ddrsa_recursive) {
3020 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
3021 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
3022 		    DS_FIND_CHILDREN));
3023 	} else {
3024 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
3025 	}
3026 	dsl_dataset_rele(hds, FTAG);
3027 }
3028 
3029 int
3030 dsl_dataset_rename_snapshot(const char *fsname,
3031     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
3032 {
3033 	dsl_dataset_rename_snapshot_arg_t ddrsa;
3034 
3035 	ddrsa.ddrsa_fsname = fsname;
3036 	ddrsa.ddrsa_oldsnapname = oldsnapname;
3037 	ddrsa.ddrsa_newsnapname = newsnapname;
3038 	ddrsa.ddrsa_recursive = recursive;
3039 
3040 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
3041 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
3042 	    1, ZFS_SPACE_CHECK_RESERVED));
3043 }
3044 
3045 /*
3046  * If we're doing an ownership handoff, we need to make sure that there is
3047  * only one long hold on the dataset.  We're not allowed to change anything here
3048  * so we don't permanently release the long hold or regular hold here.  We want
3049  * to do this only when syncing to avoid the dataset unexpectedly going away
3050  * when we release the long hold.
3051  */
3052 static int
3053 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
3054 {
3055 	boolean_t held = B_FALSE;
3056 
3057 	if (!dmu_tx_is_syncing(tx))
3058 		return (0);
3059 
3060 	dsl_dir_t *dd = ds->ds_dir;
3061 	mutex_enter(&dd->dd_activity_lock);
3062 	uint64_t holds = zfs_refcount_count(&ds->ds_longholds) -
3063 	    (owner != NULL ? 1 : 0);
3064 	/*
3065 	 * The value of dd_activity_waiters can chance as soon as we drop the
3066 	 * lock, but we're fine with that; new waiters coming in or old
3067 	 * waiters leaving doesn't cause problems, since we're going to cancel
3068 	 * waiters later anyway. The goal of this check is to verify that no
3069 	 * non-waiters have long-holds, and all new long-holds will be
3070 	 * prevented because we're holding the pool config as writer.
3071 	 */
3072 	if (holds != dd->dd_activity_waiters)
3073 		held = B_TRUE;
3074 	mutex_exit(&dd->dd_activity_lock);
3075 
3076 	if (held)
3077 		return (SET_ERROR(EBUSY));
3078 
3079 	return (0);
3080 }
3081 
3082 int
3083 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
3084 {
3085 	dsl_dataset_rollback_arg_t *ddra = arg;
3086 	dsl_pool_t *dp = dmu_tx_pool(tx);
3087 	dsl_dataset_t *ds;
3088 	int64_t unused_refres_delta;
3089 	int error;
3090 
3091 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
3092 	if (error != 0)
3093 		return (error);
3094 
3095 	/* must not be a snapshot */
3096 	if (ds->ds_is_snapshot) {
3097 		dsl_dataset_rele(ds, FTAG);
3098 		return (SET_ERROR(EINVAL));
3099 	}
3100 
3101 	/* must have a most recent snapshot */
3102 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
3103 		dsl_dataset_rele(ds, FTAG);
3104 		return (SET_ERROR(ESRCH));
3105 	}
3106 
3107 	/*
3108 	 * No rollback to a snapshot created in the current txg, because
3109 	 * the rollback may dirty the dataset and create blocks that are
3110 	 * not reachable from the rootbp while having a birth txg that
3111 	 * falls into the snapshot's range.
3112 	 */
3113 	if (dmu_tx_is_syncing(tx) &&
3114 	    dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
3115 		dsl_dataset_rele(ds, FTAG);
3116 		return (SET_ERROR(EAGAIN));
3117 	}
3118 
3119 	/*
3120 	 * If the expected target snapshot is specified, then check that
3121 	 * the latest snapshot is it.
3122 	 */
3123 	if (ddra->ddra_tosnap != NULL) {
3124 		dsl_dataset_t *snapds;
3125 
3126 		/* Check if the target snapshot exists at all. */
3127 		error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
3128 		if (error != 0) {
3129 			/*
3130 			 * ESRCH is used to signal that the target snapshot does
3131 			 * not exist, while ENOENT is used to report that
3132 			 * the rolled back dataset does not exist.
3133 			 * ESRCH is also used to cover other cases where the
3134 			 * target snapshot is not related to the dataset being
3135 			 * rolled back such as being in a different pool.
3136 			 */
3137 			if (error == ENOENT || error == EXDEV)
3138 				error = SET_ERROR(ESRCH);
3139 			dsl_dataset_rele(ds, FTAG);
3140 			return (error);
3141 		}
3142 		ASSERT(snapds->ds_is_snapshot);
3143 
3144 		/* Check if the snapshot is the latest snapshot indeed. */
3145 		if (snapds != ds->ds_prev) {
3146 			/*
3147 			 * Distinguish between the case where the only problem
3148 			 * is intervening snapshots (EEXIST) vs the snapshot
3149 			 * not being a valid target for rollback (ESRCH).
3150 			 */
3151 			if (snapds->ds_dir == ds->ds_dir ||
3152 			    (dsl_dir_is_clone(ds->ds_dir) &&
3153 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
3154 			    snapds->ds_object)) {
3155 				error = SET_ERROR(EEXIST);
3156 			} else {
3157 				error = SET_ERROR(ESRCH);
3158 			}
3159 			dsl_dataset_rele(snapds, FTAG);
3160 			dsl_dataset_rele(ds, FTAG);
3161 			return (error);
3162 		}
3163 		dsl_dataset_rele(snapds, FTAG);
3164 	}
3165 
3166 	/* must not have any bookmarks after the most recent snapshot */
3167 	if (dsl_bookmark_latest_txg(ds) >
3168 	    dsl_dataset_phys(ds)->ds_prev_snap_txg) {
3169 		dsl_dataset_rele(ds, FTAG);
3170 		return (SET_ERROR(EEXIST));
3171 	}
3172 
3173 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
3174 	if (error != 0) {
3175 		dsl_dataset_rele(ds, FTAG);
3176 		return (error);
3177 	}
3178 
3179 	/*
3180 	 * Check if the snap we are rolling back to uses more than
3181 	 * the refquota.
3182 	 */
3183 	if (ds->ds_quota != 0 &&
3184 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
3185 		dsl_dataset_rele(ds, FTAG);
3186 		return (SET_ERROR(EDQUOT));
3187 	}
3188 
3189 	/*
3190 	 * When we do the clone swap, we will temporarily use more space
3191 	 * due to the refreservation (the head will no longer have any
3192 	 * unique space, so the entire amount of the refreservation will need
3193 	 * to be free).  We will immediately destroy the clone, freeing
3194 	 * this space, but the freeing happens over many txg's.
3195 	 */
3196 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
3197 	    dsl_dataset_phys(ds)->ds_unique_bytes);
3198 
3199 	if (unused_refres_delta > 0 &&
3200 	    unused_refres_delta >
3201 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
3202 		dsl_dataset_rele(ds, FTAG);
3203 		return (SET_ERROR(ENOSPC));
3204 	}
3205 
3206 	dsl_dataset_rele(ds, FTAG);
3207 	return (0);
3208 }
3209 
3210 void
3211 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
3212 {
3213 	dsl_dataset_rollback_arg_t *ddra = arg;
3214 	dsl_pool_t *dp = dmu_tx_pool(tx);
3215 	dsl_dataset_t *ds, *clone;
3216 	uint64_t cloneobj;
3217 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
3218 
3219 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
3220 
3221 	dsl_dataset_name(ds->ds_prev, namebuf);
3222 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
3223 
3224 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
3225 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx);
3226 
3227 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
3228 
3229 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
3230 	dsl_dataset_zero_zil(ds, tx);
3231 
3232 	dsl_destroy_head_sync_impl(clone, tx);
3233 
3234 	dsl_dataset_rele(clone, FTAG);
3235 	dsl_dataset_rele(ds, FTAG);
3236 }
3237 
3238 /*
3239  * Rolls back the given filesystem or volume to the most recent snapshot.
3240  * The name of the most recent snapshot will be returned under key "target"
3241  * in the result nvlist.
3242  *
3243  * If owner != NULL:
3244  * - The existing dataset MUST be owned by the specified owner at entry
3245  * - Upon return, dataset will still be held by the same owner, whether we
3246  *   succeed or not.
3247  *
3248  * This mode is required any time the existing filesystem is mounted.  See
3249  * notes above zfs_suspend_fs() for further details.
3250  */
3251 int
3252 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
3253     nvlist_t *result)
3254 {
3255 	dsl_dataset_rollback_arg_t ddra;
3256 
3257 	ddra.ddra_fsname = fsname;
3258 	ddra.ddra_tosnap = tosnap;
3259 	ddra.ddra_owner = owner;
3260 	ddra.ddra_result = result;
3261 
3262 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
3263 	    dsl_dataset_rollback_sync, &ddra,
3264 	    1, ZFS_SPACE_CHECK_RESERVED));
3265 }
3266 
3267 struct promotenode {
3268 	list_node_t link;
3269 	dsl_dataset_t *ds;
3270 };
3271 
3272 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
3273 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
3274     const void *tag);
3275 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag);
3276 
3277 int
3278 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
3279 {
3280 	dsl_dataset_promote_arg_t *ddpa = arg;
3281 	dsl_pool_t *dp = dmu_tx_pool(tx);
3282 	dsl_dataset_t *hds;
3283 	struct promotenode *snap;
3284 	int err;
3285 	uint64_t unused;
3286 	uint64_t ss_mv_cnt;
3287 	size_t max_snap_len;
3288 	boolean_t conflicting_snaps;
3289 
3290 	err = promote_hold(ddpa, dp, FTAG);
3291 	if (err != 0)
3292 		return (err);
3293 
3294 	hds = ddpa->ddpa_clone;
3295 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
3296 
3297 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
3298 		promote_rele(ddpa, FTAG);
3299 		return (SET_ERROR(EXDEV));
3300 	}
3301 
3302 	snap = list_head(&ddpa->shared_snaps);
3303 	if (snap == NULL) {
3304 		err = SET_ERROR(ENOENT);
3305 		goto out;
3306 	}
3307 	dsl_dataset_t *const origin_ds = snap->ds;
3308 
3309 	/*
3310 	 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir.
3311 	 * When doing a promote we must make sure the encryption root for
3312 	 * both the target and the target's origin does not change to avoid
3313 	 * needing to rewrap encryption keys
3314 	 */
3315 	err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir);
3316 	if (err != 0)
3317 		goto out;
3318 
3319 	/*
3320 	 * Compute and check the amount of space to transfer.  Since this is
3321 	 * so expensive, don't do the preliminary check.
3322 	 */
3323 	if (!dmu_tx_is_syncing(tx)) {
3324 		promote_rele(ddpa, FTAG);
3325 		return (0);
3326 	}
3327 
3328 	/* compute origin's new unique space */
3329 	snap = list_tail(&ddpa->clone_snaps);
3330 	ASSERT(snap != NULL);
3331 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3332 	    origin_ds->ds_object);
3333 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3334 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
3335 	    &ddpa->unique, &unused, &unused);
3336 
3337 	/*
3338 	 * Walk the snapshots that we are moving
3339 	 *
3340 	 * Compute space to transfer.  Consider the incremental changes
3341 	 * to used by each snapshot:
3342 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
3343 	 * So each snapshot gave birth to:
3344 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
3345 	 * So a sequence would look like:
3346 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
3347 	 * Which simplifies to:
3348 	 * uN + kN + kN-1 + ... + k1 + k0
3349 	 * Note however, if we stop before we reach the ORIGIN we get:
3350 	 * uN + kN + kN-1 + ... + kM - uM-1
3351 	 */
3352 	conflicting_snaps = B_FALSE;
3353 	ss_mv_cnt = 0;
3354 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
3355 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
3356 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
3357 	for (snap = list_head(&ddpa->shared_snaps); snap;
3358 	    snap = list_next(&ddpa->shared_snaps, snap)) {
3359 		uint64_t val, dlused, dlcomp, dluncomp;
3360 		dsl_dataset_t *ds = snap->ds;
3361 
3362 		ss_mv_cnt++;
3363 
3364 		/*
3365 		 * If there are long holds, we won't be able to evict
3366 		 * the objset.
3367 		 */
3368 		if (dsl_dataset_long_held(ds)) {
3369 			err = SET_ERROR(EBUSY);
3370 			goto out;
3371 		}
3372 
3373 		/* Check that the snapshot name does not conflict */
3374 		VERIFY0(dsl_dataset_get_snapname(ds));
3375 		if (strlen(ds->ds_snapname) >= max_snap_len) {
3376 			err = SET_ERROR(ENAMETOOLONG);
3377 			goto out;
3378 		}
3379 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
3380 		if (err == 0) {
3381 			fnvlist_add_boolean(ddpa->err_ds,
3382 			    snap->ds->ds_snapname);
3383 			conflicting_snaps = B_TRUE;
3384 		} else if (err != ENOENT) {
3385 			goto out;
3386 		}
3387 
3388 		/* The very first snapshot does not have a deadlist */
3389 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
3390 			continue;
3391 
3392 		dsl_deadlist_space(&ds->ds_deadlist,
3393 		    &dlused, &dlcomp, &dluncomp);
3394 		ddpa->used += dlused;
3395 		ddpa->comp += dlcomp;
3396 		ddpa->uncomp += dluncomp;
3397 	}
3398 
3399 	/*
3400 	 * Check that bookmarks that are being transferred don't have
3401 	 * name conflicts.
3402 	 */
3403 	for (dsl_bookmark_node_t *dbn = avl_first(&origin_ds->ds_bookmarks);
3404 	    dbn != NULL && dbn->dbn_phys.zbm_creation_txg <=
3405 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3406 	    dbn = AVL_NEXT(&origin_ds->ds_bookmarks, dbn)) {
3407 		if (strlen(dbn->dbn_name) >= max_snap_len) {
3408 			err = SET_ERROR(ENAMETOOLONG);
3409 			goto out;
3410 		}
3411 		zfs_bookmark_phys_t bm;
3412 		err = dsl_bookmark_lookup_impl(ddpa->ddpa_clone,
3413 		    dbn->dbn_name, &bm);
3414 
3415 		if (err == 0) {
3416 			fnvlist_add_boolean(ddpa->err_ds, dbn->dbn_name);
3417 			conflicting_snaps = B_TRUE;
3418 		} else if (err == ESRCH) {
3419 			err = 0;
3420 		}
3421 		if (err != 0) {
3422 			goto out;
3423 		}
3424 	}
3425 
3426 	/*
3427 	 * In order to return the full list of conflicting snapshots, we check
3428 	 * whether there was a conflict after traversing all of them.
3429 	 */
3430 	if (conflicting_snaps) {
3431 		err = SET_ERROR(EEXIST);
3432 		goto out;
3433 	}
3434 
3435 	/*
3436 	 * If we are a clone of a clone then we never reached ORIGIN,
3437 	 * so we need to subtract out the clone origin's used space.
3438 	 */
3439 	if (ddpa->origin_origin) {
3440 		ddpa->used -=
3441 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
3442 		ddpa->comp -=
3443 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
3444 		ddpa->uncomp -=
3445 		    dsl_dataset_phys(ddpa->origin_origin)->
3446 		    ds_uncompressed_bytes;
3447 	}
3448 
3449 	/* Check that there is enough space and limit headroom here */
3450 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
3451 	    0, ss_mv_cnt, ddpa->used, ddpa->cr, ddpa->proc);
3452 	if (err != 0)
3453 		goto out;
3454 
3455 	/*
3456 	 * Compute the amounts of space that will be used by snapshots
3457 	 * after the promotion (for both origin and clone).  For each,
3458 	 * it is the amount of space that will be on all of their
3459 	 * deadlists (that was not born before their new origin).
3460 	 */
3461 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
3462 		uint64_t space;
3463 
3464 		/*
3465 		 * Note, typically this will not be a clone of a clone,
3466 		 * so dd_origin_txg will be < TXG_INITIAL, so
3467 		 * these snaplist_space() -> dsl_deadlist_space_range()
3468 		 * calls will be fast because they do not have to
3469 		 * iterate over all bps.
3470 		 */
3471 		snap = list_head(&ddpa->origin_snaps);
3472 		if (snap == NULL) {
3473 			err = SET_ERROR(ENOENT);
3474 			goto out;
3475 		}
3476 		err = snaplist_space(&ddpa->shared_snaps,
3477 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
3478 		if (err != 0)
3479 			goto out;
3480 
3481 		err = snaplist_space(&ddpa->clone_snaps,
3482 		    snap->ds->ds_dir->dd_origin_txg, &space);
3483 		if (err != 0)
3484 			goto out;
3485 		ddpa->cloneusedsnap += space;
3486 	}
3487 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
3488 	    DD_FLAG_USED_BREAKDOWN) {
3489 		err = snaplist_space(&ddpa->origin_snaps,
3490 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
3491 		    &ddpa->originusedsnap);
3492 		if (err != 0)
3493 			goto out;
3494 	}
3495 
3496 out:
3497 	promote_rele(ddpa, FTAG);
3498 	return (err);
3499 }
3500 
3501 void
3502 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
3503 {
3504 	dsl_dataset_promote_arg_t *ddpa = arg;
3505 	dsl_pool_t *dp = dmu_tx_pool(tx);
3506 	dsl_dataset_t *hds;
3507 	struct promotenode *snap;
3508 	dsl_dataset_t *origin_ds;
3509 	dsl_dataset_t *origin_head;
3510 	dsl_dir_t *dd;
3511 	dsl_dir_t *odd = NULL;
3512 	uint64_t oldnext_obj;
3513 	int64_t delta;
3514 
3515 	ASSERT(nvlist_empty(ddpa->err_ds));
3516 
3517 	VERIFY0(promote_hold(ddpa, dp, FTAG));
3518 	hds = ddpa->ddpa_clone;
3519 
3520 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
3521 
3522 	snap = list_head(&ddpa->shared_snaps);
3523 	origin_ds = snap->ds;
3524 	dd = hds->ds_dir;
3525 
3526 	snap = list_head(&ddpa->origin_snaps);
3527 	origin_head = snap->ds;
3528 
3529 	/*
3530 	 * We need to explicitly open odd, since origin_ds's dd will be
3531 	 * changing.
3532 	 */
3533 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
3534 	    NULL, FTAG, &odd));
3535 
3536 	dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx);
3537 
3538 	/* change origin's next snap */
3539 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
3540 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
3541 	snap = list_tail(&ddpa->clone_snaps);
3542 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3543 	    origin_ds->ds_object);
3544 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
3545 
3546 	/* change the origin's next clone */
3547 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
3548 		dsl_dataset_remove_from_next_clones(origin_ds,
3549 		    snap->ds->ds_object, tx);
3550 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3551 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
3552 		    oldnext_obj, tx));
3553 	}
3554 
3555 	/* change origin */
3556 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
3557 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
3558 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
3559 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
3560 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
3561 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
3562 	origin_head->ds_dir->dd_origin_txg =
3563 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3564 
3565 	/* change dd_clone entries */
3566 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3567 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3568 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
3569 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3570 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3571 		    hds->ds_object, tx));
3572 
3573 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3574 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3575 		    origin_head->ds_object, tx));
3576 		if (dsl_dir_phys(dd)->dd_clones == 0) {
3577 			dsl_dir_phys(dd)->dd_clones =
3578 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
3579 			    DMU_OT_NONE, 0, tx);
3580 		}
3581 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3582 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
3583 	}
3584 
3585 	/*
3586 	 * Move bookmarks to this dir.
3587 	 */
3588 	dsl_bookmark_node_t *dbn_next;
3589 	for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks);
3590 	    dbn != NULL && dbn->dbn_phys.zbm_creation_txg <=
3591 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3592 	    dbn = dbn_next) {
3593 		dbn_next = AVL_NEXT(&origin_head->ds_bookmarks, dbn);
3594 
3595 		avl_remove(&origin_head->ds_bookmarks, dbn);
3596 		VERIFY0(zap_remove(dp->dp_meta_objset,
3597 		    origin_head->ds_bookmarks_obj, dbn->dbn_name, tx));
3598 
3599 		dsl_bookmark_node_add(hds, dbn, tx);
3600 	}
3601 
3602 	dsl_bookmark_next_changed(hds, origin_ds, tx);
3603 
3604 	/* move snapshots to this dir */
3605 	for (snap = list_head(&ddpa->shared_snaps); snap;
3606 	    snap = list_next(&ddpa->shared_snaps, snap)) {
3607 		dsl_dataset_t *ds = snap->ds;
3608 
3609 		/*
3610 		 * Property callbacks are registered to a particular
3611 		 * dsl_dir.  Since ours is changing, evict the objset
3612 		 * so that they will be unregistered from the old dsl_dir.
3613 		 */
3614 		if (ds->ds_objset) {
3615 			dmu_objset_evict(ds->ds_objset);
3616 			ds->ds_objset = NULL;
3617 		}
3618 
3619 		/* move snap name entry */
3620 		VERIFY0(dsl_dataset_get_snapname(ds));
3621 		VERIFY0(dsl_dataset_snap_remove(origin_head,
3622 		    ds->ds_snapname, tx, B_TRUE));
3623 		VERIFY0(zap_add(dp->dp_meta_objset,
3624 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3625 		    8, 1, &ds->ds_object, tx));
3626 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3627 		    DD_FIELD_SNAPSHOT_COUNT, tx);
3628 
3629 		/* change containing dsl_dir */
3630 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3631 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3632 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3633 		ASSERT3P(ds->ds_dir, ==, odd);
3634 		dsl_dir_rele(ds->ds_dir, ds);
3635 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3636 		    NULL, ds, &ds->ds_dir));
3637 
3638 		/* move any clone references */
3639 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3640 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3641 			zap_cursor_t zc;
3642 			zap_attribute_t za;
3643 
3644 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
3645 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
3646 			    zap_cursor_retrieve(&zc, &za) == 0;
3647 			    zap_cursor_advance(&zc)) {
3648 				dsl_dataset_t *cnds;
3649 				uint64_t o;
3650 
3651 				if (za.za_first_integer == oldnext_obj) {
3652 					/*
3653 					 * We've already moved the
3654 					 * origin's reference.
3655 					 */
3656 					continue;
3657 				}
3658 
3659 				VERIFY0(dsl_dataset_hold_obj(dp,
3660 				    za.za_first_integer, FTAG, &cnds));
3661 				o = dsl_dir_phys(cnds->ds_dir)->
3662 				    dd_head_dataset_obj;
3663 
3664 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
3665 				    dsl_dir_phys(odd)->dd_clones, o, tx));
3666 				VERIFY0(zap_add_int(dp->dp_meta_objset,
3667 				    dsl_dir_phys(dd)->dd_clones, o, tx));
3668 				dsl_dataset_rele(cnds, FTAG);
3669 			}
3670 			zap_cursor_fini(&zc);
3671 		}
3672 
3673 		ASSERT(!dsl_prop_hascb(ds));
3674 	}
3675 
3676 	/*
3677 	 * Change space accounting.
3678 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3679 	 * both be valid, or both be 0 (resulting in delta == 0).  This
3680 	 * is true for each of {clone,origin} independently.
3681 	 */
3682 
3683 	delta = ddpa->cloneusedsnap -
3684 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3685 	ASSERT3S(delta, >=, 0);
3686 	ASSERT3U(ddpa->used, >=, delta);
3687 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3688 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
3689 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3690 
3691 	delta = ddpa->originusedsnap -
3692 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3693 	ASSERT3S(delta, <=, 0);
3694 	ASSERT3U(ddpa->used, >=, -delta);
3695 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3696 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
3697 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3698 
3699 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3700 
3701 	/*
3702 	 * Since livelists are specific to a clone's origin txg, they
3703 	 * are no longer accurate. Destroy the livelist from the clone being
3704 	 * promoted. If the origin dataset is a clone, destroy its livelist
3705 	 * as well.
3706 	 */
3707 	dsl_dir_remove_livelist(dd, tx, B_TRUE);
3708 	dsl_dir_remove_livelist(odd, tx, B_TRUE);
3709 
3710 	/* log history record */
3711 	spa_history_log_internal_ds(hds, "promote", tx, " ");
3712 
3713 	dsl_dir_rele(odd, FTAG);
3714 	promote_rele(ddpa, FTAG);
3715 
3716 	/*
3717 	 * Transfer common error blocks from old head to new head.
3718 	 */
3719 	if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_HEAD_ERRLOG)) {
3720 		uint64_t old_head = origin_head->ds_object;
3721 		uint64_t new_head = hds->ds_object;
3722 		spa_swap_errlog(dp->dp_spa, new_head, old_head, tx);
3723 	}
3724 }
3725 
3726 /*
3727  * Make a list of dsl_dataset_t's for the snapshots between first_obj
3728  * (exclusive) and last_obj (inclusive).  The list will be in reverse
3729  * order (last_obj will be the list_head()).  If first_obj == 0, do all
3730  * snapshots back to this dataset's origin.
3731  */
3732 static int
3733 snaplist_make(dsl_pool_t *dp,
3734     uint64_t first_obj, uint64_t last_obj, list_t *l, const void *tag)
3735 {
3736 	uint64_t obj = last_obj;
3737 
3738 	list_create(l, sizeof (struct promotenode),
3739 	    offsetof(struct promotenode, link));
3740 
3741 	while (obj != first_obj) {
3742 		dsl_dataset_t *ds;
3743 		struct promotenode *snap;
3744 		int err;
3745 
3746 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3747 		ASSERT(err != ENOENT);
3748 		if (err != 0)
3749 			return (err);
3750 
3751 		if (first_obj == 0)
3752 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3753 
3754 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3755 		snap->ds = ds;
3756 		list_insert_tail(l, snap);
3757 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3758 	}
3759 
3760 	return (0);
3761 }
3762 
3763 static int
3764 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3765 {
3766 	struct promotenode *snap;
3767 
3768 	*spacep = 0;
3769 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3770 		uint64_t used, comp, uncomp;
3771 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3772 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
3773 		*spacep += used;
3774 	}
3775 	return (0);
3776 }
3777 
3778 static void
3779 snaplist_destroy(list_t *l, const void *tag)
3780 {
3781 	struct promotenode *snap;
3782 
3783 	if (l == NULL || !list_link_active(&l->list_head))
3784 		return;
3785 
3786 	while ((snap = list_remove_tail(l)) != NULL) {
3787 		dsl_dataset_rele(snap->ds, tag);
3788 		kmem_free(snap, sizeof (*snap));
3789 	}
3790 	list_destroy(l);
3791 }
3792 
3793 static int
3794 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, const void *tag)
3795 {
3796 	int error;
3797 	dsl_dir_t *dd;
3798 	struct promotenode *snap;
3799 
3800 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3801 	    &ddpa->ddpa_clone);
3802 	if (error != 0)
3803 		return (error);
3804 	dd = ddpa->ddpa_clone->ds_dir;
3805 
3806 	if (ddpa->ddpa_clone->ds_is_snapshot ||
3807 	    !dsl_dir_is_clone(dd)) {
3808 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
3809 		return (SET_ERROR(EINVAL));
3810 	}
3811 
3812 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3813 	    &ddpa->shared_snaps, tag);
3814 	if (error != 0)
3815 		goto out;
3816 
3817 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3818 	    &ddpa->clone_snaps, tag);
3819 	if (error != 0)
3820 		goto out;
3821 
3822 	snap = list_head(&ddpa->shared_snaps);
3823 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3824 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3825 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3826 	    &ddpa->origin_snaps, tag);
3827 	if (error != 0)
3828 		goto out;
3829 
3830 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3831 		error = dsl_dataset_hold_obj(dp,
3832 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3833 		    tag, &ddpa->origin_origin);
3834 		if (error != 0)
3835 			goto out;
3836 	}
3837 out:
3838 	if (error != 0)
3839 		promote_rele(ddpa, tag);
3840 	return (error);
3841 }
3842 
3843 static void
3844 promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag)
3845 {
3846 	snaplist_destroy(&ddpa->shared_snaps, tag);
3847 	snaplist_destroy(&ddpa->clone_snaps, tag);
3848 	snaplist_destroy(&ddpa->origin_snaps, tag);
3849 	if (ddpa->origin_origin != NULL)
3850 		dsl_dataset_rele(ddpa->origin_origin, tag);
3851 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
3852 }
3853 
3854 /*
3855  * Promote a clone.
3856  *
3857  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3858  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3859  */
3860 int
3861 dsl_dataset_promote(const char *name, char *conflsnap)
3862 {
3863 	dsl_dataset_promote_arg_t ddpa = { 0 };
3864 	uint64_t numsnaps;
3865 	int error;
3866 	nvpair_t *snap_pair;
3867 	objset_t *os;
3868 
3869 	/*
3870 	 * We will modify space proportional to the number of
3871 	 * snapshots.  Compute numsnaps.
3872 	 */
3873 	error = dmu_objset_hold(name, FTAG, &os);
3874 	if (error != 0)
3875 		return (error);
3876 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3877 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3878 	    &numsnaps);
3879 	dmu_objset_rele(os, FTAG);
3880 	if (error != 0)
3881 		return (error);
3882 
3883 	ddpa.ddpa_clonename = name;
3884 	ddpa.err_ds = fnvlist_alloc();
3885 	ddpa.cr = CRED();
3886 	ddpa.proc = curproc;
3887 
3888 	error = dsl_sync_task(name, dsl_dataset_promote_check,
3889 	    dsl_dataset_promote_sync, &ddpa,
3890 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3891 
3892 	/*
3893 	 * Return the first conflicting snapshot found.
3894 	 */
3895 	snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3896 	if (snap_pair != NULL && conflsnap != NULL)
3897 		(void) strlcpy(conflsnap, nvpair_name(snap_pair),
3898 		    ZFS_MAX_DATASET_NAME_LEN);
3899 
3900 	fnvlist_free(ddpa.err_ds);
3901 	return (error);
3902 }
3903 
3904 int
3905 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3906     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3907 {
3908 	/*
3909 	 * "slack" factor for received datasets with refquota set on them.
3910 	 * See the bottom of this function for details on its use.
3911 	 */
3912 	uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS *
3913 	    spa_asize_inflation;
3914 	int64_t unused_refres_delta;
3915 
3916 	/* they should both be heads */
3917 	if (clone->ds_is_snapshot ||
3918 	    origin_head->ds_is_snapshot)
3919 		return (SET_ERROR(EINVAL));
3920 
3921 	/* if we are not forcing, the branch point should be just before them */
3922 	if (!force && clone->ds_prev != origin_head->ds_prev)
3923 		return (SET_ERROR(EINVAL));
3924 
3925 	/* clone should be the clone (unless they are unrelated) */
3926 	if (clone->ds_prev != NULL &&
3927 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3928 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
3929 		return (SET_ERROR(EINVAL));
3930 
3931 	/* the clone should be a child of the origin */
3932 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3933 		return (SET_ERROR(EINVAL));
3934 
3935 	/* origin_head shouldn't be modified unless 'force' */
3936 	if (!force &&
3937 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3938 		return (SET_ERROR(ETXTBSY));
3939 
3940 	/* origin_head should have no long holds (e.g. is not mounted) */
3941 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
3942 		return (SET_ERROR(EBUSY));
3943 
3944 	/* check amount of any unconsumed refreservation */
3945 	unused_refres_delta =
3946 	    (int64_t)MIN(origin_head->ds_reserved,
3947 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3948 	    (int64_t)MIN(origin_head->ds_reserved,
3949 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3950 
3951 	if (unused_refres_delta > 0 &&
3952 	    unused_refres_delta >
3953 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3954 		return (SET_ERROR(ENOSPC));
3955 
3956 	/*
3957 	 * The clone can't be too much over the head's refquota.
3958 	 *
3959 	 * To ensure that the entire refquota can be used, we allow one
3960 	 * transaction to exceed the refquota.  Therefore, this check
3961 	 * needs to also allow for the space referenced to be more than the
3962 	 * refquota.  The maximum amount of space that one transaction can use
3963 	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
3964 	 * overage ensures that we are able to receive a filesystem that
3965 	 * exceeds the refquota on the source system.
3966 	 *
3967 	 * So that overage is the refquota_slack we use below.
3968 	 */
3969 	if (origin_head->ds_quota != 0 &&
3970 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
3971 	    origin_head->ds_quota + refquota_slack)
3972 		return (SET_ERROR(EDQUOT));
3973 
3974 	return (0);
3975 }
3976 
3977 static void
3978 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3979     dsl_dataset_t *origin, dmu_tx_t *tx)
3980 {
3981 	uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3982 	dsl_pool_t *dp = dmu_tx_pool(tx);
3983 
3984 	ASSERT(dsl_pool_sync_context(dp));
3985 
3986 	clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3987 	origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3988 
3989 	if (clone_remap_dl_obj != 0) {
3990 		dsl_deadlist_close(&clone->ds_remap_deadlist);
3991 		dsl_dataset_unset_remap_deadlist_object(clone, tx);
3992 	}
3993 	if (origin_remap_dl_obj != 0) {
3994 		dsl_deadlist_close(&origin->ds_remap_deadlist);
3995 		dsl_dataset_unset_remap_deadlist_object(origin, tx);
3996 	}
3997 
3998 	if (clone_remap_dl_obj != 0) {
3999 		dsl_dataset_set_remap_deadlist_object(origin,
4000 		    clone_remap_dl_obj, tx);
4001 		dsl_deadlist_open(&origin->ds_remap_deadlist,
4002 		    dp->dp_meta_objset, clone_remap_dl_obj);
4003 	}
4004 	if (origin_remap_dl_obj != 0) {
4005 		dsl_dataset_set_remap_deadlist_object(clone,
4006 		    origin_remap_dl_obj, tx);
4007 		dsl_deadlist_open(&clone->ds_remap_deadlist,
4008 		    dp->dp_meta_objset, origin_remap_dl_obj);
4009 	}
4010 }
4011 
4012 void
4013 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
4014     dsl_dataset_t *origin_head, dmu_tx_t *tx)
4015 {
4016 	dsl_pool_t *dp = dmu_tx_pool(tx);
4017 	int64_t unused_refres_delta;
4018 
4019 	ASSERT(clone->ds_reserved == 0);
4020 	/*
4021 	 * NOTE: On DEBUG kernels there could be a race between this and
4022 	 * the check function if spa_asize_inflation is adjusted...
4023 	 */
4024 	ASSERT(origin_head->ds_quota == 0 ||
4025 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
4026 	    DMU_MAX_ACCESS * spa_asize_inflation);
4027 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
4028 
4029 	dsl_dir_cancel_waiters(origin_head->ds_dir);
4030 
4031 	/*
4032 	 * Swap per-dataset feature flags.
4033 	 */
4034 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
4035 		if (!(spa_feature_table[f].fi_flags &
4036 		    ZFEATURE_FLAG_PER_DATASET)) {
4037 			ASSERT(!dsl_dataset_feature_is_active(clone, f));
4038 			ASSERT(!dsl_dataset_feature_is_active(origin_head, f));
4039 			continue;
4040 		}
4041 
4042 		boolean_t clone_inuse = dsl_dataset_feature_is_active(clone, f);
4043 		void *clone_feature = clone->ds_feature[f];
4044 		boolean_t origin_head_inuse =
4045 		    dsl_dataset_feature_is_active(origin_head, f);
4046 		void *origin_head_feature = origin_head->ds_feature[f];
4047 
4048 		if (clone_inuse)
4049 			dsl_dataset_deactivate_feature_impl(clone, f, tx);
4050 		if (origin_head_inuse)
4051 			dsl_dataset_deactivate_feature_impl(origin_head, f, tx);
4052 
4053 		if (clone_inuse) {
4054 			dsl_dataset_activate_feature(origin_head->ds_object, f,
4055 			    clone_feature, tx);
4056 			origin_head->ds_feature[f] = clone_feature;
4057 		}
4058 		if (origin_head_inuse) {
4059 			dsl_dataset_activate_feature(clone->ds_object, f,
4060 			    origin_head_feature, tx);
4061 			clone->ds_feature[f] = origin_head_feature;
4062 		}
4063 	}
4064 
4065 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
4066 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
4067 
4068 	if (clone->ds_objset != NULL) {
4069 		dmu_objset_evict(clone->ds_objset);
4070 		clone->ds_objset = NULL;
4071 	}
4072 
4073 	if (origin_head->ds_objset != NULL) {
4074 		dmu_objset_evict(origin_head->ds_objset);
4075 		origin_head->ds_objset = NULL;
4076 	}
4077 
4078 	unused_refres_delta =
4079 	    (int64_t)MIN(origin_head->ds_reserved,
4080 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
4081 	    (int64_t)MIN(origin_head->ds_reserved,
4082 	    dsl_dataset_phys(clone)->ds_unique_bytes);
4083 
4084 	/*
4085 	 * Reset origin's unique bytes.
4086 	 */
4087 	{
4088 		dsl_dataset_t *origin = clone->ds_prev;
4089 		uint64_t comp, uncomp;
4090 
4091 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
4092 		dsl_deadlist_space_range(&clone->ds_deadlist,
4093 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
4094 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
4095 	}
4096 
4097 	/* swap blkptrs */
4098 	{
4099 		rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
4100 		rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
4101 		blkptr_t tmp;
4102 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
4103 		dsl_dataset_phys(origin_head)->ds_bp =
4104 		    dsl_dataset_phys(clone)->ds_bp;
4105 		dsl_dataset_phys(clone)->ds_bp = tmp;
4106 		rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
4107 		rrw_exit(&clone->ds_bp_rwlock, FTAG);
4108 	}
4109 
4110 	/* set dd_*_bytes */
4111 	{
4112 		int64_t dused, dcomp, duncomp;
4113 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
4114 		uint64_t odl_used, odl_comp, odl_uncomp;
4115 
4116 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
4117 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
4118 
4119 		dsl_deadlist_space(&clone->ds_deadlist,
4120 		    &cdl_used, &cdl_comp, &cdl_uncomp);
4121 		dsl_deadlist_space(&origin_head->ds_deadlist,
4122 		    &odl_used, &odl_comp, &odl_uncomp);
4123 
4124 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
4125 		    cdl_used -
4126 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
4127 		    odl_used);
4128 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
4129 		    cdl_comp -
4130 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
4131 		    odl_comp);
4132 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
4133 		    cdl_uncomp -
4134 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
4135 		    odl_uncomp);
4136 
4137 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
4138 		    dused, dcomp, duncomp, tx);
4139 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
4140 		    -dused, -dcomp, -duncomp, tx);
4141 
4142 		/*
4143 		 * The difference in the space used by snapshots is the
4144 		 * difference in snapshot space due to the head's
4145 		 * deadlist (since that's the only thing that's
4146 		 * changing that affects the snapused).
4147 		 */
4148 		dsl_deadlist_space_range(&clone->ds_deadlist,
4149 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
4150 		    &cdl_used, &cdl_comp, &cdl_uncomp);
4151 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
4152 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
4153 		    &odl_used, &odl_comp, &odl_uncomp);
4154 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
4155 		    DD_USED_HEAD, DD_USED_SNAP, tx);
4156 	}
4157 
4158 	/* swap ds_*_bytes */
4159 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
4160 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
4161 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
4162 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
4163 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
4164 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
4165 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
4166 	    dsl_dataset_phys(clone)->ds_unique_bytes);
4167 
4168 	/* apply any parent delta for change in unconsumed refreservation */
4169 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
4170 	    unused_refres_delta, 0, 0, tx);
4171 
4172 	/*
4173 	 * Swap deadlists.
4174 	 */
4175 	dsl_deadlist_close(&clone->ds_deadlist);
4176 	dsl_deadlist_close(&origin_head->ds_deadlist);
4177 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
4178 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
4179 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
4180 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
4181 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
4182 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
4183 	dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
4184 
4185 	/*
4186 	 * If there is a bookmark at the origin, its "next dataset" is
4187 	 * changing, so we need to reset its FBN.
4188 	 */
4189 	dsl_bookmark_next_changed(origin_head, origin_head->ds_prev, tx);
4190 
4191 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
4192 
4193 	/*
4194 	 * Destroy any livelists associated with the clone or the origin,
4195 	 * since after the swap the corresponding livelists are no longer
4196 	 * valid.
4197 	 */
4198 	dsl_dir_remove_livelist(clone->ds_dir, tx, B_TRUE);
4199 	dsl_dir_remove_livelist(origin_head->ds_dir, tx, B_TRUE);
4200 
4201 	spa_history_log_internal_ds(clone, "clone swap", tx,
4202 	    "parent=%s", origin_head->ds_dir->dd_myname);
4203 }
4204 
4205 /*
4206  * Given a pool name and a dataset object number in that pool,
4207  * return the name of that dataset.
4208  */
4209 int
4210 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
4211 {
4212 	dsl_pool_t *dp;
4213 	dsl_dataset_t *ds;
4214 	int error;
4215 
4216 	error = dsl_pool_hold(pname, FTAG, &dp);
4217 	if (error != 0)
4218 		return (error);
4219 
4220 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
4221 	if (error == 0) {
4222 		dsl_dataset_name(ds, buf);
4223 		dsl_dataset_rele(ds, FTAG);
4224 	}
4225 	dsl_pool_rele(dp, FTAG);
4226 
4227 	return (error);
4228 }
4229 
4230 int
4231 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
4232     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
4233 {
4234 	int error = 0;
4235 
4236 	ASSERT3S(asize, >, 0);
4237 
4238 	/*
4239 	 * *ref_rsrv is the portion of asize that will come from any
4240 	 * unconsumed refreservation space.
4241 	 */
4242 	*ref_rsrv = 0;
4243 
4244 	mutex_enter(&ds->ds_lock);
4245 	/*
4246 	 * Make a space adjustment for reserved bytes.
4247 	 */
4248 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
4249 		ASSERT3U(*used, >=,
4250 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
4251 		*used -=
4252 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
4253 		*ref_rsrv =
4254 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
4255 	}
4256 
4257 	if (!check_quota || ds->ds_quota == 0) {
4258 		mutex_exit(&ds->ds_lock);
4259 		return (0);
4260 	}
4261 	/*
4262 	 * If they are requesting more space, and our current estimate
4263 	 * is over quota, they get to try again unless the actual
4264 	 * on-disk is over quota and there are no pending changes (which
4265 	 * may free up space for us).
4266 	 */
4267 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
4268 	    ds->ds_quota) {
4269 		if (inflight > 0 ||
4270 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
4271 			error = SET_ERROR(ERESTART);
4272 		else
4273 			error = SET_ERROR(EDQUOT);
4274 	}
4275 	mutex_exit(&ds->ds_lock);
4276 
4277 	return (error);
4278 }
4279 
4280 typedef struct dsl_dataset_set_qr_arg {
4281 	const char *ddsqra_name;
4282 	zprop_source_t ddsqra_source;
4283 	uint64_t ddsqra_value;
4284 } dsl_dataset_set_qr_arg_t;
4285 
4286 
4287 static int
4288 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
4289 {
4290 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4291 	dsl_pool_t *dp = dmu_tx_pool(tx);
4292 	dsl_dataset_t *ds;
4293 	int error;
4294 	uint64_t newval;
4295 
4296 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
4297 		return (SET_ERROR(ENOTSUP));
4298 
4299 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
4300 	if (error != 0)
4301 		return (error);
4302 
4303 	if (ds->ds_is_snapshot) {
4304 		dsl_dataset_rele(ds, FTAG);
4305 		return (SET_ERROR(EINVAL));
4306 	}
4307 
4308 	error = dsl_prop_predict(ds->ds_dir,
4309 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
4310 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
4311 	if (error != 0) {
4312 		dsl_dataset_rele(ds, FTAG);
4313 		return (error);
4314 	}
4315 
4316 	if (newval == 0) {
4317 		dsl_dataset_rele(ds, FTAG);
4318 		return (0);
4319 	}
4320 
4321 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
4322 	    newval < ds->ds_reserved) {
4323 		dsl_dataset_rele(ds, FTAG);
4324 		return (SET_ERROR(ENOSPC));
4325 	}
4326 
4327 	dsl_dataset_rele(ds, FTAG);
4328 	return (0);
4329 }
4330 
4331 static void
4332 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
4333 {
4334 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4335 	dsl_pool_t *dp = dmu_tx_pool(tx);
4336 	dsl_dataset_t *ds = NULL;
4337 	uint64_t newval;
4338 
4339 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
4340 
4341 	dsl_prop_set_sync_impl(ds,
4342 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
4343 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
4344 	    &ddsqra->ddsqra_value, tx);
4345 
4346 	VERIFY0(dsl_prop_get_int_ds(ds,
4347 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
4348 
4349 	if (ds->ds_quota != newval) {
4350 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
4351 		ds->ds_quota = newval;
4352 	}
4353 	dsl_dataset_rele(ds, FTAG);
4354 }
4355 
4356 int
4357 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
4358     uint64_t refquota)
4359 {
4360 	dsl_dataset_set_qr_arg_t ddsqra;
4361 
4362 	ddsqra.ddsqra_name = dsname;
4363 	ddsqra.ddsqra_source = source;
4364 	ddsqra.ddsqra_value = refquota;
4365 
4366 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
4367 	    dsl_dataset_set_refquota_sync, &ddsqra, 0,
4368 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
4369 }
4370 
4371 static int
4372 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
4373 {
4374 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4375 	dsl_pool_t *dp = dmu_tx_pool(tx);
4376 	dsl_dataset_t *ds;
4377 	int error;
4378 	uint64_t newval, unique;
4379 
4380 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
4381 		return (SET_ERROR(ENOTSUP));
4382 
4383 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
4384 	if (error != 0)
4385 		return (error);
4386 
4387 	if (ds->ds_is_snapshot) {
4388 		dsl_dataset_rele(ds, FTAG);
4389 		return (SET_ERROR(EINVAL));
4390 	}
4391 
4392 	error = dsl_prop_predict(ds->ds_dir,
4393 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
4394 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
4395 	if (error != 0) {
4396 		dsl_dataset_rele(ds, FTAG);
4397 		return (error);
4398 	}
4399 
4400 	/*
4401 	 * If we are doing the preliminary check in open context, the
4402 	 * space estimates may be inaccurate.
4403 	 */
4404 	if (!dmu_tx_is_syncing(tx)) {
4405 		dsl_dataset_rele(ds, FTAG);
4406 		return (0);
4407 	}
4408 
4409 	mutex_enter(&ds->ds_lock);
4410 	if (!DS_UNIQUE_IS_ACCURATE(ds))
4411 		dsl_dataset_recalc_head_uniq(ds);
4412 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
4413 	mutex_exit(&ds->ds_lock);
4414 
4415 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
4416 		uint64_t delta = MAX(unique, newval) -
4417 		    MAX(unique, ds->ds_reserved);
4418 
4419 		if (delta >
4420 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
4421 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
4422 			dsl_dataset_rele(ds, FTAG);
4423 			return (SET_ERROR(ENOSPC));
4424 		}
4425 	}
4426 
4427 	dsl_dataset_rele(ds, FTAG);
4428 	return (0);
4429 }
4430 
4431 void
4432 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
4433     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
4434 {
4435 	uint64_t newval;
4436 	uint64_t unique;
4437 	int64_t delta;
4438 
4439 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
4440 	    source, sizeof (value), 1, &value, tx);
4441 
4442 	VERIFY0(dsl_prop_get_int_ds(ds,
4443 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
4444 
4445 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
4446 	mutex_enter(&ds->ds_dir->dd_lock);
4447 	mutex_enter(&ds->ds_lock);
4448 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
4449 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
4450 	delta = MAX(0, (int64_t)(newval - unique)) -
4451 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
4452 	ds->ds_reserved = newval;
4453 	mutex_exit(&ds->ds_lock);
4454 
4455 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
4456 	mutex_exit(&ds->ds_dir->dd_lock);
4457 }
4458 
4459 static void
4460 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
4461 {
4462 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4463 	dsl_pool_t *dp = dmu_tx_pool(tx);
4464 	dsl_dataset_t *ds = NULL;
4465 
4466 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
4467 	dsl_dataset_set_refreservation_sync_impl(ds,
4468 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
4469 	dsl_dataset_rele(ds, FTAG);
4470 }
4471 
4472 int
4473 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
4474     uint64_t refreservation)
4475 {
4476 	dsl_dataset_set_qr_arg_t ddsqra;
4477 
4478 	ddsqra.ddsqra_name = dsname;
4479 	ddsqra.ddsqra_source = source;
4480 	ddsqra.ddsqra_value = refreservation;
4481 
4482 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
4483 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0,
4484 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
4485 }
4486 
4487 typedef struct dsl_dataset_set_compression_arg {
4488 	const char *ddsca_name;
4489 	zprop_source_t ddsca_source;
4490 	uint64_t ddsca_value;
4491 } dsl_dataset_set_compression_arg_t;
4492 
4493 static int
4494 dsl_dataset_set_compression_check(void *arg, dmu_tx_t *tx)
4495 {
4496 	dsl_dataset_set_compression_arg_t *ddsca = arg;
4497 	dsl_pool_t *dp = dmu_tx_pool(tx);
4498 
4499 	uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value);
4500 	spa_feature_t f = zio_compress_to_feature(compval);
4501 
4502 	if (f == SPA_FEATURE_NONE)
4503 		return (SET_ERROR(EINVAL));
4504 
4505 	if (!spa_feature_is_enabled(dp->dp_spa, f))
4506 		return (SET_ERROR(ENOTSUP));
4507 
4508 	return (0);
4509 }
4510 
4511 static void
4512 dsl_dataset_set_compression_sync(void *arg, dmu_tx_t *tx)
4513 {
4514 	dsl_dataset_set_compression_arg_t *ddsca = arg;
4515 	dsl_pool_t *dp = dmu_tx_pool(tx);
4516 	dsl_dataset_t *ds = NULL;
4517 
4518 	uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value);
4519 	spa_feature_t f = zio_compress_to_feature(compval);
4520 	ASSERT3S(f, !=, SPA_FEATURE_NONE);
4521 	ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN);
4522 
4523 	VERIFY0(dsl_dataset_hold(dp, ddsca->ddsca_name, FTAG, &ds));
4524 	if (zfeature_active(f, ds->ds_feature[f]) != B_TRUE) {
4525 		ds->ds_feature_activation[f] = (void *)B_TRUE;
4526 		dsl_dataset_activate_feature(ds->ds_object, f,
4527 		    ds->ds_feature_activation[f], tx);
4528 		ds->ds_feature[f] = ds->ds_feature_activation[f];
4529 	}
4530 	dsl_dataset_rele(ds, FTAG);
4531 }
4532 
4533 int
4534 dsl_dataset_set_compression(const char *dsname, zprop_source_t source,
4535     uint64_t compression)
4536 {
4537 	dsl_dataset_set_compression_arg_t ddsca;
4538 
4539 	/*
4540 	 * The sync task is only required for zstd in order to activate
4541 	 * the feature flag when the property is first set.
4542 	 */
4543 	if (ZIO_COMPRESS_ALGO(compression) != ZIO_COMPRESS_ZSTD)
4544 		return (0);
4545 
4546 	ddsca.ddsca_name = dsname;
4547 	ddsca.ddsca_source = source;
4548 	ddsca.ddsca_value = compression;
4549 
4550 	return (dsl_sync_task(dsname, dsl_dataset_set_compression_check,
4551 	    dsl_dataset_set_compression_sync, &ddsca, 0,
4552 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
4553 }
4554 
4555 /*
4556  * Return (in *usedp) the amount of space referenced by "new" that was not
4557  * referenced at the time the bookmark corresponds to.  "New" may be a
4558  * snapshot or a head.  The bookmark must be before new, in
4559  * new's filesystem (or its origin) -- caller verifies this.
4560  *
4561  * The written space is calculated by considering two components:  First, we
4562  * ignore any freed space, and calculate the written as new's used space
4563  * minus old's used space.  Next, we add in the amount of space that was freed
4564  * between the two time points, thus reducing new's used space relative to
4565  * old's. Specifically, this is the space that was born before
4566  * zbm_creation_txg, and freed before new (ie. on new's deadlist or a
4567  * previous deadlist).
4568  *
4569  * space freed                         [---------------------]
4570  * snapshots                       ---O-------O--------O-------O------
4571  *                                         bookmark           new
4572  *
4573  * Note, the bookmark's zbm_*_bytes_refd must be valid, but if the HAS_FBN
4574  * flag is not set, we will calculate the freed_before_next based on the
4575  * next snapshot's deadlist, rather than using zbm_*_freed_before_next_snap.
4576  */
4577 static int
4578 dsl_dataset_space_written_impl(zfs_bookmark_phys_t *bmp,
4579     dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4580 {
4581 	int err = 0;
4582 	dsl_pool_t *dp = new->ds_dir->dd_pool;
4583 
4584 	ASSERT(dsl_pool_config_held(dp));
4585 	if (dsl_dataset_is_snapshot(new)) {
4586 		ASSERT3U(bmp->zbm_creation_txg, <,
4587 		    dsl_dataset_phys(new)->ds_creation_txg);
4588 	}
4589 
4590 	*usedp = 0;
4591 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
4592 	*usedp -= bmp->zbm_referenced_bytes_refd;
4593 
4594 	*compp = 0;
4595 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
4596 	*compp -= bmp->zbm_compressed_bytes_refd;
4597 
4598 	*uncompp = 0;
4599 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
4600 	*uncompp -= bmp->zbm_uncompressed_bytes_refd;
4601 
4602 	dsl_dataset_t *snap = new;
4603 
4604 	while (dsl_dataset_phys(snap)->ds_prev_snap_txg >
4605 	    bmp->zbm_creation_txg) {
4606 		uint64_t used, comp, uncomp;
4607 
4608 		dsl_deadlist_space_range(&snap->ds_deadlist,
4609 		    0, bmp->zbm_creation_txg,
4610 		    &used, &comp, &uncomp);
4611 		*usedp += used;
4612 		*compp += comp;
4613 		*uncompp += uncomp;
4614 
4615 		uint64_t snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
4616 		if (snap != new)
4617 			dsl_dataset_rele(snap, FTAG);
4618 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4619 		if (err != 0)
4620 			break;
4621 	}
4622 
4623 	/*
4624 	 * We might not have the FBN if we are calculating written from
4625 	 * a snapshot (because we didn't know the correct "next" snapshot
4626 	 * until now).
4627 	 */
4628 	if (bmp->zbm_flags & ZBM_FLAG_HAS_FBN) {
4629 		*usedp += bmp->zbm_referenced_freed_before_next_snap;
4630 		*compp += bmp->zbm_compressed_freed_before_next_snap;
4631 		*uncompp += bmp->zbm_uncompressed_freed_before_next_snap;
4632 	} else {
4633 		ASSERT3U(dsl_dataset_phys(snap)->ds_prev_snap_txg, ==,
4634 		    bmp->zbm_creation_txg);
4635 		uint64_t used, comp, uncomp;
4636 		dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp);
4637 		*usedp += used;
4638 		*compp += comp;
4639 		*uncompp += uncomp;
4640 	}
4641 	if (snap != new)
4642 		dsl_dataset_rele(snap, FTAG);
4643 	return (err);
4644 }
4645 
4646 /*
4647  * Return (in *usedp) the amount of space written in new that was not
4648  * present at the time the bookmark corresponds to.  New may be a
4649  * snapshot or the head.  Old must be a bookmark before new, in
4650  * new's filesystem (or its origin) -- caller verifies this.
4651  */
4652 int
4653 dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t *bmp,
4654     dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4655 {
4656 	if (!(bmp->zbm_flags & ZBM_FLAG_HAS_FBN))
4657 		return (SET_ERROR(ENOTSUP));
4658 	return (dsl_dataset_space_written_impl(bmp, new,
4659 	    usedp, compp, uncompp));
4660 }
4661 
4662 /*
4663  * Return (in *usedp) the amount of space written in new that is not
4664  * present in oldsnap.  New may be a snapshot or the head.  Old must be
4665  * a snapshot before new, in new's filesystem (or its origin).  If not then
4666  * fail and return EINVAL.
4667  */
4668 int
4669 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4670     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4671 {
4672 	if (!dsl_dataset_is_before(new, oldsnap, 0))
4673 		return (SET_ERROR(EINVAL));
4674 
4675 	zfs_bookmark_phys_t zbm = { 0 };
4676 	dsl_dataset_phys_t *dsp = dsl_dataset_phys(oldsnap);
4677 	zbm.zbm_guid = dsp->ds_guid;
4678 	zbm.zbm_creation_txg = dsp->ds_creation_txg;
4679 	zbm.zbm_creation_time = dsp->ds_creation_time;
4680 	zbm.zbm_referenced_bytes_refd = dsp->ds_referenced_bytes;
4681 	zbm.zbm_compressed_bytes_refd = dsp->ds_compressed_bytes;
4682 	zbm.zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes;
4683 
4684 	/*
4685 	 * If oldsnap is the origin (or origin's origin, ...) of new,
4686 	 * we can't easily calculate the effective FBN.  Therefore,
4687 	 * we do not set ZBM_FLAG_HAS_FBN, so that the _impl will calculate
4688 	 * it relative to the correct "next": the next snapshot towards "new",
4689 	 * rather than the next snapshot in oldsnap's dsl_dir.
4690 	 */
4691 	return (dsl_dataset_space_written_impl(&zbm, new,
4692 	    usedp, compp, uncompp));
4693 }
4694 
4695 /*
4696  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4697  * lastsnap, and all snapshots in between are deleted.
4698  *
4699  * blocks that would be freed            [---------------------------]
4700  * snapshots                       ---O-------O--------O-------O--------O
4701  *                                        firstsnap        lastsnap
4702  *
4703  * This is the set of blocks that were born after the snap before firstsnap,
4704  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4705  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4706  * We calculate this by iterating over the relevant deadlists (from the snap
4707  * after lastsnap, backward to the snap after firstsnap), summing up the
4708  * space on the deadlist that was born after the snap before firstsnap.
4709  */
4710 int
4711 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4712     dsl_dataset_t *lastsnap,
4713     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4714 {
4715 	int err = 0;
4716 	uint64_t snapobj;
4717 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4718 
4719 	ASSERT(firstsnap->ds_is_snapshot);
4720 	ASSERT(lastsnap->ds_is_snapshot);
4721 
4722 	/*
4723 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
4724 	 * is before lastsnap.
4725 	 */
4726 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
4727 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
4728 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
4729 		return (SET_ERROR(EINVAL));
4730 
4731 	*usedp = *compp = *uncompp = 0;
4732 
4733 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
4734 	while (snapobj != firstsnap->ds_object) {
4735 		dsl_dataset_t *ds;
4736 		uint64_t used, comp, uncomp;
4737 
4738 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4739 		if (err != 0)
4740 			break;
4741 
4742 		dsl_deadlist_space_range(&ds->ds_deadlist,
4743 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
4744 		    &used, &comp, &uncomp);
4745 		*usedp += used;
4746 		*compp += comp;
4747 		*uncompp += uncomp;
4748 
4749 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4750 		ASSERT3U(snapobj, !=, 0);
4751 		dsl_dataset_rele(ds, FTAG);
4752 	}
4753 	return (err);
4754 }
4755 
4756 /*
4757  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
4758  * For example, they could both be snapshots of the same filesystem, and
4759  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
4760  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
4761  * filesystem.  Or 'earlier' could be the origin's origin.
4762  *
4763  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4764  */
4765 boolean_t
4766 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4767     uint64_t earlier_txg)
4768 {
4769 	dsl_pool_t *dp = later->ds_dir->dd_pool;
4770 	int error;
4771 	boolean_t ret;
4772 
4773 	ASSERT(dsl_pool_config_held(dp));
4774 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4775 
4776 	if (earlier_txg == 0)
4777 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4778 
4779 	if (later->ds_is_snapshot &&
4780 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4781 		return (B_FALSE);
4782 
4783 	if (later->ds_dir == earlier->ds_dir)
4784 		return (B_TRUE);
4785 
4786 	/*
4787 	 * We check dd_origin_obj explicitly here rather than using
4788 	 * dsl_dir_is_clone() so that we will return TRUE if "earlier"
4789 	 * is $ORIGIN@$ORIGIN.  dsl_dataset_space_written() depends on
4790 	 * this behavior.
4791 	 */
4792 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == 0)
4793 		return (B_FALSE);
4794 
4795 	dsl_dataset_t *origin;
4796 	error = dsl_dataset_hold_obj(dp,
4797 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4798 	if (error != 0)
4799 		return (B_FALSE);
4800 	if (dsl_dataset_phys(origin)->ds_creation_txg == earlier_txg &&
4801 	    origin->ds_dir == earlier->ds_dir) {
4802 		dsl_dataset_rele(origin, FTAG);
4803 		return (B_TRUE);
4804 	}
4805 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4806 	dsl_dataset_rele(origin, FTAG);
4807 	return (ret);
4808 }
4809 
4810 void
4811 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4812 {
4813 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4814 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4815 }
4816 
4817 boolean_t
4818 dsl_dataset_is_zapified(dsl_dataset_t *ds)
4819 {
4820 	dmu_object_info_t doi;
4821 
4822 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
4823 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4824 }
4825 
4826 boolean_t
4827 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4828 {
4829 	return (dsl_dataset_is_zapified(ds) &&
4830 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4831 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4832 }
4833 
4834 uint64_t
4835 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4836 {
4837 	uint64_t remap_deadlist_obj;
4838 	int err;
4839 
4840 	if (!dsl_dataset_is_zapified(ds))
4841 		return (0);
4842 
4843 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4844 	    DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4845 	    &remap_deadlist_obj);
4846 
4847 	if (err != 0) {
4848 		VERIFY3S(err, ==, ENOENT);
4849 		return (0);
4850 	}
4851 
4852 	ASSERT(remap_deadlist_obj != 0);
4853 	return (remap_deadlist_obj);
4854 }
4855 
4856 boolean_t
4857 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4858 {
4859 	EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4860 	    dsl_dataset_get_remap_deadlist_object(ds) != 0);
4861 	return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4862 }
4863 
4864 static void
4865 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4866     dmu_tx_t *tx)
4867 {
4868 	ASSERT(obj != 0);
4869 	dsl_dataset_zapify(ds, tx);
4870 	VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4871 	    DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4872 }
4873 
4874 static void
4875 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4876 {
4877 	VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4878 	    ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4879 }
4880 
4881 void
4882 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4883 {
4884 	uint64_t remap_deadlist_object;
4885 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4886 
4887 	ASSERT(dmu_tx_is_syncing(tx));
4888 	ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4889 
4890 	remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4891 	dsl_deadlist_close(&ds->ds_remap_deadlist);
4892 	dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4893 	dsl_dataset_unset_remap_deadlist_object(ds, tx);
4894 	spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4895 }
4896 
4897 void
4898 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4899 {
4900 	uint64_t remap_deadlist_obj;
4901 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4902 
4903 	ASSERT(dmu_tx_is_syncing(tx));
4904 	ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4905 	/*
4906 	 * Currently we only create remap deadlists when there are indirect
4907 	 * vdevs with referenced mappings.
4908 	 */
4909 	ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4910 
4911 	remap_deadlist_obj = dsl_deadlist_clone(
4912 	    &ds->ds_deadlist, UINT64_MAX,
4913 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4914 	dsl_dataset_set_remap_deadlist_object(ds,
4915 	    remap_deadlist_obj, tx);
4916 	dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4917 	    remap_deadlist_obj);
4918 	spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4919 }
4920 
4921 void
4922 dsl_dataset_activate_redaction(dsl_dataset_t *ds, uint64_t *redact_snaps,
4923     uint64_t num_redact_snaps, dmu_tx_t *tx)
4924 {
4925 	uint64_t dsobj = ds->ds_object;
4926 	struct feature_type_uint64_array_arg *ftuaa =
4927 	    kmem_zalloc(sizeof (*ftuaa), KM_SLEEP);
4928 	ftuaa->length = (int64_t)num_redact_snaps;
4929 	if (num_redact_snaps > 0) {
4930 		ftuaa->array = kmem_alloc(num_redact_snaps * sizeof (uint64_t),
4931 		    KM_SLEEP);
4932 		memcpy(ftuaa->array, redact_snaps, num_redact_snaps *
4933 		    sizeof (uint64_t));
4934 	}
4935 	dsl_dataset_activate_feature(dsobj, SPA_FEATURE_REDACTED_DATASETS,
4936 	    ftuaa, tx);
4937 	ds->ds_feature[SPA_FEATURE_REDACTED_DATASETS] = ftuaa;
4938 }
4939 
4940 /*
4941  * Find and return (in *oldest_dsobj) the oldest snapshot of the dsobj
4942  * dataset whose birth time is >= min_txg.
4943  */
4944 int
4945 dsl_dataset_oldest_snapshot(spa_t *spa, uint64_t head_ds, uint64_t min_txg,
4946     uint64_t *oldest_dsobj)
4947 {
4948 	dsl_dataset_t *ds;
4949 	dsl_pool_t *dp = spa->spa_dsl_pool;
4950 
4951 	int error = dsl_dataset_hold_obj(dp, head_ds, FTAG, &ds);
4952 	if (error != 0)
4953 		return (error);
4954 
4955 	uint64_t prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4956 	uint64_t prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
4957 
4958 	while (prev_obj != 0 && min_txg < prev_obj_txg) {
4959 		dsl_dataset_rele(ds, FTAG);
4960 		if ((error = dsl_dataset_hold_obj(dp, prev_obj,
4961 		    FTAG, &ds)) != 0)
4962 			return (error);
4963 		prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
4964 		prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4965 	}
4966 	*oldest_dsobj = ds->ds_object;
4967 	dsl_dataset_rele(ds, FTAG);
4968 	return (0);
4969 }
4970 
4971 ZFS_MODULE_PARAM(zfs, zfs_, max_recordsize, UINT, ZMOD_RW,
4972 	"Max allowed record size");
4973 
4974 ZFS_MODULE_PARAM(zfs, zfs_, allow_redacted_dataset_mount, INT, ZMOD_RW,
4975 	"Allow mounting of redacted datasets");
4976 
4977 ZFS_MODULE_PARAM(zfs, zfs_, snapshot_history_enabled, INT, ZMOD_RW,
4978 	"Include snapshot events in pool history/events");
4979 
4980 EXPORT_SYMBOL(dsl_dataset_hold);
4981 EXPORT_SYMBOL(dsl_dataset_hold_flags);
4982 EXPORT_SYMBOL(dsl_dataset_hold_obj);
4983 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags);
4984 EXPORT_SYMBOL(dsl_dataset_own);
4985 EXPORT_SYMBOL(dsl_dataset_own_obj);
4986 EXPORT_SYMBOL(dsl_dataset_name);
4987 EXPORT_SYMBOL(dsl_dataset_rele);
4988 EXPORT_SYMBOL(dsl_dataset_rele_flags);
4989 EXPORT_SYMBOL(dsl_dataset_disown);
4990 EXPORT_SYMBOL(dsl_dataset_tryown);
4991 EXPORT_SYMBOL(dsl_dataset_create_sync);
4992 EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
4993 EXPORT_SYMBOL(dsl_dataset_snapshot_check);
4994 EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
4995 EXPORT_SYMBOL(dsl_dataset_promote);
4996 EXPORT_SYMBOL(dsl_dataset_user_hold);
4997 EXPORT_SYMBOL(dsl_dataset_user_release);
4998 EXPORT_SYMBOL(dsl_dataset_get_holds);
4999 EXPORT_SYMBOL(dsl_dataset_get_blkptr);
5000 EXPORT_SYMBOL(dsl_dataset_get_spa);
5001 EXPORT_SYMBOL(dsl_dataset_modified_since_snap);
5002 EXPORT_SYMBOL(dsl_dataset_space_written);
5003 EXPORT_SYMBOL(dsl_dataset_space_wouldfree);
5004 EXPORT_SYMBOL(dsl_dataset_sync);
5005 EXPORT_SYMBOL(dsl_dataset_block_born);
5006 EXPORT_SYMBOL(dsl_dataset_block_kill);
5007 EXPORT_SYMBOL(dsl_dataset_dirty);
5008 EXPORT_SYMBOL(dsl_dataset_stats);
5009 EXPORT_SYMBOL(dsl_dataset_fast_stat);
5010 EXPORT_SYMBOL(dsl_dataset_space);
5011 EXPORT_SYMBOL(dsl_dataset_fsid_guid);
5012 EXPORT_SYMBOL(dsl_dsobj_to_dsname);
5013 EXPORT_SYMBOL(dsl_dataset_check_quota);
5014 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl);
5015 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl);
5016