xref: /freebsd/sys/contrib/openzfs/module/zfs/dsl_dataset.c (revision 1323ec571215a77ddd21294f0871979d5ad6b992)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 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 do not allow the recordsize to be set larger than zfs_max_recordsize
77  * (default 1MB).  Larger blocks can be created by changing this tunable,
78  * and pools with larger blocks can always be imported and used, regardless
79  * of this setting.
80  */
81 int zfs_max_recordsize = 1 * 1024 * 1024;
82 static int zfs_allow_redacted_dataset_mount = 0;
83 
84 #define	SWITCH64(x, y) \
85 	{ \
86 		uint64_t __tmp = (x); \
87 		(x) = (y); \
88 		(y) = __tmp; \
89 	}
90 
91 #define	DS_REF_MAX	(1ULL << 62)
92 
93 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
94     uint64_t obj, dmu_tx_t *tx);
95 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
96     dmu_tx_t *tx);
97 
98 static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f);
99 
100 extern int spa_asize_inflation;
101 
102 static zil_header_t zero_zil;
103 
104 /*
105  * Figure out how much of this delta should be propagated to the dsl_dir
106  * layer.  If there's a refreservation, that space has already been
107  * partially accounted for in our ancestors.
108  */
109 static int64_t
110 parent_delta(dsl_dataset_t *ds, int64_t delta)
111 {
112 	dsl_dataset_phys_t *ds_phys;
113 	uint64_t old_bytes, new_bytes;
114 
115 	if (ds->ds_reserved == 0)
116 		return (delta);
117 
118 	ds_phys = dsl_dataset_phys(ds);
119 	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
120 	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
121 
122 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
123 	return (new_bytes - old_bytes);
124 }
125 
126 void
127 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
128 {
129 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
130 	int used = bp_get_dsize_sync(spa, bp);
131 	int compressed = BP_GET_PSIZE(bp);
132 	int uncompressed = BP_GET_UCSIZE(bp);
133 	int64_t delta;
134 	spa_feature_t f;
135 
136 	dprintf_bp(bp, "ds=%p", ds);
137 
138 	ASSERT(dmu_tx_is_syncing(tx));
139 	/* It could have been compressed away to nothing */
140 	if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
141 		return;
142 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
143 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
144 	if (ds == NULL) {
145 		dsl_pool_mos_diduse_space(tx->tx_pool,
146 		    used, compressed, uncompressed);
147 		return;
148 	}
149 
150 	ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
151 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
152 	mutex_enter(&ds->ds_lock);
153 	delta = parent_delta(ds, used);
154 	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
155 	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
156 	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
157 	dsl_dataset_phys(ds)->ds_unique_bytes += used;
158 
159 	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
160 		ds->ds_feature_activation[SPA_FEATURE_LARGE_BLOCKS] =
161 		    (void *)B_TRUE;
162 	}
163 
164 
165 	f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
166 	if (f != SPA_FEATURE_NONE) {
167 		ASSERT3S(spa_feature_table[f].fi_type, ==,
168 		    ZFEATURE_TYPE_BOOLEAN);
169 		ds->ds_feature_activation[f] = (void *)B_TRUE;
170 	}
171 
172 	f = zio_compress_to_feature(BP_GET_COMPRESS(bp));
173 	if (f != SPA_FEATURE_NONE) {
174 		ASSERT3S(spa_feature_table[f].fi_type, ==,
175 		    ZFEATURE_TYPE_BOOLEAN);
176 		ds->ds_feature_activation[f] = (void *)B_TRUE;
177 	}
178 
179 	/*
180 	 * Track block for livelist, but ignore embedded blocks because
181 	 * they do not need to be freed.
182 	 */
183 	if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
184 	    bp->blk_birth > ds->ds_dir->dd_origin_txg &&
185 	    !(BP_IS_EMBEDDED(bp))) {
186 		ASSERT(dsl_dir_is_clone(ds->ds_dir));
187 		ASSERT(spa_feature_is_enabled(spa,
188 		    SPA_FEATURE_LIVELIST));
189 		bplist_append(&ds->ds_dir->dd_pending_allocs, bp);
190 	}
191 
192 	mutex_exit(&ds->ds_lock);
193 	dsl_dir_diduse_transfer_space(ds->ds_dir, delta,
194 	    compressed, uncompressed, used,
195 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
196 }
197 
198 /*
199  * Called when the specified segment has been remapped, and is thus no
200  * longer referenced in the head dataset.  The vdev must be indirect.
201  *
202  * If the segment is referenced by a snapshot, put it on the remap deadlist.
203  * Otherwise, add this segment to the obsolete spacemap.
204  */
205 void
206 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset,
207     uint64_t size, uint64_t birth, dmu_tx_t *tx)
208 {
209 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
210 
211 	ASSERT(dmu_tx_is_syncing(tx));
212 	ASSERT(birth <= tx->tx_txg);
213 	ASSERT(!ds->ds_is_snapshot);
214 
215 	if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
216 		spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
217 	} else {
218 		blkptr_t fakebp;
219 		dva_t *dva = &fakebp.blk_dva[0];
220 
221 		ASSERT(ds != NULL);
222 
223 		mutex_enter(&ds->ds_remap_deadlist_lock);
224 		if (!dsl_dataset_remap_deadlist_exists(ds)) {
225 			dsl_dataset_create_remap_deadlist(ds, tx);
226 		}
227 		mutex_exit(&ds->ds_remap_deadlist_lock);
228 
229 		BP_ZERO(&fakebp);
230 		fakebp.blk_birth = birth;
231 		DVA_SET_VDEV(dva, vdev);
232 		DVA_SET_OFFSET(dva, offset);
233 		DVA_SET_ASIZE(dva, size);
234 		dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, B_FALSE,
235 		    tx);
236 	}
237 }
238 
239 int
240 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
241     boolean_t async)
242 {
243 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
244 
245 	int used = bp_get_dsize_sync(spa, bp);
246 	int compressed = BP_GET_PSIZE(bp);
247 	int uncompressed = BP_GET_UCSIZE(bp);
248 
249 	if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
250 		return (0);
251 
252 	ASSERT(dmu_tx_is_syncing(tx));
253 	ASSERT(bp->blk_birth <= tx->tx_txg);
254 
255 	if (ds == NULL) {
256 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
257 		dsl_pool_mos_diduse_space(tx->tx_pool,
258 		    -used, -compressed, -uncompressed);
259 		return (used);
260 	}
261 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
262 
263 	ASSERT(!ds->ds_is_snapshot);
264 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
265 
266 	/*
267 	 * Track block for livelist, but ignore embedded blocks because
268 	 * they do not need to be freed.
269 	 */
270 	if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
271 	    bp->blk_birth > ds->ds_dir->dd_origin_txg &&
272 	    !(BP_IS_EMBEDDED(bp))) {
273 		ASSERT(dsl_dir_is_clone(ds->ds_dir));
274 		ASSERT(spa_feature_is_enabled(spa,
275 		    SPA_FEATURE_LIVELIST));
276 		bplist_append(&ds->ds_dir->dd_pending_frees, bp);
277 	}
278 
279 	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
280 		int64_t delta;
281 
282 		dprintf_bp(bp, "freeing ds=%llu", (u_longlong_t)ds->ds_object);
283 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
284 
285 		mutex_enter(&ds->ds_lock);
286 		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
287 		    !DS_UNIQUE_IS_ACCURATE(ds));
288 		delta = parent_delta(ds, -used);
289 		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
290 		mutex_exit(&ds->ds_lock);
291 		dsl_dir_diduse_transfer_space(ds->ds_dir,
292 		    delta, -compressed, -uncompressed, -used,
293 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
294 	} else {
295 		dprintf_bp(bp, "putting on dead list: %s", "");
296 		if (async) {
297 			/*
298 			 * We are here as part of zio's write done callback,
299 			 * which means we're a zio interrupt thread.  We can't
300 			 * call dsl_deadlist_insert() now because it may block
301 			 * waiting for I/O.  Instead, put bp on the deferred
302 			 * queue and let dsl_pool_sync() finish the job.
303 			 */
304 			bplist_append(&ds->ds_pending_deadlist, bp);
305 		} else {
306 			dsl_deadlist_insert(&ds->ds_deadlist, bp, B_FALSE, tx);
307 		}
308 		ASSERT3U(ds->ds_prev->ds_object, ==,
309 		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
310 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
311 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
312 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
313 		    ds->ds_object && bp->blk_birth >
314 		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
315 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
316 			mutex_enter(&ds->ds_prev->ds_lock);
317 			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
318 			mutex_exit(&ds->ds_prev->ds_lock);
319 		}
320 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
321 			dsl_dir_transfer_space(ds->ds_dir, used,
322 			    DD_USED_HEAD, DD_USED_SNAP, tx);
323 		}
324 	}
325 
326 	dsl_bookmark_block_killed(ds, bp, tx);
327 
328 	mutex_enter(&ds->ds_lock);
329 	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
330 	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
331 	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
332 	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
333 	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
334 	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
335 	mutex_exit(&ds->ds_lock);
336 
337 	return (used);
338 }
339 
340 struct feature_type_uint64_array_arg {
341 	uint64_t length;
342 	uint64_t *array;
343 };
344 
345 static void
346 unload_zfeature(dsl_dataset_t *ds, spa_feature_t f)
347 {
348 	switch (spa_feature_table[f].fi_type) {
349 	case ZFEATURE_TYPE_BOOLEAN:
350 		break;
351 	case ZFEATURE_TYPE_UINT64_ARRAY:
352 	{
353 		struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f];
354 		kmem_free(ftuaa->array, ftuaa->length * sizeof (uint64_t));
355 		kmem_free(ftuaa, sizeof (*ftuaa));
356 		break;
357 	}
358 	default:
359 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
360 	}
361 }
362 
363 static int
364 load_zfeature(objset_t *mos, dsl_dataset_t *ds, spa_feature_t f)
365 {
366 	int err = 0;
367 	switch (spa_feature_table[f].fi_type) {
368 	case ZFEATURE_TYPE_BOOLEAN:
369 		err = zap_contains(mos, ds->ds_object,
370 		    spa_feature_table[f].fi_guid);
371 		if (err == 0) {
372 			ds->ds_feature[f] = (void *)B_TRUE;
373 		} else {
374 			ASSERT3U(err, ==, ENOENT);
375 			err = 0;
376 		}
377 		break;
378 	case ZFEATURE_TYPE_UINT64_ARRAY:
379 	{
380 		uint64_t int_size, num_int;
381 		uint64_t *data;
382 		err = zap_length(mos, ds->ds_object,
383 		    spa_feature_table[f].fi_guid, &int_size, &num_int);
384 		if (err != 0) {
385 			ASSERT3U(err, ==, ENOENT);
386 			err = 0;
387 			break;
388 		}
389 		ASSERT3U(int_size, ==, sizeof (uint64_t));
390 		data = kmem_alloc(int_size * num_int, KM_SLEEP);
391 		VERIFY0(zap_lookup(mos, ds->ds_object,
392 		    spa_feature_table[f].fi_guid, int_size, num_int, data));
393 		struct feature_type_uint64_array_arg *ftuaa =
394 		    kmem_alloc(sizeof (*ftuaa), KM_SLEEP);
395 		ftuaa->length = num_int;
396 		ftuaa->array = data;
397 		ds->ds_feature[f] = ftuaa;
398 		break;
399 	}
400 	default:
401 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
402 	}
403 	return (err);
404 }
405 
406 /*
407  * We have to release the fsid synchronously or we risk that a subsequent
408  * mount of the same dataset will fail to unique_insert the fsid.  This
409  * failure would manifest itself as the fsid of this dataset changing
410  * between mounts which makes NFS clients quite unhappy.
411  */
412 static void
413 dsl_dataset_evict_sync(void *dbu)
414 {
415 	dsl_dataset_t *ds = dbu;
416 
417 	ASSERT(ds->ds_owner == NULL);
418 
419 	unique_remove(ds->ds_fsid_guid);
420 }
421 
422 static void
423 dsl_dataset_evict_async(void *dbu)
424 {
425 	dsl_dataset_t *ds = dbu;
426 
427 	ASSERT(ds->ds_owner == NULL);
428 
429 	ds->ds_dbuf = NULL;
430 
431 	if (ds->ds_objset != NULL)
432 		dmu_objset_evict(ds->ds_objset);
433 
434 	if (ds->ds_prev) {
435 		dsl_dataset_rele(ds->ds_prev, ds);
436 		ds->ds_prev = NULL;
437 	}
438 
439 	dsl_bookmark_fini_ds(ds);
440 
441 	bplist_destroy(&ds->ds_pending_deadlist);
442 	if (dsl_deadlist_is_open(&ds->ds_deadlist))
443 		dsl_deadlist_close(&ds->ds_deadlist);
444 	if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
445 		dsl_deadlist_close(&ds->ds_remap_deadlist);
446 	if (ds->ds_dir)
447 		dsl_dir_async_rele(ds->ds_dir, ds);
448 
449 	ASSERT(!list_link_active(&ds->ds_synced_link));
450 
451 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
452 		if (dsl_dataset_feature_is_active(ds, f))
453 			unload_zfeature(ds, f);
454 	}
455 
456 	list_destroy(&ds->ds_prop_cbs);
457 	mutex_destroy(&ds->ds_lock);
458 	mutex_destroy(&ds->ds_opening_lock);
459 	mutex_destroy(&ds->ds_sendstream_lock);
460 	mutex_destroy(&ds->ds_remap_deadlist_lock);
461 	zfs_refcount_destroy(&ds->ds_longholds);
462 	rrw_destroy(&ds->ds_bp_rwlock);
463 
464 	kmem_free(ds, sizeof (dsl_dataset_t));
465 }
466 
467 int
468 dsl_dataset_get_snapname(dsl_dataset_t *ds)
469 {
470 	dsl_dataset_phys_t *headphys;
471 	int err;
472 	dmu_buf_t *headdbuf;
473 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
474 	objset_t *mos = dp->dp_meta_objset;
475 
476 	if (ds->ds_snapname[0])
477 		return (0);
478 	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
479 		return (0);
480 
481 	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
482 	    FTAG, &headdbuf);
483 	if (err != 0)
484 		return (err);
485 	headphys = headdbuf->db_data;
486 	err = zap_value_search(dp->dp_meta_objset,
487 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
488 	if (err != 0 && zfs_recover == B_TRUE) {
489 		err = 0;
490 		(void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname),
491 		    "SNAPOBJ=%llu-ERR=%d",
492 		    (unsigned long long)ds->ds_object, err);
493 	}
494 	dmu_buf_rele(headdbuf, FTAG);
495 	return (err);
496 }
497 
498 int
499 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
500 {
501 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
502 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
503 	matchtype_t mt = 0;
504 	int err;
505 
506 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
507 		mt = MT_NORMALIZE;
508 
509 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
510 	    value, mt, NULL, 0, NULL);
511 	if (err == ENOTSUP && (mt & MT_NORMALIZE))
512 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
513 	return (err);
514 }
515 
516 int
517 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
518     boolean_t adj_cnt)
519 {
520 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
521 	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
522 	matchtype_t mt = 0;
523 	int err;
524 
525 	dsl_dir_snap_cmtime_update(ds->ds_dir);
526 
527 	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
528 		mt = MT_NORMALIZE;
529 
530 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
531 	if (err == ENOTSUP && (mt & MT_NORMALIZE))
532 		err = zap_remove(mos, snapobj, name, tx);
533 
534 	if (err == 0 && adj_cnt)
535 		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
536 		    DD_FIELD_SNAPSHOT_COUNT, tx);
537 
538 	return (err);
539 }
540 
541 boolean_t
542 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
543 {
544 	dmu_buf_t *dbuf = ds->ds_dbuf;
545 	boolean_t result = B_FALSE;
546 
547 	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
548 	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
549 
550 		if (ds == dmu_buf_get_user(dbuf))
551 			result = B_TRUE;
552 		else
553 			dmu_buf_rele(dbuf, tag);
554 	}
555 
556 	return (result);
557 }
558 
559 int
560 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
561     dsl_dataset_t **dsp)
562 {
563 	objset_t *mos = dp->dp_meta_objset;
564 	dmu_buf_t *dbuf;
565 	dsl_dataset_t *ds;
566 	int err;
567 	dmu_object_info_t doi;
568 
569 	ASSERT(dsl_pool_config_held(dp));
570 
571 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
572 	if (err != 0)
573 		return (err);
574 
575 	/* Make sure dsobj has the correct object type. */
576 	dmu_object_info_from_db(dbuf, &doi);
577 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
578 		dmu_buf_rele(dbuf, tag);
579 		return (SET_ERROR(EINVAL));
580 	}
581 
582 	ds = dmu_buf_get_user(dbuf);
583 	if (ds == NULL) {
584 		dsl_dataset_t *winner = NULL;
585 
586 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
587 		ds->ds_dbuf = dbuf;
588 		ds->ds_object = dsobj;
589 		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
590 		list_link_init(&ds->ds_synced_link);
591 
592 		err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj,
593 		    NULL, ds, &ds->ds_dir);
594 		if (err != 0) {
595 			kmem_free(ds, sizeof (dsl_dataset_t));
596 			dmu_buf_rele(dbuf, tag);
597 			return (err);
598 		}
599 
600 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
601 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
602 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
603 		mutex_init(&ds->ds_remap_deadlist_lock,
604 		    NULL, MUTEX_DEFAULT, NULL);
605 		rrw_init(&ds->ds_bp_rwlock, B_FALSE);
606 		zfs_refcount_create(&ds->ds_longholds);
607 
608 		bplist_create(&ds->ds_pending_deadlist);
609 
610 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendstatus_t),
611 		    offsetof(dmu_sendstatus_t, dss_link));
612 
613 		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
614 		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
615 
616 		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
617 			spa_feature_t f;
618 
619 			for (f = 0; f < SPA_FEATURES; f++) {
620 				if (!(spa_feature_table[f].fi_flags &
621 				    ZFEATURE_FLAG_PER_DATASET))
622 					continue;
623 				err = load_zfeature(mos, ds, f);
624 			}
625 		}
626 
627 		if (!ds->ds_is_snapshot) {
628 			ds->ds_snapname[0] = '\0';
629 			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
630 				err = dsl_dataset_hold_obj(dp,
631 				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
632 				    ds, &ds->ds_prev);
633 			}
634 			err = dsl_bookmark_init_ds(ds);
635 		} else {
636 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
637 				err = dsl_dataset_get_snapname(ds);
638 			if (err == 0 &&
639 			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
640 				err = zap_count(
641 				    ds->ds_dir->dd_pool->dp_meta_objset,
642 				    dsl_dataset_phys(ds)->ds_userrefs_obj,
643 				    &ds->ds_userrefs);
644 			}
645 		}
646 
647 		if (err == 0 && !ds->ds_is_snapshot) {
648 			err = dsl_prop_get_int_ds(ds,
649 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
650 			    &ds->ds_reserved);
651 			if (err == 0) {
652 				err = dsl_prop_get_int_ds(ds,
653 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
654 				    &ds->ds_quota);
655 			}
656 		} else {
657 			ds->ds_reserved = ds->ds_quota = 0;
658 		}
659 
660 		if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 &&
661 		    ds->ds_is_snapshot &&
662 		    zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) {
663 			dp->dp_spa->spa_errata =
664 			    ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
665 		}
666 
667 		dsl_deadlist_open(&ds->ds_deadlist,
668 		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
669 		uint64_t remap_deadlist_obj =
670 		    dsl_dataset_get_remap_deadlist_object(ds);
671 		if (remap_deadlist_obj != 0) {
672 			dsl_deadlist_open(&ds->ds_remap_deadlist, mos,
673 			    remap_deadlist_obj);
674 		}
675 
676 		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
677 		    dsl_dataset_evict_async, &ds->ds_dbuf);
678 		if (err == 0)
679 			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
680 
681 		if (err != 0 || winner != NULL) {
682 			bplist_destroy(&ds->ds_pending_deadlist);
683 			dsl_deadlist_close(&ds->ds_deadlist);
684 			if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
685 				dsl_deadlist_close(&ds->ds_remap_deadlist);
686 			dsl_bookmark_fini_ds(ds);
687 			if (ds->ds_prev)
688 				dsl_dataset_rele(ds->ds_prev, ds);
689 			dsl_dir_rele(ds->ds_dir, ds);
690 			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
691 				if (dsl_dataset_feature_is_active(ds, f))
692 					unload_zfeature(ds, f);
693 			}
694 
695 			list_destroy(&ds->ds_prop_cbs);
696 			list_destroy(&ds->ds_sendstreams);
697 			mutex_destroy(&ds->ds_lock);
698 			mutex_destroy(&ds->ds_opening_lock);
699 			mutex_destroy(&ds->ds_sendstream_lock);
700 			mutex_destroy(&ds->ds_remap_deadlist_lock);
701 			zfs_refcount_destroy(&ds->ds_longholds);
702 			rrw_destroy(&ds->ds_bp_rwlock);
703 			kmem_free(ds, sizeof (dsl_dataset_t));
704 			if (err != 0) {
705 				dmu_buf_rele(dbuf, tag);
706 				return (err);
707 			}
708 			ds = winner;
709 		} else {
710 			ds->ds_fsid_guid =
711 			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
712 			if (ds->ds_fsid_guid !=
713 			    dsl_dataset_phys(ds)->ds_fsid_guid) {
714 				zfs_dbgmsg("ds_fsid_guid changed from "
715 				    "%llx to %llx for pool %s dataset id %llu",
716 				    (long long)
717 				    dsl_dataset_phys(ds)->ds_fsid_guid,
718 				    (long long)ds->ds_fsid_guid,
719 				    spa_name(dp->dp_spa),
720 				    (u_longlong_t)dsobj);
721 			}
722 		}
723 	}
724 
725 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
726 	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
727 	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
728 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
729 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
730 	*dsp = ds;
731 
732 	return (0);
733 }
734 
735 int
736 dsl_dataset_create_key_mapping(dsl_dataset_t *ds)
737 {
738 	dsl_dir_t *dd = ds->ds_dir;
739 
740 	if (dd->dd_crypto_obj == 0)
741 		return (0);
742 
743 	return (spa_keystore_create_mapping(dd->dd_pool->dp_spa,
744 	    ds, ds, &ds->ds_key_mapping));
745 }
746 
747 int
748 dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj,
749     ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp)
750 {
751 	int err;
752 
753 	err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
754 	if (err != 0)
755 		return (err);
756 
757 	ASSERT3P(*dsp, !=, NULL);
758 
759 	if (flags & DS_HOLD_FLAG_DECRYPT) {
760 		err = dsl_dataset_create_key_mapping(*dsp);
761 		if (err != 0)
762 			dsl_dataset_rele(*dsp, tag);
763 	}
764 
765 	return (err);
766 }
767 
768 int
769 dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
770     void *tag, dsl_dataset_t **dsp)
771 {
772 	dsl_dir_t *dd;
773 	const char *snapname;
774 	uint64_t obj;
775 	int err = 0;
776 	dsl_dataset_t *ds;
777 
778 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
779 	if (err != 0)
780 		return (err);
781 
782 	ASSERT(dsl_pool_config_held(dp));
783 	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
784 	if (obj != 0)
785 		err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds);
786 	else
787 		err = SET_ERROR(ENOENT);
788 
789 	/* we may be looking for a snapshot */
790 	if (err == 0 && snapname != NULL) {
791 		dsl_dataset_t *snap_ds;
792 
793 		if (*snapname++ != '@') {
794 			dsl_dataset_rele_flags(ds, flags, tag);
795 			dsl_dir_rele(dd, FTAG);
796 			return (SET_ERROR(ENOENT));
797 		}
798 
799 		dprintf("looking for snapshot '%s'\n", snapname);
800 		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
801 		if (err == 0) {
802 			err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag,
803 			    &snap_ds);
804 		}
805 		dsl_dataset_rele_flags(ds, flags, tag);
806 
807 		if (err == 0) {
808 			mutex_enter(&snap_ds->ds_lock);
809 			if (snap_ds->ds_snapname[0] == 0)
810 				(void) strlcpy(snap_ds->ds_snapname, snapname,
811 				    sizeof (snap_ds->ds_snapname));
812 			mutex_exit(&snap_ds->ds_lock);
813 			ds = snap_ds;
814 		}
815 	}
816 	if (err == 0)
817 		*dsp = ds;
818 	dsl_dir_rele(dd, FTAG);
819 	return (err);
820 }
821 
822 int
823 dsl_dataset_hold(dsl_pool_t *dp, const char *name, void *tag,
824     dsl_dataset_t **dsp)
825 {
826 	return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp));
827 }
828 
829 static int
830 dsl_dataset_own_obj_impl(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
831     void *tag, boolean_t override, dsl_dataset_t **dsp)
832 {
833 	int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp);
834 	if (err != 0)
835 		return (err);
836 	if (!dsl_dataset_tryown(*dsp, tag, override)) {
837 		dsl_dataset_rele_flags(*dsp, flags, tag);
838 		*dsp = NULL;
839 		return (SET_ERROR(EBUSY));
840 	}
841 	return (0);
842 }
843 
844 
845 int
846 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
847     void *tag, dsl_dataset_t **dsp)
848 {
849 	return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_FALSE, dsp));
850 }
851 
852 int
853 dsl_dataset_own_obj_force(dsl_pool_t *dp, uint64_t dsobj,
854     ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp)
855 {
856 	return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_TRUE, dsp));
857 }
858 
859 static int
860 dsl_dataset_own_impl(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
861     void *tag, boolean_t override, dsl_dataset_t **dsp)
862 {
863 	int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp);
864 	if (err != 0)
865 		return (err);
866 	if (!dsl_dataset_tryown(*dsp, tag, override)) {
867 		dsl_dataset_rele_flags(*dsp, flags, tag);
868 		return (SET_ERROR(EBUSY));
869 	}
870 	return (0);
871 }
872 
873 int
874 dsl_dataset_own_force(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
875     void *tag, dsl_dataset_t **dsp)
876 {
877 	return (dsl_dataset_own_impl(dp, name, flags, tag, B_TRUE, dsp));
878 }
879 
880 int
881 dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
882     void *tag, dsl_dataset_t **dsp)
883 {
884 	return (dsl_dataset_own_impl(dp, name, flags, tag, B_FALSE, dsp));
885 }
886 
887 /*
888  * See the comment above dsl_pool_hold() for details.  In summary, a long
889  * hold is used to prevent destruction of a dataset while the pool hold
890  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
891  *
892  * The dataset and pool must be held when this function is called.  After it
893  * is called, the pool hold may be released while the dataset is still held
894  * and accessed.
895  */
896 void
897 dsl_dataset_long_hold(dsl_dataset_t *ds, const void *tag)
898 {
899 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
900 	(void) zfs_refcount_add(&ds->ds_longholds, tag);
901 }
902 
903 void
904 dsl_dataset_long_rele(dsl_dataset_t *ds, const void *tag)
905 {
906 	(void) zfs_refcount_remove(&ds->ds_longholds, tag);
907 }
908 
909 /* Return B_TRUE if there are any long holds on this dataset. */
910 boolean_t
911 dsl_dataset_long_held(dsl_dataset_t *ds)
912 {
913 	return (!zfs_refcount_is_zero(&ds->ds_longholds));
914 }
915 
916 void
917 dsl_dataset_name(dsl_dataset_t *ds, char *name)
918 {
919 	if (ds == NULL) {
920 		(void) strlcpy(name, "mos", ZFS_MAX_DATASET_NAME_LEN);
921 	} else {
922 		dsl_dir_name(ds->ds_dir, name);
923 		VERIFY0(dsl_dataset_get_snapname(ds));
924 		if (ds->ds_snapname[0]) {
925 			VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
926 			    <, ZFS_MAX_DATASET_NAME_LEN);
927 			/*
928 			 * We use a "recursive" mutex so that we
929 			 * can call dprintf_ds() with ds_lock held.
930 			 */
931 			if (!MUTEX_HELD(&ds->ds_lock)) {
932 				mutex_enter(&ds->ds_lock);
933 				VERIFY3U(strlcat(name, ds->ds_snapname,
934 				    ZFS_MAX_DATASET_NAME_LEN), <,
935 				    ZFS_MAX_DATASET_NAME_LEN);
936 				mutex_exit(&ds->ds_lock);
937 			} else {
938 				VERIFY3U(strlcat(name, ds->ds_snapname,
939 				    ZFS_MAX_DATASET_NAME_LEN), <,
940 				    ZFS_MAX_DATASET_NAME_LEN);
941 			}
942 		}
943 	}
944 }
945 
946 int
947 dsl_dataset_namelen(dsl_dataset_t *ds)
948 {
949 	VERIFY0(dsl_dataset_get_snapname(ds));
950 	mutex_enter(&ds->ds_lock);
951 	int len = strlen(ds->ds_snapname);
952 	mutex_exit(&ds->ds_lock);
953 	/* add '@' if ds is a snap */
954 	if (len > 0)
955 		len++;
956 	len += dsl_dir_namelen(ds->ds_dir);
957 	return (len);
958 }
959 
960 void
961 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
962 {
963 	dmu_buf_rele(ds->ds_dbuf, tag);
964 }
965 
966 void
967 dsl_dataset_remove_key_mapping(dsl_dataset_t *ds)
968 {
969 	dsl_dir_t *dd = ds->ds_dir;
970 
971 	if (dd == NULL || dd->dd_crypto_obj == 0)
972 		return;
973 
974 	(void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa,
975 	    ds->ds_object, ds);
976 }
977 
978 void
979 dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag)
980 {
981 	if (flags & DS_HOLD_FLAG_DECRYPT)
982 		dsl_dataset_remove_key_mapping(ds);
983 
984 	dsl_dataset_rele(ds, tag);
985 }
986 
987 void
988 dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag)
989 {
990 	ASSERT3P(ds->ds_owner, ==, tag);
991 	ASSERT(ds->ds_dbuf != NULL);
992 
993 	mutex_enter(&ds->ds_lock);
994 	ds->ds_owner = NULL;
995 	mutex_exit(&ds->ds_lock);
996 	dsl_dataset_long_rele(ds, tag);
997 	dsl_dataset_rele_flags(ds, flags, tag);
998 }
999 
1000 boolean_t
1001 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag, boolean_t override)
1002 {
1003 	boolean_t gotit = FALSE;
1004 
1005 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1006 	mutex_enter(&ds->ds_lock);
1007 	if (ds->ds_owner == NULL && (override || !(DS_IS_INCONSISTENT(ds) ||
1008 	    (dsl_dataset_feature_is_active(ds,
1009 	    SPA_FEATURE_REDACTED_DATASETS) &&
1010 	    !zfs_allow_redacted_dataset_mount)))) {
1011 		ds->ds_owner = tag;
1012 		dsl_dataset_long_hold(ds, tag);
1013 		gotit = TRUE;
1014 	}
1015 	mutex_exit(&ds->ds_lock);
1016 	return (gotit);
1017 }
1018 
1019 boolean_t
1020 dsl_dataset_has_owner(dsl_dataset_t *ds)
1021 {
1022 	boolean_t rv;
1023 	mutex_enter(&ds->ds_lock);
1024 	rv = (ds->ds_owner != NULL);
1025 	mutex_exit(&ds->ds_lock);
1026 	return (rv);
1027 }
1028 
1029 static boolean_t
1030 zfeature_active(spa_feature_t f, void *arg)
1031 {
1032 	switch (spa_feature_table[f].fi_type) {
1033 	case ZFEATURE_TYPE_BOOLEAN: {
1034 		boolean_t val = (boolean_t)(uintptr_t)arg;
1035 		ASSERT(val == B_FALSE || val == B_TRUE);
1036 		return (val);
1037 	}
1038 	case ZFEATURE_TYPE_UINT64_ARRAY:
1039 		/*
1040 		 * In this case, arg is a uint64_t array.  The feature is active
1041 		 * if the array is non-null.
1042 		 */
1043 		return (arg != NULL);
1044 	default:
1045 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
1046 		return (B_FALSE);
1047 	}
1048 }
1049 
1050 boolean_t
1051 dsl_dataset_feature_is_active(dsl_dataset_t *ds, spa_feature_t f)
1052 {
1053 	return (zfeature_active(f, ds->ds_feature[f]));
1054 }
1055 
1056 /*
1057  * The buffers passed out by this function are references to internal buffers;
1058  * they should not be freed by callers of this function, and they should not be
1059  * used after the dataset has been released.
1060  */
1061 boolean_t
1062 dsl_dataset_get_uint64_array_feature(dsl_dataset_t *ds, spa_feature_t f,
1063     uint64_t *outlength, uint64_t **outp)
1064 {
1065 	VERIFY(spa_feature_table[f].fi_type & ZFEATURE_TYPE_UINT64_ARRAY);
1066 	if (!dsl_dataset_feature_is_active(ds, f)) {
1067 		return (B_FALSE);
1068 	}
1069 	struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f];
1070 	*outp = ftuaa->array;
1071 	*outlength = ftuaa->length;
1072 	return (B_TRUE);
1073 }
1074 
1075 void
1076 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, void *arg,
1077     dmu_tx_t *tx)
1078 {
1079 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1080 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
1081 	uint64_t zero = 0;
1082 
1083 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
1084 
1085 	spa_feature_incr(spa, f, tx);
1086 	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1087 
1088 	switch (spa_feature_table[f].fi_type) {
1089 	case ZFEATURE_TYPE_BOOLEAN:
1090 		ASSERT3S((boolean_t)(uintptr_t)arg, ==, B_TRUE);
1091 		VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
1092 		    sizeof (zero), 1, &zero, tx));
1093 		break;
1094 	case ZFEATURE_TYPE_UINT64_ARRAY:
1095 	{
1096 		struct feature_type_uint64_array_arg *ftuaa = arg;
1097 		VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
1098 		    sizeof (uint64_t), ftuaa->length, ftuaa->array, tx));
1099 		break;
1100 	}
1101 	default:
1102 		panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
1103 	}
1104 }
1105 
1106 static void
1107 dsl_dataset_deactivate_feature_impl(dsl_dataset_t *ds, spa_feature_t f,
1108     dmu_tx_t *tx)
1109 {
1110 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1111 	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
1112 	uint64_t dsobj = ds->ds_object;
1113 
1114 	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
1115 
1116 	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
1117 	spa_feature_decr(spa, f, tx);
1118 	ds->ds_feature[f] = NULL;
1119 }
1120 
1121 void
1122 dsl_dataset_deactivate_feature(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx)
1123 {
1124 	unload_zfeature(ds, f);
1125 	dsl_dataset_deactivate_feature_impl(ds, f, tx);
1126 }
1127 
1128 uint64_t
1129 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
1130     dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx)
1131 {
1132 	dsl_pool_t *dp = dd->dd_pool;
1133 	dmu_buf_t *dbuf;
1134 	dsl_dataset_phys_t *dsphys;
1135 	uint64_t dsobj;
1136 	objset_t *mos = dp->dp_meta_objset;
1137 
1138 	if (origin == NULL)
1139 		origin = dp->dp_origin_snap;
1140 
1141 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
1142 	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
1143 	ASSERT(dmu_tx_is_syncing(tx));
1144 	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1145 
1146 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1147 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1148 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1149 	dmu_buf_will_dirty(dbuf, tx);
1150 	dsphys = dbuf->db_data;
1151 	memset(dsphys, 0, sizeof (dsl_dataset_phys_t));
1152 	dsphys->ds_dir_obj = dd->dd_object;
1153 	dsphys->ds_flags = flags;
1154 	dsphys->ds_fsid_guid = unique_create();
1155 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1156 	    sizeof (dsphys->ds_guid));
1157 	dsphys->ds_snapnames_zapobj =
1158 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
1159 	    DMU_OT_NONE, 0, tx);
1160 	dsphys->ds_creation_time = gethrestime_sec();
1161 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
1162 
1163 	if (origin == NULL) {
1164 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
1165 	} else {
1166 		dsl_dataset_t *ohds; /* head of the origin snapshot */
1167 
1168 		dsphys->ds_prev_snap_obj = origin->ds_object;
1169 		dsphys->ds_prev_snap_txg =
1170 		    dsl_dataset_phys(origin)->ds_creation_txg;
1171 		dsphys->ds_referenced_bytes =
1172 		    dsl_dataset_phys(origin)->ds_referenced_bytes;
1173 		dsphys->ds_compressed_bytes =
1174 		    dsl_dataset_phys(origin)->ds_compressed_bytes;
1175 		dsphys->ds_uncompressed_bytes =
1176 		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
1177 		rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
1178 		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
1179 		rrw_exit(&origin->ds_bp_rwlock, FTAG);
1180 
1181 		/*
1182 		 * Inherit flags that describe the dataset's contents
1183 		 * (INCONSISTENT) or properties (Case Insensitive).
1184 		 */
1185 		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
1186 		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
1187 
1188 		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1189 			if (zfeature_active(f, origin->ds_feature[f])) {
1190 				dsl_dataset_activate_feature(dsobj, f,
1191 				    origin->ds_feature[f], tx);
1192 			}
1193 		}
1194 
1195 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
1196 		dsl_dataset_phys(origin)->ds_num_children++;
1197 
1198 		VERIFY0(dsl_dataset_hold_obj(dp,
1199 		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
1200 		    FTAG, &ohds));
1201 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
1202 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
1203 		dsl_dataset_rele(ohds, FTAG);
1204 
1205 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
1206 			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
1207 				dsl_dataset_phys(origin)->ds_next_clones_obj =
1208 				    zap_create(mos,
1209 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
1210 			}
1211 			VERIFY0(zap_add_int(mos,
1212 			    dsl_dataset_phys(origin)->ds_next_clones_obj,
1213 			    dsobj, tx));
1214 		}
1215 
1216 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
1217 		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
1218 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1219 			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
1220 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
1221 				dsl_dir_phys(origin->ds_dir)->dd_clones =
1222 				    zap_create(mos,
1223 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
1224 			}
1225 			VERIFY0(zap_add_int(mos,
1226 			    dsl_dir_phys(origin->ds_dir)->dd_clones,
1227 			    dsobj, tx));
1228 		}
1229 	}
1230 
1231 	/* handle encryption */
1232 	dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx);
1233 
1234 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1235 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1236 
1237 	dmu_buf_rele(dbuf, FTAG);
1238 
1239 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1240 	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
1241 
1242 	return (dsobj);
1243 }
1244 
1245 static void
1246 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
1247 {
1248 	objset_t *os;
1249 
1250 	VERIFY0(dmu_objset_from_ds(ds, &os));
1251 	if (memcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
1252 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1253 		zio_t *zio;
1254 
1255 		memset(&os->os_zil_header, 0, sizeof (os->os_zil_header));
1256 		if (os->os_encrypted)
1257 			os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE;
1258 
1259 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1260 		dsl_dataset_sync(ds, zio, tx);
1261 		VERIFY0(zio_wait(zio));
1262 
1263 		/* dsl_dataset_sync_done will drop this reference. */
1264 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1265 		dsl_dataset_sync_done(ds, tx);
1266 	}
1267 }
1268 
1269 uint64_t
1270 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
1271     dsl_dataset_t *origin, uint64_t flags, cred_t *cr,
1272     dsl_crypto_params_t *dcp, dmu_tx_t *tx)
1273 {
1274 	dsl_pool_t *dp = pdd->dd_pool;
1275 	uint64_t dsobj, ddobj;
1276 	dsl_dir_t *dd;
1277 
1278 	ASSERT(dmu_tx_is_syncing(tx));
1279 	ASSERT(lastname[0] != '@');
1280 	/*
1281 	 * Filesystems will eventually have their origin set to dp_origin_snap,
1282 	 * but that's taken care of in dsl_dataset_create_sync_dd. When
1283 	 * creating a filesystem, this function is called with origin equal to
1284 	 * NULL.
1285 	 */
1286 	if (origin != NULL)
1287 		ASSERT3P(origin, !=, dp->dp_origin_snap);
1288 
1289 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
1290 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
1291 
1292 	dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp,
1293 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
1294 
1295 	dsl_deleg_set_create_perms(dd, tx, cr);
1296 
1297 	/*
1298 	 * If we are creating a clone and the livelist feature is enabled,
1299 	 * add the entry DD_FIELD_LIVELIST to ZAP.
1300 	 */
1301 	if (origin != NULL &&
1302 	    spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LIVELIST)) {
1303 		objset_t *mos = dd->dd_pool->dp_meta_objset;
1304 		dsl_dir_zapify(dd, tx);
1305 		uint64_t obj = dsl_deadlist_alloc(mos, tx);
1306 		VERIFY0(zap_add(mos, dd->dd_object, DD_FIELD_LIVELIST,
1307 		    sizeof (uint64_t), 1, &obj, tx));
1308 		spa_feature_incr(dp->dp_spa, SPA_FEATURE_LIVELIST, tx);
1309 	}
1310 
1311 	/*
1312 	 * Since we're creating a new node we know it's a leaf, so we can
1313 	 * initialize the counts if the limit feature is active.
1314 	 */
1315 	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
1316 		uint64_t cnt = 0;
1317 		objset_t *os = dd->dd_pool->dp_meta_objset;
1318 
1319 		dsl_dir_zapify(dd, tx);
1320 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1321 		    sizeof (cnt), 1, &cnt, tx));
1322 		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1323 		    sizeof (cnt), 1, &cnt, tx));
1324 	}
1325 
1326 	dsl_dir_rele(dd, FTAG);
1327 
1328 	/*
1329 	 * If we are creating a clone, make sure we zero out any stale
1330 	 * data from the origin snapshots zil header.
1331 	 */
1332 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
1333 		dsl_dataset_t *ds;
1334 
1335 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1336 		dsl_dataset_zero_zil(ds, tx);
1337 		dsl_dataset_rele(ds, FTAG);
1338 	}
1339 
1340 	return (dsobj);
1341 }
1342 
1343 /*
1344  * The unique space in the head dataset can be calculated by subtracting
1345  * the space used in the most recent snapshot, that is still being used
1346  * in this file system, from the space currently in use.  To figure out
1347  * the space in the most recent snapshot still in use, we need to take
1348  * the total space used in the snapshot and subtract out the space that
1349  * has been freed up since the snapshot was taken.
1350  */
1351 void
1352 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1353 {
1354 	uint64_t mrs_used;
1355 	uint64_t dlused, dlcomp, dluncomp;
1356 
1357 	ASSERT(!ds->ds_is_snapshot);
1358 
1359 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1360 		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1361 	else
1362 		mrs_used = 0;
1363 
1364 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1365 
1366 	ASSERT3U(dlused, <=, mrs_used);
1367 	dsl_dataset_phys(ds)->ds_unique_bytes =
1368 	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1369 
1370 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1371 	    SPA_VERSION_UNIQUE_ACCURATE)
1372 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1373 }
1374 
1375 void
1376 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1377     dmu_tx_t *tx)
1378 {
1379 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1380 	uint64_t count __maybe_unused;
1381 	int err;
1382 
1383 	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1384 	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1385 	    obj, tx);
1386 	/*
1387 	 * The err should not be ENOENT, but a bug in a previous version
1388 	 * of the code could cause upgrade_clones_cb() to not set
1389 	 * ds_next_snap_obj when it should, leading to a missing entry.
1390 	 * If we knew that the pool was created after
1391 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1392 	 * ENOENT.  However, at least we can check that we don't have
1393 	 * too many entries in the next_clones_obj even after failing to
1394 	 * remove this one.
1395 	 */
1396 	if (err != ENOENT)
1397 		VERIFY0(err);
1398 	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1399 	    &count));
1400 	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1401 }
1402 
1403 
1404 blkptr_t *
1405 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1406 {
1407 	return (&dsl_dataset_phys(ds)->ds_bp);
1408 }
1409 
1410 spa_t *
1411 dsl_dataset_get_spa(dsl_dataset_t *ds)
1412 {
1413 	return (ds->ds_dir->dd_pool->dp_spa);
1414 }
1415 
1416 void
1417 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1418 {
1419 	dsl_pool_t *dp;
1420 
1421 	if (ds == NULL) /* this is the meta-objset */
1422 		return;
1423 
1424 	ASSERT(ds->ds_objset != NULL);
1425 
1426 	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1427 		panic("dirtying snapshot!");
1428 
1429 	/* Must not dirty a dataset in the same txg where it got snapshotted. */
1430 	ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1431 
1432 	dp = ds->ds_dir->dd_pool;
1433 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1434 		objset_t *os = ds->ds_objset;
1435 
1436 		/* up the hold count until we can be written out */
1437 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1438 
1439 		/* if this dataset is encrypted, grab a reference to the DCK */
1440 		if (ds->ds_dir->dd_crypto_obj != 0 &&
1441 		    !os->os_raw_receive &&
1442 		    !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
1443 			ASSERT3P(ds->ds_key_mapping, !=, NULL);
1444 			key_mapping_add_ref(ds->ds_key_mapping, ds);
1445 		}
1446 	}
1447 }
1448 
1449 static int
1450 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1451 {
1452 	uint64_t asize;
1453 
1454 	if (!dmu_tx_is_syncing(tx))
1455 		return (0);
1456 
1457 	/*
1458 	 * If there's an fs-only reservation, any blocks that might become
1459 	 * owned by the snapshot dataset must be accommodated by space
1460 	 * outside of the reservation.
1461 	 */
1462 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1463 	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1464 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1465 		return (SET_ERROR(ENOSPC));
1466 
1467 	/*
1468 	 * Propagate any reserved space for this snapshot to other
1469 	 * snapshot checks in this sync group.
1470 	 */
1471 	if (asize > 0)
1472 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1473 
1474 	return (0);
1475 }
1476 
1477 int
1478 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1479     dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr, proc_t *proc)
1480 {
1481 	int error;
1482 	uint64_t value;
1483 
1484 	ds->ds_trysnap_txg = tx->tx_txg;
1485 
1486 	if (!dmu_tx_is_syncing(tx))
1487 		return (0);
1488 
1489 	/*
1490 	 * We don't allow multiple snapshots of the same txg.  If there
1491 	 * is already one, try again.
1492 	 */
1493 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1494 		return (SET_ERROR(EAGAIN));
1495 
1496 	/*
1497 	 * Check for conflicting snapshot name.
1498 	 */
1499 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1500 	if (error == 0)
1501 		return (SET_ERROR(EEXIST));
1502 	if (error != ENOENT)
1503 		return (error);
1504 
1505 	/*
1506 	 * We don't allow taking snapshots of inconsistent datasets, such as
1507 	 * those into which we are currently receiving.  However, if we are
1508 	 * creating this snapshot as part of a receive, this check will be
1509 	 * executed atomically with respect to the completion of the receive
1510 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1511 	 * case we ignore this, knowing it will be fixed up for us shortly in
1512 	 * dmu_recv_end_sync().
1513 	 */
1514 	if (!recv && DS_IS_INCONSISTENT(ds))
1515 		return (SET_ERROR(EBUSY));
1516 
1517 	/*
1518 	 * Skip the check for temporary snapshots or if we have already checked
1519 	 * the counts in dsl_dataset_snapshot_check. This means we really only
1520 	 * check the count here when we're receiving a stream.
1521 	 */
1522 	if (cnt != 0 && cr != NULL) {
1523 		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1524 		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr, proc);
1525 		if (error != 0)
1526 			return (error);
1527 	}
1528 
1529 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1530 	if (error != 0)
1531 		return (error);
1532 
1533 	return (0);
1534 }
1535 
1536 int
1537 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1538 {
1539 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1540 	dsl_pool_t *dp = dmu_tx_pool(tx);
1541 	nvpair_t *pair;
1542 	int rv = 0;
1543 
1544 	/*
1545 	 * Pre-compute how many total new snapshots will be created for each
1546 	 * level in the tree and below. This is needed for validating the
1547 	 * snapshot limit when either taking a recursive snapshot or when
1548 	 * taking multiple snapshots.
1549 	 *
1550 	 * The problem is that the counts are not actually adjusted when
1551 	 * we are checking, only when we finally sync. For a single snapshot,
1552 	 * this is easy, the count will increase by 1 at each node up the tree,
1553 	 * but its more complicated for the recursive/multiple snapshot case.
1554 	 *
1555 	 * The dsl_fs_ss_limit_check function does recursively check the count
1556 	 * at each level up the tree but since it is validating each snapshot
1557 	 * independently we need to be sure that we are validating the complete
1558 	 * count for the entire set of snapshots. We do this by rolling up the
1559 	 * counts for each component of the name into an nvlist and then
1560 	 * checking each of those cases with the aggregated count.
1561 	 *
1562 	 * This approach properly handles not only the recursive snapshot
1563 	 * case (where we get all of those on the ddsa_snaps list) but also
1564 	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1565 	 * validate the limit on 'a' using a count of 2).
1566 	 *
1567 	 * We validate the snapshot names in the third loop and only report
1568 	 * name errors once.
1569 	 */
1570 	if (dmu_tx_is_syncing(tx)) {
1571 		char *nm;
1572 		nvlist_t *cnt_track = NULL;
1573 		cnt_track = fnvlist_alloc();
1574 
1575 		nm = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1576 
1577 		/* Rollup aggregated counts into the cnt_track list */
1578 		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1579 		    pair != NULL;
1580 		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1581 			char *pdelim;
1582 			uint64_t val;
1583 
1584 			(void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN);
1585 			pdelim = strchr(nm, '@');
1586 			if (pdelim == NULL)
1587 				continue;
1588 			*pdelim = '\0';
1589 
1590 			do {
1591 				if (nvlist_lookup_uint64(cnt_track, nm,
1592 				    &val) == 0) {
1593 					/* update existing entry */
1594 					fnvlist_add_uint64(cnt_track, nm,
1595 					    val + 1);
1596 				} else {
1597 					/* add to list */
1598 					fnvlist_add_uint64(cnt_track, nm, 1);
1599 				}
1600 
1601 				pdelim = strrchr(nm, '/');
1602 				if (pdelim != NULL)
1603 					*pdelim = '\0';
1604 			} while (pdelim != NULL);
1605 		}
1606 
1607 		kmem_free(nm, MAXPATHLEN);
1608 
1609 		/* Check aggregated counts at each level */
1610 		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1611 		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1612 			int error = 0;
1613 			char *name;
1614 			uint64_t cnt = 0;
1615 			dsl_dataset_t *ds;
1616 
1617 			name = nvpair_name(pair);
1618 			cnt = fnvpair_value_uint64(pair);
1619 			ASSERT(cnt > 0);
1620 
1621 			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1622 			if (error == 0) {
1623 				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1624 				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1625 				    ddsa->ddsa_cr, ddsa->ddsa_proc);
1626 				dsl_dataset_rele(ds, FTAG);
1627 			}
1628 
1629 			if (error != 0) {
1630 				if (ddsa->ddsa_errors != NULL)
1631 					fnvlist_add_int32(ddsa->ddsa_errors,
1632 					    name, error);
1633 				rv = error;
1634 				/* only report one error for this check */
1635 				break;
1636 			}
1637 		}
1638 		nvlist_free(cnt_track);
1639 	}
1640 
1641 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1642 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1643 		int error = 0;
1644 		dsl_dataset_t *ds;
1645 		char *name, *atp = NULL;
1646 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1647 
1648 		name = nvpair_name(pair);
1649 		if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1650 			error = SET_ERROR(ENAMETOOLONG);
1651 		if (error == 0) {
1652 			atp = strchr(name, '@');
1653 			if (atp == NULL)
1654 				error = SET_ERROR(EINVAL);
1655 			if (error == 0)
1656 				(void) strlcpy(dsname, name, atp - name + 1);
1657 		}
1658 		if (error == 0)
1659 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1660 		if (error == 0) {
1661 			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1662 			error = dsl_dataset_snapshot_check_impl(ds,
1663 			    atp + 1, tx, B_FALSE, 0, NULL, NULL);
1664 			dsl_dataset_rele(ds, FTAG);
1665 		}
1666 
1667 		if (error != 0) {
1668 			if (ddsa->ddsa_errors != NULL) {
1669 				fnvlist_add_int32(ddsa->ddsa_errors,
1670 				    name, error);
1671 			}
1672 			rv = error;
1673 		}
1674 	}
1675 
1676 	return (rv);
1677 }
1678 
1679 void
1680 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1681     dmu_tx_t *tx)
1682 {
1683 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1684 	dmu_buf_t *dbuf;
1685 	dsl_dataset_phys_t *dsphys;
1686 	uint64_t dsobj, crtxg;
1687 	objset_t *mos = dp->dp_meta_objset;
1688 	static zil_header_t zero_zil __maybe_unused;
1689 	objset_t *os __maybe_unused;
1690 
1691 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1692 
1693 	/*
1694 	 * If we are on an old pool, the zil must not be active, in which
1695 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1696 	 */
1697 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1698 	    dmu_objset_from_ds(ds, &os) != 0 ||
1699 	    memcmp(&os->os_phys->os_zil_header, &zero_zil,
1700 	    sizeof (zero_zil)) == 0);
1701 
1702 	/* Should not snapshot a dirty dataset. */
1703 	ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1704 	    ds, tx->tx_txg));
1705 
1706 	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1707 
1708 	/*
1709 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1710 	 */
1711 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1712 		crtxg = 1;
1713 	else
1714 		crtxg = tx->tx_txg;
1715 
1716 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1717 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1718 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1719 	dmu_buf_will_dirty(dbuf, tx);
1720 	dsphys = dbuf->db_data;
1721 	memset(dsphys, 0, sizeof (dsl_dataset_phys_t));
1722 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1723 	dsphys->ds_fsid_guid = unique_create();
1724 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1725 	    sizeof (dsphys->ds_guid));
1726 	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1727 	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1728 	dsphys->ds_next_snap_obj = ds->ds_object;
1729 	dsphys->ds_num_children = 1;
1730 	dsphys->ds_creation_time = gethrestime_sec();
1731 	dsphys->ds_creation_txg = crtxg;
1732 	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1733 	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1734 	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1735 	dsphys->ds_uncompressed_bytes =
1736 	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1737 	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1738 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1739 	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1740 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
1741 	dmu_buf_rele(dbuf, FTAG);
1742 
1743 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1744 		if (zfeature_active(f, ds->ds_feature[f])) {
1745 			dsl_dataset_activate_feature(dsobj, f,
1746 			    ds->ds_feature[f], tx);
1747 		}
1748 	}
1749 
1750 	ASSERT3U(ds->ds_prev != 0, ==,
1751 	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1752 	if (ds->ds_prev) {
1753 		uint64_t next_clones_obj =
1754 		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1755 		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1756 		    ds->ds_object ||
1757 		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1758 		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1759 		    ds->ds_object) {
1760 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1761 			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1762 			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1763 			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1764 		} else if (next_clones_obj != 0) {
1765 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1766 			    dsphys->ds_next_snap_obj, tx);
1767 			VERIFY0(zap_add_int(mos,
1768 			    next_clones_obj, dsobj, tx));
1769 		}
1770 	}
1771 
1772 	/*
1773 	 * If we have a reference-reservation on this dataset, we will
1774 	 * need to increase the amount of refreservation being charged
1775 	 * since our unique space is going to zero.
1776 	 */
1777 	if (ds->ds_reserved) {
1778 		int64_t delta;
1779 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1780 		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1781 		    ds->ds_reserved);
1782 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1783 		    delta, 0, 0, tx);
1784 	}
1785 
1786 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1787 	dsl_dataset_phys(ds)->ds_deadlist_obj =
1788 	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1789 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1790 	dsl_deadlist_close(&ds->ds_deadlist);
1791 	dsl_deadlist_open(&ds->ds_deadlist, mos,
1792 	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1793 	dsl_deadlist_add_key(&ds->ds_deadlist,
1794 	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1795 	dsl_bookmark_snapshotted(ds, tx);
1796 
1797 	if (dsl_dataset_remap_deadlist_exists(ds)) {
1798 		uint64_t remap_deadlist_obj =
1799 		    dsl_dataset_get_remap_deadlist_object(ds);
1800 		/*
1801 		 * Move the remap_deadlist to the snapshot.  The head
1802 		 * will create a new remap deadlist on demand, from
1803 		 * dsl_dataset_block_remapped().
1804 		 */
1805 		dsl_dataset_unset_remap_deadlist_object(ds, tx);
1806 		dsl_deadlist_close(&ds->ds_remap_deadlist);
1807 
1808 		dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1809 		VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST,
1810 		    sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx));
1811 	}
1812 
1813 	/*
1814 	 * Create a ivset guid for this snapshot if the dataset is
1815 	 * encrypted. This may be overridden by a raw receive. A
1816 	 * previous implementation of this code did not have this
1817 	 * field as part of the on-disk format for ZFS encryption
1818 	 * (see errata #4). As part of the remediation for this
1819 	 * issue, we ask the user to enable the bookmark_v2 feature
1820 	 * which is now a dependency of the encryption feature. We
1821 	 * use this as a heuristic to determine when the user has
1822 	 * elected to correct any datasets created with the old code.
1823 	 * As a result, we only do this step if the bookmark_v2
1824 	 * feature is enabled, which limits the number of states a
1825 	 * given pool / dataset can be in with regards to terms of
1826 	 * correcting the issue.
1827 	 */
1828 	if (ds->ds_dir->dd_crypto_obj != 0 &&
1829 	    spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) {
1830 		uint64_t ivset_guid = unique_create();
1831 
1832 		dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1833 		VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID,
1834 		    sizeof (ivset_guid), 1, &ivset_guid, tx));
1835 	}
1836 
1837 	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1838 	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1839 	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1840 	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1841 
1842 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1843 		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1844 
1845 	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1846 	    snapname, 8, 1, &dsobj, tx));
1847 
1848 	if (ds->ds_prev)
1849 		dsl_dataset_rele(ds->ds_prev, ds);
1850 	VERIFY0(dsl_dataset_hold_obj(dp,
1851 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1852 
1853 	dsl_scan_ds_snapshotted(ds, tx);
1854 
1855 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1856 
1857 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, " ");
1858 }
1859 
1860 void
1861 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1862 {
1863 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1864 	dsl_pool_t *dp = dmu_tx_pool(tx);
1865 	nvpair_t *pair;
1866 
1867 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1868 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1869 		dsl_dataset_t *ds;
1870 		char *name, *atp;
1871 		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1872 
1873 		name = nvpair_name(pair);
1874 		atp = strchr(name, '@');
1875 		(void) strlcpy(dsname, name, atp - name + 1);
1876 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1877 
1878 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1879 		if (ddsa->ddsa_props != NULL) {
1880 			dsl_props_set_sync_impl(ds->ds_prev,
1881 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1882 		}
1883 		dsl_dataset_rele(ds, FTAG);
1884 	}
1885 }
1886 
1887 /*
1888  * The snapshots must all be in the same pool.
1889  * All-or-nothing: if there are any failures, nothing will be modified.
1890  */
1891 int
1892 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1893 {
1894 	dsl_dataset_snapshot_arg_t ddsa;
1895 	nvpair_t *pair;
1896 	boolean_t needsuspend;
1897 	int error;
1898 	spa_t *spa;
1899 	char *firstname;
1900 	nvlist_t *suspended = NULL;
1901 
1902 	pair = nvlist_next_nvpair(snaps, NULL);
1903 	if (pair == NULL)
1904 		return (0);
1905 	firstname = nvpair_name(pair);
1906 
1907 	error = spa_open(firstname, &spa, FTAG);
1908 	if (error != 0)
1909 		return (error);
1910 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1911 	spa_close(spa, FTAG);
1912 
1913 	if (needsuspend) {
1914 		suspended = fnvlist_alloc();
1915 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1916 		    pair = nvlist_next_nvpair(snaps, pair)) {
1917 			char fsname[ZFS_MAX_DATASET_NAME_LEN];
1918 			char *snapname = nvpair_name(pair);
1919 			char *atp;
1920 			void *cookie;
1921 
1922 			atp = strchr(snapname, '@');
1923 			if (atp == NULL) {
1924 				error = SET_ERROR(EINVAL);
1925 				break;
1926 			}
1927 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1928 
1929 			error = zil_suspend(fsname, &cookie);
1930 			if (error != 0)
1931 				break;
1932 			fnvlist_add_uint64(suspended, fsname,
1933 			    (uintptr_t)cookie);
1934 		}
1935 	}
1936 
1937 	ddsa.ddsa_snaps = snaps;
1938 	ddsa.ddsa_props = props;
1939 	ddsa.ddsa_errors = errors;
1940 	ddsa.ddsa_cr = CRED();
1941 	ddsa.ddsa_proc = curproc;
1942 
1943 	if (error == 0) {
1944 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1945 		    dsl_dataset_snapshot_sync, &ddsa,
1946 		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1947 	}
1948 
1949 	if (suspended != NULL) {
1950 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1951 		    pair = nvlist_next_nvpair(suspended, pair)) {
1952 			zil_resume((void *)(uintptr_t)
1953 			    fnvpair_value_uint64(pair));
1954 		}
1955 		fnvlist_free(suspended);
1956 	}
1957 
1958 	if (error == 0) {
1959 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1960 		    pair = nvlist_next_nvpair(snaps, pair)) {
1961 			zvol_create_minor(nvpair_name(pair));
1962 		}
1963 	}
1964 
1965 	return (error);
1966 }
1967 
1968 typedef struct dsl_dataset_snapshot_tmp_arg {
1969 	const char *ddsta_fsname;
1970 	const char *ddsta_snapname;
1971 	minor_t ddsta_cleanup_minor;
1972 	const char *ddsta_htag;
1973 } dsl_dataset_snapshot_tmp_arg_t;
1974 
1975 static int
1976 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1977 {
1978 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1979 	dsl_pool_t *dp = dmu_tx_pool(tx);
1980 	dsl_dataset_t *ds;
1981 	int error;
1982 
1983 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1984 	if (error != 0)
1985 		return (error);
1986 
1987 	/* NULL cred means no limit check for tmp snapshot */
1988 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1989 	    tx, B_FALSE, 0, NULL, NULL);
1990 	if (error != 0) {
1991 		dsl_dataset_rele(ds, FTAG);
1992 		return (error);
1993 	}
1994 
1995 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1996 		dsl_dataset_rele(ds, FTAG);
1997 		return (SET_ERROR(ENOTSUP));
1998 	}
1999 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
2000 	    B_TRUE, tx);
2001 	if (error != 0) {
2002 		dsl_dataset_rele(ds, FTAG);
2003 		return (error);
2004 	}
2005 
2006 	dsl_dataset_rele(ds, FTAG);
2007 	return (0);
2008 }
2009 
2010 static void
2011 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
2012 {
2013 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
2014 	dsl_pool_t *dp = dmu_tx_pool(tx);
2015 	dsl_dataset_t *ds = NULL;
2016 
2017 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
2018 
2019 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
2020 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
2021 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
2022 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
2023 
2024 	dsl_dataset_rele(ds, FTAG);
2025 }
2026 
2027 int
2028 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
2029     minor_t cleanup_minor, const char *htag)
2030 {
2031 	dsl_dataset_snapshot_tmp_arg_t ddsta;
2032 	int error;
2033 	spa_t *spa;
2034 	boolean_t needsuspend;
2035 	void *cookie;
2036 
2037 	ddsta.ddsta_fsname = fsname;
2038 	ddsta.ddsta_snapname = snapname;
2039 	ddsta.ddsta_cleanup_minor = cleanup_minor;
2040 	ddsta.ddsta_htag = htag;
2041 
2042 	error = spa_open(fsname, &spa, FTAG);
2043 	if (error != 0)
2044 		return (error);
2045 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
2046 	spa_close(spa, FTAG);
2047 
2048 	if (needsuspend) {
2049 		error = zil_suspend(fsname, &cookie);
2050 		if (error != 0)
2051 			return (error);
2052 	}
2053 
2054 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
2055 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
2056 
2057 	if (needsuspend)
2058 		zil_resume(cookie);
2059 	return (error);
2060 }
2061 
2062 void
2063 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2064 {
2065 	ASSERT(dmu_tx_is_syncing(tx));
2066 	ASSERT(ds->ds_objset != NULL);
2067 	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
2068 
2069 	/*
2070 	 * in case we had to change ds_fsid_guid when we opened it,
2071 	 * sync it out now.
2072 	 */
2073 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2074 	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
2075 
2076 	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
2077 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2078 		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
2079 		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
2080 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2081 		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
2082 		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
2083 		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2084 		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
2085 		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
2086 		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
2087 		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
2088 		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
2089 	}
2090 
2091 	dmu_objset_sync(ds->ds_objset, zio, tx);
2092 
2093 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2094 		if (zfeature_active(f, ds->ds_feature_activation[f])) {
2095 			if (zfeature_active(f, ds->ds_feature[f]))
2096 				continue;
2097 			dsl_dataset_activate_feature(ds->ds_object, f,
2098 			    ds->ds_feature_activation[f], tx);
2099 			ds->ds_feature[f] = ds->ds_feature_activation[f];
2100 		}
2101 	}
2102 }
2103 
2104 /*
2105  * Check if the percentage of blocks shared between the clone and the
2106  * snapshot (as opposed to those that are clone only) is below a certain
2107  * threshold
2108  */
2109 static boolean_t
2110 dsl_livelist_should_disable(dsl_dataset_t *ds)
2111 {
2112 	uint64_t used, referenced;
2113 	int percent_shared;
2114 
2115 	used = dsl_dir_get_usedds(ds->ds_dir);
2116 	referenced = dsl_get_referenced(ds);
2117 	ASSERT3U(referenced, >=, 0);
2118 	ASSERT3U(used, >=, 0);
2119 	if (referenced == 0)
2120 		return (B_FALSE);
2121 	percent_shared = (100 * (referenced - used)) / referenced;
2122 	if (percent_shared <= zfs_livelist_min_percent_shared)
2123 		return (B_TRUE);
2124 	return (B_FALSE);
2125 }
2126 
2127 /*
2128  *  Check if it is possible to combine two livelist entries into one.
2129  *  This is the case if the combined number of 'live' blkptrs (ALLOCs that
2130  *  don't have a matching FREE) is under the maximum sublist size.
2131  *  We check this by subtracting twice the total number of frees from the total
2132  *  number of blkptrs. FREEs are counted twice because each FREE blkptr
2133  *  will cancel out an ALLOC blkptr when the livelist is processed.
2134  */
2135 static boolean_t
2136 dsl_livelist_should_condense(dsl_deadlist_entry_t *first,
2137     dsl_deadlist_entry_t *next)
2138 {
2139 	uint64_t total_free = first->dle_bpobj.bpo_phys->bpo_num_freed +
2140 	    next->dle_bpobj.bpo_phys->bpo_num_freed;
2141 	uint64_t total_entries = first->dle_bpobj.bpo_phys->bpo_num_blkptrs +
2142 	    next->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2143 	if ((total_entries - (2 * total_free)) < zfs_livelist_max_entries)
2144 		return (B_TRUE);
2145 	return (B_FALSE);
2146 }
2147 
2148 typedef struct try_condense_arg {
2149 	spa_t *spa;
2150 	dsl_dataset_t *ds;
2151 } try_condense_arg_t;
2152 
2153 /*
2154  * Iterate over the livelist entries, searching for a pair to condense.
2155  * A nonzero return value means stop, 0 means keep looking.
2156  */
2157 static int
2158 dsl_livelist_try_condense(void *arg, dsl_deadlist_entry_t *first)
2159 {
2160 	try_condense_arg_t *tca = arg;
2161 	spa_t *spa = tca->spa;
2162 	dsl_dataset_t *ds = tca->ds;
2163 	dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist;
2164 	dsl_deadlist_entry_t *next;
2165 
2166 	/* The condense thread has not yet been created at import */
2167 	if (spa->spa_livelist_condense_zthr == NULL)
2168 		return (1);
2169 
2170 	/* A condense is already in progress */
2171 	if (spa->spa_to_condense.ds != NULL)
2172 		return (1);
2173 
2174 	next = AVL_NEXT(&ll->dl_tree, &first->dle_node);
2175 	/* The livelist has only one entry - don't condense it */
2176 	if (next == NULL)
2177 		return (1);
2178 
2179 	/* Next is the newest entry - don't condense it */
2180 	if (AVL_NEXT(&ll->dl_tree, &next->dle_node) == NULL)
2181 		return (1);
2182 
2183 	/* This pair is not ready to condense but keep looking */
2184 	if (!dsl_livelist_should_condense(first, next))
2185 		return (0);
2186 
2187 	/*
2188 	 * Add a ref to prevent the dataset from being evicted while
2189 	 * the condense zthr or synctask are running. Ref will be
2190 	 * released at the end of the condense synctask
2191 	 */
2192 	dmu_buf_add_ref(ds->ds_dbuf, spa);
2193 
2194 	spa->spa_to_condense.ds = ds;
2195 	spa->spa_to_condense.first = first;
2196 	spa->spa_to_condense.next = next;
2197 	spa->spa_to_condense.syncing = B_FALSE;
2198 	spa->spa_to_condense.cancelled = B_FALSE;
2199 
2200 	zthr_wakeup(spa->spa_livelist_condense_zthr);
2201 	return (1);
2202 }
2203 
2204 static void
2205 dsl_flush_pending_livelist(dsl_dataset_t *ds, dmu_tx_t *tx)
2206 {
2207 	dsl_dir_t *dd = ds->ds_dir;
2208 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
2209 	dsl_deadlist_entry_t *last = dsl_deadlist_last(&dd->dd_livelist);
2210 
2211 	/* Check if we need to add a new sub-livelist */
2212 	if (last == NULL) {
2213 		/* The livelist is empty */
2214 		dsl_deadlist_add_key(&dd->dd_livelist,
2215 		    tx->tx_txg - 1, tx);
2216 	} else if (spa_sync_pass(spa) == 1) {
2217 		/*
2218 		 * Check if the newest entry is full. If it is, make a new one.
2219 		 * We only do this once per sync because we could overfill a
2220 		 * sublist in one sync pass and don't want to add another entry
2221 		 * for a txg that is already represented. This ensures that
2222 		 * blkptrs born in the same txg are stored in the same sublist.
2223 		 */
2224 		bpobj_t bpobj = last->dle_bpobj;
2225 		uint64_t all = bpobj.bpo_phys->bpo_num_blkptrs;
2226 		uint64_t free = bpobj.bpo_phys->bpo_num_freed;
2227 		uint64_t alloc = all - free;
2228 		if (alloc > zfs_livelist_max_entries) {
2229 			dsl_deadlist_add_key(&dd->dd_livelist,
2230 			    tx->tx_txg - 1, tx);
2231 		}
2232 	}
2233 
2234 	/* Insert each entry into the on-disk livelist */
2235 	bplist_iterate(&dd->dd_pending_allocs,
2236 	    dsl_deadlist_insert_alloc_cb, &dd->dd_livelist, tx);
2237 	bplist_iterate(&dd->dd_pending_frees,
2238 	    dsl_deadlist_insert_free_cb, &dd->dd_livelist, tx);
2239 
2240 	/* Attempt to condense every pair of adjacent entries */
2241 	try_condense_arg_t arg = {
2242 	    .spa = spa,
2243 	    .ds = ds
2244 	};
2245 	dsl_deadlist_iterate(&dd->dd_livelist, dsl_livelist_try_condense,
2246 	    &arg);
2247 }
2248 
2249 void
2250 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
2251 {
2252 	objset_t *os = ds->ds_objset;
2253 
2254 	bplist_iterate(&ds->ds_pending_deadlist,
2255 	    dsl_deadlist_insert_alloc_cb, &ds->ds_deadlist, tx);
2256 
2257 	if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) {
2258 		dsl_flush_pending_livelist(ds, tx);
2259 		if (dsl_livelist_should_disable(ds)) {
2260 			dsl_dir_remove_livelist(ds->ds_dir, tx, B_TRUE);
2261 		}
2262 	}
2263 
2264 	dsl_bookmark_sync_done(ds, tx);
2265 
2266 	multilist_destroy(&os->os_synced_dnodes);
2267 
2268 	if (os->os_encrypted)
2269 		os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE;
2270 	else
2271 		ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]);
2272 
2273 	ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
2274 
2275 	dmu_buf_rele(ds->ds_dbuf, ds);
2276 }
2277 
2278 int
2279 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val)
2280 {
2281 	uint64_t count = 0;
2282 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2283 	zap_cursor_t zc;
2284 	zap_attribute_t za;
2285 
2286 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
2287 
2288 	/*
2289 	 * There may be missing entries in ds_next_clones_obj
2290 	 * due to a bug in a previous version of the code.
2291 	 * Only trust it if it has the right number of entries.
2292 	 */
2293 	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
2294 		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
2295 		    &count));
2296 	}
2297 	if (count != dsl_dataset_phys(ds)->ds_num_children - 1) {
2298 		return (SET_ERROR(ENOENT));
2299 	}
2300 	for (zap_cursor_init(&zc, mos,
2301 	    dsl_dataset_phys(ds)->ds_next_clones_obj);
2302 	    zap_cursor_retrieve(&zc, &za) == 0;
2303 	    zap_cursor_advance(&zc)) {
2304 		dsl_dataset_t *clone;
2305 		char buf[ZFS_MAX_DATASET_NAME_LEN];
2306 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2307 		    za.za_first_integer, FTAG, &clone));
2308 		dsl_dir_name(clone->ds_dir, buf);
2309 		fnvlist_add_boolean(val, buf);
2310 		dsl_dataset_rele(clone, FTAG);
2311 	}
2312 	zap_cursor_fini(&zc);
2313 	return (0);
2314 }
2315 
2316 void
2317 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2318 {
2319 	nvlist_t *propval = fnvlist_alloc();
2320 	nvlist_t *val = fnvlist_alloc();
2321 
2322 	if (get_clones_stat_impl(ds, val) == 0) {
2323 		fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
2324 		fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2325 		    propval);
2326 	}
2327 
2328 	nvlist_free(val);
2329 	nvlist_free(propval);
2330 }
2331 
2332 static char *
2333 get_receive_resume_token_impl(dsl_dataset_t *ds)
2334 {
2335 	if (!dsl_dataset_has_resume_receive_state(ds))
2336 		return (NULL);
2337 
2338 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2339 	char *str;
2340 	void *packed;
2341 	uint8_t *compressed;
2342 	uint64_t val;
2343 	nvlist_t *token_nv = fnvlist_alloc();
2344 	size_t packed_size, compressed_size;
2345 
2346 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2347 	    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
2348 		fnvlist_add_uint64(token_nv, "fromguid", val);
2349 	}
2350 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2351 	    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
2352 		fnvlist_add_uint64(token_nv, "object", val);
2353 	}
2354 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2355 	    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
2356 		fnvlist_add_uint64(token_nv, "offset", val);
2357 	}
2358 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2359 	    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
2360 		fnvlist_add_uint64(token_nv, "bytes", val);
2361 	}
2362 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2363 	    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
2364 		fnvlist_add_uint64(token_nv, "toguid", val);
2365 	}
2366 	char buf[MAXNAMELEN];
2367 	if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2368 	    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
2369 		fnvlist_add_string(token_nv, "toname", buf);
2370 	}
2371 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2372 	    DS_FIELD_RESUME_LARGEBLOCK) == 0) {
2373 		fnvlist_add_boolean(token_nv, "largeblockok");
2374 	}
2375 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2376 	    DS_FIELD_RESUME_EMBEDOK) == 0) {
2377 		fnvlist_add_boolean(token_nv, "embedok");
2378 	}
2379 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2380 	    DS_FIELD_RESUME_COMPRESSOK) == 0) {
2381 		fnvlist_add_boolean(token_nv, "compressok");
2382 	}
2383 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2384 	    DS_FIELD_RESUME_RAWOK) == 0) {
2385 		fnvlist_add_boolean(token_nv, "rawok");
2386 	}
2387 	if (dsl_dataset_feature_is_active(ds,
2388 	    SPA_FEATURE_REDACTED_DATASETS)) {
2389 		uint64_t num_redact_snaps = 0;
2390 		uint64_t *redact_snaps = NULL;
2391 		VERIFY3B(dsl_dataset_get_uint64_array_feature(ds,
2392 		    SPA_FEATURE_REDACTED_DATASETS, &num_redact_snaps,
2393 		    &redact_snaps), ==, B_TRUE);
2394 		fnvlist_add_uint64_array(token_nv, "redact_snaps",
2395 		    redact_snaps, num_redact_snaps);
2396 	}
2397 	if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2398 	    DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS) == 0) {
2399 		uint64_t num_redact_snaps = 0, int_size = 0;
2400 		uint64_t *redact_snaps = NULL;
2401 		VERIFY0(zap_length(dp->dp_meta_objset, ds->ds_object,
2402 		    DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, &int_size,
2403 		    &num_redact_snaps));
2404 		ASSERT3U(int_size, ==, sizeof (uint64_t));
2405 
2406 		redact_snaps = kmem_alloc(int_size * num_redact_snaps,
2407 		    KM_SLEEP);
2408 		VERIFY0(zap_lookup(dp->dp_meta_objset, ds->ds_object,
2409 		    DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, int_size,
2410 		    num_redact_snaps, redact_snaps));
2411 		fnvlist_add_uint64_array(token_nv, "book_redact_snaps",
2412 		    redact_snaps, num_redact_snaps);
2413 		kmem_free(redact_snaps, int_size * num_redact_snaps);
2414 	}
2415 	packed = fnvlist_pack(token_nv, &packed_size);
2416 	fnvlist_free(token_nv);
2417 	compressed = kmem_alloc(packed_size, KM_SLEEP);
2418 
2419 	compressed_size = gzip_compress(packed, compressed,
2420 	    packed_size, packed_size, 6);
2421 
2422 	zio_cksum_t cksum;
2423 	fletcher_4_native_varsize(compressed, compressed_size, &cksum);
2424 
2425 	size_t alloc_size = compressed_size * 2 + 1;
2426 	str = kmem_alloc(alloc_size, KM_SLEEP);
2427 	for (int i = 0; i < compressed_size; i++) {
2428 		size_t offset = i * 2;
2429 		(void) snprintf(str + offset, alloc_size - offset,
2430 	    "%02x", compressed[i]);
2431 	}
2432 	str[compressed_size * 2] = '\0';
2433 	char *propval = kmem_asprintf("%u-%llx-%llx-%s",
2434 	    ZFS_SEND_RESUME_TOKEN_VERSION,
2435 	    (longlong_t)cksum.zc_word[0],
2436 	    (longlong_t)packed_size, str);
2437 	kmem_free(packed, packed_size);
2438 	kmem_free(str, alloc_size);
2439 	kmem_free(compressed, packed_size);
2440 	return (propval);
2441 }
2442 
2443 /*
2444  * Returns a string that represents the receive resume state token. It should
2445  * be freed with strfree(). NULL is returned if no resume state is present.
2446  */
2447 char *
2448 get_receive_resume_token(dsl_dataset_t *ds)
2449 {
2450 	/*
2451 	 * A failed "newfs" (e.g. full) resumable receive leaves
2452 	 * the stats set on this dataset.  Check here for the prop.
2453 	 */
2454 	char *token = get_receive_resume_token_impl(ds);
2455 	if (token != NULL)
2456 		return (token);
2457 	/*
2458 	 * A failed incremental resumable receive leaves the
2459 	 * stats set on our child named "%recv".  Check the child
2460 	 * for the prop.
2461 	 */
2462 	/* 6 extra bytes for /%recv */
2463 	char name[ZFS_MAX_DATASET_NAME_LEN + 6];
2464 	dsl_dataset_t *recv_ds;
2465 	dsl_dataset_name(ds, name);
2466 	if (strlcat(name, "/", sizeof (name)) < sizeof (name) &&
2467 	    strlcat(name, recv_clone_name, sizeof (name)) < sizeof (name) &&
2468 	    dsl_dataset_hold(ds->ds_dir->dd_pool, name, FTAG, &recv_ds) == 0) {
2469 		token = get_receive_resume_token_impl(recv_ds);
2470 		dsl_dataset_rele(recv_ds, FTAG);
2471 	}
2472 	return (token);
2473 }
2474 
2475 uint64_t
2476 dsl_get_refratio(dsl_dataset_t *ds)
2477 {
2478 	uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
2479 	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
2480 	    dsl_dataset_phys(ds)->ds_compressed_bytes);
2481 	return (ratio);
2482 }
2483 
2484 uint64_t
2485 dsl_get_logicalreferenced(dsl_dataset_t *ds)
2486 {
2487 	return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
2488 }
2489 
2490 uint64_t
2491 dsl_get_compressratio(dsl_dataset_t *ds)
2492 {
2493 	if (ds->ds_is_snapshot) {
2494 		return (dsl_get_refratio(ds));
2495 	} else {
2496 		dsl_dir_t *dd = ds->ds_dir;
2497 		mutex_enter(&dd->dd_lock);
2498 		uint64_t val = dsl_dir_get_compressratio(dd);
2499 		mutex_exit(&dd->dd_lock);
2500 		return (val);
2501 	}
2502 }
2503 
2504 uint64_t
2505 dsl_get_used(dsl_dataset_t *ds)
2506 {
2507 	if (ds->ds_is_snapshot) {
2508 		return (dsl_dataset_phys(ds)->ds_unique_bytes);
2509 	} else {
2510 		dsl_dir_t *dd = ds->ds_dir;
2511 		mutex_enter(&dd->dd_lock);
2512 		uint64_t val = dsl_dir_get_used(dd);
2513 		mutex_exit(&dd->dd_lock);
2514 		return (val);
2515 	}
2516 }
2517 
2518 uint64_t
2519 dsl_get_creation(dsl_dataset_t *ds)
2520 {
2521 	return (dsl_dataset_phys(ds)->ds_creation_time);
2522 }
2523 
2524 uint64_t
2525 dsl_get_creationtxg(dsl_dataset_t *ds)
2526 {
2527 	return (dsl_dataset_phys(ds)->ds_creation_txg);
2528 }
2529 
2530 uint64_t
2531 dsl_get_refquota(dsl_dataset_t *ds)
2532 {
2533 	return (ds->ds_quota);
2534 }
2535 
2536 uint64_t
2537 dsl_get_refreservation(dsl_dataset_t *ds)
2538 {
2539 	return (ds->ds_reserved);
2540 }
2541 
2542 uint64_t
2543 dsl_get_guid(dsl_dataset_t *ds)
2544 {
2545 	return (dsl_dataset_phys(ds)->ds_guid);
2546 }
2547 
2548 uint64_t
2549 dsl_get_unique(dsl_dataset_t *ds)
2550 {
2551 	return (dsl_dataset_phys(ds)->ds_unique_bytes);
2552 }
2553 
2554 uint64_t
2555 dsl_get_objsetid(dsl_dataset_t *ds)
2556 {
2557 	return (ds->ds_object);
2558 }
2559 
2560 uint64_t
2561 dsl_get_userrefs(dsl_dataset_t *ds)
2562 {
2563 	return (ds->ds_userrefs);
2564 }
2565 
2566 uint64_t
2567 dsl_get_defer_destroy(dsl_dataset_t *ds)
2568 {
2569 	return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2570 }
2571 
2572 uint64_t
2573 dsl_get_referenced(dsl_dataset_t *ds)
2574 {
2575 	return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2576 }
2577 
2578 uint64_t
2579 dsl_get_numclones(dsl_dataset_t *ds)
2580 {
2581 	ASSERT(ds->ds_is_snapshot);
2582 	return (dsl_dataset_phys(ds)->ds_num_children - 1);
2583 }
2584 
2585 uint64_t
2586 dsl_get_inconsistent(dsl_dataset_t *ds)
2587 {
2588 	return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2589 	    1 : 0);
2590 }
2591 
2592 uint64_t
2593 dsl_get_redacted(dsl_dataset_t *ds)
2594 {
2595 	return (dsl_dataset_feature_is_active(ds,
2596 	    SPA_FEATURE_REDACTED_DATASETS));
2597 }
2598 
2599 uint64_t
2600 dsl_get_available(dsl_dataset_t *ds)
2601 {
2602 	uint64_t refdbytes = dsl_get_referenced(ds);
2603 	uint64_t availbytes = dsl_dir_space_available(ds->ds_dir,
2604 	    NULL, 0, TRUE);
2605 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2606 		availbytes +=
2607 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2608 	}
2609 	if (ds->ds_quota != 0) {
2610 		/*
2611 		 * Adjust available bytes according to refquota
2612 		 */
2613 		if (refdbytes < ds->ds_quota) {
2614 			availbytes = MIN(availbytes,
2615 			    ds->ds_quota - refdbytes);
2616 		} else {
2617 			availbytes = 0;
2618 		}
2619 	}
2620 	return (availbytes);
2621 }
2622 
2623 int
2624 dsl_get_written(dsl_dataset_t *ds, uint64_t *written)
2625 {
2626 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2627 	dsl_dataset_t *prev;
2628 	int err = dsl_dataset_hold_obj(dp,
2629 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2630 	if (err == 0) {
2631 		uint64_t comp, uncomp;
2632 		err = dsl_dataset_space_written(prev, ds, written,
2633 		    &comp, &uncomp);
2634 		dsl_dataset_rele(prev, FTAG);
2635 	}
2636 	return (err);
2637 }
2638 
2639 /*
2640  * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2641  */
2642 int
2643 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap)
2644 {
2645 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2646 	if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
2647 		dsl_dataset_name(ds->ds_prev, snap);
2648 		return (0);
2649 	} else {
2650 		return (SET_ERROR(ENOENT));
2651 	}
2652 }
2653 
2654 void
2655 dsl_get_redact_snaps(dsl_dataset_t *ds, nvlist_t *propval)
2656 {
2657 	uint64_t nsnaps;
2658 	uint64_t *snaps;
2659 	if (dsl_dataset_get_uint64_array_feature(ds,
2660 	    SPA_FEATURE_REDACTED_DATASETS, &nsnaps, &snaps)) {
2661 		fnvlist_add_uint64_array(propval, ZPROP_VALUE, snaps,
2662 		    nsnaps);
2663 	}
2664 }
2665 
2666 /*
2667  * Returns the mountpoint property and source for the given dataset in the value
2668  * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2669  * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2670  * Returns 0 on success and an error on failure.
2671  */
2672 int
2673 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value,
2674     char *source)
2675 {
2676 	int error;
2677 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2678 
2679 	/* Retrieve the mountpoint value stored in the zap object */
2680 	error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1,
2681 	    ZAP_MAXVALUELEN, value, source);
2682 	if (error != 0) {
2683 		return (error);
2684 	}
2685 
2686 	/*
2687 	 * Process the dsname and source to find the full mountpoint string.
2688 	 * Can be skipped for 'legacy' or 'none'.
2689 	 */
2690 	if (value[0] == '/') {
2691 		char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
2692 		char *root = buf;
2693 		const char *relpath;
2694 
2695 		/*
2696 		 * If we inherit the mountpoint, even from a dataset
2697 		 * with a received value, the source will be the path of
2698 		 * the dataset we inherit from. If source is
2699 		 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2700 		 * inherited.
2701 		 */
2702 		if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) {
2703 			relpath = "";
2704 		} else {
2705 			ASSERT0(strncmp(dsname, source, strlen(source)));
2706 			relpath = dsname + strlen(source);
2707 			if (relpath[0] == '/')
2708 				relpath++;
2709 		}
2710 
2711 		spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN);
2712 
2713 		/*
2714 		 * Special case an alternate root of '/'. This will
2715 		 * avoid having multiple leading slashes in the
2716 		 * mountpoint path.
2717 		 */
2718 		if (strcmp(root, "/") == 0)
2719 			root++;
2720 
2721 		/*
2722 		 * If the mountpoint is '/' then skip over this
2723 		 * if we are obtaining either an alternate root or
2724 		 * an inherited mountpoint.
2725 		 */
2726 		char *mnt = value;
2727 		if (value[1] == '\0' && (root[0] != '\0' ||
2728 		    relpath[0] != '\0'))
2729 			mnt = value + 1;
2730 
2731 		if (relpath[0] == '\0') {
2732 			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s",
2733 			    root, mnt);
2734 		} else {
2735 			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s",
2736 			    root, mnt, relpath[0] == '@' ? "" : "/",
2737 			    relpath);
2738 		}
2739 		kmem_free(buf, ZAP_MAXVALUELEN);
2740 	}
2741 
2742 	return (0);
2743 }
2744 
2745 void
2746 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2747 {
2748 	dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2749 
2750 	ASSERT(dsl_pool_config_held(dp));
2751 
2752 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO,
2753 	    dsl_get_refratio(ds));
2754 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
2755 	    dsl_get_logicalreferenced(ds));
2756 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2757 	    dsl_get_compressratio(ds));
2758 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2759 	    dsl_get_used(ds));
2760 
2761 	if (ds->ds_is_snapshot) {
2762 		get_clones_stat(ds, nv);
2763 	} else {
2764 		char buf[ZFS_MAX_DATASET_NAME_LEN];
2765 		if (dsl_get_prev_snap(ds, buf) == 0)
2766 			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP,
2767 			    buf);
2768 		dsl_dir_stats(ds->ds_dir, nv);
2769 	}
2770 
2771 	nvlist_t *propval = fnvlist_alloc();
2772 	dsl_get_redact_snaps(ds, propval);
2773 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS),
2774 	    propval);
2775 	nvlist_free(propval);
2776 
2777 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE,
2778 	    dsl_get_available(ds));
2779 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
2780 	    dsl_get_referenced(ds));
2781 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2782 	    dsl_get_creation(ds));
2783 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2784 	    dsl_get_creationtxg(ds));
2785 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2786 	    dsl_get_refquota(ds));
2787 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2788 	    dsl_get_refreservation(ds));
2789 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2790 	    dsl_get_guid(ds));
2791 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2792 	    dsl_get_unique(ds));
2793 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2794 	    dsl_get_objsetid(ds));
2795 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2796 	    dsl_get_userrefs(ds));
2797 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2798 	    dsl_get_defer_destroy(ds));
2799 	dsl_dataset_crypt_stats(ds, nv);
2800 
2801 	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
2802 		uint64_t written;
2803 		if (dsl_get_written(ds, &written) == 0) {
2804 			dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2805 			    written);
2806 		}
2807 	}
2808 
2809 	if (!dsl_dataset_is_snapshot(ds)) {
2810 		char *token = get_receive_resume_token(ds);
2811 		if (token != NULL) {
2812 			dsl_prop_nvlist_add_string(nv,
2813 			    ZFS_PROP_RECEIVE_RESUME_TOKEN, token);
2814 			kmem_strfree(token);
2815 		}
2816 	}
2817 }
2818 
2819 void
2820 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2821 {
2822 	dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2823 	ASSERT(dsl_pool_config_held(dp));
2824 
2825 	stat->dds_creation_txg = dsl_get_creationtxg(ds);
2826 	stat->dds_inconsistent = dsl_get_inconsistent(ds);
2827 	stat->dds_guid = dsl_get_guid(ds);
2828 	stat->dds_redacted = dsl_get_redacted(ds);
2829 	stat->dds_origin[0] = '\0';
2830 	if (ds->ds_is_snapshot) {
2831 		stat->dds_is_snapshot = B_TRUE;
2832 		stat->dds_num_clones = dsl_get_numclones(ds);
2833 	} else {
2834 		stat->dds_is_snapshot = B_FALSE;
2835 		stat->dds_num_clones = 0;
2836 
2837 		if (dsl_dir_is_clone(ds->ds_dir)) {
2838 			dsl_dir_get_origin(ds->ds_dir, stat->dds_origin);
2839 		}
2840 	}
2841 }
2842 
2843 uint64_t
2844 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2845 {
2846 	return (ds->ds_fsid_guid);
2847 }
2848 
2849 void
2850 dsl_dataset_space(dsl_dataset_t *ds,
2851     uint64_t *refdbytesp, uint64_t *availbytesp,
2852     uint64_t *usedobjsp, uint64_t *availobjsp)
2853 {
2854 	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2855 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2856 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2857 		*availbytesp +=
2858 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2859 	if (ds->ds_quota != 0) {
2860 		/*
2861 		 * Adjust available bytes according to refquota
2862 		 */
2863 		if (*refdbytesp < ds->ds_quota)
2864 			*availbytesp = MIN(*availbytesp,
2865 			    ds->ds_quota - *refdbytesp);
2866 		else
2867 			*availbytesp = 0;
2868 	}
2869 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2870 	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2871 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2872 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2873 }
2874 
2875 boolean_t
2876 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2877 {
2878 	dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2879 	uint64_t birth;
2880 
2881 	ASSERT(dsl_pool_config_held(dp));
2882 	if (snap == NULL)
2883 		return (B_FALSE);
2884 	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2885 	birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2886 	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2887 	if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2888 		objset_t *os, *os_snap;
2889 		/*
2890 		 * It may be that only the ZIL differs, because it was
2891 		 * reset in the head.  Don't count that as being
2892 		 * modified.
2893 		 */
2894 		if (dmu_objset_from_ds(ds, &os) != 0)
2895 			return (B_TRUE);
2896 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
2897 			return (B_TRUE);
2898 		return (memcmp(&os->os_phys->os_meta_dnode,
2899 		    &os_snap->os_phys->os_meta_dnode,
2900 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
2901 	}
2902 	return (B_FALSE);
2903 }
2904 
2905 typedef struct dsl_dataset_rename_snapshot_arg {
2906 	const char *ddrsa_fsname;
2907 	const char *ddrsa_oldsnapname;
2908 	const char *ddrsa_newsnapname;
2909 	boolean_t ddrsa_recursive;
2910 	dmu_tx_t *ddrsa_tx;
2911 } dsl_dataset_rename_snapshot_arg_t;
2912 
2913 static int
2914 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2915     dsl_dataset_t *hds, void *arg)
2916 {
2917 	(void) dp;
2918 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2919 	int error;
2920 	uint64_t val;
2921 
2922 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2923 	if (error != 0) {
2924 		/* ignore nonexistent snapshots */
2925 		return (error == ENOENT ? 0 : error);
2926 	}
2927 
2928 	/* new name should not exist */
2929 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2930 	if (error == 0)
2931 		error = SET_ERROR(EEXIST);
2932 	else if (error == ENOENT)
2933 		error = 0;
2934 
2935 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2936 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
2937 	    strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2938 		error = SET_ERROR(ENAMETOOLONG);
2939 
2940 	return (error);
2941 }
2942 
2943 static int
2944 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2945 {
2946 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2947 	dsl_pool_t *dp = dmu_tx_pool(tx);
2948 	dsl_dataset_t *hds;
2949 	int error;
2950 
2951 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2952 	if (error != 0)
2953 		return (error);
2954 
2955 	if (ddrsa->ddrsa_recursive) {
2956 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2957 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2958 		    DS_FIND_CHILDREN);
2959 	} else {
2960 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2961 	}
2962 	dsl_dataset_rele(hds, FTAG);
2963 	return (error);
2964 }
2965 
2966 static int
2967 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2968     dsl_dataset_t *hds, void *arg)
2969 {
2970 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2971 	dsl_dataset_t *ds;
2972 	uint64_t val;
2973 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2974 	int error;
2975 
2976 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2977 	ASSERT(error == 0 || error == ENOENT);
2978 	if (error == ENOENT) {
2979 		/* ignore nonexistent snapshots */
2980 		return (0);
2981 	}
2982 
2983 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2984 
2985 	/* log before we change the name */
2986 	spa_history_log_internal_ds(ds, "rename", tx,
2987 	    "-> @%s", ddrsa->ddrsa_newsnapname);
2988 
2989 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2990 	    B_FALSE));
2991 	mutex_enter(&ds->ds_lock);
2992 	(void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname,
2993 	    sizeof (ds->ds_snapname));
2994 	mutex_exit(&ds->ds_lock);
2995 	VERIFY0(zap_add(dp->dp_meta_objset,
2996 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2997 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2998 	zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname,
2999 	    ddrsa->ddrsa_newsnapname, B_TRUE);
3000 
3001 	dsl_dataset_rele(ds, FTAG);
3002 	return (0);
3003 }
3004 
3005 static void
3006 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
3007 {
3008 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
3009 	dsl_pool_t *dp = dmu_tx_pool(tx);
3010 	dsl_dataset_t *hds = NULL;
3011 
3012 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
3013 	ddrsa->ddrsa_tx = tx;
3014 	if (ddrsa->ddrsa_recursive) {
3015 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
3016 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
3017 		    DS_FIND_CHILDREN));
3018 	} else {
3019 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
3020 	}
3021 	dsl_dataset_rele(hds, FTAG);
3022 }
3023 
3024 int
3025 dsl_dataset_rename_snapshot(const char *fsname,
3026     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
3027 {
3028 	dsl_dataset_rename_snapshot_arg_t ddrsa;
3029 
3030 	ddrsa.ddrsa_fsname = fsname;
3031 	ddrsa.ddrsa_oldsnapname = oldsnapname;
3032 	ddrsa.ddrsa_newsnapname = newsnapname;
3033 	ddrsa.ddrsa_recursive = recursive;
3034 
3035 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
3036 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
3037 	    1, ZFS_SPACE_CHECK_RESERVED));
3038 }
3039 
3040 /*
3041  * If we're doing an ownership handoff, we need to make sure that there is
3042  * only one long hold on the dataset.  We're not allowed to change anything here
3043  * so we don't permanently release the long hold or regular hold here.  We want
3044  * to do this only when syncing to avoid the dataset unexpectedly going away
3045  * when we release the long hold.
3046  */
3047 static int
3048 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
3049 {
3050 	boolean_t held = B_FALSE;
3051 
3052 	if (!dmu_tx_is_syncing(tx))
3053 		return (0);
3054 
3055 	dsl_dir_t *dd = ds->ds_dir;
3056 	mutex_enter(&dd->dd_activity_lock);
3057 	uint64_t holds = zfs_refcount_count(&ds->ds_longholds) -
3058 	    (owner != NULL ? 1 : 0);
3059 	/*
3060 	 * The value of dd_activity_waiters can chance as soon as we drop the
3061 	 * lock, but we're fine with that; new waiters coming in or old
3062 	 * waiters leaving doesn't cause problems, since we're going to cancel
3063 	 * waiters later anyway. The goal of this check is to verify that no
3064 	 * non-waiters have long-holds, and all new long-holds will be
3065 	 * prevented because we're holding the pool config as writer.
3066 	 */
3067 	if (holds != dd->dd_activity_waiters)
3068 		held = B_TRUE;
3069 	mutex_exit(&dd->dd_activity_lock);
3070 
3071 	if (held)
3072 		return (SET_ERROR(EBUSY));
3073 
3074 	return (0);
3075 }
3076 
3077 int
3078 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
3079 {
3080 	dsl_dataset_rollback_arg_t *ddra = arg;
3081 	dsl_pool_t *dp = dmu_tx_pool(tx);
3082 	dsl_dataset_t *ds;
3083 	int64_t unused_refres_delta;
3084 	int error;
3085 
3086 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
3087 	if (error != 0)
3088 		return (error);
3089 
3090 	/* must not be a snapshot */
3091 	if (ds->ds_is_snapshot) {
3092 		dsl_dataset_rele(ds, FTAG);
3093 		return (SET_ERROR(EINVAL));
3094 	}
3095 
3096 	/* must have a most recent snapshot */
3097 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
3098 		dsl_dataset_rele(ds, FTAG);
3099 		return (SET_ERROR(ESRCH));
3100 	}
3101 
3102 	/*
3103 	 * No rollback to a snapshot created in the current txg, because
3104 	 * the rollback may dirty the dataset and create blocks that are
3105 	 * not reachable from the rootbp while having a birth txg that
3106 	 * falls into the snapshot's range.
3107 	 */
3108 	if (dmu_tx_is_syncing(tx) &&
3109 	    dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
3110 		dsl_dataset_rele(ds, FTAG);
3111 		return (SET_ERROR(EAGAIN));
3112 	}
3113 
3114 	/*
3115 	 * If the expected target snapshot is specified, then check that
3116 	 * the latest snapshot is it.
3117 	 */
3118 	if (ddra->ddra_tosnap != NULL) {
3119 		dsl_dataset_t *snapds;
3120 
3121 		/* Check if the target snapshot exists at all. */
3122 		error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
3123 		if (error != 0) {
3124 			/*
3125 			 * ESRCH is used to signal that the target snapshot does
3126 			 * not exist, while ENOENT is used to report that
3127 			 * the rolled back dataset does not exist.
3128 			 * ESRCH is also used to cover other cases where the
3129 			 * target snapshot is not related to the dataset being
3130 			 * rolled back such as being in a different pool.
3131 			 */
3132 			if (error == ENOENT || error == EXDEV)
3133 				error = SET_ERROR(ESRCH);
3134 			dsl_dataset_rele(ds, FTAG);
3135 			return (error);
3136 		}
3137 		ASSERT(snapds->ds_is_snapshot);
3138 
3139 		/* Check if the snapshot is the latest snapshot indeed. */
3140 		if (snapds != ds->ds_prev) {
3141 			/*
3142 			 * Distinguish between the case where the only problem
3143 			 * is intervening snapshots (EEXIST) vs the snapshot
3144 			 * not being a valid target for rollback (ESRCH).
3145 			 */
3146 			if (snapds->ds_dir == ds->ds_dir ||
3147 			    (dsl_dir_is_clone(ds->ds_dir) &&
3148 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
3149 			    snapds->ds_object)) {
3150 				error = SET_ERROR(EEXIST);
3151 			} else {
3152 				error = SET_ERROR(ESRCH);
3153 			}
3154 			dsl_dataset_rele(snapds, FTAG);
3155 			dsl_dataset_rele(ds, FTAG);
3156 			return (error);
3157 		}
3158 		dsl_dataset_rele(snapds, FTAG);
3159 	}
3160 
3161 	/* must not have any bookmarks after the most recent snapshot */
3162 	if (dsl_bookmark_latest_txg(ds) >
3163 	    dsl_dataset_phys(ds)->ds_prev_snap_txg) {
3164 		dsl_dataset_rele(ds, FTAG);
3165 		return (SET_ERROR(EEXIST));
3166 	}
3167 
3168 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
3169 	if (error != 0) {
3170 		dsl_dataset_rele(ds, FTAG);
3171 		return (error);
3172 	}
3173 
3174 	/*
3175 	 * Check if the snap we are rolling back to uses more than
3176 	 * the refquota.
3177 	 */
3178 	if (ds->ds_quota != 0 &&
3179 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
3180 		dsl_dataset_rele(ds, FTAG);
3181 		return (SET_ERROR(EDQUOT));
3182 	}
3183 
3184 	/*
3185 	 * When we do the clone swap, we will temporarily use more space
3186 	 * due to the refreservation (the head will no longer have any
3187 	 * unique space, so the entire amount of the refreservation will need
3188 	 * to be free).  We will immediately destroy the clone, freeing
3189 	 * this space, but the freeing happens over many txg's.
3190 	 */
3191 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
3192 	    dsl_dataset_phys(ds)->ds_unique_bytes);
3193 
3194 	if (unused_refres_delta > 0 &&
3195 	    unused_refres_delta >
3196 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
3197 		dsl_dataset_rele(ds, FTAG);
3198 		return (SET_ERROR(ENOSPC));
3199 	}
3200 
3201 	dsl_dataset_rele(ds, FTAG);
3202 	return (0);
3203 }
3204 
3205 void
3206 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
3207 {
3208 	dsl_dataset_rollback_arg_t *ddra = arg;
3209 	dsl_pool_t *dp = dmu_tx_pool(tx);
3210 	dsl_dataset_t *ds, *clone;
3211 	uint64_t cloneobj;
3212 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
3213 
3214 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
3215 
3216 	dsl_dataset_name(ds->ds_prev, namebuf);
3217 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
3218 
3219 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
3220 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx);
3221 
3222 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
3223 
3224 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
3225 	dsl_dataset_zero_zil(ds, tx);
3226 
3227 	dsl_destroy_head_sync_impl(clone, tx);
3228 
3229 	dsl_dataset_rele(clone, FTAG);
3230 	dsl_dataset_rele(ds, FTAG);
3231 }
3232 
3233 /*
3234  * Rolls back the given filesystem or volume to the most recent snapshot.
3235  * The name of the most recent snapshot will be returned under key "target"
3236  * in the result nvlist.
3237  *
3238  * If owner != NULL:
3239  * - The existing dataset MUST be owned by the specified owner at entry
3240  * - Upon return, dataset will still be held by the same owner, whether we
3241  *   succeed or not.
3242  *
3243  * This mode is required any time the existing filesystem is mounted.  See
3244  * notes above zfs_suspend_fs() for further details.
3245  */
3246 int
3247 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
3248     nvlist_t *result)
3249 {
3250 	dsl_dataset_rollback_arg_t ddra;
3251 
3252 	ddra.ddra_fsname = fsname;
3253 	ddra.ddra_tosnap = tosnap;
3254 	ddra.ddra_owner = owner;
3255 	ddra.ddra_result = result;
3256 
3257 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
3258 	    dsl_dataset_rollback_sync, &ddra,
3259 	    1, ZFS_SPACE_CHECK_RESERVED));
3260 }
3261 
3262 struct promotenode {
3263 	list_node_t link;
3264 	dsl_dataset_t *ds;
3265 };
3266 
3267 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
3268 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
3269     void *tag);
3270 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
3271 
3272 int
3273 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
3274 {
3275 	dsl_dataset_promote_arg_t *ddpa = arg;
3276 	dsl_pool_t *dp = dmu_tx_pool(tx);
3277 	dsl_dataset_t *hds;
3278 	struct promotenode *snap;
3279 	dsl_dataset_t *origin_ds, *origin_head;
3280 	int err;
3281 	uint64_t unused;
3282 	uint64_t ss_mv_cnt;
3283 	size_t max_snap_len;
3284 	boolean_t conflicting_snaps;
3285 
3286 	err = promote_hold(ddpa, dp, FTAG);
3287 	if (err != 0)
3288 		return (err);
3289 
3290 	hds = ddpa->ddpa_clone;
3291 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
3292 
3293 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
3294 		promote_rele(ddpa, FTAG);
3295 		return (SET_ERROR(EXDEV));
3296 	}
3297 
3298 	snap = list_head(&ddpa->shared_snaps);
3299 	origin_head = snap->ds;
3300 	if (snap == NULL) {
3301 		err = SET_ERROR(ENOENT);
3302 		goto out;
3303 	}
3304 	origin_ds = snap->ds;
3305 
3306 	/*
3307 	 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir.
3308 	 * When doing a promote we must make sure the encryption root for
3309 	 * both the target and the target's origin does not change to avoid
3310 	 * needing to rewrap encryption keys
3311 	 */
3312 	err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir);
3313 	if (err != 0)
3314 		goto out;
3315 
3316 	/*
3317 	 * Compute and check the amount of space to transfer.  Since this is
3318 	 * so expensive, don't do the preliminary check.
3319 	 */
3320 	if (!dmu_tx_is_syncing(tx)) {
3321 		promote_rele(ddpa, FTAG);
3322 		return (0);
3323 	}
3324 
3325 	/* compute origin's new unique space */
3326 	snap = list_tail(&ddpa->clone_snaps);
3327 	ASSERT(snap != NULL);
3328 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3329 	    origin_ds->ds_object);
3330 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3331 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
3332 	    &ddpa->unique, &unused, &unused);
3333 
3334 	/*
3335 	 * Walk the snapshots that we are moving
3336 	 *
3337 	 * Compute space to transfer.  Consider the incremental changes
3338 	 * to used by each snapshot:
3339 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
3340 	 * So each snapshot gave birth to:
3341 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
3342 	 * So a sequence would look like:
3343 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
3344 	 * Which simplifies to:
3345 	 * uN + kN + kN-1 + ... + k1 + k0
3346 	 * Note however, if we stop before we reach the ORIGIN we get:
3347 	 * uN + kN + kN-1 + ... + kM - uM-1
3348 	 */
3349 	conflicting_snaps = B_FALSE;
3350 	ss_mv_cnt = 0;
3351 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
3352 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
3353 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
3354 	for (snap = list_head(&ddpa->shared_snaps); snap;
3355 	    snap = list_next(&ddpa->shared_snaps, snap)) {
3356 		uint64_t val, dlused, dlcomp, dluncomp;
3357 		dsl_dataset_t *ds = snap->ds;
3358 
3359 		ss_mv_cnt++;
3360 
3361 		/*
3362 		 * If there are long holds, we won't be able to evict
3363 		 * the objset.
3364 		 */
3365 		if (dsl_dataset_long_held(ds)) {
3366 			err = SET_ERROR(EBUSY);
3367 			goto out;
3368 		}
3369 
3370 		/* Check that the snapshot name does not conflict */
3371 		VERIFY0(dsl_dataset_get_snapname(ds));
3372 		if (strlen(ds->ds_snapname) >= max_snap_len) {
3373 			err = SET_ERROR(ENAMETOOLONG);
3374 			goto out;
3375 		}
3376 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
3377 		if (err == 0) {
3378 			fnvlist_add_boolean(ddpa->err_ds,
3379 			    snap->ds->ds_snapname);
3380 			conflicting_snaps = B_TRUE;
3381 		} else if (err != ENOENT) {
3382 			goto out;
3383 		}
3384 
3385 		/* The very first snapshot does not have a deadlist */
3386 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
3387 			continue;
3388 
3389 		dsl_deadlist_space(&ds->ds_deadlist,
3390 		    &dlused, &dlcomp, &dluncomp);
3391 		ddpa->used += dlused;
3392 		ddpa->comp += dlcomp;
3393 		ddpa->uncomp += dluncomp;
3394 	}
3395 
3396 	/*
3397 	 * Check that bookmarks that are being transferred don't have
3398 	 * name conflicts.
3399 	 */
3400 	for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks);
3401 	    dbn != NULL && dbn->dbn_phys.zbm_creation_txg <=
3402 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3403 	    dbn = AVL_NEXT(&origin_head->ds_bookmarks, dbn)) {
3404 		if (strlen(dbn->dbn_name) >= max_snap_len) {
3405 			err = SET_ERROR(ENAMETOOLONG);
3406 			goto out;
3407 		}
3408 		zfs_bookmark_phys_t bm;
3409 		err = dsl_bookmark_lookup_impl(ddpa->ddpa_clone,
3410 		    dbn->dbn_name, &bm);
3411 
3412 		if (err == 0) {
3413 			fnvlist_add_boolean(ddpa->err_ds, dbn->dbn_name);
3414 			conflicting_snaps = B_TRUE;
3415 		} else if (err == ESRCH) {
3416 			err = 0;
3417 		} else if (err != 0) {
3418 			goto out;
3419 		}
3420 	}
3421 
3422 	/*
3423 	 * In order to return the full list of conflicting snapshots, we check
3424 	 * whether there was a conflict after traversing all of them.
3425 	 */
3426 	if (conflicting_snaps) {
3427 		err = SET_ERROR(EEXIST);
3428 		goto out;
3429 	}
3430 
3431 	/*
3432 	 * If we are a clone of a clone then we never reached ORIGIN,
3433 	 * so we need to subtract out the clone origin's used space.
3434 	 */
3435 	if (ddpa->origin_origin) {
3436 		ddpa->used -=
3437 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
3438 		ddpa->comp -=
3439 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
3440 		ddpa->uncomp -=
3441 		    dsl_dataset_phys(ddpa->origin_origin)->
3442 		    ds_uncompressed_bytes;
3443 	}
3444 
3445 	/* Check that there is enough space and limit headroom here */
3446 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
3447 	    0, ss_mv_cnt, ddpa->used, ddpa->cr, ddpa->proc);
3448 	if (err != 0)
3449 		goto out;
3450 
3451 	/*
3452 	 * Compute the amounts of space that will be used by snapshots
3453 	 * after the promotion (for both origin and clone).  For each,
3454 	 * it is the amount of space that will be on all of their
3455 	 * deadlists (that was not born before their new origin).
3456 	 */
3457 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
3458 		uint64_t space;
3459 
3460 		/*
3461 		 * Note, typically this will not be a clone of a clone,
3462 		 * so dd_origin_txg will be < TXG_INITIAL, so
3463 		 * these snaplist_space() -> dsl_deadlist_space_range()
3464 		 * calls will be fast because they do not have to
3465 		 * iterate over all bps.
3466 		 */
3467 		snap = list_head(&ddpa->origin_snaps);
3468 		if (snap == NULL) {
3469 			err = SET_ERROR(ENOENT);
3470 			goto out;
3471 		}
3472 		err = snaplist_space(&ddpa->shared_snaps,
3473 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
3474 		if (err != 0)
3475 			goto out;
3476 
3477 		err = snaplist_space(&ddpa->clone_snaps,
3478 		    snap->ds->ds_dir->dd_origin_txg, &space);
3479 		if (err != 0)
3480 			goto out;
3481 		ddpa->cloneusedsnap += space;
3482 	}
3483 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
3484 	    DD_FLAG_USED_BREAKDOWN) {
3485 		err = snaplist_space(&ddpa->origin_snaps,
3486 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
3487 		    &ddpa->originusedsnap);
3488 		if (err != 0)
3489 			goto out;
3490 	}
3491 
3492 out:
3493 	promote_rele(ddpa, FTAG);
3494 	return (err);
3495 }
3496 
3497 void
3498 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
3499 {
3500 	dsl_dataset_promote_arg_t *ddpa = arg;
3501 	dsl_pool_t *dp = dmu_tx_pool(tx);
3502 	dsl_dataset_t *hds;
3503 	struct promotenode *snap;
3504 	dsl_dataset_t *origin_ds;
3505 	dsl_dataset_t *origin_head;
3506 	dsl_dir_t *dd;
3507 	dsl_dir_t *odd = NULL;
3508 	uint64_t oldnext_obj;
3509 	int64_t delta;
3510 
3511 	ASSERT(nvlist_empty(ddpa->err_ds));
3512 
3513 	VERIFY0(promote_hold(ddpa, dp, FTAG));
3514 	hds = ddpa->ddpa_clone;
3515 
3516 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
3517 
3518 	snap = list_head(&ddpa->shared_snaps);
3519 	origin_ds = snap->ds;
3520 	dd = hds->ds_dir;
3521 
3522 	snap = list_head(&ddpa->origin_snaps);
3523 	origin_head = snap->ds;
3524 
3525 	/*
3526 	 * We need to explicitly open odd, since origin_ds's dd will be
3527 	 * changing.
3528 	 */
3529 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
3530 	    NULL, FTAG, &odd));
3531 
3532 	dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx);
3533 
3534 	/* change origin's next snap */
3535 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
3536 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
3537 	snap = list_tail(&ddpa->clone_snaps);
3538 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3539 	    origin_ds->ds_object);
3540 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
3541 
3542 	/* change the origin's next clone */
3543 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
3544 		dsl_dataset_remove_from_next_clones(origin_ds,
3545 		    snap->ds->ds_object, tx);
3546 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3547 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
3548 		    oldnext_obj, tx));
3549 	}
3550 
3551 	/* change origin */
3552 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
3553 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
3554 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
3555 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
3556 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
3557 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
3558 	origin_head->ds_dir->dd_origin_txg =
3559 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3560 
3561 	/* change dd_clone entries */
3562 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3563 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3564 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
3565 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3566 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3567 		    hds->ds_object, tx));
3568 
3569 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3570 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3571 		    origin_head->ds_object, tx));
3572 		if (dsl_dir_phys(dd)->dd_clones == 0) {
3573 			dsl_dir_phys(dd)->dd_clones =
3574 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
3575 			    DMU_OT_NONE, 0, tx);
3576 		}
3577 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3578 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
3579 	}
3580 
3581 	/*
3582 	 * Move bookmarks to this dir.
3583 	 */
3584 	dsl_bookmark_node_t *dbn_next;
3585 	for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks);
3586 	    dbn != NULL && dbn->dbn_phys.zbm_creation_txg <=
3587 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3588 	    dbn = dbn_next) {
3589 		dbn_next = AVL_NEXT(&origin_head->ds_bookmarks, dbn);
3590 
3591 		avl_remove(&origin_head->ds_bookmarks, dbn);
3592 		VERIFY0(zap_remove(dp->dp_meta_objset,
3593 		    origin_head->ds_bookmarks_obj, dbn->dbn_name, tx));
3594 
3595 		dsl_bookmark_node_add(hds, dbn, tx);
3596 	}
3597 
3598 	dsl_bookmark_next_changed(hds, origin_ds, tx);
3599 
3600 	/* move snapshots to this dir */
3601 	for (snap = list_head(&ddpa->shared_snaps); snap;
3602 	    snap = list_next(&ddpa->shared_snaps, snap)) {
3603 		dsl_dataset_t *ds = snap->ds;
3604 
3605 		/*
3606 		 * Property callbacks are registered to a particular
3607 		 * dsl_dir.  Since ours is changing, evict the objset
3608 		 * so that they will be unregistered from the old dsl_dir.
3609 		 */
3610 		if (ds->ds_objset) {
3611 			dmu_objset_evict(ds->ds_objset);
3612 			ds->ds_objset = NULL;
3613 		}
3614 
3615 		/* move snap name entry */
3616 		VERIFY0(dsl_dataset_get_snapname(ds));
3617 		VERIFY0(dsl_dataset_snap_remove(origin_head,
3618 		    ds->ds_snapname, tx, B_TRUE));
3619 		VERIFY0(zap_add(dp->dp_meta_objset,
3620 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3621 		    8, 1, &ds->ds_object, tx));
3622 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3623 		    DD_FIELD_SNAPSHOT_COUNT, tx);
3624 
3625 		/* change containing dsl_dir */
3626 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3627 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3628 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3629 		ASSERT3P(ds->ds_dir, ==, odd);
3630 		dsl_dir_rele(ds->ds_dir, ds);
3631 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3632 		    NULL, ds, &ds->ds_dir));
3633 
3634 		/* move any clone references */
3635 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3636 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3637 			zap_cursor_t zc;
3638 			zap_attribute_t za;
3639 
3640 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
3641 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
3642 			    zap_cursor_retrieve(&zc, &za) == 0;
3643 			    zap_cursor_advance(&zc)) {
3644 				dsl_dataset_t *cnds;
3645 				uint64_t o;
3646 
3647 				if (za.za_first_integer == oldnext_obj) {
3648 					/*
3649 					 * We've already moved the
3650 					 * origin's reference.
3651 					 */
3652 					continue;
3653 				}
3654 
3655 				VERIFY0(dsl_dataset_hold_obj(dp,
3656 				    za.za_first_integer, FTAG, &cnds));
3657 				o = dsl_dir_phys(cnds->ds_dir)->
3658 				    dd_head_dataset_obj;
3659 
3660 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
3661 				    dsl_dir_phys(odd)->dd_clones, o, tx));
3662 				VERIFY0(zap_add_int(dp->dp_meta_objset,
3663 				    dsl_dir_phys(dd)->dd_clones, o, tx));
3664 				dsl_dataset_rele(cnds, FTAG);
3665 			}
3666 			zap_cursor_fini(&zc);
3667 		}
3668 
3669 		ASSERT(!dsl_prop_hascb(ds));
3670 	}
3671 
3672 	/*
3673 	 * Change space accounting.
3674 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3675 	 * both be valid, or both be 0 (resulting in delta == 0).  This
3676 	 * is true for each of {clone,origin} independently.
3677 	 */
3678 
3679 	delta = ddpa->cloneusedsnap -
3680 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3681 	ASSERT3S(delta, >=, 0);
3682 	ASSERT3U(ddpa->used, >=, delta);
3683 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3684 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
3685 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3686 
3687 	delta = ddpa->originusedsnap -
3688 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3689 	ASSERT3S(delta, <=, 0);
3690 	ASSERT3U(ddpa->used, >=, -delta);
3691 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3692 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
3693 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3694 
3695 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3696 
3697 	/*
3698 	 * Since livelists are specific to a clone's origin txg, they
3699 	 * are no longer accurate. Destroy the livelist from the clone being
3700 	 * promoted. If the origin dataset is a clone, destroy its livelist
3701 	 * as well.
3702 	 */
3703 	dsl_dir_remove_livelist(dd, tx, B_TRUE);
3704 	dsl_dir_remove_livelist(odd, tx, B_TRUE);
3705 
3706 	/* log history record */
3707 	spa_history_log_internal_ds(hds, "promote", tx, " ");
3708 
3709 	dsl_dir_rele(odd, FTAG);
3710 	promote_rele(ddpa, FTAG);
3711 }
3712 
3713 /*
3714  * Make a list of dsl_dataset_t's for the snapshots between first_obj
3715  * (exclusive) and last_obj (inclusive).  The list will be in reverse
3716  * order (last_obj will be the list_head()).  If first_obj == 0, do all
3717  * snapshots back to this dataset's origin.
3718  */
3719 static int
3720 snaplist_make(dsl_pool_t *dp,
3721     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
3722 {
3723 	uint64_t obj = last_obj;
3724 
3725 	list_create(l, sizeof (struct promotenode),
3726 	    offsetof(struct promotenode, link));
3727 
3728 	while (obj != first_obj) {
3729 		dsl_dataset_t *ds;
3730 		struct promotenode *snap;
3731 		int err;
3732 
3733 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3734 		ASSERT(err != ENOENT);
3735 		if (err != 0)
3736 			return (err);
3737 
3738 		if (first_obj == 0)
3739 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3740 
3741 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3742 		snap->ds = ds;
3743 		list_insert_tail(l, snap);
3744 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3745 	}
3746 
3747 	return (0);
3748 }
3749 
3750 static int
3751 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3752 {
3753 	struct promotenode *snap;
3754 
3755 	*spacep = 0;
3756 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3757 		uint64_t used, comp, uncomp;
3758 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3759 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
3760 		*spacep += used;
3761 	}
3762 	return (0);
3763 }
3764 
3765 static void
3766 snaplist_destroy(list_t *l, void *tag)
3767 {
3768 	struct promotenode *snap;
3769 
3770 	if (l == NULL || !list_link_active(&l->list_head))
3771 		return;
3772 
3773 	while ((snap = list_tail(l)) != NULL) {
3774 		list_remove(l, snap);
3775 		dsl_dataset_rele(snap->ds, tag);
3776 		kmem_free(snap, sizeof (*snap));
3777 	}
3778 	list_destroy(l);
3779 }
3780 
3781 static int
3782 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
3783 {
3784 	int error;
3785 	dsl_dir_t *dd;
3786 	struct promotenode *snap;
3787 
3788 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3789 	    &ddpa->ddpa_clone);
3790 	if (error != 0)
3791 		return (error);
3792 	dd = ddpa->ddpa_clone->ds_dir;
3793 
3794 	if (ddpa->ddpa_clone->ds_is_snapshot ||
3795 	    !dsl_dir_is_clone(dd)) {
3796 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
3797 		return (SET_ERROR(EINVAL));
3798 	}
3799 
3800 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3801 	    &ddpa->shared_snaps, tag);
3802 	if (error != 0)
3803 		goto out;
3804 
3805 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3806 	    &ddpa->clone_snaps, tag);
3807 	if (error != 0)
3808 		goto out;
3809 
3810 	snap = list_head(&ddpa->shared_snaps);
3811 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3812 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3813 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3814 	    &ddpa->origin_snaps, tag);
3815 	if (error != 0)
3816 		goto out;
3817 
3818 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3819 		error = dsl_dataset_hold_obj(dp,
3820 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3821 		    tag, &ddpa->origin_origin);
3822 		if (error != 0)
3823 			goto out;
3824 	}
3825 out:
3826 	if (error != 0)
3827 		promote_rele(ddpa, tag);
3828 	return (error);
3829 }
3830 
3831 static void
3832 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
3833 {
3834 	snaplist_destroy(&ddpa->shared_snaps, tag);
3835 	snaplist_destroy(&ddpa->clone_snaps, tag);
3836 	snaplist_destroy(&ddpa->origin_snaps, tag);
3837 	if (ddpa->origin_origin != NULL)
3838 		dsl_dataset_rele(ddpa->origin_origin, tag);
3839 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
3840 }
3841 
3842 /*
3843  * Promote a clone.
3844  *
3845  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3846  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3847  */
3848 int
3849 dsl_dataset_promote(const char *name, char *conflsnap)
3850 {
3851 	dsl_dataset_promote_arg_t ddpa = { 0 };
3852 	uint64_t numsnaps;
3853 	int error;
3854 	nvpair_t *snap_pair;
3855 	objset_t *os;
3856 
3857 	/*
3858 	 * We will modify space proportional to the number of
3859 	 * snapshots.  Compute numsnaps.
3860 	 */
3861 	error = dmu_objset_hold(name, FTAG, &os);
3862 	if (error != 0)
3863 		return (error);
3864 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3865 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3866 	    &numsnaps);
3867 	dmu_objset_rele(os, FTAG);
3868 	if (error != 0)
3869 		return (error);
3870 
3871 	ddpa.ddpa_clonename = name;
3872 	ddpa.err_ds = fnvlist_alloc();
3873 	ddpa.cr = CRED();
3874 	ddpa.proc = curproc;
3875 
3876 	error = dsl_sync_task(name, dsl_dataset_promote_check,
3877 	    dsl_dataset_promote_sync, &ddpa,
3878 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3879 
3880 	/*
3881 	 * Return the first conflicting snapshot found.
3882 	 */
3883 	snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3884 	if (snap_pair != NULL && conflsnap != NULL)
3885 		(void) strlcpy(conflsnap, nvpair_name(snap_pair),
3886 		    ZFS_MAX_DATASET_NAME_LEN);
3887 
3888 	fnvlist_free(ddpa.err_ds);
3889 	return (error);
3890 }
3891 
3892 int
3893 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3894     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3895 {
3896 	/*
3897 	 * "slack" factor for received datasets with refquota set on them.
3898 	 * See the bottom of this function for details on its use.
3899 	 */
3900 	uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS *
3901 	    spa_asize_inflation;
3902 	int64_t unused_refres_delta;
3903 
3904 	/* they should both be heads */
3905 	if (clone->ds_is_snapshot ||
3906 	    origin_head->ds_is_snapshot)
3907 		return (SET_ERROR(EINVAL));
3908 
3909 	/* if we are not forcing, the branch point should be just before them */
3910 	if (!force && clone->ds_prev != origin_head->ds_prev)
3911 		return (SET_ERROR(EINVAL));
3912 
3913 	/* clone should be the clone (unless they are unrelated) */
3914 	if (clone->ds_prev != NULL &&
3915 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3916 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
3917 		return (SET_ERROR(EINVAL));
3918 
3919 	/* the clone should be a child of the origin */
3920 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3921 		return (SET_ERROR(EINVAL));
3922 
3923 	/* origin_head shouldn't be modified unless 'force' */
3924 	if (!force &&
3925 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3926 		return (SET_ERROR(ETXTBSY));
3927 
3928 	/* origin_head should have no long holds (e.g. is not mounted) */
3929 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
3930 		return (SET_ERROR(EBUSY));
3931 
3932 	/* check amount of any unconsumed refreservation */
3933 	unused_refres_delta =
3934 	    (int64_t)MIN(origin_head->ds_reserved,
3935 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3936 	    (int64_t)MIN(origin_head->ds_reserved,
3937 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3938 
3939 	if (unused_refres_delta > 0 &&
3940 	    unused_refres_delta >
3941 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3942 		return (SET_ERROR(ENOSPC));
3943 
3944 	/*
3945 	 * The clone can't be too much over the head's refquota.
3946 	 *
3947 	 * To ensure that the entire refquota can be used, we allow one
3948 	 * transaction to exceed the refquota.  Therefore, this check
3949 	 * needs to also allow for the space referenced to be more than the
3950 	 * refquota.  The maximum amount of space that one transaction can use
3951 	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
3952 	 * overage ensures that we are able to receive a filesystem that
3953 	 * exceeds the refquota on the source system.
3954 	 *
3955 	 * So that overage is the refquota_slack we use below.
3956 	 */
3957 	if (origin_head->ds_quota != 0 &&
3958 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
3959 	    origin_head->ds_quota + refquota_slack)
3960 		return (SET_ERROR(EDQUOT));
3961 
3962 	return (0);
3963 }
3964 
3965 static void
3966 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3967     dsl_dataset_t *origin, dmu_tx_t *tx)
3968 {
3969 	uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3970 	dsl_pool_t *dp = dmu_tx_pool(tx);
3971 
3972 	ASSERT(dsl_pool_sync_context(dp));
3973 
3974 	clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3975 	origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3976 
3977 	if (clone_remap_dl_obj != 0) {
3978 		dsl_deadlist_close(&clone->ds_remap_deadlist);
3979 		dsl_dataset_unset_remap_deadlist_object(clone, tx);
3980 	}
3981 	if (origin_remap_dl_obj != 0) {
3982 		dsl_deadlist_close(&origin->ds_remap_deadlist);
3983 		dsl_dataset_unset_remap_deadlist_object(origin, tx);
3984 	}
3985 
3986 	if (clone_remap_dl_obj != 0) {
3987 		dsl_dataset_set_remap_deadlist_object(origin,
3988 		    clone_remap_dl_obj, tx);
3989 		dsl_deadlist_open(&origin->ds_remap_deadlist,
3990 		    dp->dp_meta_objset, clone_remap_dl_obj);
3991 	}
3992 	if (origin_remap_dl_obj != 0) {
3993 		dsl_dataset_set_remap_deadlist_object(clone,
3994 		    origin_remap_dl_obj, tx);
3995 		dsl_deadlist_open(&clone->ds_remap_deadlist,
3996 		    dp->dp_meta_objset, origin_remap_dl_obj);
3997 	}
3998 }
3999 
4000 void
4001 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
4002     dsl_dataset_t *origin_head, dmu_tx_t *tx)
4003 {
4004 	dsl_pool_t *dp = dmu_tx_pool(tx);
4005 	int64_t unused_refres_delta;
4006 
4007 	ASSERT(clone->ds_reserved == 0);
4008 	/*
4009 	 * NOTE: On DEBUG kernels there could be a race between this and
4010 	 * the check function if spa_asize_inflation is adjusted...
4011 	 */
4012 	ASSERT(origin_head->ds_quota == 0 ||
4013 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
4014 	    DMU_MAX_ACCESS * spa_asize_inflation);
4015 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
4016 
4017 	dsl_dir_cancel_waiters(origin_head->ds_dir);
4018 
4019 	/*
4020 	 * Swap per-dataset feature flags.
4021 	 */
4022 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
4023 		if (!(spa_feature_table[f].fi_flags &
4024 		    ZFEATURE_FLAG_PER_DATASET)) {
4025 			ASSERT(!dsl_dataset_feature_is_active(clone, f));
4026 			ASSERT(!dsl_dataset_feature_is_active(origin_head, f));
4027 			continue;
4028 		}
4029 
4030 		boolean_t clone_inuse = dsl_dataset_feature_is_active(clone, f);
4031 		void *clone_feature = clone->ds_feature[f];
4032 		boolean_t origin_head_inuse =
4033 		    dsl_dataset_feature_is_active(origin_head, f);
4034 		void *origin_head_feature = origin_head->ds_feature[f];
4035 
4036 		if (clone_inuse)
4037 			dsl_dataset_deactivate_feature_impl(clone, f, tx);
4038 		if (origin_head_inuse)
4039 			dsl_dataset_deactivate_feature_impl(origin_head, f, tx);
4040 
4041 		if (clone_inuse) {
4042 			dsl_dataset_activate_feature(origin_head->ds_object, f,
4043 			    clone_feature, tx);
4044 			origin_head->ds_feature[f] = clone_feature;
4045 		}
4046 		if (origin_head_inuse) {
4047 			dsl_dataset_activate_feature(clone->ds_object, f,
4048 			    origin_head_feature, tx);
4049 			clone->ds_feature[f] = origin_head_feature;
4050 		}
4051 	}
4052 
4053 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
4054 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
4055 
4056 	if (clone->ds_objset != NULL) {
4057 		dmu_objset_evict(clone->ds_objset);
4058 		clone->ds_objset = NULL;
4059 	}
4060 
4061 	if (origin_head->ds_objset != NULL) {
4062 		dmu_objset_evict(origin_head->ds_objset);
4063 		origin_head->ds_objset = NULL;
4064 	}
4065 
4066 	unused_refres_delta =
4067 	    (int64_t)MIN(origin_head->ds_reserved,
4068 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
4069 	    (int64_t)MIN(origin_head->ds_reserved,
4070 	    dsl_dataset_phys(clone)->ds_unique_bytes);
4071 
4072 	/*
4073 	 * Reset origin's unique bytes.
4074 	 */
4075 	{
4076 		dsl_dataset_t *origin = clone->ds_prev;
4077 		uint64_t comp, uncomp;
4078 
4079 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
4080 		dsl_deadlist_space_range(&clone->ds_deadlist,
4081 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
4082 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
4083 	}
4084 
4085 	/* swap blkptrs */
4086 	{
4087 		rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
4088 		rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
4089 		blkptr_t tmp;
4090 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
4091 		dsl_dataset_phys(origin_head)->ds_bp =
4092 		    dsl_dataset_phys(clone)->ds_bp;
4093 		dsl_dataset_phys(clone)->ds_bp = tmp;
4094 		rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
4095 		rrw_exit(&clone->ds_bp_rwlock, FTAG);
4096 	}
4097 
4098 	/* set dd_*_bytes */
4099 	{
4100 		int64_t dused, dcomp, duncomp;
4101 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
4102 		uint64_t odl_used, odl_comp, odl_uncomp;
4103 
4104 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
4105 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
4106 
4107 		dsl_deadlist_space(&clone->ds_deadlist,
4108 		    &cdl_used, &cdl_comp, &cdl_uncomp);
4109 		dsl_deadlist_space(&origin_head->ds_deadlist,
4110 		    &odl_used, &odl_comp, &odl_uncomp);
4111 
4112 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
4113 		    cdl_used -
4114 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
4115 		    odl_used);
4116 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
4117 		    cdl_comp -
4118 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
4119 		    odl_comp);
4120 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
4121 		    cdl_uncomp -
4122 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
4123 		    odl_uncomp);
4124 
4125 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
4126 		    dused, dcomp, duncomp, tx);
4127 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
4128 		    -dused, -dcomp, -duncomp, tx);
4129 
4130 		/*
4131 		 * The difference in the space used by snapshots is the
4132 		 * difference in snapshot space due to the head's
4133 		 * deadlist (since that's the only thing that's
4134 		 * changing that affects the snapused).
4135 		 */
4136 		dsl_deadlist_space_range(&clone->ds_deadlist,
4137 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
4138 		    &cdl_used, &cdl_comp, &cdl_uncomp);
4139 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
4140 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
4141 		    &odl_used, &odl_comp, &odl_uncomp);
4142 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
4143 		    DD_USED_HEAD, DD_USED_SNAP, tx);
4144 	}
4145 
4146 	/* swap ds_*_bytes */
4147 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
4148 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
4149 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
4150 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
4151 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
4152 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
4153 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
4154 	    dsl_dataset_phys(clone)->ds_unique_bytes);
4155 
4156 	/* apply any parent delta for change in unconsumed refreservation */
4157 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
4158 	    unused_refres_delta, 0, 0, tx);
4159 
4160 	/*
4161 	 * Swap deadlists.
4162 	 */
4163 	dsl_deadlist_close(&clone->ds_deadlist);
4164 	dsl_deadlist_close(&origin_head->ds_deadlist);
4165 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
4166 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
4167 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
4168 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
4169 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
4170 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
4171 	dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
4172 
4173 	/*
4174 	 * If there is a bookmark at the origin, its "next dataset" is
4175 	 * changing, so we need to reset its FBN.
4176 	 */
4177 	dsl_bookmark_next_changed(origin_head, origin_head->ds_prev, tx);
4178 
4179 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
4180 
4181 	/*
4182 	 * Destroy any livelists associated with the clone or the origin,
4183 	 * since after the swap the corresponding livelists are no longer
4184 	 * valid.
4185 	 */
4186 	dsl_dir_remove_livelist(clone->ds_dir, tx, B_TRUE);
4187 	dsl_dir_remove_livelist(origin_head->ds_dir, tx, B_TRUE);
4188 
4189 	spa_history_log_internal_ds(clone, "clone swap", tx,
4190 	    "parent=%s", origin_head->ds_dir->dd_myname);
4191 }
4192 
4193 /*
4194  * Given a pool name and a dataset object number in that pool,
4195  * return the name of that dataset.
4196  */
4197 int
4198 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
4199 {
4200 	dsl_pool_t *dp;
4201 	dsl_dataset_t *ds;
4202 	int error;
4203 
4204 	error = dsl_pool_hold(pname, FTAG, &dp);
4205 	if (error != 0)
4206 		return (error);
4207 
4208 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
4209 	if (error == 0) {
4210 		dsl_dataset_name(ds, buf);
4211 		dsl_dataset_rele(ds, FTAG);
4212 	}
4213 	dsl_pool_rele(dp, FTAG);
4214 
4215 	return (error);
4216 }
4217 
4218 int
4219 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
4220     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
4221 {
4222 	int error = 0;
4223 
4224 	ASSERT3S(asize, >, 0);
4225 
4226 	/*
4227 	 * *ref_rsrv is the portion of asize that will come from any
4228 	 * unconsumed refreservation space.
4229 	 */
4230 	*ref_rsrv = 0;
4231 
4232 	mutex_enter(&ds->ds_lock);
4233 	/*
4234 	 * Make a space adjustment for reserved bytes.
4235 	 */
4236 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
4237 		ASSERT3U(*used, >=,
4238 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
4239 		*used -=
4240 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
4241 		*ref_rsrv =
4242 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
4243 	}
4244 
4245 	if (!check_quota || ds->ds_quota == 0) {
4246 		mutex_exit(&ds->ds_lock);
4247 		return (0);
4248 	}
4249 	/*
4250 	 * If they are requesting more space, and our current estimate
4251 	 * is over quota, they get to try again unless the actual
4252 	 * on-disk is over quota and there are no pending changes (which
4253 	 * may free up space for us).
4254 	 */
4255 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
4256 	    ds->ds_quota) {
4257 		if (inflight > 0 ||
4258 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
4259 			error = SET_ERROR(ERESTART);
4260 		else
4261 			error = SET_ERROR(EDQUOT);
4262 	}
4263 	mutex_exit(&ds->ds_lock);
4264 
4265 	return (error);
4266 }
4267 
4268 typedef struct dsl_dataset_set_qr_arg {
4269 	const char *ddsqra_name;
4270 	zprop_source_t ddsqra_source;
4271 	uint64_t ddsqra_value;
4272 } dsl_dataset_set_qr_arg_t;
4273 
4274 
4275 static int
4276 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
4277 {
4278 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4279 	dsl_pool_t *dp = dmu_tx_pool(tx);
4280 	dsl_dataset_t *ds;
4281 	int error;
4282 	uint64_t newval;
4283 
4284 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
4285 		return (SET_ERROR(ENOTSUP));
4286 
4287 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
4288 	if (error != 0)
4289 		return (error);
4290 
4291 	if (ds->ds_is_snapshot) {
4292 		dsl_dataset_rele(ds, FTAG);
4293 		return (SET_ERROR(EINVAL));
4294 	}
4295 
4296 	error = dsl_prop_predict(ds->ds_dir,
4297 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
4298 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
4299 	if (error != 0) {
4300 		dsl_dataset_rele(ds, FTAG);
4301 		return (error);
4302 	}
4303 
4304 	if (newval == 0) {
4305 		dsl_dataset_rele(ds, FTAG);
4306 		return (0);
4307 	}
4308 
4309 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
4310 	    newval < ds->ds_reserved) {
4311 		dsl_dataset_rele(ds, FTAG);
4312 		return (SET_ERROR(ENOSPC));
4313 	}
4314 
4315 	dsl_dataset_rele(ds, FTAG);
4316 	return (0);
4317 }
4318 
4319 static void
4320 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
4321 {
4322 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4323 	dsl_pool_t *dp = dmu_tx_pool(tx);
4324 	dsl_dataset_t *ds = NULL;
4325 	uint64_t newval;
4326 
4327 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
4328 
4329 	dsl_prop_set_sync_impl(ds,
4330 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
4331 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
4332 	    &ddsqra->ddsqra_value, tx);
4333 
4334 	VERIFY0(dsl_prop_get_int_ds(ds,
4335 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
4336 
4337 	if (ds->ds_quota != newval) {
4338 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
4339 		ds->ds_quota = newval;
4340 	}
4341 	dsl_dataset_rele(ds, FTAG);
4342 }
4343 
4344 int
4345 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
4346     uint64_t refquota)
4347 {
4348 	dsl_dataset_set_qr_arg_t ddsqra;
4349 
4350 	ddsqra.ddsqra_name = dsname;
4351 	ddsqra.ddsqra_source = source;
4352 	ddsqra.ddsqra_value = refquota;
4353 
4354 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
4355 	    dsl_dataset_set_refquota_sync, &ddsqra, 0,
4356 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
4357 }
4358 
4359 static int
4360 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
4361 {
4362 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4363 	dsl_pool_t *dp = dmu_tx_pool(tx);
4364 	dsl_dataset_t *ds;
4365 	int error;
4366 	uint64_t newval, unique;
4367 
4368 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
4369 		return (SET_ERROR(ENOTSUP));
4370 
4371 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
4372 	if (error != 0)
4373 		return (error);
4374 
4375 	if (ds->ds_is_snapshot) {
4376 		dsl_dataset_rele(ds, FTAG);
4377 		return (SET_ERROR(EINVAL));
4378 	}
4379 
4380 	error = dsl_prop_predict(ds->ds_dir,
4381 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
4382 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
4383 	if (error != 0) {
4384 		dsl_dataset_rele(ds, FTAG);
4385 		return (error);
4386 	}
4387 
4388 	/*
4389 	 * If we are doing the preliminary check in open context, the
4390 	 * space estimates may be inaccurate.
4391 	 */
4392 	if (!dmu_tx_is_syncing(tx)) {
4393 		dsl_dataset_rele(ds, FTAG);
4394 		return (0);
4395 	}
4396 
4397 	mutex_enter(&ds->ds_lock);
4398 	if (!DS_UNIQUE_IS_ACCURATE(ds))
4399 		dsl_dataset_recalc_head_uniq(ds);
4400 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
4401 	mutex_exit(&ds->ds_lock);
4402 
4403 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
4404 		uint64_t delta = MAX(unique, newval) -
4405 		    MAX(unique, ds->ds_reserved);
4406 
4407 		if (delta >
4408 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
4409 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
4410 			dsl_dataset_rele(ds, FTAG);
4411 			return (SET_ERROR(ENOSPC));
4412 		}
4413 	}
4414 
4415 	dsl_dataset_rele(ds, FTAG);
4416 	return (0);
4417 }
4418 
4419 void
4420 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
4421     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
4422 {
4423 	uint64_t newval;
4424 	uint64_t unique;
4425 	int64_t delta;
4426 
4427 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
4428 	    source, sizeof (value), 1, &value, tx);
4429 
4430 	VERIFY0(dsl_prop_get_int_ds(ds,
4431 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
4432 
4433 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
4434 	mutex_enter(&ds->ds_dir->dd_lock);
4435 	mutex_enter(&ds->ds_lock);
4436 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
4437 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
4438 	delta = MAX(0, (int64_t)(newval - unique)) -
4439 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
4440 	ds->ds_reserved = newval;
4441 	mutex_exit(&ds->ds_lock);
4442 
4443 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
4444 	mutex_exit(&ds->ds_dir->dd_lock);
4445 }
4446 
4447 static void
4448 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
4449 {
4450 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
4451 	dsl_pool_t *dp = dmu_tx_pool(tx);
4452 	dsl_dataset_t *ds = NULL;
4453 
4454 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
4455 	dsl_dataset_set_refreservation_sync_impl(ds,
4456 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
4457 	dsl_dataset_rele(ds, FTAG);
4458 }
4459 
4460 int
4461 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
4462     uint64_t refreservation)
4463 {
4464 	dsl_dataset_set_qr_arg_t ddsqra;
4465 
4466 	ddsqra.ddsqra_name = dsname;
4467 	ddsqra.ddsqra_source = source;
4468 	ddsqra.ddsqra_value = refreservation;
4469 
4470 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
4471 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0,
4472 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
4473 }
4474 
4475 typedef struct dsl_dataset_set_compression_arg {
4476 	const char *ddsca_name;
4477 	zprop_source_t ddsca_source;
4478 	uint64_t ddsca_value;
4479 } dsl_dataset_set_compression_arg_t;
4480 
4481 static int
4482 dsl_dataset_set_compression_check(void *arg, dmu_tx_t *tx)
4483 {
4484 	dsl_dataset_set_compression_arg_t *ddsca = arg;
4485 	dsl_pool_t *dp = dmu_tx_pool(tx);
4486 
4487 	uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value);
4488 	spa_feature_t f = zio_compress_to_feature(compval);
4489 
4490 	if (f == SPA_FEATURE_NONE)
4491 		return (SET_ERROR(EINVAL));
4492 
4493 	if (!spa_feature_is_enabled(dp->dp_spa, f))
4494 		return (SET_ERROR(ENOTSUP));
4495 
4496 	return (0);
4497 }
4498 
4499 static void
4500 dsl_dataset_set_compression_sync(void *arg, dmu_tx_t *tx)
4501 {
4502 	dsl_dataset_set_compression_arg_t *ddsca = arg;
4503 	dsl_pool_t *dp = dmu_tx_pool(tx);
4504 	dsl_dataset_t *ds = NULL;
4505 
4506 	uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value);
4507 	spa_feature_t f = zio_compress_to_feature(compval);
4508 	ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN);
4509 
4510 	VERIFY0(dsl_dataset_hold(dp, ddsca->ddsca_name, FTAG, &ds));
4511 	if (zfeature_active(f, ds->ds_feature[f]) != B_TRUE) {
4512 		ds->ds_feature_activation[f] = (void *)B_TRUE;
4513 		dsl_dataset_activate_feature(ds->ds_object, f,
4514 		    ds->ds_feature_activation[f], tx);
4515 		ds->ds_feature[f] = ds->ds_feature_activation[f];
4516 	}
4517 	dsl_dataset_rele(ds, FTAG);
4518 }
4519 
4520 int
4521 dsl_dataset_set_compression(const char *dsname, zprop_source_t source,
4522     uint64_t compression)
4523 {
4524 	dsl_dataset_set_compression_arg_t ddsca;
4525 
4526 	/*
4527 	 * The sync task is only required for zstd in order to activate
4528 	 * the feature flag when the property is first set.
4529 	 */
4530 	if (ZIO_COMPRESS_ALGO(compression) != ZIO_COMPRESS_ZSTD)
4531 		return (0);
4532 
4533 	ddsca.ddsca_name = dsname;
4534 	ddsca.ddsca_source = source;
4535 	ddsca.ddsca_value = compression;
4536 
4537 	return (dsl_sync_task(dsname, dsl_dataset_set_compression_check,
4538 	    dsl_dataset_set_compression_sync, &ddsca, 0,
4539 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
4540 }
4541 
4542 /*
4543  * Return (in *usedp) the amount of space referenced by "new" that was not
4544  * referenced at the time the bookmark corresponds to.  "New" may be a
4545  * snapshot or a head.  The bookmark must be before new, in
4546  * new's filesystem (or its origin) -- caller verifies this.
4547  *
4548  * The written space is calculated by considering two components:  First, we
4549  * ignore any freed space, and calculate the written as new's used space
4550  * minus old's used space.  Next, we add in the amount of space that was freed
4551  * between the two time points, thus reducing new's used space relative to
4552  * old's. Specifically, this is the space that was born before
4553  * zbm_creation_txg, and freed before new (ie. on new's deadlist or a
4554  * previous deadlist).
4555  *
4556  * space freed                         [---------------------]
4557  * snapshots                       ---O-------O--------O-------O------
4558  *                                         bookmark           new
4559  *
4560  * Note, the bookmark's zbm_*_bytes_refd must be valid, but if the HAS_FBN
4561  * flag is not set, we will calculate the freed_before_next based on the
4562  * next snapshot's deadlist, rather than using zbm_*_freed_before_next_snap.
4563  */
4564 static int
4565 dsl_dataset_space_written_impl(zfs_bookmark_phys_t *bmp,
4566     dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4567 {
4568 	int err = 0;
4569 	dsl_pool_t *dp = new->ds_dir->dd_pool;
4570 
4571 	ASSERT(dsl_pool_config_held(dp));
4572 	if (dsl_dataset_is_snapshot(new)) {
4573 		ASSERT3U(bmp->zbm_creation_txg, <,
4574 		    dsl_dataset_phys(new)->ds_creation_txg);
4575 	}
4576 
4577 	*usedp = 0;
4578 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
4579 	*usedp -= bmp->zbm_referenced_bytes_refd;
4580 
4581 	*compp = 0;
4582 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
4583 	*compp -= bmp->zbm_compressed_bytes_refd;
4584 
4585 	*uncompp = 0;
4586 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
4587 	*uncompp -= bmp->zbm_uncompressed_bytes_refd;
4588 
4589 	dsl_dataset_t *snap = new;
4590 
4591 	while (dsl_dataset_phys(snap)->ds_prev_snap_txg >
4592 	    bmp->zbm_creation_txg) {
4593 		uint64_t used, comp, uncomp;
4594 
4595 		dsl_deadlist_space_range(&snap->ds_deadlist,
4596 		    0, bmp->zbm_creation_txg,
4597 		    &used, &comp, &uncomp);
4598 		*usedp += used;
4599 		*compp += comp;
4600 		*uncompp += uncomp;
4601 
4602 		uint64_t snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
4603 		if (snap != new)
4604 			dsl_dataset_rele(snap, FTAG);
4605 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4606 		if (err != 0)
4607 			break;
4608 	}
4609 
4610 	/*
4611 	 * We might not have the FBN if we are calculating written from
4612 	 * a snapshot (because we didn't know the correct "next" snapshot
4613 	 * until now).
4614 	 */
4615 	if (bmp->zbm_flags & ZBM_FLAG_HAS_FBN) {
4616 		*usedp += bmp->zbm_referenced_freed_before_next_snap;
4617 		*compp += bmp->zbm_compressed_freed_before_next_snap;
4618 		*uncompp += bmp->zbm_uncompressed_freed_before_next_snap;
4619 	} else {
4620 		ASSERT3U(dsl_dataset_phys(snap)->ds_prev_snap_txg, ==,
4621 		    bmp->zbm_creation_txg);
4622 		uint64_t used, comp, uncomp;
4623 		dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp);
4624 		*usedp += used;
4625 		*compp += comp;
4626 		*uncompp += uncomp;
4627 	}
4628 	if (snap != new)
4629 		dsl_dataset_rele(snap, FTAG);
4630 	return (err);
4631 }
4632 
4633 /*
4634  * Return (in *usedp) the amount of space written in new that was not
4635  * present at the time the bookmark corresponds to.  New may be a
4636  * snapshot or the head.  Old must be a bookmark before new, in
4637  * new's filesystem (or its origin) -- caller verifies this.
4638  */
4639 int
4640 dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t *bmp,
4641     dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4642 {
4643 	if (!(bmp->zbm_flags & ZBM_FLAG_HAS_FBN))
4644 		return (SET_ERROR(ENOTSUP));
4645 	return (dsl_dataset_space_written_impl(bmp, new,
4646 	    usedp, compp, uncompp));
4647 }
4648 
4649 /*
4650  * Return (in *usedp) the amount of space written in new that is not
4651  * present in oldsnap.  New may be a snapshot or the head.  Old must be
4652  * a snapshot before new, in new's filesystem (or its origin).  If not then
4653  * fail and return EINVAL.
4654  */
4655 int
4656 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4657     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4658 {
4659 	if (!dsl_dataset_is_before(new, oldsnap, 0))
4660 		return (SET_ERROR(EINVAL));
4661 
4662 	zfs_bookmark_phys_t zbm = { 0 };
4663 	dsl_dataset_phys_t *dsp = dsl_dataset_phys(oldsnap);
4664 	zbm.zbm_guid = dsp->ds_guid;
4665 	zbm.zbm_creation_txg = dsp->ds_creation_txg;
4666 	zbm.zbm_creation_time = dsp->ds_creation_time;
4667 	zbm.zbm_referenced_bytes_refd = dsp->ds_referenced_bytes;
4668 	zbm.zbm_compressed_bytes_refd = dsp->ds_compressed_bytes;
4669 	zbm.zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes;
4670 
4671 	/*
4672 	 * If oldsnap is the origin (or origin's origin, ...) of new,
4673 	 * we can't easily calculate the effective FBN.  Therefore,
4674 	 * we do not set ZBM_FLAG_HAS_FBN, so that the _impl will calculate
4675 	 * it relative to the correct "next": the next snapshot towards "new",
4676 	 * rather than the next snapshot in oldsnap's dsl_dir.
4677 	 */
4678 	return (dsl_dataset_space_written_impl(&zbm, new,
4679 	    usedp, compp, uncompp));
4680 }
4681 
4682 /*
4683  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4684  * lastsnap, and all snapshots in between are deleted.
4685  *
4686  * blocks that would be freed            [---------------------------]
4687  * snapshots                       ---O-------O--------O-------O--------O
4688  *                                        firstsnap        lastsnap
4689  *
4690  * This is the set of blocks that were born after the snap before firstsnap,
4691  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4692  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4693  * We calculate this by iterating over the relevant deadlists (from the snap
4694  * after lastsnap, backward to the snap after firstsnap), summing up the
4695  * space on the deadlist that was born after the snap before firstsnap.
4696  */
4697 int
4698 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4699     dsl_dataset_t *lastsnap,
4700     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4701 {
4702 	int err = 0;
4703 	uint64_t snapobj;
4704 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4705 
4706 	ASSERT(firstsnap->ds_is_snapshot);
4707 	ASSERT(lastsnap->ds_is_snapshot);
4708 
4709 	/*
4710 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
4711 	 * is before lastsnap.
4712 	 */
4713 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
4714 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
4715 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
4716 		return (SET_ERROR(EINVAL));
4717 
4718 	*usedp = *compp = *uncompp = 0;
4719 
4720 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
4721 	while (snapobj != firstsnap->ds_object) {
4722 		dsl_dataset_t *ds;
4723 		uint64_t used, comp, uncomp;
4724 
4725 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4726 		if (err != 0)
4727 			break;
4728 
4729 		dsl_deadlist_space_range(&ds->ds_deadlist,
4730 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
4731 		    &used, &comp, &uncomp);
4732 		*usedp += used;
4733 		*compp += comp;
4734 		*uncompp += uncomp;
4735 
4736 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4737 		ASSERT3U(snapobj, !=, 0);
4738 		dsl_dataset_rele(ds, FTAG);
4739 	}
4740 	return (err);
4741 }
4742 
4743 /*
4744  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
4745  * For example, they could both be snapshots of the same filesystem, and
4746  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
4747  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
4748  * filesystem.  Or 'earlier' could be the origin's origin.
4749  *
4750  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4751  */
4752 boolean_t
4753 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4754     uint64_t earlier_txg)
4755 {
4756 	dsl_pool_t *dp = later->ds_dir->dd_pool;
4757 	int error;
4758 	boolean_t ret;
4759 
4760 	ASSERT(dsl_pool_config_held(dp));
4761 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4762 
4763 	if (earlier_txg == 0)
4764 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4765 
4766 	if (later->ds_is_snapshot &&
4767 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4768 		return (B_FALSE);
4769 
4770 	if (later->ds_dir == earlier->ds_dir)
4771 		return (B_TRUE);
4772 
4773 	/*
4774 	 * We check dd_origin_obj explicitly here rather than using
4775 	 * dsl_dir_is_clone() so that we will return TRUE if "earlier"
4776 	 * is $ORIGIN@$ORIGIN.  dsl_dataset_space_written() depends on
4777 	 * this behavior.
4778 	 */
4779 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == 0)
4780 		return (B_FALSE);
4781 
4782 	dsl_dataset_t *origin;
4783 	error = dsl_dataset_hold_obj(dp,
4784 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4785 	if (error != 0)
4786 		return (B_FALSE);
4787 	if (dsl_dataset_phys(origin)->ds_creation_txg == earlier_txg &&
4788 	    origin->ds_dir == earlier->ds_dir) {
4789 		dsl_dataset_rele(origin, FTAG);
4790 		return (B_TRUE);
4791 	}
4792 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4793 	dsl_dataset_rele(origin, FTAG);
4794 	return (ret);
4795 }
4796 
4797 void
4798 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4799 {
4800 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4801 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4802 }
4803 
4804 boolean_t
4805 dsl_dataset_is_zapified(dsl_dataset_t *ds)
4806 {
4807 	dmu_object_info_t doi;
4808 
4809 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
4810 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4811 }
4812 
4813 boolean_t
4814 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4815 {
4816 	return (dsl_dataset_is_zapified(ds) &&
4817 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4818 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4819 }
4820 
4821 uint64_t
4822 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4823 {
4824 	uint64_t remap_deadlist_obj;
4825 	int err;
4826 
4827 	if (!dsl_dataset_is_zapified(ds))
4828 		return (0);
4829 
4830 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4831 	    DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4832 	    &remap_deadlist_obj);
4833 
4834 	if (err != 0) {
4835 		VERIFY3S(err, ==, ENOENT);
4836 		return (0);
4837 	}
4838 
4839 	ASSERT(remap_deadlist_obj != 0);
4840 	return (remap_deadlist_obj);
4841 }
4842 
4843 boolean_t
4844 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4845 {
4846 	EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4847 	    dsl_dataset_get_remap_deadlist_object(ds) != 0);
4848 	return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4849 }
4850 
4851 static void
4852 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4853     dmu_tx_t *tx)
4854 {
4855 	ASSERT(obj != 0);
4856 	dsl_dataset_zapify(ds, tx);
4857 	VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4858 	    DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4859 }
4860 
4861 static void
4862 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4863 {
4864 	VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4865 	    ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4866 }
4867 
4868 void
4869 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4870 {
4871 	uint64_t remap_deadlist_object;
4872 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4873 
4874 	ASSERT(dmu_tx_is_syncing(tx));
4875 	ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4876 
4877 	remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4878 	dsl_deadlist_close(&ds->ds_remap_deadlist);
4879 	dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4880 	dsl_dataset_unset_remap_deadlist_object(ds, tx);
4881 	spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4882 }
4883 
4884 void
4885 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4886 {
4887 	uint64_t remap_deadlist_obj;
4888 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4889 
4890 	ASSERT(dmu_tx_is_syncing(tx));
4891 	ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4892 	/*
4893 	 * Currently we only create remap deadlists when there are indirect
4894 	 * vdevs with referenced mappings.
4895 	 */
4896 	ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4897 
4898 	remap_deadlist_obj = dsl_deadlist_clone(
4899 	    &ds->ds_deadlist, UINT64_MAX,
4900 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4901 	dsl_dataset_set_remap_deadlist_object(ds,
4902 	    remap_deadlist_obj, tx);
4903 	dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4904 	    remap_deadlist_obj);
4905 	spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4906 }
4907 
4908 void
4909 dsl_dataset_activate_redaction(dsl_dataset_t *ds, uint64_t *redact_snaps,
4910     uint64_t num_redact_snaps, dmu_tx_t *tx)
4911 {
4912 	uint64_t dsobj = ds->ds_object;
4913 	struct feature_type_uint64_array_arg *ftuaa =
4914 	    kmem_zalloc(sizeof (*ftuaa), KM_SLEEP);
4915 	ftuaa->length = (int64_t)num_redact_snaps;
4916 	if (num_redact_snaps > 0) {
4917 		ftuaa->array = kmem_alloc(num_redact_snaps * sizeof (uint64_t),
4918 		    KM_SLEEP);
4919 		memcpy(ftuaa->array, redact_snaps, num_redact_snaps *
4920 		    sizeof (uint64_t));
4921 	}
4922 	dsl_dataset_activate_feature(dsobj, SPA_FEATURE_REDACTED_DATASETS,
4923 	    ftuaa, tx);
4924 	ds->ds_feature[SPA_FEATURE_REDACTED_DATASETS] = ftuaa;
4925 }
4926 
4927 #if defined(_LP64)
4928 #define	RECORDSIZE_PERM ZMOD_RW
4929 #else
4930 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
4931 #define	RECORDSIZE_PERM ZMOD_RD
4932 #endif
4933 ZFS_MODULE_PARAM(zfs, zfs_, max_recordsize, INT, RECORDSIZE_PERM,
4934 	"Max allowed record size");
4935 
4936 ZFS_MODULE_PARAM(zfs, zfs_, allow_redacted_dataset_mount, INT, ZMOD_RW,
4937 	"Allow mounting of redacted datasets");
4938 
4939 EXPORT_SYMBOL(dsl_dataset_hold);
4940 EXPORT_SYMBOL(dsl_dataset_hold_flags);
4941 EXPORT_SYMBOL(dsl_dataset_hold_obj);
4942 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags);
4943 EXPORT_SYMBOL(dsl_dataset_own);
4944 EXPORT_SYMBOL(dsl_dataset_own_obj);
4945 EXPORT_SYMBOL(dsl_dataset_name);
4946 EXPORT_SYMBOL(dsl_dataset_rele);
4947 EXPORT_SYMBOL(dsl_dataset_rele_flags);
4948 EXPORT_SYMBOL(dsl_dataset_disown);
4949 EXPORT_SYMBOL(dsl_dataset_tryown);
4950 EXPORT_SYMBOL(dsl_dataset_create_sync);
4951 EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
4952 EXPORT_SYMBOL(dsl_dataset_snapshot_check);
4953 EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
4954 EXPORT_SYMBOL(dsl_dataset_promote);
4955 EXPORT_SYMBOL(dsl_dataset_user_hold);
4956 EXPORT_SYMBOL(dsl_dataset_user_release);
4957 EXPORT_SYMBOL(dsl_dataset_get_holds);
4958 EXPORT_SYMBOL(dsl_dataset_get_blkptr);
4959 EXPORT_SYMBOL(dsl_dataset_get_spa);
4960 EXPORT_SYMBOL(dsl_dataset_modified_since_snap);
4961 EXPORT_SYMBOL(dsl_dataset_space_written);
4962 EXPORT_SYMBOL(dsl_dataset_space_wouldfree);
4963 EXPORT_SYMBOL(dsl_dataset_sync);
4964 EXPORT_SYMBOL(dsl_dataset_block_born);
4965 EXPORT_SYMBOL(dsl_dataset_block_kill);
4966 EXPORT_SYMBOL(dsl_dataset_dirty);
4967 EXPORT_SYMBOL(dsl_dataset_stats);
4968 EXPORT_SYMBOL(dsl_dataset_fast_stat);
4969 EXPORT_SYMBOL(dsl_dataset_space);
4970 EXPORT_SYMBOL(dsl_dataset_fsid_guid);
4971 EXPORT_SYMBOL(dsl_dsobj_to_dsname);
4972 EXPORT_SYMBOL(dsl_dataset_check_quota);
4973 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl);
4974 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl);
4975