xref: /titanic_50/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision cb8a054b1ab30d5caa746e6c44f29d4c9d3071c1)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 #include <sys/dmu_objset.h>
26 #include <sys/dsl_dataset.h>
27 #include <sys/dsl_dir.h>
28 #include <sys/dsl_prop.h>
29 #include <sys/dsl_synctask.h>
30 #include <sys/dmu_traverse.h>
31 #include <sys/dmu_tx.h>
32 #include <sys/arc.h>
33 #include <sys/zio.h>
34 #include <sys/zap.h>
35 #include <sys/unique.h>
36 #include <sys/zfs_context.h>
37 #include <sys/zfs_ioctl.h>
38 #include <sys/spa.h>
39 #include <sys/zfs_znode.h>
40 #include <sys/zfs_onexit.h>
41 #include <sys/zvol.h>
42 #include <sys/dsl_scan.h>
43 #include <sys/dsl_deadlist.h>
44 
45 /*
46  * Enable/disable prefetching of dedup-ed blocks which are going to be freed.
47  */
48 int zfs_dedup_prefetch = 1;
49 
50 static char *dsl_reaper = "the grim reaper";
51 
52 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
53 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
54 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
55 
56 #define	SWITCH64(x, y) \
57 	{ \
58 		uint64_t __tmp = (x); \
59 		(x) = (y); \
60 		(y) = __tmp; \
61 	}
62 
63 #define	DS_REF_MAX	(1ULL << 62)
64 
65 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
66 
67 #define	DSL_DATASET_IS_DESTROYED(ds)	((ds)->ds_owner == dsl_reaper)
68 
69 
70 /*
71  * Figure out how much of this delta should be propogated to the dsl_dir
72  * layer.  If there's a refreservation, that space has already been
73  * partially accounted for in our ancestors.
74  */
75 static int64_t
76 parent_delta(dsl_dataset_t *ds, int64_t delta)
77 {
78 	uint64_t old_bytes, new_bytes;
79 
80 	if (ds->ds_reserved == 0)
81 		return (delta);
82 
83 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
84 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
85 
86 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
87 	return (new_bytes - old_bytes);
88 }
89 
90 void
91 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
92 {
93 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
94 	int compressed = BP_GET_PSIZE(bp);
95 	int uncompressed = BP_GET_UCSIZE(bp);
96 	int64_t delta;
97 
98 	dprintf_bp(bp, "ds=%p", ds);
99 
100 	ASSERT(dmu_tx_is_syncing(tx));
101 	/* It could have been compressed away to nothing */
102 	if (BP_IS_HOLE(bp))
103 		return;
104 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
105 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
106 	if (ds == NULL) {
107 		/*
108 		 * Account for the meta-objset space in its placeholder
109 		 * dsl_dir.
110 		 */
111 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
112 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
113 		    used, compressed, uncompressed, tx);
114 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
115 		return;
116 	}
117 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
118 
119 	mutex_enter(&ds->ds_dir->dd_lock);
120 	mutex_enter(&ds->ds_lock);
121 	delta = parent_delta(ds, used);
122 	ds->ds_phys->ds_used_bytes += used;
123 	ds->ds_phys->ds_compressed_bytes += compressed;
124 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
125 	ds->ds_phys->ds_unique_bytes += used;
126 	mutex_exit(&ds->ds_lock);
127 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
128 	    compressed, uncompressed, tx);
129 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
130 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
131 	mutex_exit(&ds->ds_dir->dd_lock);
132 }
133 
134 int
135 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
136     boolean_t async)
137 {
138 	if (BP_IS_HOLE(bp))
139 		return (0);
140 
141 	ASSERT(dmu_tx_is_syncing(tx));
142 	ASSERT(bp->blk_birth <= tx->tx_txg);
143 
144 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
145 	int compressed = BP_GET_PSIZE(bp);
146 	int uncompressed = BP_GET_UCSIZE(bp);
147 
148 	ASSERT(used > 0);
149 	if (ds == NULL) {
150 		/*
151 		 * Account for the meta-objset space in its placeholder
152 		 * dataset.
153 		 */
154 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
155 
156 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
157 		    -used, -compressed, -uncompressed, tx);
158 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
159 		return (used);
160 	}
161 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
162 
163 	ASSERT(!dsl_dataset_is_snapshot(ds));
164 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
165 
166 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
167 		int64_t delta;
168 
169 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
170 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
171 
172 		mutex_enter(&ds->ds_dir->dd_lock);
173 		mutex_enter(&ds->ds_lock);
174 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
175 		    !DS_UNIQUE_IS_ACCURATE(ds));
176 		delta = parent_delta(ds, -used);
177 		ds->ds_phys->ds_unique_bytes -= used;
178 		mutex_exit(&ds->ds_lock);
179 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
180 		    delta, -compressed, -uncompressed, tx);
181 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
182 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
183 		mutex_exit(&ds->ds_dir->dd_lock);
184 	} else {
185 		dprintf_bp(bp, "putting on dead list: %s", "");
186 		if (async) {
187 			/*
188 			 * We are here as part of zio's write done callback,
189 			 * which means we're a zio interrupt thread.  We can't
190 			 * call dsl_deadlist_insert() now because it may block
191 			 * waiting for I/O.  Instead, put bp on the deferred
192 			 * queue and let dsl_pool_sync() finish the job.
193 			 */
194 			bplist_append(&ds->ds_pending_deadlist, bp);
195 		} else {
196 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
197 		}
198 		ASSERT3U(ds->ds_prev->ds_object, ==,
199 		    ds->ds_phys->ds_prev_snap_obj);
200 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
201 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
202 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
203 		    ds->ds_object && bp->blk_birth >
204 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
205 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
206 			mutex_enter(&ds->ds_prev->ds_lock);
207 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
208 			mutex_exit(&ds->ds_prev->ds_lock);
209 		}
210 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
211 			dsl_dir_transfer_space(ds->ds_dir, used,
212 			    DD_USED_HEAD, DD_USED_SNAP, tx);
213 		}
214 	}
215 	mutex_enter(&ds->ds_lock);
216 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
217 	ds->ds_phys->ds_used_bytes -= used;
218 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
219 	ds->ds_phys->ds_compressed_bytes -= compressed;
220 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
221 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
222 	mutex_exit(&ds->ds_lock);
223 
224 	return (used);
225 }
226 
227 uint64_t
228 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
229 {
230 	uint64_t trysnap = 0;
231 
232 	if (ds == NULL)
233 		return (0);
234 	/*
235 	 * The snapshot creation could fail, but that would cause an
236 	 * incorrect FALSE return, which would only result in an
237 	 * overestimation of the amount of space that an operation would
238 	 * consume, which is OK.
239 	 *
240 	 * There's also a small window where we could miss a pending
241 	 * snapshot, because we could set the sync task in the quiescing
242 	 * phase.  So this should only be used as a guess.
243 	 */
244 	if (ds->ds_trysnap_txg >
245 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
246 		trysnap = ds->ds_trysnap_txg;
247 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
248 }
249 
250 boolean_t
251 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
252     uint64_t blk_birth)
253 {
254 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
255 		return (B_FALSE);
256 
257 	if (zfs_dedup_prefetch && bp && BP_GET_DEDUP(bp))
258 		ddt_prefetch(dsl_dataset_get_spa(ds), bp);
259 
260 	return (B_TRUE);
261 }
262 
263 /* ARGSUSED */
264 static void
265 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
266 {
267 	dsl_dataset_t *ds = dsv;
268 
269 	ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
270 
271 	unique_remove(ds->ds_fsid_guid);
272 
273 	if (ds->ds_objset != NULL)
274 		dmu_objset_evict(ds->ds_objset);
275 
276 	if (ds->ds_prev) {
277 		dsl_dataset_drop_ref(ds->ds_prev, ds);
278 		ds->ds_prev = NULL;
279 	}
280 
281 	bplist_destroy(&ds->ds_pending_deadlist);
282 	if (db != NULL) {
283 		dsl_deadlist_close(&ds->ds_deadlist);
284 	} else {
285 		ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
286 		ASSERT(!ds->ds_deadlist.dl_oldfmt);
287 	}
288 	if (ds->ds_dir)
289 		dsl_dir_close(ds->ds_dir, ds);
290 
291 	ASSERT(!list_link_active(&ds->ds_synced_link));
292 
293 	mutex_destroy(&ds->ds_lock);
294 	mutex_destroy(&ds->ds_recvlock);
295 	mutex_destroy(&ds->ds_opening_lock);
296 	rw_destroy(&ds->ds_rwlock);
297 	cv_destroy(&ds->ds_exclusive_cv);
298 
299 	kmem_free(ds, sizeof (dsl_dataset_t));
300 }
301 
302 static int
303 dsl_dataset_get_snapname(dsl_dataset_t *ds)
304 {
305 	dsl_dataset_phys_t *headphys;
306 	int err;
307 	dmu_buf_t *headdbuf;
308 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
309 	objset_t *mos = dp->dp_meta_objset;
310 
311 	if (ds->ds_snapname[0])
312 		return (0);
313 	if (ds->ds_phys->ds_next_snap_obj == 0)
314 		return (0);
315 
316 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
317 	    FTAG, &headdbuf);
318 	if (err)
319 		return (err);
320 	headphys = headdbuf->db_data;
321 	err = zap_value_search(dp->dp_meta_objset,
322 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
323 	dmu_buf_rele(headdbuf, FTAG);
324 	return (err);
325 }
326 
327 static int
328 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
329 {
330 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
331 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
332 	matchtype_t mt;
333 	int err;
334 
335 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
336 		mt = MT_FIRST;
337 	else
338 		mt = MT_EXACT;
339 
340 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
341 	    value, mt, NULL, 0, NULL);
342 	if (err == ENOTSUP && mt == MT_FIRST)
343 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
344 	return (err);
345 }
346 
347 static int
348 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
349 {
350 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
351 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
352 	matchtype_t mt;
353 	int err;
354 
355 	dsl_dir_snap_cmtime_update(ds->ds_dir);
356 
357 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
358 		mt = MT_FIRST;
359 	else
360 		mt = MT_EXACT;
361 
362 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
363 	if (err == ENOTSUP && mt == MT_FIRST)
364 		err = zap_remove(mos, snapobj, name, tx);
365 	return (err);
366 }
367 
368 static int
369 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
370     dsl_dataset_t **dsp)
371 {
372 	objset_t *mos = dp->dp_meta_objset;
373 	dmu_buf_t *dbuf;
374 	dsl_dataset_t *ds;
375 	int err;
376 
377 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
378 	    dsl_pool_sync_context(dp));
379 
380 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
381 	if (err)
382 		return (err);
383 	ds = dmu_buf_get_user(dbuf);
384 	if (ds == NULL) {
385 		dsl_dataset_t *winner;
386 
387 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
388 		ds->ds_dbuf = dbuf;
389 		ds->ds_object = dsobj;
390 		ds->ds_phys = dbuf->db_data;
391 
392 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
393 		mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
394 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
395 		rw_init(&ds->ds_rwlock, 0, 0, 0);
396 		cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
397 
398 		bplist_create(&ds->ds_pending_deadlist);
399 		dsl_deadlist_open(&ds->ds_deadlist,
400 		    mos, ds->ds_phys->ds_deadlist_obj);
401 
402 		if (err == 0) {
403 			err = dsl_dir_open_obj(dp,
404 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
405 		}
406 		if (err) {
407 			mutex_destroy(&ds->ds_lock);
408 			mutex_destroy(&ds->ds_recvlock);
409 			mutex_destroy(&ds->ds_opening_lock);
410 			rw_destroy(&ds->ds_rwlock);
411 			cv_destroy(&ds->ds_exclusive_cv);
412 			bplist_destroy(&ds->ds_pending_deadlist);
413 			dsl_deadlist_close(&ds->ds_deadlist);
414 			kmem_free(ds, sizeof (dsl_dataset_t));
415 			dmu_buf_rele(dbuf, tag);
416 			return (err);
417 		}
418 
419 		if (!dsl_dataset_is_snapshot(ds)) {
420 			ds->ds_snapname[0] = '\0';
421 			if (ds->ds_phys->ds_prev_snap_obj) {
422 				err = dsl_dataset_get_ref(dp,
423 				    ds->ds_phys->ds_prev_snap_obj,
424 				    ds, &ds->ds_prev);
425 			}
426 		} else {
427 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
428 				err = dsl_dataset_get_snapname(ds);
429 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
430 				err = zap_count(
431 				    ds->ds_dir->dd_pool->dp_meta_objset,
432 				    ds->ds_phys->ds_userrefs_obj,
433 				    &ds->ds_userrefs);
434 			}
435 		}
436 
437 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
438 			/*
439 			 * In sync context, we're called with either no lock
440 			 * or with the write lock.  If we're not syncing,
441 			 * we're always called with the read lock held.
442 			 */
443 			boolean_t need_lock =
444 			    !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
445 			    dsl_pool_sync_context(dp);
446 
447 			if (need_lock)
448 				rw_enter(&dp->dp_config_rwlock, RW_READER);
449 
450 			err = dsl_prop_get_ds(ds,
451 			    "refreservation", sizeof (uint64_t), 1,
452 			    &ds->ds_reserved, NULL);
453 			if (err == 0) {
454 				err = dsl_prop_get_ds(ds,
455 				    "refquota", sizeof (uint64_t), 1,
456 				    &ds->ds_quota, NULL);
457 			}
458 
459 			if (need_lock)
460 				rw_exit(&dp->dp_config_rwlock);
461 		} else {
462 			ds->ds_reserved = ds->ds_quota = 0;
463 		}
464 
465 		if (err == 0) {
466 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
467 			    dsl_dataset_evict);
468 		}
469 		if (err || winner) {
470 			bplist_destroy(&ds->ds_pending_deadlist);
471 			dsl_deadlist_close(&ds->ds_deadlist);
472 			if (ds->ds_prev)
473 				dsl_dataset_drop_ref(ds->ds_prev, ds);
474 			dsl_dir_close(ds->ds_dir, ds);
475 			mutex_destroy(&ds->ds_lock);
476 			mutex_destroy(&ds->ds_recvlock);
477 			mutex_destroy(&ds->ds_opening_lock);
478 			rw_destroy(&ds->ds_rwlock);
479 			cv_destroy(&ds->ds_exclusive_cv);
480 			kmem_free(ds, sizeof (dsl_dataset_t));
481 			if (err) {
482 				dmu_buf_rele(dbuf, tag);
483 				return (err);
484 			}
485 			ds = winner;
486 		} else {
487 			ds->ds_fsid_guid =
488 			    unique_insert(ds->ds_phys->ds_fsid_guid);
489 		}
490 	}
491 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
492 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
493 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
494 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
495 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
496 	mutex_enter(&ds->ds_lock);
497 	if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
498 		mutex_exit(&ds->ds_lock);
499 		dmu_buf_rele(ds->ds_dbuf, tag);
500 		return (ENOENT);
501 	}
502 	mutex_exit(&ds->ds_lock);
503 	*dsp = ds;
504 	return (0);
505 }
506 
507 static int
508 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
509 {
510 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
511 
512 	/*
513 	 * In syncing context we don't want the rwlock lock: there
514 	 * may be an existing writer waiting for sync phase to
515 	 * finish.  We don't need to worry about such writers, since
516 	 * sync phase is single-threaded, so the writer can't be
517 	 * doing anything while we are active.
518 	 */
519 	if (dsl_pool_sync_context(dp)) {
520 		ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
521 		return (0);
522 	}
523 
524 	/*
525 	 * Normal users will hold the ds_rwlock as a READER until they
526 	 * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
527 	 * drop their READER lock after they set the ds_owner field.
528 	 *
529 	 * If the dataset is being destroyed, the destroy thread will
530 	 * obtain a WRITER lock for exclusive access after it's done its
531 	 * open-context work and then change the ds_owner to
532 	 * dsl_reaper once destruction is assured.  So threads
533 	 * may block here temporarily, until the "destructability" of
534 	 * the dataset is determined.
535 	 */
536 	ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
537 	mutex_enter(&ds->ds_lock);
538 	while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
539 		rw_exit(&dp->dp_config_rwlock);
540 		cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
541 		if (DSL_DATASET_IS_DESTROYED(ds)) {
542 			mutex_exit(&ds->ds_lock);
543 			dsl_dataset_drop_ref(ds, tag);
544 			rw_enter(&dp->dp_config_rwlock, RW_READER);
545 			return (ENOENT);
546 		}
547 		/*
548 		 * The dp_config_rwlock lives above the ds_lock. And
549 		 * we need to check DSL_DATASET_IS_DESTROYED() while
550 		 * holding the ds_lock, so we have to drop and reacquire
551 		 * the ds_lock here.
552 		 */
553 		mutex_exit(&ds->ds_lock);
554 		rw_enter(&dp->dp_config_rwlock, RW_READER);
555 		mutex_enter(&ds->ds_lock);
556 	}
557 	mutex_exit(&ds->ds_lock);
558 	return (0);
559 }
560 
561 int
562 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
563     dsl_dataset_t **dsp)
564 {
565 	int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
566 
567 	if (err)
568 		return (err);
569 	return (dsl_dataset_hold_ref(*dsp, tag));
570 }
571 
572 int
573 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
574     void *tag, dsl_dataset_t **dsp)
575 {
576 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
577 	if (err)
578 		return (err);
579 	if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
580 		dsl_dataset_rele(*dsp, tag);
581 		*dsp = NULL;
582 		return (EBUSY);
583 	}
584 	return (0);
585 }
586 
587 int
588 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
589 {
590 	dsl_dir_t *dd;
591 	dsl_pool_t *dp;
592 	const char *snapname;
593 	uint64_t obj;
594 	int err = 0;
595 
596 	err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
597 	if (err)
598 		return (err);
599 
600 	dp = dd->dd_pool;
601 	obj = dd->dd_phys->dd_head_dataset_obj;
602 	rw_enter(&dp->dp_config_rwlock, RW_READER);
603 	if (obj)
604 		err = dsl_dataset_get_ref(dp, obj, tag, dsp);
605 	else
606 		err = ENOENT;
607 	if (err)
608 		goto out;
609 
610 	err = dsl_dataset_hold_ref(*dsp, tag);
611 
612 	/* we may be looking for a snapshot */
613 	if (err == 0 && snapname != NULL) {
614 		dsl_dataset_t *ds = NULL;
615 
616 		if (*snapname++ != '@') {
617 			dsl_dataset_rele(*dsp, tag);
618 			err = ENOENT;
619 			goto out;
620 		}
621 
622 		dprintf("looking for snapshot '%s'\n", snapname);
623 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
624 		if (err == 0)
625 			err = dsl_dataset_get_ref(dp, obj, tag, &ds);
626 		dsl_dataset_rele(*dsp, tag);
627 
628 		ASSERT3U((err == 0), ==, (ds != NULL));
629 
630 		if (ds) {
631 			mutex_enter(&ds->ds_lock);
632 			if (ds->ds_snapname[0] == 0)
633 				(void) strlcpy(ds->ds_snapname, snapname,
634 				    sizeof (ds->ds_snapname));
635 			mutex_exit(&ds->ds_lock);
636 			err = dsl_dataset_hold_ref(ds, tag);
637 			*dsp = err ? NULL : ds;
638 		}
639 	}
640 out:
641 	rw_exit(&dp->dp_config_rwlock);
642 	dsl_dir_close(dd, FTAG);
643 	return (err);
644 }
645 
646 int
647 dsl_dataset_own(const char *name, boolean_t inconsistentok,
648     void *tag, dsl_dataset_t **dsp)
649 {
650 	int err = dsl_dataset_hold(name, tag, dsp);
651 	if (err)
652 		return (err);
653 	if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
654 		dsl_dataset_rele(*dsp, tag);
655 		return (EBUSY);
656 	}
657 	return (0);
658 }
659 
660 void
661 dsl_dataset_name(dsl_dataset_t *ds, char *name)
662 {
663 	if (ds == NULL) {
664 		(void) strcpy(name, "mos");
665 	} else {
666 		dsl_dir_name(ds->ds_dir, name);
667 		VERIFY(0 == dsl_dataset_get_snapname(ds));
668 		if (ds->ds_snapname[0]) {
669 			(void) strcat(name, "@");
670 			/*
671 			 * We use a "recursive" mutex so that we
672 			 * can call dprintf_ds() with ds_lock held.
673 			 */
674 			if (!MUTEX_HELD(&ds->ds_lock)) {
675 				mutex_enter(&ds->ds_lock);
676 				(void) strcat(name, ds->ds_snapname);
677 				mutex_exit(&ds->ds_lock);
678 			} else {
679 				(void) strcat(name, ds->ds_snapname);
680 			}
681 		}
682 	}
683 }
684 
685 static int
686 dsl_dataset_namelen(dsl_dataset_t *ds)
687 {
688 	int result;
689 
690 	if (ds == NULL) {
691 		result = 3;	/* "mos" */
692 	} else {
693 		result = dsl_dir_namelen(ds->ds_dir);
694 		VERIFY(0 == dsl_dataset_get_snapname(ds));
695 		if (ds->ds_snapname[0]) {
696 			++result;	/* adding one for the @-sign */
697 			if (!MUTEX_HELD(&ds->ds_lock)) {
698 				mutex_enter(&ds->ds_lock);
699 				result += strlen(ds->ds_snapname);
700 				mutex_exit(&ds->ds_lock);
701 			} else {
702 				result += strlen(ds->ds_snapname);
703 			}
704 		}
705 	}
706 
707 	return (result);
708 }
709 
710 void
711 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
712 {
713 	dmu_buf_rele(ds->ds_dbuf, tag);
714 }
715 
716 void
717 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
718 {
719 	if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
720 		rw_exit(&ds->ds_rwlock);
721 	}
722 	dsl_dataset_drop_ref(ds, tag);
723 }
724 
725 void
726 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
727 {
728 	ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
729 	    (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
730 
731 	mutex_enter(&ds->ds_lock);
732 	ds->ds_owner = NULL;
733 	if (RW_WRITE_HELD(&ds->ds_rwlock)) {
734 		rw_exit(&ds->ds_rwlock);
735 		cv_broadcast(&ds->ds_exclusive_cv);
736 	}
737 	mutex_exit(&ds->ds_lock);
738 	if (ds->ds_dbuf)
739 		dsl_dataset_drop_ref(ds, tag);
740 	else
741 		dsl_dataset_evict(NULL, ds);
742 }
743 
744 boolean_t
745 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
746 {
747 	boolean_t gotit = FALSE;
748 
749 	mutex_enter(&ds->ds_lock);
750 	if (ds->ds_owner == NULL &&
751 	    (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
752 		ds->ds_owner = tag;
753 		if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
754 			rw_exit(&ds->ds_rwlock);
755 		gotit = TRUE;
756 	}
757 	mutex_exit(&ds->ds_lock);
758 	return (gotit);
759 }
760 
761 void
762 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
763 {
764 	ASSERT3P(owner, ==, ds->ds_owner);
765 	if (!RW_WRITE_HELD(&ds->ds_rwlock))
766 		rw_enter(&ds->ds_rwlock, RW_WRITER);
767 }
768 
769 uint64_t
770 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
771     uint64_t flags, dmu_tx_t *tx)
772 {
773 	dsl_pool_t *dp = dd->dd_pool;
774 	dmu_buf_t *dbuf;
775 	dsl_dataset_phys_t *dsphys;
776 	uint64_t dsobj;
777 	objset_t *mos = dp->dp_meta_objset;
778 
779 	if (origin == NULL)
780 		origin = dp->dp_origin_snap;
781 
782 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
783 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
784 	ASSERT(dmu_tx_is_syncing(tx));
785 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
786 
787 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
788 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
789 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
790 	dmu_buf_will_dirty(dbuf, tx);
791 	dsphys = dbuf->db_data;
792 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
793 	dsphys->ds_dir_obj = dd->dd_object;
794 	dsphys->ds_flags = flags;
795 	dsphys->ds_fsid_guid = unique_create();
796 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
797 	    sizeof (dsphys->ds_guid));
798 	dsphys->ds_snapnames_zapobj =
799 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
800 	    DMU_OT_NONE, 0, tx);
801 	dsphys->ds_creation_time = gethrestime_sec();
802 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
803 
804 	if (origin == NULL) {
805 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
806 	} else {
807 		dsl_dataset_t *ohds;
808 
809 		dsphys->ds_prev_snap_obj = origin->ds_object;
810 		dsphys->ds_prev_snap_txg =
811 		    origin->ds_phys->ds_creation_txg;
812 		dsphys->ds_used_bytes =
813 		    origin->ds_phys->ds_used_bytes;
814 		dsphys->ds_compressed_bytes =
815 		    origin->ds_phys->ds_compressed_bytes;
816 		dsphys->ds_uncompressed_bytes =
817 		    origin->ds_phys->ds_uncompressed_bytes;
818 		dsphys->ds_bp = origin->ds_phys->ds_bp;
819 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
820 
821 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
822 		origin->ds_phys->ds_num_children++;
823 
824 		VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
825 		    origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
826 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
827 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
828 		dsl_dataset_rele(ohds, FTAG);
829 
830 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
831 			if (origin->ds_phys->ds_next_clones_obj == 0) {
832 				origin->ds_phys->ds_next_clones_obj =
833 				    zap_create(mos,
834 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
835 			}
836 			VERIFY(0 == zap_add_int(mos,
837 			    origin->ds_phys->ds_next_clones_obj,
838 			    dsobj, tx));
839 		}
840 
841 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
842 		dd->dd_phys->dd_origin_obj = origin->ds_object;
843 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
844 			if (origin->ds_dir->dd_phys->dd_clones == 0) {
845 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
846 				origin->ds_dir->dd_phys->dd_clones =
847 				    zap_create(mos,
848 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
849 			}
850 			VERIFY3U(0, ==, zap_add_int(mos,
851 			    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
852 		}
853 	}
854 
855 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
856 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
857 
858 	dmu_buf_rele(dbuf, FTAG);
859 
860 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
861 	dd->dd_phys->dd_head_dataset_obj = dsobj;
862 
863 	return (dsobj);
864 }
865 
866 uint64_t
867 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
868     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
869 {
870 	dsl_pool_t *dp = pdd->dd_pool;
871 	uint64_t dsobj, ddobj;
872 	dsl_dir_t *dd;
873 
874 	ASSERT(lastname[0] != '@');
875 
876 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
877 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
878 
879 	dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
880 
881 	dsl_deleg_set_create_perms(dd, tx, cr);
882 
883 	dsl_dir_close(dd, FTAG);
884 
885 	return (dsobj);
886 }
887 
888 struct destroyarg {
889 	dsl_sync_task_group_t *dstg;
890 	char *snapname;
891 	char *failed;
892 	boolean_t defer;
893 };
894 
895 static int
896 dsl_snapshot_destroy_one(const char *name, void *arg)
897 {
898 	struct destroyarg *da = arg;
899 	dsl_dataset_t *ds;
900 	int err;
901 	char *dsname;
902 
903 	dsname = kmem_asprintf("%s@%s", name, da->snapname);
904 	err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds);
905 	strfree(dsname);
906 	if (err == 0) {
907 		struct dsl_ds_destroyarg *dsda;
908 
909 		dsl_dataset_make_exclusive(ds, da->dstg);
910 		dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
911 		dsda->ds = ds;
912 		dsda->defer = da->defer;
913 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
914 		    dsl_dataset_destroy_sync, dsda, da->dstg, 0);
915 	} else if (err == ENOENT) {
916 		err = 0;
917 	} else {
918 		(void) strcpy(da->failed, name);
919 	}
920 	return (err);
921 }
922 
923 /*
924  * Destroy 'snapname' in all descendants of 'fsname'.
925  */
926 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
927 int
928 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
929 {
930 	int err;
931 	struct destroyarg da;
932 	dsl_sync_task_t *dst;
933 	spa_t *spa;
934 
935 	err = spa_open(fsname, &spa, FTAG);
936 	if (err)
937 		return (err);
938 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
939 	da.snapname = snapname;
940 	da.failed = fsname;
941 	da.defer = defer;
942 
943 	err = dmu_objset_find(fsname,
944 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
945 
946 	if (err == 0)
947 		err = dsl_sync_task_group_wait(da.dstg);
948 
949 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
950 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
951 		struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
952 		dsl_dataset_t *ds = dsda->ds;
953 
954 		/*
955 		 * Return the file system name that triggered the error
956 		 */
957 		if (dst->dst_err) {
958 			dsl_dataset_name(ds, fsname);
959 			*strchr(fsname, '@') = '\0';
960 		}
961 		ASSERT3P(dsda->rm_origin, ==, NULL);
962 		dsl_dataset_disown(ds, da.dstg);
963 		kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
964 	}
965 
966 	dsl_sync_task_group_destroy(da.dstg);
967 	spa_close(spa, FTAG);
968 	return (err);
969 }
970 
971 static boolean_t
972 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
973 {
974 	boolean_t might_destroy = B_FALSE;
975 
976 	mutex_enter(&ds->ds_lock);
977 	if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
978 	    DS_IS_DEFER_DESTROY(ds))
979 		might_destroy = B_TRUE;
980 	mutex_exit(&ds->ds_lock);
981 
982 	return (might_destroy);
983 }
984 
985 /*
986  * If we're removing a clone, and these three conditions are true:
987  *	1) the clone's origin has no other children
988  *	2) the clone's origin has no user references
989  *	3) the clone's origin has been marked for deferred destruction
990  * Then, prepare to remove the origin as part of this sync task group.
991  */
992 static int
993 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
994 {
995 	dsl_dataset_t *ds = dsda->ds;
996 	dsl_dataset_t *origin = ds->ds_prev;
997 
998 	if (dsl_dataset_might_destroy_origin(origin)) {
999 		char *name;
1000 		int namelen;
1001 		int error;
1002 
1003 		namelen = dsl_dataset_namelen(origin) + 1;
1004 		name = kmem_alloc(namelen, KM_SLEEP);
1005 		dsl_dataset_name(origin, name);
1006 #ifdef _KERNEL
1007 		error = zfs_unmount_snap(name, NULL);
1008 		if (error) {
1009 			kmem_free(name, namelen);
1010 			return (error);
1011 		}
1012 #endif
1013 		error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1014 		kmem_free(name, namelen);
1015 		if (error)
1016 			return (error);
1017 		dsda->rm_origin = origin;
1018 		dsl_dataset_make_exclusive(origin, tag);
1019 	}
1020 
1021 	return (0);
1022 }
1023 
1024 /*
1025  * ds must be opened as OWNER.  On return (whether successful or not),
1026  * ds will be closed and caller can no longer dereference it.
1027  */
1028 int
1029 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1030 {
1031 	int err;
1032 	dsl_sync_task_group_t *dstg;
1033 	objset_t *os;
1034 	dsl_dir_t *dd;
1035 	uint64_t obj;
1036 	struct dsl_ds_destroyarg dsda = { 0 };
1037 	dsl_dataset_t dummy_ds = { 0 };
1038 
1039 	dsda.ds = ds;
1040 
1041 	if (dsl_dataset_is_snapshot(ds)) {
1042 		/* Destroying a snapshot is simpler */
1043 		dsl_dataset_make_exclusive(ds, tag);
1044 
1045 		dsda.defer = defer;
1046 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1047 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1048 		    &dsda, tag, 0);
1049 		ASSERT3P(dsda.rm_origin, ==, NULL);
1050 		goto out;
1051 	} else if (defer) {
1052 		err = EINVAL;
1053 		goto out;
1054 	}
1055 
1056 	dd = ds->ds_dir;
1057 	dummy_ds.ds_dir = dd;
1058 	dummy_ds.ds_object = ds->ds_object;
1059 
1060 	/*
1061 	 * Check for errors and mark this ds as inconsistent, in
1062 	 * case we crash while freeing the objects.
1063 	 */
1064 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1065 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1066 	if (err)
1067 		goto out;
1068 
1069 	err = dmu_objset_from_ds(ds, &os);
1070 	if (err)
1071 		goto out;
1072 
1073 	/*
1074 	 * remove the objects in open context, so that we won't
1075 	 * have too much to do in syncing context.
1076 	 */
1077 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1078 	    ds->ds_phys->ds_prev_snap_txg)) {
1079 		/*
1080 		 * Ignore errors, if there is not enough disk space
1081 		 * we will deal with it in dsl_dataset_destroy_sync().
1082 		 */
1083 		(void) dmu_free_object(os, obj);
1084 	}
1085 
1086 	/*
1087 	 * We need to sync out all in-flight IO before we try to evict
1088 	 * (the dataset evict func is trying to clear the cached entries
1089 	 * for this dataset in the ARC).
1090 	 */
1091 	txg_wait_synced(dd->dd_pool, 0);
1092 
1093 	/*
1094 	 * If we managed to free all the objects in open
1095 	 * context, the user space accounting should be zero.
1096 	 */
1097 	if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1098 	    dmu_objset_userused_enabled(os)) {
1099 		uint64_t count;
1100 
1101 		ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1102 		    count == 0);
1103 		ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1104 		    count == 0);
1105 	}
1106 
1107 	if (err != ESRCH)
1108 		goto out;
1109 
1110 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1111 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1112 	rw_exit(&dd->dd_pool->dp_config_rwlock);
1113 
1114 	if (err)
1115 		goto out;
1116 
1117 	/*
1118 	 * Blow away the dsl_dir + head dataset.
1119 	 */
1120 	dsl_dataset_make_exclusive(ds, tag);
1121 	/*
1122 	 * If we're removing a clone, we might also need to remove its
1123 	 * origin.
1124 	 */
1125 	do {
1126 		dsda.need_prep = B_FALSE;
1127 		if (dsl_dir_is_clone(dd)) {
1128 			err = dsl_dataset_origin_rm_prep(&dsda, tag);
1129 			if (err) {
1130 				dsl_dir_close(dd, FTAG);
1131 				goto out;
1132 			}
1133 		}
1134 
1135 		dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1136 		dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1137 		    dsl_dataset_destroy_sync, &dsda, tag, 0);
1138 		dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1139 		    dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1140 		err = dsl_sync_task_group_wait(dstg);
1141 		dsl_sync_task_group_destroy(dstg);
1142 
1143 		/*
1144 		 * We could be racing against 'zfs release' or 'zfs destroy -d'
1145 		 * on the origin snap, in which case we can get EBUSY if we
1146 		 * needed to destroy the origin snap but were not ready to
1147 		 * do so.
1148 		 */
1149 		if (dsda.need_prep) {
1150 			ASSERT(err == EBUSY);
1151 			ASSERT(dsl_dir_is_clone(dd));
1152 			ASSERT(dsda.rm_origin == NULL);
1153 		}
1154 	} while (dsda.need_prep);
1155 
1156 	if (dsda.rm_origin != NULL)
1157 		dsl_dataset_disown(dsda.rm_origin, tag);
1158 
1159 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
1160 	if (err)
1161 		dsl_dir_close(dd, FTAG);
1162 out:
1163 	dsl_dataset_disown(ds, tag);
1164 	return (err);
1165 }
1166 
1167 blkptr_t *
1168 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1169 {
1170 	return (&ds->ds_phys->ds_bp);
1171 }
1172 
1173 void
1174 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1175 {
1176 	ASSERT(dmu_tx_is_syncing(tx));
1177 	/* If it's the meta-objset, set dp_meta_rootbp */
1178 	if (ds == NULL) {
1179 		tx->tx_pool->dp_meta_rootbp = *bp;
1180 	} else {
1181 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1182 		ds->ds_phys->ds_bp = *bp;
1183 	}
1184 }
1185 
1186 spa_t *
1187 dsl_dataset_get_spa(dsl_dataset_t *ds)
1188 {
1189 	return (ds->ds_dir->dd_pool->dp_spa);
1190 }
1191 
1192 void
1193 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1194 {
1195 	dsl_pool_t *dp;
1196 
1197 	if (ds == NULL) /* this is the meta-objset */
1198 		return;
1199 
1200 	ASSERT(ds->ds_objset != NULL);
1201 
1202 	if (ds->ds_phys->ds_next_snap_obj != 0)
1203 		panic("dirtying snapshot!");
1204 
1205 	dp = ds->ds_dir->dd_pool;
1206 
1207 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1208 		/* up the hold count until we can be written out */
1209 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1210 	}
1211 }
1212 
1213 /*
1214  * The unique space in the head dataset can be calculated by subtracting
1215  * the space used in the most recent snapshot, that is still being used
1216  * in this file system, from the space currently in use.  To figure out
1217  * the space in the most recent snapshot still in use, we need to take
1218  * the total space used in the snapshot and subtract out the space that
1219  * has been freed up since the snapshot was taken.
1220  */
1221 static void
1222 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1223 {
1224 	uint64_t mrs_used;
1225 	uint64_t dlused, dlcomp, dluncomp;
1226 
1227 	ASSERT(!dsl_dataset_is_snapshot(ds));
1228 
1229 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1230 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1231 	else
1232 		mrs_used = 0;
1233 
1234 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1235 
1236 	ASSERT3U(dlused, <=, mrs_used);
1237 	ds->ds_phys->ds_unique_bytes =
1238 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1239 
1240 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1241 	    SPA_VERSION_UNIQUE_ACCURATE)
1242 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1243 }
1244 
1245 struct killarg {
1246 	dsl_dataset_t *ds;
1247 	dmu_tx_t *tx;
1248 };
1249 
1250 /* ARGSUSED */
1251 static int
1252 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1253     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1254 {
1255 	struct killarg *ka = arg;
1256 	dmu_tx_t *tx = ka->tx;
1257 
1258 	if (bp == NULL)
1259 		return (0);
1260 
1261 	if (zb->zb_level == ZB_ZIL_LEVEL) {
1262 		ASSERT(zilog != NULL);
1263 		/*
1264 		 * It's a block in the intent log.  It has no
1265 		 * accounting, so just free it.
1266 		 */
1267 		dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1268 	} else {
1269 		ASSERT(zilog == NULL);
1270 		ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1271 		(void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1272 	}
1273 
1274 	return (0);
1275 }
1276 
1277 /* ARGSUSED */
1278 static int
1279 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1280 {
1281 	dsl_dataset_t *ds = arg1;
1282 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1283 	uint64_t count;
1284 	int err;
1285 
1286 	/*
1287 	 * Can't delete a head dataset if there are snapshots of it.
1288 	 * (Except if the only snapshots are from the branch we cloned
1289 	 * from.)
1290 	 */
1291 	if (ds->ds_prev != NULL &&
1292 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1293 		return (EBUSY);
1294 
1295 	/*
1296 	 * This is really a dsl_dir thing, but check it here so that
1297 	 * we'll be less likely to leave this dataset inconsistent &
1298 	 * nearly destroyed.
1299 	 */
1300 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1301 	if (err)
1302 		return (err);
1303 	if (count != 0)
1304 		return (EEXIST);
1305 
1306 	return (0);
1307 }
1308 
1309 /* ARGSUSED */
1310 static void
1311 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1312 {
1313 	dsl_dataset_t *ds = arg1;
1314 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1315 
1316 	/* Mark it as inconsistent on-disk, in case we crash */
1317 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1318 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1319 
1320 	spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1321 	    "dataset = %llu", ds->ds_object);
1322 }
1323 
1324 static int
1325 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1326     dmu_tx_t *tx)
1327 {
1328 	dsl_dataset_t *ds = dsda->ds;
1329 	dsl_dataset_t *ds_prev = ds->ds_prev;
1330 
1331 	if (dsl_dataset_might_destroy_origin(ds_prev)) {
1332 		struct dsl_ds_destroyarg ndsda = {0};
1333 
1334 		/*
1335 		 * If we're not prepared to remove the origin, don't remove
1336 		 * the clone either.
1337 		 */
1338 		if (dsda->rm_origin == NULL) {
1339 			dsda->need_prep = B_TRUE;
1340 			return (EBUSY);
1341 		}
1342 
1343 		ndsda.ds = ds_prev;
1344 		ndsda.is_origin_rm = B_TRUE;
1345 		return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1346 	}
1347 
1348 	/*
1349 	 * If we're not going to remove the origin after all,
1350 	 * undo the open context setup.
1351 	 */
1352 	if (dsda->rm_origin != NULL) {
1353 		dsl_dataset_disown(dsda->rm_origin, tag);
1354 		dsda->rm_origin = NULL;
1355 	}
1356 
1357 	return (0);
1358 }
1359 
1360 /* ARGSUSED */
1361 int
1362 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1363 {
1364 	struct dsl_ds_destroyarg *dsda = arg1;
1365 	dsl_dataset_t *ds = dsda->ds;
1366 
1367 	/* we have an owner hold, so noone else can destroy us */
1368 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1369 
1370 	/*
1371 	 * Only allow deferred destroy on pools that support it.
1372 	 * NOTE: deferred destroy is only supported on snapshots.
1373 	 */
1374 	if (dsda->defer) {
1375 		if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1376 		    SPA_VERSION_USERREFS)
1377 			return (ENOTSUP);
1378 		ASSERT(dsl_dataset_is_snapshot(ds));
1379 		return (0);
1380 	}
1381 
1382 	/*
1383 	 * Can't delete a head dataset if there are snapshots of it.
1384 	 * (Except if the only snapshots are from the branch we cloned
1385 	 * from.)
1386 	 */
1387 	if (ds->ds_prev != NULL &&
1388 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1389 		return (EBUSY);
1390 
1391 	/*
1392 	 * If we made changes this txg, traverse_dsl_dataset won't find
1393 	 * them.  Try again.
1394 	 */
1395 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1396 		return (EAGAIN);
1397 
1398 	if (dsl_dataset_is_snapshot(ds)) {
1399 		/*
1400 		 * If this snapshot has an elevated user reference count,
1401 		 * we can't destroy it yet.
1402 		 */
1403 		if (ds->ds_userrefs > 0 && !dsda->releasing)
1404 			return (EBUSY);
1405 
1406 		mutex_enter(&ds->ds_lock);
1407 		/*
1408 		 * Can't delete a branch point. However, if we're destroying
1409 		 * a clone and removing its origin due to it having a user
1410 		 * hold count of 0 and having been marked for deferred destroy,
1411 		 * it's OK for the origin to have a single clone.
1412 		 */
1413 		if (ds->ds_phys->ds_num_children >
1414 		    (dsda->is_origin_rm ? 2 : 1)) {
1415 			mutex_exit(&ds->ds_lock);
1416 			return (EEXIST);
1417 		}
1418 		mutex_exit(&ds->ds_lock);
1419 	} else if (dsl_dir_is_clone(ds->ds_dir)) {
1420 		return (dsl_dataset_origin_check(dsda, arg2, tx));
1421 	}
1422 
1423 	/* XXX we should do some i/o error checking... */
1424 	return (0);
1425 }
1426 
1427 struct refsarg {
1428 	kmutex_t lock;
1429 	boolean_t gone;
1430 	kcondvar_t cv;
1431 };
1432 
1433 /* ARGSUSED */
1434 static void
1435 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1436 {
1437 	struct refsarg *arg = argv;
1438 
1439 	mutex_enter(&arg->lock);
1440 	arg->gone = TRUE;
1441 	cv_signal(&arg->cv);
1442 	mutex_exit(&arg->lock);
1443 }
1444 
1445 static void
1446 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1447 {
1448 	struct refsarg arg;
1449 
1450 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1451 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1452 	arg.gone = FALSE;
1453 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1454 	    dsl_dataset_refs_gone);
1455 	dmu_buf_rele(ds->ds_dbuf, tag);
1456 	mutex_enter(&arg.lock);
1457 	while (!arg.gone)
1458 		cv_wait(&arg.cv, &arg.lock);
1459 	ASSERT(arg.gone);
1460 	mutex_exit(&arg.lock);
1461 	ds->ds_dbuf = NULL;
1462 	ds->ds_phys = NULL;
1463 	mutex_destroy(&arg.lock);
1464 	cv_destroy(&arg.cv);
1465 }
1466 
1467 static void
1468 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1469 {
1470 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1471 	uint64_t count;
1472 	int err;
1473 
1474 	ASSERT(ds->ds_phys->ds_num_children >= 2);
1475 	err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1476 	/*
1477 	 * The err should not be ENOENT, but a bug in a previous version
1478 	 * of the code could cause upgrade_clones_cb() to not set
1479 	 * ds_next_snap_obj when it should, leading to a missing entry.
1480 	 * If we knew that the pool was created after
1481 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1482 	 * ENOENT.  However, at least we can check that we don't have
1483 	 * too many entries in the next_clones_obj even after failing to
1484 	 * remove this one.
1485 	 */
1486 	if (err != ENOENT) {
1487 		VERIFY3U(err, ==, 0);
1488 	}
1489 	ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1490 	    &count));
1491 	ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1492 }
1493 
1494 static void
1495 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1496 {
1497 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1498 	zap_cursor_t zc;
1499 	zap_attribute_t za;
1500 
1501 	/*
1502 	 * If it is the old version, dd_clones doesn't exist so we can't
1503 	 * find the clones, but deadlist_remove_key() is a no-op so it
1504 	 * doesn't matter.
1505 	 */
1506 	if (ds->ds_dir->dd_phys->dd_clones == 0)
1507 		return;
1508 
1509 	for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1510 	    zap_cursor_retrieve(&zc, &za) == 0;
1511 	    zap_cursor_advance(&zc)) {
1512 		dsl_dataset_t *clone;
1513 
1514 		VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1515 		    za.za_first_integer, FTAG, &clone));
1516 		if (clone->ds_dir->dd_origin_txg > mintxg) {
1517 			dsl_deadlist_remove_key(&clone->ds_deadlist,
1518 			    mintxg, tx);
1519 			dsl_dataset_remove_clones_key(clone, mintxg, tx);
1520 		}
1521 		dsl_dataset_rele(clone, FTAG);
1522 	}
1523 	zap_cursor_fini(&zc);
1524 }
1525 
1526 struct process_old_arg {
1527 	dsl_dataset_t *ds;
1528 	dsl_dataset_t *ds_prev;
1529 	boolean_t after_branch_point;
1530 	zio_t *pio;
1531 	uint64_t used, comp, uncomp;
1532 };
1533 
1534 static int
1535 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1536 {
1537 	struct process_old_arg *poa = arg;
1538 	dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1539 
1540 	if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1541 		dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1542 		if (poa->ds_prev && !poa->after_branch_point &&
1543 		    bp->blk_birth >
1544 		    poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1545 			poa->ds_prev->ds_phys->ds_unique_bytes +=
1546 			    bp_get_dsize_sync(dp->dp_spa, bp);
1547 		}
1548 	} else {
1549 		poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1550 		poa->comp += BP_GET_PSIZE(bp);
1551 		poa->uncomp += BP_GET_UCSIZE(bp);
1552 		dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1553 	}
1554 	return (0);
1555 }
1556 
1557 static void
1558 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1559     dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1560 {
1561 	struct process_old_arg poa = { 0 };
1562 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1563 	objset_t *mos = dp->dp_meta_objset;
1564 
1565 	ASSERT(ds->ds_deadlist.dl_oldfmt);
1566 	ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1567 
1568 	poa.ds = ds;
1569 	poa.ds_prev = ds_prev;
1570 	poa.after_branch_point = after_branch_point;
1571 	poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1572 	VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1573 	    process_old_cb, &poa, tx));
1574 	VERIFY3U(zio_wait(poa.pio), ==, 0);
1575 	ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1576 
1577 	/* change snapused */
1578 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1579 	    -poa.used, -poa.comp, -poa.uncomp, tx);
1580 
1581 	/* swap next's deadlist to our deadlist */
1582 	dsl_deadlist_close(&ds->ds_deadlist);
1583 	dsl_deadlist_close(&ds_next->ds_deadlist);
1584 	SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1585 	    ds->ds_phys->ds_deadlist_obj);
1586 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1587 	dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1588 	    ds_next->ds_phys->ds_deadlist_obj);
1589 }
1590 
1591 void
1592 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1593 {
1594 	struct dsl_ds_destroyarg *dsda = arg1;
1595 	dsl_dataset_t *ds = dsda->ds;
1596 	int err;
1597 	int after_branch_point = FALSE;
1598 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1599 	objset_t *mos = dp->dp_meta_objset;
1600 	dsl_dataset_t *ds_prev = NULL;
1601 	uint64_t obj;
1602 
1603 	ASSERT(ds->ds_owner);
1604 	ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1605 	ASSERT(ds->ds_prev == NULL ||
1606 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1607 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1608 
1609 	if (dsda->defer) {
1610 		ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1611 		if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) {
1612 			dmu_buf_will_dirty(ds->ds_dbuf, tx);
1613 			ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1614 			return;
1615 		}
1616 	}
1617 
1618 	/* signal any waiters that this dataset is going away */
1619 	mutex_enter(&ds->ds_lock);
1620 	ds->ds_owner = dsl_reaper;
1621 	cv_broadcast(&ds->ds_exclusive_cv);
1622 	mutex_exit(&ds->ds_lock);
1623 
1624 	if (ds->ds_objset) {
1625 		dmu_objset_evict(ds->ds_objset);
1626 		ds->ds_objset = NULL;
1627 	}
1628 
1629 	/* Remove our reservation */
1630 	if (ds->ds_reserved != 0) {
1631 		dsl_prop_setarg_t psa;
1632 		uint64_t value = 0;
1633 
1634 		dsl_prop_setarg_init_uint64(&psa, "refreservation",
1635 		    (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1636 		    &value);
1637 		psa.psa_effective_value = 0;	/* predict default value */
1638 
1639 		dsl_dataset_set_reservation_sync(ds, &psa, tx);
1640 		ASSERT3U(ds->ds_reserved, ==, 0);
1641 	}
1642 
1643 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1644 
1645 	dsl_scan_ds_destroyed(ds, tx);
1646 
1647 	obj = ds->ds_object;
1648 
1649 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1650 		if (ds->ds_prev) {
1651 			ds_prev = ds->ds_prev;
1652 		} else {
1653 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1654 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1655 		}
1656 		after_branch_point =
1657 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1658 
1659 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1660 		if (after_branch_point &&
1661 		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
1662 			remove_from_next_clones(ds_prev, obj, tx);
1663 			if (ds->ds_phys->ds_next_snap_obj != 0) {
1664 				VERIFY(0 == zap_add_int(mos,
1665 				    ds_prev->ds_phys->ds_next_clones_obj,
1666 				    ds->ds_phys->ds_next_snap_obj, tx));
1667 			}
1668 		}
1669 		if (after_branch_point &&
1670 		    ds->ds_phys->ds_next_snap_obj == 0) {
1671 			/* This clone is toast. */
1672 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1673 			ds_prev->ds_phys->ds_num_children--;
1674 
1675 			/*
1676 			 * If the clone's origin has no other clones, no
1677 			 * user holds, and has been marked for deferred
1678 			 * deletion, then we should have done the necessary
1679 			 * destroy setup for it.
1680 			 */
1681 			if (ds_prev->ds_phys->ds_num_children == 1 &&
1682 			    ds_prev->ds_userrefs == 0 &&
1683 			    DS_IS_DEFER_DESTROY(ds_prev)) {
1684 				ASSERT3P(dsda->rm_origin, !=, NULL);
1685 			} else {
1686 				ASSERT3P(dsda->rm_origin, ==, NULL);
1687 			}
1688 		} else if (!after_branch_point) {
1689 			ds_prev->ds_phys->ds_next_snap_obj =
1690 			    ds->ds_phys->ds_next_snap_obj;
1691 		}
1692 	}
1693 
1694 	if (dsl_dataset_is_snapshot(ds)) {
1695 		dsl_dataset_t *ds_next;
1696 		uint64_t old_unique;
1697 		uint64_t used = 0, comp = 0, uncomp = 0;
1698 
1699 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1700 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1701 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1702 
1703 		old_unique = ds_next->ds_phys->ds_unique_bytes;
1704 
1705 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1706 		ds_next->ds_phys->ds_prev_snap_obj =
1707 		    ds->ds_phys->ds_prev_snap_obj;
1708 		ds_next->ds_phys->ds_prev_snap_txg =
1709 		    ds->ds_phys->ds_prev_snap_txg;
1710 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1711 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1712 
1713 
1714 		if (ds_next->ds_deadlist.dl_oldfmt) {
1715 			process_old_deadlist(ds, ds_prev, ds_next,
1716 			    after_branch_point, tx);
1717 		} else {
1718 			/* Adjust prev's unique space. */
1719 			if (ds_prev && !after_branch_point) {
1720 				dsl_deadlist_space_range(&ds_next->ds_deadlist,
1721 				    ds_prev->ds_phys->ds_prev_snap_txg,
1722 				    ds->ds_phys->ds_prev_snap_txg,
1723 				    &used, &comp, &uncomp);
1724 				ds_prev->ds_phys->ds_unique_bytes += used;
1725 			}
1726 
1727 			/* Adjust snapused. */
1728 			dsl_deadlist_space_range(&ds_next->ds_deadlist,
1729 			    ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1730 			    &used, &comp, &uncomp);
1731 			dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1732 			    -used, -comp, -uncomp, tx);
1733 
1734 			/* Move blocks to be freed to pool's free list. */
1735 			dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1736 			    &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1737 			    tx);
1738 			dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1739 			    DD_USED_HEAD, used, comp, uncomp, tx);
1740 			dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx);
1741 
1742 			/* Merge our deadlist into next's and free it. */
1743 			dsl_deadlist_merge(&ds_next->ds_deadlist,
1744 			    ds->ds_phys->ds_deadlist_obj, tx);
1745 		}
1746 		dsl_deadlist_close(&ds->ds_deadlist);
1747 		dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1748 
1749 		/* Collapse range in clone heads */
1750 		dsl_dataset_remove_clones_key(ds,
1751 		    ds->ds_phys->ds_creation_txg, tx);
1752 
1753 		if (dsl_dataset_is_snapshot(ds_next)) {
1754 			dsl_dataset_t *ds_nextnext;
1755 
1756 			/*
1757 			 * Update next's unique to include blocks which
1758 			 * were previously shared by only this snapshot
1759 			 * and it.  Those blocks will be born after the
1760 			 * prev snap and before this snap, and will have
1761 			 * died after the next snap and before the one
1762 			 * after that (ie. be on the snap after next's
1763 			 * deadlist).
1764 			 */
1765 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1766 			    ds_next->ds_phys->ds_next_snap_obj,
1767 			    FTAG, &ds_nextnext));
1768 			dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1769 			    ds->ds_phys->ds_prev_snap_txg,
1770 			    ds->ds_phys->ds_creation_txg,
1771 			    &used, &comp, &uncomp);
1772 			ds_next->ds_phys->ds_unique_bytes += used;
1773 			dsl_dataset_rele(ds_nextnext, FTAG);
1774 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1775 
1776 			/* Collapse range in this head. */
1777 			dsl_dataset_t *hds;
1778 			VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1779 			    ds->ds_dir->dd_phys->dd_head_dataset_obj,
1780 			    FTAG, &hds));
1781 			dsl_deadlist_remove_key(&hds->ds_deadlist,
1782 			    ds->ds_phys->ds_creation_txg, tx);
1783 			dsl_dataset_rele(hds, FTAG);
1784 
1785 		} else {
1786 			ASSERT3P(ds_next->ds_prev, ==, ds);
1787 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1788 			ds_next->ds_prev = NULL;
1789 			if (ds_prev) {
1790 				VERIFY(0 == dsl_dataset_get_ref(dp,
1791 				    ds->ds_phys->ds_prev_snap_obj,
1792 				    ds_next, &ds_next->ds_prev));
1793 			}
1794 
1795 			dsl_dataset_recalc_head_uniq(ds_next);
1796 
1797 			/*
1798 			 * Reduce the amount of our unconsmed refreservation
1799 			 * being charged to our parent by the amount of
1800 			 * new unique data we have gained.
1801 			 */
1802 			if (old_unique < ds_next->ds_reserved) {
1803 				int64_t mrsdelta;
1804 				uint64_t new_unique =
1805 				    ds_next->ds_phys->ds_unique_bytes;
1806 
1807 				ASSERT(old_unique <= new_unique);
1808 				mrsdelta = MIN(new_unique - old_unique,
1809 				    ds_next->ds_reserved - old_unique);
1810 				dsl_dir_diduse_space(ds->ds_dir,
1811 				    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1812 			}
1813 		}
1814 		dsl_dataset_rele(ds_next, FTAG);
1815 	} else {
1816 		/*
1817 		 * There's no next snapshot, so this is a head dataset.
1818 		 * Destroy the deadlist.  Unless it's a clone, the
1819 		 * deadlist should be empty.  (If it's a clone, it's
1820 		 * safe to ignore the deadlist contents.)
1821 		 */
1822 		struct killarg ka;
1823 
1824 		dsl_deadlist_close(&ds->ds_deadlist);
1825 		dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1826 		ds->ds_phys->ds_deadlist_obj = 0;
1827 
1828 		/*
1829 		 * Free everything that we point to (that's born after
1830 		 * the previous snapshot, if we are a clone)
1831 		 *
1832 		 * NB: this should be very quick, because we already
1833 		 * freed all the objects in open context.
1834 		 */
1835 		ka.ds = ds;
1836 		ka.tx = tx;
1837 		err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1838 		    TRAVERSE_POST, kill_blkptr, &ka);
1839 		ASSERT3U(err, ==, 0);
1840 		ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1841 		    ds->ds_phys->ds_unique_bytes == 0);
1842 
1843 		if (ds->ds_prev != NULL) {
1844 			if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1845 				VERIFY3U(0, ==, zap_remove_int(mos,
1846 				    ds->ds_prev->ds_dir->dd_phys->dd_clones,
1847 				    ds->ds_object, tx));
1848 			}
1849 			dsl_dataset_rele(ds->ds_prev, ds);
1850 			ds->ds_prev = ds_prev = NULL;
1851 		}
1852 	}
1853 
1854 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1855 		/* Erase the link in the dir */
1856 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1857 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1858 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1859 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1860 		ASSERT(err == 0);
1861 	} else {
1862 		/* remove from snapshot namespace */
1863 		dsl_dataset_t *ds_head;
1864 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1865 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1866 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1867 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1868 #ifdef ZFS_DEBUG
1869 		{
1870 			uint64_t val;
1871 
1872 			err = dsl_dataset_snap_lookup(ds_head,
1873 			    ds->ds_snapname, &val);
1874 			ASSERT3U(err, ==, 0);
1875 			ASSERT3U(val, ==, obj);
1876 		}
1877 #endif
1878 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1879 		ASSERT(err == 0);
1880 		dsl_dataset_rele(ds_head, FTAG);
1881 	}
1882 
1883 	if (ds_prev && ds->ds_prev != ds_prev)
1884 		dsl_dataset_rele(ds_prev, FTAG);
1885 
1886 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1887 	spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1888 	    "dataset = %llu", ds->ds_object);
1889 
1890 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1891 		uint64_t count;
1892 		ASSERT(0 == zap_count(mos,
1893 		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1894 		VERIFY(0 == dmu_object_free(mos,
1895 		    ds->ds_phys->ds_next_clones_obj, tx));
1896 	}
1897 	if (ds->ds_phys->ds_props_obj != 0)
1898 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1899 	if (ds->ds_phys->ds_userrefs_obj != 0)
1900 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1901 	dsl_dir_close(ds->ds_dir, ds);
1902 	ds->ds_dir = NULL;
1903 	dsl_dataset_drain_refs(ds, tag);
1904 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1905 
1906 	if (dsda->rm_origin) {
1907 		/*
1908 		 * Remove the origin of the clone we just destroyed.
1909 		 */
1910 		struct dsl_ds_destroyarg ndsda = {0};
1911 
1912 		ndsda.ds = dsda->rm_origin;
1913 		dsl_dataset_destroy_sync(&ndsda, tag, tx);
1914 	}
1915 }
1916 
1917 static int
1918 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1919 {
1920 	uint64_t asize;
1921 
1922 	if (!dmu_tx_is_syncing(tx))
1923 		return (0);
1924 
1925 	/*
1926 	 * If there's an fs-only reservation, any blocks that might become
1927 	 * owned by the snapshot dataset must be accommodated by space
1928 	 * outside of the reservation.
1929 	 */
1930 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1931 	asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1932 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1933 		return (ENOSPC);
1934 
1935 	/*
1936 	 * Propogate any reserved space for this snapshot to other
1937 	 * snapshot checks in this sync group.
1938 	 */
1939 	if (asize > 0)
1940 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1941 
1942 	return (0);
1943 }
1944 
1945 int
1946 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1947 {
1948 	dsl_dataset_t *ds = arg1;
1949 	const char *snapname = arg2;
1950 	int err;
1951 	uint64_t value;
1952 
1953 	/*
1954 	 * We don't allow multiple snapshots of the same txg.  If there
1955 	 * is already one, try again.
1956 	 */
1957 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1958 		return (EAGAIN);
1959 
1960 	/*
1961 	 * Check for conflicting name snapshot name.
1962 	 */
1963 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1964 	if (err == 0)
1965 		return (EEXIST);
1966 	if (err != ENOENT)
1967 		return (err);
1968 
1969 	/*
1970 	 * Check that the dataset's name is not too long.  Name consists
1971 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1972 	 */
1973 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1974 		return (ENAMETOOLONG);
1975 
1976 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1977 	if (err)
1978 		return (err);
1979 
1980 	ds->ds_trysnap_txg = tx->tx_txg;
1981 	return (0);
1982 }
1983 
1984 void
1985 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1986 {
1987 	dsl_dataset_t *ds = arg1;
1988 	const char *snapname = arg2;
1989 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1990 	dmu_buf_t *dbuf;
1991 	dsl_dataset_phys_t *dsphys;
1992 	uint64_t dsobj, crtxg;
1993 	objset_t *mos = dp->dp_meta_objset;
1994 	int err;
1995 
1996 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1997 
1998 	/*
1999 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
2000 	 */
2001 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2002 		crtxg = 1;
2003 	else
2004 		crtxg = tx->tx_txg;
2005 
2006 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2007 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2008 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2009 	dmu_buf_will_dirty(dbuf, tx);
2010 	dsphys = dbuf->db_data;
2011 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
2012 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2013 	dsphys->ds_fsid_guid = unique_create();
2014 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2015 	    sizeof (dsphys->ds_guid));
2016 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2017 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2018 	dsphys->ds_next_snap_obj = ds->ds_object;
2019 	dsphys->ds_num_children = 1;
2020 	dsphys->ds_creation_time = gethrestime_sec();
2021 	dsphys->ds_creation_txg = crtxg;
2022 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2023 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
2024 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2025 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2026 	dsphys->ds_flags = ds->ds_phys->ds_flags;
2027 	dsphys->ds_bp = ds->ds_phys->ds_bp;
2028 	dmu_buf_rele(dbuf, FTAG);
2029 
2030 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2031 	if (ds->ds_prev) {
2032 		uint64_t next_clones_obj =
2033 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
2034 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2035 		    ds->ds_object ||
2036 		    ds->ds_prev->ds_phys->ds_num_children > 1);
2037 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2038 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2039 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2040 			    ds->ds_prev->ds_phys->ds_creation_txg);
2041 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2042 		} else if (next_clones_obj != 0) {
2043 			remove_from_next_clones(ds->ds_prev,
2044 			    dsphys->ds_next_snap_obj, tx);
2045 			VERIFY3U(0, ==, zap_add_int(mos,
2046 			    next_clones_obj, dsobj, tx));
2047 		}
2048 	}
2049 
2050 	/*
2051 	 * If we have a reference-reservation on this dataset, we will
2052 	 * need to increase the amount of refreservation being charged
2053 	 * since our unique space is going to zero.
2054 	 */
2055 	if (ds->ds_reserved) {
2056 		int64_t delta;
2057 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2058 		delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2059 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2060 		    delta, 0, 0, tx);
2061 	}
2062 
2063 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2064 	zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2065 	    ds->ds_dir->dd_myname, snapname, dsobj,
2066 	    ds->ds_phys->ds_prev_snap_txg);
2067 	ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2068 	    UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2069 	dsl_deadlist_close(&ds->ds_deadlist);
2070 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2071 	dsl_deadlist_add_key(&ds->ds_deadlist,
2072 	    ds->ds_phys->ds_prev_snap_txg, tx);
2073 
2074 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2075 	ds->ds_phys->ds_prev_snap_obj = dsobj;
2076 	ds->ds_phys->ds_prev_snap_txg = crtxg;
2077 	ds->ds_phys->ds_unique_bytes = 0;
2078 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2079 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2080 
2081 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2082 	    snapname, 8, 1, &dsobj, tx);
2083 	ASSERT(err == 0);
2084 
2085 	if (ds->ds_prev)
2086 		dsl_dataset_drop_ref(ds->ds_prev, ds);
2087 	VERIFY(0 == dsl_dataset_get_ref(dp,
2088 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2089 
2090 	dsl_scan_ds_snapshotted(ds, tx);
2091 
2092 	dsl_dir_snap_cmtime_update(ds->ds_dir);
2093 
2094 	spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2095 	    "dataset = %llu", dsobj);
2096 }
2097 
2098 void
2099 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2100 {
2101 	ASSERT(dmu_tx_is_syncing(tx));
2102 	ASSERT(ds->ds_objset != NULL);
2103 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2104 
2105 	/*
2106 	 * in case we had to change ds_fsid_guid when we opened it,
2107 	 * sync it out now.
2108 	 */
2109 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2110 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2111 
2112 	dsl_dir_dirty(ds->ds_dir, tx);
2113 	dmu_objset_sync(ds->ds_objset, zio, tx);
2114 }
2115 
2116 void
2117 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2118 {
2119 	uint64_t refd, avail, uobjs, aobjs;
2120 
2121 	dsl_dir_stats(ds->ds_dir, nv);
2122 
2123 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2124 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2125 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2126 
2127 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2128 	    ds->ds_phys->ds_creation_time);
2129 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2130 	    ds->ds_phys->ds_creation_txg);
2131 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2132 	    ds->ds_quota);
2133 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2134 	    ds->ds_reserved);
2135 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2136 	    ds->ds_phys->ds_guid);
2137 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2138 	    ds->ds_phys->ds_unique_bytes);
2139 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2140 	    ds->ds_object);
2141 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2142 	    ds->ds_userrefs);
2143 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2144 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2145 
2146 	if (ds->ds_phys->ds_next_snap_obj) {
2147 		/*
2148 		 * This is a snapshot; override the dd's space used with
2149 		 * our unique space and compression ratio.
2150 		 */
2151 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2152 		    ds->ds_phys->ds_unique_bytes);
2153 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2154 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2155 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
2156 		    ds->ds_phys->ds_compressed_bytes));
2157 	}
2158 }
2159 
2160 void
2161 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2162 {
2163 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2164 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2165 	stat->dds_guid = ds->ds_phys->ds_guid;
2166 	if (ds->ds_phys->ds_next_snap_obj) {
2167 		stat->dds_is_snapshot = B_TRUE;
2168 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2169 	} else {
2170 		stat->dds_is_snapshot = B_FALSE;
2171 		stat->dds_num_clones = 0;
2172 	}
2173 
2174 	/* clone origin is really a dsl_dir thing... */
2175 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2176 	if (dsl_dir_is_clone(ds->ds_dir)) {
2177 		dsl_dataset_t *ods;
2178 
2179 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2180 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2181 		dsl_dataset_name(ods, stat->dds_origin);
2182 		dsl_dataset_drop_ref(ods, FTAG);
2183 	} else {
2184 		stat->dds_origin[0] = '\0';
2185 	}
2186 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2187 }
2188 
2189 uint64_t
2190 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2191 {
2192 	return (ds->ds_fsid_guid);
2193 }
2194 
2195 void
2196 dsl_dataset_space(dsl_dataset_t *ds,
2197     uint64_t *refdbytesp, uint64_t *availbytesp,
2198     uint64_t *usedobjsp, uint64_t *availobjsp)
2199 {
2200 	*refdbytesp = ds->ds_phys->ds_used_bytes;
2201 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2202 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2203 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2204 	if (ds->ds_quota != 0) {
2205 		/*
2206 		 * Adjust available bytes according to refquota
2207 		 */
2208 		if (*refdbytesp < ds->ds_quota)
2209 			*availbytesp = MIN(*availbytesp,
2210 			    ds->ds_quota - *refdbytesp);
2211 		else
2212 			*availbytesp = 0;
2213 	}
2214 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2215 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2216 }
2217 
2218 boolean_t
2219 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2220 {
2221 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2222 
2223 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2224 	    dsl_pool_sync_context(dp));
2225 	if (ds->ds_prev == NULL)
2226 		return (B_FALSE);
2227 	if (ds->ds_phys->ds_bp.blk_birth >
2228 	    ds->ds_prev->ds_phys->ds_creation_txg)
2229 		return (B_TRUE);
2230 	return (B_FALSE);
2231 }
2232 
2233 /* ARGSUSED */
2234 static int
2235 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2236 {
2237 	dsl_dataset_t *ds = arg1;
2238 	char *newsnapname = arg2;
2239 	dsl_dir_t *dd = ds->ds_dir;
2240 	dsl_dataset_t *hds;
2241 	uint64_t val;
2242 	int err;
2243 
2244 	err = dsl_dataset_hold_obj(dd->dd_pool,
2245 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2246 	if (err)
2247 		return (err);
2248 
2249 	/* new name better not be in use */
2250 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2251 	dsl_dataset_rele(hds, FTAG);
2252 
2253 	if (err == 0)
2254 		err = EEXIST;
2255 	else if (err == ENOENT)
2256 		err = 0;
2257 
2258 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2259 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2260 		err = ENAMETOOLONG;
2261 
2262 	return (err);
2263 }
2264 
2265 static void
2266 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2267 {
2268 	dsl_dataset_t *ds = arg1;
2269 	const char *newsnapname = arg2;
2270 	dsl_dir_t *dd = ds->ds_dir;
2271 	objset_t *mos = dd->dd_pool->dp_meta_objset;
2272 	dsl_dataset_t *hds;
2273 	int err;
2274 
2275 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2276 
2277 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2278 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2279 
2280 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2281 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2282 	ASSERT3U(err, ==, 0);
2283 	mutex_enter(&ds->ds_lock);
2284 	(void) strcpy(ds->ds_snapname, newsnapname);
2285 	mutex_exit(&ds->ds_lock);
2286 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2287 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2288 	ASSERT3U(err, ==, 0);
2289 
2290 	spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2291 	    "dataset = %llu", ds->ds_object);
2292 	dsl_dataset_rele(hds, FTAG);
2293 }
2294 
2295 struct renamesnaparg {
2296 	dsl_sync_task_group_t *dstg;
2297 	char failed[MAXPATHLEN];
2298 	char *oldsnap;
2299 	char *newsnap;
2300 };
2301 
2302 static int
2303 dsl_snapshot_rename_one(const char *name, void *arg)
2304 {
2305 	struct renamesnaparg *ra = arg;
2306 	dsl_dataset_t *ds = NULL;
2307 	char *snapname;
2308 	int err;
2309 
2310 	snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2311 	(void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2312 
2313 	/*
2314 	 * For recursive snapshot renames the parent won't be changing
2315 	 * so we just pass name for both the to/from argument.
2316 	 */
2317 	err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2318 	if (err != 0) {
2319 		strfree(snapname);
2320 		return (err == ENOENT ? 0 : err);
2321 	}
2322 
2323 #ifdef _KERNEL
2324 	/*
2325 	 * For all filesystems undergoing rename, we'll need to unmount it.
2326 	 */
2327 	(void) zfs_unmount_snap(snapname, NULL);
2328 #endif
2329 	err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2330 	strfree(snapname);
2331 	if (err != 0)
2332 		return (err == ENOENT ? 0 : err);
2333 
2334 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2335 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2336 
2337 	return (0);
2338 }
2339 
2340 static int
2341 dsl_recursive_rename(char *oldname, const char *newname)
2342 {
2343 	int err;
2344 	struct renamesnaparg *ra;
2345 	dsl_sync_task_t *dst;
2346 	spa_t *spa;
2347 	char *cp, *fsname = spa_strdup(oldname);
2348 	int len = strlen(oldname) + 1;
2349 
2350 	/* truncate the snapshot name to get the fsname */
2351 	cp = strchr(fsname, '@');
2352 	*cp = '\0';
2353 
2354 	err = spa_open(fsname, &spa, FTAG);
2355 	if (err) {
2356 		kmem_free(fsname, len);
2357 		return (err);
2358 	}
2359 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2360 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2361 
2362 	ra->oldsnap = strchr(oldname, '@') + 1;
2363 	ra->newsnap = strchr(newname, '@') + 1;
2364 	*ra->failed = '\0';
2365 
2366 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2367 	    DS_FIND_CHILDREN);
2368 	kmem_free(fsname, len);
2369 
2370 	if (err == 0) {
2371 		err = dsl_sync_task_group_wait(ra->dstg);
2372 	}
2373 
2374 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2375 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2376 		dsl_dataset_t *ds = dst->dst_arg1;
2377 		if (dst->dst_err) {
2378 			dsl_dir_name(ds->ds_dir, ra->failed);
2379 			(void) strlcat(ra->failed, "@", sizeof (ra->failed));
2380 			(void) strlcat(ra->failed, ra->newsnap,
2381 			    sizeof (ra->failed));
2382 		}
2383 		dsl_dataset_rele(ds, ra->dstg);
2384 	}
2385 
2386 	if (err)
2387 		(void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2388 
2389 	dsl_sync_task_group_destroy(ra->dstg);
2390 	kmem_free(ra, sizeof (struct renamesnaparg));
2391 	spa_close(spa, FTAG);
2392 	return (err);
2393 }
2394 
2395 static int
2396 dsl_valid_rename(const char *oldname, void *arg)
2397 {
2398 	int delta = *(int *)arg;
2399 
2400 	if (strlen(oldname) + delta >= MAXNAMELEN)
2401 		return (ENAMETOOLONG);
2402 
2403 	return (0);
2404 }
2405 
2406 #pragma weak dmu_objset_rename = dsl_dataset_rename
2407 int
2408 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2409 {
2410 	dsl_dir_t *dd;
2411 	dsl_dataset_t *ds;
2412 	const char *tail;
2413 	int err;
2414 
2415 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2416 	if (err)
2417 		return (err);
2418 
2419 	if (tail == NULL) {
2420 		int delta = strlen(newname) - strlen(oldname);
2421 
2422 		/* if we're growing, validate child name lengths */
2423 		if (delta > 0)
2424 			err = dmu_objset_find(oldname, dsl_valid_rename,
2425 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2426 
2427 		if (err == 0)
2428 			err = dsl_dir_rename(dd, newname);
2429 		dsl_dir_close(dd, FTAG);
2430 		return (err);
2431 	}
2432 
2433 	if (tail[0] != '@') {
2434 		/* the name ended in a nonexistent component */
2435 		dsl_dir_close(dd, FTAG);
2436 		return (ENOENT);
2437 	}
2438 
2439 	dsl_dir_close(dd, FTAG);
2440 
2441 	/* new name must be snapshot in same filesystem */
2442 	tail = strchr(newname, '@');
2443 	if (tail == NULL)
2444 		return (EINVAL);
2445 	tail++;
2446 	if (strncmp(oldname, newname, tail - newname) != 0)
2447 		return (EXDEV);
2448 
2449 	if (recursive) {
2450 		err = dsl_recursive_rename(oldname, newname);
2451 	} else {
2452 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2453 		if (err)
2454 			return (err);
2455 
2456 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2457 		    dsl_dataset_snapshot_rename_check,
2458 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2459 
2460 		dsl_dataset_rele(ds, FTAG);
2461 	}
2462 
2463 	return (err);
2464 }
2465 
2466 struct promotenode {
2467 	list_node_t link;
2468 	dsl_dataset_t *ds;
2469 };
2470 
2471 struct promotearg {
2472 	list_t shared_snaps, origin_snaps, clone_snaps;
2473 	dsl_dataset_t *origin_origin;
2474 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2475 	char *err_ds;
2476 };
2477 
2478 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2479 static boolean_t snaplist_unstable(list_t *l);
2480 
2481 static int
2482 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2483 {
2484 	dsl_dataset_t *hds = arg1;
2485 	struct promotearg *pa = arg2;
2486 	struct promotenode *snap = list_head(&pa->shared_snaps);
2487 	dsl_dataset_t *origin_ds = snap->ds;
2488 	int err;
2489 	uint64_t unused;
2490 
2491 	/* Check that it is a real clone */
2492 	if (!dsl_dir_is_clone(hds->ds_dir))
2493 		return (EINVAL);
2494 
2495 	/* Since this is so expensive, don't do the preliminary check */
2496 	if (!dmu_tx_is_syncing(tx))
2497 		return (0);
2498 
2499 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2500 		return (EXDEV);
2501 
2502 	/* compute origin's new unique space */
2503 	snap = list_tail(&pa->clone_snaps);
2504 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2505 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2506 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2507 	    &pa->unique, &unused, &unused);
2508 
2509 	/*
2510 	 * Walk the snapshots that we are moving
2511 	 *
2512 	 * Compute space to transfer.  Consider the incremental changes
2513 	 * to used for each snapshot:
2514 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2515 	 * So each snapshot gave birth to:
2516 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2517 	 * So a sequence would look like:
2518 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2519 	 * Which simplifies to:
2520 	 * uN + kN + kN-1 + ... + k1 + k0
2521 	 * Note however, if we stop before we reach the ORIGIN we get:
2522 	 * uN + kN + kN-1 + ... + kM - uM-1
2523 	 */
2524 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2525 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2526 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2527 	for (snap = list_head(&pa->shared_snaps); snap;
2528 	    snap = list_next(&pa->shared_snaps, snap)) {
2529 		uint64_t val, dlused, dlcomp, dluncomp;
2530 		dsl_dataset_t *ds = snap->ds;
2531 
2532 		/* Check that the snapshot name does not conflict */
2533 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2534 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2535 		if (err == 0) {
2536 			err = EEXIST;
2537 			goto out;
2538 		}
2539 		if (err != ENOENT)
2540 			goto out;
2541 
2542 		/* The very first snapshot does not have a deadlist */
2543 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2544 			continue;
2545 
2546 		dsl_deadlist_space(&ds->ds_deadlist,
2547 		    &dlused, &dlcomp, &dluncomp);
2548 		pa->used += dlused;
2549 		pa->comp += dlcomp;
2550 		pa->uncomp += dluncomp;
2551 	}
2552 
2553 	/*
2554 	 * If we are a clone of a clone then we never reached ORIGIN,
2555 	 * so we need to subtract out the clone origin's used space.
2556 	 */
2557 	if (pa->origin_origin) {
2558 		pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2559 		pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2560 		pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2561 	}
2562 
2563 	/* Check that there is enough space here */
2564 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2565 	    pa->used);
2566 	if (err)
2567 		return (err);
2568 
2569 	/*
2570 	 * Compute the amounts of space that will be used by snapshots
2571 	 * after the promotion (for both origin and clone).  For each,
2572 	 * it is the amount of space that will be on all of their
2573 	 * deadlists (that was not born before their new origin).
2574 	 */
2575 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2576 		uint64_t space;
2577 
2578 		/*
2579 		 * Note, typically this will not be a clone of a clone,
2580 		 * so dd_origin_txg will be < TXG_INITIAL, so
2581 		 * these snaplist_space() -> dsl_deadlist_space_range()
2582 		 * calls will be fast because they do not have to
2583 		 * iterate over all bps.
2584 		 */
2585 		snap = list_head(&pa->origin_snaps);
2586 		err = snaplist_space(&pa->shared_snaps,
2587 		    snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2588 		if (err)
2589 			return (err);
2590 
2591 		err = snaplist_space(&pa->clone_snaps,
2592 		    snap->ds->ds_dir->dd_origin_txg, &space);
2593 		if (err)
2594 			return (err);
2595 		pa->cloneusedsnap += space;
2596 	}
2597 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2598 		err = snaplist_space(&pa->origin_snaps,
2599 		    origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2600 		if (err)
2601 			return (err);
2602 	}
2603 
2604 	return (0);
2605 out:
2606 	pa->err_ds =  snap->ds->ds_snapname;
2607 	return (err);
2608 }
2609 
2610 static void
2611 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2612 {
2613 	dsl_dataset_t *hds = arg1;
2614 	struct promotearg *pa = arg2;
2615 	struct promotenode *snap = list_head(&pa->shared_snaps);
2616 	dsl_dataset_t *origin_ds = snap->ds;
2617 	dsl_dataset_t *origin_head;
2618 	dsl_dir_t *dd = hds->ds_dir;
2619 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2620 	dsl_dir_t *odd = NULL;
2621 	uint64_t oldnext_obj;
2622 	int64_t delta;
2623 
2624 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2625 
2626 	snap = list_head(&pa->origin_snaps);
2627 	origin_head = snap->ds;
2628 
2629 	/*
2630 	 * We need to explicitly open odd, since origin_ds's dd will be
2631 	 * changing.
2632 	 */
2633 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2634 	    NULL, FTAG, &odd));
2635 
2636 	/* change origin's next snap */
2637 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2638 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2639 	snap = list_tail(&pa->clone_snaps);
2640 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2641 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2642 
2643 	/* change the origin's next clone */
2644 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2645 		remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2646 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2647 		    origin_ds->ds_phys->ds_next_clones_obj,
2648 		    oldnext_obj, tx));
2649 	}
2650 
2651 	/* change origin */
2652 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2653 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2654 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2655 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2656 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2657 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2658 	origin_head->ds_dir->dd_origin_txg =
2659 	    origin_ds->ds_phys->ds_creation_txg;
2660 
2661 	/* change dd_clone entries */
2662 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2663 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2664 		    odd->dd_phys->dd_clones, hds->ds_object, tx));
2665 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2666 		    pa->origin_origin->ds_dir->dd_phys->dd_clones,
2667 		    hds->ds_object, tx));
2668 
2669 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2670 		    pa->origin_origin->ds_dir->dd_phys->dd_clones,
2671 		    origin_head->ds_object, tx));
2672 		if (dd->dd_phys->dd_clones == 0) {
2673 			dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2674 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2675 		}
2676 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2677 		    dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2678 
2679 	}
2680 
2681 	/* move snapshots to this dir */
2682 	for (snap = list_head(&pa->shared_snaps); snap;
2683 	    snap = list_next(&pa->shared_snaps, snap)) {
2684 		dsl_dataset_t *ds = snap->ds;
2685 
2686 		/* unregister props as dsl_dir is changing */
2687 		if (ds->ds_objset) {
2688 			dmu_objset_evict(ds->ds_objset);
2689 			ds->ds_objset = NULL;
2690 		}
2691 		/* move snap name entry */
2692 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2693 		VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2694 		    ds->ds_snapname, tx));
2695 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2696 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2697 		    8, 1, &ds->ds_object, tx));
2698 
2699 		/* change containing dsl_dir */
2700 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2701 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2702 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2703 		ASSERT3P(ds->ds_dir, ==, odd);
2704 		dsl_dir_close(ds->ds_dir, ds);
2705 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2706 		    NULL, ds, &ds->ds_dir));
2707 
2708 		/* move any clone references */
2709 		if (ds->ds_phys->ds_next_clones_obj &&
2710 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2711 			zap_cursor_t zc;
2712 			zap_attribute_t za;
2713 
2714 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2715 			    ds->ds_phys->ds_next_clones_obj);
2716 			    zap_cursor_retrieve(&zc, &za) == 0;
2717 			    zap_cursor_advance(&zc)) {
2718 				dsl_dataset_t *cnds;
2719 				uint64_t o;
2720 
2721 				if (za.za_first_integer == oldnext_obj) {
2722 					/*
2723 					 * We've already moved the
2724 					 * origin's reference.
2725 					 */
2726 					continue;
2727 				}
2728 
2729 				VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2730 				    za.za_first_integer, FTAG, &cnds));
2731 				o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2732 
2733 				VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2734 				    odd->dd_phys->dd_clones, o, tx), ==, 0);
2735 				VERIFY3U(zap_add_int(dp->dp_meta_objset,
2736 				    dd->dd_phys->dd_clones, o, tx), ==, 0);
2737 				dsl_dataset_rele(cnds, FTAG);
2738 			}
2739 			zap_cursor_fini(&zc);
2740 		}
2741 
2742 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2743 	}
2744 
2745 	/*
2746 	 * Change space accounting.
2747 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2748 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2749 	 * is true for each of {clone,origin} independently.
2750 	 */
2751 
2752 	delta = pa->cloneusedsnap -
2753 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2754 	ASSERT3S(delta, >=, 0);
2755 	ASSERT3U(pa->used, >=, delta);
2756 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2757 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2758 	    pa->used - delta, pa->comp, pa->uncomp, tx);
2759 
2760 	delta = pa->originusedsnap -
2761 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2762 	ASSERT3S(delta, <=, 0);
2763 	ASSERT3U(pa->used, >=, -delta);
2764 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2765 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2766 	    -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2767 
2768 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2769 
2770 	/* log history record */
2771 	spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2772 	    "dataset = %llu", hds->ds_object);
2773 
2774 	dsl_dir_close(odd, FTAG);
2775 }
2776 
2777 static char *snaplist_tag = "snaplist";
2778 /*
2779  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2780  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2781  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2782  * snapshots back to this dataset's origin.
2783  */
2784 static int
2785 snaplist_make(dsl_pool_t *dp, boolean_t own,
2786     uint64_t first_obj, uint64_t last_obj, list_t *l)
2787 {
2788 	uint64_t obj = last_obj;
2789 
2790 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2791 
2792 	list_create(l, sizeof (struct promotenode),
2793 	    offsetof(struct promotenode, link));
2794 
2795 	while (obj != first_obj) {
2796 		dsl_dataset_t *ds;
2797 		struct promotenode *snap;
2798 		int err;
2799 
2800 		if (own) {
2801 			err = dsl_dataset_own_obj(dp, obj,
2802 			    0, snaplist_tag, &ds);
2803 			if (err == 0)
2804 				dsl_dataset_make_exclusive(ds, snaplist_tag);
2805 		} else {
2806 			err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2807 		}
2808 		if (err == ENOENT) {
2809 			/* lost race with snapshot destroy */
2810 			struct promotenode *last = list_tail(l);
2811 			ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2812 			obj = last->ds->ds_phys->ds_prev_snap_obj;
2813 			continue;
2814 		} else if (err) {
2815 			return (err);
2816 		}
2817 
2818 		if (first_obj == 0)
2819 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2820 
2821 		snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2822 		snap->ds = ds;
2823 		list_insert_tail(l, snap);
2824 		obj = ds->ds_phys->ds_prev_snap_obj;
2825 	}
2826 
2827 	return (0);
2828 }
2829 
2830 static int
2831 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2832 {
2833 	struct promotenode *snap;
2834 
2835 	*spacep = 0;
2836 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2837 		uint64_t used, comp, uncomp;
2838 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2839 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2840 		*spacep += used;
2841 	}
2842 	return (0);
2843 }
2844 
2845 static void
2846 snaplist_destroy(list_t *l, boolean_t own)
2847 {
2848 	struct promotenode *snap;
2849 
2850 	if (!l || !list_link_active(&l->list_head))
2851 		return;
2852 
2853 	while ((snap = list_tail(l)) != NULL) {
2854 		list_remove(l, snap);
2855 		if (own)
2856 			dsl_dataset_disown(snap->ds, snaplist_tag);
2857 		else
2858 			dsl_dataset_rele(snap->ds, snaplist_tag);
2859 		kmem_free(snap, sizeof (struct promotenode));
2860 	}
2861 	list_destroy(l);
2862 }
2863 
2864 /*
2865  * Promote a clone.  Nomenclature note:
2866  * "clone" or "cds": the original clone which is being promoted
2867  * "origin" or "ods": the snapshot which is originally clone's origin
2868  * "origin head" or "ohds": the dataset which is the head
2869  * (filesystem/volume) for the origin
2870  * "origin origin": the origin of the origin's filesystem (typically
2871  * NULL, indicating that the clone is not a clone of a clone).
2872  */
2873 int
2874 dsl_dataset_promote(const char *name, char *conflsnap)
2875 {
2876 	dsl_dataset_t *ds;
2877 	dsl_dir_t *dd;
2878 	dsl_pool_t *dp;
2879 	dmu_object_info_t doi;
2880 	struct promotearg pa = { 0 };
2881 	struct promotenode *snap;
2882 	int err;
2883 
2884 	err = dsl_dataset_hold(name, FTAG, &ds);
2885 	if (err)
2886 		return (err);
2887 	dd = ds->ds_dir;
2888 	dp = dd->dd_pool;
2889 
2890 	err = dmu_object_info(dp->dp_meta_objset,
2891 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2892 	if (err) {
2893 		dsl_dataset_rele(ds, FTAG);
2894 		return (err);
2895 	}
2896 
2897 	if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2898 		dsl_dataset_rele(ds, FTAG);
2899 		return (EINVAL);
2900 	}
2901 
2902 	/*
2903 	 * We are going to inherit all the snapshots taken before our
2904 	 * origin (i.e., our new origin will be our parent's origin).
2905 	 * Take ownership of them so that we can rename them into our
2906 	 * namespace.
2907 	 */
2908 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2909 
2910 	err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2911 	    &pa.shared_snaps);
2912 	if (err != 0)
2913 		goto out;
2914 
2915 	err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2916 	if (err != 0)
2917 		goto out;
2918 
2919 	snap = list_head(&pa.shared_snaps);
2920 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2921 	err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2922 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2923 	if (err != 0)
2924 		goto out;
2925 
2926 	if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2927 		err = dsl_dataset_hold_obj(dp,
2928 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2929 		    FTAG, &pa.origin_origin);
2930 		if (err != 0)
2931 			goto out;
2932 	}
2933 
2934 out:
2935 	rw_exit(&dp->dp_config_rwlock);
2936 
2937 	/*
2938 	 * Add in 128x the snapnames zapobj size, since we will be moving
2939 	 * a bunch of snapnames to the promoted ds, and dirtying their
2940 	 * bonus buffers.
2941 	 */
2942 	if (err == 0) {
2943 		err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2944 		    dsl_dataset_promote_sync, ds, &pa,
2945 		    2 + 2 * doi.doi_physical_blocks_512);
2946 		if (err && pa.err_ds && conflsnap)
2947 			(void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
2948 	}
2949 
2950 	snaplist_destroy(&pa.shared_snaps, B_TRUE);
2951 	snaplist_destroy(&pa.clone_snaps, B_FALSE);
2952 	snaplist_destroy(&pa.origin_snaps, B_FALSE);
2953 	if (pa.origin_origin)
2954 		dsl_dataset_rele(pa.origin_origin, FTAG);
2955 	dsl_dataset_rele(ds, FTAG);
2956 	return (err);
2957 }
2958 
2959 struct cloneswaparg {
2960 	dsl_dataset_t *cds; /* clone dataset */
2961 	dsl_dataset_t *ohds; /* origin's head dataset */
2962 	boolean_t force;
2963 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2964 };
2965 
2966 /* ARGSUSED */
2967 static int
2968 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2969 {
2970 	struct cloneswaparg *csa = arg1;
2971 
2972 	/* they should both be heads */
2973 	if (dsl_dataset_is_snapshot(csa->cds) ||
2974 	    dsl_dataset_is_snapshot(csa->ohds))
2975 		return (EINVAL);
2976 
2977 	/* the branch point should be just before them */
2978 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2979 		return (EINVAL);
2980 
2981 	/* cds should be the clone (unless they are unrelated) */
2982 	if (csa->cds->ds_prev != NULL &&
2983 	    csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
2984 	    csa->ohds->ds_object !=
2985 	    csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
2986 		return (EINVAL);
2987 
2988 	/* the clone should be a child of the origin */
2989 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2990 		return (EINVAL);
2991 
2992 	/* ohds shouldn't be modified unless 'force' */
2993 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2994 		return (ETXTBSY);
2995 
2996 	/* adjust amount of any unconsumed refreservation */
2997 	csa->unused_refres_delta =
2998 	    (int64_t)MIN(csa->ohds->ds_reserved,
2999 	    csa->ohds->ds_phys->ds_unique_bytes) -
3000 	    (int64_t)MIN(csa->ohds->ds_reserved,
3001 	    csa->cds->ds_phys->ds_unique_bytes);
3002 
3003 	if (csa->unused_refres_delta > 0 &&
3004 	    csa->unused_refres_delta >
3005 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3006 		return (ENOSPC);
3007 
3008 	if (csa->ohds->ds_quota != 0 &&
3009 	    csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3010 		return (EDQUOT);
3011 
3012 	return (0);
3013 }
3014 
3015 /* ARGSUSED */
3016 static void
3017 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3018 {
3019 	struct cloneswaparg *csa = arg1;
3020 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3021 
3022 	ASSERT(csa->cds->ds_reserved == 0);
3023 	ASSERT(csa->ohds->ds_quota == 0 ||
3024 	    csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3025 
3026 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3027 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3028 
3029 	if (csa->cds->ds_objset != NULL) {
3030 		dmu_objset_evict(csa->cds->ds_objset);
3031 		csa->cds->ds_objset = NULL;
3032 	}
3033 
3034 	if (csa->ohds->ds_objset != NULL) {
3035 		dmu_objset_evict(csa->ohds->ds_objset);
3036 		csa->ohds->ds_objset = NULL;
3037 	}
3038 
3039 	/*
3040 	 * Reset origin's unique bytes, if it exists.
3041 	 */
3042 	if (csa->cds->ds_prev) {
3043 		dsl_dataset_t *origin = csa->cds->ds_prev;
3044 		uint64_t comp, uncomp;
3045 
3046 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
3047 		dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3048 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3049 		    &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3050 	}
3051 
3052 	/* swap blkptrs */
3053 	{
3054 		blkptr_t tmp;
3055 		tmp = csa->ohds->ds_phys->ds_bp;
3056 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3057 		csa->cds->ds_phys->ds_bp = tmp;
3058 	}
3059 
3060 	/* set dd_*_bytes */
3061 	{
3062 		int64_t dused, dcomp, duncomp;
3063 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
3064 		uint64_t odl_used, odl_comp, odl_uncomp;
3065 
3066 		ASSERT3U(csa->cds->ds_dir->dd_phys->
3067 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
3068 
3069 		dsl_deadlist_space(&csa->cds->ds_deadlist,
3070 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3071 		dsl_deadlist_space(&csa->ohds->ds_deadlist,
3072 		    &odl_used, &odl_comp, &odl_uncomp);
3073 
3074 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
3075 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
3076 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3077 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3078 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3079 		    cdl_uncomp -
3080 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3081 
3082 		dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3083 		    dused, dcomp, duncomp, tx);
3084 		dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3085 		    -dused, -dcomp, -duncomp, tx);
3086 
3087 		/*
3088 		 * The difference in the space used by snapshots is the
3089 		 * difference in snapshot space due to the head's
3090 		 * deadlist (since that's the only thing that's
3091 		 * changing that affects the snapused).
3092 		 */
3093 		dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3094 		    csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3095 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3096 		dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3097 		    csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3098 		    &odl_used, &odl_comp, &odl_uncomp);
3099 		dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3100 		    DD_USED_HEAD, DD_USED_SNAP, tx);
3101 	}
3102 
3103 	/* swap ds_*_bytes */
3104 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
3105 	    csa->cds->ds_phys->ds_used_bytes);
3106 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3107 	    csa->cds->ds_phys->ds_compressed_bytes);
3108 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3109 	    csa->cds->ds_phys->ds_uncompressed_bytes);
3110 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3111 	    csa->cds->ds_phys->ds_unique_bytes);
3112 
3113 	/* apply any parent delta for change in unconsumed refreservation */
3114 	dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3115 	    csa->unused_refres_delta, 0, 0, tx);
3116 
3117 	/*
3118 	 * Swap deadlists.
3119 	 */
3120 	dsl_deadlist_close(&csa->cds->ds_deadlist);
3121 	dsl_deadlist_close(&csa->ohds->ds_deadlist);
3122 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3123 	    csa->cds->ds_phys->ds_deadlist_obj);
3124 	dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3125 	    csa->cds->ds_phys->ds_deadlist_obj);
3126 	dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3127 	    csa->ohds->ds_phys->ds_deadlist_obj);
3128 
3129 	dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3130 }
3131 
3132 /*
3133  * Swap 'clone' with its origin head datasets.  Used at the end of "zfs
3134  * recv" into an existing fs to swizzle the file system to the new
3135  * version, and by "zfs rollback".  Can also be used to swap two
3136  * independent head datasets if neither has any snapshots.
3137  */
3138 int
3139 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3140     boolean_t force)
3141 {
3142 	struct cloneswaparg csa;
3143 	int error;
3144 
3145 	ASSERT(clone->ds_owner);
3146 	ASSERT(origin_head->ds_owner);
3147 retry:
3148 	/* Need exclusive access for the swap */
3149 	rw_enter(&clone->ds_rwlock, RW_WRITER);
3150 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3151 		rw_exit(&clone->ds_rwlock);
3152 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3153 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3154 			rw_exit(&origin_head->ds_rwlock);
3155 			goto retry;
3156 		}
3157 	}
3158 	csa.cds = clone;
3159 	csa.ohds = origin_head;
3160 	csa.force = force;
3161 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3162 	    dsl_dataset_clone_swap_check,
3163 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3164 	return (error);
3165 }
3166 
3167 /*
3168  * Given a pool name and a dataset object number in that pool,
3169  * return the name of that dataset.
3170  */
3171 int
3172 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3173 {
3174 	spa_t *spa;
3175 	dsl_pool_t *dp;
3176 	dsl_dataset_t *ds;
3177 	int error;
3178 
3179 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
3180 		return (error);
3181 	dp = spa_get_dsl(spa);
3182 	rw_enter(&dp->dp_config_rwlock, RW_READER);
3183 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3184 		dsl_dataset_name(ds, buf);
3185 		dsl_dataset_rele(ds, FTAG);
3186 	}
3187 	rw_exit(&dp->dp_config_rwlock);
3188 	spa_close(spa, FTAG);
3189 
3190 	return (error);
3191 }
3192 
3193 int
3194 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3195     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3196 {
3197 	int error = 0;
3198 
3199 	ASSERT3S(asize, >, 0);
3200 
3201 	/*
3202 	 * *ref_rsrv is the portion of asize that will come from any
3203 	 * unconsumed refreservation space.
3204 	 */
3205 	*ref_rsrv = 0;
3206 
3207 	mutex_enter(&ds->ds_lock);
3208 	/*
3209 	 * Make a space adjustment for reserved bytes.
3210 	 */
3211 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3212 		ASSERT3U(*used, >=,
3213 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3214 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3215 		*ref_rsrv =
3216 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3217 	}
3218 
3219 	if (!check_quota || ds->ds_quota == 0) {
3220 		mutex_exit(&ds->ds_lock);
3221 		return (0);
3222 	}
3223 	/*
3224 	 * If they are requesting more space, and our current estimate
3225 	 * is over quota, they get to try again unless the actual
3226 	 * on-disk is over quota and there are no pending changes (which
3227 	 * may free up space for us).
3228 	 */
3229 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3230 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3231 			error = ERESTART;
3232 		else
3233 			error = EDQUOT;
3234 	}
3235 	mutex_exit(&ds->ds_lock);
3236 
3237 	return (error);
3238 }
3239 
3240 /* ARGSUSED */
3241 static int
3242 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3243 {
3244 	dsl_dataset_t *ds = arg1;
3245 	dsl_prop_setarg_t *psa = arg2;
3246 	int err;
3247 
3248 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3249 		return (ENOTSUP);
3250 
3251 	if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3252 		return (err);
3253 
3254 	if (psa->psa_effective_value == 0)
3255 		return (0);
3256 
3257 	if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes ||
3258 	    psa->psa_effective_value < ds->ds_reserved)
3259 		return (ENOSPC);
3260 
3261 	return (0);
3262 }
3263 
3264 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3265 
3266 void
3267 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3268 {
3269 	dsl_dataset_t *ds = arg1;
3270 	dsl_prop_setarg_t *psa = arg2;
3271 	uint64_t effective_value = psa->psa_effective_value;
3272 
3273 	dsl_prop_set_sync(ds, psa, tx);
3274 	DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3275 
3276 	if (ds->ds_quota != effective_value) {
3277 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3278 		ds->ds_quota = effective_value;
3279 
3280 		spa_history_log_internal(LOG_DS_REFQUOTA,
3281 		    ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3282 		    (longlong_t)ds->ds_quota, ds->ds_object);
3283 	}
3284 }
3285 
3286 int
3287 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3288 {
3289 	dsl_dataset_t *ds;
3290 	dsl_prop_setarg_t psa;
3291 	int err;
3292 
3293 	dsl_prop_setarg_init_uint64(&psa, "refquota", source, &quota);
3294 
3295 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3296 	if (err)
3297 		return (err);
3298 
3299 	/*
3300 	 * If someone removes a file, then tries to set the quota, we
3301 	 * want to make sure the file freeing takes effect.
3302 	 */
3303 	txg_wait_open(ds->ds_dir->dd_pool, 0);
3304 
3305 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3306 	    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3307 	    ds, &psa, 0);
3308 
3309 	dsl_dataset_rele(ds, FTAG);
3310 	return (err);
3311 }
3312 
3313 static int
3314 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3315 {
3316 	dsl_dataset_t *ds = arg1;
3317 	dsl_prop_setarg_t *psa = arg2;
3318 	uint64_t effective_value;
3319 	uint64_t unique;
3320 	int err;
3321 
3322 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3323 	    SPA_VERSION_REFRESERVATION)
3324 		return (ENOTSUP);
3325 
3326 	if (dsl_dataset_is_snapshot(ds))
3327 		return (EINVAL);
3328 
3329 	if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3330 		return (err);
3331 
3332 	effective_value = psa->psa_effective_value;
3333 
3334 	/*
3335 	 * If we are doing the preliminary check in open context, the
3336 	 * space estimates may be inaccurate.
3337 	 */
3338 	if (!dmu_tx_is_syncing(tx))
3339 		return (0);
3340 
3341 	mutex_enter(&ds->ds_lock);
3342 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3343 		dsl_dataset_recalc_head_uniq(ds);
3344 	unique = ds->ds_phys->ds_unique_bytes;
3345 	mutex_exit(&ds->ds_lock);
3346 
3347 	if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3348 		uint64_t delta = MAX(unique, effective_value) -
3349 		    MAX(unique, ds->ds_reserved);
3350 
3351 		if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3352 			return (ENOSPC);
3353 		if (ds->ds_quota > 0 &&
3354 		    effective_value > ds->ds_quota)
3355 			return (ENOSPC);
3356 	}
3357 
3358 	return (0);
3359 }
3360 
3361 static void
3362 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3363 {
3364 	dsl_dataset_t *ds = arg1;
3365 	dsl_prop_setarg_t *psa = arg2;
3366 	uint64_t effective_value = psa->psa_effective_value;
3367 	uint64_t unique;
3368 	int64_t delta;
3369 
3370 	dsl_prop_set_sync(ds, psa, tx);
3371 	DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3372 
3373 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3374 
3375 	mutex_enter(&ds->ds_dir->dd_lock);
3376 	mutex_enter(&ds->ds_lock);
3377 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3378 	unique = ds->ds_phys->ds_unique_bytes;
3379 	delta = MAX(0, (int64_t)(effective_value - unique)) -
3380 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3381 	ds->ds_reserved = effective_value;
3382 	mutex_exit(&ds->ds_lock);
3383 
3384 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3385 	mutex_exit(&ds->ds_dir->dd_lock);
3386 
3387 	spa_history_log_internal(LOG_DS_REFRESERV,
3388 	    ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3389 	    (longlong_t)effective_value, ds->ds_object);
3390 }
3391 
3392 int
3393 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3394     uint64_t reservation)
3395 {
3396 	dsl_dataset_t *ds;
3397 	dsl_prop_setarg_t psa;
3398 	int err;
3399 
3400 	dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3401 	    &reservation);
3402 
3403 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3404 	if (err)
3405 		return (err);
3406 
3407 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3408 	    dsl_dataset_set_reservation_check,
3409 	    dsl_dataset_set_reservation_sync, ds, &psa, 0);
3410 
3411 	dsl_dataset_rele(ds, FTAG);
3412 	return (err);
3413 }
3414 
3415 struct dsl_ds_holdarg {
3416 	dsl_sync_task_group_t *dstg;
3417 	char *htag;
3418 	char *snapname;
3419 	boolean_t recursive;
3420 	boolean_t gotone;
3421 	boolean_t temphold;
3422 	char failed[MAXPATHLEN];
3423 };
3424 
3425 typedef struct zfs_hold_cleanup_arg {
3426 	char dsname[MAXNAMELEN];
3427 	char snapname[MAXNAMELEN];
3428 	char htag[MAXNAMELEN];
3429 	boolean_t recursive;
3430 } zfs_hold_cleanup_arg_t;
3431 
3432 static void
3433 dsl_dataset_user_release_onexit(void *arg)
3434 {
3435 	zfs_hold_cleanup_arg_t *ca = arg;
3436 
3437 	(void) dsl_dataset_user_release(ca->dsname, ca->snapname,
3438 	    ca->htag, ca->recursive);
3439 	kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3440 }
3441 
3442 /*
3443  * The max length of a temporary tag prefix is the number of hex digits
3444  * required to express UINT64_MAX plus one for the hyphen.
3445  */
3446 #define	MAX_TAG_PREFIX_LEN	17
3447 
3448 static int
3449 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3450 {
3451 	dsl_dataset_t *ds = arg1;
3452 	struct dsl_ds_holdarg *ha = arg2;
3453 	char *htag = ha->htag;
3454 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3455 	int error = 0;
3456 
3457 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3458 		return (ENOTSUP);
3459 
3460 	if (!dsl_dataset_is_snapshot(ds))
3461 		return (EINVAL);
3462 
3463 	/* tags must be unique */
3464 	mutex_enter(&ds->ds_lock);
3465 	if (ds->ds_phys->ds_userrefs_obj) {
3466 		error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3467 		    8, 1, tx);
3468 		if (error == 0)
3469 			error = EEXIST;
3470 		else if (error == ENOENT)
3471 			error = 0;
3472 	}
3473 	mutex_exit(&ds->ds_lock);
3474 
3475 	if (error == 0 && ha->temphold &&
3476 	    strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3477 		error = E2BIG;
3478 
3479 	return (error);
3480 }
3481 
3482 static void
3483 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3484 {
3485 	dsl_dataset_t *ds = arg1;
3486 	struct dsl_ds_holdarg *ha = arg2;
3487 	char *htag = ha->htag;
3488 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
3489 	objset_t *mos = dp->dp_meta_objset;
3490 	uint64_t now = gethrestime_sec();
3491 	uint64_t zapobj;
3492 
3493 	mutex_enter(&ds->ds_lock);
3494 	if (ds->ds_phys->ds_userrefs_obj == 0) {
3495 		/*
3496 		 * This is the first user hold for this dataset.  Create
3497 		 * the userrefs zap object.
3498 		 */
3499 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3500 		zapobj = ds->ds_phys->ds_userrefs_obj =
3501 		    zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3502 	} else {
3503 		zapobj = ds->ds_phys->ds_userrefs_obj;
3504 	}
3505 	ds->ds_userrefs++;
3506 	mutex_exit(&ds->ds_lock);
3507 
3508 	VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3509 
3510 	if (ha->temphold) {
3511 		VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3512 		    htag, &now, tx));
3513 	}
3514 
3515 	spa_history_log_internal(LOG_DS_USER_HOLD,
3516 	    dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3517 	    (int)ha->temphold, ds->ds_object);
3518 }
3519 
3520 static int
3521 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3522 {
3523 	struct dsl_ds_holdarg *ha = arg;
3524 	dsl_dataset_t *ds;
3525 	int error;
3526 	char *name;
3527 
3528 	/* alloc a buffer to hold dsname@snapname plus terminating NULL */
3529 	name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3530 	error = dsl_dataset_hold(name, ha->dstg, &ds);
3531 	strfree(name);
3532 	if (error == 0) {
3533 		ha->gotone = B_TRUE;
3534 		dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3535 		    dsl_dataset_user_hold_sync, ds, ha, 0);
3536 	} else if (error == ENOENT && ha->recursive) {
3537 		error = 0;
3538 	} else {
3539 		(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3540 	}
3541 	return (error);
3542 }
3543 
3544 int
3545 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3546     boolean_t recursive, boolean_t temphold, int cleanup_fd)
3547 {
3548 	struct dsl_ds_holdarg *ha;
3549 	dsl_sync_task_t *dst;
3550 	spa_t *spa;
3551 	int error;
3552 
3553 	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3554 
3555 	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3556 
3557 	error = spa_open(dsname, &spa, FTAG);
3558 	if (error) {
3559 		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3560 		return (error);
3561 	}
3562 
3563 	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3564 	ha->htag = htag;
3565 	ha->snapname = snapname;
3566 	ha->recursive = recursive;
3567 	ha->temphold = temphold;
3568 
3569 	if (recursive) {
3570 		error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3571 		    ha, DS_FIND_CHILDREN);
3572 	} else {
3573 		error = dsl_dataset_user_hold_one(dsname, ha);
3574 	}
3575 	if (error == 0)
3576 		error = dsl_sync_task_group_wait(ha->dstg);
3577 
3578 	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3579 	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3580 		dsl_dataset_t *ds = dst->dst_arg1;
3581 
3582 		if (dst->dst_err) {
3583 			dsl_dataset_name(ds, ha->failed);
3584 			*strchr(ha->failed, '@') = '\0';
3585 		}
3586 		dsl_dataset_rele(ds, ha->dstg);
3587 	}
3588 
3589 	if (error == 0 && recursive && !ha->gotone)
3590 		error = ENOENT;
3591 
3592 	if (error)
3593 		(void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3594 
3595 	dsl_sync_task_group_destroy(ha->dstg);
3596 
3597 	/*
3598 	 * If this set of temporary holds is to be removed upon process exit,
3599 	 * register that action now.
3600 	 */
3601 	if (error == 0 && cleanup_fd != -1 && temphold) {
3602 		zfs_hold_cleanup_arg_t *ca;
3603 		uint64_t action_handle;
3604 
3605 		ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3606 		(void) strlcpy(ca->dsname, dsname, sizeof (ca->dsname));
3607 		(void) strlcpy(ca->snapname, snapname, sizeof (ca->snapname));
3608 		(void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3609 		ca->recursive = recursive;
3610 		ASSERT3U(0, ==, zfs_onexit_add_cb(cleanup_fd,
3611 		    dsl_dataset_user_release_onexit, ca, &action_handle));
3612 	}
3613 
3614 	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3615 	spa_close(spa, FTAG);
3616 	return (error);
3617 }
3618 
3619 struct dsl_ds_releasearg {
3620 	dsl_dataset_t *ds;
3621 	const char *htag;
3622 	boolean_t own;		/* do we own or just hold ds? */
3623 };
3624 
3625 static int
3626 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3627     boolean_t *might_destroy)
3628 {
3629 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3630 	uint64_t zapobj;
3631 	uint64_t tmp;
3632 	int error;
3633 
3634 	*might_destroy = B_FALSE;
3635 
3636 	mutex_enter(&ds->ds_lock);
3637 	zapobj = ds->ds_phys->ds_userrefs_obj;
3638 	if (zapobj == 0) {
3639 		/* The tag can't possibly exist */
3640 		mutex_exit(&ds->ds_lock);
3641 		return (ESRCH);
3642 	}
3643 
3644 	/* Make sure the tag exists */
3645 	error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3646 	if (error) {
3647 		mutex_exit(&ds->ds_lock);
3648 		if (error == ENOENT)
3649 			error = ESRCH;
3650 		return (error);
3651 	}
3652 
3653 	if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3654 	    DS_IS_DEFER_DESTROY(ds))
3655 		*might_destroy = B_TRUE;
3656 
3657 	mutex_exit(&ds->ds_lock);
3658 	return (0);
3659 }
3660 
3661 static int
3662 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3663 {
3664 	struct dsl_ds_releasearg *ra = arg1;
3665 	dsl_dataset_t *ds = ra->ds;
3666 	boolean_t might_destroy;
3667 	int error;
3668 
3669 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3670 		return (ENOTSUP);
3671 
3672 	error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3673 	if (error)
3674 		return (error);
3675 
3676 	if (might_destroy) {
3677 		struct dsl_ds_destroyarg dsda = {0};
3678 
3679 		if (dmu_tx_is_syncing(tx)) {
3680 			/*
3681 			 * If we're not prepared to remove the snapshot,
3682 			 * we can't allow the release to happen right now.
3683 			 */
3684 			if (!ra->own)
3685 				return (EBUSY);
3686 		}
3687 		dsda.ds = ds;
3688 		dsda.releasing = B_TRUE;
3689 		return (dsl_dataset_destroy_check(&dsda, tag, tx));
3690 	}
3691 
3692 	return (0);
3693 }
3694 
3695 static void
3696 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3697 {
3698 	struct dsl_ds_releasearg *ra = arg1;
3699 	dsl_dataset_t *ds = ra->ds;
3700 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
3701 	objset_t *mos = dp->dp_meta_objset;
3702 	uint64_t zapobj;
3703 	uint64_t dsobj = ds->ds_object;
3704 	uint64_t refs;
3705 	int error;
3706 
3707 	if (ds->ds_objset) {
3708 		dmu_objset_evict(ds->ds_objset);
3709 		ds->ds_objset = NULL;
3710 	}
3711 
3712 	mutex_enter(&ds->ds_lock);
3713 	ds->ds_userrefs--;
3714 	refs = ds->ds_userrefs;
3715 	mutex_exit(&ds->ds_lock);
3716 	error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3717 	VERIFY(error == 0 || error == ENOENT);
3718 	zapobj = ds->ds_phys->ds_userrefs_obj;
3719 	VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3720 	if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3721 	    DS_IS_DEFER_DESTROY(ds)) {
3722 		struct dsl_ds_destroyarg dsda = {0};
3723 
3724 		ASSERT(ra->own);
3725 		dsda.ds = ds;
3726 		dsda.releasing = B_TRUE;
3727 		/* We already did the destroy_check */
3728 		dsl_dataset_destroy_sync(&dsda, tag, tx);
3729 	}
3730 
3731 	spa_history_log_internal(LOG_DS_USER_RELEASE,
3732 	    dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3733 	    ra->htag, (longlong_t)refs, dsobj);
3734 }
3735 
3736 static int
3737 dsl_dataset_user_release_one(const char *dsname, void *arg)
3738 {
3739 	struct dsl_ds_holdarg *ha = arg;
3740 	struct dsl_ds_releasearg *ra;
3741 	dsl_dataset_t *ds;
3742 	int error;
3743 	void *dtag = ha->dstg;
3744 	char *name;
3745 	boolean_t own = B_FALSE;
3746 	boolean_t might_destroy;
3747 
3748 	/* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3749 	name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3750 	error = dsl_dataset_hold(name, dtag, &ds);
3751 	strfree(name);
3752 	if (error == ENOENT && ha->recursive)
3753 		return (0);
3754 	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3755 	if (error)
3756 		return (error);
3757 
3758 	ha->gotone = B_TRUE;
3759 
3760 	ASSERT(dsl_dataset_is_snapshot(ds));
3761 
3762 	error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3763 	if (error) {
3764 		dsl_dataset_rele(ds, dtag);
3765 		return (error);
3766 	}
3767 
3768 	if (might_destroy) {
3769 #ifdef _KERNEL
3770 		name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3771 		error = zfs_unmount_snap(name, NULL);
3772 		strfree(name);
3773 		if (error) {
3774 			dsl_dataset_rele(ds, dtag);
3775 			return (error);
3776 		}
3777 #endif
3778 		if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3779 			dsl_dataset_rele(ds, dtag);
3780 			return (EBUSY);
3781 		} else {
3782 			own = B_TRUE;
3783 			dsl_dataset_make_exclusive(ds, dtag);
3784 		}
3785 	}
3786 
3787 	ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3788 	ra->ds = ds;
3789 	ra->htag = ha->htag;
3790 	ra->own = own;
3791 	dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3792 	    dsl_dataset_user_release_sync, ra, dtag, 0);
3793 
3794 	return (0);
3795 }
3796 
3797 int
3798 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3799     boolean_t recursive)
3800 {
3801 	struct dsl_ds_holdarg *ha;
3802 	dsl_sync_task_t *dst;
3803 	spa_t *spa;
3804 	int error;
3805 
3806 top:
3807 	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3808 
3809 	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3810 
3811 	error = spa_open(dsname, &spa, FTAG);
3812 	if (error) {
3813 		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3814 		return (error);
3815 	}
3816 
3817 	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3818 	ha->htag = htag;
3819 	ha->snapname = snapname;
3820 	ha->recursive = recursive;
3821 	if (recursive) {
3822 		error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3823 		    ha, DS_FIND_CHILDREN);
3824 	} else {
3825 		error = dsl_dataset_user_release_one(dsname, ha);
3826 	}
3827 	if (error == 0)
3828 		error = dsl_sync_task_group_wait(ha->dstg);
3829 
3830 	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3831 	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3832 		struct dsl_ds_releasearg *ra = dst->dst_arg1;
3833 		dsl_dataset_t *ds = ra->ds;
3834 
3835 		if (dst->dst_err)
3836 			dsl_dataset_name(ds, ha->failed);
3837 
3838 		if (ra->own)
3839 			dsl_dataset_disown(ds, ha->dstg);
3840 		else
3841 			dsl_dataset_rele(ds, ha->dstg);
3842 
3843 		kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3844 	}
3845 
3846 	if (error == 0 && recursive && !ha->gotone)
3847 		error = ENOENT;
3848 
3849 	if (error && error != EBUSY)
3850 		(void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3851 
3852 	dsl_sync_task_group_destroy(ha->dstg);
3853 	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3854 	spa_close(spa, FTAG);
3855 
3856 	/*
3857 	 * We can get EBUSY if we were racing with deferred destroy and
3858 	 * dsl_dataset_user_release_check() hadn't done the necessary
3859 	 * open context setup.  We can also get EBUSY if we're racing
3860 	 * with destroy and that thread is the ds_owner.  Either way
3861 	 * the busy condition should be transient, and we should retry
3862 	 * the release operation.
3863 	 */
3864 	if (error == EBUSY)
3865 		goto top;
3866 
3867 	return (error);
3868 }
3869 
3870 /*
3871  * Called at spa_load time to release a stale temporary user hold.
3872  */
3873 int
3874 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag)
3875 {
3876 	dsl_dataset_t *ds;
3877 	char *snap;
3878 	char *name;
3879 	int namelen;
3880 	int error;
3881 
3882 	rw_enter(&dp->dp_config_rwlock, RW_READER);
3883 	error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
3884 	rw_exit(&dp->dp_config_rwlock);
3885 	if (error)
3886 		return (error);
3887 	namelen = dsl_dataset_namelen(ds)+1;
3888 	name = kmem_alloc(namelen, KM_SLEEP);
3889 	dsl_dataset_name(ds, name);
3890 	dsl_dataset_rele(ds, FTAG);
3891 
3892 	snap = strchr(name, '@');
3893 	*snap = '\0';
3894 	++snap;
3895 	return (dsl_dataset_user_release(name, snap, htag, B_FALSE));
3896 }
3897 
3898 int
3899 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
3900 {
3901 	dsl_dataset_t *ds;
3902 	int err;
3903 
3904 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3905 	if (err)
3906 		return (err);
3907 
3908 	VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
3909 	if (ds->ds_phys->ds_userrefs_obj != 0) {
3910 		zap_attribute_t *za;
3911 		zap_cursor_t zc;
3912 
3913 		za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
3914 		for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
3915 		    ds->ds_phys->ds_userrefs_obj);
3916 		    zap_cursor_retrieve(&zc, za) == 0;
3917 		    zap_cursor_advance(&zc)) {
3918 			VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
3919 			    za->za_first_integer));
3920 		}
3921 		zap_cursor_fini(&zc);
3922 		kmem_free(za, sizeof (zap_attribute_t));
3923 	}
3924 	dsl_dataset_rele(ds, FTAG);
3925 	return (0);
3926 }
3927 
3928 /*
3929  * Note, this fuction is used as the callback for dmu_objset_find().  We
3930  * always return 0 so that we will continue to find and process
3931  * inconsistent datasets, even if we encounter an error trying to
3932  * process one of them.
3933  */
3934 /* ARGSUSED */
3935 int
3936 dsl_destroy_inconsistent(const char *dsname, void *arg)
3937 {
3938 	dsl_dataset_t *ds;
3939 
3940 	if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
3941 		if (DS_IS_INCONSISTENT(ds))
3942 			(void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
3943 		else
3944 			dsl_dataset_disown(ds, FTAG);
3945 	}
3946 	return (0);
3947 }
3948