xref: /titanic_44/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision fcdc8680ec9ba9c5b19abb1bfa18a8eaa0d06416)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/dmu_objset.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_dir.h>
29 #include <sys/dsl_prop.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dmu_traverse.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/arc.h>
34 #include <sys/zio.h>
35 #include <sys/zap.h>
36 #include <sys/unique.h>
37 #include <sys/zfs_context.h>
38 #include <sys/zfs_ioctl.h>
39 #include <sys/spa.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/sunddi.h>
42 #include <sys/zvol.h>
43 
44 static char *dsl_reaper = "the grim reaper";
45 
46 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
47 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
48 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
49 
50 #define	DS_REF_MAX	(1ULL << 62)
51 
52 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
53 
54 #define	DSL_DATASET_IS_DESTROYED(ds)	((ds)->ds_owner == dsl_reaper)
55 
56 
57 /*
58  * Figure out how much of this delta should be propogated to the dsl_dir
59  * layer.  If there's a refreservation, that space has already been
60  * partially accounted for in our ancestors.
61  */
62 static int64_t
63 parent_delta(dsl_dataset_t *ds, int64_t delta)
64 {
65 	uint64_t old_bytes, new_bytes;
66 
67 	if (ds->ds_reserved == 0)
68 		return (delta);
69 
70 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
71 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
72 
73 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
74 	return (new_bytes - old_bytes);
75 }
76 
77 void
78 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
79 {
80 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
81 	int compressed = BP_GET_PSIZE(bp);
82 	int uncompressed = BP_GET_UCSIZE(bp);
83 	int64_t delta;
84 
85 	dprintf_bp(bp, "born, ds=%p\n", ds);
86 
87 	ASSERT(dmu_tx_is_syncing(tx));
88 	/* It could have been compressed away to nothing */
89 	if (BP_IS_HOLE(bp))
90 		return;
91 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
92 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
93 	if (ds == NULL) {
94 		/*
95 		 * Account for the meta-objset space in its placeholder
96 		 * dsl_dir.
97 		 */
98 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
99 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
100 		    used, compressed, uncompressed, tx);
101 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
102 		return;
103 	}
104 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
105 	mutex_enter(&ds->ds_dir->dd_lock);
106 	mutex_enter(&ds->ds_lock);
107 	delta = parent_delta(ds, used);
108 	ds->ds_phys->ds_used_bytes += used;
109 	ds->ds_phys->ds_compressed_bytes += compressed;
110 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
111 	ds->ds_phys->ds_unique_bytes += used;
112 	mutex_exit(&ds->ds_lock);
113 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
114 	    compressed, uncompressed, tx);
115 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
116 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
117 	mutex_exit(&ds->ds_dir->dd_lock);
118 }
119 
120 int
121 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
122     dmu_tx_t *tx)
123 {
124 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
125 	int compressed = BP_GET_PSIZE(bp);
126 	int uncompressed = BP_GET_UCSIZE(bp);
127 
128 	ASSERT(pio != NULL);
129 	ASSERT(dmu_tx_is_syncing(tx));
130 	/* No block pointer => nothing to free */
131 	if (BP_IS_HOLE(bp))
132 		return (0);
133 
134 	ASSERT(used > 0);
135 	if (ds == NULL) {
136 		int err;
137 		/*
138 		 * Account for the meta-objset space in its placeholder
139 		 * dataset.
140 		 */
141 		err = dsl_free(pio, tx->tx_pool,
142 		    tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
143 		ASSERT(err == 0);
144 
145 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
146 		    -used, -compressed, -uncompressed, tx);
147 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
148 		return (used);
149 	}
150 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
151 
152 	ASSERT(!dsl_dataset_is_snapshot(ds));
153 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
154 
155 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
156 		int err;
157 		int64_t delta;
158 
159 		dprintf_bp(bp, "freeing: %s", "");
160 		err = dsl_free(pio, tx->tx_pool,
161 		    tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
162 		ASSERT(err == 0);
163 
164 		mutex_enter(&ds->ds_dir->dd_lock);
165 		mutex_enter(&ds->ds_lock);
166 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
167 		    !DS_UNIQUE_IS_ACCURATE(ds));
168 		delta = parent_delta(ds, -used);
169 		ds->ds_phys->ds_unique_bytes -= used;
170 		mutex_exit(&ds->ds_lock);
171 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
172 		    delta, -compressed, -uncompressed, tx);
173 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
174 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
175 		mutex_exit(&ds->ds_dir->dd_lock);
176 	} else {
177 		dprintf_bp(bp, "putting on dead list: %s", "");
178 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
179 		ASSERT3U(ds->ds_prev->ds_object, ==,
180 		    ds->ds_phys->ds_prev_snap_obj);
181 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
182 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
183 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
184 		    ds->ds_object && bp->blk_birth >
185 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
186 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
187 			mutex_enter(&ds->ds_prev->ds_lock);
188 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
189 			mutex_exit(&ds->ds_prev->ds_lock);
190 		}
191 		if (bp->blk_birth > ds->ds_origin_txg) {
192 			dsl_dir_transfer_space(ds->ds_dir, used,
193 			    DD_USED_HEAD, DD_USED_SNAP, tx);
194 		}
195 	}
196 	mutex_enter(&ds->ds_lock);
197 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
198 	ds->ds_phys->ds_used_bytes -= used;
199 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
200 	ds->ds_phys->ds_compressed_bytes -= compressed;
201 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
202 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
203 	mutex_exit(&ds->ds_lock);
204 
205 	return (used);
206 }
207 
208 uint64_t
209 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
210 {
211 	uint64_t trysnap = 0;
212 
213 	if (ds == NULL)
214 		return (0);
215 	/*
216 	 * The snapshot creation could fail, but that would cause an
217 	 * incorrect FALSE return, which would only result in an
218 	 * overestimation of the amount of space that an operation would
219 	 * consume, which is OK.
220 	 *
221 	 * There's also a small window where we could miss a pending
222 	 * snapshot, because we could set the sync task in the quiescing
223 	 * phase.  So this should only be used as a guess.
224 	 */
225 	if (ds->ds_trysnap_txg >
226 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
227 		trysnap = ds->ds_trysnap_txg;
228 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
229 }
230 
231 boolean_t
232 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
233 {
234 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
235 }
236 
237 /* ARGSUSED */
238 static void
239 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
240 {
241 	dsl_dataset_t *ds = dsv;
242 
243 	ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
244 
245 	unique_remove(ds->ds_fsid_guid);
246 
247 	if (ds->ds_user_ptr != NULL)
248 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
249 
250 	if (ds->ds_prev) {
251 		dsl_dataset_drop_ref(ds->ds_prev, ds);
252 		ds->ds_prev = NULL;
253 	}
254 
255 	bplist_close(&ds->ds_deadlist);
256 	if (ds->ds_dir)
257 		dsl_dir_close(ds->ds_dir, ds);
258 
259 	ASSERT(!list_link_active(&ds->ds_synced_link));
260 
261 	mutex_destroy(&ds->ds_lock);
262 	mutex_destroy(&ds->ds_recvlock);
263 	mutex_destroy(&ds->ds_opening_lock);
264 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
265 	rw_destroy(&ds->ds_rwlock);
266 	cv_destroy(&ds->ds_exclusive_cv);
267 
268 	kmem_free(ds, sizeof (dsl_dataset_t));
269 }
270 
271 static int
272 dsl_dataset_get_snapname(dsl_dataset_t *ds)
273 {
274 	dsl_dataset_phys_t *headphys;
275 	int err;
276 	dmu_buf_t *headdbuf;
277 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
278 	objset_t *mos = dp->dp_meta_objset;
279 
280 	if (ds->ds_snapname[0])
281 		return (0);
282 	if (ds->ds_phys->ds_next_snap_obj == 0)
283 		return (0);
284 
285 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
286 	    FTAG, &headdbuf);
287 	if (err)
288 		return (err);
289 	headphys = headdbuf->db_data;
290 	err = zap_value_search(dp->dp_meta_objset,
291 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
292 	dmu_buf_rele(headdbuf, FTAG);
293 	return (err);
294 }
295 
296 static int
297 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
298 {
299 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
300 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
301 	matchtype_t mt;
302 	int err;
303 
304 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
305 		mt = MT_FIRST;
306 	else
307 		mt = MT_EXACT;
308 
309 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
310 	    value, mt, NULL, 0, NULL);
311 	if (err == ENOTSUP && mt == MT_FIRST)
312 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
313 	return (err);
314 }
315 
316 static int
317 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
318 {
319 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
320 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
321 	matchtype_t mt;
322 	int err;
323 
324 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
325 		mt = MT_FIRST;
326 	else
327 		mt = MT_EXACT;
328 
329 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
330 	if (err == ENOTSUP && mt == MT_FIRST)
331 		err = zap_remove(mos, snapobj, name, tx);
332 	return (err);
333 }
334 
335 static int
336 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
337     dsl_dataset_t **dsp)
338 {
339 	objset_t *mos = dp->dp_meta_objset;
340 	dmu_buf_t *dbuf;
341 	dsl_dataset_t *ds;
342 	int err;
343 
344 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
345 	    dsl_pool_sync_context(dp));
346 
347 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
348 	if (err)
349 		return (err);
350 	ds = dmu_buf_get_user(dbuf);
351 	if (ds == NULL) {
352 		dsl_dataset_t *winner;
353 
354 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
355 		ds->ds_dbuf = dbuf;
356 		ds->ds_object = dsobj;
357 		ds->ds_phys = dbuf->db_data;
358 
359 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
360 		mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
361 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
362 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
363 		    NULL);
364 		rw_init(&ds->ds_rwlock, 0, 0, 0);
365 		cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
366 
367 		err = bplist_open(&ds->ds_deadlist,
368 		    mos, ds->ds_phys->ds_deadlist_obj);
369 		if (err == 0) {
370 			err = dsl_dir_open_obj(dp,
371 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
372 		}
373 		if (err) {
374 			/*
375 			 * we don't really need to close the blist if we
376 			 * just opened it.
377 			 */
378 			mutex_destroy(&ds->ds_lock);
379 			mutex_destroy(&ds->ds_recvlock);
380 			mutex_destroy(&ds->ds_opening_lock);
381 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
382 			rw_destroy(&ds->ds_rwlock);
383 			cv_destroy(&ds->ds_exclusive_cv);
384 			kmem_free(ds, sizeof (dsl_dataset_t));
385 			dmu_buf_rele(dbuf, tag);
386 			return (err);
387 		}
388 
389 		if (!dsl_dataset_is_snapshot(ds)) {
390 			ds->ds_snapname[0] = '\0';
391 			if (ds->ds_phys->ds_prev_snap_obj) {
392 				err = dsl_dataset_get_ref(dp,
393 				    ds->ds_phys->ds_prev_snap_obj,
394 				    ds, &ds->ds_prev);
395 			}
396 
397 			if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) {
398 				dsl_dataset_t *origin;
399 
400 				err = dsl_dataset_hold_obj(dp,
401 				    ds->ds_dir->dd_phys->dd_origin_obj,
402 				    FTAG, &origin);
403 				if (err == 0) {
404 					ds->ds_origin_txg =
405 					    origin->ds_phys->ds_creation_txg;
406 					dsl_dataset_rele(origin, FTAG);
407 				}
408 			}
409 		} else {
410 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
411 				err = dsl_dataset_get_snapname(ds);
412 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
413 				err = zap_count(
414 				    ds->ds_dir->dd_pool->dp_meta_objset,
415 				    ds->ds_phys->ds_userrefs_obj,
416 				    &ds->ds_userrefs);
417 			}
418 		}
419 
420 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
421 			/*
422 			 * In sync context, we're called with either no lock
423 			 * or with the write lock.  If we're not syncing,
424 			 * we're always called with the read lock held.
425 			 */
426 			boolean_t need_lock =
427 			    !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
428 			    dsl_pool_sync_context(dp);
429 
430 			if (need_lock)
431 				rw_enter(&dp->dp_config_rwlock, RW_READER);
432 
433 			err = dsl_prop_get_ds(ds,
434 			    "refreservation", sizeof (uint64_t), 1,
435 			    &ds->ds_reserved, NULL);
436 			if (err == 0) {
437 				err = dsl_prop_get_ds(ds,
438 				    "refquota", sizeof (uint64_t), 1,
439 				    &ds->ds_quota, NULL);
440 			}
441 
442 			if (need_lock)
443 				rw_exit(&dp->dp_config_rwlock);
444 		} else {
445 			ds->ds_reserved = ds->ds_quota = 0;
446 		}
447 
448 		if (err == 0) {
449 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
450 			    dsl_dataset_evict);
451 		}
452 		if (err || winner) {
453 			bplist_close(&ds->ds_deadlist);
454 			if (ds->ds_prev)
455 				dsl_dataset_drop_ref(ds->ds_prev, ds);
456 			dsl_dir_close(ds->ds_dir, ds);
457 			mutex_destroy(&ds->ds_lock);
458 			mutex_destroy(&ds->ds_recvlock);
459 			mutex_destroy(&ds->ds_opening_lock);
460 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
461 			rw_destroy(&ds->ds_rwlock);
462 			cv_destroy(&ds->ds_exclusive_cv);
463 			kmem_free(ds, sizeof (dsl_dataset_t));
464 			if (err) {
465 				dmu_buf_rele(dbuf, tag);
466 				return (err);
467 			}
468 			ds = winner;
469 		} else {
470 			ds->ds_fsid_guid =
471 			    unique_insert(ds->ds_phys->ds_fsid_guid);
472 		}
473 	}
474 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
475 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
476 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
477 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
478 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
479 	mutex_enter(&ds->ds_lock);
480 	if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
481 		mutex_exit(&ds->ds_lock);
482 		dmu_buf_rele(ds->ds_dbuf, tag);
483 		return (ENOENT);
484 	}
485 	mutex_exit(&ds->ds_lock);
486 	*dsp = ds;
487 	return (0);
488 }
489 
490 static int
491 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
492 {
493 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
494 
495 	/*
496 	 * In syncing context we don't want the rwlock lock: there
497 	 * may be an existing writer waiting for sync phase to
498 	 * finish.  We don't need to worry about such writers, since
499 	 * sync phase is single-threaded, so the writer can't be
500 	 * doing anything while we are active.
501 	 */
502 	if (dsl_pool_sync_context(dp)) {
503 		ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
504 		return (0);
505 	}
506 
507 	/*
508 	 * Normal users will hold the ds_rwlock as a READER until they
509 	 * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
510 	 * drop their READER lock after they set the ds_owner field.
511 	 *
512 	 * If the dataset is being destroyed, the destroy thread will
513 	 * obtain a WRITER lock for exclusive access after it's done its
514 	 * open-context work and then change the ds_owner to
515 	 * dsl_reaper once destruction is assured.  So threads
516 	 * may block here temporarily, until the "destructability" of
517 	 * the dataset is determined.
518 	 */
519 	ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
520 	mutex_enter(&ds->ds_lock);
521 	while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
522 		rw_exit(&dp->dp_config_rwlock);
523 		cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
524 		if (DSL_DATASET_IS_DESTROYED(ds)) {
525 			mutex_exit(&ds->ds_lock);
526 			dsl_dataset_drop_ref(ds, tag);
527 			rw_enter(&dp->dp_config_rwlock, RW_READER);
528 			return (ENOENT);
529 		}
530 		rw_enter(&dp->dp_config_rwlock, RW_READER);
531 	}
532 	mutex_exit(&ds->ds_lock);
533 	return (0);
534 }
535 
536 int
537 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
538     dsl_dataset_t **dsp)
539 {
540 	int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
541 
542 	if (err)
543 		return (err);
544 	return (dsl_dataset_hold_ref(*dsp, tag));
545 }
546 
547 int
548 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, int flags, void *owner,
549     dsl_dataset_t **dsp)
550 {
551 	int err = dsl_dataset_hold_obj(dp, dsobj, owner, dsp);
552 
553 	ASSERT(DS_MODE_TYPE(flags) != DS_MODE_USER);
554 
555 	if (err)
556 		return (err);
557 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
558 		dsl_dataset_rele(*dsp, owner);
559 		*dsp = NULL;
560 		return (EBUSY);
561 	}
562 	return (0);
563 }
564 
565 int
566 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
567 {
568 	dsl_dir_t *dd;
569 	dsl_pool_t *dp;
570 	const char *snapname;
571 	uint64_t obj;
572 	int err = 0;
573 
574 	err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
575 	if (err)
576 		return (err);
577 
578 	dp = dd->dd_pool;
579 	obj = dd->dd_phys->dd_head_dataset_obj;
580 	rw_enter(&dp->dp_config_rwlock, RW_READER);
581 	if (obj)
582 		err = dsl_dataset_get_ref(dp, obj, tag, dsp);
583 	else
584 		err = ENOENT;
585 	if (err)
586 		goto out;
587 
588 	err = dsl_dataset_hold_ref(*dsp, tag);
589 
590 	/* we may be looking for a snapshot */
591 	if (err == 0 && snapname != NULL) {
592 		dsl_dataset_t *ds = NULL;
593 
594 		if (*snapname++ != '@') {
595 			dsl_dataset_rele(*dsp, tag);
596 			err = ENOENT;
597 			goto out;
598 		}
599 
600 		dprintf("looking for snapshot '%s'\n", snapname);
601 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
602 		if (err == 0)
603 			err = dsl_dataset_get_ref(dp, obj, tag, &ds);
604 		dsl_dataset_rele(*dsp, tag);
605 
606 		ASSERT3U((err == 0), ==, (ds != NULL));
607 
608 		if (ds) {
609 			mutex_enter(&ds->ds_lock);
610 			if (ds->ds_snapname[0] == 0)
611 				(void) strlcpy(ds->ds_snapname, snapname,
612 				    sizeof (ds->ds_snapname));
613 			mutex_exit(&ds->ds_lock);
614 			err = dsl_dataset_hold_ref(ds, tag);
615 			*dsp = err ? NULL : ds;
616 		}
617 	}
618 out:
619 	rw_exit(&dp->dp_config_rwlock);
620 	dsl_dir_close(dd, FTAG);
621 	return (err);
622 }
623 
624 int
625 dsl_dataset_own(const char *name, int flags, void *owner, dsl_dataset_t **dsp)
626 {
627 	int err = dsl_dataset_hold(name, owner, dsp);
628 	if (err)
629 		return (err);
630 	if ((*dsp)->ds_phys->ds_num_children > 0 &&
631 	    !DS_MODE_IS_READONLY(flags)) {
632 		dsl_dataset_rele(*dsp, owner);
633 		return (EROFS);
634 	}
635 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
636 		dsl_dataset_rele(*dsp, owner);
637 		return (EBUSY);
638 	}
639 	return (0);
640 }
641 
642 void
643 dsl_dataset_name(dsl_dataset_t *ds, char *name)
644 {
645 	if (ds == NULL) {
646 		(void) strcpy(name, "mos");
647 	} else {
648 		dsl_dir_name(ds->ds_dir, name);
649 		VERIFY(0 == dsl_dataset_get_snapname(ds));
650 		if (ds->ds_snapname[0]) {
651 			(void) strcat(name, "@");
652 			/*
653 			 * We use a "recursive" mutex so that we
654 			 * can call dprintf_ds() with ds_lock held.
655 			 */
656 			if (!MUTEX_HELD(&ds->ds_lock)) {
657 				mutex_enter(&ds->ds_lock);
658 				(void) strcat(name, ds->ds_snapname);
659 				mutex_exit(&ds->ds_lock);
660 			} else {
661 				(void) strcat(name, ds->ds_snapname);
662 			}
663 		}
664 	}
665 }
666 
667 static int
668 dsl_dataset_namelen(dsl_dataset_t *ds)
669 {
670 	int result;
671 
672 	if (ds == NULL) {
673 		result = 3;	/* "mos" */
674 	} else {
675 		result = dsl_dir_namelen(ds->ds_dir);
676 		VERIFY(0 == dsl_dataset_get_snapname(ds));
677 		if (ds->ds_snapname[0]) {
678 			++result;	/* adding one for the @-sign */
679 			if (!MUTEX_HELD(&ds->ds_lock)) {
680 				mutex_enter(&ds->ds_lock);
681 				result += strlen(ds->ds_snapname);
682 				mutex_exit(&ds->ds_lock);
683 			} else {
684 				result += strlen(ds->ds_snapname);
685 			}
686 		}
687 	}
688 
689 	return (result);
690 }
691 
692 void
693 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
694 {
695 	dmu_buf_rele(ds->ds_dbuf, tag);
696 }
697 
698 void
699 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
700 {
701 	if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
702 		rw_exit(&ds->ds_rwlock);
703 	}
704 	dsl_dataset_drop_ref(ds, tag);
705 }
706 
707 void
708 dsl_dataset_disown(dsl_dataset_t *ds, void *owner)
709 {
710 	ASSERT((ds->ds_owner == owner && ds->ds_dbuf) ||
711 	    (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
712 
713 	mutex_enter(&ds->ds_lock);
714 	ds->ds_owner = NULL;
715 	if (RW_WRITE_HELD(&ds->ds_rwlock)) {
716 		rw_exit(&ds->ds_rwlock);
717 		cv_broadcast(&ds->ds_exclusive_cv);
718 	}
719 	mutex_exit(&ds->ds_lock);
720 	if (ds->ds_dbuf)
721 		dsl_dataset_drop_ref(ds, owner);
722 	else
723 		dsl_dataset_evict(ds->ds_dbuf, ds);
724 }
725 
726 boolean_t
727 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *owner)
728 {
729 	boolean_t gotit = FALSE;
730 
731 	mutex_enter(&ds->ds_lock);
732 	if (ds->ds_owner == NULL &&
733 	    (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
734 		ds->ds_owner = owner;
735 		if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
736 			rw_exit(&ds->ds_rwlock);
737 		gotit = TRUE;
738 	}
739 	mutex_exit(&ds->ds_lock);
740 	return (gotit);
741 }
742 
743 void
744 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
745 {
746 	ASSERT3P(owner, ==, ds->ds_owner);
747 	if (!RW_WRITE_HELD(&ds->ds_rwlock))
748 		rw_enter(&ds->ds_rwlock, RW_WRITER);
749 }
750 
751 uint64_t
752 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
753     uint64_t flags, dmu_tx_t *tx)
754 {
755 	dsl_pool_t *dp = dd->dd_pool;
756 	dmu_buf_t *dbuf;
757 	dsl_dataset_phys_t *dsphys;
758 	uint64_t dsobj;
759 	objset_t *mos = dp->dp_meta_objset;
760 
761 	if (origin == NULL)
762 		origin = dp->dp_origin_snap;
763 
764 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
765 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
766 	ASSERT(dmu_tx_is_syncing(tx));
767 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
768 
769 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
770 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
771 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
772 	dmu_buf_will_dirty(dbuf, tx);
773 	dsphys = dbuf->db_data;
774 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
775 	dsphys->ds_dir_obj = dd->dd_object;
776 	dsphys->ds_flags = flags;
777 	dsphys->ds_fsid_guid = unique_create();
778 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
779 	    sizeof (dsphys->ds_guid));
780 	dsphys->ds_snapnames_zapobj =
781 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
782 	    DMU_OT_NONE, 0, tx);
783 	dsphys->ds_creation_time = gethrestime_sec();
784 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
785 	dsphys->ds_deadlist_obj =
786 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
787 
788 	if (origin) {
789 		dsphys->ds_prev_snap_obj = origin->ds_object;
790 		dsphys->ds_prev_snap_txg =
791 		    origin->ds_phys->ds_creation_txg;
792 		dsphys->ds_used_bytes =
793 		    origin->ds_phys->ds_used_bytes;
794 		dsphys->ds_compressed_bytes =
795 		    origin->ds_phys->ds_compressed_bytes;
796 		dsphys->ds_uncompressed_bytes =
797 		    origin->ds_phys->ds_uncompressed_bytes;
798 		dsphys->ds_bp = origin->ds_phys->ds_bp;
799 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
800 
801 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
802 		origin->ds_phys->ds_num_children++;
803 
804 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
805 			if (origin->ds_phys->ds_next_clones_obj == 0) {
806 				origin->ds_phys->ds_next_clones_obj =
807 				    zap_create(mos,
808 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
809 			}
810 			VERIFY(0 == zap_add_int(mos,
811 			    origin->ds_phys->ds_next_clones_obj,
812 			    dsobj, tx));
813 		}
814 
815 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
816 		dd->dd_phys->dd_origin_obj = origin->ds_object;
817 	}
818 
819 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
820 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
821 
822 	dmu_buf_rele(dbuf, FTAG);
823 
824 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
825 	dd->dd_phys->dd_head_dataset_obj = dsobj;
826 
827 	return (dsobj);
828 }
829 
830 uint64_t
831 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
832     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
833 {
834 	dsl_pool_t *dp = pdd->dd_pool;
835 	uint64_t dsobj, ddobj;
836 	dsl_dir_t *dd;
837 
838 	ASSERT(lastname[0] != '@');
839 
840 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
841 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
842 
843 	dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
844 
845 	dsl_deleg_set_create_perms(dd, tx, cr);
846 
847 	dsl_dir_close(dd, FTAG);
848 
849 	return (dsobj);
850 }
851 
852 struct destroyarg {
853 	dsl_sync_task_group_t *dstg;
854 	char *snapname;
855 	char *failed;
856 	boolean_t defer;
857 };
858 
859 static int
860 dsl_snapshot_destroy_one(char *name, void *arg)
861 {
862 	struct destroyarg *da = arg;
863 	dsl_dataset_t *ds;
864 	int err;
865 	char *dsname;
866 
867 	dsname = kmem_asprintf("%s@%s", name, da->snapname);
868 	err = dsl_dataset_own(dsname, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
869 	    da->dstg, &ds);
870 	strfree(dsname);
871 	if (err == 0) {
872 		struct dsl_ds_destroyarg *dsda;
873 
874 		dsl_dataset_make_exclusive(ds, da->dstg);
875 		if (ds->ds_user_ptr) {
876 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
877 			ds->ds_user_ptr = NULL;
878 		}
879 		dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
880 		dsda->ds = ds;
881 		dsda->defer = da->defer;
882 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
883 		    dsl_dataset_destroy_sync, dsda, da->dstg, 0);
884 	} else if (err == ENOENT) {
885 		err = 0;
886 	} else {
887 		(void) strcpy(da->failed, name);
888 	}
889 	return (err);
890 }
891 
892 /*
893  * Destroy 'snapname' in all descendants of 'fsname'.
894  */
895 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
896 int
897 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
898 {
899 	int err;
900 	struct destroyarg da;
901 	dsl_sync_task_t *dst;
902 	spa_t *spa;
903 
904 	err = spa_open(fsname, &spa, FTAG);
905 	if (err)
906 		return (err);
907 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
908 	da.snapname = snapname;
909 	da.failed = fsname;
910 	da.defer = defer;
911 
912 	err = dmu_objset_find(fsname,
913 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
914 
915 	if (err == 0)
916 		err = dsl_sync_task_group_wait(da.dstg);
917 
918 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
919 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
920 		struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
921 		dsl_dataset_t *ds = dsda->ds;
922 
923 		/*
924 		 * Return the file system name that triggered the error
925 		 */
926 		if (dst->dst_err) {
927 			dsl_dataset_name(ds, fsname);
928 			*strchr(fsname, '@') = '\0';
929 		}
930 		ASSERT3P(dsda->rm_origin, ==, NULL);
931 		dsl_dataset_disown(ds, da.dstg);
932 		kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
933 	}
934 
935 	dsl_sync_task_group_destroy(da.dstg);
936 	spa_close(spa, FTAG);
937 	return (err);
938 }
939 
940 static boolean_t
941 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
942 {
943 	boolean_t might_destroy = B_FALSE;
944 
945 	mutex_enter(&ds->ds_lock);
946 	if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
947 	    DS_IS_DEFER_DESTROY(ds))
948 		might_destroy = B_TRUE;
949 	mutex_exit(&ds->ds_lock);
950 
951 	return (might_destroy);
952 }
953 
954 #ifdef _KERNEL
955 static int
956 dsl_dataset_zvol_cleanup(dsl_dataset_t *ds, const char *name)
957 {
958 	int error;
959 	objset_t *os;
960 
961 	error = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
962 	if (error)
963 		return (error);
964 
965 	if (dmu_objset_type(os) == DMU_OST_ZVOL)
966 		error = zvol_remove_minor(name);
967 	dmu_objset_close(os);
968 
969 	return (error);
970 }
971 #endif
972 
973 /*
974  * If we're removing a clone, and these three conditions are true:
975  *	1) the clone's origin has no other children
976  *	2) the clone's origin has no user references
977  *	3) the clone's origin has been marked for deferred destruction
978  * Then, prepare to remove the origin as part of this sync task group.
979  */
980 static int
981 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
982 {
983 	dsl_dataset_t *ds = dsda->ds;
984 	dsl_dataset_t *origin = ds->ds_prev;
985 
986 	if (dsl_dataset_might_destroy_origin(origin)) {
987 		char *name;
988 		int namelen;
989 		int error;
990 
991 		namelen = dsl_dataset_namelen(origin) + 1;
992 		name = kmem_alloc(namelen, KM_SLEEP);
993 		dsl_dataset_name(origin, name);
994 #ifdef _KERNEL
995 		error = zfs_unmount_snap(name, NULL);
996 		if (error) {
997 			kmem_free(name, namelen);
998 			return (error);
999 		}
1000 		error = dsl_dataset_zvol_cleanup(origin, name);
1001 		if (error) {
1002 			kmem_free(name, namelen);
1003 			return (error);
1004 		}
1005 #endif
1006 		error = dsl_dataset_own(name,
1007 		    DS_MODE_READONLY | DS_MODE_INCONSISTENT,
1008 		    tag, &origin);
1009 		kmem_free(name, namelen);
1010 		if (error)
1011 			return (error);
1012 		dsda->rm_origin = origin;
1013 		dsl_dataset_make_exclusive(origin, tag);
1014 	}
1015 
1016 	return (0);
1017 }
1018 
1019 /*
1020  * ds must be opened as OWNER.  On return (whether successful or not),
1021  * ds will be closed and caller can no longer dereference it.
1022  */
1023 int
1024 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1025 {
1026 	int err;
1027 	dsl_sync_task_group_t *dstg;
1028 	objset_t *os;
1029 	dsl_dir_t *dd;
1030 	uint64_t obj;
1031 	struct dsl_ds_destroyarg dsda = {0};
1032 
1033 	dsda.ds = ds;
1034 
1035 	if (dsl_dataset_is_snapshot(ds)) {
1036 		/* Destroying a snapshot is simpler */
1037 		dsl_dataset_make_exclusive(ds, tag);
1038 
1039 		if (ds->ds_user_ptr) {
1040 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1041 			ds->ds_user_ptr = NULL;
1042 		}
1043 		/* NOTE: defer is always B_FALSE for non-snapshots */
1044 		dsda.defer = defer;
1045 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1046 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1047 		    &dsda, tag, 0);
1048 		ASSERT3P(dsda.rm_origin, ==, NULL);
1049 		goto out;
1050 	}
1051 
1052 	dd = ds->ds_dir;
1053 
1054 	/*
1055 	 * Check for errors and mark this ds as inconsistent, in
1056 	 * case we crash while freeing the objects.
1057 	 */
1058 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1059 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1060 	if (err)
1061 		goto out;
1062 
1063 	err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
1064 	if (err)
1065 		goto out;
1066 
1067 	/*
1068 	 * remove the objects in open context, so that we won't
1069 	 * have too much to do in syncing context.
1070 	 */
1071 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1072 	    ds->ds_phys->ds_prev_snap_txg)) {
1073 		/*
1074 		 * Ignore errors, if there is not enough disk space
1075 		 * we will deal with it in dsl_dataset_destroy_sync().
1076 		 */
1077 		(void) dmu_free_object(os, obj);
1078 	}
1079 
1080 	/*
1081 	 * We need to sync out all in-flight IO before we try to evict
1082 	 * (the dataset evict func is trying to clear the cached entries
1083 	 * for this dataset in the ARC).
1084 	 */
1085 	txg_wait_synced(dd->dd_pool, 0);
1086 
1087 	/*
1088 	 * If we managed to free all the objects in open
1089 	 * context, the user space accounting should be zero.
1090 	 */
1091 	if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1092 	    dmu_objset_userused_enabled(os->os)) {
1093 		uint64_t count;
1094 
1095 		ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1096 		    count == 0);
1097 		ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1098 		    count == 0);
1099 	}
1100 
1101 	dmu_objset_close(os);
1102 	if (err != ESRCH)
1103 		goto out;
1104 
1105 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1106 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1107 	rw_exit(&dd->dd_pool->dp_config_rwlock);
1108 
1109 	if (err)
1110 		goto out;
1111 
1112 	if (ds->ds_user_ptr) {
1113 		/*
1114 		 * We need to sync out all in-flight IO before we try
1115 		 * to evict (the dataset evict func is trying to clear
1116 		 * the cached entries for this dataset in the ARC).
1117 		 */
1118 		txg_wait_synced(dd->dd_pool, 0);
1119 	}
1120 
1121 	/*
1122 	 * Blow away the dsl_dir + head dataset.
1123 	 */
1124 	dsl_dataset_make_exclusive(ds, tag);
1125 	if (ds->ds_user_ptr) {
1126 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1127 		ds->ds_user_ptr = NULL;
1128 	}
1129 
1130 	/*
1131 	 * If we're removing a clone, we might also need to remove its
1132 	 * origin.
1133 	 */
1134 	do {
1135 		dsda.need_prep = B_FALSE;
1136 		if (dsl_dir_is_clone(dd)) {
1137 			err = dsl_dataset_origin_rm_prep(&dsda, tag);
1138 			if (err) {
1139 				dsl_dir_close(dd, FTAG);
1140 				goto out;
1141 			}
1142 		}
1143 
1144 		dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1145 		dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1146 		    dsl_dataset_destroy_sync, &dsda, tag, 0);
1147 		dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1148 		    dsl_dir_destroy_sync, dd, FTAG, 0);
1149 		err = dsl_sync_task_group_wait(dstg);
1150 		dsl_sync_task_group_destroy(dstg);
1151 
1152 		/*
1153 		 * We could be racing against 'zfs release' or 'zfs destroy -d'
1154 		 * on the origin snap, in which case we can get EBUSY if we
1155 		 * needed to destroy the origin snap but were not ready to
1156 		 * do so.
1157 		 */
1158 		if (dsda.need_prep) {
1159 			ASSERT(err == EBUSY);
1160 			ASSERT(dsl_dir_is_clone(dd));
1161 			ASSERT(dsda.rm_origin == NULL);
1162 		}
1163 	} while (dsda.need_prep);
1164 
1165 	if (dsda.rm_origin != NULL)
1166 		dsl_dataset_disown(dsda.rm_origin, tag);
1167 
1168 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
1169 	if (err)
1170 		dsl_dir_close(dd, FTAG);
1171 out:
1172 	dsl_dataset_disown(ds, tag);
1173 	return (err);
1174 }
1175 
1176 void *
1177 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
1178     void *p, dsl_dataset_evict_func_t func)
1179 {
1180 	void *old;
1181 
1182 	mutex_enter(&ds->ds_lock);
1183 	old = ds->ds_user_ptr;
1184 	if (old == NULL) {
1185 		ds->ds_user_ptr = p;
1186 		ds->ds_user_evict_func = func;
1187 	}
1188 	mutex_exit(&ds->ds_lock);
1189 	return (old);
1190 }
1191 
1192 void *
1193 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
1194 {
1195 	return (ds->ds_user_ptr);
1196 }
1197 
1198 blkptr_t *
1199 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1200 {
1201 	return (&ds->ds_phys->ds_bp);
1202 }
1203 
1204 void
1205 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1206 {
1207 	ASSERT(dmu_tx_is_syncing(tx));
1208 	/* If it's the meta-objset, set dp_meta_rootbp */
1209 	if (ds == NULL) {
1210 		tx->tx_pool->dp_meta_rootbp = *bp;
1211 	} else {
1212 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1213 		ds->ds_phys->ds_bp = *bp;
1214 	}
1215 }
1216 
1217 spa_t *
1218 dsl_dataset_get_spa(dsl_dataset_t *ds)
1219 {
1220 	return (ds->ds_dir->dd_pool->dp_spa);
1221 }
1222 
1223 void
1224 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1225 {
1226 	dsl_pool_t *dp;
1227 
1228 	if (ds == NULL) /* this is the meta-objset */
1229 		return;
1230 
1231 	ASSERT(ds->ds_user_ptr != NULL);
1232 
1233 	if (ds->ds_phys->ds_next_snap_obj != 0)
1234 		panic("dirtying snapshot!");
1235 
1236 	dp = ds->ds_dir->dd_pool;
1237 
1238 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1239 		/* up the hold count until we can be written out */
1240 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1241 	}
1242 }
1243 
1244 /*
1245  * The unique space in the head dataset can be calculated by subtracting
1246  * the space used in the most recent snapshot, that is still being used
1247  * in this file system, from the space currently in use.  To figure out
1248  * the space in the most recent snapshot still in use, we need to take
1249  * the total space used in the snapshot and subtract out the space that
1250  * has been freed up since the snapshot was taken.
1251  */
1252 static void
1253 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1254 {
1255 	uint64_t mrs_used;
1256 	uint64_t dlused, dlcomp, dluncomp;
1257 
1258 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1259 
1260 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1261 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1262 	else
1263 		mrs_used = 0;
1264 
1265 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1266 	    &dluncomp));
1267 
1268 	ASSERT3U(dlused, <=, mrs_used);
1269 	ds->ds_phys->ds_unique_bytes =
1270 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1271 
1272 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1273 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1274 	    SPA_VERSION_UNIQUE_ACCURATE)
1275 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1276 }
1277 
1278 static uint64_t
1279 dsl_dataset_unique(dsl_dataset_t *ds)
1280 {
1281 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1282 		dsl_dataset_recalc_head_uniq(ds);
1283 
1284 	return (ds->ds_phys->ds_unique_bytes);
1285 }
1286 
1287 struct killarg {
1288 	dsl_dataset_t *ds;
1289 	zio_t *zio;
1290 	dmu_tx_t *tx;
1291 };
1292 
1293 /* ARGSUSED */
1294 static int
1295 kill_blkptr(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
1296     const dnode_phys_t *dnp, void *arg)
1297 {
1298 	struct killarg *ka = arg;
1299 
1300 	if (bp == NULL)
1301 		return (0);
1302 
1303 	if ((zb->zb_level == -1ULL && zb->zb_blkid != 0) ||
1304 	    (zb->zb_object != 0 && dnp == NULL)) {
1305 		/*
1306 		 * It's a block in the intent log.  It has no
1307 		 * accounting, so just free it.
1308 		 */
1309 		VERIFY3U(0, ==, dsl_free(ka->zio, ka->tx->tx_pool,
1310 		    ka->tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT));
1311 	} else {
1312 		ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1313 		(void) dsl_dataset_block_kill(ka->ds, bp, ka->zio, ka->tx);
1314 	}
1315 
1316 	return (0);
1317 }
1318 
1319 /* ARGSUSED */
1320 static int
1321 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1322 {
1323 	dsl_dataset_t *ds = arg1;
1324 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1325 	uint64_t count;
1326 	int err;
1327 
1328 	/*
1329 	 * Can't delete a head dataset if there are snapshots of it.
1330 	 * (Except if the only snapshots are from the branch we cloned
1331 	 * from.)
1332 	 */
1333 	if (ds->ds_prev != NULL &&
1334 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1335 		return (EINVAL);
1336 
1337 	/*
1338 	 * This is really a dsl_dir thing, but check it here so that
1339 	 * we'll be less likely to leave this dataset inconsistent &
1340 	 * nearly destroyed.
1341 	 */
1342 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1343 	if (err)
1344 		return (err);
1345 	if (count != 0)
1346 		return (EEXIST);
1347 
1348 	return (0);
1349 }
1350 
1351 /* ARGSUSED */
1352 static void
1353 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1354 {
1355 	dsl_dataset_t *ds = arg1;
1356 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1357 
1358 	/* Mark it as inconsistent on-disk, in case we crash */
1359 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1360 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1361 
1362 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1363 	    cr, "dataset = %llu", ds->ds_object);
1364 }
1365 
1366 static int
1367 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1368     dmu_tx_t *tx)
1369 {
1370 	dsl_dataset_t *ds = dsda->ds;
1371 	dsl_dataset_t *ds_prev = ds->ds_prev;
1372 
1373 	if (dsl_dataset_might_destroy_origin(ds_prev)) {
1374 		struct dsl_ds_destroyarg ndsda = {0};
1375 
1376 		/*
1377 		 * If we're not prepared to remove the origin, don't remove
1378 		 * the clone either.
1379 		 */
1380 		if (dsda->rm_origin == NULL) {
1381 			dsda->need_prep = B_TRUE;
1382 			return (EBUSY);
1383 		}
1384 
1385 		ndsda.ds = ds_prev;
1386 		ndsda.is_origin_rm = B_TRUE;
1387 		return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1388 	}
1389 
1390 	/*
1391 	 * If we're not going to remove the origin after all,
1392 	 * undo the open context setup.
1393 	 */
1394 	if (dsda->rm_origin != NULL) {
1395 		dsl_dataset_disown(dsda->rm_origin, tag);
1396 		dsda->rm_origin = NULL;
1397 	}
1398 
1399 	return (0);
1400 }
1401 
1402 /* ARGSUSED */
1403 int
1404 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1405 {
1406 	struct dsl_ds_destroyarg *dsda = arg1;
1407 	dsl_dataset_t *ds = dsda->ds;
1408 
1409 	/* we have an owner hold, so noone else can destroy us */
1410 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1411 
1412 	/*
1413 	 * Only allow deferred destroy on pools that support it.
1414 	 * NOTE: deferred destroy is only supported on snapshots.
1415 	 */
1416 	if (dsda->defer) {
1417 		if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1418 		    SPA_VERSION_USERREFS)
1419 			return (ENOTSUP);
1420 		ASSERT(dsl_dataset_is_snapshot(ds));
1421 		return (0);
1422 	}
1423 
1424 	/*
1425 	 * Can't delete a head dataset if there are snapshots of it.
1426 	 * (Except if the only snapshots are from the branch we cloned
1427 	 * from.)
1428 	 */
1429 	if (ds->ds_prev != NULL &&
1430 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1431 		return (EINVAL);
1432 
1433 	/*
1434 	 * If we made changes this txg, traverse_dsl_dataset won't find
1435 	 * them.  Try again.
1436 	 */
1437 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1438 		return (EAGAIN);
1439 
1440 	if (dsl_dataset_is_snapshot(ds)) {
1441 		/*
1442 		 * If this snapshot has an elevated user reference count,
1443 		 * we can't destroy it yet.
1444 		 */
1445 		if (ds->ds_userrefs > 0 && !dsda->releasing)
1446 			return (EBUSY);
1447 
1448 		mutex_enter(&ds->ds_lock);
1449 		/*
1450 		 * Can't delete a branch point. However, if we're destroying
1451 		 * a clone and removing its origin due to it having a user
1452 		 * hold count of 0 and having been marked for deferred destroy,
1453 		 * it's OK for the origin to have a single clone.
1454 		 */
1455 		if (ds->ds_phys->ds_num_children >
1456 		    (dsda->is_origin_rm ? 2 : 1)) {
1457 			mutex_exit(&ds->ds_lock);
1458 			return (EEXIST);
1459 		}
1460 		mutex_exit(&ds->ds_lock);
1461 	} else if (dsl_dir_is_clone(ds->ds_dir)) {
1462 		return (dsl_dataset_origin_check(dsda, arg2, tx));
1463 	}
1464 
1465 	/* XXX we should do some i/o error checking... */
1466 	return (0);
1467 }
1468 
1469 struct refsarg {
1470 	kmutex_t lock;
1471 	boolean_t gone;
1472 	kcondvar_t cv;
1473 };
1474 
1475 /* ARGSUSED */
1476 static void
1477 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1478 {
1479 	struct refsarg *arg = argv;
1480 
1481 	mutex_enter(&arg->lock);
1482 	arg->gone = TRUE;
1483 	cv_signal(&arg->cv);
1484 	mutex_exit(&arg->lock);
1485 }
1486 
1487 static void
1488 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1489 {
1490 	struct refsarg arg;
1491 
1492 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1493 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1494 	arg.gone = FALSE;
1495 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1496 	    dsl_dataset_refs_gone);
1497 	dmu_buf_rele(ds->ds_dbuf, tag);
1498 	mutex_enter(&arg.lock);
1499 	while (!arg.gone)
1500 		cv_wait(&arg.cv, &arg.lock);
1501 	ASSERT(arg.gone);
1502 	mutex_exit(&arg.lock);
1503 	ds->ds_dbuf = NULL;
1504 	ds->ds_phys = NULL;
1505 	mutex_destroy(&arg.lock);
1506 	cv_destroy(&arg.cv);
1507 }
1508 
1509 void
1510 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1511 {
1512 	struct dsl_ds_destroyarg *dsda = arg1;
1513 	dsl_dataset_t *ds = dsda->ds;
1514 	zio_t *zio;
1515 	int err;
1516 	int after_branch_point = FALSE;
1517 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1518 	objset_t *mos = dp->dp_meta_objset;
1519 	dsl_dataset_t *ds_prev = NULL;
1520 	uint64_t obj;
1521 
1522 	ASSERT(ds->ds_owner);
1523 	ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1524 	ASSERT(ds->ds_prev == NULL ||
1525 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1526 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1527 
1528 	if (dsda->defer) {
1529 		ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1530 		if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) {
1531 			dmu_buf_will_dirty(ds->ds_dbuf, tx);
1532 			ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1533 			return;
1534 		}
1535 	}
1536 
1537 	/* signal any waiters that this dataset is going away */
1538 	mutex_enter(&ds->ds_lock);
1539 	ds->ds_owner = dsl_reaper;
1540 	cv_broadcast(&ds->ds_exclusive_cv);
1541 	mutex_exit(&ds->ds_lock);
1542 
1543 	/* Remove our reservation */
1544 	if (ds->ds_reserved != 0) {
1545 		uint64_t val = 0;
1546 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1547 		ASSERT3U(ds->ds_reserved, ==, 0);
1548 	}
1549 
1550 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1551 
1552 	dsl_pool_ds_destroyed(ds, tx);
1553 
1554 	obj = ds->ds_object;
1555 
1556 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1557 		if (ds->ds_prev) {
1558 			ds_prev = ds->ds_prev;
1559 		} else {
1560 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1561 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1562 		}
1563 		after_branch_point =
1564 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1565 
1566 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1567 		if (after_branch_point &&
1568 		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
1569 			VERIFY3U(0, ==, zap_remove_int(mos,
1570 			    ds_prev->ds_phys->ds_next_clones_obj, obj, tx));
1571 			if (ds->ds_phys->ds_next_snap_obj != 0) {
1572 				VERIFY(0 == zap_add_int(mos,
1573 				    ds_prev->ds_phys->ds_next_clones_obj,
1574 				    ds->ds_phys->ds_next_snap_obj, tx));
1575 			}
1576 		}
1577 		if (after_branch_point &&
1578 		    ds->ds_phys->ds_next_snap_obj == 0) {
1579 			/* This clone is toast. */
1580 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1581 			ds_prev->ds_phys->ds_num_children--;
1582 
1583 			/*
1584 			 * If the clone's origin has no other clones, no
1585 			 * user holds, and has been marked for deferred
1586 			 * deletion, then we should have done the necessary
1587 			 * destroy setup for it.
1588 			 */
1589 			if (ds_prev->ds_phys->ds_num_children == 1 &&
1590 			    ds_prev->ds_userrefs == 0 &&
1591 			    DS_IS_DEFER_DESTROY(ds_prev)) {
1592 				ASSERT3P(dsda->rm_origin, !=, NULL);
1593 			} else {
1594 				ASSERT3P(dsda->rm_origin, ==, NULL);
1595 			}
1596 		} else if (!after_branch_point) {
1597 			ds_prev->ds_phys->ds_next_snap_obj =
1598 			    ds->ds_phys->ds_next_snap_obj;
1599 		}
1600 	}
1601 
1602 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1603 
1604 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1605 		blkptr_t bp;
1606 		dsl_dataset_t *ds_next;
1607 		uint64_t itor = 0;
1608 		uint64_t old_unique;
1609 		int64_t used = 0, compressed = 0, uncompressed = 0;
1610 
1611 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1612 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1613 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1614 
1615 		old_unique = dsl_dataset_unique(ds_next);
1616 
1617 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1618 		ds_next->ds_phys->ds_prev_snap_obj =
1619 		    ds->ds_phys->ds_prev_snap_obj;
1620 		ds_next->ds_phys->ds_prev_snap_txg =
1621 		    ds->ds_phys->ds_prev_snap_txg;
1622 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1623 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1624 
1625 		/*
1626 		 * Transfer to our deadlist (which will become next's
1627 		 * new deadlist) any entries from next's current
1628 		 * deadlist which were born before prev, and free the
1629 		 * other entries.
1630 		 *
1631 		 * XXX we're doing this long task with the config lock held
1632 		 */
1633 		while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1634 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1635 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1636 				    &bp, tx));
1637 				if (ds_prev && !after_branch_point &&
1638 				    bp.blk_birth >
1639 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1640 					ds_prev->ds_phys->ds_unique_bytes +=
1641 					    bp_get_dasize(dp->dp_spa, &bp);
1642 				}
1643 			} else {
1644 				used += bp_get_dasize(dp->dp_spa, &bp);
1645 				compressed += BP_GET_PSIZE(&bp);
1646 				uncompressed += BP_GET_UCSIZE(&bp);
1647 				/* XXX check return value? */
1648 				(void) dsl_free(zio, dp, tx->tx_txg,
1649 				    &bp, NULL, NULL, ARC_NOWAIT);
1650 			}
1651 		}
1652 
1653 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1654 
1655 		/* change snapused */
1656 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1657 		    -used, -compressed, -uncompressed, tx);
1658 
1659 		/* free next's deadlist */
1660 		bplist_close(&ds_next->ds_deadlist);
1661 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1662 
1663 		/* set next's deadlist to our deadlist */
1664 		bplist_close(&ds->ds_deadlist);
1665 		ds_next->ds_phys->ds_deadlist_obj =
1666 		    ds->ds_phys->ds_deadlist_obj;
1667 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1668 		    ds_next->ds_phys->ds_deadlist_obj));
1669 		ds->ds_phys->ds_deadlist_obj = 0;
1670 
1671 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1672 			/*
1673 			 * Update next's unique to include blocks which
1674 			 * were previously shared by only this snapshot
1675 			 * and it.  Those blocks will be born after the
1676 			 * prev snap and before this snap, and will have
1677 			 * died after the next snap and before the one
1678 			 * after that (ie. be on the snap after next's
1679 			 * deadlist).
1680 			 *
1681 			 * XXX we're doing this long task with the
1682 			 * config lock held
1683 			 */
1684 			dsl_dataset_t *ds_after_next;
1685 			uint64_t space;
1686 
1687 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1688 			    ds_next->ds_phys->ds_next_snap_obj,
1689 			    FTAG, &ds_after_next));
1690 
1691 			VERIFY(0 ==
1692 			    bplist_space_birthrange(&ds_after_next->ds_deadlist,
1693 			    ds->ds_phys->ds_prev_snap_txg,
1694 			    ds->ds_phys->ds_creation_txg, &space));
1695 			ds_next->ds_phys->ds_unique_bytes += space;
1696 
1697 			dsl_dataset_rele(ds_after_next, FTAG);
1698 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1699 		} else {
1700 			ASSERT3P(ds_next->ds_prev, ==, ds);
1701 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1702 			ds_next->ds_prev = NULL;
1703 			if (ds_prev) {
1704 				VERIFY(0 == dsl_dataset_get_ref(dp,
1705 				    ds->ds_phys->ds_prev_snap_obj,
1706 				    ds_next, &ds_next->ds_prev));
1707 			}
1708 
1709 			dsl_dataset_recalc_head_uniq(ds_next);
1710 
1711 			/*
1712 			 * Reduce the amount of our unconsmed refreservation
1713 			 * being charged to our parent by the amount of
1714 			 * new unique data we have gained.
1715 			 */
1716 			if (old_unique < ds_next->ds_reserved) {
1717 				int64_t mrsdelta;
1718 				uint64_t new_unique =
1719 				    ds_next->ds_phys->ds_unique_bytes;
1720 
1721 				ASSERT(old_unique <= new_unique);
1722 				mrsdelta = MIN(new_unique - old_unique,
1723 				    ds_next->ds_reserved - old_unique);
1724 				dsl_dir_diduse_space(ds->ds_dir,
1725 				    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1726 			}
1727 		}
1728 		dsl_dataset_rele(ds_next, FTAG);
1729 	} else {
1730 		/*
1731 		 * There's no next snapshot, so this is a head dataset.
1732 		 * Destroy the deadlist.  Unless it's a clone, the
1733 		 * deadlist should be empty.  (If it's a clone, it's
1734 		 * safe to ignore the deadlist contents.)
1735 		 */
1736 		struct killarg ka;
1737 
1738 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1739 		bplist_close(&ds->ds_deadlist);
1740 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1741 		ds->ds_phys->ds_deadlist_obj = 0;
1742 
1743 		/*
1744 		 * Free everything that we point to (that's born after
1745 		 * the previous snapshot, if we are a clone)
1746 		 *
1747 		 * NB: this should be very quick, because we already
1748 		 * freed all the objects in open context.
1749 		 */
1750 		ka.ds = ds;
1751 		ka.zio = zio;
1752 		ka.tx = tx;
1753 		err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1754 		    TRAVERSE_POST, kill_blkptr, &ka);
1755 		ASSERT3U(err, ==, 0);
1756 		ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1757 		    ds->ds_phys->ds_unique_bytes == 0);
1758 	}
1759 
1760 	err = zio_wait(zio);
1761 	ASSERT3U(err, ==, 0);
1762 
1763 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1764 		/* Erase the link in the dir */
1765 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1766 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1767 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1768 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1769 		ASSERT(err == 0);
1770 	} else {
1771 		/* remove from snapshot namespace */
1772 		dsl_dataset_t *ds_head;
1773 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1774 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1775 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1776 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1777 #ifdef ZFS_DEBUG
1778 		{
1779 			uint64_t val;
1780 
1781 			err = dsl_dataset_snap_lookup(ds_head,
1782 			    ds->ds_snapname, &val);
1783 			ASSERT3U(err, ==, 0);
1784 			ASSERT3U(val, ==, obj);
1785 		}
1786 #endif
1787 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1788 		ASSERT(err == 0);
1789 		dsl_dataset_rele(ds_head, FTAG);
1790 	}
1791 
1792 	if (ds_prev && ds->ds_prev != ds_prev)
1793 		dsl_dataset_rele(ds_prev, FTAG);
1794 
1795 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1796 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1797 	    cr, "dataset = %llu", ds->ds_object);
1798 
1799 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1800 		uint64_t count;
1801 		ASSERT(0 == zap_count(mos,
1802 		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1803 		VERIFY(0 == dmu_object_free(mos,
1804 		    ds->ds_phys->ds_next_clones_obj, tx));
1805 	}
1806 	if (ds->ds_phys->ds_props_obj != 0)
1807 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1808 	if (ds->ds_phys->ds_userrefs_obj != 0)
1809 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1810 	dsl_dir_close(ds->ds_dir, ds);
1811 	ds->ds_dir = NULL;
1812 	dsl_dataset_drain_refs(ds, tag);
1813 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1814 
1815 	if (dsda->rm_origin) {
1816 		/*
1817 		 * Remove the origin of the clone we just destroyed.
1818 		 */
1819 		dsl_dataset_t *origin = ds->ds_prev;
1820 		struct dsl_ds_destroyarg ndsda = {0};
1821 
1822 		ASSERT3P(origin, ==, dsda->rm_origin);
1823 		if (origin->ds_user_ptr) {
1824 			origin->ds_user_evict_func(origin, origin->ds_user_ptr);
1825 			origin->ds_user_ptr = NULL;
1826 		}
1827 
1828 		dsl_dataset_rele(origin, ds);
1829 		ds->ds_prev = NULL;
1830 
1831 		ndsda.ds = origin;
1832 		dsl_dataset_destroy_sync(&ndsda, tag, cr, tx);
1833 	}
1834 }
1835 
1836 static int
1837 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1838 {
1839 	uint64_t asize;
1840 
1841 	if (!dmu_tx_is_syncing(tx))
1842 		return (0);
1843 
1844 	/*
1845 	 * If there's an fs-only reservation, any blocks that might become
1846 	 * owned by the snapshot dataset must be accommodated by space
1847 	 * outside of the reservation.
1848 	 */
1849 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1850 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1851 		return (ENOSPC);
1852 
1853 	/*
1854 	 * Propogate any reserved space for this snapshot to other
1855 	 * snapshot checks in this sync group.
1856 	 */
1857 	if (asize > 0)
1858 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1859 
1860 	return (0);
1861 }
1862 
1863 /* ARGSUSED */
1864 int
1865 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1866 {
1867 	dsl_dataset_t *ds = arg1;
1868 	const char *snapname = arg2;
1869 	int err;
1870 	uint64_t value;
1871 
1872 	/*
1873 	 * We don't allow multiple snapshots of the same txg.  If there
1874 	 * is already one, try again.
1875 	 */
1876 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1877 		return (EAGAIN);
1878 
1879 	/*
1880 	 * Check for conflicting name snapshot name.
1881 	 */
1882 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1883 	if (err == 0)
1884 		return (EEXIST);
1885 	if (err != ENOENT)
1886 		return (err);
1887 
1888 	/*
1889 	 * Check that the dataset's name is not too long.  Name consists
1890 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1891 	 */
1892 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1893 		return (ENAMETOOLONG);
1894 
1895 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1896 	if (err)
1897 		return (err);
1898 
1899 	ds->ds_trysnap_txg = tx->tx_txg;
1900 	return (0);
1901 }
1902 
1903 void
1904 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1905 {
1906 	dsl_dataset_t *ds = arg1;
1907 	const char *snapname = arg2;
1908 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1909 	dmu_buf_t *dbuf;
1910 	dsl_dataset_phys_t *dsphys;
1911 	uint64_t dsobj, crtxg;
1912 	objset_t *mos = dp->dp_meta_objset;
1913 	int err;
1914 
1915 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1916 
1917 	/*
1918 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1919 	 */
1920 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1921 		crtxg = 1;
1922 	else
1923 		crtxg = tx->tx_txg;
1924 
1925 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1926 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1927 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1928 	dmu_buf_will_dirty(dbuf, tx);
1929 	dsphys = dbuf->db_data;
1930 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1931 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1932 	dsphys->ds_fsid_guid = unique_create();
1933 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1934 	    sizeof (dsphys->ds_guid));
1935 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1936 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1937 	dsphys->ds_next_snap_obj = ds->ds_object;
1938 	dsphys->ds_num_children = 1;
1939 	dsphys->ds_creation_time = gethrestime_sec();
1940 	dsphys->ds_creation_txg = crtxg;
1941 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1942 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1943 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1944 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1945 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1946 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1947 	dmu_buf_rele(dbuf, FTAG);
1948 
1949 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1950 	if (ds->ds_prev) {
1951 		uint64_t next_clones_obj =
1952 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1953 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1954 		    ds->ds_object ||
1955 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1956 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1957 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1958 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1959 			    ds->ds_prev->ds_phys->ds_creation_txg);
1960 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1961 		} else if (next_clones_obj != 0) {
1962 			VERIFY3U(0, ==, zap_remove_int(mos,
1963 			    next_clones_obj, dsphys->ds_next_snap_obj, tx));
1964 			VERIFY3U(0, ==, zap_add_int(mos,
1965 			    next_clones_obj, dsobj, tx));
1966 		}
1967 	}
1968 
1969 	/*
1970 	 * If we have a reference-reservation on this dataset, we will
1971 	 * need to increase the amount of refreservation being charged
1972 	 * since our unique space is going to zero.
1973 	 */
1974 	if (ds->ds_reserved) {
1975 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1976 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1977 		    add, 0, 0, tx);
1978 	}
1979 
1980 	bplist_close(&ds->ds_deadlist);
1981 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1982 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1983 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1984 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1985 	ds->ds_phys->ds_unique_bytes = 0;
1986 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1987 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1988 	ds->ds_phys->ds_deadlist_obj =
1989 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1990 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1991 	    ds->ds_phys->ds_deadlist_obj));
1992 
1993 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1994 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1995 	    snapname, 8, 1, &dsobj, tx);
1996 	ASSERT(err == 0);
1997 
1998 	if (ds->ds_prev)
1999 		dsl_dataset_drop_ref(ds->ds_prev, ds);
2000 	VERIFY(0 == dsl_dataset_get_ref(dp,
2001 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2002 
2003 	dsl_pool_ds_snapshotted(ds, tx);
2004 
2005 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
2006 	    "dataset = %llu", dsobj);
2007 }
2008 
2009 void
2010 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2011 {
2012 	ASSERT(dmu_tx_is_syncing(tx));
2013 	ASSERT(ds->ds_user_ptr != NULL);
2014 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2015 
2016 	/*
2017 	 * in case we had to change ds_fsid_guid when we opened it,
2018 	 * sync it out now.
2019 	 */
2020 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2021 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2022 
2023 	dsl_dir_dirty(ds->ds_dir, tx);
2024 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
2025 }
2026 
2027 void
2028 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2029 {
2030 	uint64_t refd, avail, uobjs, aobjs;
2031 
2032 	dsl_dir_stats(ds->ds_dir, nv);
2033 
2034 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2035 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2036 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2037 
2038 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2039 	    ds->ds_phys->ds_creation_time);
2040 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2041 	    ds->ds_phys->ds_creation_txg);
2042 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2043 	    ds->ds_quota);
2044 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2045 	    ds->ds_reserved);
2046 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2047 	    ds->ds_phys->ds_guid);
2048 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, ds->ds_userrefs);
2049 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2050 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2051 
2052 	if (ds->ds_phys->ds_next_snap_obj) {
2053 		/*
2054 		 * This is a snapshot; override the dd's space used with
2055 		 * our unique space and compression ratio.
2056 		 */
2057 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2058 		    ds->ds_phys->ds_unique_bytes);
2059 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2060 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2061 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
2062 		    ds->ds_phys->ds_compressed_bytes));
2063 	}
2064 }
2065 
2066 void
2067 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2068 {
2069 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2070 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2071 	stat->dds_guid = ds->ds_phys->ds_guid;
2072 	if (ds->ds_phys->ds_next_snap_obj) {
2073 		stat->dds_is_snapshot = B_TRUE;
2074 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2075 	} else {
2076 		stat->dds_is_snapshot = B_FALSE;
2077 		stat->dds_num_clones = 0;
2078 	}
2079 
2080 	/* clone origin is really a dsl_dir thing... */
2081 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2082 	if (dsl_dir_is_clone(ds->ds_dir)) {
2083 		dsl_dataset_t *ods;
2084 
2085 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2086 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2087 		dsl_dataset_name(ods, stat->dds_origin);
2088 		dsl_dataset_drop_ref(ods, FTAG);
2089 	} else {
2090 		stat->dds_origin[0] = '\0';
2091 	}
2092 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2093 }
2094 
2095 uint64_t
2096 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2097 {
2098 	return (ds->ds_fsid_guid);
2099 }
2100 
2101 void
2102 dsl_dataset_space(dsl_dataset_t *ds,
2103     uint64_t *refdbytesp, uint64_t *availbytesp,
2104     uint64_t *usedobjsp, uint64_t *availobjsp)
2105 {
2106 	*refdbytesp = ds->ds_phys->ds_used_bytes;
2107 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2108 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2109 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2110 	if (ds->ds_quota != 0) {
2111 		/*
2112 		 * Adjust available bytes according to refquota
2113 		 */
2114 		if (*refdbytesp < ds->ds_quota)
2115 			*availbytesp = MIN(*availbytesp,
2116 			    ds->ds_quota - *refdbytesp);
2117 		else
2118 			*availbytesp = 0;
2119 	}
2120 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2121 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2122 }
2123 
2124 boolean_t
2125 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2126 {
2127 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2128 
2129 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2130 	    dsl_pool_sync_context(dp));
2131 	if (ds->ds_prev == NULL)
2132 		return (B_FALSE);
2133 	if (ds->ds_phys->ds_bp.blk_birth >
2134 	    ds->ds_prev->ds_phys->ds_creation_txg)
2135 		return (B_TRUE);
2136 	return (B_FALSE);
2137 }
2138 
2139 /* ARGSUSED */
2140 static int
2141 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2142 {
2143 	dsl_dataset_t *ds = arg1;
2144 	char *newsnapname = arg2;
2145 	dsl_dir_t *dd = ds->ds_dir;
2146 	dsl_dataset_t *hds;
2147 	uint64_t val;
2148 	int err;
2149 
2150 	err = dsl_dataset_hold_obj(dd->dd_pool,
2151 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2152 	if (err)
2153 		return (err);
2154 
2155 	/* new name better not be in use */
2156 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2157 	dsl_dataset_rele(hds, FTAG);
2158 
2159 	if (err == 0)
2160 		err = EEXIST;
2161 	else if (err == ENOENT)
2162 		err = 0;
2163 
2164 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2165 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2166 		err = ENAMETOOLONG;
2167 
2168 	return (err);
2169 }
2170 
2171 static void
2172 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
2173     cred_t *cr, dmu_tx_t *tx)
2174 {
2175 	dsl_dataset_t *ds = arg1;
2176 	const char *newsnapname = arg2;
2177 	dsl_dir_t *dd = ds->ds_dir;
2178 	objset_t *mos = dd->dd_pool->dp_meta_objset;
2179 	dsl_dataset_t *hds;
2180 	int err;
2181 
2182 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2183 
2184 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2185 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2186 
2187 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2188 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2189 	ASSERT3U(err, ==, 0);
2190 	mutex_enter(&ds->ds_lock);
2191 	(void) strcpy(ds->ds_snapname, newsnapname);
2192 	mutex_exit(&ds->ds_lock);
2193 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2194 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2195 	ASSERT3U(err, ==, 0);
2196 
2197 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2198 	    cr, "dataset = %llu", ds->ds_object);
2199 	dsl_dataset_rele(hds, FTAG);
2200 }
2201 
2202 struct renamesnaparg {
2203 	dsl_sync_task_group_t *dstg;
2204 	char failed[MAXPATHLEN];
2205 	char *oldsnap;
2206 	char *newsnap;
2207 };
2208 
2209 static int
2210 dsl_snapshot_rename_one(char *name, void *arg)
2211 {
2212 	struct renamesnaparg *ra = arg;
2213 	dsl_dataset_t *ds = NULL;
2214 	char *cp;
2215 	int err;
2216 
2217 	cp = name + strlen(name);
2218 	*cp = '@';
2219 	(void) strcpy(cp + 1, ra->oldsnap);
2220 
2221 	/*
2222 	 * For recursive snapshot renames the parent won't be changing
2223 	 * so we just pass name for both the to/from argument.
2224 	 */
2225 	err = zfs_secpolicy_rename_perms(name, name, CRED());
2226 	if (err == ENOENT) {
2227 		return (0);
2228 	} else if (err) {
2229 		(void) strcpy(ra->failed, name);
2230 		return (err);
2231 	}
2232 
2233 #ifdef _KERNEL
2234 	/*
2235 	 * For all filesystems undergoing rename, we'll need to unmount it.
2236 	 */
2237 	(void) zfs_unmount_snap(name, NULL);
2238 #endif
2239 	err = dsl_dataset_hold(name, ra->dstg, &ds);
2240 	*cp = '\0';
2241 	if (err == ENOENT) {
2242 		return (0);
2243 	} else if (err) {
2244 		(void) strcpy(ra->failed, name);
2245 		return (err);
2246 	}
2247 
2248 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2249 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2250 
2251 	return (0);
2252 }
2253 
2254 static int
2255 dsl_recursive_rename(char *oldname, const char *newname)
2256 {
2257 	int err;
2258 	struct renamesnaparg *ra;
2259 	dsl_sync_task_t *dst;
2260 	spa_t *spa;
2261 	char *cp, *fsname = spa_strdup(oldname);
2262 	int len = strlen(oldname);
2263 
2264 	/* truncate the snapshot name to get the fsname */
2265 	cp = strchr(fsname, '@');
2266 	*cp = '\0';
2267 
2268 	err = spa_open(fsname, &spa, FTAG);
2269 	if (err) {
2270 		kmem_free(fsname, len + 1);
2271 		return (err);
2272 	}
2273 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2274 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2275 
2276 	ra->oldsnap = strchr(oldname, '@') + 1;
2277 	ra->newsnap = strchr(newname, '@') + 1;
2278 	*ra->failed = '\0';
2279 
2280 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2281 	    DS_FIND_CHILDREN);
2282 	kmem_free(fsname, len + 1);
2283 
2284 	if (err == 0) {
2285 		err = dsl_sync_task_group_wait(ra->dstg);
2286 	}
2287 
2288 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2289 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2290 		dsl_dataset_t *ds = dst->dst_arg1;
2291 		if (dst->dst_err) {
2292 			dsl_dir_name(ds->ds_dir, ra->failed);
2293 			(void) strcat(ra->failed, "@");
2294 			(void) strcat(ra->failed, ra->newsnap);
2295 		}
2296 		dsl_dataset_rele(ds, ra->dstg);
2297 	}
2298 
2299 	if (err)
2300 		(void) strcpy(oldname, ra->failed);
2301 
2302 	dsl_sync_task_group_destroy(ra->dstg);
2303 	kmem_free(ra, sizeof (struct renamesnaparg));
2304 	spa_close(spa, FTAG);
2305 	return (err);
2306 }
2307 
2308 static int
2309 dsl_valid_rename(char *oldname, void *arg)
2310 {
2311 	int delta = *(int *)arg;
2312 
2313 	if (strlen(oldname) + delta >= MAXNAMELEN)
2314 		return (ENAMETOOLONG);
2315 
2316 	return (0);
2317 }
2318 
2319 #pragma weak dmu_objset_rename = dsl_dataset_rename
2320 int
2321 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2322 {
2323 	dsl_dir_t *dd;
2324 	dsl_dataset_t *ds;
2325 	const char *tail;
2326 	int err;
2327 
2328 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2329 	if (err)
2330 		return (err);
2331 	/*
2332 	 * If there are more than 2 references there may be holds
2333 	 * hanging around that haven't been cleared out yet.
2334 	 */
2335 	if (dmu_buf_refcount(dd->dd_dbuf) > 2)
2336 		txg_wait_synced(dd->dd_pool, 0);
2337 	if (tail == NULL) {
2338 		int delta = strlen(newname) - strlen(oldname);
2339 
2340 		/* if we're growing, validate child name lengths */
2341 		if (delta > 0)
2342 			err = dmu_objset_find(oldname, dsl_valid_rename,
2343 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2344 
2345 		if (!err)
2346 			err = dsl_dir_rename(dd, newname);
2347 		dsl_dir_close(dd, FTAG);
2348 		return (err);
2349 	}
2350 	if (tail[0] != '@') {
2351 		/* the name ended in a nonexistant component */
2352 		dsl_dir_close(dd, FTAG);
2353 		return (ENOENT);
2354 	}
2355 
2356 	dsl_dir_close(dd, FTAG);
2357 
2358 	/* new name must be snapshot in same filesystem */
2359 	tail = strchr(newname, '@');
2360 	if (tail == NULL)
2361 		return (EINVAL);
2362 	tail++;
2363 	if (strncmp(oldname, newname, tail - newname) != 0)
2364 		return (EXDEV);
2365 
2366 	if (recursive) {
2367 		err = dsl_recursive_rename(oldname, newname);
2368 	} else {
2369 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2370 		if (err)
2371 			return (err);
2372 
2373 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2374 		    dsl_dataset_snapshot_rename_check,
2375 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2376 
2377 		dsl_dataset_rele(ds, FTAG);
2378 	}
2379 
2380 	return (err);
2381 }
2382 
2383 struct promotenode {
2384 	list_node_t link;
2385 	dsl_dataset_t *ds;
2386 };
2387 
2388 struct promotearg {
2389 	list_t shared_snaps, origin_snaps, clone_snaps;
2390 	dsl_dataset_t *origin_origin, *origin_head;
2391 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2392 };
2393 
2394 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2395 
2396 /* ARGSUSED */
2397 static int
2398 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2399 {
2400 	dsl_dataset_t *hds = arg1;
2401 	struct promotearg *pa = arg2;
2402 	struct promotenode *snap = list_head(&pa->shared_snaps);
2403 	dsl_dataset_t *origin_ds = snap->ds;
2404 	int err;
2405 
2406 	/* Check that it is a real clone */
2407 	if (!dsl_dir_is_clone(hds->ds_dir))
2408 		return (EINVAL);
2409 
2410 	/* Since this is so expensive, don't do the preliminary check */
2411 	if (!dmu_tx_is_syncing(tx))
2412 		return (0);
2413 
2414 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2415 		return (EXDEV);
2416 
2417 	/* compute origin's new unique space */
2418 	snap = list_tail(&pa->clone_snaps);
2419 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2420 	err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2421 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique);
2422 	if (err)
2423 		return (err);
2424 
2425 	/*
2426 	 * Walk the snapshots that we are moving
2427 	 *
2428 	 * Compute space to transfer.  Consider the incremental changes
2429 	 * to used for each snapshot:
2430 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2431 	 * So each snapshot gave birth to:
2432 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2433 	 * So a sequence would look like:
2434 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2435 	 * Which simplifies to:
2436 	 * uN + kN + kN-1 + ... + k1 + k0
2437 	 * Note however, if we stop before we reach the ORIGIN we get:
2438 	 * uN + kN + kN-1 + ... + kM - uM-1
2439 	 */
2440 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2441 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2442 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2443 	for (snap = list_head(&pa->shared_snaps); snap;
2444 	    snap = list_next(&pa->shared_snaps, snap)) {
2445 		uint64_t val, dlused, dlcomp, dluncomp;
2446 		dsl_dataset_t *ds = snap->ds;
2447 
2448 		/* Check that the snapshot name does not conflict */
2449 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2450 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2451 		if (err == 0)
2452 			return (EEXIST);
2453 		if (err != ENOENT)
2454 			return (err);
2455 
2456 		/* The very first snapshot does not have a deadlist */
2457 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2458 			continue;
2459 
2460 		if (err = bplist_space(&ds->ds_deadlist,
2461 		    &dlused, &dlcomp, &dluncomp))
2462 			return (err);
2463 		pa->used += dlused;
2464 		pa->comp += dlcomp;
2465 		pa->uncomp += dluncomp;
2466 	}
2467 
2468 	/*
2469 	 * If we are a clone of a clone then we never reached ORIGIN,
2470 	 * so we need to subtract out the clone origin's used space.
2471 	 */
2472 	if (pa->origin_origin) {
2473 		pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2474 		pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2475 		pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2476 	}
2477 
2478 	/* Check that there is enough space here */
2479 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2480 	    pa->used);
2481 	if (err)
2482 		return (err);
2483 
2484 	/*
2485 	 * Compute the amounts of space that will be used by snapshots
2486 	 * after the promotion (for both origin and clone).  For each,
2487 	 * it is the amount of space that will be on all of their
2488 	 * deadlists (that was not born before their new origin).
2489 	 */
2490 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2491 		uint64_t space;
2492 
2493 		/*
2494 		 * Note, typically this will not be a clone of a clone,
2495 		 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so
2496 		 * these snaplist_space() -> bplist_space_birthrange()
2497 		 * calls will be fast because they do not have to
2498 		 * iterate over all bps.
2499 		 */
2500 		snap = list_head(&pa->origin_snaps);
2501 		err = snaplist_space(&pa->shared_snaps,
2502 		    snap->ds->ds_origin_txg, &pa->cloneusedsnap);
2503 		if (err)
2504 			return (err);
2505 
2506 		err = snaplist_space(&pa->clone_snaps,
2507 		    snap->ds->ds_origin_txg, &space);
2508 		if (err)
2509 			return (err);
2510 		pa->cloneusedsnap += space;
2511 	}
2512 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2513 		err = snaplist_space(&pa->origin_snaps,
2514 		    origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2515 		if (err)
2516 			return (err);
2517 	}
2518 
2519 	return (0);
2520 }
2521 
2522 static void
2523 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2524 {
2525 	dsl_dataset_t *hds = arg1;
2526 	struct promotearg *pa = arg2;
2527 	struct promotenode *snap = list_head(&pa->shared_snaps);
2528 	dsl_dataset_t *origin_ds = snap->ds;
2529 	dsl_dataset_t *origin_head;
2530 	dsl_dir_t *dd = hds->ds_dir;
2531 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2532 	dsl_dir_t *odd = NULL;
2533 	uint64_t oldnext_obj;
2534 	int64_t delta;
2535 
2536 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2537 
2538 	snap = list_head(&pa->origin_snaps);
2539 	origin_head = snap->ds;
2540 
2541 	/*
2542 	 * We need to explicitly open odd, since origin_ds's dd will be
2543 	 * changing.
2544 	 */
2545 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2546 	    NULL, FTAG, &odd));
2547 
2548 	/* change origin's next snap */
2549 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2550 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2551 	snap = list_tail(&pa->clone_snaps);
2552 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2553 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2554 
2555 	/* change the origin's next clone */
2556 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2557 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2558 		    origin_ds->ds_phys->ds_next_clones_obj,
2559 		    origin_ds->ds_phys->ds_next_snap_obj, tx));
2560 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2561 		    origin_ds->ds_phys->ds_next_clones_obj,
2562 		    oldnext_obj, tx));
2563 	}
2564 
2565 	/* change origin */
2566 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2567 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2568 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2569 	hds->ds_origin_txg = origin_head->ds_origin_txg;
2570 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2571 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2572 	origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg;
2573 
2574 	/* move snapshots to this dir */
2575 	for (snap = list_head(&pa->shared_snaps); snap;
2576 	    snap = list_next(&pa->shared_snaps, snap)) {
2577 		dsl_dataset_t *ds = snap->ds;
2578 
2579 		/* unregister props as dsl_dir is changing */
2580 		if (ds->ds_user_ptr) {
2581 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
2582 			ds->ds_user_ptr = NULL;
2583 		}
2584 		/* move snap name entry */
2585 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2586 		VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2587 		    ds->ds_snapname, tx));
2588 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2589 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2590 		    8, 1, &ds->ds_object, tx));
2591 		/* change containing dsl_dir */
2592 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2593 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2594 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2595 		ASSERT3P(ds->ds_dir, ==, odd);
2596 		dsl_dir_close(ds->ds_dir, ds);
2597 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2598 		    NULL, ds, &ds->ds_dir));
2599 
2600 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2601 	}
2602 
2603 	/*
2604 	 * Change space accounting.
2605 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2606 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2607 	 * is true for each of {clone,origin} independently.
2608 	 */
2609 
2610 	delta = pa->cloneusedsnap -
2611 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2612 	ASSERT3S(delta, >=, 0);
2613 	ASSERT3U(pa->used, >=, delta);
2614 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2615 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2616 	    pa->used - delta, pa->comp, pa->uncomp, tx);
2617 
2618 	delta = pa->originusedsnap -
2619 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2620 	ASSERT3S(delta, <=, 0);
2621 	ASSERT3U(pa->used, >=, -delta);
2622 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2623 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2624 	    -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2625 
2626 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2627 
2628 	/* log history record */
2629 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2630 	    cr, "dataset = %llu", hds->ds_object);
2631 
2632 	dsl_dir_close(odd, FTAG);
2633 }
2634 
2635 static char *snaplist_tag = "snaplist";
2636 /*
2637  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2638  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2639  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2640  * snapshots back to this dataset's origin.
2641  */
2642 static int
2643 snaplist_make(dsl_pool_t *dp, boolean_t own,
2644     uint64_t first_obj, uint64_t last_obj, list_t *l)
2645 {
2646 	uint64_t obj = last_obj;
2647 
2648 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2649 
2650 	list_create(l, sizeof (struct promotenode),
2651 	    offsetof(struct promotenode, link));
2652 
2653 	while (obj != first_obj) {
2654 		dsl_dataset_t *ds;
2655 		struct promotenode *snap;
2656 		int err;
2657 
2658 		if (own) {
2659 			err = dsl_dataset_own_obj(dp, obj,
2660 			    0, snaplist_tag, &ds);
2661 			if (err == 0)
2662 				dsl_dataset_make_exclusive(ds, snaplist_tag);
2663 		} else {
2664 			err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2665 		}
2666 		if (err == ENOENT) {
2667 			/* lost race with snapshot destroy */
2668 			struct promotenode *last = list_tail(l);
2669 			ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2670 			obj = last->ds->ds_phys->ds_prev_snap_obj;
2671 			continue;
2672 		} else if (err) {
2673 			return (err);
2674 		}
2675 
2676 		if (first_obj == 0)
2677 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2678 
2679 		snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2680 		snap->ds = ds;
2681 		list_insert_tail(l, snap);
2682 		obj = ds->ds_phys->ds_prev_snap_obj;
2683 	}
2684 
2685 	return (0);
2686 }
2687 
2688 static int
2689 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2690 {
2691 	struct promotenode *snap;
2692 
2693 	*spacep = 0;
2694 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2695 		uint64_t used;
2696 		int err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2697 		    mintxg, UINT64_MAX, &used);
2698 		if (err)
2699 			return (err);
2700 		*spacep += used;
2701 	}
2702 	return (0);
2703 }
2704 
2705 static void
2706 snaplist_destroy(list_t *l, boolean_t own)
2707 {
2708 	struct promotenode *snap;
2709 
2710 	if (!l || !list_link_active(&l->list_head))
2711 		return;
2712 
2713 	while ((snap = list_tail(l)) != NULL) {
2714 		list_remove(l, snap);
2715 		if (own)
2716 			dsl_dataset_disown(snap->ds, snaplist_tag);
2717 		else
2718 			dsl_dataset_rele(snap->ds, snaplist_tag);
2719 		kmem_free(snap, sizeof (struct promotenode));
2720 	}
2721 	list_destroy(l);
2722 }
2723 
2724 /*
2725  * Promote a clone.  Nomenclature note:
2726  * "clone" or "cds": the original clone which is being promoted
2727  * "origin" or "ods": the snapshot which is originally clone's origin
2728  * "origin head" or "ohds": the dataset which is the head
2729  * (filesystem/volume) for the origin
2730  * "origin origin": the origin of the origin's filesystem (typically
2731  * NULL, indicating that the clone is not a clone of a clone).
2732  */
2733 int
2734 dsl_dataset_promote(const char *name)
2735 {
2736 	dsl_dataset_t *ds;
2737 	dsl_dir_t *dd;
2738 	dsl_pool_t *dp;
2739 	dmu_object_info_t doi;
2740 	struct promotearg pa = { 0 };
2741 	struct promotenode *snap;
2742 	int err;
2743 
2744 	err = dsl_dataset_hold(name, FTAG, &ds);
2745 	if (err)
2746 		return (err);
2747 	dd = ds->ds_dir;
2748 	dp = dd->dd_pool;
2749 
2750 	err = dmu_object_info(dp->dp_meta_objset,
2751 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2752 	if (err) {
2753 		dsl_dataset_rele(ds, FTAG);
2754 		return (err);
2755 	}
2756 
2757 	if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2758 		dsl_dataset_rele(ds, FTAG);
2759 		return (EINVAL);
2760 	}
2761 
2762 	/*
2763 	 * We are going to inherit all the snapshots taken before our
2764 	 * origin (i.e., our new origin will be our parent's origin).
2765 	 * Take ownership of them so that we can rename them into our
2766 	 * namespace.
2767 	 */
2768 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2769 
2770 	err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2771 	    &pa.shared_snaps);
2772 	if (err != 0)
2773 		goto out;
2774 
2775 	err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2776 	if (err != 0)
2777 		goto out;
2778 
2779 	snap = list_head(&pa.shared_snaps);
2780 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2781 	err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2782 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2783 	if (err != 0)
2784 		goto out;
2785 
2786 	if (dsl_dir_is_clone(snap->ds->ds_dir)) {
2787 		err = dsl_dataset_own_obj(dp,
2788 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2789 		    0, FTAG, &pa.origin_origin);
2790 		if (err != 0)
2791 			goto out;
2792 	}
2793 
2794 out:
2795 	rw_exit(&dp->dp_config_rwlock);
2796 
2797 	/*
2798 	 * Add in 128x the snapnames zapobj size, since we will be moving
2799 	 * a bunch of snapnames to the promoted ds, and dirtying their
2800 	 * bonus buffers.
2801 	 */
2802 	if (err == 0) {
2803 		err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2804 		    dsl_dataset_promote_sync, ds, &pa,
2805 		    2 + 2 * doi.doi_physical_blks);
2806 	}
2807 
2808 	snaplist_destroy(&pa.shared_snaps, B_TRUE);
2809 	snaplist_destroy(&pa.clone_snaps, B_FALSE);
2810 	snaplist_destroy(&pa.origin_snaps, B_FALSE);
2811 	if (pa.origin_origin)
2812 		dsl_dataset_disown(pa.origin_origin, FTAG);
2813 	dsl_dataset_rele(ds, FTAG);
2814 	return (err);
2815 }
2816 
2817 struct cloneswaparg {
2818 	dsl_dataset_t *cds; /* clone dataset */
2819 	dsl_dataset_t *ohds; /* origin's head dataset */
2820 	boolean_t force;
2821 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2822 };
2823 
2824 /* ARGSUSED */
2825 static int
2826 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2827 {
2828 	struct cloneswaparg *csa = arg1;
2829 
2830 	/* they should both be heads */
2831 	if (dsl_dataset_is_snapshot(csa->cds) ||
2832 	    dsl_dataset_is_snapshot(csa->ohds))
2833 		return (EINVAL);
2834 
2835 	/* the branch point should be just before them */
2836 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2837 		return (EINVAL);
2838 
2839 	/* cds should be the clone (unless they are unrelated) */
2840 	if (csa->cds->ds_prev != NULL &&
2841 	    csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
2842 	    csa->ohds->ds_object !=
2843 	    csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
2844 		return (EINVAL);
2845 
2846 	/* the clone should be a child of the origin */
2847 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2848 		return (EINVAL);
2849 
2850 	/* ohds shouldn't be modified unless 'force' */
2851 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2852 		return (ETXTBSY);
2853 
2854 	/* adjust amount of any unconsumed refreservation */
2855 	csa->unused_refres_delta =
2856 	    (int64_t)MIN(csa->ohds->ds_reserved,
2857 	    csa->ohds->ds_phys->ds_unique_bytes) -
2858 	    (int64_t)MIN(csa->ohds->ds_reserved,
2859 	    csa->cds->ds_phys->ds_unique_bytes);
2860 
2861 	if (csa->unused_refres_delta > 0 &&
2862 	    csa->unused_refres_delta >
2863 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2864 		return (ENOSPC);
2865 
2866 	return (0);
2867 }
2868 
2869 /* ARGSUSED */
2870 static void
2871 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2872 {
2873 	struct cloneswaparg *csa = arg1;
2874 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2875 
2876 	ASSERT(csa->cds->ds_reserved == 0);
2877 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2878 
2879 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2880 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2881 
2882 	if (csa->cds->ds_user_ptr != NULL) {
2883 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2884 		csa->cds->ds_user_ptr = NULL;
2885 	}
2886 
2887 	if (csa->ohds->ds_user_ptr != NULL) {
2888 		csa->ohds->ds_user_evict_func(csa->ohds,
2889 		    csa->ohds->ds_user_ptr);
2890 		csa->ohds->ds_user_ptr = NULL;
2891 	}
2892 
2893 	/*
2894 	 * Reset origin's unique bytes, if it exists.
2895 	 */
2896 	if (csa->cds->ds_prev) {
2897 		dsl_dataset_t *origin = csa->cds->ds_prev;
2898 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2899 		VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2900 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2901 		    &origin->ds_phys->ds_unique_bytes));
2902 	}
2903 
2904 	/* swap blkptrs */
2905 	{
2906 		blkptr_t tmp;
2907 		tmp = csa->ohds->ds_phys->ds_bp;
2908 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2909 		csa->cds->ds_phys->ds_bp = tmp;
2910 	}
2911 
2912 	/* set dd_*_bytes */
2913 	{
2914 		int64_t dused, dcomp, duncomp;
2915 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2916 		uint64_t odl_used, odl_comp, odl_uncomp;
2917 
2918 		ASSERT3U(csa->cds->ds_dir->dd_phys->
2919 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2920 
2921 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2922 		    &cdl_comp, &cdl_uncomp));
2923 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2924 		    &odl_comp, &odl_uncomp));
2925 
2926 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2927 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2928 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2929 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2930 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2931 		    cdl_uncomp -
2932 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2933 
2934 		dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
2935 		    dused, dcomp, duncomp, tx);
2936 		dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
2937 		    -dused, -dcomp, -duncomp, tx);
2938 
2939 		/*
2940 		 * The difference in the space used by snapshots is the
2941 		 * difference in snapshot space due to the head's
2942 		 * deadlist (since that's the only thing that's
2943 		 * changing that affects the snapused).
2944 		 */
2945 		VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2946 		    csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used));
2947 		VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist,
2948 		    csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used));
2949 		dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
2950 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2951 	}
2952 
2953 #define	SWITCH64(x, y) \
2954 	{ \
2955 		uint64_t __tmp = (x); \
2956 		(x) = (y); \
2957 		(y) = __tmp; \
2958 	}
2959 
2960 	/* swap ds_*_bytes */
2961 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2962 	    csa->cds->ds_phys->ds_used_bytes);
2963 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2964 	    csa->cds->ds_phys->ds_compressed_bytes);
2965 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2966 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2967 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2968 	    csa->cds->ds_phys->ds_unique_bytes);
2969 
2970 	/* apply any parent delta for change in unconsumed refreservation */
2971 	dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
2972 	    csa->unused_refres_delta, 0, 0, tx);
2973 
2974 	/* swap deadlists */
2975 	bplist_close(&csa->cds->ds_deadlist);
2976 	bplist_close(&csa->ohds->ds_deadlist);
2977 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2978 	    csa->cds->ds_phys->ds_deadlist_obj);
2979 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2980 	    csa->cds->ds_phys->ds_deadlist_obj));
2981 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2982 	    csa->ohds->ds_phys->ds_deadlist_obj));
2983 
2984 	dsl_pool_ds_clone_swapped(csa->ohds, csa->cds, tx);
2985 }
2986 
2987 /*
2988  * Swap 'clone' with its origin head datasets.  Used at the end of "zfs
2989  * recv" into an existing fs to swizzle the file system to the new
2990  * version, and by "zfs rollback".  Can also be used to swap two
2991  * independent head datasets if neither has any snapshots.
2992  */
2993 int
2994 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2995     boolean_t force)
2996 {
2997 	struct cloneswaparg csa;
2998 	int error;
2999 
3000 	ASSERT(clone->ds_owner);
3001 	ASSERT(origin_head->ds_owner);
3002 retry:
3003 	/* Need exclusive access for the swap */
3004 	rw_enter(&clone->ds_rwlock, RW_WRITER);
3005 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3006 		rw_exit(&clone->ds_rwlock);
3007 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3008 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3009 			rw_exit(&origin_head->ds_rwlock);
3010 			goto retry;
3011 		}
3012 	}
3013 	csa.cds = clone;
3014 	csa.ohds = origin_head;
3015 	csa.force = force;
3016 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3017 	    dsl_dataset_clone_swap_check,
3018 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3019 	return (error);
3020 }
3021 
3022 /*
3023  * Given a pool name and a dataset object number in that pool,
3024  * return the name of that dataset.
3025  */
3026 int
3027 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3028 {
3029 	spa_t *spa;
3030 	dsl_pool_t *dp;
3031 	dsl_dataset_t *ds;
3032 	int error;
3033 
3034 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
3035 		return (error);
3036 	dp = spa_get_dsl(spa);
3037 	rw_enter(&dp->dp_config_rwlock, RW_READER);
3038 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3039 		dsl_dataset_name(ds, buf);
3040 		dsl_dataset_rele(ds, FTAG);
3041 	}
3042 	rw_exit(&dp->dp_config_rwlock);
3043 	spa_close(spa, FTAG);
3044 
3045 	return (error);
3046 }
3047 
3048 int
3049 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3050     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3051 {
3052 	int error = 0;
3053 
3054 	ASSERT3S(asize, >, 0);
3055 
3056 	/*
3057 	 * *ref_rsrv is the portion of asize that will come from any
3058 	 * unconsumed refreservation space.
3059 	 */
3060 	*ref_rsrv = 0;
3061 
3062 	mutex_enter(&ds->ds_lock);
3063 	/*
3064 	 * Make a space adjustment for reserved bytes.
3065 	 */
3066 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3067 		ASSERT3U(*used, >=,
3068 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3069 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3070 		*ref_rsrv =
3071 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3072 	}
3073 
3074 	if (!check_quota || ds->ds_quota == 0) {
3075 		mutex_exit(&ds->ds_lock);
3076 		return (0);
3077 	}
3078 	/*
3079 	 * If they are requesting more space, and our current estimate
3080 	 * is over quota, they get to try again unless the actual
3081 	 * on-disk is over quota and there are no pending changes (which
3082 	 * may free up space for us).
3083 	 */
3084 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3085 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3086 			error = ERESTART;
3087 		else
3088 			error = EDQUOT;
3089 	}
3090 	mutex_exit(&ds->ds_lock);
3091 
3092 	return (error);
3093 }
3094 
3095 /* ARGSUSED */
3096 static int
3097 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3098 {
3099 	dsl_dataset_t *ds = arg1;
3100 	uint64_t *quotap = arg2;
3101 	uint64_t new_quota = *quotap;
3102 
3103 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3104 		return (ENOTSUP);
3105 
3106 	if (new_quota == 0)
3107 		return (0);
3108 
3109 	if (new_quota < ds->ds_phys->ds_used_bytes ||
3110 	    new_quota < ds->ds_reserved)
3111 		return (ENOSPC);
3112 
3113 	return (0);
3114 }
3115 
3116 /* ARGSUSED */
3117 void
3118 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3119 {
3120 	dsl_dataset_t *ds = arg1;
3121 	uint64_t *quotap = arg2;
3122 	uint64_t new_quota = *quotap;
3123 
3124 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3125 
3126 	ds->ds_quota = new_quota;
3127 
3128 	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
3129 
3130 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
3131 	    tx, cr, "%lld dataset = %llu ",
3132 	    (longlong_t)new_quota, ds->ds_object);
3133 }
3134 
3135 int
3136 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
3137 {
3138 	dsl_dataset_t *ds;
3139 	int err;
3140 
3141 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3142 	if (err)
3143 		return (err);
3144 
3145 	if (quota != ds->ds_quota) {
3146 		/*
3147 		 * If someone removes a file, then tries to set the quota, we
3148 		 * want to make sure the file freeing takes effect.
3149 		 */
3150 		txg_wait_open(ds->ds_dir->dd_pool, 0);
3151 
3152 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3153 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3154 		    ds, &quota, 0);
3155 	}
3156 	dsl_dataset_rele(ds, FTAG);
3157 	return (err);
3158 }
3159 
3160 static int
3161 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3162 {
3163 	dsl_dataset_t *ds = arg1;
3164 	uint64_t *reservationp = arg2;
3165 	uint64_t new_reservation = *reservationp;
3166 	uint64_t unique;
3167 
3168 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3169 	    SPA_VERSION_REFRESERVATION)
3170 		return (ENOTSUP);
3171 
3172 	if (dsl_dataset_is_snapshot(ds))
3173 		return (EINVAL);
3174 
3175 	/*
3176 	 * If we are doing the preliminary check in open context, the
3177 	 * space estimates may be inaccurate.
3178 	 */
3179 	if (!dmu_tx_is_syncing(tx))
3180 		return (0);
3181 
3182 	mutex_enter(&ds->ds_lock);
3183 	unique = dsl_dataset_unique(ds);
3184 	mutex_exit(&ds->ds_lock);
3185 
3186 	if (MAX(unique, new_reservation) > MAX(unique, ds->ds_reserved)) {
3187 		uint64_t delta = MAX(unique, new_reservation) -
3188 		    MAX(unique, ds->ds_reserved);
3189 
3190 		if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3191 			return (ENOSPC);
3192 		if (ds->ds_quota > 0 &&
3193 		    new_reservation > ds->ds_quota)
3194 			return (ENOSPC);
3195 	}
3196 
3197 	return (0);
3198 }
3199 
3200 /* ARGSUSED */
3201 static void
3202 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
3203     dmu_tx_t *tx)
3204 {
3205 	dsl_dataset_t *ds = arg1;
3206 	uint64_t *reservationp = arg2;
3207 	uint64_t new_reservation = *reservationp;
3208 	uint64_t unique;
3209 	int64_t delta;
3210 
3211 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3212 
3213 	mutex_enter(&ds->ds_dir->dd_lock);
3214 	mutex_enter(&ds->ds_lock);
3215 	unique = dsl_dataset_unique(ds);
3216 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
3217 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3218 	ds->ds_reserved = new_reservation;
3219 	mutex_exit(&ds->ds_lock);
3220 
3221 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3222 	mutex_exit(&ds->ds_dir->dd_lock);
3223 	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refreservation",
3224 	    new_reservation, cr, tx);
3225 
3226 	spa_history_internal_log(LOG_DS_REFRESERV,
3227 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
3228 	    (longlong_t)new_reservation, ds->ds_object);
3229 }
3230 
3231 int
3232 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
3233 {
3234 	dsl_dataset_t *ds;
3235 	int err;
3236 
3237 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3238 	if (err)
3239 		return (err);
3240 
3241 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3242 	    dsl_dataset_set_reservation_check,
3243 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
3244 	dsl_dataset_rele(ds, FTAG);
3245 	return (err);
3246 }
3247 
3248 static int
3249 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3250 {
3251 	dsl_dataset_t *ds = arg1;
3252 	char *htag = arg2;
3253 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3254 	int error = 0;
3255 
3256 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3257 		return (ENOTSUP);
3258 
3259 	if (!dsl_dataset_is_snapshot(ds))
3260 		return (EINVAL);
3261 
3262 	if (strlen(htag) >= ZAP_MAXNAMELEN)
3263 		return (ENAMETOOLONG);
3264 
3265 	/* tags must be unique */
3266 	mutex_enter(&ds->ds_lock);
3267 	if (ds->ds_phys->ds_userrefs_obj) {
3268 		error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3269 		    8, 1, tx);
3270 		if (error == 0)
3271 			error = EEXIST;
3272 		else if (error == ENOENT)
3273 			error = 0;
3274 	}
3275 	mutex_exit(&ds->ds_lock);
3276 
3277 	return (error);
3278 }
3279 
3280 static void
3281 dsl_dataset_user_hold_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3282 {
3283 	dsl_dataset_t *ds = arg1;
3284 	char *htag = arg2;
3285 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3286 	time_t now = gethrestime_sec();
3287 	uint64_t zapobj;
3288 
3289 	mutex_enter(&ds->ds_lock);
3290 	if (ds->ds_phys->ds_userrefs_obj == 0) {
3291 		/*
3292 		 * This is the first user hold for this dataset.  Create
3293 		 * the userrefs zap object.
3294 		 */
3295 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3296 		zapobj = ds->ds_phys->ds_userrefs_obj =
3297 		    zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3298 	} else {
3299 		zapobj = ds->ds_phys->ds_userrefs_obj;
3300 	}
3301 	ds->ds_userrefs++;
3302 	mutex_exit(&ds->ds_lock);
3303 
3304 	VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3305 
3306 	spa_history_internal_log(LOG_DS_USER_HOLD,
3307 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "<%s> dataset = %llu",
3308 	    htag, ds->ds_object);
3309 }
3310 
3311 struct dsl_ds_holdarg {
3312 	dsl_sync_task_group_t *dstg;
3313 	char *htag;
3314 	char *snapname;
3315 	boolean_t recursive;
3316 	boolean_t gotone;
3317 	char failed[MAXPATHLEN];
3318 };
3319 
3320 static int
3321 dsl_dataset_user_hold_one(char *dsname, void *arg)
3322 {
3323 	struct dsl_ds_holdarg *ha = arg;
3324 	dsl_dataset_t *ds;
3325 	int error;
3326 	char *name;
3327 
3328 	/* alloc a buffer to hold dsname@snapname plus terminating NULL */
3329 	name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3330 	error = dsl_dataset_hold(name, ha->dstg, &ds);
3331 	strfree(name);
3332 	if (error == 0) {
3333 		ha->gotone = B_TRUE;
3334 		dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3335 		    dsl_dataset_user_hold_sync, ds, ha->htag, 0);
3336 	} else if (error == ENOENT && ha->recursive) {
3337 		error = 0;
3338 	} else {
3339 		(void) strcpy(ha->failed, dsname);
3340 	}
3341 	return (error);
3342 }
3343 
3344 int
3345 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3346     boolean_t recursive)
3347 {
3348 	struct dsl_ds_holdarg *ha;
3349 	dsl_sync_task_t *dst;
3350 	spa_t *spa;
3351 	int error;
3352 
3353 	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3354 
3355 	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3356 
3357 	error = spa_open(dsname, &spa, FTAG);
3358 	if (error) {
3359 		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3360 		return (error);
3361 	}
3362 
3363 	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3364 	ha->htag = htag;
3365 	ha->snapname = snapname;
3366 	ha->recursive = recursive;
3367 	if (recursive) {
3368 		error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3369 		    ha, DS_FIND_CHILDREN);
3370 	} else {
3371 		error = dsl_dataset_user_hold_one(dsname, ha);
3372 	}
3373 	if (error == 0)
3374 		error = dsl_sync_task_group_wait(ha->dstg);
3375 
3376 	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3377 	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3378 		dsl_dataset_t *ds = dst->dst_arg1;
3379 
3380 		if (dst->dst_err) {
3381 			dsl_dataset_name(ds, ha->failed);
3382 			*strchr(ha->failed, '@') = '\0';
3383 		}
3384 		dsl_dataset_rele(ds, ha->dstg);
3385 	}
3386 
3387 	if (error == 0 && recursive && !ha->gotone)
3388 		error = ENOENT;
3389 
3390 	if (error)
3391 		(void) strcpy(dsname, ha->failed);
3392 
3393 	dsl_sync_task_group_destroy(ha->dstg);
3394 	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3395 	spa_close(spa, FTAG);
3396 	return (error);
3397 }
3398 
3399 struct dsl_ds_releasearg {
3400 	dsl_dataset_t *ds;
3401 	const char *htag;
3402 	boolean_t own;		/* do we own or just hold ds? */
3403 };
3404 
3405 static int
3406 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3407     boolean_t *might_destroy)
3408 {
3409 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3410 	uint64_t zapobj;
3411 	uint64_t tmp;
3412 	int error;
3413 
3414 	*might_destroy = B_FALSE;
3415 
3416 	mutex_enter(&ds->ds_lock);
3417 	zapobj = ds->ds_phys->ds_userrefs_obj;
3418 	if (zapobj == 0) {
3419 		/* The tag can't possibly exist */
3420 		mutex_exit(&ds->ds_lock);
3421 		return (ESRCH);
3422 	}
3423 
3424 	/* Make sure the tag exists */
3425 	error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3426 	if (error) {
3427 		mutex_exit(&ds->ds_lock);
3428 		if (error == ENOENT)
3429 			error = ESRCH;
3430 		return (error);
3431 	}
3432 
3433 	if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3434 	    DS_IS_DEFER_DESTROY(ds))
3435 		*might_destroy = B_TRUE;
3436 
3437 	mutex_exit(&ds->ds_lock);
3438 	return (0);
3439 }
3440 
3441 static int
3442 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3443 {
3444 	struct dsl_ds_releasearg *ra = arg1;
3445 	dsl_dataset_t *ds = ra->ds;
3446 	boolean_t might_destroy;
3447 	int error;
3448 
3449 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3450 		return (ENOTSUP);
3451 
3452 	error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3453 	if (error)
3454 		return (error);
3455 
3456 	if (might_destroy) {
3457 		struct dsl_ds_destroyarg dsda = {0};
3458 
3459 		if (dmu_tx_is_syncing(tx)) {
3460 			/*
3461 			 * If we're not prepared to remove the snapshot,
3462 			 * we can't allow the release to happen right now.
3463 			 */
3464 			if (!ra->own)
3465 				return (EBUSY);
3466 			if (ds->ds_user_ptr) {
3467 				ds->ds_user_evict_func(ds, ds->ds_user_ptr);
3468 				ds->ds_user_ptr = NULL;
3469 			}
3470 		}
3471 		dsda.ds = ds;
3472 		dsda.releasing = B_TRUE;
3473 		return (dsl_dataset_destroy_check(&dsda, tag, tx));
3474 	}
3475 
3476 	return (0);
3477 }
3478 
3479 static void
3480 dsl_dataset_user_release_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
3481 {
3482 	struct dsl_ds_releasearg *ra = arg1;
3483 	dsl_dataset_t *ds = ra->ds;
3484 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
3485 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3486 	uint64_t zapobj;
3487 	uint64_t dsobj = ds->ds_object;
3488 	uint64_t refs;
3489 
3490 	mutex_enter(&ds->ds_lock);
3491 	ds->ds_userrefs--;
3492 	refs = ds->ds_userrefs;
3493 	mutex_exit(&ds->ds_lock);
3494 	zapobj = ds->ds_phys->ds_userrefs_obj;
3495 	VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3496 	if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3497 	    DS_IS_DEFER_DESTROY(ds)) {
3498 		struct dsl_ds_destroyarg dsda = {0};
3499 
3500 		ASSERT(ra->own);
3501 		dsda.ds = ds;
3502 		dsda.releasing = B_TRUE;
3503 		/* We already did the destroy_check */
3504 		dsl_dataset_destroy_sync(&dsda, tag, cr, tx);
3505 	}
3506 
3507 	spa_history_internal_log(LOG_DS_USER_RELEASE,
3508 	    spa, tx, cr, "<%s> %lld dataset = %llu",
3509 	    ra->htag, (longlong_t)refs, dsobj);
3510 }
3511 
3512 static int
3513 dsl_dataset_user_release_one(char *dsname, void *arg)
3514 {
3515 	struct dsl_ds_holdarg *ha = arg;
3516 	struct dsl_ds_releasearg *ra;
3517 	dsl_dataset_t *ds;
3518 	int error;
3519 	void *dtag = ha->dstg;
3520 	char *name;
3521 	boolean_t own = B_FALSE;
3522 	boolean_t might_destroy;
3523 
3524 	if (strlen(ha->htag) >= ZAP_MAXNAMELEN)
3525 		return (ENAMETOOLONG);
3526 
3527 	/* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3528 	name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3529 	error = dsl_dataset_hold(name, dtag, &ds);
3530 	strfree(name);
3531 	if (error == ENOENT && ha->recursive)
3532 		return (0);
3533 	(void) strcpy(ha->failed, dsname);
3534 	if (error)
3535 		return (error);
3536 
3537 	ha->gotone = B_TRUE;
3538 
3539 	ASSERT(dsl_dataset_is_snapshot(ds));
3540 
3541 	error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3542 	if (error) {
3543 		dsl_dataset_rele(ds, dtag);
3544 		return (error);
3545 	}
3546 
3547 	if (might_destroy) {
3548 #ifdef _KERNEL
3549 		error = zfs_unmount_snap(name, NULL);
3550 		if (error) {
3551 			dsl_dataset_rele(ds, dtag);
3552 			return (error);
3553 		}
3554 		error = dsl_dataset_zvol_cleanup(ds, name);
3555 		if (error) {
3556 			dsl_dataset_rele(ds, dtag);
3557 			return (error);
3558 		}
3559 #endif
3560 		if (!dsl_dataset_tryown(ds,
3561 		    DS_MODE_READONLY | DS_MODE_INCONSISTENT, dtag)) {
3562 			dsl_dataset_rele(ds, dtag);
3563 			return (EBUSY);
3564 		} else {
3565 			own = B_TRUE;
3566 			dsl_dataset_make_exclusive(ds, dtag);
3567 		}
3568 	}
3569 
3570 	ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3571 	ra->ds = ds;
3572 	ra->htag = ha->htag;
3573 	ra->own = own;
3574 	dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3575 	    dsl_dataset_user_release_sync, ra, dtag, 0);
3576 
3577 	return (0);
3578 }
3579 
3580 int
3581 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3582     boolean_t recursive)
3583 {
3584 	struct dsl_ds_holdarg *ha;
3585 	dsl_sync_task_t *dst;
3586 	spa_t *spa;
3587 	int error;
3588 
3589 	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3590 
3591 	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3592 
3593 	error = spa_open(dsname, &spa, FTAG);
3594 	if (error) {
3595 		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3596 		return (error);
3597 	}
3598 
3599 	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3600 	ha->htag = htag;
3601 	ha->snapname = snapname;
3602 	ha->recursive = recursive;
3603 	if (recursive) {
3604 		error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3605 		    ha, DS_FIND_CHILDREN);
3606 	} else {
3607 		error = dsl_dataset_user_release_one(dsname, ha);
3608 	}
3609 	if (error == 0)
3610 		error = dsl_sync_task_group_wait(ha->dstg);
3611 
3612 	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3613 	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3614 		struct dsl_ds_releasearg *ra = dst->dst_arg1;
3615 		dsl_dataset_t *ds = ra->ds;
3616 
3617 		if (dst->dst_err)
3618 			dsl_dataset_name(ds, ha->failed);
3619 
3620 		if (ra->own)
3621 			dsl_dataset_disown(ds, ha->dstg);
3622 		else
3623 			dsl_dataset_rele(ds, ha->dstg);
3624 
3625 		kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3626 	}
3627 
3628 	if (error == 0 && recursive && !ha->gotone)
3629 		error = ENOENT;
3630 
3631 	if (error)
3632 		(void) strcpy(dsname, ha->failed);
3633 
3634 	dsl_sync_task_group_destroy(ha->dstg);
3635 	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3636 	spa_close(spa, FTAG);
3637 	return (error);
3638 }
3639 
3640 int
3641 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
3642 {
3643 	dsl_dataset_t *ds;
3644 	int err;
3645 
3646 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3647 	if (err)
3648 		return (err);
3649 
3650 	VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
3651 	if (ds->ds_phys->ds_userrefs_obj != 0) {
3652 		zap_attribute_t *za;
3653 		zap_cursor_t zc;
3654 
3655 		za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
3656 		for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
3657 		    ds->ds_phys->ds_userrefs_obj);
3658 		    zap_cursor_retrieve(&zc, za) == 0;
3659 		    zap_cursor_advance(&zc)) {
3660 			VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
3661 			    za->za_first_integer));
3662 		}
3663 		zap_cursor_fini(&zc);
3664 		kmem_free(za, sizeof (zap_attribute_t));
3665 	}
3666 	dsl_dataset_rele(ds, FTAG);
3667 	return (0);
3668 }
3669