xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision aebdfce800eec0973de8dbcd7ec522064210ee21)
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_objset != NULL)
248 		dmu_objset_evict(ds->ds_objset);
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 	dsl_dir_snap_cmtime_update(ds->ds_dir);
325 
326 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
327 		mt = MT_FIRST;
328 	else
329 		mt = MT_EXACT;
330 
331 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
332 	if (err == ENOTSUP && mt == MT_FIRST)
333 		err = zap_remove(mos, snapobj, name, tx);
334 	return (err);
335 }
336 
337 static int
338 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
339     dsl_dataset_t **dsp)
340 {
341 	objset_t *mos = dp->dp_meta_objset;
342 	dmu_buf_t *dbuf;
343 	dsl_dataset_t *ds;
344 	int err;
345 
346 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
347 	    dsl_pool_sync_context(dp));
348 
349 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
350 	if (err)
351 		return (err);
352 	ds = dmu_buf_get_user(dbuf);
353 	if (ds == NULL) {
354 		dsl_dataset_t *winner;
355 
356 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
357 		ds->ds_dbuf = dbuf;
358 		ds->ds_object = dsobj;
359 		ds->ds_phys = dbuf->db_data;
360 
361 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
362 		mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
363 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
364 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
365 		    NULL);
366 		rw_init(&ds->ds_rwlock, 0, 0, 0);
367 		cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
368 
369 		err = bplist_open(&ds->ds_deadlist,
370 		    mos, ds->ds_phys->ds_deadlist_obj);
371 		if (err == 0) {
372 			err = dsl_dir_open_obj(dp,
373 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
374 		}
375 		if (err) {
376 			/*
377 			 * we don't really need to close the blist if we
378 			 * just opened it.
379 			 */
380 			mutex_destroy(&ds->ds_lock);
381 			mutex_destroy(&ds->ds_recvlock);
382 			mutex_destroy(&ds->ds_opening_lock);
383 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
384 			rw_destroy(&ds->ds_rwlock);
385 			cv_destroy(&ds->ds_exclusive_cv);
386 			kmem_free(ds, sizeof (dsl_dataset_t));
387 			dmu_buf_rele(dbuf, tag);
388 			return (err);
389 		}
390 
391 		if (!dsl_dataset_is_snapshot(ds)) {
392 			ds->ds_snapname[0] = '\0';
393 			if (ds->ds_phys->ds_prev_snap_obj) {
394 				err = dsl_dataset_get_ref(dp,
395 				    ds->ds_phys->ds_prev_snap_obj,
396 				    ds, &ds->ds_prev);
397 			}
398 
399 			if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) {
400 				dsl_dataset_t *origin;
401 
402 				err = dsl_dataset_hold_obj(dp,
403 				    ds->ds_dir->dd_phys->dd_origin_obj,
404 				    FTAG, &origin);
405 				if (err == 0) {
406 					ds->ds_origin_txg =
407 					    origin->ds_phys->ds_creation_txg;
408 					dsl_dataset_rele(origin, FTAG);
409 				}
410 			}
411 		} else {
412 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
413 				err = dsl_dataset_get_snapname(ds);
414 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
415 				err = zap_count(
416 				    ds->ds_dir->dd_pool->dp_meta_objset,
417 				    ds->ds_phys->ds_userrefs_obj,
418 				    &ds->ds_userrefs);
419 			}
420 		}
421 
422 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
423 			/*
424 			 * In sync context, we're called with either no lock
425 			 * or with the write lock.  If we're not syncing,
426 			 * we're always called with the read lock held.
427 			 */
428 			boolean_t need_lock =
429 			    !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
430 			    dsl_pool_sync_context(dp);
431 
432 			if (need_lock)
433 				rw_enter(&dp->dp_config_rwlock, RW_READER);
434 
435 			err = dsl_prop_get_ds(ds,
436 			    "refreservation", sizeof (uint64_t), 1,
437 			    &ds->ds_reserved, NULL);
438 			if (err == 0) {
439 				err = dsl_prop_get_ds(ds,
440 				    "refquota", sizeof (uint64_t), 1,
441 				    &ds->ds_quota, NULL);
442 			}
443 
444 			if (need_lock)
445 				rw_exit(&dp->dp_config_rwlock);
446 		} else {
447 			ds->ds_reserved = ds->ds_quota = 0;
448 		}
449 
450 		if (err == 0) {
451 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
452 			    dsl_dataset_evict);
453 		}
454 		if (err || winner) {
455 			bplist_close(&ds->ds_deadlist);
456 			if (ds->ds_prev)
457 				dsl_dataset_drop_ref(ds->ds_prev, ds);
458 			dsl_dir_close(ds->ds_dir, ds);
459 			mutex_destroy(&ds->ds_lock);
460 			mutex_destroy(&ds->ds_recvlock);
461 			mutex_destroy(&ds->ds_opening_lock);
462 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
463 			rw_destroy(&ds->ds_rwlock);
464 			cv_destroy(&ds->ds_exclusive_cv);
465 			kmem_free(ds, sizeof (dsl_dataset_t));
466 			if (err) {
467 				dmu_buf_rele(dbuf, tag);
468 				return (err);
469 			}
470 			ds = winner;
471 		} else {
472 			ds->ds_fsid_guid =
473 			    unique_insert(ds->ds_phys->ds_fsid_guid);
474 		}
475 	}
476 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
477 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
478 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
479 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
480 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
481 	mutex_enter(&ds->ds_lock);
482 	if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
483 		mutex_exit(&ds->ds_lock);
484 		dmu_buf_rele(ds->ds_dbuf, tag);
485 		return (ENOENT);
486 	}
487 	mutex_exit(&ds->ds_lock);
488 	*dsp = ds;
489 	return (0);
490 }
491 
492 static int
493 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
494 {
495 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
496 
497 	/*
498 	 * In syncing context we don't want the rwlock lock: there
499 	 * may be an existing writer waiting for sync phase to
500 	 * finish.  We don't need to worry about such writers, since
501 	 * sync phase is single-threaded, so the writer can't be
502 	 * doing anything while we are active.
503 	 */
504 	if (dsl_pool_sync_context(dp)) {
505 		ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
506 		return (0);
507 	}
508 
509 	/*
510 	 * Normal users will hold the ds_rwlock as a READER until they
511 	 * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
512 	 * drop their READER lock after they set the ds_owner field.
513 	 *
514 	 * If the dataset is being destroyed, the destroy thread will
515 	 * obtain a WRITER lock for exclusive access after it's done its
516 	 * open-context work and then change the ds_owner to
517 	 * dsl_reaper once destruction is assured.  So threads
518 	 * may block here temporarily, until the "destructability" of
519 	 * the dataset is determined.
520 	 */
521 	ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
522 	mutex_enter(&ds->ds_lock);
523 	while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
524 		rw_exit(&dp->dp_config_rwlock);
525 		cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
526 		if (DSL_DATASET_IS_DESTROYED(ds)) {
527 			mutex_exit(&ds->ds_lock);
528 			dsl_dataset_drop_ref(ds, tag);
529 			rw_enter(&dp->dp_config_rwlock, RW_READER);
530 			return (ENOENT);
531 		}
532 		rw_enter(&dp->dp_config_rwlock, RW_READER);
533 	}
534 	mutex_exit(&ds->ds_lock);
535 	return (0);
536 }
537 
538 int
539 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
540     dsl_dataset_t **dsp)
541 {
542 	int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
543 
544 	if (err)
545 		return (err);
546 	return (dsl_dataset_hold_ref(*dsp, tag));
547 }
548 
549 int
550 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
551     void *tag, dsl_dataset_t **dsp)
552 {
553 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
554 	if (err)
555 		return (err);
556 	if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
557 		dsl_dataset_rele(*dsp, tag);
558 		*dsp = NULL;
559 		return (EBUSY);
560 	}
561 	return (0);
562 }
563 
564 int
565 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
566 {
567 	dsl_dir_t *dd;
568 	dsl_pool_t *dp;
569 	const char *snapname;
570 	uint64_t obj;
571 	int err = 0;
572 
573 	err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
574 	if (err)
575 		return (err);
576 
577 	dp = dd->dd_pool;
578 	obj = dd->dd_phys->dd_head_dataset_obj;
579 	rw_enter(&dp->dp_config_rwlock, RW_READER);
580 	if (obj)
581 		err = dsl_dataset_get_ref(dp, obj, tag, dsp);
582 	else
583 		err = ENOENT;
584 	if (err)
585 		goto out;
586 
587 	err = dsl_dataset_hold_ref(*dsp, tag);
588 
589 	/* we may be looking for a snapshot */
590 	if (err == 0 && snapname != NULL) {
591 		dsl_dataset_t *ds = NULL;
592 
593 		if (*snapname++ != '@') {
594 			dsl_dataset_rele(*dsp, tag);
595 			err = ENOENT;
596 			goto out;
597 		}
598 
599 		dprintf("looking for snapshot '%s'\n", snapname);
600 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
601 		if (err == 0)
602 			err = dsl_dataset_get_ref(dp, obj, tag, &ds);
603 		dsl_dataset_rele(*dsp, tag);
604 
605 		ASSERT3U((err == 0), ==, (ds != NULL));
606 
607 		if (ds) {
608 			mutex_enter(&ds->ds_lock);
609 			if (ds->ds_snapname[0] == 0)
610 				(void) strlcpy(ds->ds_snapname, snapname,
611 				    sizeof (ds->ds_snapname));
612 			mutex_exit(&ds->ds_lock);
613 			err = dsl_dataset_hold_ref(ds, tag);
614 			*dsp = err ? NULL : ds;
615 		}
616 	}
617 out:
618 	rw_exit(&dp->dp_config_rwlock);
619 	dsl_dir_close(dd, FTAG);
620 	return (err);
621 }
622 
623 int
624 dsl_dataset_own(const char *name, boolean_t inconsistentok,
625     void *tag, dsl_dataset_t **dsp)
626 {
627 	int err = dsl_dataset_hold(name, tag, dsp);
628 	if (err)
629 		return (err);
630 	if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
631 		dsl_dataset_rele(*dsp, tag);
632 		return (EBUSY);
633 	}
634 	return (0);
635 }
636 
637 void
638 dsl_dataset_name(dsl_dataset_t *ds, char *name)
639 {
640 	if (ds == NULL) {
641 		(void) strcpy(name, "mos");
642 	} else {
643 		dsl_dir_name(ds->ds_dir, name);
644 		VERIFY(0 == dsl_dataset_get_snapname(ds));
645 		if (ds->ds_snapname[0]) {
646 			(void) strcat(name, "@");
647 			/*
648 			 * We use a "recursive" mutex so that we
649 			 * can call dprintf_ds() with ds_lock held.
650 			 */
651 			if (!MUTEX_HELD(&ds->ds_lock)) {
652 				mutex_enter(&ds->ds_lock);
653 				(void) strcat(name, ds->ds_snapname);
654 				mutex_exit(&ds->ds_lock);
655 			} else {
656 				(void) strcat(name, ds->ds_snapname);
657 			}
658 		}
659 	}
660 }
661 
662 static int
663 dsl_dataset_namelen(dsl_dataset_t *ds)
664 {
665 	int result;
666 
667 	if (ds == NULL) {
668 		result = 3;	/* "mos" */
669 	} else {
670 		result = dsl_dir_namelen(ds->ds_dir);
671 		VERIFY(0 == dsl_dataset_get_snapname(ds));
672 		if (ds->ds_snapname[0]) {
673 			++result;	/* adding one for the @-sign */
674 			if (!MUTEX_HELD(&ds->ds_lock)) {
675 				mutex_enter(&ds->ds_lock);
676 				result += strlen(ds->ds_snapname);
677 				mutex_exit(&ds->ds_lock);
678 			} else {
679 				result += strlen(ds->ds_snapname);
680 			}
681 		}
682 	}
683 
684 	return (result);
685 }
686 
687 void
688 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
689 {
690 	dmu_buf_rele(ds->ds_dbuf, tag);
691 }
692 
693 void
694 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
695 {
696 	if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
697 		rw_exit(&ds->ds_rwlock);
698 	}
699 	dsl_dataset_drop_ref(ds, tag);
700 }
701 
702 void
703 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
704 {
705 	ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
706 	    (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
707 
708 	mutex_enter(&ds->ds_lock);
709 	ds->ds_owner = NULL;
710 	if (RW_WRITE_HELD(&ds->ds_rwlock)) {
711 		rw_exit(&ds->ds_rwlock);
712 		cv_broadcast(&ds->ds_exclusive_cv);
713 	}
714 	mutex_exit(&ds->ds_lock);
715 	if (ds->ds_dbuf)
716 		dsl_dataset_drop_ref(ds, tag);
717 	else
718 		dsl_dataset_evict(ds->ds_dbuf, ds);
719 }
720 
721 boolean_t
722 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
723 {
724 	boolean_t gotit = FALSE;
725 
726 	mutex_enter(&ds->ds_lock);
727 	if (ds->ds_owner == NULL &&
728 	    (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
729 		ds->ds_owner = tag;
730 		if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
731 			rw_exit(&ds->ds_rwlock);
732 		gotit = TRUE;
733 	}
734 	mutex_exit(&ds->ds_lock);
735 	return (gotit);
736 }
737 
738 void
739 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
740 {
741 	ASSERT3P(owner, ==, ds->ds_owner);
742 	if (!RW_WRITE_HELD(&ds->ds_rwlock))
743 		rw_enter(&ds->ds_rwlock, RW_WRITER);
744 }
745 
746 uint64_t
747 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
748     uint64_t flags, dmu_tx_t *tx)
749 {
750 	dsl_pool_t *dp = dd->dd_pool;
751 	dmu_buf_t *dbuf;
752 	dsl_dataset_phys_t *dsphys;
753 	uint64_t dsobj;
754 	objset_t *mos = dp->dp_meta_objset;
755 
756 	if (origin == NULL)
757 		origin = dp->dp_origin_snap;
758 
759 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
760 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
761 	ASSERT(dmu_tx_is_syncing(tx));
762 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
763 
764 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
765 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
766 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
767 	dmu_buf_will_dirty(dbuf, tx);
768 	dsphys = dbuf->db_data;
769 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
770 	dsphys->ds_dir_obj = dd->dd_object;
771 	dsphys->ds_flags = flags;
772 	dsphys->ds_fsid_guid = unique_create();
773 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
774 	    sizeof (dsphys->ds_guid));
775 	dsphys->ds_snapnames_zapobj =
776 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
777 	    DMU_OT_NONE, 0, tx);
778 	dsphys->ds_creation_time = gethrestime_sec();
779 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
780 	dsphys->ds_deadlist_obj =
781 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
782 
783 	if (origin) {
784 		dsphys->ds_prev_snap_obj = origin->ds_object;
785 		dsphys->ds_prev_snap_txg =
786 		    origin->ds_phys->ds_creation_txg;
787 		dsphys->ds_used_bytes =
788 		    origin->ds_phys->ds_used_bytes;
789 		dsphys->ds_compressed_bytes =
790 		    origin->ds_phys->ds_compressed_bytes;
791 		dsphys->ds_uncompressed_bytes =
792 		    origin->ds_phys->ds_uncompressed_bytes;
793 		dsphys->ds_bp = origin->ds_phys->ds_bp;
794 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
795 
796 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
797 		origin->ds_phys->ds_num_children++;
798 
799 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
800 			if (origin->ds_phys->ds_next_clones_obj == 0) {
801 				origin->ds_phys->ds_next_clones_obj =
802 				    zap_create(mos,
803 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
804 			}
805 			VERIFY(0 == zap_add_int(mos,
806 			    origin->ds_phys->ds_next_clones_obj,
807 			    dsobj, tx));
808 		}
809 
810 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
811 		dd->dd_phys->dd_origin_obj = origin->ds_object;
812 	}
813 
814 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
815 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
816 
817 	dmu_buf_rele(dbuf, FTAG);
818 
819 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
820 	dd->dd_phys->dd_head_dataset_obj = dsobj;
821 
822 	return (dsobj);
823 }
824 
825 uint64_t
826 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
827     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
828 {
829 	dsl_pool_t *dp = pdd->dd_pool;
830 	uint64_t dsobj, ddobj;
831 	dsl_dir_t *dd;
832 
833 	ASSERT(lastname[0] != '@');
834 
835 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
836 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
837 
838 	dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
839 
840 	dsl_deleg_set_create_perms(dd, tx, cr);
841 
842 	dsl_dir_close(dd, FTAG);
843 
844 	return (dsobj);
845 }
846 
847 struct destroyarg {
848 	dsl_sync_task_group_t *dstg;
849 	char *snapname;
850 	char *failed;
851 	boolean_t defer;
852 };
853 
854 static int
855 dsl_snapshot_destroy_one(char *name, void *arg)
856 {
857 	struct destroyarg *da = arg;
858 	dsl_dataset_t *ds;
859 	int err;
860 	char *dsname;
861 
862 	dsname = kmem_asprintf("%s@%s", name, da->snapname);
863 	err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds);
864 	strfree(dsname);
865 	if (err == 0) {
866 		struct dsl_ds_destroyarg *dsda;
867 
868 		dsl_dataset_make_exclusive(ds, da->dstg);
869 		if (ds->ds_objset != NULL) {
870 			dmu_objset_evict(ds->ds_objset);
871 			ds->ds_objset = NULL;
872 		}
873 		dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
874 		dsda->ds = ds;
875 		dsda->defer = da->defer;
876 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
877 		    dsl_dataset_destroy_sync, dsda, da->dstg, 0);
878 	} else if (err == ENOENT) {
879 		err = 0;
880 	} else {
881 		(void) strcpy(da->failed, name);
882 	}
883 	return (err);
884 }
885 
886 /*
887  * Destroy 'snapname' in all descendants of 'fsname'.
888  */
889 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
890 int
891 dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
892 {
893 	int err;
894 	struct destroyarg da;
895 	dsl_sync_task_t *dst;
896 	spa_t *spa;
897 
898 	err = spa_open(fsname, &spa, FTAG);
899 	if (err)
900 		return (err);
901 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
902 	da.snapname = snapname;
903 	da.failed = fsname;
904 	da.defer = defer;
905 
906 	err = dmu_objset_find(fsname,
907 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
908 
909 	if (err == 0)
910 		err = dsl_sync_task_group_wait(da.dstg);
911 
912 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
913 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
914 		struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
915 		dsl_dataset_t *ds = dsda->ds;
916 
917 		/*
918 		 * Return the file system name that triggered the error
919 		 */
920 		if (dst->dst_err) {
921 			dsl_dataset_name(ds, fsname);
922 			*strchr(fsname, '@') = '\0';
923 		}
924 		ASSERT3P(dsda->rm_origin, ==, NULL);
925 		dsl_dataset_disown(ds, da.dstg);
926 		kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
927 	}
928 
929 	dsl_sync_task_group_destroy(da.dstg);
930 	spa_close(spa, FTAG);
931 	return (err);
932 }
933 
934 static boolean_t
935 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
936 {
937 	boolean_t might_destroy = B_FALSE;
938 
939 	mutex_enter(&ds->ds_lock);
940 	if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
941 	    DS_IS_DEFER_DESTROY(ds))
942 		might_destroy = B_TRUE;
943 	mutex_exit(&ds->ds_lock);
944 
945 	return (might_destroy);
946 }
947 
948 #ifdef _KERNEL
949 static int
950 dsl_dataset_zvol_cleanup(dsl_dataset_t *ds, const char *name)
951 {
952 	int error;
953 	objset_t *os;
954 
955 	error = dmu_objset_from_ds(ds, &os);
956 	if (error)
957 		return (error);
958 
959 	if (dmu_objset_type(os) == DMU_OST_ZVOL)
960 		error = zvol_remove_minor(name);
961 
962 	return (error);
963 }
964 #endif
965 
966 /*
967  * If we're removing a clone, and these three conditions are true:
968  *	1) the clone's origin has no other children
969  *	2) the clone's origin has no user references
970  *	3) the clone's origin has been marked for deferred destruction
971  * Then, prepare to remove the origin as part of this sync task group.
972  */
973 static int
974 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
975 {
976 	dsl_dataset_t *ds = dsda->ds;
977 	dsl_dataset_t *origin = ds->ds_prev;
978 
979 	if (dsl_dataset_might_destroy_origin(origin)) {
980 		char *name;
981 		int namelen;
982 		int error;
983 
984 		namelen = dsl_dataset_namelen(origin) + 1;
985 		name = kmem_alloc(namelen, KM_SLEEP);
986 		dsl_dataset_name(origin, name);
987 #ifdef _KERNEL
988 		error = zfs_unmount_snap(name, NULL);
989 		if (error) {
990 			kmem_free(name, namelen);
991 			return (error);
992 		}
993 		error = dsl_dataset_zvol_cleanup(origin, name);
994 		if (error) {
995 			kmem_free(name, namelen);
996 			return (error);
997 		}
998 #endif
999 		error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1000 		kmem_free(name, namelen);
1001 		if (error)
1002 			return (error);
1003 		dsda->rm_origin = origin;
1004 		dsl_dataset_make_exclusive(origin, tag);
1005 
1006 		if (origin->ds_objset != NULL) {
1007 			dmu_objset_evict(origin->ds_objset);
1008 			origin->ds_objset = NULL;
1009 		}
1010 	}
1011 
1012 	return (0);
1013 }
1014 
1015 /*
1016  * ds must be opened as OWNER.  On return (whether successful or not),
1017  * ds will be closed and caller can no longer dereference it.
1018  */
1019 int
1020 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1021 {
1022 	int err;
1023 	dsl_sync_task_group_t *dstg;
1024 	objset_t *os;
1025 	dsl_dir_t *dd;
1026 	uint64_t obj;
1027 	struct dsl_ds_destroyarg dsda = {0};
1028 
1029 	dsda.ds = ds;
1030 
1031 	if (dsl_dataset_is_snapshot(ds)) {
1032 		/* Destroying a snapshot is simpler */
1033 		dsl_dataset_make_exclusive(ds, tag);
1034 
1035 		if (ds->ds_objset != NULL) {
1036 			dmu_objset_evict(ds->ds_objset);
1037 			ds->ds_objset = NULL;
1038 		}
1039 		/* NOTE: defer is always B_FALSE for non-snapshots */
1040 		dsda.defer = defer;
1041 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1042 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1043 		    &dsda, tag, 0);
1044 		ASSERT3P(dsda.rm_origin, ==, NULL);
1045 		goto out;
1046 	}
1047 
1048 	dd = ds->ds_dir;
1049 
1050 	/*
1051 	 * Check for errors and mark this ds as inconsistent, in
1052 	 * case we crash while freeing the objects.
1053 	 */
1054 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1055 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1056 	if (err)
1057 		goto out;
1058 
1059 	err = dmu_objset_from_ds(ds, &os);
1060 	if (err)
1061 		goto out;
1062 
1063 	/*
1064 	 * remove the objects in open context, so that we won't
1065 	 * have too much to do in syncing context.
1066 	 */
1067 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1068 	    ds->ds_phys->ds_prev_snap_txg)) {
1069 		/*
1070 		 * Ignore errors, if there is not enough disk space
1071 		 * we will deal with it in dsl_dataset_destroy_sync().
1072 		 */
1073 		(void) dmu_free_object(os, obj);
1074 	}
1075 
1076 	/*
1077 	 * We need to sync out all in-flight IO before we try to evict
1078 	 * (the dataset evict func is trying to clear the cached entries
1079 	 * for this dataset in the ARC).
1080 	 */
1081 	txg_wait_synced(dd->dd_pool, 0);
1082 
1083 	/*
1084 	 * If we managed to free all the objects in open
1085 	 * context, the user space accounting should be zero.
1086 	 */
1087 	if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1088 	    dmu_objset_userused_enabled(os)) {
1089 		uint64_t count;
1090 
1091 		ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1092 		    count == 0);
1093 		ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1094 		    count == 0);
1095 	}
1096 
1097 	if (err != ESRCH)
1098 		goto out;
1099 
1100 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1101 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1102 	rw_exit(&dd->dd_pool->dp_config_rwlock);
1103 
1104 	if (err)
1105 		goto out;
1106 
1107 	if (ds->ds_objset) {
1108 		/*
1109 		 * We need to sync out all in-flight IO before we try
1110 		 * to evict (the dataset evict func is trying to clear
1111 		 * the cached entries for this dataset in the ARC).
1112 		 */
1113 		txg_wait_synced(dd->dd_pool, 0);
1114 	}
1115 
1116 	/*
1117 	 * Blow away the dsl_dir + head dataset.
1118 	 */
1119 	dsl_dataset_make_exclusive(ds, tag);
1120 	if (ds->ds_objset) {
1121 		dmu_objset_evict(ds->ds_objset);
1122 		ds->ds_objset = NULL;
1123 	}
1124 
1125 	/*
1126 	 * If we're removing a clone, we might also need to remove its
1127 	 * origin.
1128 	 */
1129 	do {
1130 		dsda.need_prep = B_FALSE;
1131 		if (dsl_dir_is_clone(dd)) {
1132 			err = dsl_dataset_origin_rm_prep(&dsda, tag);
1133 			if (err) {
1134 				dsl_dir_close(dd, FTAG);
1135 				goto out;
1136 			}
1137 		}
1138 
1139 		dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1140 		dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1141 		    dsl_dataset_destroy_sync, &dsda, tag, 0);
1142 		dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1143 		    dsl_dir_destroy_sync, dd, FTAG, 0);
1144 		err = dsl_sync_task_group_wait(dstg);
1145 		dsl_sync_task_group_destroy(dstg);
1146 
1147 		/*
1148 		 * We could be racing against 'zfs release' or 'zfs destroy -d'
1149 		 * on the origin snap, in which case we can get EBUSY if we
1150 		 * needed to destroy the origin snap but were not ready to
1151 		 * do so.
1152 		 */
1153 		if (dsda.need_prep) {
1154 			ASSERT(err == EBUSY);
1155 			ASSERT(dsl_dir_is_clone(dd));
1156 			ASSERT(dsda.rm_origin == NULL);
1157 		}
1158 	} while (dsda.need_prep);
1159 
1160 	if (dsda.rm_origin != NULL)
1161 		dsl_dataset_disown(dsda.rm_origin, tag);
1162 
1163 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
1164 	if (err)
1165 		dsl_dir_close(dd, FTAG);
1166 out:
1167 	dsl_dataset_disown(ds, tag);
1168 	return (err);
1169 }
1170 
1171 blkptr_t *
1172 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1173 {
1174 	return (&ds->ds_phys->ds_bp);
1175 }
1176 
1177 void
1178 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1179 {
1180 	ASSERT(dmu_tx_is_syncing(tx));
1181 	/* If it's the meta-objset, set dp_meta_rootbp */
1182 	if (ds == NULL) {
1183 		tx->tx_pool->dp_meta_rootbp = *bp;
1184 	} else {
1185 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1186 		ds->ds_phys->ds_bp = *bp;
1187 	}
1188 }
1189 
1190 spa_t *
1191 dsl_dataset_get_spa(dsl_dataset_t *ds)
1192 {
1193 	return (ds->ds_dir->dd_pool->dp_spa);
1194 }
1195 
1196 void
1197 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1198 {
1199 	dsl_pool_t *dp;
1200 
1201 	if (ds == NULL) /* this is the meta-objset */
1202 		return;
1203 
1204 	ASSERT(ds->ds_objset != NULL);
1205 
1206 	if (ds->ds_phys->ds_next_snap_obj != 0)
1207 		panic("dirtying snapshot!");
1208 
1209 	dp = ds->ds_dir->dd_pool;
1210 
1211 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1212 		/* up the hold count until we can be written out */
1213 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1214 	}
1215 }
1216 
1217 /*
1218  * The unique space in the head dataset can be calculated by subtracting
1219  * the space used in the most recent snapshot, that is still being used
1220  * in this file system, from the space currently in use.  To figure out
1221  * the space in the most recent snapshot still in use, we need to take
1222  * the total space used in the snapshot and subtract out the space that
1223  * has been freed up since the snapshot was taken.
1224  */
1225 static void
1226 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1227 {
1228 	uint64_t mrs_used;
1229 	uint64_t dlused, dlcomp, dluncomp;
1230 
1231 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1232 
1233 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1234 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1235 	else
1236 		mrs_used = 0;
1237 
1238 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1239 	    &dluncomp));
1240 
1241 	ASSERT3U(dlused, <=, mrs_used);
1242 	ds->ds_phys->ds_unique_bytes =
1243 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1244 
1245 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1246 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1247 	    SPA_VERSION_UNIQUE_ACCURATE)
1248 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1249 }
1250 
1251 static uint64_t
1252 dsl_dataset_unique(dsl_dataset_t *ds)
1253 {
1254 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1255 		dsl_dataset_recalc_head_uniq(ds);
1256 
1257 	return (ds->ds_phys->ds_unique_bytes);
1258 }
1259 
1260 struct killarg {
1261 	dsl_dataset_t *ds;
1262 	zio_t *zio;
1263 	dmu_tx_t *tx;
1264 };
1265 
1266 /* ARGSUSED */
1267 static int
1268 kill_blkptr(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
1269     const dnode_phys_t *dnp, void *arg)
1270 {
1271 	struct killarg *ka = arg;
1272 
1273 	if (bp == NULL)
1274 		return (0);
1275 
1276 	if ((zb->zb_level == -1ULL && zb->zb_blkid != 0) ||
1277 	    (zb->zb_object != 0 && dnp == NULL)) {
1278 		/*
1279 		 * It's a block in the intent log.  It has no
1280 		 * accounting, so just free it.
1281 		 */
1282 		VERIFY3U(0, ==, dsl_free(ka->zio, ka->tx->tx_pool,
1283 		    ka->tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT));
1284 	} else {
1285 		ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1286 		(void) dsl_dataset_block_kill(ka->ds, bp, ka->zio, ka->tx);
1287 	}
1288 
1289 	return (0);
1290 }
1291 
1292 /* ARGSUSED */
1293 static int
1294 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1295 {
1296 	dsl_dataset_t *ds = arg1;
1297 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1298 	uint64_t count;
1299 	int err;
1300 
1301 	/*
1302 	 * Can't delete a head dataset if there are snapshots of it.
1303 	 * (Except if the only snapshots are from the branch we cloned
1304 	 * from.)
1305 	 */
1306 	if (ds->ds_prev != NULL &&
1307 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1308 		return (EINVAL);
1309 
1310 	/*
1311 	 * This is really a dsl_dir thing, but check it here so that
1312 	 * we'll be less likely to leave this dataset inconsistent &
1313 	 * nearly destroyed.
1314 	 */
1315 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1316 	if (err)
1317 		return (err);
1318 	if (count != 0)
1319 		return (EEXIST);
1320 
1321 	return (0);
1322 }
1323 
1324 /* ARGSUSED */
1325 static void
1326 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1327 {
1328 	dsl_dataset_t *ds = arg1;
1329 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1330 
1331 	/* Mark it as inconsistent on-disk, in case we crash */
1332 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1333 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1334 
1335 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1336 	    cr, "dataset = %llu", ds->ds_object);
1337 }
1338 
1339 static int
1340 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1341     dmu_tx_t *tx)
1342 {
1343 	dsl_dataset_t *ds = dsda->ds;
1344 	dsl_dataset_t *ds_prev = ds->ds_prev;
1345 
1346 	if (dsl_dataset_might_destroy_origin(ds_prev)) {
1347 		struct dsl_ds_destroyarg ndsda = {0};
1348 
1349 		/*
1350 		 * If we're not prepared to remove the origin, don't remove
1351 		 * the clone either.
1352 		 */
1353 		if (dsda->rm_origin == NULL) {
1354 			dsda->need_prep = B_TRUE;
1355 			return (EBUSY);
1356 		}
1357 
1358 		ndsda.ds = ds_prev;
1359 		ndsda.is_origin_rm = B_TRUE;
1360 		return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1361 	}
1362 
1363 	/*
1364 	 * If we're not going to remove the origin after all,
1365 	 * undo the open context setup.
1366 	 */
1367 	if (dsda->rm_origin != NULL) {
1368 		dsl_dataset_disown(dsda->rm_origin, tag);
1369 		dsda->rm_origin = NULL;
1370 	}
1371 
1372 	return (0);
1373 }
1374 
1375 /* ARGSUSED */
1376 int
1377 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1378 {
1379 	struct dsl_ds_destroyarg *dsda = arg1;
1380 	dsl_dataset_t *ds = dsda->ds;
1381 
1382 	/* we have an owner hold, so noone else can destroy us */
1383 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1384 
1385 	/*
1386 	 * Only allow deferred destroy on pools that support it.
1387 	 * NOTE: deferred destroy is only supported on snapshots.
1388 	 */
1389 	if (dsda->defer) {
1390 		if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1391 		    SPA_VERSION_USERREFS)
1392 			return (ENOTSUP);
1393 		ASSERT(dsl_dataset_is_snapshot(ds));
1394 		return (0);
1395 	}
1396 
1397 	/*
1398 	 * Can't delete a head dataset if there are snapshots of it.
1399 	 * (Except if the only snapshots are from the branch we cloned
1400 	 * from.)
1401 	 */
1402 	if (ds->ds_prev != NULL &&
1403 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1404 		return (EINVAL);
1405 
1406 	/*
1407 	 * If we made changes this txg, traverse_dsl_dataset won't find
1408 	 * them.  Try again.
1409 	 */
1410 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1411 		return (EAGAIN);
1412 
1413 	if (dsl_dataset_is_snapshot(ds)) {
1414 		/*
1415 		 * If this snapshot has an elevated user reference count,
1416 		 * we can't destroy it yet.
1417 		 */
1418 		if (ds->ds_userrefs > 0 && !dsda->releasing)
1419 			return (EBUSY);
1420 
1421 		mutex_enter(&ds->ds_lock);
1422 		/*
1423 		 * Can't delete a branch point. However, if we're destroying
1424 		 * a clone and removing its origin due to it having a user
1425 		 * hold count of 0 and having been marked for deferred destroy,
1426 		 * it's OK for the origin to have a single clone.
1427 		 */
1428 		if (ds->ds_phys->ds_num_children >
1429 		    (dsda->is_origin_rm ? 2 : 1)) {
1430 			mutex_exit(&ds->ds_lock);
1431 			return (EEXIST);
1432 		}
1433 		mutex_exit(&ds->ds_lock);
1434 	} else if (dsl_dir_is_clone(ds->ds_dir)) {
1435 		return (dsl_dataset_origin_check(dsda, arg2, tx));
1436 	}
1437 
1438 	/* XXX we should do some i/o error checking... */
1439 	return (0);
1440 }
1441 
1442 struct refsarg {
1443 	kmutex_t lock;
1444 	boolean_t gone;
1445 	kcondvar_t cv;
1446 };
1447 
1448 /* ARGSUSED */
1449 static void
1450 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1451 {
1452 	struct refsarg *arg = argv;
1453 
1454 	mutex_enter(&arg->lock);
1455 	arg->gone = TRUE;
1456 	cv_signal(&arg->cv);
1457 	mutex_exit(&arg->lock);
1458 }
1459 
1460 static void
1461 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1462 {
1463 	struct refsarg arg;
1464 
1465 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1466 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1467 	arg.gone = FALSE;
1468 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1469 	    dsl_dataset_refs_gone);
1470 	dmu_buf_rele(ds->ds_dbuf, tag);
1471 	mutex_enter(&arg.lock);
1472 	while (!arg.gone)
1473 		cv_wait(&arg.cv, &arg.lock);
1474 	ASSERT(arg.gone);
1475 	mutex_exit(&arg.lock);
1476 	ds->ds_dbuf = NULL;
1477 	ds->ds_phys = NULL;
1478 	mutex_destroy(&arg.lock);
1479 	cv_destroy(&arg.cv);
1480 }
1481 
1482 void
1483 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1484 {
1485 	struct dsl_ds_destroyarg *dsda = arg1;
1486 	dsl_dataset_t *ds = dsda->ds;
1487 	zio_t *zio;
1488 	int err;
1489 	int after_branch_point = FALSE;
1490 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1491 	objset_t *mos = dp->dp_meta_objset;
1492 	dsl_dataset_t *ds_prev = NULL;
1493 	uint64_t obj;
1494 
1495 	ASSERT(ds->ds_owner);
1496 	ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1497 	ASSERT(ds->ds_prev == NULL ||
1498 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1499 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1500 
1501 	if (dsda->defer) {
1502 		ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1503 		if (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1) {
1504 			dmu_buf_will_dirty(ds->ds_dbuf, tx);
1505 			ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1506 			return;
1507 		}
1508 	}
1509 
1510 	/* signal any waiters that this dataset is going away */
1511 	mutex_enter(&ds->ds_lock);
1512 	ds->ds_owner = dsl_reaper;
1513 	cv_broadcast(&ds->ds_exclusive_cv);
1514 	mutex_exit(&ds->ds_lock);
1515 
1516 	/* Remove our reservation */
1517 	if (ds->ds_reserved != 0) {
1518 		uint64_t val = 0;
1519 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1520 		ASSERT3U(ds->ds_reserved, ==, 0);
1521 	}
1522 
1523 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1524 
1525 	dsl_pool_ds_destroyed(ds, tx);
1526 
1527 	obj = ds->ds_object;
1528 
1529 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1530 		if (ds->ds_prev) {
1531 			ds_prev = ds->ds_prev;
1532 		} else {
1533 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1534 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1535 		}
1536 		after_branch_point =
1537 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1538 
1539 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1540 		if (after_branch_point &&
1541 		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
1542 			VERIFY3U(0, ==, zap_remove_int(mos,
1543 			    ds_prev->ds_phys->ds_next_clones_obj, obj, tx));
1544 			if (ds->ds_phys->ds_next_snap_obj != 0) {
1545 				VERIFY(0 == zap_add_int(mos,
1546 				    ds_prev->ds_phys->ds_next_clones_obj,
1547 				    ds->ds_phys->ds_next_snap_obj, tx));
1548 			}
1549 		}
1550 		if (after_branch_point &&
1551 		    ds->ds_phys->ds_next_snap_obj == 0) {
1552 			/* This clone is toast. */
1553 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1554 			ds_prev->ds_phys->ds_num_children--;
1555 
1556 			/*
1557 			 * If the clone's origin has no other clones, no
1558 			 * user holds, and has been marked for deferred
1559 			 * deletion, then we should have done the necessary
1560 			 * destroy setup for it.
1561 			 */
1562 			if (ds_prev->ds_phys->ds_num_children == 1 &&
1563 			    ds_prev->ds_userrefs == 0 &&
1564 			    DS_IS_DEFER_DESTROY(ds_prev)) {
1565 				ASSERT3P(dsda->rm_origin, !=, NULL);
1566 			} else {
1567 				ASSERT3P(dsda->rm_origin, ==, NULL);
1568 			}
1569 		} else if (!after_branch_point) {
1570 			ds_prev->ds_phys->ds_next_snap_obj =
1571 			    ds->ds_phys->ds_next_snap_obj;
1572 		}
1573 	}
1574 
1575 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1576 
1577 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1578 		blkptr_t bp;
1579 		dsl_dataset_t *ds_next;
1580 		uint64_t itor = 0;
1581 		uint64_t old_unique;
1582 		int64_t used = 0, compressed = 0, uncompressed = 0;
1583 
1584 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1585 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1586 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1587 
1588 		old_unique = dsl_dataset_unique(ds_next);
1589 
1590 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1591 		ds_next->ds_phys->ds_prev_snap_obj =
1592 		    ds->ds_phys->ds_prev_snap_obj;
1593 		ds_next->ds_phys->ds_prev_snap_txg =
1594 		    ds->ds_phys->ds_prev_snap_txg;
1595 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1596 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1597 
1598 		/*
1599 		 * Transfer to our deadlist (which will become next's
1600 		 * new deadlist) any entries from next's current
1601 		 * deadlist which were born before prev, and free the
1602 		 * other entries.
1603 		 *
1604 		 * XXX we're doing this long task with the config lock held
1605 		 */
1606 		while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1607 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1608 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1609 				    &bp, tx));
1610 				if (ds_prev && !after_branch_point &&
1611 				    bp.blk_birth >
1612 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1613 					ds_prev->ds_phys->ds_unique_bytes +=
1614 					    bp_get_dasize(dp->dp_spa, &bp);
1615 				}
1616 			} else {
1617 				used += bp_get_dasize(dp->dp_spa, &bp);
1618 				compressed += BP_GET_PSIZE(&bp);
1619 				uncompressed += BP_GET_UCSIZE(&bp);
1620 				/* XXX check return value? */
1621 				(void) dsl_free(zio, dp, tx->tx_txg,
1622 				    &bp, NULL, NULL, ARC_NOWAIT);
1623 			}
1624 		}
1625 
1626 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1627 
1628 		/* change snapused */
1629 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1630 		    -used, -compressed, -uncompressed, tx);
1631 
1632 		/* free next's deadlist */
1633 		bplist_close(&ds_next->ds_deadlist);
1634 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1635 
1636 		/* set next's deadlist to our deadlist */
1637 		bplist_close(&ds->ds_deadlist);
1638 		ds_next->ds_phys->ds_deadlist_obj =
1639 		    ds->ds_phys->ds_deadlist_obj;
1640 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1641 		    ds_next->ds_phys->ds_deadlist_obj));
1642 		ds->ds_phys->ds_deadlist_obj = 0;
1643 
1644 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1645 			/*
1646 			 * Update next's unique to include blocks which
1647 			 * were previously shared by only this snapshot
1648 			 * and it.  Those blocks will be born after the
1649 			 * prev snap and before this snap, and will have
1650 			 * died after the next snap and before the one
1651 			 * after that (ie. be on the snap after next's
1652 			 * deadlist).
1653 			 *
1654 			 * XXX we're doing this long task with the
1655 			 * config lock held
1656 			 */
1657 			dsl_dataset_t *ds_after_next;
1658 			uint64_t space;
1659 
1660 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1661 			    ds_next->ds_phys->ds_next_snap_obj,
1662 			    FTAG, &ds_after_next));
1663 
1664 			VERIFY(0 ==
1665 			    bplist_space_birthrange(&ds_after_next->ds_deadlist,
1666 			    ds->ds_phys->ds_prev_snap_txg,
1667 			    ds->ds_phys->ds_creation_txg, &space));
1668 			ds_next->ds_phys->ds_unique_bytes += space;
1669 
1670 			dsl_dataset_rele(ds_after_next, FTAG);
1671 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1672 		} else {
1673 			ASSERT3P(ds_next->ds_prev, ==, ds);
1674 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1675 			ds_next->ds_prev = NULL;
1676 			if (ds_prev) {
1677 				VERIFY(0 == dsl_dataset_get_ref(dp,
1678 				    ds->ds_phys->ds_prev_snap_obj,
1679 				    ds_next, &ds_next->ds_prev));
1680 			}
1681 
1682 			dsl_dataset_recalc_head_uniq(ds_next);
1683 
1684 			/*
1685 			 * Reduce the amount of our unconsmed refreservation
1686 			 * being charged to our parent by the amount of
1687 			 * new unique data we have gained.
1688 			 */
1689 			if (old_unique < ds_next->ds_reserved) {
1690 				int64_t mrsdelta;
1691 				uint64_t new_unique =
1692 				    ds_next->ds_phys->ds_unique_bytes;
1693 
1694 				ASSERT(old_unique <= new_unique);
1695 				mrsdelta = MIN(new_unique - old_unique,
1696 				    ds_next->ds_reserved - old_unique);
1697 				dsl_dir_diduse_space(ds->ds_dir,
1698 				    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1699 			}
1700 		}
1701 		dsl_dataset_rele(ds_next, FTAG);
1702 	} else {
1703 		/*
1704 		 * There's no next snapshot, so this is a head dataset.
1705 		 * Destroy the deadlist.  Unless it's a clone, the
1706 		 * deadlist should be empty.  (If it's a clone, it's
1707 		 * safe to ignore the deadlist contents.)
1708 		 */
1709 		struct killarg ka;
1710 
1711 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1712 		bplist_close(&ds->ds_deadlist);
1713 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1714 		ds->ds_phys->ds_deadlist_obj = 0;
1715 
1716 		/*
1717 		 * Free everything that we point to (that's born after
1718 		 * the previous snapshot, if we are a clone)
1719 		 *
1720 		 * NB: this should be very quick, because we already
1721 		 * freed all the objects in open context.
1722 		 */
1723 		ka.ds = ds;
1724 		ka.zio = zio;
1725 		ka.tx = tx;
1726 		err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1727 		    TRAVERSE_POST, kill_blkptr, &ka);
1728 		ASSERT3U(err, ==, 0);
1729 		ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1730 		    ds->ds_phys->ds_unique_bytes == 0);
1731 
1732 		if (ds->ds_prev != NULL) {
1733 			dsl_dataset_rele(ds->ds_prev, ds);
1734 			ds->ds_prev = ds_prev = NULL;
1735 		}
1736 	}
1737 
1738 	err = zio_wait(zio);
1739 	ASSERT3U(err, ==, 0);
1740 
1741 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1742 		/* Erase the link in the dir */
1743 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1744 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1745 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1746 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1747 		ASSERT(err == 0);
1748 	} else {
1749 		/* remove from snapshot namespace */
1750 		dsl_dataset_t *ds_head;
1751 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1752 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1753 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1754 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1755 #ifdef ZFS_DEBUG
1756 		{
1757 			uint64_t val;
1758 
1759 			err = dsl_dataset_snap_lookup(ds_head,
1760 			    ds->ds_snapname, &val);
1761 			ASSERT3U(err, ==, 0);
1762 			ASSERT3U(val, ==, obj);
1763 		}
1764 #endif
1765 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1766 		ASSERT(err == 0);
1767 		dsl_dataset_rele(ds_head, FTAG);
1768 	}
1769 
1770 	if (ds_prev && ds->ds_prev != ds_prev)
1771 		dsl_dataset_rele(ds_prev, FTAG);
1772 
1773 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1774 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1775 	    cr, "dataset = %llu", ds->ds_object);
1776 
1777 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1778 		uint64_t count;
1779 		ASSERT(0 == zap_count(mos,
1780 		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1781 		VERIFY(0 == dmu_object_free(mos,
1782 		    ds->ds_phys->ds_next_clones_obj, tx));
1783 	}
1784 	if (ds->ds_phys->ds_props_obj != 0)
1785 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1786 	if (ds->ds_phys->ds_userrefs_obj != 0)
1787 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1788 	dsl_dir_close(ds->ds_dir, ds);
1789 	ds->ds_dir = NULL;
1790 	dsl_dataset_drain_refs(ds, tag);
1791 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1792 
1793 	if (dsda->rm_origin) {
1794 		/*
1795 		 * Remove the origin of the clone we just destroyed.
1796 		 */
1797 		struct dsl_ds_destroyarg ndsda = {0};
1798 
1799 		ndsda.ds = dsda->rm_origin;
1800 		dsl_dataset_destroy_sync(&ndsda, tag, cr, tx);
1801 	}
1802 }
1803 
1804 static int
1805 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1806 {
1807 	uint64_t asize;
1808 
1809 	if (!dmu_tx_is_syncing(tx))
1810 		return (0);
1811 
1812 	/*
1813 	 * If there's an fs-only reservation, any blocks that might become
1814 	 * owned by the snapshot dataset must be accommodated by space
1815 	 * outside of the reservation.
1816 	 */
1817 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1818 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1819 		return (ENOSPC);
1820 
1821 	/*
1822 	 * Propogate any reserved space for this snapshot to other
1823 	 * snapshot checks in this sync group.
1824 	 */
1825 	if (asize > 0)
1826 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1827 
1828 	return (0);
1829 }
1830 
1831 /* ARGSUSED */
1832 int
1833 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1834 {
1835 	dsl_dataset_t *ds = arg1;
1836 	const char *snapname = arg2;
1837 	int err;
1838 	uint64_t value;
1839 
1840 	/*
1841 	 * We don't allow multiple snapshots of the same txg.  If there
1842 	 * is already one, try again.
1843 	 */
1844 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1845 		return (EAGAIN);
1846 
1847 	/*
1848 	 * Check for conflicting name snapshot name.
1849 	 */
1850 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1851 	if (err == 0)
1852 		return (EEXIST);
1853 	if (err != ENOENT)
1854 		return (err);
1855 
1856 	/*
1857 	 * Check that the dataset's name is not too long.  Name consists
1858 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1859 	 */
1860 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1861 		return (ENAMETOOLONG);
1862 
1863 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1864 	if (err)
1865 		return (err);
1866 
1867 	ds->ds_trysnap_txg = tx->tx_txg;
1868 	return (0);
1869 }
1870 
1871 void
1872 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1873 {
1874 	dsl_dataset_t *ds = arg1;
1875 	const char *snapname = arg2;
1876 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1877 	dmu_buf_t *dbuf;
1878 	dsl_dataset_phys_t *dsphys;
1879 	uint64_t dsobj, crtxg;
1880 	objset_t *mos = dp->dp_meta_objset;
1881 	int err;
1882 
1883 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1884 
1885 	/*
1886 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1887 	 */
1888 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1889 		crtxg = 1;
1890 	else
1891 		crtxg = tx->tx_txg;
1892 
1893 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1894 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1895 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1896 	dmu_buf_will_dirty(dbuf, tx);
1897 	dsphys = dbuf->db_data;
1898 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1899 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1900 	dsphys->ds_fsid_guid = unique_create();
1901 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1902 	    sizeof (dsphys->ds_guid));
1903 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1904 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1905 	dsphys->ds_next_snap_obj = ds->ds_object;
1906 	dsphys->ds_num_children = 1;
1907 	dsphys->ds_creation_time = gethrestime_sec();
1908 	dsphys->ds_creation_txg = crtxg;
1909 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1910 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1911 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1912 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1913 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1914 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1915 	dmu_buf_rele(dbuf, FTAG);
1916 
1917 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1918 	if (ds->ds_prev) {
1919 		uint64_t next_clones_obj =
1920 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1921 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1922 		    ds->ds_object ||
1923 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1924 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1925 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1926 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1927 			    ds->ds_prev->ds_phys->ds_creation_txg);
1928 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1929 		} else if (next_clones_obj != 0) {
1930 			VERIFY3U(0, ==, zap_remove_int(mos,
1931 			    next_clones_obj, dsphys->ds_next_snap_obj, tx));
1932 			VERIFY3U(0, ==, zap_add_int(mos,
1933 			    next_clones_obj, dsobj, tx));
1934 		}
1935 	}
1936 
1937 	/*
1938 	 * If we have a reference-reservation on this dataset, we will
1939 	 * need to increase the amount of refreservation being charged
1940 	 * since our unique space is going to zero.
1941 	 */
1942 	if (ds->ds_reserved) {
1943 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1944 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1945 		    add, 0, 0, tx);
1946 	}
1947 
1948 	bplist_close(&ds->ds_deadlist);
1949 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1950 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1951 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1952 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1953 	ds->ds_phys->ds_unique_bytes = 0;
1954 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1955 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1956 	ds->ds_phys->ds_deadlist_obj =
1957 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1958 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1959 	    ds->ds_phys->ds_deadlist_obj));
1960 
1961 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1962 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1963 	    snapname, 8, 1, &dsobj, tx);
1964 	ASSERT(err == 0);
1965 
1966 	if (ds->ds_prev)
1967 		dsl_dataset_drop_ref(ds->ds_prev, ds);
1968 	VERIFY(0 == dsl_dataset_get_ref(dp,
1969 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1970 
1971 	dsl_pool_ds_snapshotted(ds, tx);
1972 
1973 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1974 
1975 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1976 	    "dataset = %llu", dsobj);
1977 }
1978 
1979 void
1980 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1981 {
1982 	ASSERT(dmu_tx_is_syncing(tx));
1983 	ASSERT(ds->ds_objset != NULL);
1984 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1985 
1986 	/*
1987 	 * in case we had to change ds_fsid_guid when we opened it,
1988 	 * sync it out now.
1989 	 */
1990 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1991 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1992 
1993 	dsl_dir_dirty(ds->ds_dir, tx);
1994 	dmu_objset_sync(ds->ds_objset, zio, tx);
1995 }
1996 
1997 void
1998 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1999 {
2000 	uint64_t refd, avail, uobjs, aobjs;
2001 
2002 	dsl_dir_stats(ds->ds_dir, nv);
2003 
2004 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2005 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2006 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2007 
2008 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2009 	    ds->ds_phys->ds_creation_time);
2010 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2011 	    ds->ds_phys->ds_creation_txg);
2012 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2013 	    ds->ds_quota);
2014 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2015 	    ds->ds_reserved);
2016 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2017 	    ds->ds_phys->ds_guid);
2018 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, ds->ds_userrefs);
2019 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2020 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2021 
2022 	if (ds->ds_phys->ds_next_snap_obj) {
2023 		/*
2024 		 * This is a snapshot; override the dd's space used with
2025 		 * our unique space and compression ratio.
2026 		 */
2027 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2028 		    ds->ds_phys->ds_unique_bytes);
2029 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2030 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2031 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
2032 		    ds->ds_phys->ds_compressed_bytes));
2033 	}
2034 }
2035 
2036 void
2037 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2038 {
2039 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2040 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2041 	stat->dds_guid = ds->ds_phys->ds_guid;
2042 	if (ds->ds_phys->ds_next_snap_obj) {
2043 		stat->dds_is_snapshot = B_TRUE;
2044 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2045 	} else {
2046 		stat->dds_is_snapshot = B_FALSE;
2047 		stat->dds_num_clones = 0;
2048 	}
2049 
2050 	/* clone origin is really a dsl_dir thing... */
2051 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2052 	if (dsl_dir_is_clone(ds->ds_dir)) {
2053 		dsl_dataset_t *ods;
2054 
2055 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2056 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2057 		dsl_dataset_name(ods, stat->dds_origin);
2058 		dsl_dataset_drop_ref(ods, FTAG);
2059 	} else {
2060 		stat->dds_origin[0] = '\0';
2061 	}
2062 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2063 }
2064 
2065 uint64_t
2066 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2067 {
2068 	return (ds->ds_fsid_guid);
2069 }
2070 
2071 void
2072 dsl_dataset_space(dsl_dataset_t *ds,
2073     uint64_t *refdbytesp, uint64_t *availbytesp,
2074     uint64_t *usedobjsp, uint64_t *availobjsp)
2075 {
2076 	*refdbytesp = ds->ds_phys->ds_used_bytes;
2077 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2078 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2079 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2080 	if (ds->ds_quota != 0) {
2081 		/*
2082 		 * Adjust available bytes according to refquota
2083 		 */
2084 		if (*refdbytesp < ds->ds_quota)
2085 			*availbytesp = MIN(*availbytesp,
2086 			    ds->ds_quota - *refdbytesp);
2087 		else
2088 			*availbytesp = 0;
2089 	}
2090 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2091 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2092 }
2093 
2094 boolean_t
2095 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2096 {
2097 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2098 
2099 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2100 	    dsl_pool_sync_context(dp));
2101 	if (ds->ds_prev == NULL)
2102 		return (B_FALSE);
2103 	if (ds->ds_phys->ds_bp.blk_birth >
2104 	    ds->ds_prev->ds_phys->ds_creation_txg)
2105 		return (B_TRUE);
2106 	return (B_FALSE);
2107 }
2108 
2109 /* ARGSUSED */
2110 static int
2111 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2112 {
2113 	dsl_dataset_t *ds = arg1;
2114 	char *newsnapname = arg2;
2115 	dsl_dir_t *dd = ds->ds_dir;
2116 	dsl_dataset_t *hds;
2117 	uint64_t val;
2118 	int err;
2119 
2120 	err = dsl_dataset_hold_obj(dd->dd_pool,
2121 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2122 	if (err)
2123 		return (err);
2124 
2125 	/* new name better not be in use */
2126 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2127 	dsl_dataset_rele(hds, FTAG);
2128 
2129 	if (err == 0)
2130 		err = EEXIST;
2131 	else if (err == ENOENT)
2132 		err = 0;
2133 
2134 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2135 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2136 		err = ENAMETOOLONG;
2137 
2138 	return (err);
2139 }
2140 
2141 static void
2142 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
2143     cred_t *cr, dmu_tx_t *tx)
2144 {
2145 	dsl_dataset_t *ds = arg1;
2146 	const char *newsnapname = arg2;
2147 	dsl_dir_t *dd = ds->ds_dir;
2148 	objset_t *mos = dd->dd_pool->dp_meta_objset;
2149 	dsl_dataset_t *hds;
2150 	int err;
2151 
2152 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2153 
2154 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2155 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2156 
2157 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2158 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2159 	ASSERT3U(err, ==, 0);
2160 	mutex_enter(&ds->ds_lock);
2161 	(void) strcpy(ds->ds_snapname, newsnapname);
2162 	mutex_exit(&ds->ds_lock);
2163 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2164 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2165 	ASSERT3U(err, ==, 0);
2166 
2167 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2168 	    cr, "dataset = %llu", ds->ds_object);
2169 	dsl_dataset_rele(hds, FTAG);
2170 }
2171 
2172 struct renamesnaparg {
2173 	dsl_sync_task_group_t *dstg;
2174 	char failed[MAXPATHLEN];
2175 	char *oldsnap;
2176 	char *newsnap;
2177 };
2178 
2179 static int
2180 dsl_snapshot_rename_one(char *name, void *arg)
2181 {
2182 	struct renamesnaparg *ra = arg;
2183 	dsl_dataset_t *ds = NULL;
2184 	char *cp;
2185 	int err;
2186 
2187 	cp = name + strlen(name);
2188 	*cp = '@';
2189 	(void) strcpy(cp + 1, ra->oldsnap);
2190 
2191 	/*
2192 	 * For recursive snapshot renames the parent won't be changing
2193 	 * so we just pass name for both the to/from argument.
2194 	 */
2195 	err = zfs_secpolicy_rename_perms(name, name, CRED());
2196 	if (err == ENOENT) {
2197 		return (0);
2198 	} else if (err) {
2199 		(void) strcpy(ra->failed, name);
2200 		return (err);
2201 	}
2202 
2203 #ifdef _KERNEL
2204 	/*
2205 	 * For all filesystems undergoing rename, we'll need to unmount it.
2206 	 */
2207 	(void) zfs_unmount_snap(name, NULL);
2208 #endif
2209 	err = dsl_dataset_hold(name, ra->dstg, &ds);
2210 	*cp = '\0';
2211 	if (err == ENOENT) {
2212 		return (0);
2213 	} else if (err) {
2214 		(void) strcpy(ra->failed, name);
2215 		return (err);
2216 	}
2217 
2218 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2219 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2220 
2221 	return (0);
2222 }
2223 
2224 static int
2225 dsl_recursive_rename(char *oldname, const char *newname)
2226 {
2227 	int err;
2228 	struct renamesnaparg *ra;
2229 	dsl_sync_task_t *dst;
2230 	spa_t *spa;
2231 	char *cp, *fsname = spa_strdup(oldname);
2232 	int len = strlen(oldname);
2233 
2234 	/* truncate the snapshot name to get the fsname */
2235 	cp = strchr(fsname, '@');
2236 	*cp = '\0';
2237 
2238 	err = spa_open(fsname, &spa, FTAG);
2239 	if (err) {
2240 		kmem_free(fsname, len + 1);
2241 		return (err);
2242 	}
2243 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2244 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2245 
2246 	ra->oldsnap = strchr(oldname, '@') + 1;
2247 	ra->newsnap = strchr(newname, '@') + 1;
2248 	*ra->failed = '\0';
2249 
2250 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2251 	    DS_FIND_CHILDREN);
2252 	kmem_free(fsname, len + 1);
2253 
2254 	if (err == 0) {
2255 		err = dsl_sync_task_group_wait(ra->dstg);
2256 	}
2257 
2258 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2259 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2260 		dsl_dataset_t *ds = dst->dst_arg1;
2261 		if (dst->dst_err) {
2262 			dsl_dir_name(ds->ds_dir, ra->failed);
2263 			(void) strcat(ra->failed, "@");
2264 			(void) strcat(ra->failed, ra->newsnap);
2265 		}
2266 		dsl_dataset_rele(ds, ra->dstg);
2267 	}
2268 
2269 	if (err)
2270 		(void) strcpy(oldname, ra->failed);
2271 
2272 	dsl_sync_task_group_destroy(ra->dstg);
2273 	kmem_free(ra, sizeof (struct renamesnaparg));
2274 	spa_close(spa, FTAG);
2275 	return (err);
2276 }
2277 
2278 static int
2279 dsl_valid_rename(char *oldname, void *arg)
2280 {
2281 	int delta = *(int *)arg;
2282 
2283 	if (strlen(oldname) + delta >= MAXNAMELEN)
2284 		return (ENAMETOOLONG);
2285 
2286 	return (0);
2287 }
2288 
2289 #pragma weak dmu_objset_rename = dsl_dataset_rename
2290 int
2291 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2292 {
2293 	dsl_dir_t *dd;
2294 	dsl_dataset_t *ds;
2295 	const char *tail;
2296 	int err;
2297 
2298 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2299 	if (err)
2300 		return (err);
2301 	/*
2302 	 * If there are more than 2 references there may be holds
2303 	 * hanging around that haven't been cleared out yet.
2304 	 */
2305 	if (dmu_buf_refcount(dd->dd_dbuf) > 2)
2306 		txg_wait_synced(dd->dd_pool, 0);
2307 	if (tail == NULL) {
2308 		int delta = strlen(newname) - strlen(oldname);
2309 
2310 		/* if we're growing, validate child name lengths */
2311 		if (delta > 0)
2312 			err = dmu_objset_find(oldname, dsl_valid_rename,
2313 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2314 
2315 		if (!err)
2316 			err = dsl_dir_rename(dd, newname);
2317 		dsl_dir_close(dd, FTAG);
2318 		return (err);
2319 	}
2320 	if (tail[0] != '@') {
2321 		/* the name ended in a nonexistant component */
2322 		dsl_dir_close(dd, FTAG);
2323 		return (ENOENT);
2324 	}
2325 
2326 	dsl_dir_close(dd, FTAG);
2327 
2328 	/* new name must be snapshot in same filesystem */
2329 	tail = strchr(newname, '@');
2330 	if (tail == NULL)
2331 		return (EINVAL);
2332 	tail++;
2333 	if (strncmp(oldname, newname, tail - newname) != 0)
2334 		return (EXDEV);
2335 
2336 	if (recursive) {
2337 		err = dsl_recursive_rename(oldname, newname);
2338 	} else {
2339 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2340 		if (err)
2341 			return (err);
2342 
2343 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2344 		    dsl_dataset_snapshot_rename_check,
2345 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2346 
2347 		dsl_dataset_rele(ds, FTAG);
2348 	}
2349 
2350 	return (err);
2351 }
2352 
2353 struct promotenode {
2354 	list_node_t link;
2355 	dsl_dataset_t *ds;
2356 };
2357 
2358 struct promotearg {
2359 	list_t shared_snaps, origin_snaps, clone_snaps;
2360 	dsl_dataset_t *origin_origin, *origin_head;
2361 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2362 };
2363 
2364 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2365 
2366 /* ARGSUSED */
2367 static int
2368 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2369 {
2370 	dsl_dataset_t *hds = arg1;
2371 	struct promotearg *pa = arg2;
2372 	struct promotenode *snap = list_head(&pa->shared_snaps);
2373 	dsl_dataset_t *origin_ds = snap->ds;
2374 	int err;
2375 
2376 	/* Check that it is a real clone */
2377 	if (!dsl_dir_is_clone(hds->ds_dir))
2378 		return (EINVAL);
2379 
2380 	/* Since this is so expensive, don't do the preliminary check */
2381 	if (!dmu_tx_is_syncing(tx))
2382 		return (0);
2383 
2384 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2385 		return (EXDEV);
2386 
2387 	/* compute origin's new unique space */
2388 	snap = list_tail(&pa->clone_snaps);
2389 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2390 	err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2391 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique);
2392 	if (err)
2393 		return (err);
2394 
2395 	/*
2396 	 * Walk the snapshots that we are moving
2397 	 *
2398 	 * Compute space to transfer.  Consider the incremental changes
2399 	 * to used for each snapshot:
2400 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2401 	 * So each snapshot gave birth to:
2402 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2403 	 * So a sequence would look like:
2404 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2405 	 * Which simplifies to:
2406 	 * uN + kN + kN-1 + ... + k1 + k0
2407 	 * Note however, if we stop before we reach the ORIGIN we get:
2408 	 * uN + kN + kN-1 + ... + kM - uM-1
2409 	 */
2410 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2411 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2412 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2413 	for (snap = list_head(&pa->shared_snaps); snap;
2414 	    snap = list_next(&pa->shared_snaps, snap)) {
2415 		uint64_t val, dlused, dlcomp, dluncomp;
2416 		dsl_dataset_t *ds = snap->ds;
2417 
2418 		/* Check that the snapshot name does not conflict */
2419 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2420 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2421 		if (err == 0)
2422 			return (EEXIST);
2423 		if (err != ENOENT)
2424 			return (err);
2425 
2426 		/* The very first snapshot does not have a deadlist */
2427 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2428 			continue;
2429 
2430 		if (err = bplist_space(&ds->ds_deadlist,
2431 		    &dlused, &dlcomp, &dluncomp))
2432 			return (err);
2433 		pa->used += dlused;
2434 		pa->comp += dlcomp;
2435 		pa->uncomp += dluncomp;
2436 	}
2437 
2438 	/*
2439 	 * If we are a clone of a clone then we never reached ORIGIN,
2440 	 * so we need to subtract out the clone origin's used space.
2441 	 */
2442 	if (pa->origin_origin) {
2443 		pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2444 		pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2445 		pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2446 	}
2447 
2448 	/* Check that there is enough space here */
2449 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2450 	    pa->used);
2451 	if (err)
2452 		return (err);
2453 
2454 	/*
2455 	 * Compute the amounts of space that will be used by snapshots
2456 	 * after the promotion (for both origin and clone).  For each,
2457 	 * it is the amount of space that will be on all of their
2458 	 * deadlists (that was not born before their new origin).
2459 	 */
2460 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2461 		uint64_t space;
2462 
2463 		/*
2464 		 * Note, typically this will not be a clone of a clone,
2465 		 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so
2466 		 * these snaplist_space() -> bplist_space_birthrange()
2467 		 * calls will be fast because they do not have to
2468 		 * iterate over all bps.
2469 		 */
2470 		snap = list_head(&pa->origin_snaps);
2471 		err = snaplist_space(&pa->shared_snaps,
2472 		    snap->ds->ds_origin_txg, &pa->cloneusedsnap);
2473 		if (err)
2474 			return (err);
2475 
2476 		err = snaplist_space(&pa->clone_snaps,
2477 		    snap->ds->ds_origin_txg, &space);
2478 		if (err)
2479 			return (err);
2480 		pa->cloneusedsnap += space;
2481 	}
2482 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2483 		err = snaplist_space(&pa->origin_snaps,
2484 		    origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2485 		if (err)
2486 			return (err);
2487 	}
2488 
2489 	return (0);
2490 }
2491 
2492 static void
2493 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2494 {
2495 	dsl_dataset_t *hds = arg1;
2496 	struct promotearg *pa = arg2;
2497 	struct promotenode *snap = list_head(&pa->shared_snaps);
2498 	dsl_dataset_t *origin_ds = snap->ds;
2499 	dsl_dataset_t *origin_head;
2500 	dsl_dir_t *dd = hds->ds_dir;
2501 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2502 	dsl_dir_t *odd = NULL;
2503 	uint64_t oldnext_obj;
2504 	int64_t delta;
2505 
2506 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2507 
2508 	snap = list_head(&pa->origin_snaps);
2509 	origin_head = snap->ds;
2510 
2511 	/*
2512 	 * We need to explicitly open odd, since origin_ds's dd will be
2513 	 * changing.
2514 	 */
2515 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2516 	    NULL, FTAG, &odd));
2517 
2518 	/* change origin's next snap */
2519 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2520 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2521 	snap = list_tail(&pa->clone_snaps);
2522 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2523 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2524 
2525 	/* change the origin's next clone */
2526 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2527 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2528 		    origin_ds->ds_phys->ds_next_clones_obj,
2529 		    origin_ds->ds_phys->ds_next_snap_obj, tx));
2530 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2531 		    origin_ds->ds_phys->ds_next_clones_obj,
2532 		    oldnext_obj, tx));
2533 	}
2534 
2535 	/* change origin */
2536 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2537 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2538 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2539 	hds->ds_origin_txg = origin_head->ds_origin_txg;
2540 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2541 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2542 	origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg;
2543 
2544 	/* move snapshots to this dir */
2545 	for (snap = list_head(&pa->shared_snaps); snap;
2546 	    snap = list_next(&pa->shared_snaps, snap)) {
2547 		dsl_dataset_t *ds = snap->ds;
2548 
2549 		/* unregister props as dsl_dir is changing */
2550 		if (ds->ds_objset) {
2551 			dmu_objset_evict(ds->ds_objset);
2552 			ds->ds_objset = NULL;
2553 		}
2554 		/* move snap name entry */
2555 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2556 		VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2557 		    ds->ds_snapname, tx));
2558 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2559 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2560 		    8, 1, &ds->ds_object, tx));
2561 		/* change containing dsl_dir */
2562 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2563 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2564 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2565 		ASSERT3P(ds->ds_dir, ==, odd);
2566 		dsl_dir_close(ds->ds_dir, ds);
2567 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2568 		    NULL, ds, &ds->ds_dir));
2569 
2570 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2571 	}
2572 
2573 	/*
2574 	 * Change space accounting.
2575 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2576 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2577 	 * is true for each of {clone,origin} independently.
2578 	 */
2579 
2580 	delta = pa->cloneusedsnap -
2581 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2582 	ASSERT3S(delta, >=, 0);
2583 	ASSERT3U(pa->used, >=, delta);
2584 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2585 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2586 	    pa->used - delta, pa->comp, pa->uncomp, tx);
2587 
2588 	delta = pa->originusedsnap -
2589 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2590 	ASSERT3S(delta, <=, 0);
2591 	ASSERT3U(pa->used, >=, -delta);
2592 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2593 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2594 	    -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2595 
2596 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2597 
2598 	/* log history record */
2599 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2600 	    cr, "dataset = %llu", hds->ds_object);
2601 
2602 	dsl_dir_close(odd, FTAG);
2603 }
2604 
2605 static char *snaplist_tag = "snaplist";
2606 /*
2607  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2608  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2609  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2610  * snapshots back to this dataset's origin.
2611  */
2612 static int
2613 snaplist_make(dsl_pool_t *dp, boolean_t own,
2614     uint64_t first_obj, uint64_t last_obj, list_t *l)
2615 {
2616 	uint64_t obj = last_obj;
2617 
2618 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2619 
2620 	list_create(l, sizeof (struct promotenode),
2621 	    offsetof(struct promotenode, link));
2622 
2623 	while (obj != first_obj) {
2624 		dsl_dataset_t *ds;
2625 		struct promotenode *snap;
2626 		int err;
2627 
2628 		if (own) {
2629 			err = dsl_dataset_own_obj(dp, obj,
2630 			    0, snaplist_tag, &ds);
2631 			if (err == 0)
2632 				dsl_dataset_make_exclusive(ds, snaplist_tag);
2633 		} else {
2634 			err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2635 		}
2636 		if (err == ENOENT) {
2637 			/* lost race with snapshot destroy */
2638 			struct promotenode *last = list_tail(l);
2639 			ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2640 			obj = last->ds->ds_phys->ds_prev_snap_obj;
2641 			continue;
2642 		} else if (err) {
2643 			return (err);
2644 		}
2645 
2646 		if (first_obj == 0)
2647 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2648 
2649 		snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2650 		snap->ds = ds;
2651 		list_insert_tail(l, snap);
2652 		obj = ds->ds_phys->ds_prev_snap_obj;
2653 	}
2654 
2655 	return (0);
2656 }
2657 
2658 static int
2659 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2660 {
2661 	struct promotenode *snap;
2662 
2663 	*spacep = 0;
2664 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2665 		uint64_t used;
2666 		int err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2667 		    mintxg, UINT64_MAX, &used);
2668 		if (err)
2669 			return (err);
2670 		*spacep += used;
2671 	}
2672 	return (0);
2673 }
2674 
2675 static void
2676 snaplist_destroy(list_t *l, boolean_t own)
2677 {
2678 	struct promotenode *snap;
2679 
2680 	if (!l || !list_link_active(&l->list_head))
2681 		return;
2682 
2683 	while ((snap = list_tail(l)) != NULL) {
2684 		list_remove(l, snap);
2685 		if (own)
2686 			dsl_dataset_disown(snap->ds, snaplist_tag);
2687 		else
2688 			dsl_dataset_rele(snap->ds, snaplist_tag);
2689 		kmem_free(snap, sizeof (struct promotenode));
2690 	}
2691 	list_destroy(l);
2692 }
2693 
2694 /*
2695  * Promote a clone.  Nomenclature note:
2696  * "clone" or "cds": the original clone which is being promoted
2697  * "origin" or "ods": the snapshot which is originally clone's origin
2698  * "origin head" or "ohds": the dataset which is the head
2699  * (filesystem/volume) for the origin
2700  * "origin origin": the origin of the origin's filesystem (typically
2701  * NULL, indicating that the clone is not a clone of a clone).
2702  */
2703 int
2704 dsl_dataset_promote(const char *name)
2705 {
2706 	dsl_dataset_t *ds;
2707 	dsl_dir_t *dd;
2708 	dsl_pool_t *dp;
2709 	dmu_object_info_t doi;
2710 	struct promotearg pa = { 0 };
2711 	struct promotenode *snap;
2712 	int err;
2713 
2714 	err = dsl_dataset_hold(name, FTAG, &ds);
2715 	if (err)
2716 		return (err);
2717 	dd = ds->ds_dir;
2718 	dp = dd->dd_pool;
2719 
2720 	err = dmu_object_info(dp->dp_meta_objset,
2721 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2722 	if (err) {
2723 		dsl_dataset_rele(ds, FTAG);
2724 		return (err);
2725 	}
2726 
2727 	if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2728 		dsl_dataset_rele(ds, FTAG);
2729 		return (EINVAL);
2730 	}
2731 
2732 	/*
2733 	 * We are going to inherit all the snapshots taken before our
2734 	 * origin (i.e., our new origin will be our parent's origin).
2735 	 * Take ownership of them so that we can rename them into our
2736 	 * namespace.
2737 	 */
2738 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2739 
2740 	err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2741 	    &pa.shared_snaps);
2742 	if (err != 0)
2743 		goto out;
2744 
2745 	err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2746 	if (err != 0)
2747 		goto out;
2748 
2749 	snap = list_head(&pa.shared_snaps);
2750 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2751 	err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2752 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2753 	if (err != 0)
2754 		goto out;
2755 
2756 	if (dsl_dir_is_clone(snap->ds->ds_dir)) {
2757 		err = dsl_dataset_own_obj(dp,
2758 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2759 		    0, FTAG, &pa.origin_origin);
2760 		if (err != 0)
2761 			goto out;
2762 	}
2763 
2764 out:
2765 	rw_exit(&dp->dp_config_rwlock);
2766 
2767 	/*
2768 	 * Add in 128x the snapnames zapobj size, since we will be moving
2769 	 * a bunch of snapnames to the promoted ds, and dirtying their
2770 	 * bonus buffers.
2771 	 */
2772 	if (err == 0) {
2773 		err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2774 		    dsl_dataset_promote_sync, ds, &pa,
2775 		    2 + 2 * doi.doi_physical_blks);
2776 	}
2777 
2778 	snaplist_destroy(&pa.shared_snaps, B_TRUE);
2779 	snaplist_destroy(&pa.clone_snaps, B_FALSE);
2780 	snaplist_destroy(&pa.origin_snaps, B_FALSE);
2781 	if (pa.origin_origin)
2782 		dsl_dataset_disown(pa.origin_origin, FTAG);
2783 	dsl_dataset_rele(ds, FTAG);
2784 	return (err);
2785 }
2786 
2787 struct cloneswaparg {
2788 	dsl_dataset_t *cds; /* clone dataset */
2789 	dsl_dataset_t *ohds; /* origin's head dataset */
2790 	boolean_t force;
2791 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2792 };
2793 
2794 /* ARGSUSED */
2795 static int
2796 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2797 {
2798 	struct cloneswaparg *csa = arg1;
2799 
2800 	/* they should both be heads */
2801 	if (dsl_dataset_is_snapshot(csa->cds) ||
2802 	    dsl_dataset_is_snapshot(csa->ohds))
2803 		return (EINVAL);
2804 
2805 	/* the branch point should be just before them */
2806 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2807 		return (EINVAL);
2808 
2809 	/* cds should be the clone (unless they are unrelated) */
2810 	if (csa->cds->ds_prev != NULL &&
2811 	    csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
2812 	    csa->ohds->ds_object !=
2813 	    csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
2814 		return (EINVAL);
2815 
2816 	/* the clone should be a child of the origin */
2817 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2818 		return (EINVAL);
2819 
2820 	/* ohds shouldn't be modified unless 'force' */
2821 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2822 		return (ETXTBSY);
2823 
2824 	/* adjust amount of any unconsumed refreservation */
2825 	csa->unused_refres_delta =
2826 	    (int64_t)MIN(csa->ohds->ds_reserved,
2827 	    csa->ohds->ds_phys->ds_unique_bytes) -
2828 	    (int64_t)MIN(csa->ohds->ds_reserved,
2829 	    csa->cds->ds_phys->ds_unique_bytes);
2830 
2831 	if (csa->unused_refres_delta > 0 &&
2832 	    csa->unused_refres_delta >
2833 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2834 		return (ENOSPC);
2835 
2836 	return (0);
2837 }
2838 
2839 /* ARGSUSED */
2840 static void
2841 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2842 {
2843 	struct cloneswaparg *csa = arg1;
2844 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2845 
2846 	ASSERT(csa->cds->ds_reserved == 0);
2847 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2848 
2849 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2850 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2851 
2852 	if (csa->cds->ds_objset != NULL) {
2853 		dmu_objset_evict(csa->cds->ds_objset);
2854 		csa->cds->ds_objset = NULL;
2855 	}
2856 
2857 	if (csa->ohds->ds_objset != NULL) {
2858 		dmu_objset_evict(csa->ohds->ds_objset);
2859 		csa->ohds->ds_objset = NULL;
2860 	}
2861 
2862 	/*
2863 	 * Reset origin's unique bytes, if it exists.
2864 	 */
2865 	if (csa->cds->ds_prev) {
2866 		dsl_dataset_t *origin = csa->cds->ds_prev;
2867 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2868 		VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2869 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2870 		    &origin->ds_phys->ds_unique_bytes));
2871 	}
2872 
2873 	/* swap blkptrs */
2874 	{
2875 		blkptr_t tmp;
2876 		tmp = csa->ohds->ds_phys->ds_bp;
2877 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2878 		csa->cds->ds_phys->ds_bp = tmp;
2879 	}
2880 
2881 	/* set dd_*_bytes */
2882 	{
2883 		int64_t dused, dcomp, duncomp;
2884 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2885 		uint64_t odl_used, odl_comp, odl_uncomp;
2886 
2887 		ASSERT3U(csa->cds->ds_dir->dd_phys->
2888 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2889 
2890 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2891 		    &cdl_comp, &cdl_uncomp));
2892 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2893 		    &odl_comp, &odl_uncomp));
2894 
2895 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2896 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2897 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2898 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2899 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2900 		    cdl_uncomp -
2901 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2902 
2903 		dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
2904 		    dused, dcomp, duncomp, tx);
2905 		dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
2906 		    -dused, -dcomp, -duncomp, tx);
2907 
2908 		/*
2909 		 * The difference in the space used by snapshots is the
2910 		 * difference in snapshot space due to the head's
2911 		 * deadlist (since that's the only thing that's
2912 		 * changing that affects the snapused).
2913 		 */
2914 		VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2915 		    csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used));
2916 		VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist,
2917 		    csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used));
2918 		dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
2919 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2920 	}
2921 
2922 #define	SWITCH64(x, y) \
2923 	{ \
2924 		uint64_t __tmp = (x); \
2925 		(x) = (y); \
2926 		(y) = __tmp; \
2927 	}
2928 
2929 	/* swap ds_*_bytes */
2930 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2931 	    csa->cds->ds_phys->ds_used_bytes);
2932 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2933 	    csa->cds->ds_phys->ds_compressed_bytes);
2934 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2935 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2936 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2937 	    csa->cds->ds_phys->ds_unique_bytes);
2938 
2939 	/* apply any parent delta for change in unconsumed refreservation */
2940 	dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
2941 	    csa->unused_refres_delta, 0, 0, tx);
2942 
2943 	/* swap deadlists */
2944 	bplist_close(&csa->cds->ds_deadlist);
2945 	bplist_close(&csa->ohds->ds_deadlist);
2946 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2947 	    csa->cds->ds_phys->ds_deadlist_obj);
2948 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2949 	    csa->cds->ds_phys->ds_deadlist_obj));
2950 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2951 	    csa->ohds->ds_phys->ds_deadlist_obj));
2952 
2953 	dsl_pool_ds_clone_swapped(csa->ohds, csa->cds, tx);
2954 }
2955 
2956 /*
2957  * Swap 'clone' with its origin head datasets.  Used at the end of "zfs
2958  * recv" into an existing fs to swizzle the file system to the new
2959  * version, and by "zfs rollback".  Can also be used to swap two
2960  * independent head datasets if neither has any snapshots.
2961  */
2962 int
2963 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2964     boolean_t force)
2965 {
2966 	struct cloneswaparg csa;
2967 	int error;
2968 
2969 	ASSERT(clone->ds_owner);
2970 	ASSERT(origin_head->ds_owner);
2971 retry:
2972 	/* Need exclusive access for the swap */
2973 	rw_enter(&clone->ds_rwlock, RW_WRITER);
2974 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
2975 		rw_exit(&clone->ds_rwlock);
2976 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
2977 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
2978 			rw_exit(&origin_head->ds_rwlock);
2979 			goto retry;
2980 		}
2981 	}
2982 	csa.cds = clone;
2983 	csa.ohds = origin_head;
2984 	csa.force = force;
2985 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
2986 	    dsl_dataset_clone_swap_check,
2987 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
2988 	return (error);
2989 }
2990 
2991 /*
2992  * Given a pool name and a dataset object number in that pool,
2993  * return the name of that dataset.
2994  */
2995 int
2996 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2997 {
2998 	spa_t *spa;
2999 	dsl_pool_t *dp;
3000 	dsl_dataset_t *ds;
3001 	int error;
3002 
3003 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
3004 		return (error);
3005 	dp = spa_get_dsl(spa);
3006 	rw_enter(&dp->dp_config_rwlock, RW_READER);
3007 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3008 		dsl_dataset_name(ds, buf);
3009 		dsl_dataset_rele(ds, FTAG);
3010 	}
3011 	rw_exit(&dp->dp_config_rwlock);
3012 	spa_close(spa, FTAG);
3013 
3014 	return (error);
3015 }
3016 
3017 int
3018 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3019     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3020 {
3021 	int error = 0;
3022 
3023 	ASSERT3S(asize, >, 0);
3024 
3025 	/*
3026 	 * *ref_rsrv is the portion of asize that will come from any
3027 	 * unconsumed refreservation space.
3028 	 */
3029 	*ref_rsrv = 0;
3030 
3031 	mutex_enter(&ds->ds_lock);
3032 	/*
3033 	 * Make a space adjustment for reserved bytes.
3034 	 */
3035 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3036 		ASSERT3U(*used, >=,
3037 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3038 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3039 		*ref_rsrv =
3040 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3041 	}
3042 
3043 	if (!check_quota || ds->ds_quota == 0) {
3044 		mutex_exit(&ds->ds_lock);
3045 		return (0);
3046 	}
3047 	/*
3048 	 * If they are requesting more space, and our current estimate
3049 	 * is over quota, they get to try again unless the actual
3050 	 * on-disk is over quota and there are no pending changes (which
3051 	 * may free up space for us).
3052 	 */
3053 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3054 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3055 			error = ERESTART;
3056 		else
3057 			error = EDQUOT;
3058 	}
3059 	mutex_exit(&ds->ds_lock);
3060 
3061 	return (error);
3062 }
3063 
3064 /* ARGSUSED */
3065 static int
3066 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3067 {
3068 	dsl_dataset_t *ds = arg1;
3069 	uint64_t *quotap = arg2;
3070 	uint64_t new_quota = *quotap;
3071 
3072 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3073 		return (ENOTSUP);
3074 
3075 	if (new_quota == 0)
3076 		return (0);
3077 
3078 	if (new_quota < ds->ds_phys->ds_used_bytes ||
3079 	    new_quota < ds->ds_reserved)
3080 		return (ENOSPC);
3081 
3082 	return (0);
3083 }
3084 
3085 /* ARGSUSED */
3086 void
3087 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3088 {
3089 	dsl_dataset_t *ds = arg1;
3090 	uint64_t *quotap = arg2;
3091 	uint64_t new_quota = *quotap;
3092 
3093 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3094 
3095 	ds->ds_quota = new_quota;
3096 
3097 	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
3098 
3099 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
3100 	    tx, cr, "%lld dataset = %llu ",
3101 	    (longlong_t)new_quota, ds->ds_object);
3102 }
3103 
3104 int
3105 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
3106 {
3107 	dsl_dataset_t *ds;
3108 	int err;
3109 
3110 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3111 	if (err)
3112 		return (err);
3113 
3114 	if (quota != ds->ds_quota) {
3115 		/*
3116 		 * If someone removes a file, then tries to set the quota, we
3117 		 * want to make sure the file freeing takes effect.
3118 		 */
3119 		txg_wait_open(ds->ds_dir->dd_pool, 0);
3120 
3121 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3122 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3123 		    ds, &quota, 0);
3124 	}
3125 	dsl_dataset_rele(ds, FTAG);
3126 	return (err);
3127 }
3128 
3129 static int
3130 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3131 {
3132 	dsl_dataset_t *ds = arg1;
3133 	uint64_t *reservationp = arg2;
3134 	uint64_t new_reservation = *reservationp;
3135 	uint64_t unique;
3136 
3137 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3138 	    SPA_VERSION_REFRESERVATION)
3139 		return (ENOTSUP);
3140 
3141 	if (dsl_dataset_is_snapshot(ds))
3142 		return (EINVAL);
3143 
3144 	/*
3145 	 * If we are doing the preliminary check in open context, the
3146 	 * space estimates may be inaccurate.
3147 	 */
3148 	if (!dmu_tx_is_syncing(tx))
3149 		return (0);
3150 
3151 	mutex_enter(&ds->ds_lock);
3152 	unique = dsl_dataset_unique(ds);
3153 	mutex_exit(&ds->ds_lock);
3154 
3155 	if (MAX(unique, new_reservation) > MAX(unique, ds->ds_reserved)) {
3156 		uint64_t delta = MAX(unique, new_reservation) -
3157 		    MAX(unique, ds->ds_reserved);
3158 
3159 		if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3160 			return (ENOSPC);
3161 		if (ds->ds_quota > 0 &&
3162 		    new_reservation > ds->ds_quota)
3163 			return (ENOSPC);
3164 	}
3165 
3166 	return (0);
3167 }
3168 
3169 /* ARGSUSED */
3170 static void
3171 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
3172     dmu_tx_t *tx)
3173 {
3174 	dsl_dataset_t *ds = arg1;
3175 	uint64_t *reservationp = arg2;
3176 	uint64_t new_reservation = *reservationp;
3177 	uint64_t unique;
3178 	int64_t delta;
3179 
3180 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3181 
3182 	mutex_enter(&ds->ds_dir->dd_lock);
3183 	mutex_enter(&ds->ds_lock);
3184 	unique = dsl_dataset_unique(ds);
3185 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
3186 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3187 	ds->ds_reserved = new_reservation;
3188 	mutex_exit(&ds->ds_lock);
3189 
3190 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3191 	mutex_exit(&ds->ds_dir->dd_lock);
3192 	dsl_dir_prop_set_uint64_sync(ds->ds_dir, "refreservation",
3193 	    new_reservation, cr, tx);
3194 
3195 	spa_history_internal_log(LOG_DS_REFRESERV,
3196 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
3197 	    (longlong_t)new_reservation, ds->ds_object);
3198 }
3199 
3200 int
3201 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
3202 {
3203 	dsl_dataset_t *ds;
3204 	int err;
3205 
3206 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3207 	if (err)
3208 		return (err);
3209 
3210 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3211 	    dsl_dataset_set_reservation_check,
3212 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
3213 	dsl_dataset_rele(ds, FTAG);
3214 	return (err);
3215 }
3216 
3217 struct dsl_ds_holdarg {
3218 	dsl_sync_task_group_t *dstg;
3219 	char *htag;
3220 	char *snapname;
3221 	boolean_t recursive;
3222 	boolean_t gotone;
3223 	boolean_t temphold;
3224 	char failed[MAXPATHLEN];
3225 };
3226 
3227 static int
3228 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3229 {
3230 	dsl_dataset_t *ds = arg1;
3231 	struct dsl_ds_holdarg *ha = arg2;
3232 	char *htag = ha->htag;
3233 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3234 	int error = 0;
3235 
3236 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3237 		return (ENOTSUP);
3238 
3239 	if (!dsl_dataset_is_snapshot(ds))
3240 		return (EINVAL);
3241 
3242 	/* tags must be unique */
3243 	mutex_enter(&ds->ds_lock);
3244 	if (ds->ds_phys->ds_userrefs_obj) {
3245 		error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3246 		    8, 1, tx);
3247 		if (error == 0)
3248 			error = EEXIST;
3249 		else if (error == ENOENT)
3250 			error = 0;
3251 	}
3252 	mutex_exit(&ds->ds_lock);
3253 
3254 	return (error);
3255 }
3256 
3257 static void
3258 dsl_dataset_user_hold_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
3259 {
3260 	dsl_dataset_t *ds = arg1;
3261 	struct dsl_ds_holdarg *ha = arg2;
3262 	char *htag = ha->htag;
3263 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
3264 	objset_t *mos = dp->dp_meta_objset;
3265 	time_t now = gethrestime_sec();
3266 	uint64_t zapobj;
3267 
3268 	mutex_enter(&ds->ds_lock);
3269 	if (ds->ds_phys->ds_userrefs_obj == 0) {
3270 		/*
3271 		 * This is the first user hold for this dataset.  Create
3272 		 * the userrefs zap object.
3273 		 */
3274 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3275 		zapobj = ds->ds_phys->ds_userrefs_obj =
3276 		    zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3277 	} else {
3278 		zapobj = ds->ds_phys->ds_userrefs_obj;
3279 	}
3280 	ds->ds_userrefs++;
3281 	mutex_exit(&ds->ds_lock);
3282 
3283 	VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3284 
3285 	if (ha->temphold) {
3286 		VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3287 		    htag, &now, tx));
3288 	}
3289 
3290 	spa_history_internal_log(LOG_DS_USER_HOLD,
3291 	    dp->dp_spa, tx, cr, "<%s> temp = %d dataset = %llu", htag,
3292 	    (int)ha->temphold, ds->ds_object);
3293 }
3294 
3295 static int
3296 dsl_dataset_user_hold_one(char *dsname, void *arg)
3297 {
3298 	struct dsl_ds_holdarg *ha = arg;
3299 	dsl_dataset_t *ds;
3300 	int error;
3301 	char *name;
3302 
3303 	/* alloc a buffer to hold dsname@snapname plus terminating NULL */
3304 	name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3305 	error = dsl_dataset_hold(name, ha->dstg, &ds);
3306 	strfree(name);
3307 	if (error == 0) {
3308 		ha->gotone = B_TRUE;
3309 		dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3310 		    dsl_dataset_user_hold_sync, ds, ha, 0);
3311 	} else if (error == ENOENT && ha->recursive) {
3312 		error = 0;
3313 	} else {
3314 		(void) strcpy(ha->failed, dsname);
3315 	}
3316 	return (error);
3317 }
3318 
3319 int
3320 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3321     boolean_t recursive, boolean_t temphold)
3322 {
3323 	struct dsl_ds_holdarg *ha;
3324 	dsl_sync_task_t *dst;
3325 	spa_t *spa;
3326 	int error;
3327 
3328 	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3329 
3330 	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3331 
3332 	error = spa_open(dsname, &spa, FTAG);
3333 	if (error) {
3334 		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3335 		return (error);
3336 	}
3337 
3338 	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3339 	ha->htag = htag;
3340 	ha->snapname = snapname;
3341 	ha->recursive = recursive;
3342 	ha->temphold = temphold;
3343 	if (recursive) {
3344 		error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3345 		    ha, DS_FIND_CHILDREN);
3346 	} else {
3347 		error = dsl_dataset_user_hold_one(dsname, ha);
3348 	}
3349 	if (error == 0)
3350 		error = dsl_sync_task_group_wait(ha->dstg);
3351 
3352 	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3353 	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3354 		dsl_dataset_t *ds = dst->dst_arg1;
3355 
3356 		if (dst->dst_err) {
3357 			dsl_dataset_name(ds, ha->failed);
3358 			*strchr(ha->failed, '@') = '\0';
3359 		}
3360 		dsl_dataset_rele(ds, ha->dstg);
3361 	}
3362 
3363 	if (error == 0 && recursive && !ha->gotone)
3364 		error = ENOENT;
3365 
3366 	if (error)
3367 		(void) strcpy(dsname, ha->failed);
3368 
3369 	dsl_sync_task_group_destroy(ha->dstg);
3370 	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3371 	spa_close(spa, FTAG);
3372 	return (error);
3373 }
3374 
3375 struct dsl_ds_releasearg {
3376 	dsl_dataset_t *ds;
3377 	const char *htag;
3378 	boolean_t own;		/* do we own or just hold ds? */
3379 };
3380 
3381 static int
3382 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3383     boolean_t *might_destroy)
3384 {
3385 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3386 	uint64_t zapobj;
3387 	uint64_t tmp;
3388 	int error;
3389 
3390 	*might_destroy = B_FALSE;
3391 
3392 	mutex_enter(&ds->ds_lock);
3393 	zapobj = ds->ds_phys->ds_userrefs_obj;
3394 	if (zapobj == 0) {
3395 		/* The tag can't possibly exist */
3396 		mutex_exit(&ds->ds_lock);
3397 		return (ESRCH);
3398 	}
3399 
3400 	/* Make sure the tag exists */
3401 	error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3402 	if (error) {
3403 		mutex_exit(&ds->ds_lock);
3404 		if (error == ENOENT)
3405 			error = ESRCH;
3406 		return (error);
3407 	}
3408 
3409 	if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3410 	    DS_IS_DEFER_DESTROY(ds))
3411 		*might_destroy = B_TRUE;
3412 
3413 	mutex_exit(&ds->ds_lock);
3414 	return (0);
3415 }
3416 
3417 static int
3418 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3419 {
3420 	struct dsl_ds_releasearg *ra = arg1;
3421 	dsl_dataset_t *ds = ra->ds;
3422 	boolean_t might_destroy;
3423 	int error;
3424 
3425 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3426 		return (ENOTSUP);
3427 
3428 	error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3429 	if (error)
3430 		return (error);
3431 
3432 	if (might_destroy) {
3433 		struct dsl_ds_destroyarg dsda = {0};
3434 
3435 		if (dmu_tx_is_syncing(tx)) {
3436 			/*
3437 			 * If we're not prepared to remove the snapshot,
3438 			 * we can't allow the release to happen right now.
3439 			 */
3440 			if (!ra->own)
3441 				return (EBUSY);
3442 			if (ds->ds_objset) {
3443 				dmu_objset_evict(ds->ds_objset);
3444 				ds->ds_objset = NULL;
3445 			}
3446 		}
3447 		dsda.ds = ds;
3448 		dsda.releasing = B_TRUE;
3449 		return (dsl_dataset_destroy_check(&dsda, tag, tx));
3450 	}
3451 
3452 	return (0);
3453 }
3454 
3455 static void
3456 dsl_dataset_user_release_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
3457 {
3458 	struct dsl_ds_releasearg *ra = arg1;
3459 	dsl_dataset_t *ds = ra->ds;
3460 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
3461 	objset_t *mos = dp->dp_meta_objset;
3462 	uint64_t zapobj;
3463 	uint64_t dsobj = ds->ds_object;
3464 	uint64_t refs;
3465 	int error;
3466 
3467 	mutex_enter(&ds->ds_lock);
3468 	ds->ds_userrefs--;
3469 	refs = ds->ds_userrefs;
3470 	mutex_exit(&ds->ds_lock);
3471 	error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3472 	VERIFY(error == 0 || error == ENOENT);
3473 	zapobj = ds->ds_phys->ds_userrefs_obj;
3474 	VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3475 	if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3476 	    DS_IS_DEFER_DESTROY(ds)) {
3477 		struct dsl_ds_destroyarg dsda = {0};
3478 
3479 		ASSERT(ra->own);
3480 		dsda.ds = ds;
3481 		dsda.releasing = B_TRUE;
3482 		/* We already did the destroy_check */
3483 		dsl_dataset_destroy_sync(&dsda, tag, cr, tx);
3484 	}
3485 
3486 	spa_history_internal_log(LOG_DS_USER_RELEASE,
3487 	    dp->dp_spa, tx, cr, "<%s> %lld dataset = %llu",
3488 	    ra->htag, (longlong_t)refs, dsobj);
3489 }
3490 
3491 static int
3492 dsl_dataset_user_release_one(char *dsname, void *arg)
3493 {
3494 	struct dsl_ds_holdarg *ha = arg;
3495 	struct dsl_ds_releasearg *ra;
3496 	dsl_dataset_t *ds;
3497 	int error;
3498 	void *dtag = ha->dstg;
3499 	char *name;
3500 	boolean_t own = B_FALSE;
3501 	boolean_t might_destroy;
3502 
3503 	/* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3504 	name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3505 	error = dsl_dataset_hold(name, dtag, &ds);
3506 	strfree(name);
3507 	if (error == ENOENT && ha->recursive)
3508 		return (0);
3509 	(void) strcpy(ha->failed, dsname);
3510 	if (error)
3511 		return (error);
3512 
3513 	ha->gotone = B_TRUE;
3514 
3515 	ASSERT(dsl_dataset_is_snapshot(ds));
3516 
3517 	error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3518 	if (error) {
3519 		dsl_dataset_rele(ds, dtag);
3520 		return (error);
3521 	}
3522 
3523 	if (might_destroy) {
3524 #ifdef _KERNEL
3525 		error = zfs_unmount_snap(name, NULL);
3526 		if (error) {
3527 			dsl_dataset_rele(ds, dtag);
3528 			return (error);
3529 		}
3530 		error = dsl_dataset_zvol_cleanup(ds, name);
3531 		if (error) {
3532 			dsl_dataset_rele(ds, dtag);
3533 			return (error);
3534 		}
3535 #endif
3536 		if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3537 			dsl_dataset_rele(ds, dtag);
3538 			return (EBUSY);
3539 		} else {
3540 			own = B_TRUE;
3541 			dsl_dataset_make_exclusive(ds, dtag);
3542 		}
3543 	}
3544 
3545 	ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3546 	ra->ds = ds;
3547 	ra->htag = ha->htag;
3548 	ra->own = own;
3549 	dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3550 	    dsl_dataset_user_release_sync, ra, dtag, 0);
3551 
3552 	return (0);
3553 }
3554 
3555 int
3556 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3557     boolean_t recursive)
3558 {
3559 	struct dsl_ds_holdarg *ha;
3560 	dsl_sync_task_t *dst;
3561 	spa_t *spa;
3562 	int error;
3563 
3564 	ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3565 
3566 	(void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3567 
3568 	error = spa_open(dsname, &spa, FTAG);
3569 	if (error) {
3570 		kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3571 		return (error);
3572 	}
3573 
3574 	ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3575 	ha->htag = htag;
3576 	ha->snapname = snapname;
3577 	ha->recursive = recursive;
3578 	if (recursive) {
3579 		error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3580 		    ha, DS_FIND_CHILDREN);
3581 	} else {
3582 		error = dsl_dataset_user_release_one(dsname, ha);
3583 	}
3584 	if (error == 0)
3585 		error = dsl_sync_task_group_wait(ha->dstg);
3586 
3587 	for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3588 	    dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3589 		struct dsl_ds_releasearg *ra = dst->dst_arg1;
3590 		dsl_dataset_t *ds = ra->ds;
3591 
3592 		if (dst->dst_err)
3593 			dsl_dataset_name(ds, ha->failed);
3594 
3595 		if (ra->own)
3596 			dsl_dataset_disown(ds, ha->dstg);
3597 		else
3598 			dsl_dataset_rele(ds, ha->dstg);
3599 
3600 		kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3601 	}
3602 
3603 	if (error == 0 && recursive && !ha->gotone)
3604 		error = ENOENT;
3605 
3606 	if (error)
3607 		(void) strcpy(dsname, ha->failed);
3608 
3609 	dsl_sync_task_group_destroy(ha->dstg);
3610 	kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3611 	spa_close(spa, FTAG);
3612 	return (error);
3613 }
3614 
3615 /*
3616  * Called at spa_load time to release a stale temporary user hold.
3617  */
3618 int
3619 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag)
3620 {
3621 	dsl_dataset_t *ds;
3622 	char *snap;
3623 	char *name;
3624 	int namelen;
3625 	int error;
3626 
3627 	rw_enter(&dp->dp_config_rwlock, RW_READER);
3628 	error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
3629 	rw_exit(&dp->dp_config_rwlock);
3630 	if (error)
3631 		return (error);
3632 	namelen = dsl_dataset_namelen(ds)+1;
3633 	name = kmem_alloc(namelen, KM_SLEEP);
3634 	dsl_dataset_name(ds, name);
3635 	dsl_dataset_rele(ds, FTAG);
3636 
3637 	snap = strchr(name, '@');
3638 	*snap = '\0';
3639 	++snap;
3640 	return (dsl_dataset_user_release(name, snap, htag, B_FALSE));
3641 }
3642 
3643 int
3644 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
3645 {
3646 	dsl_dataset_t *ds;
3647 	int err;
3648 
3649 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3650 	if (err)
3651 		return (err);
3652 
3653 	VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
3654 	if (ds->ds_phys->ds_userrefs_obj != 0) {
3655 		zap_attribute_t *za;
3656 		zap_cursor_t zc;
3657 
3658 		za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
3659 		for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
3660 		    ds->ds_phys->ds_userrefs_obj);
3661 		    zap_cursor_retrieve(&zc, za) == 0;
3662 		    zap_cursor_advance(&zc)) {
3663 			VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
3664 			    za->za_first_integer));
3665 		}
3666 		zap_cursor_fini(&zc);
3667 		kmem_free(za, sizeof (zap_attribute_t));
3668 	}
3669 	dsl_dataset_rele(ds, FTAG);
3670 	return (0);
3671 }
3672 
3673 /*
3674  * Note, this fuction is used as the callback for dmu_objset_find().  We
3675  * always return 0 so that we will continue to find and process
3676  * inconsistent datasets, even if we encounter an error trying to
3677  * process one of them.
3678  */
3679 /* ARGSUSED */
3680 int
3681 dsl_destroy_inconsistent(char *dsname, void *arg)
3682 {
3683 	dsl_dataset_t *ds;
3684 
3685 	if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
3686 		if (DS_IS_INCONSISTENT(ds))
3687 			(void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
3688 		else
3689 			dsl_dataset_disown(ds, FTAG);
3690 	}
3691 	return (0);
3692 }
3693