xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision cc1a9a89a73172cc2db053635fab3b1b91691657)
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 2008 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 
43 static char *dsl_reaper = "the grim reaper";
44 
45 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
46 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
47 static dsl_checkfunc_t dsl_dataset_rollback_check;
48 static dsl_syncfunc_t dsl_dataset_rollback_sync;
49 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
50 
51 #define	DS_REF_MAX	(1ULL << 62)
52 
53 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
54 
55 #define	DSL_DATASET_IS_DESTROYED(ds)	((ds)->ds_owner == dsl_reaper)
56 
57 
58 /*
59  * Figure out how much of this delta should be propogated to the dsl_dir
60  * layer.  If there's a refreservation, that space has already been
61  * partially accounted for in our ancestors.
62  */
63 static int64_t
64 parent_delta(dsl_dataset_t *ds, int64_t delta)
65 {
66 	uint64_t old_bytes, new_bytes;
67 
68 	if (ds->ds_reserved == 0)
69 		return (delta);
70 
71 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
72 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
73 
74 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
75 	return (new_bytes - old_bytes);
76 }
77 
78 void
79 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
80 {
81 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
82 	int compressed = BP_GET_PSIZE(bp);
83 	int uncompressed = BP_GET_UCSIZE(bp);
84 	int64_t delta;
85 
86 	dprintf_bp(bp, "born, ds=%p\n", ds);
87 
88 	ASSERT(dmu_tx_is_syncing(tx));
89 	/* It could have been compressed away to nothing */
90 	if (BP_IS_HOLE(bp))
91 		return;
92 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
93 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
94 	if (ds == NULL) {
95 		/*
96 		 * Account for the meta-objset space in its placeholder
97 		 * dsl_dir.
98 		 */
99 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
100 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
101 		    used, compressed, uncompressed, tx);
102 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
103 		return;
104 	}
105 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
106 	mutex_enter(&ds->ds_dir->dd_lock);
107 	mutex_enter(&ds->ds_lock);
108 	delta = parent_delta(ds, used);
109 	ds->ds_phys->ds_used_bytes += used;
110 	ds->ds_phys->ds_compressed_bytes += compressed;
111 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
112 	ds->ds_phys->ds_unique_bytes += used;
113 	mutex_exit(&ds->ds_lock);
114 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
115 	    compressed, uncompressed, tx);
116 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
117 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
118 	mutex_exit(&ds->ds_dir->dd_lock);
119 }
120 
121 int
122 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
123     dmu_tx_t *tx)
124 {
125 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
126 	int compressed = BP_GET_PSIZE(bp);
127 	int uncompressed = BP_GET_UCSIZE(bp);
128 
129 	ASSERT(pio != NULL);
130 	ASSERT(dmu_tx_is_syncing(tx));
131 	/* No block pointer => nothing to free */
132 	if (BP_IS_HOLE(bp))
133 		return (0);
134 
135 	ASSERT(used > 0);
136 	if (ds == NULL) {
137 		int err;
138 		/*
139 		 * Account for the meta-objset space in its placeholder
140 		 * dataset.
141 		 */
142 		err = dsl_free(pio, tx->tx_pool,
143 		    tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
144 		ASSERT(err == 0);
145 
146 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
147 		    -used, -compressed, -uncompressed, tx);
148 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
149 		return (used);
150 	}
151 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
152 
153 	ASSERT(!dsl_dataset_is_snapshot(ds));
154 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
155 
156 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
157 		int err;
158 		int64_t delta;
159 
160 		dprintf_bp(bp, "freeing: %s", "");
161 		err = dsl_free(pio, tx->tx_pool,
162 		    tx->tx_txg, bp, NULL, NULL, ARC_NOWAIT);
163 		ASSERT(err == 0);
164 
165 		mutex_enter(&ds->ds_dir->dd_lock);
166 		mutex_enter(&ds->ds_lock);
167 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
168 		    !DS_UNIQUE_IS_ACCURATE(ds));
169 		delta = parent_delta(ds, -used);
170 		ds->ds_phys->ds_unique_bytes -= used;
171 		mutex_exit(&ds->ds_lock);
172 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
173 		    delta, -compressed, -uncompressed, tx);
174 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
175 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
176 		mutex_exit(&ds->ds_dir->dd_lock);
177 	} else {
178 		dprintf_bp(bp, "putting on dead list: %s", "");
179 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
180 		ASSERT3U(ds->ds_prev->ds_object, ==,
181 		    ds->ds_phys->ds_prev_snap_obj);
182 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
183 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
184 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
185 		    ds->ds_object && bp->blk_birth >
186 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
187 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
188 			mutex_enter(&ds->ds_prev->ds_lock);
189 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
190 			mutex_exit(&ds->ds_prev->ds_lock);
191 		}
192 		if (bp->blk_birth > ds->ds_origin_txg) {
193 			dsl_dir_transfer_space(ds->ds_dir, used,
194 			    DD_USED_HEAD, DD_USED_SNAP, tx);
195 		}
196 	}
197 	mutex_enter(&ds->ds_lock);
198 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
199 	ds->ds_phys->ds_used_bytes -= used;
200 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
201 	ds->ds_phys->ds_compressed_bytes -= compressed;
202 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
203 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
204 	mutex_exit(&ds->ds_lock);
205 
206 	return (used);
207 }
208 
209 uint64_t
210 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
211 {
212 	uint64_t trysnap = 0;
213 
214 	if (ds == NULL)
215 		return (0);
216 	/*
217 	 * The snapshot creation could fail, but that would cause an
218 	 * incorrect FALSE return, which would only result in an
219 	 * overestimation of the amount of space that an operation would
220 	 * consume, which is OK.
221 	 *
222 	 * There's also a small window where we could miss a pending
223 	 * snapshot, because we could set the sync task in the quiescing
224 	 * phase.  So this should only be used as a guess.
225 	 */
226 	if (ds->ds_trysnap_txg >
227 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
228 		trysnap = ds->ds_trysnap_txg;
229 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
230 }
231 
232 int
233 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
234 {
235 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
236 }
237 
238 /* ARGSUSED */
239 static void
240 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
241 {
242 	dsl_dataset_t *ds = dsv;
243 
244 	ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
245 
246 	dprintf_ds(ds, "evicting %s\n", "");
247 
248 	unique_remove(ds->ds_fsid_guid);
249 
250 	if (ds->ds_user_ptr != NULL)
251 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
252 
253 	if (ds->ds_prev) {
254 		dsl_dataset_drop_ref(ds->ds_prev, ds);
255 		ds->ds_prev = NULL;
256 	}
257 
258 	bplist_close(&ds->ds_deadlist);
259 	if (ds->ds_dir)
260 		dsl_dir_close(ds->ds_dir, ds);
261 
262 	ASSERT(!list_link_active(&ds->ds_synced_link));
263 
264 	mutex_destroy(&ds->ds_lock);
265 	mutex_destroy(&ds->ds_opening_lock);
266 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
267 	rw_destroy(&ds->ds_rwlock);
268 	cv_destroy(&ds->ds_exclusive_cv);
269 
270 	kmem_free(ds, sizeof (dsl_dataset_t));
271 }
272 
273 static int
274 dsl_dataset_get_snapname(dsl_dataset_t *ds)
275 {
276 	dsl_dataset_phys_t *headphys;
277 	int err;
278 	dmu_buf_t *headdbuf;
279 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
280 	objset_t *mos = dp->dp_meta_objset;
281 
282 	if (ds->ds_snapname[0])
283 		return (0);
284 	if (ds->ds_phys->ds_next_snap_obj == 0)
285 		return (0);
286 
287 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
288 	    FTAG, &headdbuf);
289 	if (err)
290 		return (err);
291 	headphys = headdbuf->db_data;
292 	err = zap_value_search(dp->dp_meta_objset,
293 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
294 	dmu_buf_rele(headdbuf, FTAG);
295 	return (err);
296 }
297 
298 static int
299 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
300 {
301 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
302 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
303 	matchtype_t mt;
304 	int err;
305 
306 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
307 		mt = MT_FIRST;
308 	else
309 		mt = MT_EXACT;
310 
311 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
312 	    value, mt, NULL, 0, NULL);
313 	if (err == ENOTSUP && mt == MT_FIRST)
314 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
315 	return (err);
316 }
317 
318 static int
319 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
320 {
321 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
322 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
323 	matchtype_t mt;
324 	int err;
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_opening_lock, NULL, MUTEX_DEFAULT, NULL);
363 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
364 		    NULL);
365 		rw_init(&ds->ds_rwlock, 0, 0, 0);
366 		cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
367 
368 		err = bplist_open(&ds->ds_deadlist,
369 		    mos, ds->ds_phys->ds_deadlist_obj);
370 		if (err == 0) {
371 			err = dsl_dir_open_obj(dp,
372 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
373 		}
374 		if (err) {
375 			/*
376 			 * we don't really need to close the blist if we
377 			 * just opened it.
378 			 */
379 			mutex_destroy(&ds->ds_lock);
380 			mutex_destroy(&ds->ds_opening_lock);
381 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
382 			rw_destroy(&ds->ds_rwlock);
383 			cv_destroy(&ds->ds_exclusive_cv);
384 			kmem_free(ds, sizeof (dsl_dataset_t));
385 			dmu_buf_rele(dbuf, tag);
386 			return (err);
387 		}
388 
389 		if (!dsl_dataset_is_snapshot(ds)) {
390 			ds->ds_snapname[0] = '\0';
391 			if (ds->ds_phys->ds_prev_snap_obj) {
392 				err = dsl_dataset_get_ref(dp,
393 				    ds->ds_phys->ds_prev_snap_obj,
394 				    ds, &ds->ds_prev);
395 			}
396 
397 			if (err == 0 && dsl_dir_is_clone(ds->ds_dir)) {
398 				dsl_dataset_t *origin;
399 
400 				err = dsl_dataset_hold_obj(dp,
401 				    ds->ds_dir->dd_phys->dd_origin_obj,
402 				    FTAG, &origin);
403 				if (err == 0) {
404 					ds->ds_origin_txg =
405 					    origin->ds_phys->ds_creation_txg;
406 					dsl_dataset_rele(origin, FTAG);
407 				}
408 			}
409 		} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
410 			err = dsl_dataset_get_snapname(ds);
411 		}
412 
413 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
414 			/*
415 			 * In sync context, we're called with either no lock
416 			 * or with the write lock.  If we're not syncing,
417 			 * we're always called with the read lock held.
418 			 */
419 			boolean_t need_lock =
420 			    !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
421 			    dsl_pool_sync_context(dp);
422 
423 			if (need_lock)
424 				rw_enter(&dp->dp_config_rwlock, RW_READER);
425 
426 			err = dsl_prop_get_ds(ds,
427 			    "refreservation", sizeof (uint64_t), 1,
428 			    &ds->ds_reserved, NULL);
429 			if (err == 0) {
430 				err = dsl_prop_get_ds(ds,
431 				    "refquota", sizeof (uint64_t), 1,
432 				    &ds->ds_quota, NULL);
433 			}
434 
435 			if (need_lock)
436 				rw_exit(&dp->dp_config_rwlock);
437 		} else {
438 			ds->ds_reserved = ds->ds_quota = 0;
439 		}
440 
441 		if (err == 0) {
442 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
443 			    dsl_dataset_evict);
444 		}
445 		if (err || winner) {
446 			bplist_close(&ds->ds_deadlist);
447 			if (ds->ds_prev)
448 				dsl_dataset_drop_ref(ds->ds_prev, ds);
449 			dsl_dir_close(ds->ds_dir, ds);
450 			mutex_destroy(&ds->ds_lock);
451 			mutex_destroy(&ds->ds_opening_lock);
452 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
453 			rw_destroy(&ds->ds_rwlock);
454 			cv_destroy(&ds->ds_exclusive_cv);
455 			kmem_free(ds, sizeof (dsl_dataset_t));
456 			if (err) {
457 				dmu_buf_rele(dbuf, tag);
458 				return (err);
459 			}
460 			ds = winner;
461 		} else {
462 			ds->ds_fsid_guid =
463 			    unique_insert(ds->ds_phys->ds_fsid_guid);
464 		}
465 	}
466 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
467 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
468 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
469 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
470 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
471 	mutex_enter(&ds->ds_lock);
472 	if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
473 		mutex_exit(&ds->ds_lock);
474 		dmu_buf_rele(ds->ds_dbuf, tag);
475 		return (ENOENT);
476 	}
477 	mutex_exit(&ds->ds_lock);
478 	*dsp = ds;
479 	return (0);
480 }
481 
482 static int
483 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
484 {
485 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
486 
487 	/*
488 	 * In syncing context we don't want the rwlock lock: there
489 	 * may be an existing writer waiting for sync phase to
490 	 * finish.  We don't need to worry about such writers, since
491 	 * sync phase is single-threaded, so the writer can't be
492 	 * doing anything while we are active.
493 	 */
494 	if (dsl_pool_sync_context(dp)) {
495 		ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
496 		return (0);
497 	}
498 
499 	/*
500 	 * Normal users will hold the ds_rwlock as a READER until they
501 	 * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
502 	 * drop their READER lock after they set the ds_owner field.
503 	 *
504 	 * If the dataset is being destroyed, the destroy thread will
505 	 * obtain a WRITER lock for exclusive access after it's done its
506 	 * open-context work and then change the ds_owner to
507 	 * dsl_reaper once destruction is assured.  So threads
508 	 * may block here temporarily, until the "destructability" of
509 	 * the dataset is determined.
510 	 */
511 	ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
512 	mutex_enter(&ds->ds_lock);
513 	while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
514 		rw_exit(&dp->dp_config_rwlock);
515 		cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
516 		if (DSL_DATASET_IS_DESTROYED(ds)) {
517 			mutex_exit(&ds->ds_lock);
518 			dsl_dataset_drop_ref(ds, tag);
519 			rw_enter(&dp->dp_config_rwlock, RW_READER);
520 			return (ENOENT);
521 		}
522 		rw_enter(&dp->dp_config_rwlock, RW_READER);
523 	}
524 	mutex_exit(&ds->ds_lock);
525 	return (0);
526 }
527 
528 int
529 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
530     dsl_dataset_t **dsp)
531 {
532 	int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
533 
534 	if (err)
535 		return (err);
536 	return (dsl_dataset_hold_ref(*dsp, tag));
537 }
538 
539 int
540 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, int flags, void *owner,
541     dsl_dataset_t **dsp)
542 {
543 	int err = dsl_dataset_hold_obj(dp, dsobj, owner, dsp);
544 
545 	ASSERT(DS_MODE_TYPE(flags) != DS_MODE_USER);
546 
547 	if (err)
548 		return (err);
549 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
550 		dsl_dataset_rele(*dsp, owner);
551 		return (EBUSY);
552 	}
553 	return (0);
554 }
555 
556 int
557 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
558 {
559 	dsl_dir_t *dd;
560 	dsl_pool_t *dp;
561 	const char *snapname;
562 	uint64_t obj;
563 	int err = 0;
564 
565 	err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
566 	if (err)
567 		return (err);
568 
569 	dp = dd->dd_pool;
570 	obj = dd->dd_phys->dd_head_dataset_obj;
571 	rw_enter(&dp->dp_config_rwlock, RW_READER);
572 	if (obj)
573 		err = dsl_dataset_get_ref(dp, obj, tag, dsp);
574 	else
575 		err = ENOENT;
576 	if (err)
577 		goto out;
578 
579 	err = dsl_dataset_hold_ref(*dsp, tag);
580 
581 	/* we may be looking for a snapshot */
582 	if (err == 0 && snapname != NULL) {
583 		dsl_dataset_t *ds = NULL;
584 
585 		if (*snapname++ != '@') {
586 			dsl_dataset_rele(*dsp, tag);
587 			err = ENOENT;
588 			goto out;
589 		}
590 
591 		dprintf("looking for snapshot '%s'\n", snapname);
592 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
593 		if (err == 0)
594 			err = dsl_dataset_get_ref(dp, obj, tag, &ds);
595 		dsl_dataset_rele(*dsp, tag);
596 
597 		ASSERT3U((err == 0), ==, (ds != NULL));
598 
599 		if (ds) {
600 			mutex_enter(&ds->ds_lock);
601 			if (ds->ds_snapname[0] == 0)
602 				(void) strlcpy(ds->ds_snapname, snapname,
603 				    sizeof (ds->ds_snapname));
604 			mutex_exit(&ds->ds_lock);
605 			err = dsl_dataset_hold_ref(ds, tag);
606 			*dsp = err ? NULL : ds;
607 		}
608 	}
609 out:
610 	rw_exit(&dp->dp_config_rwlock);
611 	dsl_dir_close(dd, FTAG);
612 	return (err);
613 }
614 
615 int
616 dsl_dataset_own(const char *name, int flags, void *owner, dsl_dataset_t **dsp)
617 {
618 	int err = dsl_dataset_hold(name, owner, dsp);
619 	if (err)
620 		return (err);
621 	if ((*dsp)->ds_phys->ds_num_children > 0 &&
622 	    !DS_MODE_IS_READONLY(flags)) {
623 		dsl_dataset_rele(*dsp, owner);
624 		return (EROFS);
625 	}
626 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
627 		dsl_dataset_rele(*dsp, owner);
628 		return (EBUSY);
629 	}
630 	return (0);
631 }
632 
633 void
634 dsl_dataset_name(dsl_dataset_t *ds, char *name)
635 {
636 	if (ds == NULL) {
637 		(void) strcpy(name, "mos");
638 	} else {
639 		dsl_dir_name(ds->ds_dir, name);
640 		VERIFY(0 == dsl_dataset_get_snapname(ds));
641 		if (ds->ds_snapname[0]) {
642 			(void) strcat(name, "@");
643 			/*
644 			 * We use a "recursive" mutex so that we
645 			 * can call dprintf_ds() with ds_lock held.
646 			 */
647 			if (!MUTEX_HELD(&ds->ds_lock)) {
648 				mutex_enter(&ds->ds_lock);
649 				(void) strcat(name, ds->ds_snapname);
650 				mutex_exit(&ds->ds_lock);
651 			} else {
652 				(void) strcat(name, ds->ds_snapname);
653 			}
654 		}
655 	}
656 }
657 
658 static int
659 dsl_dataset_namelen(dsl_dataset_t *ds)
660 {
661 	int result;
662 
663 	if (ds == NULL) {
664 		result = 3;	/* "mos" */
665 	} else {
666 		result = dsl_dir_namelen(ds->ds_dir);
667 		VERIFY(0 == dsl_dataset_get_snapname(ds));
668 		if (ds->ds_snapname[0]) {
669 			++result;	/* adding one for the @-sign */
670 			if (!MUTEX_HELD(&ds->ds_lock)) {
671 				mutex_enter(&ds->ds_lock);
672 				result += strlen(ds->ds_snapname);
673 				mutex_exit(&ds->ds_lock);
674 			} else {
675 				result += strlen(ds->ds_snapname);
676 			}
677 		}
678 	}
679 
680 	return (result);
681 }
682 
683 void
684 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
685 {
686 	dmu_buf_rele(ds->ds_dbuf, tag);
687 }
688 
689 void
690 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
691 {
692 	if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
693 		rw_exit(&ds->ds_rwlock);
694 	}
695 	dsl_dataset_drop_ref(ds, tag);
696 }
697 
698 void
699 dsl_dataset_disown(dsl_dataset_t *ds, void *owner)
700 {
701 	ASSERT((ds->ds_owner == owner && ds->ds_dbuf) ||
702 	    (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
703 
704 	mutex_enter(&ds->ds_lock);
705 	ds->ds_owner = NULL;
706 	if (RW_WRITE_HELD(&ds->ds_rwlock)) {
707 		rw_exit(&ds->ds_rwlock);
708 		cv_broadcast(&ds->ds_exclusive_cv);
709 	}
710 	mutex_exit(&ds->ds_lock);
711 	if (ds->ds_dbuf)
712 		dsl_dataset_drop_ref(ds, owner);
713 	else
714 		dsl_dataset_evict(ds->ds_dbuf, ds);
715 }
716 
717 boolean_t
718 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *owner)
719 {
720 	boolean_t gotit = FALSE;
721 
722 	mutex_enter(&ds->ds_lock);
723 	if (ds->ds_owner == NULL &&
724 	    (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
725 		ds->ds_owner = owner;
726 		if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
727 			rw_exit(&ds->ds_rwlock);
728 		gotit = TRUE;
729 	}
730 	mutex_exit(&ds->ds_lock);
731 	return (gotit);
732 }
733 
734 void
735 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
736 {
737 	ASSERT3P(owner, ==, ds->ds_owner);
738 	if (!RW_WRITE_HELD(&ds->ds_rwlock))
739 		rw_enter(&ds->ds_rwlock, RW_WRITER);
740 }
741 
742 uint64_t
743 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
744     uint64_t flags, dmu_tx_t *tx)
745 {
746 	dsl_pool_t *dp = dd->dd_pool;
747 	dmu_buf_t *dbuf;
748 	dsl_dataset_phys_t *dsphys;
749 	uint64_t dsobj;
750 	objset_t *mos = dp->dp_meta_objset;
751 
752 	if (origin == NULL)
753 		origin = dp->dp_origin_snap;
754 
755 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
756 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
757 	ASSERT(dmu_tx_is_syncing(tx));
758 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
759 
760 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
761 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
762 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
763 	dmu_buf_will_dirty(dbuf, tx);
764 	dsphys = dbuf->db_data;
765 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
766 	dsphys->ds_dir_obj = dd->dd_object;
767 	dsphys->ds_flags = flags;
768 	dsphys->ds_fsid_guid = unique_create();
769 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
770 	    sizeof (dsphys->ds_guid));
771 	dsphys->ds_snapnames_zapobj =
772 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
773 	    DMU_OT_NONE, 0, tx);
774 	dsphys->ds_creation_time = gethrestime_sec();
775 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
776 	dsphys->ds_deadlist_obj =
777 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
778 
779 	if (origin) {
780 		dsphys->ds_prev_snap_obj = origin->ds_object;
781 		dsphys->ds_prev_snap_txg =
782 		    origin->ds_phys->ds_creation_txg;
783 		dsphys->ds_used_bytes =
784 		    origin->ds_phys->ds_used_bytes;
785 		dsphys->ds_compressed_bytes =
786 		    origin->ds_phys->ds_compressed_bytes;
787 		dsphys->ds_uncompressed_bytes =
788 		    origin->ds_phys->ds_uncompressed_bytes;
789 		dsphys->ds_bp = origin->ds_phys->ds_bp;
790 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
791 
792 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
793 		origin->ds_phys->ds_num_children++;
794 
795 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
796 			if (origin->ds_phys->ds_next_clones_obj == 0) {
797 				origin->ds_phys->ds_next_clones_obj =
798 				    zap_create(mos,
799 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
800 			}
801 			VERIFY(0 == zap_add_int(mos,
802 			    origin->ds_phys->ds_next_clones_obj,
803 			    dsobj, tx));
804 		}
805 
806 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
807 		dd->dd_phys->dd_origin_obj = origin->ds_object;
808 	}
809 
810 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
811 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
812 
813 	dmu_buf_rele(dbuf, FTAG);
814 
815 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
816 	dd->dd_phys->dd_head_dataset_obj = dsobj;
817 
818 	return (dsobj);
819 }
820 
821 uint64_t
822 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
823     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
824 {
825 	dsl_pool_t *dp = pdd->dd_pool;
826 	uint64_t dsobj, ddobj;
827 	dsl_dir_t *dd;
828 
829 	ASSERT(lastname[0] != '@');
830 
831 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
832 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
833 
834 	dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
835 
836 	dsl_deleg_set_create_perms(dd, tx, cr);
837 
838 	dsl_dir_close(dd, FTAG);
839 
840 	return (dsobj);
841 }
842 
843 struct destroyarg {
844 	dsl_sync_task_group_t *dstg;
845 	char *snapname;
846 	char *failed;
847 };
848 
849 static int
850 dsl_snapshot_destroy_one(char *name, void *arg)
851 {
852 	struct destroyarg *da = arg;
853 	dsl_dataset_t *ds;
854 	char *cp;
855 	int err;
856 
857 	(void) strcat(name, "@");
858 	(void) strcat(name, da->snapname);
859 	err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
860 	    da->dstg, &ds);
861 	cp = strchr(name, '@');
862 	*cp = '\0';
863 	if (err == 0) {
864 		dsl_dataset_make_exclusive(ds, da->dstg);
865 		if (ds->ds_user_ptr) {
866 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
867 			ds->ds_user_ptr = NULL;
868 		}
869 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
870 		    dsl_dataset_destroy_sync, ds, da->dstg, 0);
871 	} else if (err == ENOENT) {
872 		err = 0;
873 	} else {
874 		(void) strcpy(da->failed, name);
875 	}
876 	return (err);
877 }
878 
879 /*
880  * Destroy 'snapname' in all descendants of 'fsname'.
881  */
882 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
883 int
884 dsl_snapshots_destroy(char *fsname, char *snapname)
885 {
886 	int err;
887 	struct destroyarg da;
888 	dsl_sync_task_t *dst;
889 	spa_t *spa;
890 
891 	err = spa_open(fsname, &spa, FTAG);
892 	if (err)
893 		return (err);
894 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
895 	da.snapname = snapname;
896 	da.failed = fsname;
897 
898 	err = dmu_objset_find(fsname,
899 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
900 
901 	if (err == 0)
902 		err = dsl_sync_task_group_wait(da.dstg);
903 
904 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
905 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
906 		dsl_dataset_t *ds = dst->dst_arg1;
907 		/*
908 		 * Return the file system name that triggered the error
909 		 */
910 		if (dst->dst_err) {
911 			dsl_dataset_name(ds, fsname);
912 			*strchr(fsname, '@') = '\0';
913 		}
914 		dsl_dataset_disown(ds, da.dstg);
915 	}
916 
917 	dsl_sync_task_group_destroy(da.dstg);
918 	spa_close(spa, FTAG);
919 	return (err);
920 }
921 
922 /*
923  * ds must be opened as OWNER.  On return (whether successful or not),
924  * ds will be closed and caller can no longer dereference it.
925  */
926 int
927 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
928 {
929 	int err;
930 	dsl_sync_task_group_t *dstg;
931 	objset_t *os;
932 	dsl_dir_t *dd;
933 	uint64_t obj;
934 
935 	if (dsl_dataset_is_snapshot(ds)) {
936 		/* Destroying a snapshot is simpler */
937 		dsl_dataset_make_exclusive(ds, tag);
938 
939 		if (ds->ds_user_ptr) {
940 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
941 			ds->ds_user_ptr = NULL;
942 		}
943 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
944 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
945 		    ds, tag, 0);
946 		goto out;
947 	}
948 
949 	dd = ds->ds_dir;
950 
951 	/*
952 	 * Check for errors and mark this ds as inconsistent, in
953 	 * case we crash while freeing the objects.
954 	 */
955 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
956 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
957 	if (err)
958 		goto out;
959 
960 	err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
961 	if (err)
962 		goto out;
963 
964 	/*
965 	 * remove the objects in open context, so that we won't
966 	 * have too much to do in syncing context.
967 	 */
968 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
969 	    ds->ds_phys->ds_prev_snap_txg)) {
970 		/*
971 		 * Ignore errors, if there is not enough disk space
972 		 * we will deal with it in dsl_dataset_destroy_sync().
973 		 */
974 		(void) dmu_free_object(os, obj);
975 	}
976 
977 	dmu_objset_close(os);
978 	if (err != ESRCH)
979 		goto out;
980 
981 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
982 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
983 	rw_exit(&dd->dd_pool->dp_config_rwlock);
984 
985 	if (err)
986 		goto out;
987 
988 	if (ds->ds_user_ptr) {
989 		/*
990 		 * We need to sync out all in-flight IO before we try
991 		 * to evict (the dataset evict func is trying to clear
992 		 * the cached entries for this dataset in the ARC).
993 		 */
994 		txg_wait_synced(dd->dd_pool, 0);
995 	}
996 
997 	/*
998 	 * Blow away the dsl_dir + head dataset.
999 	 */
1000 	dsl_dataset_make_exclusive(ds, tag);
1001 	if (ds->ds_user_ptr) {
1002 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1003 		ds->ds_user_ptr = NULL;
1004 	}
1005 	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1006 	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1007 	    dsl_dataset_destroy_sync, ds, tag, 0);
1008 	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1009 	    dsl_dir_destroy_sync, dd, FTAG, 0);
1010 	err = dsl_sync_task_group_wait(dstg);
1011 	dsl_sync_task_group_destroy(dstg);
1012 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
1013 	if (err)
1014 		dsl_dir_close(dd, FTAG);
1015 out:
1016 	dsl_dataset_disown(ds, tag);
1017 	return (err);
1018 }
1019 
1020 int
1021 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
1022 {
1023 	int err;
1024 
1025 	ASSERT(ds->ds_owner);
1026 
1027 	dsl_dataset_make_exclusive(ds, ds->ds_owner);
1028 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1029 	    dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
1030 	    ds, &ost, 0);
1031 	/* drop exclusive access */
1032 	mutex_enter(&ds->ds_lock);
1033 	rw_exit(&ds->ds_rwlock);
1034 	cv_broadcast(&ds->ds_exclusive_cv);
1035 	mutex_exit(&ds->ds_lock);
1036 	return (err);
1037 }
1038 
1039 void *
1040 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
1041     void *p, dsl_dataset_evict_func_t func)
1042 {
1043 	void *old;
1044 
1045 	mutex_enter(&ds->ds_lock);
1046 	old = ds->ds_user_ptr;
1047 	if (old == NULL) {
1048 		ds->ds_user_ptr = p;
1049 		ds->ds_user_evict_func = func;
1050 	}
1051 	mutex_exit(&ds->ds_lock);
1052 	return (old);
1053 }
1054 
1055 void *
1056 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
1057 {
1058 	return (ds->ds_user_ptr);
1059 }
1060 
1061 
1062 blkptr_t *
1063 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1064 {
1065 	return (&ds->ds_phys->ds_bp);
1066 }
1067 
1068 void
1069 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1070 {
1071 	ASSERT(dmu_tx_is_syncing(tx));
1072 	/* If it's the meta-objset, set dp_meta_rootbp */
1073 	if (ds == NULL) {
1074 		tx->tx_pool->dp_meta_rootbp = *bp;
1075 	} else {
1076 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1077 		ds->ds_phys->ds_bp = *bp;
1078 	}
1079 }
1080 
1081 spa_t *
1082 dsl_dataset_get_spa(dsl_dataset_t *ds)
1083 {
1084 	return (ds->ds_dir->dd_pool->dp_spa);
1085 }
1086 
1087 void
1088 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1089 {
1090 	dsl_pool_t *dp;
1091 
1092 	if (ds == NULL) /* this is the meta-objset */
1093 		return;
1094 
1095 	ASSERT(ds->ds_user_ptr != NULL);
1096 
1097 	if (ds->ds_phys->ds_next_snap_obj != 0)
1098 		panic("dirtying snapshot!");
1099 
1100 	dp = ds->ds_dir->dd_pool;
1101 
1102 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1103 		/* up the hold count until we can be written out */
1104 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1105 	}
1106 }
1107 
1108 /*
1109  * The unique space in the head dataset can be calculated by subtracting
1110  * the space used in the most recent snapshot, that is still being used
1111  * in this file system, from the space currently in use.  To figure out
1112  * the space in the most recent snapshot still in use, we need to take
1113  * the total space used in the snapshot and subtract out the space that
1114  * has been freed up since the snapshot was taken.
1115  */
1116 static void
1117 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1118 {
1119 	uint64_t mrs_used;
1120 	uint64_t dlused, dlcomp, dluncomp;
1121 
1122 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1123 
1124 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1125 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1126 	else
1127 		mrs_used = 0;
1128 
1129 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1130 	    &dluncomp));
1131 
1132 	ASSERT3U(dlused, <=, mrs_used);
1133 	ds->ds_phys->ds_unique_bytes =
1134 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1135 
1136 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1137 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1138 	    SPA_VERSION_UNIQUE_ACCURATE)
1139 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1140 }
1141 
1142 static uint64_t
1143 dsl_dataset_unique(dsl_dataset_t *ds)
1144 {
1145 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1146 		dsl_dataset_recalc_head_uniq(ds);
1147 
1148 	return (ds->ds_phys->ds_unique_bytes);
1149 }
1150 
1151 struct killarg {
1152 	dsl_dataset_t *ds;
1153 	zio_t *zio;
1154 	dmu_tx_t *tx;
1155 };
1156 
1157 /* ARGSUSED */
1158 static int
1159 kill_blkptr(spa_t *spa, blkptr_t *bp, const zbookmark_t *zb,
1160     const dnode_phys_t *dnp, void *arg)
1161 {
1162 	struct killarg *ka = arg;
1163 
1164 	if (bp == NULL)
1165 		return (0);
1166 
1167 	ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1168 	(void) dsl_dataset_block_kill(ka->ds, bp, ka->zio, ka->tx);
1169 
1170 	return (0);
1171 }
1172 
1173 /* ARGSUSED */
1174 static int
1175 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1176 {
1177 	dsl_dataset_t *ds = arg1;
1178 	dmu_objset_type_t *ost = arg2;
1179 
1180 	/*
1181 	 * We can only roll back to emptyness if it is a ZPL objset.
1182 	 */
1183 	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1184 		return (EINVAL);
1185 
1186 	/*
1187 	 * This must not be a snapshot.
1188 	 */
1189 	if (ds->ds_phys->ds_next_snap_obj != 0)
1190 		return (EINVAL);
1191 
1192 	/*
1193 	 * If we made changes this txg, traverse_dataset won't find
1194 	 * them.  Try again.
1195 	 */
1196 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1197 		return (EAGAIN);
1198 
1199 	return (0);
1200 }
1201 
1202 /* ARGSUSED */
1203 static void
1204 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1205 {
1206 	dsl_dataset_t *ds = arg1;
1207 	dmu_objset_type_t *ost = arg2;
1208 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1209 
1210 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1211 
1212 	/*
1213 	 * Before the roll back destroy the zil.
1214 	 */
1215 	if (ds->ds_user_ptr != NULL) {
1216 		zil_rollback_destroy(
1217 		    ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx);
1218 
1219 		/*
1220 		 * We need to make sure that the objset_impl_t is reopened after
1221 		 * we do the rollback, otherwise it will have the wrong
1222 		 * objset_phys_t.  Normally this would happen when this
1223 		 * dataset-open is closed, thus causing the
1224 		 * dataset to be immediately evicted.  But when doing "zfs recv
1225 		 * -F", we reopen the objset before that, so that there is no
1226 		 * window where the dataset is closed and inconsistent.
1227 		 */
1228 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1229 		ds->ds_user_ptr = NULL;
1230 	}
1231 
1232 	/* Transfer space that was freed since last snap back to the head. */
1233 	{
1234 		uint64_t used;
1235 
1236 		VERIFY(0 == bplist_space_birthrange(&ds->ds_deadlist,
1237 		    ds->ds_origin_txg, UINT64_MAX, &used));
1238 		dsl_dir_transfer_space(ds->ds_dir, used,
1239 		    DD_USED_SNAP, DD_USED_HEAD, tx);
1240 	}
1241 
1242 	/* Zero out the deadlist. */
1243 	bplist_close(&ds->ds_deadlist);
1244 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1245 	ds->ds_phys->ds_deadlist_obj =
1246 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1247 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1248 	    ds->ds_phys->ds_deadlist_obj));
1249 
1250 	{
1251 		/* Free blkptrs that we gave birth to */
1252 		zio_t *zio;
1253 		struct killarg ka;
1254 
1255 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1256 		    ZIO_FLAG_MUSTSUCCEED);
1257 		ka.ds = ds;
1258 		ka.zio = zio;
1259 		ka.tx = tx;
1260 		(void) traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1261 		    TRAVERSE_POST, kill_blkptr, &ka);
1262 		(void) zio_wait(zio);
1263 	}
1264 
1265 	ASSERT(!(ds->ds_phys->ds_flags & DS_FLAG_UNIQUE_ACCURATE) ||
1266 	    ds->ds_phys->ds_unique_bytes == 0);
1267 
1268 	if (ds->ds_prev && ds->ds_prev != ds->ds_dir->dd_pool->dp_origin_snap) {
1269 		/* Change our contents to that of the prev snapshot */
1270 
1271 		ASSERT3U(ds->ds_prev->ds_object, ==,
1272 		    ds->ds_phys->ds_prev_snap_obj);
1273 		ASSERT3U(ds->ds_phys->ds_used_bytes, <=,
1274 		    ds->ds_prev->ds_phys->ds_used_bytes);
1275 
1276 		ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1277 		ds->ds_phys->ds_used_bytes =
1278 		    ds->ds_prev->ds_phys->ds_used_bytes;
1279 		ds->ds_phys->ds_compressed_bytes =
1280 		    ds->ds_prev->ds_phys->ds_compressed_bytes;
1281 		ds->ds_phys->ds_uncompressed_bytes =
1282 		    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1283 		ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1284 
1285 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1286 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1287 			ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1288 		}
1289 	} else {
1290 		objset_impl_t *osi;
1291 
1292 		ASSERT3U(ds->ds_phys->ds_used_bytes, ==, 0);
1293 		ASSERT3U(ds->ds_phys->ds_compressed_bytes, ==, 0);
1294 		ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, ==, 0);
1295 
1296 		bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1297 		ds->ds_phys->ds_flags = 0;
1298 		ds->ds_phys->ds_unique_bytes = 0;
1299 		if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1300 		    SPA_VERSION_UNIQUE_ACCURATE)
1301 			ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1302 
1303 		osi = dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1304 		    &ds->ds_phys->ds_bp, *ost, tx);
1305 #ifdef _KERNEL
1306 		zfs_create_fs(&osi->os, kcred, NULL, tx);
1307 #endif
1308 	}
1309 
1310 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1311 	    tx, cr, "dataset = %llu", ds->ds_object);
1312 }
1313 
1314 /* ARGSUSED */
1315 static int
1316 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1317 {
1318 	dsl_dataset_t *ds = arg1;
1319 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1320 	uint64_t count;
1321 	int err;
1322 
1323 	/*
1324 	 * Can't delete a head dataset if there are snapshots of it.
1325 	 * (Except if the only snapshots are from the branch we cloned
1326 	 * from.)
1327 	 */
1328 	if (ds->ds_prev != NULL &&
1329 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1330 		return (EINVAL);
1331 
1332 	/*
1333 	 * This is really a dsl_dir thing, but check it here so that
1334 	 * we'll be less likely to leave this dataset inconsistent &
1335 	 * nearly destroyed.
1336 	 */
1337 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1338 	if (err)
1339 		return (err);
1340 	if (count != 0)
1341 		return (EEXIST);
1342 
1343 	return (0);
1344 }
1345 
1346 /* ARGSUSED */
1347 static void
1348 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1349 {
1350 	dsl_dataset_t *ds = arg1;
1351 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1352 
1353 	/* Mark it as inconsistent on-disk, in case we crash */
1354 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1355 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1356 
1357 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1358 	    cr, "dataset = %llu", ds->ds_object);
1359 }
1360 
1361 /* ARGSUSED */
1362 int
1363 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1364 {
1365 	dsl_dataset_t *ds = arg1;
1366 
1367 	/* we have an owner hold, so noone else can destroy us */
1368 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1369 
1370 	/* Can't delete a branch point. */
1371 	if (ds->ds_phys->ds_num_children > 1)
1372 		return (EEXIST);
1373 
1374 	/*
1375 	 * Can't delete a head dataset if there are snapshots of it.
1376 	 * (Except if the only snapshots are from the branch we cloned
1377 	 * from.)
1378 	 */
1379 	if (ds->ds_prev != NULL &&
1380 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1381 		return (EINVAL);
1382 
1383 	/*
1384 	 * If we made changes this txg, traverse_dsl_dataset won't find
1385 	 * them.  Try again.
1386 	 */
1387 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1388 		return (EAGAIN);
1389 
1390 	/* XXX we should do some i/o error checking... */
1391 	return (0);
1392 }
1393 
1394 struct refsarg {
1395 	kmutex_t lock;
1396 	boolean_t gone;
1397 	kcondvar_t cv;
1398 };
1399 
1400 /* ARGSUSED */
1401 static void
1402 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1403 {
1404 	struct refsarg *arg = argv;
1405 
1406 	mutex_enter(&arg->lock);
1407 	arg->gone = TRUE;
1408 	cv_signal(&arg->cv);
1409 	mutex_exit(&arg->lock);
1410 }
1411 
1412 static void
1413 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1414 {
1415 	struct refsarg arg;
1416 
1417 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1418 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1419 	arg.gone = FALSE;
1420 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1421 	    dsl_dataset_refs_gone);
1422 	dmu_buf_rele(ds->ds_dbuf, tag);
1423 	mutex_enter(&arg.lock);
1424 	while (!arg.gone)
1425 		cv_wait(&arg.cv, &arg.lock);
1426 	ASSERT(arg.gone);
1427 	mutex_exit(&arg.lock);
1428 	ds->ds_dbuf = NULL;
1429 	ds->ds_phys = NULL;
1430 	mutex_destroy(&arg.lock);
1431 	cv_destroy(&arg.cv);
1432 }
1433 
1434 void
1435 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1436 {
1437 	dsl_dataset_t *ds = arg1;
1438 	zio_t *zio;
1439 	int err;
1440 	int after_branch_point = FALSE;
1441 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1442 	objset_t *mos = dp->dp_meta_objset;
1443 	dsl_dataset_t *ds_prev = NULL;
1444 	uint64_t obj;
1445 
1446 	ASSERT(ds->ds_owner);
1447 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1448 	ASSERT(ds->ds_prev == NULL ||
1449 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1450 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1451 
1452 	/* signal any waiters that this dataset is going away */
1453 	mutex_enter(&ds->ds_lock);
1454 	ds->ds_owner = dsl_reaper;
1455 	cv_broadcast(&ds->ds_exclusive_cv);
1456 	mutex_exit(&ds->ds_lock);
1457 
1458 	/* Remove our reservation */
1459 	if (ds->ds_reserved != 0) {
1460 		uint64_t val = 0;
1461 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1462 		ASSERT3U(ds->ds_reserved, ==, 0);
1463 	}
1464 
1465 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1466 
1467 	dsl_pool_ds_destroyed(ds, tx);
1468 
1469 	obj = ds->ds_object;
1470 
1471 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1472 		if (ds->ds_prev) {
1473 			ds_prev = ds->ds_prev;
1474 		} else {
1475 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1476 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1477 		}
1478 		after_branch_point =
1479 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1480 
1481 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1482 		if (after_branch_point &&
1483 		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
1484 			VERIFY(0 == zap_remove_int(mos,
1485 			    ds_prev->ds_phys->ds_next_clones_obj, obj, tx));
1486 			if (ds->ds_phys->ds_next_snap_obj != 0) {
1487 				VERIFY(0 == zap_add_int(mos,
1488 				    ds_prev->ds_phys->ds_next_clones_obj,
1489 				    ds->ds_phys->ds_next_snap_obj, tx));
1490 			}
1491 		}
1492 		if (after_branch_point &&
1493 		    ds->ds_phys->ds_next_snap_obj == 0) {
1494 			/* This clone is toast. */
1495 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1496 			ds_prev->ds_phys->ds_num_children--;
1497 		} else if (!after_branch_point) {
1498 			ds_prev->ds_phys->ds_next_snap_obj =
1499 			    ds->ds_phys->ds_next_snap_obj;
1500 		}
1501 	}
1502 
1503 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1504 
1505 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1506 		blkptr_t bp;
1507 		dsl_dataset_t *ds_next;
1508 		uint64_t itor = 0;
1509 		uint64_t old_unique;
1510 		int64_t used = 0, compressed = 0, uncompressed = 0;
1511 
1512 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1513 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1514 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1515 
1516 		old_unique = dsl_dataset_unique(ds_next);
1517 
1518 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1519 		ds_next->ds_phys->ds_prev_snap_obj =
1520 		    ds->ds_phys->ds_prev_snap_obj;
1521 		ds_next->ds_phys->ds_prev_snap_txg =
1522 		    ds->ds_phys->ds_prev_snap_txg;
1523 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1524 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1525 
1526 		/*
1527 		 * Transfer to our deadlist (which will become next's
1528 		 * new deadlist) any entries from next's current
1529 		 * deadlist which were born before prev, and free the
1530 		 * other entries.
1531 		 *
1532 		 * XXX we're doing this long task with the config lock held
1533 		 */
1534 		while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1535 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1536 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1537 				    &bp, tx));
1538 				if (ds_prev && !after_branch_point &&
1539 				    bp.blk_birth >
1540 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1541 					ds_prev->ds_phys->ds_unique_bytes +=
1542 					    bp_get_dasize(dp->dp_spa, &bp);
1543 				}
1544 			} else {
1545 				used += bp_get_dasize(dp->dp_spa, &bp);
1546 				compressed += BP_GET_PSIZE(&bp);
1547 				uncompressed += BP_GET_UCSIZE(&bp);
1548 				/* XXX check return value? */
1549 				(void) dsl_free(zio, dp, tx->tx_txg,
1550 				    &bp, NULL, NULL, ARC_NOWAIT);
1551 			}
1552 		}
1553 
1554 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1555 
1556 		/* change snapused */
1557 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1558 		    -used, -compressed, -uncompressed, tx);
1559 
1560 		/* free next's deadlist */
1561 		bplist_close(&ds_next->ds_deadlist);
1562 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1563 
1564 		/* set next's deadlist to our deadlist */
1565 		bplist_close(&ds->ds_deadlist);
1566 		ds_next->ds_phys->ds_deadlist_obj =
1567 		    ds->ds_phys->ds_deadlist_obj;
1568 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1569 		    ds_next->ds_phys->ds_deadlist_obj));
1570 		ds->ds_phys->ds_deadlist_obj = 0;
1571 
1572 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1573 			/*
1574 			 * Update next's unique to include blocks which
1575 			 * were previously shared by only this snapshot
1576 			 * and it.  Those blocks will be born after the
1577 			 * prev snap and before this snap, and will have
1578 			 * died after the next snap and before the one
1579 			 * after that (ie. be on the snap after next's
1580 			 * deadlist).
1581 			 *
1582 			 * XXX we're doing this long task with the
1583 			 * config lock held
1584 			 */
1585 			dsl_dataset_t *ds_after_next;
1586 			uint64_t space;
1587 
1588 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1589 			    ds_next->ds_phys->ds_next_snap_obj,
1590 			    FTAG, &ds_after_next));
1591 
1592 			VERIFY(0 ==
1593 			    bplist_space_birthrange(&ds_after_next->ds_deadlist,
1594 			    ds->ds_phys->ds_prev_snap_txg,
1595 			    ds->ds_phys->ds_creation_txg, &space));
1596 			ds_next->ds_phys->ds_unique_bytes += space;
1597 
1598 			dsl_dataset_rele(ds_after_next, FTAG);
1599 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1600 		} else {
1601 			ASSERT3P(ds_next->ds_prev, ==, ds);
1602 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1603 			ds_next->ds_prev = NULL;
1604 			if (ds_prev) {
1605 				VERIFY(0 == dsl_dataset_get_ref(dp,
1606 				    ds->ds_phys->ds_prev_snap_obj,
1607 				    ds_next, &ds_next->ds_prev));
1608 			}
1609 
1610 			dsl_dataset_recalc_head_uniq(ds_next);
1611 
1612 			/*
1613 			 * Reduce the amount of our unconsmed refreservation
1614 			 * being charged to our parent by the amount of
1615 			 * new unique data we have gained.
1616 			 */
1617 			if (old_unique < ds_next->ds_reserved) {
1618 				int64_t mrsdelta;
1619 				uint64_t new_unique =
1620 				    ds_next->ds_phys->ds_unique_bytes;
1621 
1622 				ASSERT(old_unique <= new_unique);
1623 				mrsdelta = MIN(new_unique - old_unique,
1624 				    ds_next->ds_reserved - old_unique);
1625 				dsl_dir_diduse_space(ds->ds_dir,
1626 				    DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1627 			}
1628 		}
1629 		dsl_dataset_rele(ds_next, FTAG);
1630 	} else {
1631 		/*
1632 		 * There's no next snapshot, so this is a head dataset.
1633 		 * Destroy the deadlist.  Unless it's a clone, the
1634 		 * deadlist should be empty.  (If it's a clone, it's
1635 		 * safe to ignore the deadlist contents.)
1636 		 */
1637 		struct killarg ka;
1638 
1639 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1640 		bplist_close(&ds->ds_deadlist);
1641 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1642 		ds->ds_phys->ds_deadlist_obj = 0;
1643 
1644 		/*
1645 		 * Free everything that we point to (that's born after
1646 		 * the previous snapshot, if we are a clone)
1647 		 *
1648 		 * NB: this should be very quick, because we already
1649 		 * freed all the objects in open context.
1650 		 */
1651 		ka.ds = ds;
1652 		ka.zio = zio;
1653 		ka.tx = tx;
1654 		err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1655 		    TRAVERSE_POST, kill_blkptr, &ka);
1656 		ASSERT3U(err, ==, 0);
1657 		ASSERT(spa_version(dp->dp_spa) < SPA_VERSION_UNIQUE_ACCURATE ||
1658 		    ds->ds_phys->ds_unique_bytes == 0);
1659 	}
1660 
1661 	err = zio_wait(zio);
1662 	ASSERT3U(err, ==, 0);
1663 
1664 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1665 		/* Erase the link in the dir */
1666 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1667 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1668 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1669 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1670 		ASSERT(err == 0);
1671 	} else {
1672 		/* remove from snapshot namespace */
1673 		dsl_dataset_t *ds_head;
1674 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1675 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1676 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1677 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1678 #ifdef ZFS_DEBUG
1679 		{
1680 			uint64_t val;
1681 
1682 			err = dsl_dataset_snap_lookup(ds_head,
1683 			    ds->ds_snapname, &val);
1684 			ASSERT3U(err, ==, 0);
1685 			ASSERT3U(val, ==, obj);
1686 		}
1687 #endif
1688 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1689 		ASSERT(err == 0);
1690 		dsl_dataset_rele(ds_head, FTAG);
1691 	}
1692 
1693 	if (ds_prev && ds->ds_prev != ds_prev)
1694 		dsl_dataset_rele(ds_prev, FTAG);
1695 
1696 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1697 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1698 	    cr, "dataset = %llu", ds->ds_object);
1699 
1700 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1701 		uint64_t count;
1702 		ASSERT(0 == zap_count(mos,
1703 		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1704 		VERIFY(0 == dmu_object_free(mos,
1705 		    ds->ds_phys->ds_next_clones_obj, tx));
1706 	}
1707 	if (ds->ds_phys->ds_props_obj != 0)
1708 		VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1709 	dsl_dir_close(ds->ds_dir, ds);
1710 	ds->ds_dir = NULL;
1711 	dsl_dataset_drain_refs(ds, tag);
1712 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1713 }
1714 
1715 static int
1716 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1717 {
1718 	uint64_t asize;
1719 
1720 	if (!dmu_tx_is_syncing(tx))
1721 		return (0);
1722 
1723 	/*
1724 	 * If there's an fs-only reservation, any blocks that might become
1725 	 * owned by the snapshot dataset must be accommodated by space
1726 	 * outside of the reservation.
1727 	 */
1728 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1729 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1730 		return (ENOSPC);
1731 
1732 	/*
1733 	 * Propogate any reserved space for this snapshot to other
1734 	 * snapshot checks in this sync group.
1735 	 */
1736 	if (asize > 0)
1737 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1738 
1739 	return (0);
1740 }
1741 
1742 /* ARGSUSED */
1743 int
1744 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1745 {
1746 	dsl_dataset_t *ds = arg1;
1747 	const char *snapname = arg2;
1748 	int err;
1749 	uint64_t value;
1750 
1751 	/*
1752 	 * We don't allow multiple snapshots of the same txg.  If there
1753 	 * is already one, try again.
1754 	 */
1755 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1756 		return (EAGAIN);
1757 
1758 	/*
1759 	 * Check for conflicting name snapshot name.
1760 	 */
1761 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1762 	if (err == 0)
1763 		return (EEXIST);
1764 	if (err != ENOENT)
1765 		return (err);
1766 
1767 	/*
1768 	 * Check that the dataset's name is not too long.  Name consists
1769 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1770 	 */
1771 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1772 		return (ENAMETOOLONG);
1773 
1774 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1775 	if (err)
1776 		return (err);
1777 
1778 	ds->ds_trysnap_txg = tx->tx_txg;
1779 	return (0);
1780 }
1781 
1782 void
1783 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1784 {
1785 	dsl_dataset_t *ds = arg1;
1786 	const char *snapname = arg2;
1787 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1788 	dmu_buf_t *dbuf;
1789 	dsl_dataset_phys_t *dsphys;
1790 	uint64_t dsobj, crtxg;
1791 	objset_t *mos = dp->dp_meta_objset;
1792 	int err;
1793 
1794 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1795 
1796 	/*
1797 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1798 	 */
1799 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1800 		crtxg = 1;
1801 	else
1802 		crtxg = tx->tx_txg;
1803 
1804 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1805 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1806 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1807 	dmu_buf_will_dirty(dbuf, tx);
1808 	dsphys = dbuf->db_data;
1809 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1810 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1811 	dsphys->ds_fsid_guid = unique_create();
1812 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1813 	    sizeof (dsphys->ds_guid));
1814 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1815 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1816 	dsphys->ds_next_snap_obj = ds->ds_object;
1817 	dsphys->ds_num_children = 1;
1818 	dsphys->ds_creation_time = gethrestime_sec();
1819 	dsphys->ds_creation_txg = crtxg;
1820 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1821 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1822 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1823 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1824 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1825 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1826 	dmu_buf_rele(dbuf, FTAG);
1827 
1828 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1829 	if (ds->ds_prev) {
1830 		uint64_t next_clones_obj =
1831 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1832 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1833 		    ds->ds_object ||
1834 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1835 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1836 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1837 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1838 			    ds->ds_prev->ds_phys->ds_creation_txg);
1839 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1840 		} else if (next_clones_obj != 0) {
1841 			VERIFY3U(0, ==, zap_remove_int(mos,
1842 			    next_clones_obj, dsphys->ds_next_snap_obj, tx));
1843 			VERIFY3U(0, ==, zap_add_int(mos,
1844 			    next_clones_obj, dsobj, tx));
1845 		}
1846 	}
1847 
1848 	/*
1849 	 * If we have a reference-reservation on this dataset, we will
1850 	 * need to increase the amount of refreservation being charged
1851 	 * since our unique space is going to zero.
1852 	 */
1853 	if (ds->ds_reserved) {
1854 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1855 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1856 		    add, 0, 0, tx);
1857 	}
1858 
1859 	bplist_close(&ds->ds_deadlist);
1860 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1861 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1862 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1863 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1864 	ds->ds_phys->ds_unique_bytes = 0;
1865 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1866 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1867 	ds->ds_phys->ds_deadlist_obj =
1868 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1869 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1870 	    ds->ds_phys->ds_deadlist_obj));
1871 
1872 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1873 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1874 	    snapname, 8, 1, &dsobj, tx);
1875 	ASSERT(err == 0);
1876 
1877 	if (ds->ds_prev)
1878 		dsl_dataset_drop_ref(ds->ds_prev, ds);
1879 	VERIFY(0 == dsl_dataset_get_ref(dp,
1880 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1881 
1882 	dsl_pool_ds_snapshotted(ds, tx);
1883 
1884 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1885 	    "dataset = %llu", dsobj);
1886 }
1887 
1888 void
1889 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1890 {
1891 	ASSERT(dmu_tx_is_syncing(tx));
1892 	ASSERT(ds->ds_user_ptr != NULL);
1893 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1894 
1895 	/*
1896 	 * in case we had to change ds_fsid_guid when we opened it,
1897 	 * sync it out now.
1898 	 */
1899 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1900 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1901 
1902 	dsl_dir_dirty(ds->ds_dir, tx);
1903 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1904 }
1905 
1906 void
1907 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1908 {
1909 	uint64_t refd, avail, uobjs, aobjs;
1910 
1911 	dsl_dir_stats(ds->ds_dir, nv);
1912 
1913 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1914 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1915 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1916 
1917 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1918 	    ds->ds_phys->ds_creation_time);
1919 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1920 	    ds->ds_phys->ds_creation_txg);
1921 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1922 	    ds->ds_quota);
1923 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1924 	    ds->ds_reserved);
1925 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1926 	    ds->ds_phys->ds_guid);
1927 
1928 	if (ds->ds_phys->ds_next_snap_obj) {
1929 		/*
1930 		 * This is a snapshot; override the dd's space used with
1931 		 * our unique space and compression ratio.
1932 		 */
1933 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1934 		    ds->ds_phys->ds_unique_bytes);
1935 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1936 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1937 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1938 		    ds->ds_phys->ds_compressed_bytes));
1939 	}
1940 }
1941 
1942 void
1943 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1944 {
1945 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1946 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1947 	stat->dds_guid = ds->ds_phys->ds_guid;
1948 	if (ds->ds_phys->ds_next_snap_obj) {
1949 		stat->dds_is_snapshot = B_TRUE;
1950 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1951 	}
1952 
1953 	/* clone origin is really a dsl_dir thing... */
1954 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1955 	if (dsl_dir_is_clone(ds->ds_dir)) {
1956 		dsl_dataset_t *ods;
1957 
1958 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
1959 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1960 		dsl_dataset_name(ods, stat->dds_origin);
1961 		dsl_dataset_drop_ref(ods, FTAG);
1962 	}
1963 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1964 }
1965 
1966 uint64_t
1967 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1968 {
1969 	return (ds->ds_fsid_guid);
1970 }
1971 
1972 void
1973 dsl_dataset_space(dsl_dataset_t *ds,
1974     uint64_t *refdbytesp, uint64_t *availbytesp,
1975     uint64_t *usedobjsp, uint64_t *availobjsp)
1976 {
1977 	*refdbytesp = ds->ds_phys->ds_used_bytes;
1978 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1979 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1980 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1981 	if (ds->ds_quota != 0) {
1982 		/*
1983 		 * Adjust available bytes according to refquota
1984 		 */
1985 		if (*refdbytesp < ds->ds_quota)
1986 			*availbytesp = MIN(*availbytesp,
1987 			    ds->ds_quota - *refdbytesp);
1988 		else
1989 			*availbytesp = 0;
1990 	}
1991 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1992 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1993 }
1994 
1995 boolean_t
1996 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1997 {
1998 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1999 
2000 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2001 	    dsl_pool_sync_context(dp));
2002 	if (ds->ds_prev == NULL)
2003 		return (B_FALSE);
2004 	if (ds->ds_phys->ds_bp.blk_birth >
2005 	    ds->ds_prev->ds_phys->ds_creation_txg)
2006 		return (B_TRUE);
2007 	return (B_FALSE);
2008 }
2009 
2010 /* ARGSUSED */
2011 static int
2012 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2013 {
2014 	dsl_dataset_t *ds = arg1;
2015 	char *newsnapname = arg2;
2016 	dsl_dir_t *dd = ds->ds_dir;
2017 	dsl_dataset_t *hds;
2018 	uint64_t val;
2019 	int err;
2020 
2021 	err = dsl_dataset_hold_obj(dd->dd_pool,
2022 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2023 	if (err)
2024 		return (err);
2025 
2026 	/* new name better not be in use */
2027 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2028 	dsl_dataset_rele(hds, FTAG);
2029 
2030 	if (err == 0)
2031 		err = EEXIST;
2032 	else if (err == ENOENT)
2033 		err = 0;
2034 
2035 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2036 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2037 		err = ENAMETOOLONG;
2038 
2039 	return (err);
2040 }
2041 
2042 static void
2043 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
2044     cred_t *cr, dmu_tx_t *tx)
2045 {
2046 	dsl_dataset_t *ds = arg1;
2047 	const char *newsnapname = arg2;
2048 	dsl_dir_t *dd = ds->ds_dir;
2049 	objset_t *mos = dd->dd_pool->dp_meta_objset;
2050 	dsl_dataset_t *hds;
2051 	int err;
2052 
2053 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2054 
2055 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2056 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2057 
2058 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2059 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2060 	ASSERT3U(err, ==, 0);
2061 	mutex_enter(&ds->ds_lock);
2062 	(void) strcpy(ds->ds_snapname, newsnapname);
2063 	mutex_exit(&ds->ds_lock);
2064 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2065 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2066 	ASSERT3U(err, ==, 0);
2067 
2068 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2069 	    cr, "dataset = %llu", ds->ds_object);
2070 	dsl_dataset_rele(hds, FTAG);
2071 }
2072 
2073 struct renamesnaparg {
2074 	dsl_sync_task_group_t *dstg;
2075 	char failed[MAXPATHLEN];
2076 	char *oldsnap;
2077 	char *newsnap;
2078 };
2079 
2080 static int
2081 dsl_snapshot_rename_one(char *name, void *arg)
2082 {
2083 	struct renamesnaparg *ra = arg;
2084 	dsl_dataset_t *ds = NULL;
2085 	char *cp;
2086 	int err;
2087 
2088 	cp = name + strlen(name);
2089 	*cp = '@';
2090 	(void) strcpy(cp + 1, ra->oldsnap);
2091 
2092 	/*
2093 	 * For recursive snapshot renames the parent won't be changing
2094 	 * so we just pass name for both the to/from argument.
2095 	 */
2096 	err = zfs_secpolicy_rename_perms(name, name, CRED());
2097 	if (err == ENOENT) {
2098 		return (0);
2099 	} else if (err) {
2100 		(void) strcpy(ra->failed, name);
2101 		return (err);
2102 	}
2103 
2104 #ifdef _KERNEL
2105 	/*
2106 	 * For all filesystems undergoing rename, we'll need to unmount it.
2107 	 */
2108 	(void) zfs_unmount_snap(name, NULL);
2109 #endif
2110 	err = dsl_dataset_hold(name, ra->dstg, &ds);
2111 	*cp = '\0';
2112 	if (err == ENOENT) {
2113 		return (0);
2114 	} else if (err) {
2115 		(void) strcpy(ra->failed, name);
2116 		return (err);
2117 	}
2118 
2119 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2120 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2121 
2122 	return (0);
2123 }
2124 
2125 static int
2126 dsl_recursive_rename(char *oldname, const char *newname)
2127 {
2128 	int err;
2129 	struct renamesnaparg *ra;
2130 	dsl_sync_task_t *dst;
2131 	spa_t *spa;
2132 	char *cp, *fsname = spa_strdup(oldname);
2133 	int len = strlen(oldname);
2134 
2135 	/* truncate the snapshot name to get the fsname */
2136 	cp = strchr(fsname, '@');
2137 	*cp = '\0';
2138 
2139 	err = spa_open(fsname, &spa, FTAG);
2140 	if (err) {
2141 		kmem_free(fsname, len + 1);
2142 		return (err);
2143 	}
2144 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2145 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2146 
2147 	ra->oldsnap = strchr(oldname, '@') + 1;
2148 	ra->newsnap = strchr(newname, '@') + 1;
2149 	*ra->failed = '\0';
2150 
2151 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2152 	    DS_FIND_CHILDREN);
2153 	kmem_free(fsname, len + 1);
2154 
2155 	if (err == 0) {
2156 		err = dsl_sync_task_group_wait(ra->dstg);
2157 	}
2158 
2159 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2160 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2161 		dsl_dataset_t *ds = dst->dst_arg1;
2162 		if (dst->dst_err) {
2163 			dsl_dir_name(ds->ds_dir, ra->failed);
2164 			(void) strcat(ra->failed, "@");
2165 			(void) strcat(ra->failed, ra->newsnap);
2166 		}
2167 		dsl_dataset_rele(ds, ra->dstg);
2168 	}
2169 
2170 	if (err)
2171 		(void) strcpy(oldname, ra->failed);
2172 
2173 	dsl_sync_task_group_destroy(ra->dstg);
2174 	kmem_free(ra, sizeof (struct renamesnaparg));
2175 	spa_close(spa, FTAG);
2176 	return (err);
2177 }
2178 
2179 static int
2180 dsl_valid_rename(char *oldname, void *arg)
2181 {
2182 	int delta = *(int *)arg;
2183 
2184 	if (strlen(oldname) + delta >= MAXNAMELEN)
2185 		return (ENAMETOOLONG);
2186 
2187 	return (0);
2188 }
2189 
2190 #pragma weak dmu_objset_rename = dsl_dataset_rename
2191 int
2192 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2193 {
2194 	dsl_dir_t *dd;
2195 	dsl_dataset_t *ds;
2196 	const char *tail;
2197 	int err;
2198 
2199 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2200 	if (err)
2201 		return (err);
2202 	if (tail == NULL) {
2203 		int delta = strlen(newname) - strlen(oldname);
2204 
2205 		/* if we're growing, validate child name lengths */
2206 		if (delta > 0)
2207 			err = dmu_objset_find(oldname, dsl_valid_rename,
2208 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2209 
2210 		if (!err)
2211 			err = dsl_dir_rename(dd, newname);
2212 		dsl_dir_close(dd, FTAG);
2213 		return (err);
2214 	}
2215 	if (tail[0] != '@') {
2216 		/* the name ended in a nonexistant component */
2217 		dsl_dir_close(dd, FTAG);
2218 		return (ENOENT);
2219 	}
2220 
2221 	dsl_dir_close(dd, FTAG);
2222 
2223 	/* new name must be snapshot in same filesystem */
2224 	tail = strchr(newname, '@');
2225 	if (tail == NULL)
2226 		return (EINVAL);
2227 	tail++;
2228 	if (strncmp(oldname, newname, tail - newname) != 0)
2229 		return (EXDEV);
2230 
2231 	if (recursive) {
2232 		err = dsl_recursive_rename(oldname, newname);
2233 	} else {
2234 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2235 		if (err)
2236 			return (err);
2237 
2238 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2239 		    dsl_dataset_snapshot_rename_check,
2240 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2241 
2242 		dsl_dataset_rele(ds, FTAG);
2243 	}
2244 
2245 	return (err);
2246 }
2247 
2248 struct promotenode {
2249 	list_node_t link;
2250 	dsl_dataset_t *ds;
2251 };
2252 
2253 struct promotearg {
2254 	list_t shared_snaps, origin_snaps, clone_snaps;
2255 	dsl_dataset_t *origin_origin, *origin_head;
2256 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2257 };
2258 
2259 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2260 
2261 /* ARGSUSED */
2262 static int
2263 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2264 {
2265 	dsl_dataset_t *hds = arg1;
2266 	struct promotearg *pa = arg2;
2267 	struct promotenode *snap = list_head(&pa->shared_snaps);
2268 	dsl_dataset_t *origin_ds = snap->ds;
2269 	int err;
2270 
2271 	/* Check that it is a real clone */
2272 	if (!dsl_dir_is_clone(hds->ds_dir))
2273 		return (EINVAL);
2274 
2275 	/* Since this is so expensive, don't do the preliminary check */
2276 	if (!dmu_tx_is_syncing(tx))
2277 		return (0);
2278 
2279 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2280 		return (EXDEV);
2281 
2282 	/* compute origin's new unique space */
2283 	snap = list_tail(&pa->clone_snaps);
2284 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2285 	err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2286 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, &pa->unique);
2287 	if (err)
2288 		return (err);
2289 
2290 	/*
2291 	 * Walk the snapshots that we are moving
2292 	 *
2293 	 * Compute space to transfer.  Consider the incremental changes
2294 	 * to used for each snapshot:
2295 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2296 	 * So each snapshot gave birth to:
2297 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2298 	 * So a sequence would look like:
2299 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2300 	 * Which simplifies to:
2301 	 * uN + kN + kN-1 + ... + k1 + k0
2302 	 * Note however, if we stop before we reach the ORIGIN we get:
2303 	 * uN + kN + kN-1 + ... + kM - uM-1
2304 	 */
2305 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2306 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2307 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2308 	for (snap = list_head(&pa->shared_snaps); snap;
2309 	    snap = list_next(&pa->shared_snaps, snap)) {
2310 		uint64_t val, dlused, dlcomp, dluncomp;
2311 		dsl_dataset_t *ds = snap->ds;
2312 
2313 		/* Check that the snapshot name does not conflict */
2314 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2315 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2316 		if (err == 0)
2317 			return (EEXIST);
2318 		if (err != ENOENT)
2319 			return (err);
2320 
2321 		/* The very first snapshot does not have a deadlist */
2322 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2323 			continue;
2324 
2325 		if (err = bplist_space(&ds->ds_deadlist,
2326 		    &dlused, &dlcomp, &dluncomp))
2327 			return (err);
2328 		pa->used += dlused;
2329 		pa->comp += dlcomp;
2330 		pa->uncomp += dluncomp;
2331 	}
2332 
2333 	/*
2334 	 * If we are a clone of a clone then we never reached ORIGIN,
2335 	 * so we need to subtract out the clone origin's used space.
2336 	 */
2337 	if (pa->origin_origin) {
2338 		pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2339 		pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2340 		pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2341 	}
2342 
2343 	/* Check that there is enough space here */
2344 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2345 	    pa->used);
2346 	if (err)
2347 		return (err);
2348 
2349 	/*
2350 	 * Compute the amounts of space that will be used by snapshots
2351 	 * after the promotion (for both origin and clone).  For each,
2352 	 * it is the amount of space that will be on all of their
2353 	 * deadlists (that was not born before their new origin).
2354 	 */
2355 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2356 		uint64_t space;
2357 
2358 		/*
2359 		 * Note, typically this will not be a clone of a clone,
2360 		 * so snap->ds->ds_origin_txg will be < TXG_INITIAL, so
2361 		 * these snaplist_space() -> bplist_space_birthrange()
2362 		 * calls will be fast because they do not have to
2363 		 * iterate over all bps.
2364 		 */
2365 		snap = list_head(&pa->origin_snaps);
2366 		err = snaplist_space(&pa->shared_snaps,
2367 		    snap->ds->ds_origin_txg, &pa->cloneusedsnap);
2368 		if (err)
2369 			return (err);
2370 
2371 		err = snaplist_space(&pa->clone_snaps,
2372 		    snap->ds->ds_origin_txg, &space);
2373 		if (err)
2374 			return (err);
2375 		pa->cloneusedsnap += space;
2376 	}
2377 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2378 		err = snaplist_space(&pa->origin_snaps,
2379 		    origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2380 		if (err)
2381 			return (err);
2382 	}
2383 
2384 	return (0);
2385 }
2386 
2387 static void
2388 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2389 {
2390 	dsl_dataset_t *hds = arg1;
2391 	struct promotearg *pa = arg2;
2392 	struct promotenode *snap = list_head(&pa->shared_snaps);
2393 	dsl_dataset_t *origin_ds = snap->ds;
2394 	dsl_dataset_t *origin_head;
2395 	dsl_dir_t *dd = hds->ds_dir;
2396 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2397 	dsl_dir_t *odd = NULL;
2398 	uint64_t oldnext_obj;
2399 	int64_t delta;
2400 
2401 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2402 
2403 	snap = list_head(&pa->origin_snaps);
2404 	origin_head = snap->ds;
2405 
2406 	/*
2407 	 * We need to explicitly open odd, since origin_ds's dd will be
2408 	 * changing.
2409 	 */
2410 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2411 	    NULL, FTAG, &odd));
2412 
2413 	/* change origin's next snap */
2414 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2415 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2416 	snap = list_tail(&pa->clone_snaps);
2417 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2418 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2419 
2420 	/* change the origin's next clone */
2421 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2422 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2423 		    origin_ds->ds_phys->ds_next_clones_obj,
2424 		    origin_ds->ds_phys->ds_next_snap_obj, tx));
2425 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2426 		    origin_ds->ds_phys->ds_next_clones_obj,
2427 		    oldnext_obj, tx));
2428 	}
2429 
2430 	/* change origin */
2431 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2432 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2433 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2434 	hds->ds_origin_txg = origin_head->ds_origin_txg;
2435 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2436 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2437 	origin_head->ds_origin_txg = origin_ds->ds_phys->ds_creation_txg;
2438 
2439 	/* move snapshots to this dir */
2440 	for (snap = list_head(&pa->shared_snaps); snap;
2441 	    snap = list_next(&pa->shared_snaps, snap)) {
2442 		dsl_dataset_t *ds = snap->ds;
2443 
2444 		/* unregister props as dsl_dir is changing */
2445 		if (ds->ds_user_ptr) {
2446 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
2447 			ds->ds_user_ptr = NULL;
2448 		}
2449 		/* move snap name entry */
2450 		VERIFY(0 == dsl_dataset_get_snapname(ds));
2451 		VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2452 		    ds->ds_snapname, tx));
2453 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2454 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2455 		    8, 1, &ds->ds_object, tx));
2456 		/* change containing dsl_dir */
2457 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2458 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2459 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2460 		ASSERT3P(ds->ds_dir, ==, odd);
2461 		dsl_dir_close(ds->ds_dir, ds);
2462 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2463 		    NULL, ds, &ds->ds_dir));
2464 
2465 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2466 	}
2467 
2468 	/*
2469 	 * Change space accounting.
2470 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2471 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2472 	 * is true for each of {clone,origin} independently.
2473 	 */
2474 
2475 	delta = pa->cloneusedsnap -
2476 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2477 	ASSERT3S(delta, >=, 0);
2478 	ASSERT3U(pa->used, >=, delta);
2479 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2480 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2481 	    pa->used - delta, pa->comp, pa->uncomp, tx);
2482 
2483 	delta = pa->originusedsnap -
2484 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2485 	ASSERT3S(delta, <=, 0);
2486 	ASSERT3U(pa->used, >=, -delta);
2487 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2488 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2489 	    -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2490 
2491 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2492 
2493 	/* log history record */
2494 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2495 	    cr, "dataset = %llu", hds->ds_object);
2496 
2497 	dsl_dir_close(odd, FTAG);
2498 }
2499 
2500 static char *snaplist_tag = "snaplist";
2501 /*
2502  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2503  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2504  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2505  * snapshots back to this dataset's origin.
2506  */
2507 static int
2508 snaplist_make(dsl_pool_t *dp, boolean_t own,
2509     uint64_t first_obj, uint64_t last_obj, list_t *l)
2510 {
2511 	uint64_t obj = last_obj;
2512 
2513 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2514 
2515 	list_create(l, sizeof (struct promotenode),
2516 	    offsetof(struct promotenode, link));
2517 
2518 	while (obj != first_obj) {
2519 		dsl_dataset_t *ds;
2520 		struct promotenode *snap;
2521 		int err;
2522 
2523 		if (own) {
2524 			err = dsl_dataset_own_obj(dp, obj,
2525 			    0, snaplist_tag, &ds);
2526 			if (err == 0)
2527 				dsl_dataset_make_exclusive(ds, snaplist_tag);
2528 		} else {
2529 			err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2530 		}
2531 		if (err == ENOENT) {
2532 			/* lost race with snapshot destroy */
2533 			struct promotenode *last = list_tail(l);
2534 			ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2535 			obj = last->ds->ds_phys->ds_prev_snap_obj;
2536 			continue;
2537 		} else if (err) {
2538 			return (err);
2539 		}
2540 
2541 		if (first_obj == 0)
2542 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2543 
2544 		snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2545 		snap->ds = ds;
2546 		list_insert_tail(l, snap);
2547 		obj = ds->ds_phys->ds_prev_snap_obj;
2548 	}
2549 
2550 	return (0);
2551 }
2552 
2553 static int
2554 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2555 {
2556 	struct promotenode *snap;
2557 
2558 	*spacep = 0;
2559 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2560 		uint64_t used;
2561 		int err = bplist_space_birthrange(&snap->ds->ds_deadlist,
2562 		    mintxg, UINT64_MAX, &used);
2563 		if (err)
2564 			return (err);
2565 		*spacep += used;
2566 	}
2567 	return (0);
2568 }
2569 
2570 static void
2571 snaplist_destroy(list_t *l, boolean_t own)
2572 {
2573 	struct promotenode *snap;
2574 
2575 	if (!list_link_active(&l->list_head))
2576 		return;
2577 
2578 	while ((snap = list_tail(l)) != NULL) {
2579 		list_remove(l, snap);
2580 		if (own)
2581 			dsl_dataset_disown(snap->ds, snaplist_tag);
2582 		else
2583 			dsl_dataset_rele(snap->ds, snaplist_tag);
2584 		kmem_free(snap, sizeof (struct promotenode));
2585 	}
2586 	list_destroy(l);
2587 }
2588 
2589 /*
2590  * Promote a clone.  Nomenclature note:
2591  * "clone" or "cds": the original clone which is being promoted
2592  * "origin" or "ods": the snapshot which is originally clone's origin
2593  * "origin head" or "ohds": the dataset which is the head
2594  * (filesystem/volume) for the origin
2595  * "origin origin": the origin of the origin's filesystem (typically
2596  * NULL, indicating that the clone is not a clone of a clone).
2597  */
2598 int
2599 dsl_dataset_promote(const char *name)
2600 {
2601 	dsl_dataset_t *ds;
2602 	dsl_dir_t *dd;
2603 	dsl_pool_t *dp;
2604 	dmu_object_info_t doi;
2605 	struct promotearg pa = { 0 };
2606 	struct promotenode *snap;
2607 	int err;
2608 
2609 	err = dsl_dataset_hold(name, FTAG, &ds);
2610 	if (err)
2611 		return (err);
2612 	dd = ds->ds_dir;
2613 	dp = dd->dd_pool;
2614 
2615 	err = dmu_object_info(dp->dp_meta_objset,
2616 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2617 	if (err) {
2618 		dsl_dataset_rele(ds, FTAG);
2619 		return (err);
2620 	}
2621 
2622 	if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2623 		dsl_dataset_rele(ds, FTAG);
2624 		return (EINVAL);
2625 	}
2626 
2627 	/*
2628 	 * We are going to inherit all the snapshots taken before our
2629 	 * origin (i.e., our new origin will be our parent's origin).
2630 	 * Take ownership of them so that we can rename them into our
2631 	 * namespace.
2632 	 */
2633 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2634 
2635 	err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2636 	    &pa.shared_snaps);
2637 	if (err != 0)
2638 		goto out;
2639 
2640 	err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2641 	if (err != 0)
2642 		goto out;
2643 
2644 	snap = list_head(&pa.shared_snaps);
2645 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2646 	err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2647 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2648 	if (err != 0)
2649 		goto out;
2650 
2651 	if (dsl_dir_is_clone(snap->ds->ds_dir)) {
2652 		err = dsl_dataset_own_obj(dp,
2653 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2654 		    0, FTAG, &pa.origin_origin);
2655 		if (err != 0)
2656 			goto out;
2657 	}
2658 
2659 out:
2660 	rw_exit(&dp->dp_config_rwlock);
2661 
2662 	/*
2663 	 * Add in 128x the snapnames zapobj size, since we will be moving
2664 	 * a bunch of snapnames to the promoted ds, and dirtying their
2665 	 * bonus buffers.
2666 	 */
2667 	if (err == 0) {
2668 		err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2669 		    dsl_dataset_promote_sync, ds, &pa,
2670 		    2 + 2 * doi.doi_physical_blks);
2671 	}
2672 
2673 	snaplist_destroy(&pa.shared_snaps, B_TRUE);
2674 	snaplist_destroy(&pa.clone_snaps, B_FALSE);
2675 	snaplist_destroy(&pa.origin_snaps, B_FALSE);
2676 	if (pa.origin_origin)
2677 		dsl_dataset_disown(pa.origin_origin, FTAG);
2678 	dsl_dataset_rele(ds, FTAG);
2679 	return (err);
2680 }
2681 
2682 struct cloneswaparg {
2683 	dsl_dataset_t *cds; /* clone dataset */
2684 	dsl_dataset_t *ohds; /* origin's head dataset */
2685 	boolean_t force;
2686 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2687 };
2688 
2689 /* ARGSUSED */
2690 static int
2691 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2692 {
2693 	struct cloneswaparg *csa = arg1;
2694 
2695 	/* they should both be heads */
2696 	if (dsl_dataset_is_snapshot(csa->cds) ||
2697 	    dsl_dataset_is_snapshot(csa->ohds))
2698 		return (EINVAL);
2699 
2700 	/* the branch point should be just before them */
2701 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2702 		return (EINVAL);
2703 
2704 	/* cds should be the clone */
2705 	if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2706 	    csa->ohds->ds_object)
2707 		return (EINVAL);
2708 
2709 	/* the clone should be a child of the origin */
2710 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2711 		return (EINVAL);
2712 
2713 	/* ohds shouldn't be modified unless 'force' */
2714 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2715 		return (ETXTBSY);
2716 
2717 	/* adjust amount of any unconsumed refreservation */
2718 	csa->unused_refres_delta =
2719 	    (int64_t)MIN(csa->ohds->ds_reserved,
2720 	    csa->ohds->ds_phys->ds_unique_bytes) -
2721 	    (int64_t)MIN(csa->ohds->ds_reserved,
2722 	    csa->cds->ds_phys->ds_unique_bytes);
2723 
2724 	if (csa->unused_refres_delta > 0 &&
2725 	    csa->unused_refres_delta >
2726 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2727 		return (ENOSPC);
2728 
2729 	return (0);
2730 }
2731 
2732 /* ARGSUSED */
2733 static void
2734 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2735 {
2736 	struct cloneswaparg *csa = arg1;
2737 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2738 
2739 	ASSERT(csa->cds->ds_reserved == 0);
2740 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2741 
2742 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2743 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2744 	dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2745 
2746 	if (csa->cds->ds_user_ptr != NULL) {
2747 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2748 		csa->cds->ds_user_ptr = NULL;
2749 	}
2750 
2751 	if (csa->ohds->ds_user_ptr != NULL) {
2752 		csa->ohds->ds_user_evict_func(csa->ohds,
2753 		    csa->ohds->ds_user_ptr);
2754 		csa->ohds->ds_user_ptr = NULL;
2755 	}
2756 
2757 	/* reset origin's unique bytes */
2758 	VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2759 	    csa->cds->ds_prev->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2760 	    &csa->cds->ds_prev->ds_phys->ds_unique_bytes));
2761 
2762 	/* swap blkptrs */
2763 	{
2764 		blkptr_t tmp;
2765 		tmp = csa->ohds->ds_phys->ds_bp;
2766 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2767 		csa->cds->ds_phys->ds_bp = tmp;
2768 	}
2769 
2770 	/* set dd_*_bytes */
2771 	{
2772 		int64_t dused, dcomp, duncomp;
2773 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2774 		uint64_t odl_used, odl_comp, odl_uncomp;
2775 
2776 		ASSERT3U(csa->cds->ds_dir->dd_phys->
2777 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2778 
2779 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2780 		    &cdl_comp, &cdl_uncomp));
2781 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2782 		    &odl_comp, &odl_uncomp));
2783 
2784 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2785 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2786 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2787 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2788 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2789 		    cdl_uncomp -
2790 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2791 
2792 		dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
2793 		    dused, dcomp, duncomp, tx);
2794 		dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
2795 		    -dused, -dcomp, -duncomp, tx);
2796 
2797 		/*
2798 		 * The difference in the space used by snapshots is the
2799 		 * difference in snapshot space due to the head's
2800 		 * deadlist (since that's the only thing that's
2801 		 * changing that affects the snapused).
2802 		 */
2803 		VERIFY(0 == bplist_space_birthrange(&csa->cds->ds_deadlist,
2804 		    csa->ohds->ds_origin_txg, UINT64_MAX, &cdl_used));
2805 		VERIFY(0 == bplist_space_birthrange(&csa->ohds->ds_deadlist,
2806 		    csa->ohds->ds_origin_txg, UINT64_MAX, &odl_used));
2807 		dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
2808 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2809 	}
2810 
2811 #define	SWITCH64(x, y) \
2812 	{ \
2813 		uint64_t __tmp = (x); \
2814 		(x) = (y); \
2815 		(y) = __tmp; \
2816 	}
2817 
2818 	/* swap ds_*_bytes */
2819 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2820 	    csa->cds->ds_phys->ds_used_bytes);
2821 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2822 	    csa->cds->ds_phys->ds_compressed_bytes);
2823 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2824 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2825 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2826 	    csa->cds->ds_phys->ds_unique_bytes);
2827 
2828 	/* apply any parent delta for change in unconsumed refreservation */
2829 	dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
2830 	    csa->unused_refres_delta, 0, 0, tx);
2831 
2832 	/* swap deadlists */
2833 	bplist_close(&csa->cds->ds_deadlist);
2834 	bplist_close(&csa->ohds->ds_deadlist);
2835 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2836 	    csa->cds->ds_phys->ds_deadlist_obj);
2837 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2838 	    csa->cds->ds_phys->ds_deadlist_obj));
2839 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2840 	    csa->ohds->ds_phys->ds_deadlist_obj));
2841 
2842 	dsl_pool_ds_clone_swapped(csa->ohds, csa->cds, tx);
2843 }
2844 
2845 /*
2846  * Swap 'clone' with its origin head file system.  Used at the end
2847  * of "online recv" to swizzle the file system to the new version.
2848  */
2849 int
2850 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2851     boolean_t force)
2852 {
2853 	struct cloneswaparg csa;
2854 	int error;
2855 
2856 	ASSERT(clone->ds_owner);
2857 	ASSERT(origin_head->ds_owner);
2858 retry:
2859 	/* Need exclusive access for the swap */
2860 	rw_enter(&clone->ds_rwlock, RW_WRITER);
2861 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
2862 		rw_exit(&clone->ds_rwlock);
2863 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
2864 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
2865 			rw_exit(&origin_head->ds_rwlock);
2866 			goto retry;
2867 		}
2868 	}
2869 	csa.cds = clone;
2870 	csa.ohds = origin_head;
2871 	csa.force = force;
2872 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
2873 	    dsl_dataset_clone_swap_check,
2874 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
2875 	return (error);
2876 }
2877 
2878 /*
2879  * Given a pool name and a dataset object number in that pool,
2880  * return the name of that dataset.
2881  */
2882 int
2883 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2884 {
2885 	spa_t *spa;
2886 	dsl_pool_t *dp;
2887 	dsl_dataset_t *ds;
2888 	int error;
2889 
2890 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2891 		return (error);
2892 	dp = spa_get_dsl(spa);
2893 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2894 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
2895 		dsl_dataset_name(ds, buf);
2896 		dsl_dataset_rele(ds, FTAG);
2897 	}
2898 	rw_exit(&dp->dp_config_rwlock);
2899 	spa_close(spa, FTAG);
2900 
2901 	return (error);
2902 }
2903 
2904 int
2905 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2906     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2907 {
2908 	int error = 0;
2909 
2910 	ASSERT3S(asize, >, 0);
2911 
2912 	/*
2913 	 * *ref_rsrv is the portion of asize that will come from any
2914 	 * unconsumed refreservation space.
2915 	 */
2916 	*ref_rsrv = 0;
2917 
2918 	mutex_enter(&ds->ds_lock);
2919 	/*
2920 	 * Make a space adjustment for reserved bytes.
2921 	 */
2922 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2923 		ASSERT3U(*used, >=,
2924 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2925 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2926 		*ref_rsrv =
2927 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2928 	}
2929 
2930 	if (!check_quota || ds->ds_quota == 0) {
2931 		mutex_exit(&ds->ds_lock);
2932 		return (0);
2933 	}
2934 	/*
2935 	 * If they are requesting more space, and our current estimate
2936 	 * is over quota, they get to try again unless the actual
2937 	 * on-disk is over quota and there are no pending changes (which
2938 	 * may free up space for us).
2939 	 */
2940 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2941 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2942 			error = ERESTART;
2943 		else
2944 			error = EDQUOT;
2945 	}
2946 	mutex_exit(&ds->ds_lock);
2947 
2948 	return (error);
2949 }
2950 
2951 /* ARGSUSED */
2952 static int
2953 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2954 {
2955 	dsl_dataset_t *ds = arg1;
2956 	uint64_t *quotap = arg2;
2957 	uint64_t new_quota = *quotap;
2958 
2959 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2960 		return (ENOTSUP);
2961 
2962 	if (new_quota == 0)
2963 		return (0);
2964 
2965 	if (new_quota < ds->ds_phys->ds_used_bytes ||
2966 	    new_quota < ds->ds_reserved)
2967 		return (ENOSPC);
2968 
2969 	return (0);
2970 }
2971 
2972 /* ARGSUSED */
2973 void
2974 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2975 {
2976 	dsl_dataset_t *ds = arg1;
2977 	uint64_t *quotap = arg2;
2978 	uint64_t new_quota = *quotap;
2979 
2980 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2981 
2982 	ds->ds_quota = new_quota;
2983 
2984 	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
2985 
2986 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
2987 	    tx, cr, "%lld dataset = %llu ",
2988 	    (longlong_t)new_quota, ds->ds_object);
2989 }
2990 
2991 int
2992 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
2993 {
2994 	dsl_dataset_t *ds;
2995 	int err;
2996 
2997 	err = dsl_dataset_hold(dsname, FTAG, &ds);
2998 	if (err)
2999 		return (err);
3000 
3001 	if (quota != ds->ds_quota) {
3002 		/*
3003 		 * If someone removes a file, then tries to set the quota, we
3004 		 * want to make sure the file freeing takes effect.
3005 		 */
3006 		txg_wait_open(ds->ds_dir->dd_pool, 0);
3007 
3008 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3009 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3010 		    ds, &quota, 0);
3011 	}
3012 	dsl_dataset_rele(ds, FTAG);
3013 	return (err);
3014 }
3015 
3016 static int
3017 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3018 {
3019 	dsl_dataset_t *ds = arg1;
3020 	uint64_t *reservationp = arg2;
3021 	uint64_t new_reservation = *reservationp;
3022 	int64_t delta;
3023 	uint64_t unique;
3024 
3025 	if (new_reservation > INT64_MAX)
3026 		return (EOVERFLOW);
3027 
3028 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3029 	    SPA_VERSION_REFRESERVATION)
3030 		return (ENOTSUP);
3031 
3032 	if (dsl_dataset_is_snapshot(ds))
3033 		return (EINVAL);
3034 
3035 	/*
3036 	 * If we are doing the preliminary check in open context, the
3037 	 * space estimates may be inaccurate.
3038 	 */
3039 	if (!dmu_tx_is_syncing(tx))
3040 		return (0);
3041 
3042 	mutex_enter(&ds->ds_lock);
3043 	unique = dsl_dataset_unique(ds);
3044 	delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved);
3045 	mutex_exit(&ds->ds_lock);
3046 
3047 	if (delta > 0 &&
3048 	    delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3049 		return (ENOSPC);
3050 	if (delta > 0 && ds->ds_quota > 0 &&
3051 	    new_reservation > ds->ds_quota)
3052 		return (ENOSPC);
3053 
3054 	return (0);
3055 }
3056 
3057 /* ARGSUSED */
3058 static void
3059 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
3060     dmu_tx_t *tx)
3061 {
3062 	dsl_dataset_t *ds = arg1;
3063 	uint64_t *reservationp = arg2;
3064 	uint64_t new_reservation = *reservationp;
3065 	uint64_t unique;
3066 	int64_t delta;
3067 
3068 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3069 
3070 	mutex_enter(&ds->ds_dir->dd_lock);
3071 	mutex_enter(&ds->ds_lock);
3072 	unique = dsl_dataset_unique(ds);
3073 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
3074 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3075 	ds->ds_reserved = new_reservation;
3076 	mutex_exit(&ds->ds_lock);
3077 
3078 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3079 	mutex_exit(&ds->ds_dir->dd_lock);
3080 	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
3081 	    new_reservation, cr, tx);
3082 
3083 	spa_history_internal_log(LOG_DS_REFRESERV,
3084 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
3085 	    (longlong_t)new_reservation, ds->ds_object);
3086 }
3087 
3088 int
3089 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
3090 {
3091 	dsl_dataset_t *ds;
3092 	int err;
3093 
3094 	err = dsl_dataset_hold(dsname, FTAG, &ds);
3095 	if (err)
3096 		return (err);
3097 
3098 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3099 	    dsl_dataset_set_reservation_check,
3100 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
3101 	dsl_dataset_rele(ds, FTAG);
3102 	return (err);
3103 }
3104