xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 0cd13cbfb4270b840b4bd22ec5f673b2b6a2c02b)
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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/dmu_objset.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_prop.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/arc.h>
36 #include <sys/zio.h>
37 #include <sys/zap.h>
38 #include <sys/unique.h>
39 #include <sys/zfs_context.h>
40 #include <sys/zfs_ioctl.h>
41 #include <sys/spa.h>
42 #include <sys/sunddi.h>
43 
44 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
45 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
46 static dsl_checkfunc_t dsl_dataset_rollback_check;
47 static dsl_syncfunc_t dsl_dataset_rollback_sync;
48 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
49 
50 #define	DS_REF_MAX	(1ULL << 62)
51 
52 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
53 
54 /*
55  * We use weighted reference counts to express the various forms of exclusion
56  * between different open modes.  A STANDARD open is 1 point, an EXCLUSIVE open
57  * is DS_REF_MAX, and a PRIMARY open is little more than half of an EXCLUSIVE.
58  * This makes the exclusion logic simple: the total refcnt for all opens cannot
59  * exceed DS_REF_MAX.  For example, EXCLUSIVE opens are exclusive because their
60  * weight (DS_REF_MAX) consumes the entire refcnt space.  PRIMARY opens consume
61  * just over half of the refcnt space, so there can't be more than one, but it
62  * can peacefully coexist with any number of STANDARD opens.
63  */
64 static uint64_t ds_refcnt_weight[DS_MODE_LEVELS] = {
65 	0,			/* DS_MODE_NONE - invalid		*/
66 	1,			/* DS_MODE_STANDARD - unlimited number	*/
67 	(DS_REF_MAX >> 1) + 1,	/* DS_MODE_PRIMARY - only one of these	*/
68 	DS_REF_MAX		/* DS_MODE_EXCLUSIVE - no other opens	*/
69 };
70 
71 /*
72  * Figure out how much of this delta should be propogated to the dsl_dir
73  * layer.  If there's a refreservation, that space has already been
74  * partially accounted for in our ancestors.
75  */
76 static int64_t
77 parent_delta(dsl_dataset_t *ds, int64_t delta)
78 {
79 	uint64_t old_bytes, new_bytes;
80 
81 	if (ds->ds_reserved == 0)
82 		return (delta);
83 
84 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
85 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
86 
87 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
88 	return (new_bytes - old_bytes);
89 }
90 
91 void
92 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
93 {
94 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
95 	int compressed = BP_GET_PSIZE(bp);
96 	int uncompressed = BP_GET_UCSIZE(bp);
97 	int64_t delta;
98 
99 	dprintf_bp(bp, "born, ds=%p\n", ds);
100 
101 	ASSERT(dmu_tx_is_syncing(tx));
102 	/* It could have been compressed away to nothing */
103 	if (BP_IS_HOLE(bp))
104 		return;
105 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
106 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
107 	if (ds == NULL) {
108 		/*
109 		 * Account for the meta-objset space in its placeholder
110 		 * dsl_dir.
111 		 */
112 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
113 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
114 		    used, compressed, uncompressed, tx);
115 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
116 		return;
117 	}
118 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
119 	mutex_enter(&ds->ds_lock);
120 	delta = parent_delta(ds, used);
121 	ds->ds_phys->ds_used_bytes += used;
122 	ds->ds_phys->ds_compressed_bytes += compressed;
123 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
124 	ds->ds_phys->ds_unique_bytes += used;
125 	mutex_exit(&ds->ds_lock);
126 	dsl_dir_diduse_space(ds->ds_dir, delta, compressed, uncompressed, tx);
127 }
128 
129 void
130 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
131     dmu_tx_t *tx)
132 {
133 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
134 	int compressed = BP_GET_PSIZE(bp);
135 	int uncompressed = BP_GET_UCSIZE(bp);
136 
137 	ASSERT(dmu_tx_is_syncing(tx));
138 	/* No block pointer => nothing to free */
139 	if (BP_IS_HOLE(bp))
140 		return;
141 
142 	ASSERT(used > 0);
143 	if (ds == NULL) {
144 		int err;
145 		/*
146 		 * Account for the meta-objset space in its placeholder
147 		 * dataset.
148 		 */
149 		err = arc_free(pio, tx->tx_pool->dp_spa,
150 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
151 		ASSERT(err == 0);
152 
153 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
154 		    -used, -compressed, -uncompressed, tx);
155 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
156 		return;
157 	}
158 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
159 
160 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
161 
162 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
163 		int err;
164 		int64_t delta;
165 
166 		dprintf_bp(bp, "freeing: %s", "");
167 		err = arc_free(pio, tx->tx_pool->dp_spa,
168 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
169 		ASSERT(err == 0);
170 
171 		mutex_enter(&ds->ds_lock);
172 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
173 		    !DS_UNIQUE_IS_ACCURATE(ds));
174 		delta = parent_delta(ds, -used);
175 		ds->ds_phys->ds_unique_bytes -= used;
176 		mutex_exit(&ds->ds_lock);
177 		dsl_dir_diduse_space(ds->ds_dir,
178 		    delta, -compressed, -uncompressed, tx);
179 	} else {
180 		dprintf_bp(bp, "putting on dead list: %s", "");
181 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
182 		ASSERT3U(ds->ds_prev->ds_object, ==,
183 		    ds->ds_phys->ds_prev_snap_obj);
184 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
185 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
186 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
187 		    ds->ds_object && bp->blk_birth >
188 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
189 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
190 			mutex_enter(&ds->ds_prev->ds_lock);
191 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
192 			mutex_exit(&ds->ds_prev->ds_lock);
193 		}
194 	}
195 	mutex_enter(&ds->ds_lock);
196 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
197 	ds->ds_phys->ds_used_bytes -= used;
198 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
199 	ds->ds_phys->ds_compressed_bytes -= compressed;
200 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
201 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
202 	mutex_exit(&ds->ds_lock);
203 }
204 
205 uint64_t
206 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
207 {
208 	uint64_t trysnap = 0;
209 
210 	if (ds == NULL)
211 		return (0);
212 	/*
213 	 * The snapshot creation could fail, but that would cause an
214 	 * incorrect FALSE return, which would only result in an
215 	 * overestimation of the amount of space that an operation would
216 	 * consume, which is OK.
217 	 *
218 	 * There's also a small window where we could miss a pending
219 	 * snapshot, because we could set the sync task in the quiescing
220 	 * phase.  So this should only be used as a guess.
221 	 */
222 	if (ds->ds_trysnap_txg >
223 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
224 		trysnap = ds->ds_trysnap_txg;
225 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
226 }
227 
228 int
229 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
230 {
231 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
232 }
233 
234 /* ARGSUSED */
235 static void
236 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
237 {
238 	dsl_dataset_t *ds = dsv;
239 
240 	/* open_refcount == DS_REF_MAX when deleting */
241 	ASSERT(ds->ds_open_refcount == 0 ||
242 	    ds->ds_open_refcount == DS_REF_MAX);
243 
244 	dprintf_ds(ds, "evicting %s\n", "");
245 
246 	unique_remove(ds->ds_fsid_guid);
247 
248 	if (ds->ds_user_ptr != NULL)
249 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
250 
251 	if (ds->ds_prev) {
252 		dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
253 		ds->ds_prev = NULL;
254 	}
255 
256 	bplist_close(&ds->ds_deadlist);
257 	dsl_dir_close(ds->ds_dir, ds);
258 
259 	ASSERT(!list_link_active(&ds->ds_synced_link));
260 
261 	mutex_destroy(&ds->ds_lock);
262 	mutex_destroy(&ds->ds_opening_lock);
263 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
264 
265 	kmem_free(ds, sizeof (dsl_dataset_t));
266 }
267 
268 static int
269 dsl_dataset_get_snapname(dsl_dataset_t *ds)
270 {
271 	dsl_dataset_phys_t *headphys;
272 	int err;
273 	dmu_buf_t *headdbuf;
274 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
275 	objset_t *mos = dp->dp_meta_objset;
276 
277 	if (ds->ds_snapname[0])
278 		return (0);
279 	if (ds->ds_phys->ds_next_snap_obj == 0)
280 		return (0);
281 
282 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
283 	    FTAG, &headdbuf);
284 	if (err)
285 		return (err);
286 	headphys = headdbuf->db_data;
287 	err = zap_value_search(dp->dp_meta_objset,
288 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
289 	dmu_buf_rele(headdbuf, FTAG);
290 	return (err);
291 }
292 
293 static int
294 dsl_dataset_snap_lookup(objset_t *os, uint64_t flags,
295     uint64_t snapnames_zapobj, const char *name, uint64_t *value)
296 {
297 	matchtype_t mt;
298 	int err;
299 
300 	if (flags & DS_FLAG_CI_DATASET)
301 		mt = MT_FIRST;
302 	else
303 		mt = MT_EXACT;
304 
305 	err = zap_lookup_norm(os, snapnames_zapobj, name, 8, 1,
306 	    value, mt, NULL, 0, NULL);
307 	if (err == ENOTSUP && mt == MT_FIRST)
308 		err = zap_lookup(os, snapnames_zapobj, name, 8, 1, value);
309 	return (err);
310 }
311 
312 static int
313 dsl_dataset_snap_remove(objset_t *os, uint64_t flags,
314     uint64_t snapnames_zapobj, char *name, dmu_tx_t *tx)
315 {
316 	matchtype_t mt;
317 	int err;
318 
319 	if (flags & DS_FLAG_CI_DATASET)
320 		mt = MT_FIRST;
321 	else
322 		mt = MT_EXACT;
323 
324 	err = zap_remove_norm(os, snapnames_zapobj, name, mt, tx);
325 	if (err == ENOTSUP && mt == MT_FIRST)
326 		err = zap_remove(os, snapnames_zapobj, name, tx);
327 	return (err);
328 }
329 
330 int
331 dsl_dataset_open_obj(dsl_pool_t *dp, uint64_t dsobj, const char *snapname,
332     int mode, void *tag, dsl_dataset_t **dsp)
333 {
334 	uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
335 	objset_t *mos = dp->dp_meta_objset;
336 	dmu_buf_t *dbuf;
337 	dsl_dataset_t *ds;
338 	int err;
339 
340 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
341 	    dsl_pool_sync_context(dp));
342 
343 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
344 	if (err)
345 		return (err);
346 	ds = dmu_buf_get_user(dbuf);
347 	if (ds == NULL) {
348 		dsl_dataset_t *winner;
349 
350 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
351 		ds->ds_dbuf = dbuf;
352 		ds->ds_object = dsobj;
353 		ds->ds_phys = dbuf->db_data;
354 
355 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
356 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
357 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
358 		    NULL);
359 
360 		err = bplist_open(&ds->ds_deadlist,
361 		    mos, ds->ds_phys->ds_deadlist_obj);
362 		if (err == 0) {
363 			err = dsl_dir_open_obj(dp,
364 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
365 		}
366 		if (err) {
367 			/*
368 			 * we don't really need to close the blist if we
369 			 * just opened it.
370 			 */
371 			mutex_destroy(&ds->ds_lock);
372 			mutex_destroy(&ds->ds_opening_lock);
373 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
374 			kmem_free(ds, sizeof (dsl_dataset_t));
375 			dmu_buf_rele(dbuf, tag);
376 			return (err);
377 		}
378 
379 		if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) {
380 			ds->ds_snapname[0] = '\0';
381 			if (ds->ds_phys->ds_prev_snap_obj) {
382 				err = dsl_dataset_open_obj(dp,
383 				    ds->ds_phys->ds_prev_snap_obj, NULL,
384 				    DS_MODE_NONE, ds, &ds->ds_prev);
385 			}
386 		} else {
387 			if (snapname) {
388 #ifdef ZFS_DEBUG
389 				dsl_dataset_phys_t *headphys;
390 				dmu_buf_t *headdbuf;
391 				err = dmu_bonus_hold(mos,
392 				    ds->ds_dir->dd_phys->dd_head_dataset_obj,
393 				    FTAG, &headdbuf);
394 				if (err == 0) {
395 					uint64_t foundobj;
396 
397 					headphys = headdbuf->db_data;
398 					err = dsl_dataset_snap_lookup(
399 					    dp->dp_meta_objset,
400 					    headphys->ds_flags,
401 					    headphys->ds_snapnames_zapobj,
402 					    snapname, &foundobj);
403 					ASSERT3U(foundobj, ==, dsobj);
404 					dmu_buf_rele(headdbuf, FTAG);
405 				}
406 #endif
407 				(void) strcat(ds->ds_snapname, snapname);
408 			} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
409 				err = dsl_dataset_get_snapname(ds);
410 			}
411 		}
412 
413 		if (!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_locked(ds->ds_dir,
427 			    "refreservation", sizeof (uint64_t), 1,
428 			    &ds->ds_reserved, NULL);
429 			if (err == 0) {
430 				err = dsl_prop_get_ds_locked(ds->ds_dir,
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_close(ds->ds_prev,
449 				    DS_MODE_NONE, ds);
450 			}
451 			dsl_dir_close(ds->ds_dir, ds);
452 			mutex_destroy(&ds->ds_lock);
453 			mutex_destroy(&ds->ds_opening_lock);
454 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
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 
469 	mutex_enter(&ds->ds_lock);
470 	if ((DS_MODE_LEVEL(mode) == DS_MODE_PRIMARY &&
471 	    (ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT) &&
472 	    !DS_MODE_IS_INCONSISTENT(mode)) ||
473 	    (ds->ds_open_refcount + weight > DS_REF_MAX)) {
474 		mutex_exit(&ds->ds_lock);
475 		dsl_dataset_close(ds, DS_MODE_NONE, tag);
476 		return (EBUSY);
477 	}
478 	ds->ds_open_refcount += weight;
479 	mutex_exit(&ds->ds_lock);
480 
481 	*dsp = ds;
482 	return (0);
483 }
484 
485 int
486 dsl_dataset_open_spa(spa_t *spa, const char *name, int mode,
487     void *tag, dsl_dataset_t **dsp)
488 {
489 	dsl_dir_t *dd;
490 	dsl_pool_t *dp;
491 	const char *tail;
492 	uint64_t obj;
493 	dsl_dataset_t *ds = NULL;
494 	int err = 0;
495 
496 	err = dsl_dir_open_spa(spa, name, FTAG, &dd, &tail);
497 	if (err)
498 		return (err);
499 
500 	dp = dd->dd_pool;
501 	obj = dd->dd_phys->dd_head_dataset_obj;
502 	rw_enter(&dp->dp_config_rwlock, RW_READER);
503 	if (obj == 0) {
504 		/* A dataset with no associated objset */
505 		err = ENOENT;
506 		goto out;
507 	}
508 
509 	if (tail != NULL) {
510 		objset_t *mos = dp->dp_meta_objset;
511 		uint64_t flags;
512 
513 		err = dsl_dataset_open_obj(dp, obj, NULL,
514 		    DS_MODE_NONE, tag, &ds);
515 		if (err)
516 			goto out;
517 		flags = ds->ds_phys->ds_flags;
518 		obj = ds->ds_phys->ds_snapnames_zapobj;
519 		dsl_dataset_close(ds, DS_MODE_NONE, tag);
520 		ds = NULL;
521 
522 		if (tail[0] != '@') {
523 			err = ENOENT;
524 			goto out;
525 		}
526 		tail++;
527 
528 		/* Look for a snapshot */
529 		if (!DS_MODE_IS_READONLY(mode)) {
530 			err = EROFS;
531 			goto out;
532 		}
533 		dprintf("looking for snapshot '%s'\n", tail);
534 		err = dsl_dataset_snap_lookup(mos, flags, obj, tail, &obj);
535 		if (err)
536 			goto out;
537 	}
538 	err = dsl_dataset_open_obj(dp, obj, tail, mode, tag, &ds);
539 
540 out:
541 	rw_exit(&dp->dp_config_rwlock);
542 	dsl_dir_close(dd, FTAG);
543 
544 	ASSERT3U((err == 0), ==, (ds != NULL));
545 	/* ASSERT(ds == NULL || strcmp(name, ds->ds_name) == 0); */
546 
547 	*dsp = ds;
548 	return (err);
549 }
550 
551 int
552 dsl_dataset_open(const char *name, int mode, void *tag, dsl_dataset_t **dsp)
553 {
554 	return (dsl_dataset_open_spa(NULL, name, mode, tag, dsp));
555 }
556 
557 void
558 dsl_dataset_name(dsl_dataset_t *ds, char *name)
559 {
560 	if (ds == NULL) {
561 		(void) strcpy(name, "mos");
562 	} else {
563 		dsl_dir_name(ds->ds_dir, name);
564 		VERIFY(0 == dsl_dataset_get_snapname(ds));
565 		if (ds->ds_snapname[0]) {
566 			(void) strcat(name, "@");
567 			if (!MUTEX_HELD(&ds->ds_lock)) {
568 				/*
569 				 * We use a "recursive" mutex so that we
570 				 * can call dprintf_ds() with ds_lock held.
571 				 */
572 				mutex_enter(&ds->ds_lock);
573 				(void) strcat(name, ds->ds_snapname);
574 				mutex_exit(&ds->ds_lock);
575 			} else {
576 				(void) strcat(name, ds->ds_snapname);
577 			}
578 		}
579 	}
580 }
581 
582 static int
583 dsl_dataset_namelen(dsl_dataset_t *ds)
584 {
585 	int result;
586 
587 	if (ds == NULL) {
588 		result = 3;	/* "mos" */
589 	} else {
590 		result = dsl_dir_namelen(ds->ds_dir);
591 		VERIFY(0 == dsl_dataset_get_snapname(ds));
592 		if (ds->ds_snapname[0]) {
593 			++result;	/* adding one for the @-sign */
594 			if (!MUTEX_HELD(&ds->ds_lock)) {
595 				/* see dsl_datset_name */
596 				mutex_enter(&ds->ds_lock);
597 				result += strlen(ds->ds_snapname);
598 				mutex_exit(&ds->ds_lock);
599 			} else {
600 				result += strlen(ds->ds_snapname);
601 			}
602 		}
603 	}
604 
605 	return (result);
606 }
607 
608 void
609 dsl_dataset_close(dsl_dataset_t *ds, int mode, void *tag)
610 {
611 	uint64_t weight = ds_refcnt_weight[DS_MODE_LEVEL(mode)];
612 	mutex_enter(&ds->ds_lock);
613 	ASSERT3U(ds->ds_open_refcount, >=, weight);
614 	ds->ds_open_refcount -= weight;
615 	mutex_exit(&ds->ds_lock);
616 
617 	dmu_buf_rele(ds->ds_dbuf, tag);
618 }
619 
620 void
621 dsl_dataset_downgrade(dsl_dataset_t *ds, int oldmode, int newmode)
622 {
623 	uint64_t oldweight = ds_refcnt_weight[DS_MODE_LEVEL(oldmode)];
624 	uint64_t newweight = ds_refcnt_weight[DS_MODE_LEVEL(newmode)];
625 	mutex_enter(&ds->ds_lock);
626 	ASSERT3U(ds->ds_open_refcount, >=, oldweight);
627 	ASSERT3U(oldweight, >=, newweight);
628 	ds->ds_open_refcount -= oldweight;
629 	ds->ds_open_refcount += newweight;
630 	mutex_exit(&ds->ds_lock);
631 }
632 
633 boolean_t
634 dsl_dataset_tryupgrade(dsl_dataset_t *ds, int oldmode, int newmode)
635 {
636 	boolean_t rv;
637 	uint64_t oldweight = ds_refcnt_weight[DS_MODE_LEVEL(oldmode)];
638 	uint64_t newweight = ds_refcnt_weight[DS_MODE_LEVEL(newmode)];
639 	mutex_enter(&ds->ds_lock);
640 	ASSERT3U(ds->ds_open_refcount, >=, oldweight);
641 	ASSERT3U(newweight, >=, oldweight);
642 	if (ds->ds_open_refcount - oldweight + newweight > DS_REF_MAX) {
643 		rv = B_FALSE;
644 	} else {
645 		ds->ds_open_refcount -= oldweight;
646 		ds->ds_open_refcount += newweight;
647 		rv = B_TRUE;
648 	}
649 	mutex_exit(&ds->ds_lock);
650 	return (rv);
651 }
652 
653 void
654 dsl_dataset_create_root(dsl_pool_t *dp, uint64_t *ddobjp, dmu_tx_t *tx)
655 {
656 	objset_t *mos = dp->dp_meta_objset;
657 	dmu_buf_t *dbuf;
658 	dsl_dataset_phys_t *dsphys;
659 	dsl_dataset_t *ds;
660 	uint64_t dsobj;
661 	dsl_dir_t *dd;
662 
663 	dsl_dir_create_root(mos, ddobjp, tx);
664 	VERIFY(0 == dsl_dir_open_obj(dp, *ddobjp, NULL, FTAG, &dd));
665 
666 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
667 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
668 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
669 	dmu_buf_will_dirty(dbuf, tx);
670 	dsphys = dbuf->db_data;
671 	dsphys->ds_dir_obj = dd->dd_object;
672 	dsphys->ds_fsid_guid = unique_create();
673 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
674 	    sizeof (dsphys->ds_guid));
675 	dsphys->ds_snapnames_zapobj =
676 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
677 	    DMU_OT_NONE, 0, tx);
678 	dsphys->ds_creation_time = gethrestime_sec();
679 	dsphys->ds_creation_txg = tx->tx_txg;
680 	dsphys->ds_deadlist_obj =
681 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
682 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
683 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
684 	dmu_buf_rele(dbuf, FTAG);
685 
686 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
687 	dd->dd_phys->dd_head_dataset_obj = dsobj;
688 	dsl_dir_close(dd, FTAG);
689 
690 	VERIFY(0 ==
691 	    dsl_dataset_open_obj(dp, dsobj, NULL, DS_MODE_NONE, FTAG, &ds));
692 	(void) dmu_objset_create_impl(dp->dp_spa, ds,
693 	    &ds->ds_phys->ds_bp, DMU_OST_ZFS, tx);
694 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
695 }
696 
697 uint64_t
698 dsl_dataset_create_sync_impl(dsl_dir_t *dd, dsl_dataset_t *origin,
699     uint64_t flags, dmu_tx_t *tx)
700 {
701 	dsl_pool_t *dp = dd->dd_pool;
702 	dmu_buf_t *dbuf;
703 	dsl_dataset_phys_t *dsphys;
704 	uint64_t dsobj;
705 	objset_t *mos = dp->dp_meta_objset;
706 
707 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
708 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
709 	ASSERT(dmu_tx_is_syncing(tx));
710 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
711 
712 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
713 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
714 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
715 	dmu_buf_will_dirty(dbuf, tx);
716 	dsphys = dbuf->db_data;
717 	dsphys->ds_dir_obj = dd->dd_object;
718 	dsphys->ds_flags = flags;
719 	dsphys->ds_fsid_guid = unique_create();
720 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
721 	    sizeof (dsphys->ds_guid));
722 	dsphys->ds_snapnames_zapobj =
723 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
724 	    DMU_OT_NONE, 0, tx);
725 	dsphys->ds_creation_time = gethrestime_sec();
726 	dsphys->ds_creation_txg = tx->tx_txg;
727 	dsphys->ds_deadlist_obj =
728 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
729 
730 	if (origin) {
731 		dsphys->ds_prev_snap_obj = origin->ds_object;
732 		dsphys->ds_prev_snap_txg =
733 		    origin->ds_phys->ds_creation_txg;
734 		dsphys->ds_used_bytes =
735 		    origin->ds_phys->ds_used_bytes;
736 		dsphys->ds_compressed_bytes =
737 		    origin->ds_phys->ds_compressed_bytes;
738 		dsphys->ds_uncompressed_bytes =
739 		    origin->ds_phys->ds_uncompressed_bytes;
740 		dsphys->ds_bp = origin->ds_phys->ds_bp;
741 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
742 
743 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
744 		origin->ds_phys->ds_num_children++;
745 
746 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
747 		dd->dd_phys->dd_origin_obj = origin->ds_object;
748 	}
749 
750 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
751 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
752 
753 	dmu_buf_rele(dbuf, FTAG);
754 
755 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
756 	dd->dd_phys->dd_head_dataset_obj = dsobj;
757 
758 	return (dsobj);
759 }
760 
761 uint64_t
762 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
763     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
764 {
765 	dsl_pool_t *dp = pdd->dd_pool;
766 	uint64_t dsobj, ddobj;
767 	dsl_dir_t *dd;
768 
769 	ASSERT(lastname[0] != '@');
770 
771 	ddobj = dsl_dir_create_sync(pdd, lastname, tx);
772 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
773 
774 	dsobj = dsl_dataset_create_sync_impl(dd, origin, flags, tx);
775 
776 	dsl_deleg_set_create_perms(dd, tx, cr);
777 
778 	dsl_dir_close(dd, FTAG);
779 
780 	return (dsobj);
781 }
782 
783 struct destroyarg {
784 	dsl_sync_task_group_t *dstg;
785 	char *snapname;
786 	char *failed;
787 };
788 
789 static int
790 dsl_snapshot_destroy_one(char *name, void *arg)
791 {
792 	struct destroyarg *da = arg;
793 	dsl_dataset_t *ds;
794 	char *cp;
795 	int err;
796 
797 	(void) strcat(name, "@");
798 	(void) strcat(name, da->snapname);
799 	err = dsl_dataset_open(name,
800 	    DS_MODE_EXCLUSIVE | DS_MODE_READONLY | DS_MODE_INCONSISTENT,
801 	    da->dstg, &ds);
802 	cp = strchr(name, '@');
803 	*cp = '\0';
804 	if (err == ENOENT)
805 		return (0);
806 	if (err) {
807 		(void) strcpy(da->failed, name);
808 		return (err);
809 	}
810 
811 	dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
812 	    dsl_dataset_destroy_sync, ds, da->dstg, 0);
813 	return (0);
814 }
815 
816 /*
817  * Destroy 'snapname' in all descendants of 'fsname'.
818  */
819 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
820 int
821 dsl_snapshots_destroy(char *fsname, char *snapname)
822 {
823 	int err;
824 	struct destroyarg da;
825 	dsl_sync_task_t *dst;
826 	spa_t *spa;
827 
828 	err = spa_open(fsname, &spa, FTAG);
829 	if (err)
830 		return (err);
831 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
832 	da.snapname = snapname;
833 	da.failed = fsname;
834 
835 	err = dmu_objset_find(fsname,
836 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
837 
838 	if (err == 0)
839 		err = dsl_sync_task_group_wait(da.dstg);
840 
841 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
842 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
843 		dsl_dataset_t *ds = dst->dst_arg1;
844 		if (dst->dst_err) {
845 			dsl_dataset_name(ds, fsname);
846 			*strchr(fsname, '@') = '\0';
847 		}
848 		/*
849 		 * If it was successful, destroy_sync would have
850 		 * closed the ds
851 		 */
852 		if (err)
853 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, da.dstg);
854 	}
855 
856 	dsl_sync_task_group_destroy(da.dstg);
857 	spa_close(spa, FTAG);
858 	return (err);
859 }
860 
861 /*
862  * ds must be opened EXCLUSIVE or PRIMARY.  on return (whether
863  * successful or not), ds will be closed and caller can no longer
864  * dereference it.
865  */
866 int
867 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
868 {
869 	int err;
870 	dsl_sync_task_group_t *dstg;
871 	objset_t *os;
872 	dsl_dir_t *dd;
873 	uint64_t obj;
874 
875 	if (ds->ds_open_refcount != DS_REF_MAX) {
876 		if (dsl_dataset_tryupgrade(ds, DS_MODE_PRIMARY,
877 		    DS_MODE_EXCLUSIVE) == 0) {
878 			dsl_dataset_close(ds, DS_MODE_PRIMARY, tag);
879 			return (EBUSY);
880 		}
881 	}
882 
883 	if (dsl_dataset_is_snapshot(ds)) {
884 		/* Destroying a snapshot is simpler */
885 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
886 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
887 		    ds, tag, 0);
888 		goto out;
889 	}
890 
891 	dd = ds->ds_dir;
892 
893 	/*
894 	 * Check for errors and mark this ds as inconsistent, in
895 	 * case we crash while freeing the objects.
896 	 */
897 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
898 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
899 	if (err)
900 		goto out;
901 
902 	err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
903 	if (err)
904 		goto out;
905 
906 	/*
907 	 * remove the objects in open context, so that we won't
908 	 * have too much to do in syncing context.
909 	 */
910 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
911 	    ds->ds_phys->ds_prev_snap_txg)) {
912 		dmu_tx_t *tx = dmu_tx_create(os);
913 		dmu_tx_hold_free(tx, obj, 0, DMU_OBJECT_END);
914 		dmu_tx_hold_bonus(tx, obj);
915 		err = dmu_tx_assign(tx, TXG_WAIT);
916 		if (err) {
917 			/*
918 			 * Perhaps there is not enough disk
919 			 * space.  Just deal with it from
920 			 * dsl_dataset_destroy_sync().
921 			 */
922 			dmu_tx_abort(tx);
923 			continue;
924 		}
925 		VERIFY(0 == dmu_object_free(os, obj, tx));
926 		dmu_tx_commit(tx);
927 	}
928 	/* Make sure it's not dirty before we finish destroying it. */
929 	txg_wait_synced(dd->dd_pool, 0);
930 
931 	dmu_objset_close(os);
932 	if (err != ESRCH)
933 		goto out;
934 
935 	if (ds->ds_user_ptr) {
936 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
937 		ds->ds_user_ptr = NULL;
938 	}
939 
940 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
941 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
942 	rw_exit(&dd->dd_pool->dp_config_rwlock);
943 
944 	if (err)
945 		goto out;
946 
947 	/*
948 	 * Blow away the dsl_dir + head dataset.
949 	 */
950 	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
951 	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
952 	    dsl_dataset_destroy_sync, ds, tag, 0);
953 	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
954 	    dsl_dir_destroy_sync, dd, FTAG, 0);
955 	err = dsl_sync_task_group_wait(dstg);
956 	dsl_sync_task_group_destroy(dstg);
957 	/* if it is successful, *destroy_sync will close the ds+dd */
958 	if (err)
959 		dsl_dir_close(dd, FTAG);
960 out:
961 	if (err)
962 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
963 	return (err);
964 }
965 
966 int
967 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
968 {
969 	ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
970 
971 	return (dsl_sync_task_do(ds->ds_dir->dd_pool,
972 	    dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
973 	    ds, &ost, 0));
974 }
975 
976 void *
977 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
978     void *p, dsl_dataset_evict_func_t func)
979 {
980 	void *old;
981 
982 	mutex_enter(&ds->ds_lock);
983 	old = ds->ds_user_ptr;
984 	if (old == NULL) {
985 		ds->ds_user_ptr = p;
986 		ds->ds_user_evict_func = func;
987 	}
988 	mutex_exit(&ds->ds_lock);
989 	return (old);
990 }
991 
992 void *
993 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
994 {
995 	return (ds->ds_user_ptr);
996 }
997 
998 
999 blkptr_t *
1000 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1001 {
1002 	return (&ds->ds_phys->ds_bp);
1003 }
1004 
1005 void
1006 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1007 {
1008 	ASSERT(dmu_tx_is_syncing(tx));
1009 	/* If it's the meta-objset, set dp_meta_rootbp */
1010 	if (ds == NULL) {
1011 		tx->tx_pool->dp_meta_rootbp = *bp;
1012 	} else {
1013 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1014 		ds->ds_phys->ds_bp = *bp;
1015 	}
1016 }
1017 
1018 spa_t *
1019 dsl_dataset_get_spa(dsl_dataset_t *ds)
1020 {
1021 	return (ds->ds_dir->dd_pool->dp_spa);
1022 }
1023 
1024 void
1025 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1026 {
1027 	dsl_pool_t *dp;
1028 
1029 	if (ds == NULL) /* this is the meta-objset */
1030 		return;
1031 
1032 	ASSERT(ds->ds_user_ptr != NULL);
1033 
1034 	if (ds->ds_phys->ds_next_snap_obj != 0)
1035 		panic("dirtying snapshot!");
1036 
1037 	dp = ds->ds_dir->dd_pool;
1038 
1039 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1040 		/* up the hold count until we can be written out */
1041 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1042 	}
1043 }
1044 
1045 /*
1046  * The unique space in the head dataset can be calculated by subtracting
1047  * the space used in the most recent snapshot, that is still being used
1048  * in this file system, from the space currently in use.  To figure out
1049  * the space in the most recent snapshot still in use, we need to take
1050  * the total space used in the snapshot and subtract out the space that
1051  * has been freed up since the snapshot was taken.
1052  */
1053 static void
1054 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1055 {
1056 	uint64_t mrs_used;
1057 	uint64_t dlused, dlcomp, dluncomp;
1058 
1059 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1060 
1061 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1062 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1063 	else
1064 		mrs_used = 0;
1065 
1066 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1067 	    &dluncomp));
1068 
1069 	ASSERT3U(dlused, <=, mrs_used);
1070 	ds->ds_phys->ds_unique_bytes =
1071 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1072 
1073 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1074 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1075 	    SPA_VERSION_UNIQUE_ACCURATE)
1076 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1077 }
1078 
1079 static uint64_t
1080 dsl_dataset_unique(dsl_dataset_t *ds)
1081 {
1082 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1083 		dsl_dataset_recalc_head_uniq(ds);
1084 
1085 	return (ds->ds_phys->ds_unique_bytes);
1086 }
1087 
1088 struct killarg {
1089 	int64_t *usedp;
1090 	int64_t *compressedp;
1091 	int64_t *uncompressedp;
1092 	zio_t *zio;
1093 	dmu_tx_t *tx;
1094 };
1095 
1096 static int
1097 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1098 {
1099 	struct killarg *ka = arg;
1100 	blkptr_t *bp = &bc->bc_blkptr;
1101 
1102 	ASSERT3U(bc->bc_errno, ==, 0);
1103 
1104 	/*
1105 	 * Since this callback is not called concurrently, no lock is
1106 	 * needed on the accounting values.
1107 	 */
1108 	*ka->usedp += bp_get_dasize(spa, bp);
1109 	*ka->compressedp += BP_GET_PSIZE(bp);
1110 	*ka->uncompressedp += BP_GET_UCSIZE(bp);
1111 	/* XXX check for EIO? */
1112 	(void) arc_free(ka->zio, spa, ka->tx->tx_txg, bp, NULL, NULL,
1113 	    ARC_NOWAIT);
1114 	return (0);
1115 }
1116 
1117 /* ARGSUSED */
1118 static int
1119 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1120 {
1121 	dsl_dataset_t *ds = arg1;
1122 	dmu_objset_type_t *ost = arg2;
1123 
1124 	/*
1125 	 * We can only roll back to emptyness if it is a ZPL objset.
1126 	 */
1127 	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1128 		return (EINVAL);
1129 
1130 	/*
1131 	 * This must not be a snapshot.
1132 	 */
1133 	if (ds->ds_phys->ds_next_snap_obj != 0)
1134 		return (EINVAL);
1135 
1136 	/*
1137 	 * If we made changes this txg, traverse_dsl_dataset won't find
1138 	 * them.  Try again.
1139 	 */
1140 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1141 		return (EAGAIN);
1142 
1143 	return (0);
1144 }
1145 
1146 /* ARGSUSED */
1147 static void
1148 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1149 {
1150 	dsl_dataset_t *ds = arg1;
1151 	dmu_objset_type_t *ost = arg2;
1152 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1153 
1154 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1155 
1156 	/*
1157 	 * Before the roll back destroy the zil.
1158 	 */
1159 	if (ds->ds_user_ptr != NULL) {
1160 		zil_rollback_destroy(
1161 		    ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx);
1162 
1163 		/*
1164 		 * We need to make sure that the objset_impl_t is reopened after
1165 		 * we do the rollback, otherwise it will have the wrong
1166 		 * objset_phys_t.  Normally this would happen when this
1167 		 * DS_MODE_EXCLUSIVE dataset-open is closed, thus causing the
1168 		 * dataset to be immediately evicted.  But when doing "zfs recv
1169 		 * -F", we reopen the objset before that, so that there is no
1170 		 * window where the dataset is closed and inconsistent.
1171 		 */
1172 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1173 		ds->ds_user_ptr = NULL;
1174 	}
1175 
1176 	/* Zero out the deadlist. */
1177 	bplist_close(&ds->ds_deadlist);
1178 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1179 	ds->ds_phys->ds_deadlist_obj =
1180 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1181 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1182 	    ds->ds_phys->ds_deadlist_obj));
1183 
1184 	{
1185 		/* Free blkptrs that we gave birth to */
1186 		zio_t *zio;
1187 		int64_t used = 0, compressed = 0, uncompressed = 0;
1188 		struct killarg ka;
1189 		int64_t delta;
1190 
1191 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1192 		    ZIO_FLAG_MUSTSUCCEED);
1193 		ka.usedp = &used;
1194 		ka.compressedp = &compressed;
1195 		ka.uncompressedp = &uncompressed;
1196 		ka.zio = zio;
1197 		ka.tx = tx;
1198 		(void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1199 		    ADVANCE_POST, kill_blkptr, &ka);
1200 		(void) zio_wait(zio);
1201 
1202 		/* only deduct space beyond any refreservation */
1203 		delta = parent_delta(ds, -used);
1204 		dsl_dir_diduse_space(ds->ds_dir,
1205 		    delta, -compressed, -uncompressed, tx);
1206 	}
1207 
1208 	if (ds->ds_prev) {
1209 		/* Change our contents to that of the prev snapshot */
1210 		ASSERT3U(ds->ds_prev->ds_object, ==,
1211 		    ds->ds_phys->ds_prev_snap_obj);
1212 		ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1213 		ds->ds_phys->ds_used_bytes =
1214 		    ds->ds_prev->ds_phys->ds_used_bytes;
1215 		ds->ds_phys->ds_compressed_bytes =
1216 		    ds->ds_prev->ds_phys->ds_compressed_bytes;
1217 		ds->ds_phys->ds_uncompressed_bytes =
1218 		    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1219 		ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1220 		ds->ds_phys->ds_unique_bytes = 0;
1221 
1222 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1223 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1224 			ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1225 		}
1226 	} else {
1227 		/* Zero out our contents, recreate objset */
1228 		bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1229 		ds->ds_phys->ds_used_bytes = 0;
1230 		ds->ds_phys->ds_compressed_bytes = 0;
1231 		ds->ds_phys->ds_uncompressed_bytes = 0;
1232 		ds->ds_phys->ds_flags = 0;
1233 		ds->ds_phys->ds_unique_bytes = 0;
1234 		(void) dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1235 		    &ds->ds_phys->ds_bp, *ost, tx);
1236 	}
1237 
1238 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1239 	    tx, cr, "dataset = %llu", ds->ds_object);
1240 }
1241 
1242 /* ARGSUSED */
1243 static int
1244 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1245 {
1246 	dsl_dataset_t *ds = arg1;
1247 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1248 	uint64_t count;
1249 	int err;
1250 
1251 	/*
1252 	 * Can't delete a head dataset if there are snapshots of it.
1253 	 * (Except if the only snapshots are from the branch we cloned
1254 	 * from.)
1255 	 */
1256 	if (ds->ds_prev != NULL &&
1257 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1258 		return (EINVAL);
1259 
1260 	/*
1261 	 * This is really a dsl_dir thing, but check it here so that
1262 	 * we'll be less likely to leave this dataset inconsistent &
1263 	 * nearly destroyed.
1264 	 */
1265 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1266 	if (err)
1267 		return (err);
1268 	if (count != 0)
1269 		return (EEXIST);
1270 
1271 	return (0);
1272 }
1273 
1274 /* ARGSUSED */
1275 static void
1276 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1277 {
1278 	dsl_dataset_t *ds = arg1;
1279 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1280 
1281 	/* Mark it as inconsistent on-disk, in case we crash */
1282 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1283 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1284 
1285 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1286 	    cr, "dataset = %llu", ds->ds_object);
1287 }
1288 
1289 /* ARGSUSED */
1290 int
1291 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1292 {
1293 	dsl_dataset_t *ds = arg1;
1294 
1295 	/* Can't delete a branch point. */
1296 	if (ds->ds_phys->ds_num_children > 1)
1297 		return (EEXIST);
1298 
1299 	/*
1300 	 * Can't delete a head dataset if there are snapshots of it.
1301 	 * (Except if the only snapshots are from the branch we cloned
1302 	 * from.)
1303 	 */
1304 	if (ds->ds_prev != NULL &&
1305 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1306 		return (EINVAL);
1307 
1308 	/*
1309 	 * If we made changes this txg, traverse_dsl_dataset won't find
1310 	 * them.  Try again.
1311 	 */
1312 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1313 		return (EAGAIN);
1314 
1315 	/* XXX we should do some i/o error checking... */
1316 	return (0);
1317 }
1318 
1319 void
1320 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1321 {
1322 	dsl_dataset_t *ds = arg1;
1323 	int64_t used = 0, compressed = 0, uncompressed = 0;
1324 	zio_t *zio;
1325 	int err;
1326 	int after_branch_point = FALSE;
1327 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1328 	objset_t *mos = dp->dp_meta_objset;
1329 	dsl_dataset_t *ds_prev = NULL;
1330 	uint64_t obj;
1331 
1332 	ASSERT3U(ds->ds_open_refcount, ==, DS_REF_MAX);
1333 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1334 	ASSERT(ds->ds_prev == NULL ||
1335 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1336 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1337 
1338 	/* Remove our reservation */
1339 	if (ds->ds_reserved != 0) {
1340 		uint64_t val = 0;
1341 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1342 		ASSERT3U(ds->ds_reserved, ==, 0);
1343 	}
1344 
1345 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1346 
1347 	obj = ds->ds_object;
1348 
1349 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1350 		if (ds->ds_prev) {
1351 			ds_prev = ds->ds_prev;
1352 		} else {
1353 			VERIFY(0 == dsl_dataset_open_obj(dp,
1354 			    ds->ds_phys->ds_prev_snap_obj, NULL,
1355 			    DS_MODE_NONE, FTAG, &ds_prev));
1356 		}
1357 		after_branch_point =
1358 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1359 
1360 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1361 		if (after_branch_point &&
1362 		    ds->ds_phys->ds_next_snap_obj == 0) {
1363 			/* This clone is toast. */
1364 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1365 			ds_prev->ds_phys->ds_num_children--;
1366 		} else if (!after_branch_point) {
1367 			ds_prev->ds_phys->ds_next_snap_obj =
1368 			    ds->ds_phys->ds_next_snap_obj;
1369 		}
1370 	}
1371 
1372 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1373 
1374 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1375 		blkptr_t bp;
1376 		dsl_dataset_t *ds_next;
1377 		uint64_t itor = 0;
1378 		uint64_t old_unique;
1379 
1380 		spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1381 
1382 		VERIFY(0 == dsl_dataset_open_obj(dp,
1383 		    ds->ds_phys->ds_next_snap_obj, NULL,
1384 		    DS_MODE_NONE, FTAG, &ds_next));
1385 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1386 
1387 		old_unique = dsl_dataset_unique(ds_next);
1388 
1389 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1390 		ds_next->ds_phys->ds_prev_snap_obj =
1391 		    ds->ds_phys->ds_prev_snap_obj;
1392 		ds_next->ds_phys->ds_prev_snap_txg =
1393 		    ds->ds_phys->ds_prev_snap_txg;
1394 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1395 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1396 
1397 		/*
1398 		 * Transfer to our deadlist (which will become next's
1399 		 * new deadlist) any entries from next's current
1400 		 * deadlist which were born before prev, and free the
1401 		 * other entries.
1402 		 *
1403 		 * XXX we're doing this long task with the config lock held
1404 		 */
1405 		while (bplist_iterate(&ds_next->ds_deadlist, &itor,
1406 		    &bp) == 0) {
1407 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1408 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1409 				    &bp, tx));
1410 				if (ds_prev && !after_branch_point &&
1411 				    bp.blk_birth >
1412 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1413 					ds_prev->ds_phys->ds_unique_bytes +=
1414 					    bp_get_dasize(dp->dp_spa, &bp);
1415 				}
1416 			} else {
1417 				used += bp_get_dasize(dp->dp_spa, &bp);
1418 				compressed += BP_GET_PSIZE(&bp);
1419 				uncompressed += BP_GET_UCSIZE(&bp);
1420 				/* XXX check return value? */
1421 				(void) arc_free(zio, dp->dp_spa, tx->tx_txg,
1422 				    &bp, NULL, NULL, ARC_NOWAIT);
1423 			}
1424 		}
1425 
1426 		/* free next's deadlist */
1427 		bplist_close(&ds_next->ds_deadlist);
1428 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1429 
1430 		/* set next's deadlist to our deadlist */
1431 		ds_next->ds_phys->ds_deadlist_obj =
1432 		    ds->ds_phys->ds_deadlist_obj;
1433 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1434 		    ds_next->ds_phys->ds_deadlist_obj));
1435 		ds->ds_phys->ds_deadlist_obj = 0;
1436 
1437 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1438 			/*
1439 			 * Update next's unique to include blocks which
1440 			 * were previously shared by only this snapshot
1441 			 * and it.  Those blocks will be born after the
1442 			 * prev snap and before this snap, and will have
1443 			 * died after the next snap and before the one
1444 			 * after that (ie. be on the snap after next's
1445 			 * deadlist).
1446 			 *
1447 			 * XXX we're doing this long task with the
1448 			 * config lock held
1449 			 */
1450 			dsl_dataset_t *ds_after_next;
1451 
1452 			VERIFY(0 == dsl_dataset_open_obj(dp,
1453 			    ds_next->ds_phys->ds_next_snap_obj, NULL,
1454 			    DS_MODE_NONE, FTAG, &ds_after_next));
1455 			itor = 0;
1456 			while (bplist_iterate(&ds_after_next->ds_deadlist,
1457 			    &itor, &bp) == 0) {
1458 				if (bp.blk_birth >
1459 				    ds->ds_phys->ds_prev_snap_txg &&
1460 				    bp.blk_birth <=
1461 				    ds->ds_phys->ds_creation_txg) {
1462 					ds_next->ds_phys->ds_unique_bytes +=
1463 					    bp_get_dasize(dp->dp_spa, &bp);
1464 				}
1465 			}
1466 
1467 			dsl_dataset_close(ds_after_next, DS_MODE_NONE, FTAG);
1468 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1469 		} else {
1470 			ASSERT3P(ds_next->ds_prev, ==, ds);
1471 			dsl_dataset_close(ds_next->ds_prev, DS_MODE_NONE,
1472 			    ds_next);
1473 			if (ds_prev) {
1474 				VERIFY(0 == dsl_dataset_open_obj(dp,
1475 				    ds->ds_phys->ds_prev_snap_obj, NULL,
1476 				    DS_MODE_NONE, ds_next, &ds_next->ds_prev));
1477 			} else {
1478 				ds_next->ds_prev = NULL;
1479 			}
1480 
1481 			dsl_dataset_recalc_head_uniq(ds_next);
1482 
1483 			/*
1484 			 * Reduce the amount of our unconsmed refreservation
1485 			 * being charged to our parent by the amount of
1486 			 * new unique data we have gained.
1487 			 */
1488 			if (old_unique < ds_next->ds_reserved) {
1489 				int64_t mrsdelta;
1490 				uint64_t new_unique =
1491 				    ds_next->ds_phys->ds_unique_bytes;
1492 
1493 				ASSERT(old_unique <= new_unique);
1494 				mrsdelta = MIN(new_unique - old_unique,
1495 				    ds_next->ds_reserved - old_unique);
1496 				dsl_dir_diduse_space(ds->ds_dir, -mrsdelta,
1497 				    0, 0, tx);
1498 			}
1499 		}
1500 		dsl_dataset_close(ds_next, DS_MODE_NONE, FTAG);
1501 
1502 		/*
1503 		 * NB: unique_bytes might not be accurate for the head objset.
1504 		 * Before SPA_VERSION 9, we didn't update its value when we
1505 		 * deleted the most recent snapshot.
1506 		 */
1507 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1508 	} else {
1509 		/*
1510 		 * There's no next snapshot, so this is a head dataset.
1511 		 * Destroy the deadlist.  Unless it's a clone, the
1512 		 * deadlist should be empty.  (If it's a clone, it's
1513 		 * safe to ignore the deadlist contents.)
1514 		 */
1515 		struct killarg ka;
1516 
1517 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1518 		bplist_close(&ds->ds_deadlist);
1519 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1520 		ds->ds_phys->ds_deadlist_obj = 0;
1521 
1522 		/*
1523 		 * Free everything that we point to (that's born after
1524 		 * the previous snapshot, if we are a clone)
1525 		 *
1526 		 * XXX we're doing this long task with the config lock held
1527 		 */
1528 		ka.usedp = &used;
1529 		ka.compressedp = &compressed;
1530 		ka.uncompressedp = &uncompressed;
1531 		ka.zio = zio;
1532 		ka.tx = tx;
1533 		err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1534 		    ADVANCE_POST, kill_blkptr, &ka);
1535 		ASSERT3U(err, ==, 0);
1536 		ASSERT(spa_version(dp->dp_spa) <
1537 		    SPA_VERSION_UNIQUE_ACCURATE ||
1538 		    used == ds->ds_phys->ds_unique_bytes);
1539 	}
1540 
1541 	err = zio_wait(zio);
1542 	ASSERT3U(err, ==, 0);
1543 
1544 	dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx);
1545 
1546 	if (ds->ds_phys->ds_snapnames_zapobj) {
1547 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1548 		ASSERT(err == 0);
1549 	}
1550 
1551 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1552 		/* Erase the link in the dataset */
1553 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1554 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1555 		/*
1556 		 * dsl_dir_sync_destroy() called us, they'll destroy
1557 		 * the dataset.
1558 		 */
1559 	} else {
1560 		/* remove from snapshot namespace */
1561 		dsl_dataset_t *ds_head;
1562 		VERIFY(0 == dsl_dataset_open_obj(dp,
1563 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, NULL,
1564 		    DS_MODE_NONE, FTAG, &ds_head));
1565 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1566 #ifdef ZFS_DEBUG
1567 		{
1568 			uint64_t val;
1569 
1570 			err = dsl_dataset_snap_lookup(mos,
1571 			    ds_head->ds_phys->ds_flags,
1572 			    ds_head->ds_phys->ds_snapnames_zapobj,
1573 			    ds->ds_snapname, &val);
1574 			ASSERT3U(err, ==, 0);
1575 			ASSERT3U(val, ==, obj);
1576 		}
1577 #endif
1578 		err = dsl_dataset_snap_remove(mos,
1579 		    ds_head->ds_phys->ds_flags,
1580 		    ds_head->ds_phys->ds_snapnames_zapobj,
1581 		    ds->ds_snapname, tx);
1582 		ASSERT(err == 0);
1583 		dsl_dataset_close(ds_head, DS_MODE_NONE, FTAG);
1584 	}
1585 
1586 	if (ds_prev && ds->ds_prev != ds_prev)
1587 		dsl_dataset_close(ds_prev, DS_MODE_NONE, FTAG);
1588 
1589 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1590 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1591 	    cr, "dataset = %llu", ds->ds_object);
1592 
1593 	dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, tag);
1594 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1595 
1596 }
1597 
1598 static int
1599 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1600 {
1601 	uint64_t asize;
1602 
1603 	if (!dmu_tx_is_syncing(tx))
1604 		return (0);
1605 
1606 	/*
1607 	 * If there's an fs-only reservation, any blocks that might become
1608 	 * owned by the snapshot dataset must be accommodated by space
1609 	 * outside of the reservation.
1610 	 */
1611 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1612 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1613 		return (ENOSPC);
1614 
1615 	/*
1616 	 * Propogate any reserved space for this snapshot to other
1617 	 * snapshot checks in this sync group.
1618 	 */
1619 	if (asize > 0)
1620 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1621 
1622 	return (0);
1623 }
1624 
1625 /* ARGSUSED */
1626 int
1627 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1628 {
1629 	dsl_dataset_t *ds = arg1;
1630 	const char *snapname = arg2;
1631 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1632 	int err;
1633 	uint64_t value;
1634 
1635 	/*
1636 	 * We don't allow multiple snapshots of the same txg.  If there
1637 	 * is already one, try again.
1638 	 */
1639 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1640 		return (EAGAIN);
1641 
1642 	/*
1643 	 * Check for conflicting name snapshot name.
1644 	 */
1645 	err = dsl_dataset_snap_lookup(mos, ds->ds_phys->ds_flags,
1646 	    ds->ds_phys->ds_snapnames_zapobj, snapname, &value);
1647 	if (err == 0)
1648 		return (EEXIST);
1649 	if (err != ENOENT)
1650 		return (err);
1651 
1652 	/*
1653 	 * Check that the dataset's name is not too long.  Name consists
1654 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1655 	 */
1656 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1657 		return (ENAMETOOLONG);
1658 
1659 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1660 	if (err)
1661 		return (err);
1662 
1663 	ds->ds_trysnap_txg = tx->tx_txg;
1664 	return (0);
1665 }
1666 
1667 void
1668 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1669 {
1670 	dsl_dataset_t *ds = arg1;
1671 	const char *snapname = arg2;
1672 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1673 	dmu_buf_t *dbuf;
1674 	dsl_dataset_phys_t *dsphys;
1675 	uint64_t dsobj;
1676 	objset_t *mos = dp->dp_meta_objset;
1677 	int err;
1678 
1679 	spa_scrub_restart(dp->dp_spa, tx->tx_txg);
1680 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1681 
1682 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1683 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1684 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1685 	dmu_buf_will_dirty(dbuf, tx);
1686 	dsphys = dbuf->db_data;
1687 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1688 	dsphys->ds_fsid_guid = unique_create();
1689 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1690 	    sizeof (dsphys->ds_guid));
1691 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1692 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1693 	dsphys->ds_next_snap_obj = ds->ds_object;
1694 	dsphys->ds_num_children = 1;
1695 	dsphys->ds_creation_time = gethrestime_sec();
1696 	dsphys->ds_creation_txg = tx->tx_txg;
1697 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1698 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1699 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1700 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1701 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1702 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1703 	dmu_buf_rele(dbuf, FTAG);
1704 
1705 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1706 	if (ds->ds_prev) {
1707 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1708 		    ds->ds_object ||
1709 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1710 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1711 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1712 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1713 			    ds->ds_prev->ds_phys->ds_creation_txg);
1714 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1715 		}
1716 	}
1717 
1718 	/*
1719 	 * If we have a reference-reservation on this dataset, we will
1720 	 * need to increase the amount of refreservation being charged
1721 	 * since our unique space is going to zero.
1722 	 */
1723 	if (ds->ds_reserved) {
1724 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1725 		dsl_dir_diduse_space(ds->ds_dir, add, 0, 0, tx);
1726 	}
1727 
1728 	bplist_close(&ds->ds_deadlist);
1729 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1730 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1731 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1732 	ds->ds_phys->ds_prev_snap_txg = tx->tx_txg;
1733 	ds->ds_phys->ds_unique_bytes = 0;
1734 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1735 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1736 	ds->ds_phys->ds_deadlist_obj =
1737 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1738 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1739 	    ds->ds_phys->ds_deadlist_obj));
1740 
1741 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1742 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1743 	    snapname, 8, 1, &dsobj, tx);
1744 	ASSERT(err == 0);
1745 
1746 	if (ds->ds_prev)
1747 		dsl_dataset_close(ds->ds_prev, DS_MODE_NONE, ds);
1748 	VERIFY(0 == dsl_dataset_open_obj(dp,
1749 	    ds->ds_phys->ds_prev_snap_obj, snapname,
1750 	    DS_MODE_NONE, ds, &ds->ds_prev));
1751 
1752 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1753 	    "dataset = %llu", dsobj);
1754 }
1755 
1756 void
1757 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1758 {
1759 	ASSERT(dmu_tx_is_syncing(tx));
1760 	ASSERT(ds->ds_user_ptr != NULL);
1761 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1762 
1763 	/*
1764 	 * in case we had to change ds_fsid_guid when we opened it,
1765 	 * sync it out now.
1766 	 */
1767 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1768 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1769 
1770 	dsl_dir_dirty(ds->ds_dir, tx);
1771 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1772 }
1773 
1774 void
1775 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1776 {
1777 	uint64_t refd, avail, uobjs, aobjs;
1778 
1779 	dsl_dir_stats(ds->ds_dir, nv);
1780 
1781 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1782 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1783 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1784 
1785 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1786 	    ds->ds_phys->ds_creation_time);
1787 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1788 	    ds->ds_phys->ds_creation_txg);
1789 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1790 	    ds->ds_quota);
1791 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1792 	    ds->ds_reserved);
1793 
1794 	if (ds->ds_phys->ds_next_snap_obj) {
1795 		/*
1796 		 * This is a snapshot; override the dd's space used with
1797 		 * our unique space and compression ratio.
1798 		 */
1799 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1800 		    ds->ds_phys->ds_unique_bytes);
1801 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1802 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1803 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1804 		    ds->ds_phys->ds_compressed_bytes));
1805 	}
1806 }
1807 
1808 void
1809 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1810 {
1811 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1812 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1813 	stat->dds_guid = ds->ds_phys->ds_guid;
1814 	if (ds->ds_phys->ds_next_snap_obj) {
1815 		stat->dds_is_snapshot = B_TRUE;
1816 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1817 	}
1818 
1819 	/* clone origin is really a dsl_dir thing... */
1820 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1821 	if (ds->ds_dir->dd_phys->dd_origin_obj) {
1822 		dsl_dataset_t *ods;
1823 
1824 		VERIFY(0 == dsl_dataset_open_obj(ds->ds_dir->dd_pool,
1825 		    ds->ds_dir->dd_phys->dd_origin_obj,
1826 		    NULL, DS_MODE_NONE, FTAG, &ods));
1827 		dsl_dataset_name(ods, stat->dds_origin);
1828 		dsl_dataset_close(ods, DS_MODE_NONE, FTAG);
1829 	}
1830 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1831 }
1832 
1833 uint64_t
1834 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1835 {
1836 	return (ds->ds_fsid_guid);
1837 }
1838 
1839 void
1840 dsl_dataset_space(dsl_dataset_t *ds,
1841     uint64_t *refdbytesp, uint64_t *availbytesp,
1842     uint64_t *usedobjsp, uint64_t *availobjsp)
1843 {
1844 	*refdbytesp = ds->ds_phys->ds_used_bytes;
1845 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1846 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1847 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1848 	if (ds->ds_quota != 0) {
1849 		/*
1850 		 * Adjust available bytes according to refquota
1851 		 */
1852 		if (*refdbytesp < ds->ds_quota)
1853 			*availbytesp = MIN(*availbytesp,
1854 			    ds->ds_quota - *refdbytesp);
1855 		else
1856 			*availbytesp = 0;
1857 	}
1858 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1859 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1860 }
1861 
1862 boolean_t
1863 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1864 {
1865 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1866 
1867 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
1868 	    dsl_pool_sync_context(dp));
1869 	if (ds->ds_prev == NULL)
1870 		return (B_FALSE);
1871 	if (ds->ds_phys->ds_bp.blk_birth >
1872 	    ds->ds_prev->ds_phys->ds_creation_txg)
1873 		return (B_TRUE);
1874 	return (B_FALSE);
1875 }
1876 
1877 /* ARGSUSED */
1878 static int
1879 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1880 {
1881 	dsl_dataset_t *ds = arg1;
1882 	char *newsnapname = arg2;
1883 	dsl_dir_t *dd = ds->ds_dir;
1884 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1885 	dsl_dataset_t *hds;
1886 	uint64_t val;
1887 	int err;
1888 
1889 	err = dsl_dataset_open_obj(dd->dd_pool,
1890 	    dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds);
1891 	if (err)
1892 		return (err);
1893 
1894 	/* new name better not be in use */
1895 	err = dsl_dataset_snap_lookup(mos, hds->ds_phys->ds_flags,
1896 	    hds->ds_phys->ds_snapnames_zapobj, newsnapname, &val);
1897 	dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1898 
1899 	if (err == 0)
1900 		err = EEXIST;
1901 	else if (err == ENOENT)
1902 		err = 0;
1903 
1904 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1905 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
1906 		err = ENAMETOOLONG;
1907 
1908 	return (err);
1909 }
1910 
1911 static void
1912 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
1913     cred_t *cr, dmu_tx_t *tx)
1914 {
1915 	dsl_dataset_t *ds = arg1;
1916 	const char *newsnapname = arg2;
1917 	dsl_dir_t *dd = ds->ds_dir;
1918 	objset_t *mos = dd->dd_pool->dp_meta_objset;
1919 	dsl_dataset_t *hds;
1920 	int err;
1921 
1922 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
1923 
1924 	VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool,
1925 	    dd->dd_phys->dd_head_dataset_obj, NULL, DS_MODE_NONE, FTAG, &hds));
1926 
1927 	VERIFY(0 == dsl_dataset_get_snapname(ds));
1928 	err = dsl_dataset_snap_remove(mos, hds->ds_phys->ds_flags,
1929 	    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, tx);
1930 	ASSERT3U(err, ==, 0);
1931 	mutex_enter(&ds->ds_lock);
1932 	(void) strcpy(ds->ds_snapname, newsnapname);
1933 	mutex_exit(&ds->ds_lock);
1934 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
1935 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
1936 	ASSERT3U(err, ==, 0);
1937 
1938 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
1939 	    cr, "dataset = %llu", ds->ds_object);
1940 	dsl_dataset_close(hds, DS_MODE_NONE, FTAG);
1941 }
1942 
1943 struct renamesnaparg {
1944 	dsl_sync_task_group_t *dstg;
1945 	char failed[MAXPATHLEN];
1946 	char *oldsnap;
1947 	char *newsnap;
1948 };
1949 
1950 static int
1951 dsl_snapshot_rename_one(char *name, void *arg)
1952 {
1953 	struct renamesnaparg *ra = arg;
1954 	dsl_dataset_t *ds = NULL;
1955 	char *cp;
1956 	int err;
1957 
1958 	cp = name + strlen(name);
1959 	*cp = '@';
1960 	(void) strcpy(cp + 1, ra->oldsnap);
1961 
1962 	/*
1963 	 * For recursive snapshot renames the parent won't be changing
1964 	 * so we just pass name for both the to/from argument.
1965 	 */
1966 	if (err = zfs_secpolicy_rename_perms(name, name, CRED())) {
1967 		(void) strcpy(ra->failed, name);
1968 		return (err);
1969 	}
1970 
1971 	err = dsl_dataset_open(name, DS_MODE_READONLY | DS_MODE_STANDARD,
1972 	    ra->dstg, &ds);
1973 	if (err == ENOENT) {
1974 		*cp = '\0';
1975 		return (0);
1976 	}
1977 	if (err) {
1978 		(void) strcpy(ra->failed, name);
1979 		*cp = '\0';
1980 		dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
1981 		return (err);
1982 	}
1983 
1984 #ifdef _KERNEL
1985 	/* for all filesystems undergoing rename, we'll need to unmount it */
1986 	(void) zfs_unmount_snap(name, NULL);
1987 #endif
1988 
1989 	*cp = '\0';
1990 
1991 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
1992 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
1993 
1994 	return (0);
1995 }
1996 
1997 static int
1998 dsl_recursive_rename(char *oldname, const char *newname)
1999 {
2000 	int err;
2001 	struct renamesnaparg *ra;
2002 	dsl_sync_task_t *dst;
2003 	spa_t *spa;
2004 	char *cp, *fsname = spa_strdup(oldname);
2005 	int len = strlen(oldname);
2006 
2007 	/* truncate the snapshot name to get the fsname */
2008 	cp = strchr(fsname, '@');
2009 	*cp = '\0';
2010 
2011 	err = spa_open(fsname, &spa, FTAG);
2012 	if (err) {
2013 		kmem_free(fsname, len + 1);
2014 		return (err);
2015 	}
2016 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2017 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2018 
2019 	ra->oldsnap = strchr(oldname, '@') + 1;
2020 	ra->newsnap = strchr(newname, '@') + 1;
2021 	*ra->failed = '\0';
2022 
2023 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2024 	    DS_FIND_CHILDREN);
2025 	kmem_free(fsname, len + 1);
2026 
2027 	if (err == 0) {
2028 		err = dsl_sync_task_group_wait(ra->dstg);
2029 	}
2030 
2031 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2032 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2033 		dsl_dataset_t *ds = dst->dst_arg1;
2034 		if (dst->dst_err) {
2035 			dsl_dir_name(ds->ds_dir, ra->failed);
2036 			(void) strcat(ra->failed, "@");
2037 			(void) strcat(ra->failed, ra->newsnap);
2038 		}
2039 		dsl_dataset_close(ds, DS_MODE_STANDARD, ra->dstg);
2040 	}
2041 
2042 	if (err)
2043 		(void) strcpy(oldname, ra->failed);
2044 
2045 	dsl_sync_task_group_destroy(ra->dstg);
2046 	kmem_free(ra, sizeof (struct renamesnaparg));
2047 	spa_close(spa, FTAG);
2048 	return (err);
2049 }
2050 
2051 static int
2052 dsl_valid_rename(char *oldname, void *arg)
2053 {
2054 	int delta = *(int *)arg;
2055 
2056 	if (strlen(oldname) + delta >= MAXNAMELEN)
2057 		return (ENAMETOOLONG);
2058 
2059 	return (0);
2060 }
2061 
2062 #pragma weak dmu_objset_rename = dsl_dataset_rename
2063 int
2064 dsl_dataset_rename(char *oldname, const char *newname,
2065     boolean_t recursive)
2066 {
2067 	dsl_dir_t *dd;
2068 	dsl_dataset_t *ds;
2069 	const char *tail;
2070 	int err;
2071 
2072 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2073 	if (err)
2074 		return (err);
2075 	if (tail == NULL) {
2076 		int delta = strlen(newname) - strlen(oldname);
2077 
2078 		/* if we're growing, validate child size lengths */
2079 		if (delta > 0)
2080 			err = dmu_objset_find(oldname, dsl_valid_rename,
2081 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2082 
2083 		if (!err)
2084 			err = dsl_dir_rename(dd, newname);
2085 		dsl_dir_close(dd, FTAG);
2086 		return (err);
2087 	}
2088 	if (tail[0] != '@') {
2089 		/* the name ended in a nonexistant component */
2090 		dsl_dir_close(dd, FTAG);
2091 		return (ENOENT);
2092 	}
2093 
2094 	dsl_dir_close(dd, FTAG);
2095 
2096 	/* new name must be snapshot in same filesystem */
2097 	tail = strchr(newname, '@');
2098 	if (tail == NULL)
2099 		return (EINVAL);
2100 	tail++;
2101 	if (strncmp(oldname, newname, tail - newname) != 0)
2102 		return (EXDEV);
2103 
2104 	if (recursive) {
2105 		err = dsl_recursive_rename(oldname, newname);
2106 	} else {
2107 		err = dsl_dataset_open(oldname,
2108 		    DS_MODE_READONLY | DS_MODE_STANDARD, FTAG, &ds);
2109 		if (err)
2110 			return (err);
2111 
2112 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2113 		    dsl_dataset_snapshot_rename_check,
2114 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2115 
2116 		dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2117 	}
2118 
2119 	return (err);
2120 }
2121 
2122 struct promotearg {
2123 	uint64_t used, comp, uncomp, unique;
2124 	uint64_t ds_flags, newnext_obj, snapnames_obj;
2125 };
2126 
2127 /* ARGSUSED */
2128 static int
2129 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2130 {
2131 	dsl_dataset_t *hds = arg1;
2132 	struct promotearg *pa = arg2;
2133 	dsl_dir_t *dd = hds->ds_dir;
2134 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2135 	dsl_dir_t *odd = NULL;
2136 	dsl_dataset_t *ds = NULL;
2137 	dsl_dataset_t *origin_ds = NULL;
2138 	dsl_dataset_t *newnext_ds = NULL;
2139 	int err;
2140 	char *name = NULL;
2141 	uint64_t itor = 0;
2142 	blkptr_t bp;
2143 
2144 	bzero(pa, sizeof (*pa));
2145 
2146 	/* Check that it is a clone */
2147 	if (dd->dd_phys->dd_origin_obj == 0)
2148 		return (EINVAL);
2149 
2150 	/* Since this is so expensive, don't do the preliminary check */
2151 	if (!dmu_tx_is_syncing(tx))
2152 		return (0);
2153 
2154 	if (err = dsl_dataset_open_obj(dp, dd->dd_phys->dd_origin_obj,
2155 	    NULL, DS_MODE_EXCLUSIVE, FTAG, &origin_ds))
2156 		goto out;
2157 	odd = origin_ds->ds_dir;
2158 
2159 	{
2160 		dsl_dataset_t *phds;
2161 		if (err = dsl_dataset_open_obj(dd->dd_pool,
2162 		    odd->dd_phys->dd_head_dataset_obj,
2163 		    NULL, DS_MODE_NONE, FTAG, &phds))
2164 			goto out;
2165 		pa->ds_flags = phds->ds_phys->ds_flags;
2166 		pa->snapnames_obj = phds->ds_phys->ds_snapnames_zapobj;
2167 		dsl_dataset_close(phds, DS_MODE_NONE, FTAG);
2168 	}
2169 
2170 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
2171 		err = EXDEV;
2172 		goto out;
2173 	}
2174 
2175 	/* find origin's new next ds */
2176 	VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, hds->ds_object,
2177 	    NULL, DS_MODE_NONE, FTAG, &newnext_ds));
2178 	while (newnext_ds->ds_phys->ds_prev_snap_obj != origin_ds->ds_object) {
2179 		dsl_dataset_t *prev;
2180 
2181 		if (err = dsl_dataset_open_obj(dd->dd_pool,
2182 		    newnext_ds->ds_phys->ds_prev_snap_obj,
2183 		    NULL, DS_MODE_NONE, FTAG, &prev))
2184 			goto out;
2185 		dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
2186 		newnext_ds = prev;
2187 	}
2188 	pa->newnext_obj = newnext_ds->ds_object;
2189 
2190 	/* compute origin's new unique space */
2191 	while ((err = bplist_iterate(&newnext_ds->ds_deadlist,
2192 	    &itor, &bp)) == 0) {
2193 		if (bp.blk_birth > origin_ds->ds_phys->ds_prev_snap_txg)
2194 			pa->unique += bp_get_dasize(dd->dd_pool->dp_spa, &bp);
2195 	}
2196 	if (err != ENOENT)
2197 		goto out;
2198 
2199 	/* Walk the snapshots that we are moving */
2200 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2201 	ds = origin_ds;
2202 	/* CONSTCOND */
2203 	while (TRUE) {
2204 		uint64_t val, dlused, dlcomp, dluncomp;
2205 		dsl_dataset_t *prev;
2206 
2207 		/* Check that the snapshot name does not conflict */
2208 		dsl_dataset_name(ds, name);
2209 		err = dsl_dataset_snap_lookup(dd->dd_pool->dp_meta_objset,
2210 		    hds->ds_phys->ds_flags, hds->ds_phys->ds_snapnames_zapobj,
2211 		    ds->ds_snapname, &val);
2212 		if (err != ENOENT) {
2213 			if (err == 0)
2214 				err = EEXIST;
2215 			goto out;
2216 		}
2217 
2218 		/*
2219 		 * compute space to transfer.  Each snapshot gave birth to:
2220 		 * (my used) - (prev's used) + (deadlist's used)
2221 		 */
2222 		pa->used += ds->ds_phys->ds_used_bytes;
2223 		pa->comp += ds->ds_phys->ds_compressed_bytes;
2224 		pa->uncomp += ds->ds_phys->ds_uncompressed_bytes;
2225 
2226 		/* If we reach the first snapshot, we're done. */
2227 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2228 			break;
2229 
2230 		if (err = bplist_space(&ds->ds_deadlist,
2231 		    &dlused, &dlcomp, &dluncomp))
2232 			goto out;
2233 		if (err = dsl_dataset_open_obj(dd->dd_pool,
2234 		    ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
2235 		    FTAG, &prev))
2236 			goto out;
2237 		pa->used += dlused - prev->ds_phys->ds_used_bytes;
2238 		pa->comp += dlcomp - prev->ds_phys->ds_compressed_bytes;
2239 		pa->uncomp += dluncomp - prev->ds_phys->ds_uncompressed_bytes;
2240 
2241 		/*
2242 		 * We could be a clone of a clone.  If we reach our
2243 		 * parent's branch point, we're done.
2244 		 */
2245 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
2246 			dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
2247 			break;
2248 		}
2249 		if (ds != origin_ds)
2250 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2251 		ds = prev;
2252 	}
2253 
2254 	/* Check that there is enough space here */
2255 	err = dsl_dir_transfer_possible(odd, dd, pa->used);
2256 
2257 out:
2258 	if (ds && ds != origin_ds)
2259 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2260 	if (origin_ds)
2261 		dsl_dataset_close(origin_ds, DS_MODE_EXCLUSIVE, FTAG);
2262 	if (newnext_ds)
2263 		dsl_dataset_close(newnext_ds, DS_MODE_NONE, FTAG);
2264 	if (name)
2265 		kmem_free(name, MAXPATHLEN);
2266 	return (err);
2267 }
2268 
2269 static void
2270 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2271 {
2272 	dsl_dataset_t *hds = arg1;
2273 	struct promotearg *pa = arg2;
2274 	dsl_dir_t *dd = hds->ds_dir;
2275 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2276 	dsl_dir_t *odd = NULL;
2277 	dsl_dataset_t *ds, *origin_ds;
2278 	char *name;
2279 
2280 	ASSERT(dd->dd_phys->dd_origin_obj != 0);
2281 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2282 
2283 	VERIFY(0 == dsl_dataset_open_obj(dp, dd->dd_phys->dd_origin_obj,
2284 	    NULL, DS_MODE_EXCLUSIVE, FTAG, &origin_ds));
2285 	/*
2286 	 * We need to explicitly open odd, since origin_ds's dd will be
2287 	 * changing.
2288 	 */
2289 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2290 	    NULL, FTAG, &odd));
2291 
2292 	/* move snapshots to this dir */
2293 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2294 	ds = origin_ds;
2295 	/* CONSTCOND */
2296 	while (TRUE) {
2297 		dsl_dataset_t *prev;
2298 
2299 		/* move snap name entry */
2300 		dsl_dataset_name(ds, name);
2301 		VERIFY(0 == dsl_dataset_snap_remove(dp->dp_meta_objset,
2302 		    pa->ds_flags, pa->snapnames_obj, ds->ds_snapname, tx));
2303 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2304 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2305 		    8, 1, &ds->ds_object, tx));
2306 
2307 		/* change containing dsl_dir */
2308 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2309 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2310 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2311 		ASSERT3P(ds->ds_dir, ==, odd);
2312 		dsl_dir_close(ds->ds_dir, ds);
2313 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2314 		    NULL, ds, &ds->ds_dir));
2315 
2316 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2317 
2318 		if (ds->ds_phys->ds_prev_snap_obj == 0)
2319 			break;
2320 
2321 		VERIFY(0 == dsl_dataset_open_obj(dp,
2322 		    ds->ds_phys->ds_prev_snap_obj, NULL, DS_MODE_EXCLUSIVE,
2323 		    FTAG, &prev));
2324 
2325 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
2326 			dsl_dataset_close(prev, DS_MODE_EXCLUSIVE, FTAG);
2327 			break;
2328 		}
2329 		if (ds != origin_ds)
2330 			dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2331 		ds = prev;
2332 	}
2333 	if (ds != origin_ds)
2334 		dsl_dataset_close(ds, DS_MODE_EXCLUSIVE, FTAG);
2335 
2336 	/* change origin's next snap */
2337 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2338 	origin_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj;
2339 
2340 	/* change origin */
2341 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2342 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2343 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2344 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2345 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2346 
2347 	/* change space accounting */
2348 	dsl_dir_diduse_space(odd, -pa->used, -pa->comp, -pa->uncomp, tx);
2349 	dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx);
2350 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2351 
2352 	/* log history record */
2353 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2354 	    cr, "dataset = %llu", ds->ds_object);
2355 
2356 	dsl_dir_close(odd, FTAG);
2357 	dsl_dataset_close(origin_ds, DS_MODE_EXCLUSIVE, FTAG);
2358 	kmem_free(name, MAXPATHLEN);
2359 }
2360 
2361 int
2362 dsl_dataset_promote(const char *name)
2363 {
2364 	dsl_dataset_t *ds;
2365 	int err;
2366 	dmu_object_info_t doi;
2367 	struct promotearg pa;
2368 
2369 	err = dsl_dataset_open(name, DS_MODE_NONE, FTAG, &ds);
2370 	if (err)
2371 		return (err);
2372 
2373 	err = dmu_object_info(ds->ds_dir->dd_pool->dp_meta_objset,
2374 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2375 	if (err) {
2376 		dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2377 		return (err);
2378 	}
2379 
2380 	/*
2381 	 * Add in 128x the snapnames zapobj size, since we will be moving
2382 	 * a bunch of snapnames to the promoted ds, and dirtying their
2383 	 * bonus buffers.
2384 	 */
2385 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2386 	    dsl_dataset_promote_check,
2387 	    dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks);
2388 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2389 	return (err);
2390 }
2391 
2392 struct cloneswaparg {
2393 	dsl_dataset_t *cds; /* clone dataset */
2394 	dsl_dataset_t *ohds; /* origin's head dataset */
2395 	boolean_t force;
2396 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2397 };
2398 
2399 /* ARGSUSED */
2400 static int
2401 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2402 {
2403 	struct cloneswaparg *csa = arg1;
2404 
2405 	/* they should both be heads */
2406 	if (dsl_dataset_is_snapshot(csa->cds) ||
2407 	    dsl_dataset_is_snapshot(csa->ohds))
2408 		return (EINVAL);
2409 
2410 	/* the branch point should be just before them */
2411 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2412 		return (EINVAL);
2413 
2414 	/* cds should be the clone */
2415 	if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2416 	    csa->ohds->ds_object)
2417 		return (EINVAL);
2418 
2419 	/* the clone should be a child of the origin */
2420 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2421 		return (EINVAL);
2422 
2423 	/* ohds shouldn't be modified unless 'force' */
2424 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2425 		return (ETXTBSY);
2426 
2427 	/* adjust amount of any unconsumed refreservation */
2428 	csa->unused_refres_delta =
2429 	    (int64_t)MIN(csa->ohds->ds_reserved,
2430 	    csa->ohds->ds_phys->ds_unique_bytes) -
2431 	    (int64_t)MIN(csa->ohds->ds_reserved,
2432 	    csa->cds->ds_phys->ds_unique_bytes);
2433 
2434 	if (csa->unused_refres_delta > 0 &&
2435 	    csa->unused_refres_delta >
2436 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2437 		return (ENOSPC);
2438 
2439 	return (0);
2440 }
2441 
2442 /* ARGSUSED */
2443 static void
2444 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2445 {
2446 	struct cloneswaparg *csa = arg1;
2447 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2448 	uint64_t itor = 0;
2449 	blkptr_t bp;
2450 	uint64_t unique = 0;
2451 	int err;
2452 
2453 	ASSERT(csa->cds->ds_reserved == 0);
2454 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2455 
2456 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2457 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2458 	dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2459 
2460 	if (csa->cds->ds_user_ptr != NULL) {
2461 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2462 		csa->cds->ds_user_ptr = NULL;
2463 	}
2464 
2465 	if (csa->ohds->ds_user_ptr != NULL) {
2466 		csa->ohds->ds_user_evict_func(csa->ohds,
2467 		    csa->ohds->ds_user_ptr);
2468 		csa->ohds->ds_user_ptr = NULL;
2469 	}
2470 
2471 	/* compute unique space */
2472 	while ((err = bplist_iterate(&csa->cds->ds_deadlist,
2473 	    &itor, &bp)) == 0) {
2474 		if (bp.blk_birth > csa->cds->ds_prev->ds_phys->ds_prev_snap_txg)
2475 			unique += bp_get_dasize(dp->dp_spa, &bp);
2476 	}
2477 	VERIFY(err == ENOENT);
2478 
2479 	/* reset origin's unique bytes */
2480 	csa->cds->ds_prev->ds_phys->ds_unique_bytes = unique;
2481 
2482 	/* swap blkptrs */
2483 	{
2484 		blkptr_t tmp;
2485 		tmp = csa->ohds->ds_phys->ds_bp;
2486 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2487 		csa->cds->ds_phys->ds_bp = tmp;
2488 	}
2489 
2490 	/* set dd_*_bytes */
2491 	{
2492 		int64_t dused, dcomp, duncomp;
2493 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2494 		uint64_t odl_used, odl_comp, odl_uncomp;
2495 
2496 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2497 		    &cdl_comp, &cdl_uncomp));
2498 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2499 		    &odl_comp, &odl_uncomp));
2500 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2501 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2502 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2503 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2504 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2505 		    cdl_uncomp -
2506 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2507 
2508 		dsl_dir_diduse_space(csa->ohds->ds_dir,
2509 		    dused, dcomp, duncomp, tx);
2510 		dsl_dir_diduse_space(csa->cds->ds_dir,
2511 		    -dused, -dcomp, -duncomp, tx);
2512 	}
2513 
2514 #define	SWITCH64(x, y) \
2515 	{ \
2516 		uint64_t __tmp = (x); \
2517 		(x) = (y); \
2518 		(y) = __tmp; \
2519 	}
2520 
2521 	/* swap ds_*_bytes */
2522 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2523 	    csa->cds->ds_phys->ds_used_bytes);
2524 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2525 	    csa->cds->ds_phys->ds_compressed_bytes);
2526 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2527 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2528 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2529 	    csa->cds->ds_phys->ds_unique_bytes);
2530 
2531 	/* apply any parent delta for change in unconsumed refreservation */
2532 	dsl_dir_diduse_space(csa->ohds->ds_dir, csa->unused_refres_delta,
2533 	    0, 0, tx);
2534 
2535 	/* swap deadlists */
2536 	bplist_close(&csa->cds->ds_deadlist);
2537 	bplist_close(&csa->ohds->ds_deadlist);
2538 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2539 	    csa->cds->ds_phys->ds_deadlist_obj);
2540 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2541 	    csa->cds->ds_phys->ds_deadlist_obj));
2542 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2543 	    csa->ohds->ds_phys->ds_deadlist_obj));
2544 }
2545 
2546 /*
2547  * Swap 'clone' with its origin head file system.
2548  */
2549 int
2550 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2551     boolean_t force)
2552 {
2553 	struct cloneswaparg csa;
2554 
2555 	ASSERT(clone->ds_open_refcount == DS_REF_MAX);
2556 	ASSERT(origin_head->ds_open_refcount == DS_REF_MAX);
2557 
2558 	csa.cds = clone;
2559 	csa.ohds = origin_head;
2560 	csa.force = force;
2561 	return (dsl_sync_task_do(clone->ds_dir->dd_pool,
2562 	    dsl_dataset_clone_swap_check,
2563 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9));
2564 }
2565 
2566 /*
2567  * Given a pool name and a dataset object number in that pool,
2568  * return the name of that dataset.
2569  */
2570 int
2571 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2572 {
2573 	spa_t *spa;
2574 	dsl_pool_t *dp;
2575 	dsl_dataset_t *ds = NULL;
2576 	int error;
2577 
2578 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2579 		return (error);
2580 	dp = spa_get_dsl(spa);
2581 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2582 	if ((error = dsl_dataset_open_obj(dp, obj,
2583 	    NULL, DS_MODE_NONE, FTAG, &ds)) != 0) {
2584 		rw_exit(&dp->dp_config_rwlock);
2585 		spa_close(spa, FTAG);
2586 		return (error);
2587 	}
2588 	dsl_dataset_name(ds, buf);
2589 	dsl_dataset_close(ds, DS_MODE_NONE, FTAG);
2590 	rw_exit(&dp->dp_config_rwlock);
2591 	spa_close(spa, FTAG);
2592 
2593 	return (0);
2594 }
2595 
2596 int
2597 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2598     uint64_t asize, uint64_t inflight, uint64_t *used,
2599     uint64_t *ref_rsrv)
2600 {
2601 	int error = 0;
2602 
2603 	ASSERT3S(asize, >, 0);
2604 
2605 	/*
2606 	 * *ref_rsrv is the portion of asize that will come from any
2607 	 * unconsumed refreservation space.
2608 	 */
2609 	*ref_rsrv = 0;
2610 
2611 	mutex_enter(&ds->ds_lock);
2612 	/*
2613 	 * Make a space adjustment for reserved bytes.
2614 	 */
2615 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2616 		ASSERT3U(*used, >=,
2617 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2618 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2619 		*ref_rsrv =
2620 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2621 	}
2622 
2623 	if (!check_quota || ds->ds_quota == 0) {
2624 		mutex_exit(&ds->ds_lock);
2625 		return (0);
2626 	}
2627 	/*
2628 	 * If they are requesting more space, and our current estimate
2629 	 * is over quota, they get to try again unless the actual
2630 	 * on-disk is over quota and there are no pending changes (which
2631 	 * may free up space for us).
2632 	 */
2633 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2634 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2635 			error = ERESTART;
2636 		else
2637 			error = EDQUOT;
2638 	}
2639 	mutex_exit(&ds->ds_lock);
2640 
2641 	return (error);
2642 }
2643 
2644 /* ARGSUSED */
2645 static int
2646 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2647 {
2648 	dsl_dataset_t *ds = arg1;
2649 	uint64_t *quotap = arg2;
2650 	uint64_t new_quota = *quotap;
2651 
2652 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2653 		return (ENOTSUP);
2654 
2655 	if (new_quota == 0)
2656 		return (0);
2657 
2658 	if (new_quota < ds->ds_phys->ds_used_bytes ||
2659 	    new_quota < ds->ds_reserved)
2660 		return (ENOSPC);
2661 
2662 	return (0);
2663 }
2664 
2665 /* ARGSUSED */
2666 void
2667 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2668 {
2669 	dsl_dataset_t *ds = arg1;
2670 	uint64_t *quotap = arg2;
2671 	uint64_t new_quota = *quotap;
2672 
2673 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2674 
2675 	mutex_enter(&ds->ds_lock);
2676 	ds->ds_quota = new_quota;
2677 	mutex_exit(&ds->ds_lock);
2678 
2679 	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
2680 
2681 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
2682 	    tx, cr, "%lld dataset = %llu ",
2683 	    (longlong_t)new_quota, ds->ds_dir->dd_phys->dd_head_dataset_obj);
2684 }
2685 
2686 int
2687 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
2688 {
2689 	dsl_dataset_t *ds;
2690 	int err;
2691 
2692 	err = dsl_dataset_open(dsname, DS_MODE_STANDARD, FTAG, &ds);
2693 	if (err)
2694 		return (err);
2695 
2696 	if (quota != ds->ds_quota) {
2697 		/*
2698 		 * If someone removes a file, then tries to set the quota, we
2699 		 * want to make sure the file freeing takes effect.
2700 		 */
2701 		txg_wait_open(ds->ds_dir->dd_pool, 0);
2702 
2703 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2704 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
2705 		    ds, &quota, 0);
2706 	}
2707 	dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2708 	return (err);
2709 }
2710 
2711 static int
2712 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
2713 {
2714 	dsl_dataset_t *ds = arg1;
2715 	uint64_t *reservationp = arg2;
2716 	uint64_t new_reservation = *reservationp;
2717 	int64_t delta;
2718 	uint64_t unique;
2719 
2720 	if (new_reservation > INT64_MAX)
2721 		return (EOVERFLOW);
2722 
2723 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
2724 	    SPA_VERSION_REFRESERVATION)
2725 		return (ENOTSUP);
2726 
2727 	if (dsl_dataset_is_snapshot(ds))
2728 		return (EINVAL);
2729 
2730 	/*
2731 	 * If we are doing the preliminary check in open context, the
2732 	 * space estimates may be inaccurate.
2733 	 */
2734 	if (!dmu_tx_is_syncing(tx))
2735 		return (0);
2736 
2737 	mutex_enter(&ds->ds_lock);
2738 	unique = dsl_dataset_unique(ds);
2739 	delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved);
2740 	mutex_exit(&ds->ds_lock);
2741 
2742 	if (delta > 0 &&
2743 	    delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2744 		return (ENOSPC);
2745 	if (delta > 0 && ds->ds_quota > 0 &&
2746 	    new_reservation > ds->ds_quota)
2747 		return (ENOSPC);
2748 
2749 	return (0);
2750 }
2751 
2752 /* ARGSUSED */
2753 static void
2754 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
2755     dmu_tx_t *tx)
2756 {
2757 	dsl_dataset_t *ds = arg1;
2758 	uint64_t *reservationp = arg2;
2759 	uint64_t new_reservation = *reservationp;
2760 	uint64_t unique;
2761 	int64_t delta;
2762 
2763 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2764 
2765 	mutex_enter(&ds->ds_lock);
2766 	unique = dsl_dataset_unique(ds);
2767 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
2768 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2769 	ds->ds_reserved = new_reservation;
2770 	mutex_exit(&ds->ds_lock);
2771 
2772 	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
2773 	    new_reservation, cr, tx);
2774 
2775 	dsl_dir_diduse_space(ds->ds_dir, delta, 0, 0, tx);
2776 
2777 	spa_history_internal_log(LOG_DS_REFRESERV,
2778 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
2779 	    (longlong_t)new_reservation,
2780 	    ds->ds_dir->dd_phys->dd_head_dataset_obj);
2781 }
2782 
2783 int
2784 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
2785 {
2786 	dsl_dataset_t *ds;
2787 	int err;
2788 
2789 	err = dsl_dataset_open(dsname, DS_MODE_STANDARD, FTAG, &ds);
2790 	if (err)
2791 		return (err);
2792 
2793 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2794 	    dsl_dataset_set_reservation_check,
2795 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
2796 	dsl_dataset_close(ds, DS_MODE_STANDARD, FTAG);
2797 	return (err);
2798 }
2799