xref: /titanic_51/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 2c20fda2feee4850b7ec4297e4120bc72c2bac09)
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/zfs_znode.h>
43 #include <sys/sunddi.h>
44 
45 static char *dsl_reaper = "the grim reaper";
46 
47 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
48 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
49 static dsl_checkfunc_t dsl_dataset_rollback_check;
50 static dsl_syncfunc_t dsl_dataset_rollback_sync;
51 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
52 
53 #define	DS_REF_MAX	(1ULL << 62)
54 
55 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
56 
57 #define	DSL_DATASET_IS_DESTROYED(ds)	((ds)->ds_owner == dsl_reaper)
58 
59 
60 /*
61  * Figure out how much of this delta should be propogated to the dsl_dir
62  * layer.  If there's a refreservation, that space has already been
63  * partially accounted for in our ancestors.
64  */
65 static int64_t
66 parent_delta(dsl_dataset_t *ds, int64_t delta)
67 {
68 	uint64_t old_bytes, new_bytes;
69 
70 	if (ds->ds_reserved == 0)
71 		return (delta);
72 
73 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
74 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
75 
76 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
77 	return (new_bytes - old_bytes);
78 }
79 
80 void
81 dsl_dataset_block_born(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
82 {
83 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
84 	int compressed = BP_GET_PSIZE(bp);
85 	int uncompressed = BP_GET_UCSIZE(bp);
86 	int64_t delta;
87 
88 	dprintf_bp(bp, "born, ds=%p\n", ds);
89 
90 	ASSERT(dmu_tx_is_syncing(tx));
91 	/* It could have been compressed away to nothing */
92 	if (BP_IS_HOLE(bp))
93 		return;
94 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
95 	ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
96 	if (ds == NULL) {
97 		/*
98 		 * Account for the meta-objset space in its placeholder
99 		 * dsl_dir.
100 		 */
101 		ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
102 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
103 		    used, compressed, uncompressed, tx);
104 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
105 		return;
106 	}
107 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
108 	mutex_enter(&ds->ds_lock);
109 	delta = parent_delta(ds, used);
110 	ds->ds_phys->ds_used_bytes += used;
111 	ds->ds_phys->ds_compressed_bytes += compressed;
112 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
113 	ds->ds_phys->ds_unique_bytes += used;
114 	mutex_exit(&ds->ds_lock);
115 	dsl_dir_diduse_space(ds->ds_dir, delta, compressed, uncompressed, tx);
116 }
117 
118 int
119 dsl_dataset_block_kill(dsl_dataset_t *ds, blkptr_t *bp, zio_t *pio,
120     dmu_tx_t *tx)
121 {
122 	int used = bp_get_dasize(tx->tx_pool->dp_spa, bp);
123 	int compressed = BP_GET_PSIZE(bp);
124 	int uncompressed = BP_GET_UCSIZE(bp);
125 
126 	ASSERT(dmu_tx_is_syncing(tx));
127 	/* No block pointer => nothing to free */
128 	if (BP_IS_HOLE(bp))
129 		return (0);
130 
131 	ASSERT(used > 0);
132 	if (ds == NULL) {
133 		int err;
134 		/*
135 		 * Account for the meta-objset space in its placeholder
136 		 * dataset.
137 		 */
138 		err = dsl_free(pio, tx->tx_pool,
139 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
140 		ASSERT(err == 0);
141 
142 		dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir,
143 		    -used, -compressed, -uncompressed, tx);
144 		dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
145 		return (used);
146 	}
147 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
148 
149 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
150 
151 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
152 		int err;
153 		int64_t delta;
154 
155 		dprintf_bp(bp, "freeing: %s", "");
156 		err = dsl_free(pio, tx->tx_pool,
157 		    tx->tx_txg, bp, NULL, NULL, pio ? ARC_NOWAIT: ARC_WAIT);
158 		ASSERT(err == 0);
159 
160 		mutex_enter(&ds->ds_lock);
161 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
162 		    !DS_UNIQUE_IS_ACCURATE(ds));
163 		delta = parent_delta(ds, -used);
164 		ds->ds_phys->ds_unique_bytes -= used;
165 		mutex_exit(&ds->ds_lock);
166 		dsl_dir_diduse_space(ds->ds_dir,
167 		    delta, -compressed, -uncompressed, tx);
168 	} else {
169 		dprintf_bp(bp, "putting on dead list: %s", "");
170 		VERIFY(0 == bplist_enqueue(&ds->ds_deadlist, bp, tx));
171 		ASSERT3U(ds->ds_prev->ds_object, ==,
172 		    ds->ds_phys->ds_prev_snap_obj);
173 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
174 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
175 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
176 		    ds->ds_object && bp->blk_birth >
177 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
178 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
179 			mutex_enter(&ds->ds_prev->ds_lock);
180 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
181 			mutex_exit(&ds->ds_prev->ds_lock);
182 		}
183 	}
184 	mutex_enter(&ds->ds_lock);
185 	ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
186 	ds->ds_phys->ds_used_bytes -= used;
187 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
188 	ds->ds_phys->ds_compressed_bytes -= compressed;
189 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
190 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
191 	mutex_exit(&ds->ds_lock);
192 
193 	return (used);
194 }
195 
196 uint64_t
197 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
198 {
199 	uint64_t trysnap = 0;
200 
201 	if (ds == NULL)
202 		return (0);
203 	/*
204 	 * The snapshot creation could fail, but that would cause an
205 	 * incorrect FALSE return, which would only result in an
206 	 * overestimation of the amount of space that an operation would
207 	 * consume, which is OK.
208 	 *
209 	 * There's also a small window where we could miss a pending
210 	 * snapshot, because we could set the sync task in the quiescing
211 	 * phase.  So this should only be used as a guess.
212 	 */
213 	if (ds->ds_trysnap_txg >
214 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
215 		trysnap = ds->ds_trysnap_txg;
216 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
217 }
218 
219 int
220 dsl_dataset_block_freeable(dsl_dataset_t *ds, uint64_t blk_birth)
221 {
222 	return (blk_birth > dsl_dataset_prev_snap_txg(ds));
223 }
224 
225 /* ARGSUSED */
226 static void
227 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
228 {
229 	dsl_dataset_t *ds = dsv;
230 
231 	ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
232 
233 	dprintf_ds(ds, "evicting %s\n", "");
234 
235 	unique_remove(ds->ds_fsid_guid);
236 
237 	if (ds->ds_user_ptr != NULL)
238 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
239 
240 	if (ds->ds_prev) {
241 		dsl_dataset_drop_ref(ds->ds_prev, ds);
242 		ds->ds_prev = NULL;
243 	}
244 
245 	bplist_close(&ds->ds_deadlist);
246 	if (ds->ds_dir)
247 		dsl_dir_close(ds->ds_dir, ds);
248 
249 	ASSERT(!list_link_active(&ds->ds_synced_link));
250 
251 	mutex_destroy(&ds->ds_lock);
252 	mutex_destroy(&ds->ds_opening_lock);
253 	mutex_destroy(&ds->ds_deadlist.bpl_lock);
254 	rw_destroy(&ds->ds_rwlock);
255 	cv_destroy(&ds->ds_exclusive_cv);
256 
257 	kmem_free(ds, sizeof (dsl_dataset_t));
258 }
259 
260 static int
261 dsl_dataset_get_snapname(dsl_dataset_t *ds)
262 {
263 	dsl_dataset_phys_t *headphys;
264 	int err;
265 	dmu_buf_t *headdbuf;
266 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
267 	objset_t *mos = dp->dp_meta_objset;
268 
269 	if (ds->ds_snapname[0])
270 		return (0);
271 	if (ds->ds_phys->ds_next_snap_obj == 0)
272 		return (0);
273 
274 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
275 	    FTAG, &headdbuf);
276 	if (err)
277 		return (err);
278 	headphys = headdbuf->db_data;
279 	err = zap_value_search(dp->dp_meta_objset,
280 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
281 	dmu_buf_rele(headdbuf, FTAG);
282 	return (err);
283 }
284 
285 static int
286 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
287 {
288 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
289 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
290 	matchtype_t mt;
291 	int err;
292 
293 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
294 		mt = MT_FIRST;
295 	else
296 		mt = MT_EXACT;
297 
298 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
299 	    value, mt, NULL, 0, NULL);
300 	if (err == ENOTSUP && mt == MT_FIRST)
301 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
302 	return (err);
303 }
304 
305 static int
306 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
307 {
308 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
309 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
310 	matchtype_t mt;
311 	int err;
312 
313 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
314 		mt = MT_FIRST;
315 	else
316 		mt = MT_EXACT;
317 
318 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
319 	if (err == ENOTSUP && mt == MT_FIRST)
320 		err = zap_remove(mos, snapobj, name, tx);
321 	return (err);
322 }
323 
324 static int
325 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
326     dsl_dataset_t **dsp)
327 {
328 	objset_t *mos = dp->dp_meta_objset;
329 	dmu_buf_t *dbuf;
330 	dsl_dataset_t *ds;
331 	int err;
332 
333 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
334 	    dsl_pool_sync_context(dp));
335 
336 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
337 	if (err)
338 		return (err);
339 	ds = dmu_buf_get_user(dbuf);
340 	if (ds == NULL) {
341 		dsl_dataset_t *winner;
342 
343 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
344 		ds->ds_dbuf = dbuf;
345 		ds->ds_object = dsobj;
346 		ds->ds_phys = dbuf->db_data;
347 
348 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
349 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
350 		mutex_init(&ds->ds_deadlist.bpl_lock, NULL, MUTEX_DEFAULT,
351 		    NULL);
352 		rw_init(&ds->ds_rwlock, 0, 0, 0);
353 		cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
354 
355 		err = bplist_open(&ds->ds_deadlist,
356 		    mos, ds->ds_phys->ds_deadlist_obj);
357 		if (err == 0) {
358 			err = dsl_dir_open_obj(dp,
359 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
360 		}
361 		if (err) {
362 			/*
363 			 * we don't really need to close the blist if we
364 			 * just opened it.
365 			 */
366 			mutex_destroy(&ds->ds_lock);
367 			mutex_destroy(&ds->ds_opening_lock);
368 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
369 			rw_destroy(&ds->ds_rwlock);
370 			cv_destroy(&ds->ds_exclusive_cv);
371 			kmem_free(ds, sizeof (dsl_dataset_t));
372 			dmu_buf_rele(dbuf, tag);
373 			return (err);
374 		}
375 
376 		if (ds->ds_dir->dd_phys->dd_head_dataset_obj == dsobj) {
377 			ds->ds_snapname[0] = '\0';
378 			if (ds->ds_phys->ds_prev_snap_obj) {
379 				err = dsl_dataset_get_ref(dp,
380 				    ds->ds_phys->ds_prev_snap_obj,
381 				    ds, &ds->ds_prev);
382 			}
383 		} else if (zfs_flags & ZFS_DEBUG_SNAPNAMES) {
384 			err = dsl_dataset_get_snapname(ds);
385 		}
386 
387 		if (!dsl_dataset_is_snapshot(ds)) {
388 			/*
389 			 * In sync context, we're called with either no lock
390 			 * or with the write lock.  If we're not syncing,
391 			 * we're always called with the read lock held.
392 			 */
393 			boolean_t need_lock =
394 			    !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
395 			    dsl_pool_sync_context(dp);
396 
397 			if (need_lock)
398 				rw_enter(&dp->dp_config_rwlock, RW_READER);
399 
400 			err = dsl_prop_get_ds_locked(ds->ds_dir,
401 			    "refreservation", sizeof (uint64_t), 1,
402 			    &ds->ds_reserved, NULL);
403 			if (err == 0) {
404 				err = dsl_prop_get_ds_locked(ds->ds_dir,
405 				    "refquota", sizeof (uint64_t), 1,
406 				    &ds->ds_quota, NULL);
407 			}
408 
409 			if (need_lock)
410 				rw_exit(&dp->dp_config_rwlock);
411 		} else {
412 			ds->ds_reserved = ds->ds_quota = 0;
413 		}
414 
415 		if (err == 0) {
416 			winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
417 			    dsl_dataset_evict);
418 		}
419 		if (err || winner) {
420 			bplist_close(&ds->ds_deadlist);
421 			if (ds->ds_prev)
422 				dsl_dataset_drop_ref(ds->ds_prev, ds);
423 			dsl_dir_close(ds->ds_dir, ds);
424 			mutex_destroy(&ds->ds_lock);
425 			mutex_destroy(&ds->ds_opening_lock);
426 			mutex_destroy(&ds->ds_deadlist.bpl_lock);
427 			rw_destroy(&ds->ds_rwlock);
428 			cv_destroy(&ds->ds_exclusive_cv);
429 			kmem_free(ds, sizeof (dsl_dataset_t));
430 			if (err) {
431 				dmu_buf_rele(dbuf, tag);
432 				return (err);
433 			}
434 			ds = winner;
435 		} else {
436 			ds->ds_fsid_guid =
437 			    unique_insert(ds->ds_phys->ds_fsid_guid);
438 		}
439 	}
440 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
441 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
442 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
443 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
444 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
445 	mutex_enter(&ds->ds_lock);
446 	if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
447 		mutex_exit(&ds->ds_lock);
448 		dmu_buf_rele(ds->ds_dbuf, tag);
449 		return (ENOENT);
450 	}
451 	mutex_exit(&ds->ds_lock);
452 	*dsp = ds;
453 	return (0);
454 }
455 
456 static int
457 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
458 {
459 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
460 
461 	/*
462 	 * In syncing context we don't want the rwlock lock: there
463 	 * may be an existing writer waiting for sync phase to
464 	 * finish.  We don't need to worry about such writers, since
465 	 * sync phase is single-threaded, so the writer can't be
466 	 * doing anything while we are active.
467 	 */
468 	if (dsl_pool_sync_context(dp)) {
469 		ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
470 		return (0);
471 	}
472 
473 	/*
474 	 * Normal users will hold the ds_rwlock as a READER until they
475 	 * are finished (i.e., call dsl_dataset_rele()).  "Owners" will
476 	 * drop their READER lock after they set the ds_owner field.
477 	 *
478 	 * If the dataset is being destroyed, the destroy thread will
479 	 * obtain a WRITER lock for exclusive access after it's done its
480 	 * open-context work and then change the ds_owner to
481 	 * dsl_reaper once destruction is assured.  So threads
482 	 * may block here temporarily, until the "destructability" of
483 	 * the dataset is determined.
484 	 */
485 	ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
486 	mutex_enter(&ds->ds_lock);
487 	while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
488 		rw_exit(&dp->dp_config_rwlock);
489 		cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
490 		if (DSL_DATASET_IS_DESTROYED(ds)) {
491 			mutex_exit(&ds->ds_lock);
492 			dsl_dataset_drop_ref(ds, tag);
493 			rw_enter(&dp->dp_config_rwlock, RW_READER);
494 			return (ENOENT);
495 		}
496 		rw_enter(&dp->dp_config_rwlock, RW_READER);
497 	}
498 	mutex_exit(&ds->ds_lock);
499 	return (0);
500 }
501 
502 int
503 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
504     dsl_dataset_t **dsp)
505 {
506 	int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
507 
508 	if (err)
509 		return (err);
510 	return (dsl_dataset_hold_ref(*dsp, tag));
511 }
512 
513 int
514 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, int flags, void *owner,
515     dsl_dataset_t **dsp)
516 {
517 	int err = dsl_dataset_hold_obj(dp, dsobj, owner, dsp);
518 
519 	ASSERT(DS_MODE_TYPE(flags) != DS_MODE_USER);
520 
521 	if (err)
522 		return (err);
523 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
524 		dsl_dataset_rele(*dsp, owner);
525 		return (EBUSY);
526 	}
527 	return (0);
528 }
529 
530 int
531 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
532 {
533 	dsl_dir_t *dd;
534 	dsl_pool_t *dp;
535 	const char *snapname;
536 	uint64_t obj;
537 	int err = 0;
538 
539 	err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
540 	if (err)
541 		return (err);
542 
543 	dp = dd->dd_pool;
544 	obj = dd->dd_phys->dd_head_dataset_obj;
545 	rw_enter(&dp->dp_config_rwlock, RW_READER);
546 	if (obj)
547 		err = dsl_dataset_get_ref(dp, obj, tag, dsp);
548 	else
549 		err = ENOENT;
550 	if (err)
551 		goto out;
552 
553 	err = dsl_dataset_hold_ref(*dsp, tag);
554 
555 	/* we may be looking for a snapshot */
556 	if (err == 0 && snapname != NULL) {
557 		dsl_dataset_t *ds = NULL;
558 
559 		if (*snapname++ != '@') {
560 			dsl_dataset_rele(*dsp, tag);
561 			err = ENOENT;
562 			goto out;
563 		}
564 
565 		dprintf("looking for snapshot '%s'\n", snapname);
566 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
567 		if (err == 0)
568 			err = dsl_dataset_get_ref(dp, obj, tag, &ds);
569 		dsl_dataset_rele(*dsp, tag);
570 
571 		ASSERT3U((err == 0), ==, (ds != NULL));
572 
573 		if (ds) {
574 			mutex_enter(&ds->ds_lock);
575 			if (ds->ds_snapname[0] == 0)
576 				(void) strlcpy(ds->ds_snapname, snapname,
577 				    sizeof (ds->ds_snapname));
578 			mutex_exit(&ds->ds_lock);
579 			err = dsl_dataset_hold_ref(ds, tag);
580 			*dsp = err ? NULL : ds;
581 		}
582 	}
583 out:
584 	rw_exit(&dp->dp_config_rwlock);
585 	dsl_dir_close(dd, FTAG);
586 	return (err);
587 }
588 
589 int
590 dsl_dataset_own(const char *name, int flags, void *owner, dsl_dataset_t **dsp)
591 {
592 	int err = dsl_dataset_hold(name, owner, dsp);
593 	if (err)
594 		return (err);
595 	if ((*dsp)->ds_phys->ds_num_children > 0 &&
596 	    !DS_MODE_IS_READONLY(flags)) {
597 		dsl_dataset_rele(*dsp, owner);
598 		return (EROFS);
599 	}
600 	if (!dsl_dataset_tryown(*dsp, DS_MODE_IS_INCONSISTENT(flags), owner)) {
601 		dsl_dataset_rele(*dsp, owner);
602 		return (EBUSY);
603 	}
604 	return (0);
605 }
606 
607 void
608 dsl_dataset_name(dsl_dataset_t *ds, char *name)
609 {
610 	if (ds == NULL) {
611 		(void) strcpy(name, "mos");
612 	} else {
613 		dsl_dir_name(ds->ds_dir, name);
614 		VERIFY(0 == dsl_dataset_get_snapname(ds));
615 		if (ds->ds_snapname[0]) {
616 			(void) strcat(name, "@");
617 			/*
618 			 * We use a "recursive" mutex so that we
619 			 * can call dprintf_ds() with ds_lock held.
620 			 */
621 			if (!MUTEX_HELD(&ds->ds_lock)) {
622 				mutex_enter(&ds->ds_lock);
623 				(void) strcat(name, ds->ds_snapname);
624 				mutex_exit(&ds->ds_lock);
625 			} else {
626 				(void) strcat(name, ds->ds_snapname);
627 			}
628 		}
629 	}
630 }
631 
632 static int
633 dsl_dataset_namelen(dsl_dataset_t *ds)
634 {
635 	int result;
636 
637 	if (ds == NULL) {
638 		result = 3;	/* "mos" */
639 	} else {
640 		result = dsl_dir_namelen(ds->ds_dir);
641 		VERIFY(0 == dsl_dataset_get_snapname(ds));
642 		if (ds->ds_snapname[0]) {
643 			++result;	/* adding one for the @-sign */
644 			if (!MUTEX_HELD(&ds->ds_lock)) {
645 				mutex_enter(&ds->ds_lock);
646 				result += strlen(ds->ds_snapname);
647 				mutex_exit(&ds->ds_lock);
648 			} else {
649 				result += strlen(ds->ds_snapname);
650 			}
651 		}
652 	}
653 
654 	return (result);
655 }
656 
657 void
658 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
659 {
660 	dmu_buf_rele(ds->ds_dbuf, tag);
661 }
662 
663 void
664 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
665 {
666 	if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
667 		rw_exit(&ds->ds_rwlock);
668 	}
669 	dsl_dataset_drop_ref(ds, tag);
670 }
671 
672 void
673 dsl_dataset_disown(dsl_dataset_t *ds, void *owner)
674 {
675 	ASSERT((ds->ds_owner == owner && ds->ds_dbuf) ||
676 	    (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
677 
678 	mutex_enter(&ds->ds_lock);
679 	ds->ds_owner = NULL;
680 	if (RW_WRITE_HELD(&ds->ds_rwlock)) {
681 		rw_exit(&ds->ds_rwlock);
682 		cv_broadcast(&ds->ds_exclusive_cv);
683 	}
684 	mutex_exit(&ds->ds_lock);
685 	if (ds->ds_dbuf)
686 		dsl_dataset_drop_ref(ds, owner);
687 	else
688 		dsl_dataset_evict(ds->ds_dbuf, ds);
689 }
690 
691 boolean_t
692 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *owner)
693 {
694 	boolean_t gotit = FALSE;
695 
696 	mutex_enter(&ds->ds_lock);
697 	if (ds->ds_owner == NULL &&
698 	    (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
699 		ds->ds_owner = owner;
700 		if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
701 			rw_exit(&ds->ds_rwlock);
702 		gotit = TRUE;
703 	}
704 	mutex_exit(&ds->ds_lock);
705 	return (gotit);
706 }
707 
708 void
709 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
710 {
711 	ASSERT3P(owner, ==, ds->ds_owner);
712 	if (!RW_WRITE_HELD(&ds->ds_rwlock))
713 		rw_enter(&ds->ds_rwlock, RW_WRITER);
714 }
715 
716 uint64_t
717 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
718     uint64_t flags, dmu_tx_t *tx)
719 {
720 	dsl_pool_t *dp = dd->dd_pool;
721 	dmu_buf_t *dbuf;
722 	dsl_dataset_phys_t *dsphys;
723 	uint64_t dsobj;
724 	objset_t *mos = dp->dp_meta_objset;
725 
726 	if (origin == NULL)
727 		origin = dp->dp_origin_snap;
728 
729 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
730 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
731 	ASSERT(dmu_tx_is_syncing(tx));
732 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
733 
734 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
735 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
736 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
737 	dmu_buf_will_dirty(dbuf, tx);
738 	dsphys = dbuf->db_data;
739 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
740 	dsphys->ds_dir_obj = dd->dd_object;
741 	dsphys->ds_flags = flags;
742 	dsphys->ds_fsid_guid = unique_create();
743 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
744 	    sizeof (dsphys->ds_guid));
745 	dsphys->ds_snapnames_zapobj =
746 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
747 	    DMU_OT_NONE, 0, tx);
748 	dsphys->ds_creation_time = gethrestime_sec();
749 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
750 	dsphys->ds_deadlist_obj =
751 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
752 
753 	if (origin) {
754 		dsphys->ds_prev_snap_obj = origin->ds_object;
755 		dsphys->ds_prev_snap_txg =
756 		    origin->ds_phys->ds_creation_txg;
757 		dsphys->ds_used_bytes =
758 		    origin->ds_phys->ds_used_bytes;
759 		dsphys->ds_compressed_bytes =
760 		    origin->ds_phys->ds_compressed_bytes;
761 		dsphys->ds_uncompressed_bytes =
762 		    origin->ds_phys->ds_uncompressed_bytes;
763 		dsphys->ds_bp = origin->ds_phys->ds_bp;
764 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
765 
766 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
767 		origin->ds_phys->ds_num_children++;
768 
769 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
770 			if (origin->ds_phys->ds_next_clones_obj == 0) {
771 				origin->ds_phys->ds_next_clones_obj =
772 				    zap_create(mos,
773 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
774 			}
775 			VERIFY(0 == zap_add_int(mos,
776 			    origin->ds_phys->ds_next_clones_obj,
777 			    dsobj, tx));
778 		}
779 
780 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
781 		dd->dd_phys->dd_origin_obj = origin->ds_object;
782 	}
783 
784 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
785 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
786 
787 	dmu_buf_rele(dbuf, FTAG);
788 
789 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
790 	dd->dd_phys->dd_head_dataset_obj = dsobj;
791 
792 	return (dsobj);
793 }
794 
795 uint64_t
796 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
797     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
798 {
799 	dsl_pool_t *dp = pdd->dd_pool;
800 	uint64_t dsobj, ddobj;
801 	dsl_dir_t *dd;
802 
803 	ASSERT(lastname[0] != '@');
804 
805 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
806 	VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
807 
808 	dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
809 
810 	dsl_deleg_set_create_perms(dd, tx, cr);
811 
812 	dsl_dir_close(dd, FTAG);
813 
814 	return (dsobj);
815 }
816 
817 struct destroyarg {
818 	dsl_sync_task_group_t *dstg;
819 	char *snapname;
820 	char *failed;
821 };
822 
823 static int
824 dsl_snapshot_destroy_one(char *name, void *arg)
825 {
826 	struct destroyarg *da = arg;
827 	dsl_dataset_t *ds;
828 	char *cp;
829 	int err;
830 
831 	(void) strcat(name, "@");
832 	(void) strcat(name, da->snapname);
833 	err = dsl_dataset_own(name, DS_MODE_READONLY | DS_MODE_INCONSISTENT,
834 	    da->dstg, &ds);
835 	cp = strchr(name, '@');
836 	*cp = '\0';
837 	if (err == 0) {
838 		dsl_dataset_make_exclusive(ds, da->dstg);
839 		if (ds->ds_user_ptr) {
840 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
841 			ds->ds_user_ptr = NULL;
842 		}
843 		dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
844 		    dsl_dataset_destroy_sync, ds, da->dstg, 0);
845 	} else if (err == ENOENT) {
846 		err = 0;
847 	} else {
848 		(void) strcpy(da->failed, name);
849 	}
850 	return (err);
851 }
852 
853 /*
854  * Destroy 'snapname' in all descendants of 'fsname'.
855  */
856 #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
857 int
858 dsl_snapshots_destroy(char *fsname, char *snapname)
859 {
860 	int err;
861 	struct destroyarg da;
862 	dsl_sync_task_t *dst;
863 	spa_t *spa;
864 
865 	err = spa_open(fsname, &spa, FTAG);
866 	if (err)
867 		return (err);
868 	da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
869 	da.snapname = snapname;
870 	da.failed = fsname;
871 
872 	err = dmu_objset_find(fsname,
873 	    dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
874 
875 	if (err == 0)
876 		err = dsl_sync_task_group_wait(da.dstg);
877 
878 	for (dst = list_head(&da.dstg->dstg_tasks); dst;
879 	    dst = list_next(&da.dstg->dstg_tasks, dst)) {
880 		dsl_dataset_t *ds = dst->dst_arg1;
881 		/*
882 		 * Return the file system name that triggered the error
883 		 */
884 		if (dst->dst_err) {
885 			dsl_dataset_name(ds, fsname);
886 			*strchr(fsname, '@') = '\0';
887 		}
888 		dsl_dataset_disown(ds, da.dstg);
889 	}
890 
891 	dsl_sync_task_group_destroy(da.dstg);
892 	spa_close(spa, FTAG);
893 	return (err);
894 }
895 
896 /*
897  * ds must be opened as OWNER.  On return (whether successful or not),
898  * ds will be closed and caller can no longer dereference it.
899  */
900 int
901 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag)
902 {
903 	int err;
904 	dsl_sync_task_group_t *dstg;
905 	objset_t *os;
906 	dsl_dir_t *dd;
907 	uint64_t obj;
908 
909 	if (dsl_dataset_is_snapshot(ds)) {
910 		/* Destroying a snapshot is simpler */
911 		dsl_dataset_make_exclusive(ds, tag);
912 
913 		if (ds->ds_user_ptr) {
914 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
915 			ds->ds_user_ptr = NULL;
916 		}
917 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
918 		    dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
919 		    ds, tag, 0);
920 		goto out;
921 	}
922 
923 	dd = ds->ds_dir;
924 
925 	/*
926 	 * Check for errors and mark this ds as inconsistent, in
927 	 * case we crash while freeing the objects.
928 	 */
929 	err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
930 	    dsl_dataset_destroy_begin_sync, ds, NULL, 0);
931 	if (err)
932 		goto out;
933 
934 	err = dmu_objset_open_ds(ds, DMU_OST_ANY, &os);
935 	if (err)
936 		goto out;
937 
938 	/*
939 	 * remove the objects in open context, so that we won't
940 	 * have too much to do in syncing context.
941 	 */
942 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
943 	    ds->ds_phys->ds_prev_snap_txg)) {
944 		/*
945 		 * Ignore errors, if there is not enough disk space
946 		 * we will deal with it in dsl_dataset_destroy_sync().
947 		 */
948 		(void) dmu_free_object(os, obj);
949 	}
950 
951 	dmu_objset_close(os);
952 	if (err != ESRCH)
953 		goto out;
954 
955 	rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
956 	err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
957 	rw_exit(&dd->dd_pool->dp_config_rwlock);
958 
959 	if (err)
960 		goto out;
961 
962 	if (ds->ds_user_ptr) {
963 		/*
964 		 * We need to sync out all in-flight IO before we try
965 		 * to evict (the dataset evict func is trying to clear
966 		 * the cached entries for this dataset in the ARC).
967 		 */
968 		txg_wait_synced(dd->dd_pool, 0);
969 	}
970 
971 	/*
972 	 * Blow away the dsl_dir + head dataset.
973 	 */
974 	dsl_dataset_make_exclusive(ds, tag);
975 	if (ds->ds_user_ptr) {
976 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
977 		ds->ds_user_ptr = NULL;
978 	}
979 	dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
980 	dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
981 	    dsl_dataset_destroy_sync, ds, tag, 0);
982 	dsl_sync_task_create(dstg, dsl_dir_destroy_check,
983 	    dsl_dir_destroy_sync, dd, FTAG, 0);
984 	err = dsl_sync_task_group_wait(dstg);
985 	dsl_sync_task_group_destroy(dstg);
986 	/* if it is successful, dsl_dir_destroy_sync will close the dd */
987 	if (err)
988 		dsl_dir_close(dd, FTAG);
989 out:
990 	dsl_dataset_disown(ds, tag);
991 	return (err);
992 }
993 
994 int
995 dsl_dataset_rollback(dsl_dataset_t *ds, dmu_objset_type_t ost)
996 {
997 	ASSERT(ds->ds_owner);
998 
999 	return (dsl_sync_task_do(ds->ds_dir->dd_pool,
1000 	    dsl_dataset_rollback_check, dsl_dataset_rollback_sync,
1001 	    ds, &ost, 0));
1002 }
1003 
1004 void *
1005 dsl_dataset_set_user_ptr(dsl_dataset_t *ds,
1006     void *p, dsl_dataset_evict_func_t func)
1007 {
1008 	void *old;
1009 
1010 	mutex_enter(&ds->ds_lock);
1011 	old = ds->ds_user_ptr;
1012 	if (old == NULL) {
1013 		ds->ds_user_ptr = p;
1014 		ds->ds_user_evict_func = func;
1015 	}
1016 	mutex_exit(&ds->ds_lock);
1017 	return (old);
1018 }
1019 
1020 void *
1021 dsl_dataset_get_user_ptr(dsl_dataset_t *ds)
1022 {
1023 	return (ds->ds_user_ptr);
1024 }
1025 
1026 
1027 blkptr_t *
1028 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1029 {
1030 	return (&ds->ds_phys->ds_bp);
1031 }
1032 
1033 void
1034 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1035 {
1036 	ASSERT(dmu_tx_is_syncing(tx));
1037 	/* If it's the meta-objset, set dp_meta_rootbp */
1038 	if (ds == NULL) {
1039 		tx->tx_pool->dp_meta_rootbp = *bp;
1040 	} else {
1041 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
1042 		ds->ds_phys->ds_bp = *bp;
1043 	}
1044 }
1045 
1046 spa_t *
1047 dsl_dataset_get_spa(dsl_dataset_t *ds)
1048 {
1049 	return (ds->ds_dir->dd_pool->dp_spa);
1050 }
1051 
1052 void
1053 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1054 {
1055 	dsl_pool_t *dp;
1056 
1057 	if (ds == NULL) /* this is the meta-objset */
1058 		return;
1059 
1060 	ASSERT(ds->ds_user_ptr != NULL);
1061 
1062 	if (ds->ds_phys->ds_next_snap_obj != 0)
1063 		panic("dirtying snapshot!");
1064 
1065 	dp = ds->ds_dir->dd_pool;
1066 
1067 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1068 		/* up the hold count until we can be written out */
1069 		dmu_buf_add_ref(ds->ds_dbuf, ds);
1070 	}
1071 }
1072 
1073 /*
1074  * The unique space in the head dataset can be calculated by subtracting
1075  * the space used in the most recent snapshot, that is still being used
1076  * in this file system, from the space currently in use.  To figure out
1077  * the space in the most recent snapshot still in use, we need to take
1078  * the total space used in the snapshot and subtract out the space that
1079  * has been freed up since the snapshot was taken.
1080  */
1081 static void
1082 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1083 {
1084 	uint64_t mrs_used;
1085 	uint64_t dlused, dlcomp, dluncomp;
1086 
1087 	ASSERT(ds->ds_object == ds->ds_dir->dd_phys->dd_head_dataset_obj);
1088 
1089 	if (ds->ds_phys->ds_prev_snap_obj != 0)
1090 		mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1091 	else
1092 		mrs_used = 0;
1093 
1094 	VERIFY(0 == bplist_space(&ds->ds_deadlist, &dlused, &dlcomp,
1095 	    &dluncomp));
1096 
1097 	ASSERT3U(dlused, <=, mrs_used);
1098 	ds->ds_phys->ds_unique_bytes =
1099 	    ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1100 
1101 	if (!DS_UNIQUE_IS_ACCURATE(ds) &&
1102 	    spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1103 	    SPA_VERSION_UNIQUE_ACCURATE)
1104 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1105 }
1106 
1107 static uint64_t
1108 dsl_dataset_unique(dsl_dataset_t *ds)
1109 {
1110 	if (!DS_UNIQUE_IS_ACCURATE(ds) && !dsl_dataset_is_snapshot(ds))
1111 		dsl_dataset_recalc_head_uniq(ds);
1112 
1113 	return (ds->ds_phys->ds_unique_bytes);
1114 }
1115 
1116 struct killarg {
1117 	int64_t *usedp;
1118 	int64_t *compressedp;
1119 	int64_t *uncompressedp;
1120 	zio_t *zio;
1121 	dmu_tx_t *tx;
1122 };
1123 
1124 static int
1125 kill_blkptr(traverse_blk_cache_t *bc, spa_t *spa, void *arg)
1126 {
1127 	struct killarg *ka = arg;
1128 	blkptr_t *bp = &bc->bc_blkptr;
1129 
1130 	ASSERT3U(bc->bc_errno, ==, 0);
1131 
1132 	/*
1133 	 * Since this callback is not called concurrently, no lock is
1134 	 * needed on the accounting values.
1135 	 */
1136 	*ka->usedp += bp_get_dasize(spa, bp);
1137 	*ka->compressedp += BP_GET_PSIZE(bp);
1138 	*ka->uncompressedp += BP_GET_UCSIZE(bp);
1139 	/* XXX check for EIO? */
1140 	(void) dsl_free(ka->zio, spa_get_dsl(spa), ka->tx->tx_txg,
1141 	    bp, NULL, NULL, ARC_NOWAIT);
1142 	return (0);
1143 }
1144 
1145 /* ARGSUSED */
1146 static int
1147 dsl_dataset_rollback_check(void *arg1, void *arg2, dmu_tx_t *tx)
1148 {
1149 	dsl_dataset_t *ds = arg1;
1150 	dmu_objset_type_t *ost = arg2;
1151 
1152 	/*
1153 	 * We can only roll back to emptyness if it is a ZPL objset.
1154 	 */
1155 	if (*ost != DMU_OST_ZFS && ds->ds_phys->ds_prev_snap_txg == 0)
1156 		return (EINVAL);
1157 
1158 	/*
1159 	 * This must not be a snapshot.
1160 	 */
1161 	if (ds->ds_phys->ds_next_snap_obj != 0)
1162 		return (EINVAL);
1163 
1164 	/*
1165 	 * If we made changes this txg, traverse_dsl_dataset won't find
1166 	 * them.  Try again.
1167 	 */
1168 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1169 		return (EAGAIN);
1170 
1171 	return (0);
1172 }
1173 
1174 /* ARGSUSED */
1175 static void
1176 dsl_dataset_rollback_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1177 {
1178 	dsl_dataset_t *ds = arg1;
1179 	dmu_objset_type_t *ost = arg2;
1180 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1181 
1182 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1183 
1184 	/*
1185 	 * Before the roll back destroy the zil.
1186 	 */
1187 	if (ds->ds_user_ptr != NULL) {
1188 		zil_rollback_destroy(
1189 		    ((objset_impl_t *)ds->ds_user_ptr)->os_zil, tx);
1190 
1191 		/*
1192 		 * We need to make sure that the objset_impl_t is reopened after
1193 		 * we do the rollback, otherwise it will have the wrong
1194 		 * objset_phys_t.  Normally this would happen when this
1195 		 * dataset-open is closed, thus causing the
1196 		 * dataset to be immediately evicted.  But when doing "zfs recv
1197 		 * -F", we reopen the objset before that, so that there is no
1198 		 * window where the dataset is closed and inconsistent.
1199 		 */
1200 		ds->ds_user_evict_func(ds, ds->ds_user_ptr);
1201 		ds->ds_user_ptr = NULL;
1202 	}
1203 
1204 	/* Zero out the deadlist. */
1205 	bplist_close(&ds->ds_deadlist);
1206 	bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1207 	ds->ds_phys->ds_deadlist_obj =
1208 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1209 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1210 	    ds->ds_phys->ds_deadlist_obj));
1211 
1212 	{
1213 		/* Free blkptrs that we gave birth to */
1214 		zio_t *zio;
1215 		int64_t used = 0, compressed = 0, uncompressed = 0;
1216 		struct killarg ka;
1217 		int64_t delta;
1218 
1219 		zio = zio_root(tx->tx_pool->dp_spa, NULL, NULL,
1220 		    ZIO_FLAG_MUSTSUCCEED);
1221 		ka.usedp = &used;
1222 		ka.compressedp = &compressed;
1223 		ka.uncompressedp = &uncompressed;
1224 		ka.zio = zio;
1225 		ka.tx = tx;
1226 		(void) traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1227 		    ADVANCE_POST, kill_blkptr, &ka);
1228 		(void) zio_wait(zio);
1229 
1230 		/* only deduct space beyond any refreservation */
1231 		delta = parent_delta(ds, -used);
1232 		dsl_dir_diduse_space(ds->ds_dir,
1233 		    delta, -compressed, -uncompressed, tx);
1234 	}
1235 
1236 	if (ds->ds_prev && ds->ds_prev != ds->ds_dir->dd_pool->dp_origin_snap) {
1237 		/* Change our contents to that of the prev snapshot */
1238 		ASSERT3U(ds->ds_prev->ds_object, ==,
1239 		    ds->ds_phys->ds_prev_snap_obj);
1240 		ds->ds_phys->ds_bp = ds->ds_prev->ds_phys->ds_bp;
1241 		ds->ds_phys->ds_used_bytes =
1242 		    ds->ds_prev->ds_phys->ds_used_bytes;
1243 		ds->ds_phys->ds_compressed_bytes =
1244 		    ds->ds_prev->ds_phys->ds_compressed_bytes;
1245 		ds->ds_phys->ds_uncompressed_bytes =
1246 		    ds->ds_prev->ds_phys->ds_uncompressed_bytes;
1247 		ds->ds_phys->ds_flags = ds->ds_prev->ds_phys->ds_flags;
1248 		ds->ds_phys->ds_unique_bytes = 0;
1249 
1250 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1251 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1252 			ds->ds_prev->ds_phys->ds_unique_bytes = 0;
1253 		}
1254 	} else {
1255 		objset_impl_t *osi;
1256 
1257 		/* Zero out our contents, recreate objset */
1258 		bzero(&ds->ds_phys->ds_bp, sizeof (blkptr_t));
1259 		ds->ds_phys->ds_used_bytes = 0;
1260 		ds->ds_phys->ds_compressed_bytes = 0;
1261 		ds->ds_phys->ds_uncompressed_bytes = 0;
1262 		ds->ds_phys->ds_flags = 0;
1263 		ds->ds_phys->ds_unique_bytes = 0;
1264 		osi = dmu_objset_create_impl(ds->ds_dir->dd_pool->dp_spa, ds,
1265 		    &ds->ds_phys->ds_bp, *ost, tx);
1266 #ifdef _KERNEL
1267 		zfs_create_fs(&osi->os, kcred, NULL, tx);
1268 #endif
1269 	}
1270 
1271 	spa_history_internal_log(LOG_DS_ROLLBACK, ds->ds_dir->dd_pool->dp_spa,
1272 	    tx, cr, "dataset = %llu", ds->ds_object);
1273 }
1274 
1275 /* ARGSUSED */
1276 static int
1277 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1278 {
1279 	dsl_dataset_t *ds = arg1;
1280 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1281 	uint64_t count;
1282 	int err;
1283 
1284 	/*
1285 	 * Can't delete a head dataset if there are snapshots of it.
1286 	 * (Except if the only snapshots are from the branch we cloned
1287 	 * from.)
1288 	 */
1289 	if (ds->ds_prev != NULL &&
1290 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1291 		return (EINVAL);
1292 
1293 	/*
1294 	 * This is really a dsl_dir thing, but check it here so that
1295 	 * we'll be less likely to leave this dataset inconsistent &
1296 	 * nearly destroyed.
1297 	 */
1298 	err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1299 	if (err)
1300 		return (err);
1301 	if (count != 0)
1302 		return (EEXIST);
1303 
1304 	return (0);
1305 }
1306 
1307 /* ARGSUSED */
1308 static void
1309 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1310 {
1311 	dsl_dataset_t *ds = arg1;
1312 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1313 
1314 	/* Mark it as inconsistent on-disk, in case we crash */
1315 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1316 	ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1317 
1318 	spa_history_internal_log(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1319 	    cr, "dataset = %llu", ds->ds_object);
1320 }
1321 
1322 /* ARGSUSED */
1323 int
1324 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1325 {
1326 	dsl_dataset_t *ds = arg1;
1327 
1328 	/* we have an owner hold, so noone else can destroy us */
1329 	ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1330 
1331 	/* Can't delete a branch point. */
1332 	if (ds->ds_phys->ds_num_children > 1)
1333 		return (EEXIST);
1334 
1335 	/*
1336 	 * Can't delete a head dataset if there are snapshots of it.
1337 	 * (Except if the only snapshots are from the branch we cloned
1338 	 * from.)
1339 	 */
1340 	if (ds->ds_prev != NULL &&
1341 	    ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1342 		return (EINVAL);
1343 
1344 	/*
1345 	 * If we made changes this txg, traverse_dsl_dataset won't find
1346 	 * them.  Try again.
1347 	 */
1348 	if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1349 		return (EAGAIN);
1350 
1351 	/* XXX we should do some i/o error checking... */
1352 	return (0);
1353 }
1354 
1355 struct refsarg {
1356 	kmutex_t lock;
1357 	boolean_t gone;
1358 	kcondvar_t cv;
1359 };
1360 
1361 /* ARGSUSED */
1362 static void
1363 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1364 {
1365 	struct refsarg *arg = argv;
1366 
1367 	mutex_enter(&arg->lock);
1368 	arg->gone = TRUE;
1369 	cv_signal(&arg->cv);
1370 	mutex_exit(&arg->lock);
1371 }
1372 
1373 static void
1374 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1375 {
1376 	struct refsarg arg;
1377 
1378 	mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1379 	cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1380 	arg.gone = FALSE;
1381 	(void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1382 	    dsl_dataset_refs_gone);
1383 	dmu_buf_rele(ds->ds_dbuf, tag);
1384 	mutex_enter(&arg.lock);
1385 	while (!arg.gone)
1386 		cv_wait(&arg.cv, &arg.lock);
1387 	ASSERT(arg.gone);
1388 	mutex_exit(&arg.lock);
1389 	ds->ds_dbuf = NULL;
1390 	ds->ds_phys = NULL;
1391 	mutex_destroy(&arg.lock);
1392 	cv_destroy(&arg.cv);
1393 }
1394 
1395 void
1396 dsl_dataset_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx)
1397 {
1398 	dsl_dataset_t *ds = arg1;
1399 	int64_t used = 0, compressed = 0, uncompressed = 0;
1400 	zio_t *zio;
1401 	int err;
1402 	int after_branch_point = FALSE;
1403 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1404 	objset_t *mos = dp->dp_meta_objset;
1405 	dsl_dataset_t *ds_prev = NULL;
1406 	uint64_t obj;
1407 
1408 	ASSERT(ds->ds_owner);
1409 	ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
1410 	ASSERT(ds->ds_prev == NULL ||
1411 	    ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1412 	ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1413 
1414 	/* signal any waiters that this dataset is going away */
1415 	mutex_enter(&ds->ds_lock);
1416 	ds->ds_owner = dsl_reaper;
1417 	cv_broadcast(&ds->ds_exclusive_cv);
1418 	mutex_exit(&ds->ds_lock);
1419 
1420 	/* Remove our reservation */
1421 	if (ds->ds_reserved != 0) {
1422 		uint64_t val = 0;
1423 		dsl_dataset_set_reservation_sync(ds, &val, cr, tx);
1424 		ASSERT3U(ds->ds_reserved, ==, 0);
1425 	}
1426 
1427 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1428 
1429 	dsl_pool_ds_destroyed(ds, tx);
1430 
1431 	obj = ds->ds_object;
1432 
1433 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1434 		if (ds->ds_prev) {
1435 			ds_prev = ds->ds_prev;
1436 		} else {
1437 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1438 			    ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1439 		}
1440 		after_branch_point =
1441 		    (ds_prev->ds_phys->ds_next_snap_obj != obj);
1442 
1443 		dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1444 		if (after_branch_point &&
1445 		    ds_prev->ds_phys->ds_next_clones_obj != 0) {
1446 			VERIFY(0 == zap_remove_int(mos,
1447 			    ds_prev->ds_phys->ds_next_clones_obj, obj, tx));
1448 			if (ds->ds_phys->ds_next_snap_obj != 0) {
1449 				VERIFY(0 == zap_add_int(mos,
1450 				    ds_prev->ds_phys->ds_next_clones_obj,
1451 				    ds->ds_phys->ds_next_snap_obj, tx));
1452 			}
1453 		}
1454 		if (after_branch_point &&
1455 		    ds->ds_phys->ds_next_snap_obj == 0) {
1456 			/* This clone is toast. */
1457 			ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1458 			ds_prev->ds_phys->ds_num_children--;
1459 		} else if (!after_branch_point) {
1460 			ds_prev->ds_phys->ds_next_snap_obj =
1461 			    ds->ds_phys->ds_next_snap_obj;
1462 		}
1463 	}
1464 
1465 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1466 
1467 	if (ds->ds_phys->ds_next_snap_obj != 0) {
1468 		blkptr_t bp;
1469 		dsl_dataset_t *ds_next;
1470 		uint64_t itor = 0;
1471 		uint64_t old_unique;
1472 
1473 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1474 		    ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1475 		ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1476 
1477 		old_unique = dsl_dataset_unique(ds_next);
1478 
1479 		dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1480 		ds_next->ds_phys->ds_prev_snap_obj =
1481 		    ds->ds_phys->ds_prev_snap_obj;
1482 		ds_next->ds_phys->ds_prev_snap_txg =
1483 		    ds->ds_phys->ds_prev_snap_txg;
1484 		ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1485 		    ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1486 
1487 		/*
1488 		 * Transfer to our deadlist (which will become next's
1489 		 * new deadlist) any entries from next's current
1490 		 * deadlist which were born before prev, and free the
1491 		 * other entries.
1492 		 *
1493 		 * XXX we're doing this long task with the config lock held
1494 		 */
1495 		while (bplist_iterate(&ds_next->ds_deadlist, &itor, &bp) == 0) {
1496 			if (bp.blk_birth <= ds->ds_phys->ds_prev_snap_txg) {
1497 				VERIFY(0 == bplist_enqueue(&ds->ds_deadlist,
1498 				    &bp, tx));
1499 				if (ds_prev && !after_branch_point &&
1500 				    bp.blk_birth >
1501 				    ds_prev->ds_phys->ds_prev_snap_txg) {
1502 					ds_prev->ds_phys->ds_unique_bytes +=
1503 					    bp_get_dasize(dp->dp_spa, &bp);
1504 				}
1505 			} else {
1506 				used += bp_get_dasize(dp->dp_spa, &bp);
1507 				compressed += BP_GET_PSIZE(&bp);
1508 				uncompressed += BP_GET_UCSIZE(&bp);
1509 				/* XXX check return value? */
1510 				(void) dsl_free(zio, dp, tx->tx_txg,
1511 				    &bp, NULL, NULL, ARC_NOWAIT);
1512 			}
1513 		}
1514 
1515 		/* free next's deadlist */
1516 		bplist_close(&ds_next->ds_deadlist);
1517 		bplist_destroy(mos, ds_next->ds_phys->ds_deadlist_obj, tx);
1518 
1519 		/* set next's deadlist to our deadlist */
1520 		bplist_close(&ds->ds_deadlist);
1521 		ds_next->ds_phys->ds_deadlist_obj =
1522 		    ds->ds_phys->ds_deadlist_obj;
1523 		VERIFY(0 == bplist_open(&ds_next->ds_deadlist, mos,
1524 		    ds_next->ds_phys->ds_deadlist_obj));
1525 		ds->ds_phys->ds_deadlist_obj = 0;
1526 
1527 		if (ds_next->ds_phys->ds_next_snap_obj != 0) {
1528 			/*
1529 			 * Update next's unique to include blocks which
1530 			 * were previously shared by only this snapshot
1531 			 * and it.  Those blocks will be born after the
1532 			 * prev snap and before this snap, and will have
1533 			 * died after the next snap and before the one
1534 			 * after that (ie. be on the snap after next's
1535 			 * deadlist).
1536 			 *
1537 			 * XXX we're doing this long task with the
1538 			 * config lock held
1539 			 */
1540 			dsl_dataset_t *ds_after_next;
1541 
1542 			VERIFY(0 == dsl_dataset_hold_obj(dp,
1543 			    ds_next->ds_phys->ds_next_snap_obj,
1544 			    FTAG, &ds_after_next));
1545 			itor = 0;
1546 			while (bplist_iterate(&ds_after_next->ds_deadlist,
1547 			    &itor, &bp) == 0) {
1548 				if (bp.blk_birth >
1549 				    ds->ds_phys->ds_prev_snap_txg &&
1550 				    bp.blk_birth <=
1551 				    ds->ds_phys->ds_creation_txg) {
1552 					ds_next->ds_phys->ds_unique_bytes +=
1553 					    bp_get_dasize(dp->dp_spa, &bp);
1554 				}
1555 			}
1556 
1557 			dsl_dataset_rele(ds_after_next, FTAG);
1558 			ASSERT3P(ds_next->ds_prev, ==, NULL);
1559 		} else {
1560 			ASSERT3P(ds_next->ds_prev, ==, ds);
1561 			dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1562 			ds_next->ds_prev = NULL;
1563 			if (ds_prev) {
1564 				VERIFY(0 == dsl_dataset_get_ref(dp,
1565 				    ds->ds_phys->ds_prev_snap_obj,
1566 				    ds_next, &ds_next->ds_prev));
1567 			}
1568 
1569 			dsl_dataset_recalc_head_uniq(ds_next);
1570 
1571 			/*
1572 			 * Reduce the amount of our unconsmed refreservation
1573 			 * being charged to our parent by the amount of
1574 			 * new unique data we have gained.
1575 			 */
1576 			if (old_unique < ds_next->ds_reserved) {
1577 				int64_t mrsdelta;
1578 				uint64_t new_unique =
1579 				    ds_next->ds_phys->ds_unique_bytes;
1580 
1581 				ASSERT(old_unique <= new_unique);
1582 				mrsdelta = MIN(new_unique - old_unique,
1583 				    ds_next->ds_reserved - old_unique);
1584 				dsl_dir_diduse_space(ds->ds_dir, -mrsdelta,
1585 				    0, 0, tx);
1586 			}
1587 		}
1588 		dsl_dataset_rele(ds_next, FTAG);
1589 
1590 		/*
1591 		 * NB: unique_bytes might not be accurate for the head objset.
1592 		 * Before SPA_VERSION 9, we didn't update its value when we
1593 		 * deleted the most recent snapshot.
1594 		 */
1595 		ASSERT3U(used, ==, ds->ds_phys->ds_unique_bytes);
1596 	} else {
1597 		/*
1598 		 * There's no next snapshot, so this is a head dataset.
1599 		 * Destroy the deadlist.  Unless it's a clone, the
1600 		 * deadlist should be empty.  (If it's a clone, it's
1601 		 * safe to ignore the deadlist contents.)
1602 		 */
1603 		struct killarg ka;
1604 
1605 		ASSERT(after_branch_point || bplist_empty(&ds->ds_deadlist));
1606 		bplist_close(&ds->ds_deadlist);
1607 		bplist_destroy(mos, ds->ds_phys->ds_deadlist_obj, tx);
1608 		ds->ds_phys->ds_deadlist_obj = 0;
1609 
1610 		/*
1611 		 * Free everything that we point to (that's born after
1612 		 * the previous snapshot, if we are a clone)
1613 		 *
1614 		 * XXX we're doing this long task with the config lock held
1615 		 */
1616 		ka.usedp = &used;
1617 		ka.compressedp = &compressed;
1618 		ka.uncompressedp = &uncompressed;
1619 		ka.zio = zio;
1620 		ka.tx = tx;
1621 		err = traverse_dsl_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1622 		    ADVANCE_POST, kill_blkptr, &ka);
1623 		ASSERT3U(err, ==, 0);
1624 		ASSERT(spa_version(dp->dp_spa) <
1625 		    SPA_VERSION_UNIQUE_ACCURATE ||
1626 		    used == ds->ds_phys->ds_unique_bytes);
1627 	}
1628 
1629 	err = zio_wait(zio);
1630 	ASSERT3U(err, ==, 0);
1631 
1632 	dsl_dir_diduse_space(ds->ds_dir, -used, -compressed, -uncompressed, tx);
1633 
1634 	if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1635 		/* Erase the link in the dir */
1636 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1637 		ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1638 		ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1639 		err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1640 		ASSERT(err == 0);
1641 	} else {
1642 		/* remove from snapshot namespace */
1643 		dsl_dataset_t *ds_head;
1644 		ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1645 		VERIFY(0 == dsl_dataset_hold_obj(dp,
1646 		    ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1647 		VERIFY(0 == dsl_dataset_get_snapname(ds));
1648 #ifdef ZFS_DEBUG
1649 		{
1650 			uint64_t val;
1651 
1652 			err = dsl_dataset_snap_lookup(ds_head,
1653 			    ds->ds_snapname, &val);
1654 			ASSERT3U(err, ==, 0);
1655 			ASSERT3U(val, ==, obj);
1656 		}
1657 #endif
1658 		err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1659 		ASSERT(err == 0);
1660 		dsl_dataset_rele(ds_head, FTAG);
1661 	}
1662 
1663 	if (ds_prev && ds->ds_prev != ds_prev)
1664 		dsl_dataset_rele(ds_prev, FTAG);
1665 
1666 	spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1667 	spa_history_internal_log(LOG_DS_DESTROY, dp->dp_spa, tx,
1668 	    cr, "dataset = %llu", ds->ds_object);
1669 
1670 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1671 		uint64_t count;
1672 		ASSERT(0 == zap_count(mos,
1673 		    ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1674 		VERIFY(0 == dmu_object_free(mos,
1675 		    ds->ds_phys->ds_next_clones_obj, tx));
1676 	}
1677 	dsl_dir_close(ds->ds_dir, ds);
1678 	ds->ds_dir = NULL;
1679 	dsl_dataset_drain_refs(ds, tag);
1680 	VERIFY(0 == dmu_object_free(mos, obj, tx));
1681 }
1682 
1683 static int
1684 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1685 {
1686 	uint64_t asize;
1687 
1688 	if (!dmu_tx_is_syncing(tx))
1689 		return (0);
1690 
1691 	/*
1692 	 * If there's an fs-only reservation, any blocks that might become
1693 	 * owned by the snapshot dataset must be accommodated by space
1694 	 * outside of the reservation.
1695 	 */
1696 	asize = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1697 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, FALSE))
1698 		return (ENOSPC);
1699 
1700 	/*
1701 	 * Propogate any reserved space for this snapshot to other
1702 	 * snapshot checks in this sync group.
1703 	 */
1704 	if (asize > 0)
1705 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1706 
1707 	return (0);
1708 }
1709 
1710 /* ARGSUSED */
1711 int
1712 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1713 {
1714 	dsl_dataset_t *ds = arg1;
1715 	const char *snapname = arg2;
1716 	int err;
1717 	uint64_t value;
1718 
1719 	/*
1720 	 * We don't allow multiple snapshots of the same txg.  If there
1721 	 * is already one, try again.
1722 	 */
1723 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1724 		return (EAGAIN);
1725 
1726 	/*
1727 	 * Check for conflicting name snapshot name.
1728 	 */
1729 	err = dsl_dataset_snap_lookup(ds, snapname, &value);
1730 	if (err == 0)
1731 		return (EEXIST);
1732 	if (err != ENOENT)
1733 		return (err);
1734 
1735 	/*
1736 	 * Check that the dataset's name is not too long.  Name consists
1737 	 * of the dataset's length + 1 for the @-sign + snapshot name's length
1738 	 */
1739 	if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1740 		return (ENAMETOOLONG);
1741 
1742 	err = dsl_dataset_snapshot_reserve_space(ds, tx);
1743 	if (err)
1744 		return (err);
1745 
1746 	ds->ds_trysnap_txg = tx->tx_txg;
1747 	return (0);
1748 }
1749 
1750 void
1751 dsl_dataset_snapshot_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
1752 {
1753 	dsl_dataset_t *ds = arg1;
1754 	const char *snapname = arg2;
1755 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1756 	dmu_buf_t *dbuf;
1757 	dsl_dataset_phys_t *dsphys;
1758 	uint64_t dsobj, crtxg;
1759 	objset_t *mos = dp->dp_meta_objset;
1760 	int err;
1761 
1762 	ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1763 
1764 	/*
1765 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1766 	 */
1767 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1768 		crtxg = 1;
1769 	else
1770 		crtxg = tx->tx_txg;
1771 
1772 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1773 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1774 	VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1775 	dmu_buf_will_dirty(dbuf, tx);
1776 	dsphys = dbuf->db_data;
1777 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1778 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1779 	dsphys->ds_fsid_guid = unique_create();
1780 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1781 	    sizeof (dsphys->ds_guid));
1782 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1783 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1784 	dsphys->ds_next_snap_obj = ds->ds_object;
1785 	dsphys->ds_num_children = 1;
1786 	dsphys->ds_creation_time = gethrestime_sec();
1787 	dsphys->ds_creation_txg = crtxg;
1788 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1789 	dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
1790 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1791 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1792 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1793 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1794 	dmu_buf_rele(dbuf, FTAG);
1795 
1796 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1797 	if (ds->ds_prev) {
1798 		uint64_t next_clones_obj =
1799 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1800 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1801 		    ds->ds_object ||
1802 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1803 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1804 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1805 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1806 			    ds->ds_prev->ds_phys->ds_creation_txg);
1807 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1808 		} else if (next_clones_obj != 0) {
1809 			VERIFY3U(0, ==, zap_remove_int(mos,
1810 			    next_clones_obj, dsphys->ds_next_snap_obj, tx));
1811 			VERIFY3U(0, ==, zap_add_int(mos,
1812 			    next_clones_obj, dsobj, tx));
1813 		}
1814 	}
1815 
1816 	/*
1817 	 * If we have a reference-reservation on this dataset, we will
1818 	 * need to increase the amount of refreservation being charged
1819 	 * since our unique space is going to zero.
1820 	 */
1821 	if (ds->ds_reserved) {
1822 		int64_t add = MIN(dsl_dataset_unique(ds), ds->ds_reserved);
1823 		dsl_dir_diduse_space(ds->ds_dir, add, 0, 0, tx);
1824 	}
1825 
1826 	bplist_close(&ds->ds_deadlist);
1827 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1828 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1829 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1830 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1831 	ds->ds_phys->ds_unique_bytes = 0;
1832 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1833 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1834 	ds->ds_phys->ds_deadlist_obj =
1835 	    bplist_create(mos, DSL_DEADLIST_BLOCKSIZE, tx);
1836 	VERIFY(0 == bplist_open(&ds->ds_deadlist, mos,
1837 	    ds->ds_phys->ds_deadlist_obj));
1838 
1839 	dprintf("snap '%s' -> obj %llu\n", snapname, dsobj);
1840 	err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1841 	    snapname, 8, 1, &dsobj, tx);
1842 	ASSERT(err == 0);
1843 
1844 	if (ds->ds_prev)
1845 		dsl_dataset_drop_ref(ds->ds_prev, ds);
1846 	VERIFY(0 == dsl_dataset_get_ref(dp,
1847 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1848 
1849 	dsl_pool_ds_snapshotted(ds, tx);
1850 
1851 	spa_history_internal_log(LOG_DS_SNAPSHOT, dp->dp_spa, tx, cr,
1852 	    "dataset = %llu", dsobj);
1853 }
1854 
1855 void
1856 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1857 {
1858 	ASSERT(dmu_tx_is_syncing(tx));
1859 	ASSERT(ds->ds_user_ptr != NULL);
1860 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1861 
1862 	/*
1863 	 * in case we had to change ds_fsid_guid when we opened it,
1864 	 * sync it out now.
1865 	 */
1866 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1867 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1868 
1869 	dsl_dir_dirty(ds->ds_dir, tx);
1870 	dmu_objset_sync(ds->ds_user_ptr, zio, tx);
1871 }
1872 
1873 void
1874 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1875 {
1876 	uint64_t refd, avail, uobjs, aobjs;
1877 
1878 	dsl_dir_stats(ds->ds_dir, nv);
1879 
1880 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1881 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1882 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1883 
1884 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1885 	    ds->ds_phys->ds_creation_time);
1886 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1887 	    ds->ds_phys->ds_creation_txg);
1888 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1889 	    ds->ds_quota);
1890 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1891 	    ds->ds_reserved);
1892 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1893 	    ds->ds_phys->ds_guid);
1894 
1895 	if (ds->ds_phys->ds_next_snap_obj) {
1896 		/*
1897 		 * This is a snapshot; override the dd's space used with
1898 		 * our unique space and compression ratio.
1899 		 */
1900 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1901 		    ds->ds_phys->ds_unique_bytes);
1902 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
1903 		    ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1904 		    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1905 		    ds->ds_phys->ds_compressed_bytes));
1906 	}
1907 }
1908 
1909 void
1910 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1911 {
1912 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1913 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1914 	stat->dds_guid = ds->ds_phys->ds_guid;
1915 	if (ds->ds_phys->ds_next_snap_obj) {
1916 		stat->dds_is_snapshot = B_TRUE;
1917 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1918 	}
1919 
1920 	/* clone origin is really a dsl_dir thing... */
1921 	rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
1922 	if (dsl_dir_is_clone(ds->ds_dir)) {
1923 		dsl_dataset_t *ods;
1924 
1925 		VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
1926 		    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1927 		dsl_dataset_name(ods, stat->dds_origin);
1928 		dsl_dataset_drop_ref(ods, FTAG);
1929 	}
1930 	rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
1931 }
1932 
1933 uint64_t
1934 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1935 {
1936 	return (ds->ds_fsid_guid);
1937 }
1938 
1939 void
1940 dsl_dataset_space(dsl_dataset_t *ds,
1941     uint64_t *refdbytesp, uint64_t *availbytesp,
1942     uint64_t *usedobjsp, uint64_t *availobjsp)
1943 {
1944 	*refdbytesp = ds->ds_phys->ds_used_bytes;
1945 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1946 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1947 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1948 	if (ds->ds_quota != 0) {
1949 		/*
1950 		 * Adjust available bytes according to refquota
1951 		 */
1952 		if (*refdbytesp < ds->ds_quota)
1953 			*availbytesp = MIN(*availbytesp,
1954 			    ds->ds_quota - *refdbytesp);
1955 		else
1956 			*availbytesp = 0;
1957 	}
1958 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1959 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1960 }
1961 
1962 boolean_t
1963 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1964 {
1965 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1966 
1967 	ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
1968 	    dsl_pool_sync_context(dp));
1969 	if (ds->ds_prev == NULL)
1970 		return (B_FALSE);
1971 	if (ds->ds_phys->ds_bp.blk_birth >
1972 	    ds->ds_prev->ds_phys->ds_creation_txg)
1973 		return (B_TRUE);
1974 	return (B_FALSE);
1975 }
1976 
1977 /* ARGSUSED */
1978 static int
1979 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
1980 {
1981 	dsl_dataset_t *ds = arg1;
1982 	char *newsnapname = arg2;
1983 	dsl_dir_t *dd = ds->ds_dir;
1984 	dsl_dataset_t *hds;
1985 	uint64_t val;
1986 	int err;
1987 
1988 	err = dsl_dataset_hold_obj(dd->dd_pool,
1989 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
1990 	if (err)
1991 		return (err);
1992 
1993 	/* new name better not be in use */
1994 	err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
1995 	dsl_dataset_rele(hds, FTAG);
1996 
1997 	if (err == 0)
1998 		err = EEXIST;
1999 	else if (err == ENOENT)
2000 		err = 0;
2001 
2002 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2003 	if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2004 		err = ENAMETOOLONG;
2005 
2006 	return (err);
2007 }
2008 
2009 static void
2010 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2,
2011     cred_t *cr, dmu_tx_t *tx)
2012 {
2013 	dsl_dataset_t *ds = arg1;
2014 	const char *newsnapname = arg2;
2015 	dsl_dir_t *dd = ds->ds_dir;
2016 	objset_t *mos = dd->dd_pool->dp_meta_objset;
2017 	dsl_dataset_t *hds;
2018 	int err;
2019 
2020 	ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2021 
2022 	VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2023 	    dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2024 
2025 	VERIFY(0 == dsl_dataset_get_snapname(ds));
2026 	err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2027 	ASSERT3U(err, ==, 0);
2028 	mutex_enter(&ds->ds_lock);
2029 	(void) strcpy(ds->ds_snapname, newsnapname);
2030 	mutex_exit(&ds->ds_lock);
2031 	err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2032 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2033 	ASSERT3U(err, ==, 0);
2034 
2035 	spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2036 	    cr, "dataset = %llu", ds->ds_object);
2037 	dsl_dataset_rele(hds, FTAG);
2038 }
2039 
2040 struct renamesnaparg {
2041 	dsl_sync_task_group_t *dstg;
2042 	char failed[MAXPATHLEN];
2043 	char *oldsnap;
2044 	char *newsnap;
2045 };
2046 
2047 static int
2048 dsl_snapshot_rename_one(char *name, void *arg)
2049 {
2050 	struct renamesnaparg *ra = arg;
2051 	dsl_dataset_t *ds = NULL;
2052 	char *cp;
2053 	int err;
2054 
2055 	cp = name + strlen(name);
2056 	*cp = '@';
2057 	(void) strcpy(cp + 1, ra->oldsnap);
2058 
2059 	/*
2060 	 * For recursive snapshot renames the parent won't be changing
2061 	 * so we just pass name for both the to/from argument.
2062 	 */
2063 	if (err = zfs_secpolicy_rename_perms(name, name, CRED())) {
2064 		(void) strcpy(ra->failed, name);
2065 		return (err);
2066 	}
2067 
2068 #ifdef _KERNEL
2069 	/*
2070 	 * For all filesystems undergoing rename, we'll need to unmount it.
2071 	 */
2072 	(void) zfs_unmount_snap(name, NULL);
2073 #endif
2074 	err = dsl_dataset_hold(name, ra->dstg, &ds);
2075 	*cp = '\0';
2076 	if (err == ENOENT) {
2077 		return (0);
2078 	} else if (err) {
2079 		(void) strcpy(ra->failed, name);
2080 		return (err);
2081 	}
2082 
2083 	dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2084 	    dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2085 
2086 	return (0);
2087 }
2088 
2089 static int
2090 dsl_recursive_rename(char *oldname, const char *newname)
2091 {
2092 	int err;
2093 	struct renamesnaparg *ra;
2094 	dsl_sync_task_t *dst;
2095 	spa_t *spa;
2096 	char *cp, *fsname = spa_strdup(oldname);
2097 	int len = strlen(oldname);
2098 
2099 	/* truncate the snapshot name to get the fsname */
2100 	cp = strchr(fsname, '@');
2101 	*cp = '\0';
2102 
2103 	err = spa_open(fsname, &spa, FTAG);
2104 	if (err) {
2105 		kmem_free(fsname, len + 1);
2106 		return (err);
2107 	}
2108 	ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2109 	ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2110 
2111 	ra->oldsnap = strchr(oldname, '@') + 1;
2112 	ra->newsnap = strchr(newname, '@') + 1;
2113 	*ra->failed = '\0';
2114 
2115 	err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2116 	    DS_FIND_CHILDREN);
2117 	kmem_free(fsname, len + 1);
2118 
2119 	if (err == 0) {
2120 		err = dsl_sync_task_group_wait(ra->dstg);
2121 	}
2122 
2123 	for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2124 	    dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2125 		dsl_dataset_t *ds = dst->dst_arg1;
2126 		if (dst->dst_err) {
2127 			dsl_dir_name(ds->ds_dir, ra->failed);
2128 			(void) strcat(ra->failed, "@");
2129 			(void) strcat(ra->failed, ra->newsnap);
2130 		}
2131 		dsl_dataset_rele(ds, ra->dstg);
2132 	}
2133 
2134 	if (err)
2135 		(void) strcpy(oldname, ra->failed);
2136 
2137 	dsl_sync_task_group_destroy(ra->dstg);
2138 	kmem_free(ra, sizeof (struct renamesnaparg));
2139 	spa_close(spa, FTAG);
2140 	return (err);
2141 }
2142 
2143 static int
2144 dsl_valid_rename(char *oldname, void *arg)
2145 {
2146 	int delta = *(int *)arg;
2147 
2148 	if (strlen(oldname) + delta >= MAXNAMELEN)
2149 		return (ENAMETOOLONG);
2150 
2151 	return (0);
2152 }
2153 
2154 #pragma weak dmu_objset_rename = dsl_dataset_rename
2155 int
2156 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2157 {
2158 	dsl_dir_t *dd;
2159 	dsl_dataset_t *ds;
2160 	const char *tail;
2161 	int err;
2162 
2163 	err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2164 	if (err)
2165 		return (err);
2166 	if (tail == NULL) {
2167 		int delta = strlen(newname) - strlen(oldname);
2168 
2169 		/* if we're growing, validate child name lengths */
2170 		if (delta > 0)
2171 			err = dmu_objset_find(oldname, dsl_valid_rename,
2172 			    &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2173 
2174 		if (!err)
2175 			err = dsl_dir_rename(dd, newname);
2176 		dsl_dir_close(dd, FTAG);
2177 		return (err);
2178 	}
2179 	if (tail[0] != '@') {
2180 		/* the name ended in a nonexistant component */
2181 		dsl_dir_close(dd, FTAG);
2182 		return (ENOENT);
2183 	}
2184 
2185 	dsl_dir_close(dd, FTAG);
2186 
2187 	/* new name must be snapshot in same filesystem */
2188 	tail = strchr(newname, '@');
2189 	if (tail == NULL)
2190 		return (EINVAL);
2191 	tail++;
2192 	if (strncmp(oldname, newname, tail - newname) != 0)
2193 		return (EXDEV);
2194 
2195 	if (recursive) {
2196 		err = dsl_recursive_rename(oldname, newname);
2197 	} else {
2198 		err = dsl_dataset_hold(oldname, FTAG, &ds);
2199 		if (err)
2200 			return (err);
2201 
2202 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2203 		    dsl_dataset_snapshot_rename_check,
2204 		    dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2205 
2206 		dsl_dataset_rele(ds, FTAG);
2207 	}
2208 
2209 	return (err);
2210 }
2211 
2212 struct promotenode {
2213 	list_node_t link;
2214 	dsl_dataset_t *ds;
2215 };
2216 
2217 struct promotearg {
2218 	list_t snap_list;
2219 	dsl_dataset_t *clone_origin, *old_head;
2220 	uint64_t used, comp, uncomp, unique;
2221 	uint64_t newnext_obj;
2222 };
2223 
2224 /* ARGSUSED */
2225 static int
2226 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2227 {
2228 	dsl_dataset_t *hds = arg1;
2229 	struct promotearg *pa = arg2;
2230 	struct promotenode *snap = list_head(&pa->snap_list);
2231 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2232 	dsl_dataset_t *origin_ds = snap->ds;
2233 	dsl_dataset_t *newnext_ds;
2234 	char *name;
2235 	uint64_t itor = 0;
2236 	blkptr_t bp;
2237 	int err;
2238 
2239 	/* Check that it is a real clone */
2240 	if (!dsl_dir_is_clone(hds->ds_dir))
2241 		return (EINVAL);
2242 
2243 	/* Since this is so expensive, don't do the preliminary check */
2244 	if (!dmu_tx_is_syncing(tx))
2245 		return (0);
2246 
2247 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2248 		return (EXDEV);
2249 
2250 	/* find origin's new next ds */
2251 	newnext_ds = hds;
2252 	while (newnext_ds->ds_phys->ds_prev_snap_obj != origin_ds->ds_object) {
2253 		dsl_dataset_t *prev;
2254 
2255 		err = dsl_dataset_hold_obj(dp,
2256 		    newnext_ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2257 		if (newnext_ds != hds)
2258 			dsl_dataset_rele(newnext_ds, FTAG);
2259 		if (err)
2260 			return (err);
2261 		newnext_ds = prev;
2262 	}
2263 	pa->newnext_obj = newnext_ds->ds_object;
2264 
2265 	/* compute origin's new unique space */
2266 	pa->unique = 0;
2267 	while ((err = bplist_iterate(&newnext_ds->ds_deadlist,
2268 	    &itor, &bp)) == 0) {
2269 		if (bp.blk_birth > origin_ds->ds_phys->ds_prev_snap_txg)
2270 			pa->unique += bp_get_dasize(dp->dp_spa, &bp);
2271 	}
2272 	if (newnext_ds != hds)
2273 		dsl_dataset_rele(newnext_ds, FTAG);
2274 	if (err != ENOENT)
2275 		return (err);
2276 
2277 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2278 
2279 	/*
2280 	 * Walk the snapshots that we are moving
2281 	 *
2282 	 * Compute space to transfer.  Each snapshot gave birth to:
2283 	 * (my used) - (prev's used) + (deadlist's used)
2284 	 * So a sequence would look like:
2285 	 * uN - u(N-1) + dN + ... + u1 - u0 + d1 + u0 - 0 + d0
2286 	 * Which simplifies to:
2287 	 * uN + dN + ... + d1 + d0
2288 	 * Note however, if we stop before we reach the ORIGIN we get:
2289 	 * uN + dN + ... + dM - uM-1
2290 	 */
2291 	pa->used = origin_ds->ds_phys->ds_used_bytes;
2292 	pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2293 	pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2294 	do {
2295 		uint64_t val, dlused, dlcomp, dluncomp;
2296 		dsl_dataset_t *ds = snap->ds;
2297 
2298 		/* Check that the snapshot name does not conflict */
2299 		dsl_dataset_name(ds, name);
2300 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2301 		if (err == 0)
2302 			err = EEXIST;
2303 		if (err != ENOENT)
2304 			break;
2305 		err = 0;
2306 
2307 		/* The very first snapshot does not have a deadlist */
2308 		if (ds->ds_phys->ds_prev_snap_obj != 0) {
2309 			if (err = bplist_space(&ds->ds_deadlist,
2310 			    &dlused, &dlcomp, &dluncomp))
2311 				break;
2312 			pa->used += dlused;
2313 			pa->comp += dlcomp;
2314 			pa->uncomp += dluncomp;
2315 		}
2316 	} while (snap = list_next(&pa->snap_list, snap));
2317 
2318 	/*
2319 	 * If we are a clone of a clone then we never reached ORIGIN,
2320 	 * so we need to subtract out the clone origin's used space.
2321 	 */
2322 	if (pa->clone_origin) {
2323 		pa->used -= pa->clone_origin->ds_phys->ds_used_bytes;
2324 		pa->comp -= pa->clone_origin->ds_phys->ds_compressed_bytes;
2325 		pa->uncomp -= pa->clone_origin->ds_phys->ds_uncompressed_bytes;
2326 	}
2327 
2328 	kmem_free(name, MAXPATHLEN);
2329 
2330 	/* Check that there is enough space here */
2331 	if (err == 0) {
2332 		dsl_dir_t *odd = origin_ds->ds_dir;
2333 		err = dsl_dir_transfer_possible(odd, hds->ds_dir, pa->used);
2334 	}
2335 
2336 	return (err);
2337 }
2338 
2339 static void
2340 dsl_dataset_promote_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2341 {
2342 	dsl_dataset_t *hds = arg1;
2343 	struct promotearg *pa = arg2;
2344 	struct promotenode *snap = list_head(&pa->snap_list);
2345 	dsl_dataset_t *origin_ds = snap->ds;
2346 	dsl_dir_t *dd = hds->ds_dir;
2347 	dsl_pool_t *dp = hds->ds_dir->dd_pool;
2348 	dsl_dir_t *odd = NULL;
2349 	char *name;
2350 	uint64_t oldnext_obj;
2351 
2352 	ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2353 
2354 	/*
2355 	 * We need to explicitly open odd, since origin_ds's dd will be
2356 	 * changing.
2357 	 */
2358 	VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2359 	    NULL, FTAG, &odd));
2360 
2361 	/* change origin's next snap */
2362 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2363 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2364 	origin_ds->ds_phys->ds_next_snap_obj = pa->newnext_obj;
2365 
2366 	/* change the origin's next clone */
2367 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2368 		VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2369 		    origin_ds->ds_phys->ds_next_clones_obj,
2370 		    pa->newnext_obj, tx));
2371 		VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2372 		    origin_ds->ds_phys->ds_next_clones_obj,
2373 		    oldnext_obj, tx));
2374 	}
2375 
2376 	/* change origin */
2377 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2378 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2379 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2380 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2381 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2382 
2383 	/* move snapshots to this dir */
2384 	name = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2385 	do {
2386 		dsl_dataset_t *ds = snap->ds;
2387 
2388 		/* unregister props as dsl_dir is changing */
2389 		if (ds->ds_user_ptr) {
2390 			ds->ds_user_evict_func(ds, ds->ds_user_ptr);
2391 			ds->ds_user_ptr = NULL;
2392 		}
2393 		/* move snap name entry */
2394 		dsl_dataset_name(ds, name);
2395 		VERIFY(0 == dsl_dataset_snap_remove(pa->old_head,
2396 		    ds->ds_snapname, tx));
2397 		VERIFY(0 == zap_add(dp->dp_meta_objset,
2398 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2399 		    8, 1, &ds->ds_object, tx));
2400 		/* change containing dsl_dir */
2401 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2402 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2403 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2404 		ASSERT3P(ds->ds_dir, ==, odd);
2405 		dsl_dir_close(ds->ds_dir, ds);
2406 		VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2407 		    NULL, ds, &ds->ds_dir));
2408 
2409 		ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2410 	} while (snap = list_next(&pa->snap_list, snap));
2411 
2412 	/* change space accounting */
2413 	dsl_dir_diduse_space(odd, -pa->used, -pa->comp, -pa->uncomp, tx);
2414 	dsl_dir_diduse_space(dd, pa->used, pa->comp, pa->uncomp, tx);
2415 	origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2416 
2417 	/* log history record */
2418 	spa_history_internal_log(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2419 	    cr, "dataset = %llu", hds->ds_object);
2420 
2421 	dsl_dir_close(odd, FTAG);
2422 	kmem_free(name, MAXPATHLEN);
2423 }
2424 
2425 int
2426 dsl_dataset_promote(const char *name)
2427 {
2428 	dsl_dataset_t *ds;
2429 	dsl_dir_t *dd;
2430 	dsl_pool_t *dp;
2431 	dmu_object_info_t doi;
2432 	struct promotearg pa;
2433 	struct promotenode *snap;
2434 	uint64_t snap_obj;
2435 	uint64_t last_snap = 0;
2436 	int err;
2437 
2438 	err = dsl_dataset_hold(name, FTAG, &ds);
2439 	if (err)
2440 		return (err);
2441 	dd = ds->ds_dir;
2442 	dp = dd->dd_pool;
2443 
2444 	err = dmu_object_info(dp->dp_meta_objset,
2445 	    ds->ds_phys->ds_snapnames_zapobj, &doi);
2446 	if (err) {
2447 		dsl_dataset_rele(ds, FTAG);
2448 		return (err);
2449 	}
2450 
2451 	/*
2452 	 * We are going to inherit all the snapshots taken before our
2453 	 * origin (i.e., our new origin will be our parent's origin).
2454 	 * Take ownership of them so that we can rename them into our
2455 	 * namespace.
2456 	 */
2457 	pa.clone_origin = NULL;
2458 	list_create(&pa.snap_list,
2459 	    sizeof (struct promotenode), offsetof(struct promotenode, link));
2460 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2461 	ASSERT(dd->dd_phys->dd_origin_obj != 0);
2462 	snap_obj = dd->dd_phys->dd_origin_obj;
2463 	while (snap_obj) {
2464 		dsl_dataset_t *snapds;
2465 
2466 		/*
2467 		 * NB: this would be handled by the below check for
2468 		 * clone of a clone, but then we'd always own_obj() the
2469 		 * $ORIGIN, thus causing unnecessary EBUSYs.  We don't
2470 		 * need to set pa.clone_origin because the $ORIGIN has
2471 		 * no data to account for.
2472 		 */
2473 		if (dp->dp_origin_snap &&
2474 		    snap_obj == dp->dp_origin_snap->ds_object)
2475 			break;
2476 
2477 		err = dsl_dataset_own_obj(dp, snap_obj, 0, FTAG, &snapds);
2478 		if (err == ENOENT) {
2479 			/* lost race with snapshot destroy */
2480 			struct promotenode *last = list_tail(&pa.snap_list);
2481 			ASSERT(snap_obj != last->ds->ds_phys->ds_prev_snap_obj);
2482 			snap_obj = last->ds->ds_phys->ds_prev_snap_obj;
2483 			continue;
2484 		} else if (err) {
2485 			rw_exit(&dp->dp_config_rwlock);
2486 			goto out;
2487 		}
2488 
2489 		/*
2490 		 * We could be a clone of a clone.  If we reach our
2491 		 * parent's branch point, we're done.
2492 		 */
2493 		if (last_snap &&
2494 		    snapds->ds_phys->ds_next_snap_obj != last_snap) {
2495 			pa.clone_origin = snapds;
2496 			break;
2497 		}
2498 
2499 		snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2500 		snap->ds = snapds;
2501 		list_insert_tail(&pa.snap_list, snap);
2502 		last_snap = snap_obj;
2503 		snap_obj = snap->ds->ds_phys->ds_prev_snap_obj;
2504 	}
2505 	snap = list_head(&pa.snap_list);
2506 	ASSERT(snap != NULL);
2507 	err = dsl_dataset_hold_obj(dp,
2508 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &pa.old_head);
2509 	rw_exit(&dp->dp_config_rwlock);
2510 
2511 	if (err)
2512 		goto out;
2513 
2514 	/*
2515 	 * Add in 128x the snapnames zapobj size, since we will be moving
2516 	 * a bunch of snapnames to the promoted ds, and dirtying their
2517 	 * bonus buffers.
2518 	 */
2519 	err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2520 	    dsl_dataset_promote_sync, ds, &pa, 2 + 2 * doi.doi_physical_blks);
2521 
2522 	dsl_dataset_rele(pa.old_head, FTAG);
2523 out:
2524 	while ((snap = list_tail(&pa.snap_list)) != NULL) {
2525 		list_remove(&pa.snap_list, snap);
2526 		dsl_dataset_disown(snap->ds, FTAG);
2527 		kmem_free(snap, sizeof (struct promotenode));
2528 	}
2529 	list_destroy(&pa.snap_list);
2530 	if (pa.clone_origin)
2531 		dsl_dataset_disown(pa.clone_origin, FTAG);
2532 	dsl_dataset_rele(ds, FTAG);
2533 	return (err);
2534 }
2535 
2536 struct cloneswaparg {
2537 	dsl_dataset_t *cds; /* clone dataset */
2538 	dsl_dataset_t *ohds; /* origin's head dataset */
2539 	boolean_t force;
2540 	int64_t unused_refres_delta; /* change in unconsumed refreservation */
2541 };
2542 
2543 /* ARGSUSED */
2544 static int
2545 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
2546 {
2547 	struct cloneswaparg *csa = arg1;
2548 
2549 	/* they should both be heads */
2550 	if (dsl_dataset_is_snapshot(csa->cds) ||
2551 	    dsl_dataset_is_snapshot(csa->ohds))
2552 		return (EINVAL);
2553 
2554 	/* the branch point should be just before them */
2555 	if (csa->cds->ds_prev != csa->ohds->ds_prev)
2556 		return (EINVAL);
2557 
2558 	/* cds should be the clone */
2559 	if (csa->cds->ds_prev->ds_phys->ds_next_snap_obj !=
2560 	    csa->ohds->ds_object)
2561 		return (EINVAL);
2562 
2563 	/* the clone should be a child of the origin */
2564 	if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
2565 		return (EINVAL);
2566 
2567 	/* ohds shouldn't be modified unless 'force' */
2568 	if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
2569 		return (ETXTBSY);
2570 
2571 	/* adjust amount of any unconsumed refreservation */
2572 	csa->unused_refres_delta =
2573 	    (int64_t)MIN(csa->ohds->ds_reserved,
2574 	    csa->ohds->ds_phys->ds_unique_bytes) -
2575 	    (int64_t)MIN(csa->ohds->ds_reserved,
2576 	    csa->cds->ds_phys->ds_unique_bytes);
2577 
2578 	if (csa->unused_refres_delta > 0 &&
2579 	    csa->unused_refres_delta >
2580 	    dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
2581 		return (ENOSPC);
2582 
2583 	return (0);
2584 }
2585 
2586 /* ARGSUSED */
2587 static void
2588 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2589 {
2590 	struct cloneswaparg *csa = arg1;
2591 	dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
2592 	uint64_t itor = 0;
2593 	blkptr_t bp;
2594 	uint64_t unique = 0;
2595 	int err;
2596 
2597 	ASSERT(csa->cds->ds_reserved == 0);
2598 	ASSERT(csa->cds->ds_quota == csa->ohds->ds_quota);
2599 
2600 	dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
2601 	dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
2602 	dmu_buf_will_dirty(csa->cds->ds_prev->ds_dbuf, tx);
2603 
2604 	if (csa->cds->ds_user_ptr != NULL) {
2605 		csa->cds->ds_user_evict_func(csa->cds, csa->cds->ds_user_ptr);
2606 		csa->cds->ds_user_ptr = NULL;
2607 	}
2608 
2609 	if (csa->ohds->ds_user_ptr != NULL) {
2610 		csa->ohds->ds_user_evict_func(csa->ohds,
2611 		    csa->ohds->ds_user_ptr);
2612 		csa->ohds->ds_user_ptr = NULL;
2613 	}
2614 
2615 	/* compute unique space */
2616 	while ((err = bplist_iterate(&csa->cds->ds_deadlist,
2617 	    &itor, &bp)) == 0) {
2618 		if (bp.blk_birth > csa->cds->ds_prev->ds_phys->ds_prev_snap_txg)
2619 			unique += bp_get_dasize(dp->dp_spa, &bp);
2620 	}
2621 	VERIFY(err == ENOENT);
2622 
2623 	/* reset origin's unique bytes */
2624 	csa->cds->ds_prev->ds_phys->ds_unique_bytes = unique;
2625 
2626 	/* swap blkptrs */
2627 	{
2628 		blkptr_t tmp;
2629 		tmp = csa->ohds->ds_phys->ds_bp;
2630 		csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
2631 		csa->cds->ds_phys->ds_bp = tmp;
2632 	}
2633 
2634 	/* set dd_*_bytes */
2635 	{
2636 		int64_t dused, dcomp, duncomp;
2637 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2638 		uint64_t odl_used, odl_comp, odl_uncomp;
2639 
2640 		VERIFY(0 == bplist_space(&csa->cds->ds_deadlist, &cdl_used,
2641 		    &cdl_comp, &cdl_uncomp));
2642 		VERIFY(0 == bplist_space(&csa->ohds->ds_deadlist, &odl_used,
2643 		    &odl_comp, &odl_uncomp));
2644 		dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
2645 		    (csa->ohds->ds_phys->ds_used_bytes + odl_used);
2646 		dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
2647 		    (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
2648 		duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
2649 		    cdl_uncomp -
2650 		    (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2651 
2652 		dsl_dir_diduse_space(csa->ohds->ds_dir,
2653 		    dused, dcomp, duncomp, tx);
2654 		dsl_dir_diduse_space(csa->cds->ds_dir,
2655 		    -dused, -dcomp, -duncomp, tx);
2656 	}
2657 
2658 #define	SWITCH64(x, y) \
2659 	{ \
2660 		uint64_t __tmp = (x); \
2661 		(x) = (y); \
2662 		(y) = __tmp; \
2663 	}
2664 
2665 	/* swap ds_*_bytes */
2666 	SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
2667 	    csa->cds->ds_phys->ds_used_bytes);
2668 	SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
2669 	    csa->cds->ds_phys->ds_compressed_bytes);
2670 	SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
2671 	    csa->cds->ds_phys->ds_uncompressed_bytes);
2672 	SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
2673 	    csa->cds->ds_phys->ds_unique_bytes);
2674 
2675 	/* apply any parent delta for change in unconsumed refreservation */
2676 	dsl_dir_diduse_space(csa->ohds->ds_dir, csa->unused_refres_delta,
2677 	    0, 0, tx);
2678 
2679 	/* swap deadlists */
2680 	bplist_close(&csa->cds->ds_deadlist);
2681 	bplist_close(&csa->ohds->ds_deadlist);
2682 	SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
2683 	    csa->cds->ds_phys->ds_deadlist_obj);
2684 	VERIFY(0 == bplist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
2685 	    csa->cds->ds_phys->ds_deadlist_obj));
2686 	VERIFY(0 == bplist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
2687 	    csa->ohds->ds_phys->ds_deadlist_obj));
2688 }
2689 
2690 /*
2691  * Swap 'clone' with its origin head file system.  Used at the end
2692  * of "online recv" to swizzle the file system to the new version.
2693  */
2694 int
2695 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
2696     boolean_t force)
2697 {
2698 	struct cloneswaparg csa;
2699 	int error;
2700 
2701 	ASSERT(clone->ds_owner);
2702 	ASSERT(origin_head->ds_owner);
2703 retry:
2704 	/* Need exclusive access for the swap */
2705 	rw_enter(&clone->ds_rwlock, RW_WRITER);
2706 	if (!rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
2707 		rw_exit(&clone->ds_rwlock);
2708 		rw_enter(&origin_head->ds_rwlock, RW_WRITER);
2709 		if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
2710 			rw_exit(&origin_head->ds_rwlock);
2711 			goto retry;
2712 		}
2713 	}
2714 	csa.cds = clone;
2715 	csa.ohds = origin_head;
2716 	csa.force = force;
2717 	error = dsl_sync_task_do(clone->ds_dir->dd_pool,
2718 	    dsl_dataset_clone_swap_check,
2719 	    dsl_dataset_clone_swap_sync, &csa, NULL, 9);
2720 	return (error);
2721 }
2722 
2723 /*
2724  * Given a pool name and a dataset object number in that pool,
2725  * return the name of that dataset.
2726  */
2727 int
2728 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2729 {
2730 	spa_t *spa;
2731 	dsl_pool_t *dp;
2732 	dsl_dataset_t *ds;
2733 	int error;
2734 
2735 	if ((error = spa_open(pname, &spa, FTAG)) != 0)
2736 		return (error);
2737 	dp = spa_get_dsl(spa);
2738 	rw_enter(&dp->dp_config_rwlock, RW_READER);
2739 	if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
2740 		dsl_dataset_name(ds, buf);
2741 		dsl_dataset_rele(ds, FTAG);
2742 	}
2743 	rw_exit(&dp->dp_config_rwlock);
2744 	spa_close(spa, FTAG);
2745 
2746 	return (error);
2747 }
2748 
2749 int
2750 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2751     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2752 {
2753 	int error = 0;
2754 
2755 	ASSERT3S(asize, >, 0);
2756 
2757 	/*
2758 	 * *ref_rsrv is the portion of asize that will come from any
2759 	 * unconsumed refreservation space.
2760 	 */
2761 	*ref_rsrv = 0;
2762 
2763 	mutex_enter(&ds->ds_lock);
2764 	/*
2765 	 * Make a space adjustment for reserved bytes.
2766 	 */
2767 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2768 		ASSERT3U(*used, >=,
2769 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2770 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2771 		*ref_rsrv =
2772 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2773 	}
2774 
2775 	if (!check_quota || ds->ds_quota == 0) {
2776 		mutex_exit(&ds->ds_lock);
2777 		return (0);
2778 	}
2779 	/*
2780 	 * If they are requesting more space, and our current estimate
2781 	 * is over quota, they get to try again unless the actual
2782 	 * on-disk is over quota and there are no pending changes (which
2783 	 * may free up space for us).
2784 	 */
2785 	if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
2786 		if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
2787 			error = ERESTART;
2788 		else
2789 			error = EDQUOT;
2790 	}
2791 	mutex_exit(&ds->ds_lock);
2792 
2793 	return (error);
2794 }
2795 
2796 /* ARGSUSED */
2797 static int
2798 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
2799 {
2800 	dsl_dataset_t *ds = arg1;
2801 	uint64_t *quotap = arg2;
2802 	uint64_t new_quota = *quotap;
2803 
2804 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
2805 		return (ENOTSUP);
2806 
2807 	if (new_quota == 0)
2808 		return (0);
2809 
2810 	if (new_quota < ds->ds_phys->ds_used_bytes ||
2811 	    new_quota < ds->ds_reserved)
2812 		return (ENOSPC);
2813 
2814 	return (0);
2815 }
2816 
2817 /* ARGSUSED */
2818 void
2819 dsl_dataset_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
2820 {
2821 	dsl_dataset_t *ds = arg1;
2822 	uint64_t *quotap = arg2;
2823 	uint64_t new_quota = *quotap;
2824 
2825 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2826 
2827 	ds->ds_quota = new_quota;
2828 
2829 	dsl_prop_set_uint64_sync(ds->ds_dir, "refquota", new_quota, cr, tx);
2830 
2831 	spa_history_internal_log(LOG_DS_REFQUOTA, ds->ds_dir->dd_pool->dp_spa,
2832 	    tx, cr, "%lld dataset = %llu ",
2833 	    (longlong_t)new_quota, ds->ds_object);
2834 }
2835 
2836 int
2837 dsl_dataset_set_quota(const char *dsname, uint64_t quota)
2838 {
2839 	dsl_dataset_t *ds;
2840 	int err;
2841 
2842 	err = dsl_dataset_hold(dsname, FTAG, &ds);
2843 	if (err)
2844 		return (err);
2845 
2846 	if (quota != ds->ds_quota) {
2847 		/*
2848 		 * If someone removes a file, then tries to set the quota, we
2849 		 * want to make sure the file freeing takes effect.
2850 		 */
2851 		txg_wait_open(ds->ds_dir->dd_pool, 0);
2852 
2853 		err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2854 		    dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
2855 		    ds, &quota, 0);
2856 	}
2857 	dsl_dataset_rele(ds, FTAG);
2858 	return (err);
2859 }
2860 
2861 static int
2862 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
2863 {
2864 	dsl_dataset_t *ds = arg1;
2865 	uint64_t *reservationp = arg2;
2866 	uint64_t new_reservation = *reservationp;
2867 	int64_t delta;
2868 	uint64_t unique;
2869 
2870 	if (new_reservation > INT64_MAX)
2871 		return (EOVERFLOW);
2872 
2873 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
2874 	    SPA_VERSION_REFRESERVATION)
2875 		return (ENOTSUP);
2876 
2877 	if (dsl_dataset_is_snapshot(ds))
2878 		return (EINVAL);
2879 
2880 	/*
2881 	 * If we are doing the preliminary check in open context, the
2882 	 * space estimates may be inaccurate.
2883 	 */
2884 	if (!dmu_tx_is_syncing(tx))
2885 		return (0);
2886 
2887 	mutex_enter(&ds->ds_lock);
2888 	unique = dsl_dataset_unique(ds);
2889 	delta = MAX(unique, new_reservation) - MAX(unique, ds->ds_reserved);
2890 	mutex_exit(&ds->ds_lock);
2891 
2892 	if (delta > 0 &&
2893 	    delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2894 		return (ENOSPC);
2895 	if (delta > 0 && ds->ds_quota > 0 &&
2896 	    new_reservation > ds->ds_quota)
2897 		return (ENOSPC);
2898 
2899 	return (0);
2900 }
2901 
2902 /* ARGSUSED */
2903 static void
2904 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, cred_t *cr,
2905     dmu_tx_t *tx)
2906 {
2907 	dsl_dataset_t *ds = arg1;
2908 	uint64_t *reservationp = arg2;
2909 	uint64_t new_reservation = *reservationp;
2910 	uint64_t unique;
2911 	int64_t delta;
2912 
2913 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2914 
2915 	mutex_enter(&ds->ds_lock);
2916 	unique = dsl_dataset_unique(ds);
2917 	delta = MAX(0, (int64_t)(new_reservation - unique)) -
2918 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2919 	ds->ds_reserved = new_reservation;
2920 	mutex_exit(&ds->ds_lock);
2921 
2922 	dsl_prop_set_uint64_sync(ds->ds_dir, "refreservation",
2923 	    new_reservation, cr, tx);
2924 
2925 	dsl_dir_diduse_space(ds->ds_dir, delta, 0, 0, tx);
2926 
2927 	spa_history_internal_log(LOG_DS_REFRESERV,
2928 	    ds->ds_dir->dd_pool->dp_spa, tx, cr, "%lld dataset = %llu",
2929 	    (longlong_t)new_reservation,
2930 	    ds->ds_dir->dd_phys->dd_head_dataset_obj);
2931 }
2932 
2933 int
2934 dsl_dataset_set_reservation(const char *dsname, uint64_t reservation)
2935 {
2936 	dsl_dataset_t *ds;
2937 	int err;
2938 
2939 	err = dsl_dataset_hold(dsname, FTAG, &ds);
2940 	if (err)
2941 		return (err);
2942 
2943 	err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2944 	    dsl_dataset_set_reservation_check,
2945 	    dsl_dataset_set_reservation_sync, ds, &reservation, 0);
2946 	dsl_dataset_rele(ds, FTAG);
2947 	return (err);
2948 }
2949