xref: /titanic_50/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision c0e7977a434048a8bc7386ea0e8befaa77a646cf)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013 by Delphix. All rights reserved.
24  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25  */
26 
27 #include <sys/dmu_objset.h>
28 #include <sys/dsl_dataset.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_prop.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_traverse.h>
33 #include <sys/dmu_impl.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/zfeature.h>
39 #include <sys/unique.h>
40 #include <sys/zfs_context.h>
41 #include <sys/zfs_ioctl.h>
42 #include <sys/spa.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_onexit.h>
45 #include <sys/zvol.h>
46 #include <sys/dsl_scan.h>
47 #include <sys/dsl_deadlist.h>
48 #include <sys/dsl_destroy.h>
49 #include <sys/dsl_userhold.h>
50 
51 #define	SWITCH64(x, y) \
52 	{ \
53 		uint64_t __tmp = (x); \
54 		(x) = (y); \
55 		(y) = __tmp; \
56 	}
57 
58 #define	DS_REF_MAX	(1ULL << 62)
59 
60 #define	DSL_DEADLIST_BLOCKSIZE	SPA_MAXBLOCKSIZE
61 
62 /*
63  * Figure out how much of this delta should be propogated to the dsl_dir
64  * layer.  If there's a refreservation, that space has already been
65  * partially accounted for in our ancestors.
66  */
67 static int64_t
68 parent_delta(dsl_dataset_t *ds, int64_t delta)
69 {
70 	uint64_t old_bytes, new_bytes;
71 
72 	if (ds->ds_reserved == 0)
73 		return (delta);
74 
75 	old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
76 	new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
77 
78 	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
79 	return (new_bytes - old_bytes);
80 }
81 
82 void
83 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
84 {
85 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
86 	int compressed = BP_GET_PSIZE(bp);
87 	int uncompressed = BP_GET_UCSIZE(bp);
88 	int64_t delta;
89 
90 	dprintf_bp(bp, "ds=%p", ds);
91 
92 	ASSERT(dmu_tx_is_syncing(tx));
93 	/* It could have been compressed away to nothing */
94 	if (BP_IS_HOLE(bp))
95 		return;
96 	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
97 	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
98 	if (ds == NULL) {
99 		dsl_pool_mos_diduse_space(tx->tx_pool,
100 		    used, compressed, uncompressed);
101 		return;
102 	}
103 
104 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
105 	mutex_enter(&ds->ds_lock);
106 	delta = parent_delta(ds, used);
107 	ds->ds_phys->ds_referenced_bytes += used;
108 	ds->ds_phys->ds_compressed_bytes += compressed;
109 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
110 	ds->ds_phys->ds_unique_bytes += used;
111 	mutex_exit(&ds->ds_lock);
112 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
113 	    compressed, uncompressed, tx);
114 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
115 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
116 }
117 
118 int
119 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
120     boolean_t async)
121 {
122 	if (BP_IS_HOLE(bp))
123 		return (0);
124 
125 	ASSERT(dmu_tx_is_syncing(tx));
126 	ASSERT(bp->blk_birth <= tx->tx_txg);
127 
128 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
129 	int compressed = BP_GET_PSIZE(bp);
130 	int uncompressed = BP_GET_UCSIZE(bp);
131 
132 	ASSERT(used > 0);
133 	if (ds == NULL) {
134 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
135 		dsl_pool_mos_diduse_space(tx->tx_pool,
136 		    -used, -compressed, -uncompressed);
137 		return (used);
138 	}
139 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
140 
141 	ASSERT(!dsl_dataset_is_snapshot(ds));
142 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
143 
144 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
145 		int64_t delta;
146 
147 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
148 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
149 
150 		mutex_enter(&ds->ds_lock);
151 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
152 		    !DS_UNIQUE_IS_ACCURATE(ds));
153 		delta = parent_delta(ds, -used);
154 		ds->ds_phys->ds_unique_bytes -= used;
155 		mutex_exit(&ds->ds_lock);
156 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
157 		    delta, -compressed, -uncompressed, tx);
158 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
159 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
160 	} else {
161 		dprintf_bp(bp, "putting on dead list: %s", "");
162 		if (async) {
163 			/*
164 			 * We are here as part of zio's write done callback,
165 			 * which means we're a zio interrupt thread.  We can't
166 			 * call dsl_deadlist_insert() now because it may block
167 			 * waiting for I/O.  Instead, put bp on the deferred
168 			 * queue and let dsl_pool_sync() finish the job.
169 			 */
170 			bplist_append(&ds->ds_pending_deadlist, bp);
171 		} else {
172 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
173 		}
174 		ASSERT3U(ds->ds_prev->ds_object, ==,
175 		    ds->ds_phys->ds_prev_snap_obj);
176 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
177 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
178 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
179 		    ds->ds_object && bp->blk_birth >
180 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
181 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
182 			mutex_enter(&ds->ds_prev->ds_lock);
183 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
184 			mutex_exit(&ds->ds_prev->ds_lock);
185 		}
186 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
187 			dsl_dir_transfer_space(ds->ds_dir, used,
188 			    DD_USED_HEAD, DD_USED_SNAP, tx);
189 		}
190 	}
191 	mutex_enter(&ds->ds_lock);
192 	ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
193 	ds->ds_phys->ds_referenced_bytes -= used;
194 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
195 	ds->ds_phys->ds_compressed_bytes -= compressed;
196 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
197 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
198 	mutex_exit(&ds->ds_lock);
199 
200 	return (used);
201 }
202 
203 uint64_t
204 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
205 {
206 	uint64_t trysnap = 0;
207 
208 	if (ds == NULL)
209 		return (0);
210 	/*
211 	 * The snapshot creation could fail, but that would cause an
212 	 * incorrect FALSE return, which would only result in an
213 	 * overestimation of the amount of space that an operation would
214 	 * consume, which is OK.
215 	 *
216 	 * There's also a small window where we could miss a pending
217 	 * snapshot, because we could set the sync task in the quiescing
218 	 * phase.  So this should only be used as a guess.
219 	 */
220 	if (ds->ds_trysnap_txg >
221 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
222 		trysnap = ds->ds_trysnap_txg;
223 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
224 }
225 
226 boolean_t
227 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
228     uint64_t blk_birth)
229 {
230 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
231 		return (B_FALSE);
232 
233 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
234 
235 	return (B_TRUE);
236 }
237 
238 /* ARGSUSED */
239 static void
240 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
241 {
242 	dsl_dataset_t *ds = dsv;
243 
244 	ASSERT(ds->ds_owner == NULL);
245 
246 	unique_remove(ds->ds_fsid_guid);
247 
248 	if (ds->ds_objset != NULL)
249 		dmu_objset_evict(ds->ds_objset);
250 
251 	if (ds->ds_prev) {
252 		dsl_dataset_rele(ds->ds_prev, ds);
253 		ds->ds_prev = NULL;
254 	}
255 
256 	bplist_destroy(&ds->ds_pending_deadlist);
257 	if (ds->ds_phys->ds_deadlist_obj != 0)
258 		dsl_deadlist_close(&ds->ds_deadlist);
259 	if (ds->ds_dir)
260 		dsl_dir_rele(ds->ds_dir, ds);
261 
262 	ASSERT(!list_link_active(&ds->ds_synced_link));
263 
264 	mutex_destroy(&ds->ds_lock);
265 	mutex_destroy(&ds->ds_opening_lock);
266 	refcount_destroy(&ds->ds_longholds);
267 
268 	kmem_free(ds, sizeof (dsl_dataset_t));
269 }
270 
271 int
272 dsl_dataset_get_snapname(dsl_dataset_t *ds)
273 {
274 	dsl_dataset_phys_t *headphys;
275 	int err;
276 	dmu_buf_t *headdbuf;
277 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
278 	objset_t *mos = dp->dp_meta_objset;
279 
280 	if (ds->ds_snapname[0])
281 		return (0);
282 	if (ds->ds_phys->ds_next_snap_obj == 0)
283 		return (0);
284 
285 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
286 	    FTAG, &headdbuf);
287 	if (err != 0)
288 		return (err);
289 	headphys = headdbuf->db_data;
290 	err = zap_value_search(dp->dp_meta_objset,
291 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
292 	dmu_buf_rele(headdbuf, FTAG);
293 	return (err);
294 }
295 
296 int
297 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
298 {
299 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
300 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
301 	matchtype_t mt;
302 	int err;
303 
304 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
305 		mt = MT_FIRST;
306 	else
307 		mt = MT_EXACT;
308 
309 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
310 	    value, mt, NULL, 0, NULL);
311 	if (err == ENOTSUP && mt == MT_FIRST)
312 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
313 	return (err);
314 }
315 
316 int
317 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx)
318 {
319 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
320 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
321 	matchtype_t mt;
322 	int err;
323 
324 	dsl_dir_snap_cmtime_update(ds->ds_dir);
325 
326 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
327 		mt = MT_FIRST;
328 	else
329 		mt = MT_EXACT;
330 
331 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
332 	if (err == ENOTSUP && mt == MT_FIRST)
333 		err = zap_remove(mos, snapobj, name, tx);
334 	return (err);
335 }
336 
337 int
338 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
339     dsl_dataset_t **dsp)
340 {
341 	objset_t *mos = dp->dp_meta_objset;
342 	dmu_buf_t *dbuf;
343 	dsl_dataset_t *ds;
344 	int err;
345 	dmu_object_info_t doi;
346 
347 	ASSERT(dsl_pool_config_held(dp));
348 
349 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
350 	if (err != 0)
351 		return (err);
352 
353 	/* Make sure dsobj has the correct object type. */
354 	dmu_object_info_from_db(dbuf, &doi);
355 	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
356 		dmu_buf_rele(dbuf, tag);
357 		return (SET_ERROR(EINVAL));
358 	}
359 
360 	ds = dmu_buf_get_user(dbuf);
361 	if (ds == NULL) {
362 		dsl_dataset_t *winner = NULL;
363 
364 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
365 		ds->ds_dbuf = dbuf;
366 		ds->ds_object = dsobj;
367 		ds->ds_phys = dbuf->db_data;
368 
369 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
370 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
371 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
372 		refcount_create(&ds->ds_longholds);
373 
374 		bplist_create(&ds->ds_pending_deadlist);
375 		dsl_deadlist_open(&ds->ds_deadlist,
376 		    mos, ds->ds_phys->ds_deadlist_obj);
377 
378 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
379 		    offsetof(dmu_sendarg_t, dsa_link));
380 
381 		if (err == 0) {
382 			err = dsl_dir_hold_obj(dp,
383 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
384 		}
385 		if (err != 0) {
386 			mutex_destroy(&ds->ds_lock);
387 			mutex_destroy(&ds->ds_opening_lock);
388 			refcount_destroy(&ds->ds_longholds);
389 			bplist_destroy(&ds->ds_pending_deadlist);
390 			dsl_deadlist_close(&ds->ds_deadlist);
391 			kmem_free(ds, sizeof (dsl_dataset_t));
392 			dmu_buf_rele(dbuf, tag);
393 			return (err);
394 		}
395 
396 		if (!dsl_dataset_is_snapshot(ds)) {
397 			ds->ds_snapname[0] = '\0';
398 			if (ds->ds_phys->ds_prev_snap_obj != 0) {
399 				err = dsl_dataset_hold_obj(dp,
400 				    ds->ds_phys->ds_prev_snap_obj,
401 				    ds, &ds->ds_prev);
402 			}
403 		} else {
404 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
405 				err = dsl_dataset_get_snapname(ds);
406 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
407 				err = zap_count(
408 				    ds->ds_dir->dd_pool->dp_meta_objset,
409 				    ds->ds_phys->ds_userrefs_obj,
410 				    &ds->ds_userrefs);
411 			}
412 		}
413 
414 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
415 			err = dsl_prop_get_int_ds(ds,
416 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
417 			    &ds->ds_reserved);
418 			if (err == 0) {
419 				err = dsl_prop_get_int_ds(ds,
420 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
421 				    &ds->ds_quota);
422 			}
423 		} else {
424 			ds->ds_reserved = ds->ds_quota = 0;
425 		}
426 
427 		if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
428 		    &ds->ds_phys, dsl_dataset_evict)) != NULL) {
429 			bplist_destroy(&ds->ds_pending_deadlist);
430 			dsl_deadlist_close(&ds->ds_deadlist);
431 			if (ds->ds_prev)
432 				dsl_dataset_rele(ds->ds_prev, ds);
433 			dsl_dir_rele(ds->ds_dir, ds);
434 			mutex_destroy(&ds->ds_lock);
435 			mutex_destroy(&ds->ds_opening_lock);
436 			refcount_destroy(&ds->ds_longholds);
437 			kmem_free(ds, sizeof (dsl_dataset_t));
438 			if (err != 0) {
439 				dmu_buf_rele(dbuf, tag);
440 				return (err);
441 			}
442 			ds = winner;
443 		} else {
444 			ds->ds_fsid_guid =
445 			    unique_insert(ds->ds_phys->ds_fsid_guid);
446 		}
447 	}
448 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
449 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
450 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
451 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
452 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
453 	*dsp = ds;
454 	return (0);
455 }
456 
457 int
458 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
459     void *tag, dsl_dataset_t **dsp)
460 {
461 	dsl_dir_t *dd;
462 	const char *snapname;
463 	uint64_t obj;
464 	int err = 0;
465 
466 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
467 	if (err != 0)
468 		return (err);
469 
470 	ASSERT(dsl_pool_config_held(dp));
471 	obj = dd->dd_phys->dd_head_dataset_obj;
472 	if (obj != 0)
473 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
474 	else
475 		err = SET_ERROR(ENOENT);
476 
477 	/* we may be looking for a snapshot */
478 	if (err == 0 && snapname != NULL) {
479 		dsl_dataset_t *ds;
480 
481 		if (*snapname++ != '@') {
482 			dsl_dataset_rele(*dsp, tag);
483 			dsl_dir_rele(dd, FTAG);
484 			return (SET_ERROR(ENOENT));
485 		}
486 
487 		dprintf("looking for snapshot '%s'\n", snapname);
488 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
489 		if (err == 0)
490 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
491 		dsl_dataset_rele(*dsp, tag);
492 
493 		if (err == 0) {
494 			mutex_enter(&ds->ds_lock);
495 			if (ds->ds_snapname[0] == 0)
496 				(void) strlcpy(ds->ds_snapname, snapname,
497 				    sizeof (ds->ds_snapname));
498 			mutex_exit(&ds->ds_lock);
499 			*dsp = ds;
500 		}
501 	}
502 
503 	dsl_dir_rele(dd, FTAG);
504 	return (err);
505 }
506 
507 int
508 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
509     void *tag, dsl_dataset_t **dsp)
510 {
511 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
512 	if (err != 0)
513 		return (err);
514 	if (!dsl_dataset_tryown(*dsp, tag)) {
515 		dsl_dataset_rele(*dsp, tag);
516 		*dsp = NULL;
517 		return (SET_ERROR(EBUSY));
518 	}
519 	return (0);
520 }
521 
522 int
523 dsl_dataset_own(dsl_pool_t *dp, const char *name,
524     void *tag, dsl_dataset_t **dsp)
525 {
526 	int err = dsl_dataset_hold(dp, name, tag, dsp);
527 	if (err != 0)
528 		return (err);
529 	if (!dsl_dataset_tryown(*dsp, tag)) {
530 		dsl_dataset_rele(*dsp, tag);
531 		return (SET_ERROR(EBUSY));
532 	}
533 	return (0);
534 }
535 
536 /*
537  * See the comment above dsl_pool_hold() for details.  In summary, a long
538  * hold is used to prevent destruction of a dataset while the pool hold
539  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
540  *
541  * The dataset and pool must be held when this function is called.  After it
542  * is called, the pool hold may be released while the dataset is still held
543  * and accessed.
544  */
545 void
546 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
547 {
548 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
549 	(void) refcount_add(&ds->ds_longholds, tag);
550 }
551 
552 void
553 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
554 {
555 	(void) refcount_remove(&ds->ds_longholds, tag);
556 }
557 
558 /* Return B_TRUE if there are any long holds on this dataset. */
559 boolean_t
560 dsl_dataset_long_held(dsl_dataset_t *ds)
561 {
562 	return (!refcount_is_zero(&ds->ds_longholds));
563 }
564 
565 void
566 dsl_dataset_name(dsl_dataset_t *ds, char *name)
567 {
568 	if (ds == NULL) {
569 		(void) strcpy(name, "mos");
570 	} else {
571 		dsl_dir_name(ds->ds_dir, name);
572 		VERIFY0(dsl_dataset_get_snapname(ds));
573 		if (ds->ds_snapname[0]) {
574 			(void) strcat(name, "@");
575 			/*
576 			 * We use a "recursive" mutex so that we
577 			 * can call dprintf_ds() with ds_lock held.
578 			 */
579 			if (!MUTEX_HELD(&ds->ds_lock)) {
580 				mutex_enter(&ds->ds_lock);
581 				(void) strcat(name, ds->ds_snapname);
582 				mutex_exit(&ds->ds_lock);
583 			} else {
584 				(void) strcat(name, ds->ds_snapname);
585 			}
586 		}
587 	}
588 }
589 
590 void
591 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
592 {
593 	dmu_buf_rele(ds->ds_dbuf, tag);
594 }
595 
596 void
597 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
598 {
599 	ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
600 
601 	mutex_enter(&ds->ds_lock);
602 	ds->ds_owner = NULL;
603 	mutex_exit(&ds->ds_lock);
604 	dsl_dataset_long_rele(ds, tag);
605 	if (ds->ds_dbuf != NULL)
606 		dsl_dataset_rele(ds, tag);
607 	else
608 		dsl_dataset_evict(NULL, ds);
609 }
610 
611 boolean_t
612 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
613 {
614 	boolean_t gotit = FALSE;
615 
616 	mutex_enter(&ds->ds_lock);
617 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
618 		ds->ds_owner = tag;
619 		dsl_dataset_long_hold(ds, tag);
620 		gotit = TRUE;
621 	}
622 	mutex_exit(&ds->ds_lock);
623 	return (gotit);
624 }
625 
626 uint64_t
627 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
628     uint64_t flags, dmu_tx_t *tx)
629 {
630 	dsl_pool_t *dp = dd->dd_pool;
631 	dmu_buf_t *dbuf;
632 	dsl_dataset_phys_t *dsphys;
633 	uint64_t dsobj;
634 	objset_t *mos = dp->dp_meta_objset;
635 
636 	if (origin == NULL)
637 		origin = dp->dp_origin_snap;
638 
639 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
640 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
641 	ASSERT(dmu_tx_is_syncing(tx));
642 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
643 
644 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
645 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
646 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
647 	dmu_buf_will_dirty(dbuf, tx);
648 	dsphys = dbuf->db_data;
649 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
650 	dsphys->ds_dir_obj = dd->dd_object;
651 	dsphys->ds_flags = flags;
652 	dsphys->ds_fsid_guid = unique_create();
653 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
654 	    sizeof (dsphys->ds_guid));
655 	dsphys->ds_snapnames_zapobj =
656 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
657 	    DMU_OT_NONE, 0, tx);
658 	dsphys->ds_creation_time = gethrestime_sec();
659 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
660 
661 	if (origin == NULL) {
662 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
663 	} else {
664 		dsl_dataset_t *ohds; /* head of the origin snapshot */
665 
666 		dsphys->ds_prev_snap_obj = origin->ds_object;
667 		dsphys->ds_prev_snap_txg =
668 		    origin->ds_phys->ds_creation_txg;
669 		dsphys->ds_referenced_bytes =
670 		    origin->ds_phys->ds_referenced_bytes;
671 		dsphys->ds_compressed_bytes =
672 		    origin->ds_phys->ds_compressed_bytes;
673 		dsphys->ds_uncompressed_bytes =
674 		    origin->ds_phys->ds_uncompressed_bytes;
675 		dsphys->ds_bp = origin->ds_phys->ds_bp;
676 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
677 
678 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
679 		origin->ds_phys->ds_num_children++;
680 
681 		VERIFY0(dsl_dataset_hold_obj(dp,
682 		    origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
683 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
684 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
685 		dsl_dataset_rele(ohds, FTAG);
686 
687 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
688 			if (origin->ds_phys->ds_next_clones_obj == 0) {
689 				origin->ds_phys->ds_next_clones_obj =
690 				    zap_create(mos,
691 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
692 			}
693 			VERIFY0(zap_add_int(mos,
694 			    origin->ds_phys->ds_next_clones_obj, dsobj, tx));
695 		}
696 
697 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
698 		dd->dd_phys->dd_origin_obj = origin->ds_object;
699 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
700 			if (origin->ds_dir->dd_phys->dd_clones == 0) {
701 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
702 				origin->ds_dir->dd_phys->dd_clones =
703 				    zap_create(mos,
704 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
705 			}
706 			VERIFY0(zap_add_int(mos,
707 			    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
708 		}
709 	}
710 
711 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
712 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
713 
714 	dmu_buf_rele(dbuf, FTAG);
715 
716 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
717 	dd->dd_phys->dd_head_dataset_obj = dsobj;
718 
719 	return (dsobj);
720 }
721 
722 static void
723 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
724 {
725 	objset_t *os;
726 
727 	VERIFY0(dmu_objset_from_ds(ds, &os));
728 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
729 	dsl_dataset_dirty(ds, tx);
730 }
731 
732 uint64_t
733 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
734     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
735 {
736 	dsl_pool_t *dp = pdd->dd_pool;
737 	uint64_t dsobj, ddobj;
738 	dsl_dir_t *dd;
739 
740 	ASSERT(dmu_tx_is_syncing(tx));
741 	ASSERT(lastname[0] != '@');
742 
743 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
744 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
745 
746 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
747 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
748 
749 	dsl_deleg_set_create_perms(dd, tx, cr);
750 
751 	dsl_dir_rele(dd, FTAG);
752 
753 	/*
754 	 * If we are creating a clone, make sure we zero out any stale
755 	 * data from the origin snapshots zil header.
756 	 */
757 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
758 		dsl_dataset_t *ds;
759 
760 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
761 		dsl_dataset_zero_zil(ds, tx);
762 		dsl_dataset_rele(ds, FTAG);
763 	}
764 
765 	return (dsobj);
766 }
767 
768 /*
769  * The unique space in the head dataset can be calculated by subtracting
770  * the space used in the most recent snapshot, that is still being used
771  * in this file system, from the space currently in use.  To figure out
772  * the space in the most recent snapshot still in use, we need to take
773  * the total space used in the snapshot and subtract out the space that
774  * has been freed up since the snapshot was taken.
775  */
776 void
777 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
778 {
779 	uint64_t mrs_used;
780 	uint64_t dlused, dlcomp, dluncomp;
781 
782 	ASSERT(!dsl_dataset_is_snapshot(ds));
783 
784 	if (ds->ds_phys->ds_prev_snap_obj != 0)
785 		mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
786 	else
787 		mrs_used = 0;
788 
789 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
790 
791 	ASSERT3U(dlused, <=, mrs_used);
792 	ds->ds_phys->ds_unique_bytes =
793 	    ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
794 
795 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
796 	    SPA_VERSION_UNIQUE_ACCURATE)
797 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
798 }
799 
800 void
801 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
802     dmu_tx_t *tx)
803 {
804 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
805 	uint64_t count;
806 	int err;
807 
808 	ASSERT(ds->ds_phys->ds_num_children >= 2);
809 	err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
810 	/*
811 	 * The err should not be ENOENT, but a bug in a previous version
812 	 * of the code could cause upgrade_clones_cb() to not set
813 	 * ds_next_snap_obj when it should, leading to a missing entry.
814 	 * If we knew that the pool was created after
815 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
816 	 * ENOENT.  However, at least we can check that we don't have
817 	 * too many entries in the next_clones_obj even after failing to
818 	 * remove this one.
819 	 */
820 	if (err != ENOENT)
821 		VERIFY0(err);
822 	ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
823 	    &count));
824 	ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
825 }
826 
827 
828 blkptr_t *
829 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
830 {
831 	return (&ds->ds_phys->ds_bp);
832 }
833 
834 void
835 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
836 {
837 	ASSERT(dmu_tx_is_syncing(tx));
838 	/* If it's the meta-objset, set dp_meta_rootbp */
839 	if (ds == NULL) {
840 		tx->tx_pool->dp_meta_rootbp = *bp;
841 	} else {
842 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
843 		ds->ds_phys->ds_bp = *bp;
844 	}
845 }
846 
847 spa_t *
848 dsl_dataset_get_spa(dsl_dataset_t *ds)
849 {
850 	return (ds->ds_dir->dd_pool->dp_spa);
851 }
852 
853 void
854 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
855 {
856 	dsl_pool_t *dp;
857 
858 	if (ds == NULL) /* this is the meta-objset */
859 		return;
860 
861 	ASSERT(ds->ds_objset != NULL);
862 
863 	if (ds->ds_phys->ds_next_snap_obj != 0)
864 		panic("dirtying snapshot!");
865 
866 	dp = ds->ds_dir->dd_pool;
867 
868 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
869 		/* up the hold count until we can be written out */
870 		dmu_buf_add_ref(ds->ds_dbuf, ds);
871 	}
872 }
873 
874 boolean_t
875 dsl_dataset_is_dirty(dsl_dataset_t *ds)
876 {
877 	for (int t = 0; t < TXG_SIZE; t++) {
878 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
879 		    ds, t))
880 			return (B_TRUE);
881 	}
882 	return (B_FALSE);
883 }
884 
885 static int
886 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
887 {
888 	uint64_t asize;
889 
890 	if (!dmu_tx_is_syncing(tx))
891 		return (0);
892 
893 	/*
894 	 * If there's an fs-only reservation, any blocks that might become
895 	 * owned by the snapshot dataset must be accommodated by space
896 	 * outside of the reservation.
897 	 */
898 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
899 	asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
900 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
901 		return (SET_ERROR(ENOSPC));
902 
903 	/*
904 	 * Propagate any reserved space for this snapshot to other
905 	 * snapshot checks in this sync group.
906 	 */
907 	if (asize > 0)
908 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
909 
910 	return (0);
911 }
912 
913 typedef struct dsl_dataset_snapshot_arg {
914 	nvlist_t *ddsa_snaps;
915 	nvlist_t *ddsa_props;
916 	nvlist_t *ddsa_errors;
917 } dsl_dataset_snapshot_arg_t;
918 
919 int
920 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
921     dmu_tx_t *tx, boolean_t recv)
922 {
923 	int error;
924 	uint64_t value;
925 
926 	ds->ds_trysnap_txg = tx->tx_txg;
927 
928 	if (!dmu_tx_is_syncing(tx))
929 		return (0);
930 
931 	/*
932 	 * We don't allow multiple snapshots of the same txg.  If there
933 	 * is already one, try again.
934 	 */
935 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
936 		return (SET_ERROR(EAGAIN));
937 
938 	/*
939 	 * Check for conflicting snapshot name.
940 	 */
941 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
942 	if (error == 0)
943 		return (SET_ERROR(EEXIST));
944 	if (error != ENOENT)
945 		return (error);
946 
947 	/*
948 	 * We don't allow taking snapshots of inconsistent datasets, such as
949 	 * those into which we are currently receiving.  However, if we are
950 	 * creating this snapshot as part of a receive, this check will be
951 	 * executed atomically with respect to the completion of the receive
952 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
953 	 * case we ignore this, knowing it will be fixed up for us shortly in
954 	 * dmu_recv_end_sync().
955 	 */
956 	if (!recv && DS_IS_INCONSISTENT(ds))
957 		return (SET_ERROR(EBUSY));
958 
959 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
960 	if (error != 0)
961 		return (error);
962 
963 	return (0);
964 }
965 
966 static int
967 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
968 {
969 	dsl_dataset_snapshot_arg_t *ddsa = arg;
970 	dsl_pool_t *dp = dmu_tx_pool(tx);
971 	nvpair_t *pair;
972 	int rv = 0;
973 
974 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
975 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
976 		int error = 0;
977 		dsl_dataset_t *ds;
978 		char *name, *atp;
979 		char dsname[MAXNAMELEN];
980 
981 		name = nvpair_name(pair);
982 		if (strlen(name) >= MAXNAMELEN)
983 			error = SET_ERROR(ENAMETOOLONG);
984 		if (error == 0) {
985 			atp = strchr(name, '@');
986 			if (atp == NULL)
987 				error = SET_ERROR(EINVAL);
988 			if (error == 0)
989 				(void) strlcpy(dsname, name, atp - name + 1);
990 		}
991 		if (error == 0)
992 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
993 		if (error == 0) {
994 			error = dsl_dataset_snapshot_check_impl(ds,
995 			    atp + 1, tx, B_FALSE);
996 			dsl_dataset_rele(ds, FTAG);
997 		}
998 
999 		if (error != 0) {
1000 			if (ddsa->ddsa_errors != NULL) {
1001 				fnvlist_add_int32(ddsa->ddsa_errors,
1002 				    name, error);
1003 			}
1004 			rv = error;
1005 		}
1006 	}
1007 	return (rv);
1008 }
1009 
1010 void
1011 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1012     dmu_tx_t *tx)
1013 {
1014 	static zil_header_t zero_zil;
1015 
1016 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1017 	dmu_buf_t *dbuf;
1018 	dsl_dataset_phys_t *dsphys;
1019 	uint64_t dsobj, crtxg;
1020 	objset_t *mos = dp->dp_meta_objset;
1021 	objset_t *os;
1022 
1023 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1024 
1025 	/*
1026 	 * If we are on an old pool, the zil must not be active, in which
1027 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1028 	 */
1029 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1030 	    dmu_objset_from_ds(ds, &os) != 0 ||
1031 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1032 	    sizeof (zero_zil)) == 0);
1033 
1034 
1035 	/*
1036 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1037 	 */
1038 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1039 		crtxg = 1;
1040 	else
1041 		crtxg = tx->tx_txg;
1042 
1043 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1044 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1045 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1046 	dmu_buf_will_dirty(dbuf, tx);
1047 	dsphys = dbuf->db_data;
1048 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1049 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1050 	dsphys->ds_fsid_guid = unique_create();
1051 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1052 	    sizeof (dsphys->ds_guid));
1053 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1054 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1055 	dsphys->ds_next_snap_obj = ds->ds_object;
1056 	dsphys->ds_num_children = 1;
1057 	dsphys->ds_creation_time = gethrestime_sec();
1058 	dsphys->ds_creation_txg = crtxg;
1059 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1060 	dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1061 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1062 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1063 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1064 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1065 	dmu_buf_rele(dbuf, FTAG);
1066 
1067 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1068 	if (ds->ds_prev) {
1069 		uint64_t next_clones_obj =
1070 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1071 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1072 		    ds->ds_object ||
1073 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1074 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1075 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1076 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1077 			    ds->ds_prev->ds_phys->ds_creation_txg);
1078 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1079 		} else if (next_clones_obj != 0) {
1080 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1081 			    dsphys->ds_next_snap_obj, tx);
1082 			VERIFY0(zap_add_int(mos,
1083 			    next_clones_obj, dsobj, tx));
1084 		}
1085 	}
1086 
1087 	/*
1088 	 * If we have a reference-reservation on this dataset, we will
1089 	 * need to increase the amount of refreservation being charged
1090 	 * since our unique space is going to zero.
1091 	 */
1092 	if (ds->ds_reserved) {
1093 		int64_t delta;
1094 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1095 		delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1096 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1097 		    delta, 0, 0, tx);
1098 	}
1099 
1100 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1101 	ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1102 	    UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1103 	dsl_deadlist_close(&ds->ds_deadlist);
1104 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1105 	dsl_deadlist_add_key(&ds->ds_deadlist,
1106 	    ds->ds_phys->ds_prev_snap_txg, tx);
1107 
1108 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1109 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1110 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1111 	ds->ds_phys->ds_unique_bytes = 0;
1112 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1113 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1114 
1115 	VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1116 	    snapname, 8, 1, &dsobj, tx));
1117 
1118 	if (ds->ds_prev)
1119 		dsl_dataset_rele(ds->ds_prev, ds);
1120 	VERIFY0(dsl_dataset_hold_obj(dp,
1121 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1122 
1123 	dsl_scan_ds_snapshotted(ds, tx);
1124 
1125 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1126 
1127 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1128 }
1129 
1130 static void
1131 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1132 {
1133 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1134 	dsl_pool_t *dp = dmu_tx_pool(tx);
1135 	nvpair_t *pair;
1136 
1137 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1138 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1139 		dsl_dataset_t *ds;
1140 		char *name, *atp;
1141 		char dsname[MAXNAMELEN];
1142 
1143 		name = nvpair_name(pair);
1144 		atp = strchr(name, '@');
1145 		(void) strlcpy(dsname, name, atp - name + 1);
1146 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1147 
1148 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1149 		if (ddsa->ddsa_props != NULL) {
1150 			dsl_props_set_sync_impl(ds->ds_prev,
1151 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1152 		}
1153 		dsl_dataset_rele(ds, FTAG);
1154 	}
1155 }
1156 
1157 /*
1158  * The snapshots must all be in the same pool.
1159  * All-or-nothing: if there are any failures, nothing will be modified.
1160  */
1161 int
1162 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1163 {
1164 	dsl_dataset_snapshot_arg_t ddsa;
1165 	nvpair_t *pair;
1166 	boolean_t needsuspend;
1167 	int error;
1168 	spa_t *spa;
1169 	char *firstname;
1170 	nvlist_t *suspended = NULL;
1171 
1172 	pair = nvlist_next_nvpair(snaps, NULL);
1173 	if (pair == NULL)
1174 		return (0);
1175 	firstname = nvpair_name(pair);
1176 
1177 	error = spa_open(firstname, &spa, FTAG);
1178 	if (error != 0)
1179 		return (error);
1180 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1181 	spa_close(spa, FTAG);
1182 
1183 	if (needsuspend) {
1184 		suspended = fnvlist_alloc();
1185 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1186 		    pair = nvlist_next_nvpair(snaps, pair)) {
1187 			char fsname[MAXNAMELEN];
1188 			char *snapname = nvpair_name(pair);
1189 			char *atp;
1190 			void *cookie;
1191 
1192 			atp = strchr(snapname, '@');
1193 			if (atp == NULL) {
1194 				error = SET_ERROR(EINVAL);
1195 				break;
1196 			}
1197 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1198 
1199 			error = zil_suspend(fsname, &cookie);
1200 			if (error != 0)
1201 				break;
1202 			fnvlist_add_uint64(suspended, fsname,
1203 			    (uintptr_t)cookie);
1204 		}
1205 	}
1206 
1207 	ddsa.ddsa_snaps = snaps;
1208 	ddsa.ddsa_props = props;
1209 	ddsa.ddsa_errors = errors;
1210 
1211 	if (error == 0) {
1212 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1213 		    dsl_dataset_snapshot_sync, &ddsa,
1214 		    fnvlist_num_pairs(snaps) * 3);
1215 	}
1216 
1217 	if (suspended != NULL) {
1218 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1219 		    pair = nvlist_next_nvpair(suspended, pair)) {
1220 			zil_resume((void *)(uintptr_t)
1221 			    fnvpair_value_uint64(pair));
1222 		}
1223 		fnvlist_free(suspended);
1224 	}
1225 
1226 	return (error);
1227 }
1228 
1229 typedef struct dsl_dataset_snapshot_tmp_arg {
1230 	const char *ddsta_fsname;
1231 	const char *ddsta_snapname;
1232 	minor_t ddsta_cleanup_minor;
1233 	const char *ddsta_htag;
1234 } dsl_dataset_snapshot_tmp_arg_t;
1235 
1236 static int
1237 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1238 {
1239 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1240 	dsl_pool_t *dp = dmu_tx_pool(tx);
1241 	dsl_dataset_t *ds;
1242 	int error;
1243 
1244 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1245 	if (error != 0)
1246 		return (error);
1247 
1248 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1249 	    tx, B_FALSE);
1250 	if (error != 0) {
1251 		dsl_dataset_rele(ds, FTAG);
1252 		return (error);
1253 	}
1254 
1255 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1256 		dsl_dataset_rele(ds, FTAG);
1257 		return (SET_ERROR(ENOTSUP));
1258 	}
1259 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1260 	    B_TRUE, tx);
1261 	if (error != 0) {
1262 		dsl_dataset_rele(ds, FTAG);
1263 		return (error);
1264 	}
1265 
1266 	dsl_dataset_rele(ds, FTAG);
1267 	return (0);
1268 }
1269 
1270 static void
1271 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1272 {
1273 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1274 	dsl_pool_t *dp = dmu_tx_pool(tx);
1275 	dsl_dataset_t *ds;
1276 
1277 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1278 
1279 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1280 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1281 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1282 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1283 
1284 	dsl_dataset_rele(ds, FTAG);
1285 }
1286 
1287 int
1288 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1289     minor_t cleanup_minor, const char *htag)
1290 {
1291 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1292 	int error;
1293 	spa_t *spa;
1294 	boolean_t needsuspend;
1295 	void *cookie;
1296 
1297 	ddsta.ddsta_fsname = fsname;
1298 	ddsta.ddsta_snapname = snapname;
1299 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1300 	ddsta.ddsta_htag = htag;
1301 
1302 	error = spa_open(fsname, &spa, FTAG);
1303 	if (error != 0)
1304 		return (error);
1305 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1306 	spa_close(spa, FTAG);
1307 
1308 	if (needsuspend) {
1309 		error = zil_suspend(fsname, &cookie);
1310 		if (error != 0)
1311 			return (error);
1312 	}
1313 
1314 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1315 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1316 
1317 	if (needsuspend)
1318 		zil_resume(cookie);
1319 	return (error);
1320 }
1321 
1322 
1323 void
1324 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1325 {
1326 	ASSERT(dmu_tx_is_syncing(tx));
1327 	ASSERT(ds->ds_objset != NULL);
1328 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1329 
1330 	/*
1331 	 * in case we had to change ds_fsid_guid when we opened it,
1332 	 * sync it out now.
1333 	 */
1334 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1335 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1336 
1337 	dmu_objset_sync(ds->ds_objset, zio, tx);
1338 }
1339 
1340 static void
1341 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1342 {
1343 	uint64_t count = 0;
1344 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1345 	zap_cursor_t zc;
1346 	zap_attribute_t za;
1347 	nvlist_t *propval = fnvlist_alloc();
1348 	nvlist_t *val = fnvlist_alloc();
1349 
1350 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1351 
1352 	/*
1353 	 * There may be missing entries in ds_next_clones_obj
1354 	 * due to a bug in a previous version of the code.
1355 	 * Only trust it if it has the right number of entries.
1356 	 */
1357 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1358 		ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1359 		    &count));
1360 	}
1361 	if (count != ds->ds_phys->ds_num_children - 1)
1362 		goto fail;
1363 	for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1364 	    zap_cursor_retrieve(&zc, &za) == 0;
1365 	    zap_cursor_advance(&zc)) {
1366 		dsl_dataset_t *clone;
1367 		char buf[ZFS_MAXNAMELEN];
1368 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1369 		    za.za_first_integer, FTAG, &clone));
1370 		dsl_dir_name(clone->ds_dir, buf);
1371 		fnvlist_add_boolean(val, buf);
1372 		dsl_dataset_rele(clone, FTAG);
1373 	}
1374 	zap_cursor_fini(&zc);
1375 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1376 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1377 fail:
1378 	nvlist_free(val);
1379 	nvlist_free(propval);
1380 }
1381 
1382 void
1383 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1384 {
1385 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1386 	uint64_t refd, avail, uobjs, aobjs, ratio;
1387 
1388 	ASSERT(dsl_pool_config_held(dp));
1389 
1390 	ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1391 	    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1392 	    ds->ds_phys->ds_compressed_bytes);
1393 
1394 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1395 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1396 	    ds->ds_phys->ds_uncompressed_bytes);
1397 
1398 	if (dsl_dataset_is_snapshot(ds)) {
1399 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1400 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1401 		    ds->ds_phys->ds_unique_bytes);
1402 		get_clones_stat(ds, nv);
1403 	} else {
1404 		dsl_dir_stats(ds->ds_dir, nv);
1405 	}
1406 
1407 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1408 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1409 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1410 
1411 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1412 	    ds->ds_phys->ds_creation_time);
1413 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1414 	    ds->ds_phys->ds_creation_txg);
1415 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1416 	    ds->ds_quota);
1417 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1418 	    ds->ds_reserved);
1419 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1420 	    ds->ds_phys->ds_guid);
1421 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1422 	    ds->ds_phys->ds_unique_bytes);
1423 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1424 	    ds->ds_object);
1425 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1426 	    ds->ds_userrefs);
1427 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1428 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1429 
1430 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1431 		uint64_t written, comp, uncomp;
1432 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1433 		dsl_dataset_t *prev;
1434 
1435 		int err = dsl_dataset_hold_obj(dp,
1436 		    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1437 		if (err == 0) {
1438 			err = dsl_dataset_space_written(prev, ds, &written,
1439 			    &comp, &uncomp);
1440 			dsl_dataset_rele(prev, FTAG);
1441 			if (err == 0) {
1442 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1443 				    written);
1444 			}
1445 		}
1446 	}
1447 }
1448 
1449 void
1450 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1451 {
1452 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1453 	ASSERT(dsl_pool_config_held(dp));
1454 
1455 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1456 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1457 	stat->dds_guid = ds->ds_phys->ds_guid;
1458 	stat->dds_origin[0] = '\0';
1459 	if (dsl_dataset_is_snapshot(ds)) {
1460 		stat->dds_is_snapshot = B_TRUE;
1461 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1462 	} else {
1463 		stat->dds_is_snapshot = B_FALSE;
1464 		stat->dds_num_clones = 0;
1465 
1466 		if (dsl_dir_is_clone(ds->ds_dir)) {
1467 			dsl_dataset_t *ods;
1468 
1469 			VERIFY0(dsl_dataset_hold_obj(dp,
1470 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1471 			dsl_dataset_name(ods, stat->dds_origin);
1472 			dsl_dataset_rele(ods, FTAG);
1473 		}
1474 	}
1475 }
1476 
1477 uint64_t
1478 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1479 {
1480 	return (ds->ds_fsid_guid);
1481 }
1482 
1483 void
1484 dsl_dataset_space(dsl_dataset_t *ds,
1485     uint64_t *refdbytesp, uint64_t *availbytesp,
1486     uint64_t *usedobjsp, uint64_t *availobjsp)
1487 {
1488 	*refdbytesp = ds->ds_phys->ds_referenced_bytes;
1489 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1490 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1491 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1492 	if (ds->ds_quota != 0) {
1493 		/*
1494 		 * Adjust available bytes according to refquota
1495 		 */
1496 		if (*refdbytesp < ds->ds_quota)
1497 			*availbytesp = MIN(*availbytesp,
1498 			    ds->ds_quota - *refdbytesp);
1499 		else
1500 			*availbytesp = 0;
1501 	}
1502 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1503 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1504 }
1505 
1506 boolean_t
1507 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1508 {
1509 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1510 
1511 	ASSERT(dsl_pool_config_held(dp));
1512 	if (snap == NULL)
1513 		return (B_FALSE);
1514 	if (ds->ds_phys->ds_bp.blk_birth >
1515 	    snap->ds_phys->ds_creation_txg) {
1516 		objset_t *os, *os_snap;
1517 		/*
1518 		 * It may be that only the ZIL differs, because it was
1519 		 * reset in the head.  Don't count that as being
1520 		 * modified.
1521 		 */
1522 		if (dmu_objset_from_ds(ds, &os) != 0)
1523 			return (B_TRUE);
1524 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1525 			return (B_TRUE);
1526 		return (bcmp(&os->os_phys->os_meta_dnode,
1527 		    &os_snap->os_phys->os_meta_dnode,
1528 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1529 	}
1530 	return (B_FALSE);
1531 }
1532 
1533 typedef struct dsl_dataset_rename_snapshot_arg {
1534 	const char *ddrsa_fsname;
1535 	const char *ddrsa_oldsnapname;
1536 	const char *ddrsa_newsnapname;
1537 	boolean_t ddrsa_recursive;
1538 	dmu_tx_t *ddrsa_tx;
1539 } dsl_dataset_rename_snapshot_arg_t;
1540 
1541 /* ARGSUSED */
1542 static int
1543 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1544     dsl_dataset_t *hds, void *arg)
1545 {
1546 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1547 	int error;
1548 	uint64_t val;
1549 
1550 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1551 	if (error != 0) {
1552 		/* ignore nonexistent snapshots */
1553 		return (error == ENOENT ? 0 : error);
1554 	}
1555 
1556 	/* new name should not exist */
1557 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1558 	if (error == 0)
1559 		error = SET_ERROR(EEXIST);
1560 	else if (error == ENOENT)
1561 		error = 0;
1562 
1563 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1564 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1565 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1566 		error = SET_ERROR(ENAMETOOLONG);
1567 
1568 	return (error);
1569 }
1570 
1571 static int
1572 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1573 {
1574 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1575 	dsl_pool_t *dp = dmu_tx_pool(tx);
1576 	dsl_dataset_t *hds;
1577 	int error;
1578 
1579 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1580 	if (error != 0)
1581 		return (error);
1582 
1583 	if (ddrsa->ddrsa_recursive) {
1584 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1585 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1586 		    DS_FIND_CHILDREN);
1587 	} else {
1588 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1589 	}
1590 	dsl_dataset_rele(hds, FTAG);
1591 	return (error);
1592 }
1593 
1594 static int
1595 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1596     dsl_dataset_t *hds, void *arg)
1597 {
1598 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1599 	dsl_dataset_t *ds;
1600 	uint64_t val;
1601 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1602 	int error;
1603 
1604 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1605 	ASSERT(error == 0 || error == ENOENT);
1606 	if (error == ENOENT) {
1607 		/* ignore nonexistent snapshots */
1608 		return (0);
1609 	}
1610 
1611 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1612 
1613 	/* log before we change the name */
1614 	spa_history_log_internal_ds(ds, "rename", tx,
1615 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1616 
1617 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx));
1618 	mutex_enter(&ds->ds_lock);
1619 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1620 	mutex_exit(&ds->ds_lock);
1621 	VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1622 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1623 
1624 	dsl_dataset_rele(ds, FTAG);
1625 	return (0);
1626 }
1627 
1628 static void
1629 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1630 {
1631 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1632 	dsl_pool_t *dp = dmu_tx_pool(tx);
1633 	dsl_dataset_t *hds;
1634 
1635 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1636 	ddrsa->ddrsa_tx = tx;
1637 	if (ddrsa->ddrsa_recursive) {
1638 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1639 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1640 		    DS_FIND_CHILDREN));
1641 	} else {
1642 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1643 	}
1644 	dsl_dataset_rele(hds, FTAG);
1645 }
1646 
1647 int
1648 dsl_dataset_rename_snapshot(const char *fsname,
1649     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1650 {
1651 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1652 
1653 	ddrsa.ddrsa_fsname = fsname;
1654 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1655 	ddrsa.ddrsa_newsnapname = newsnapname;
1656 	ddrsa.ddrsa_recursive = recursive;
1657 
1658 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1659 	    dsl_dataset_rename_snapshot_sync, &ddrsa, 1));
1660 }
1661 
1662 /*
1663  * If we're doing an ownership handoff, we need to make sure that there is
1664  * only one long hold on the dataset.  We're not allowed to change anything here
1665  * so we don't permanently release the long hold or regular hold here.  We want
1666  * to do this only when syncing to avoid the dataset unexpectedly going away
1667  * when we release the long hold.
1668  */
1669 static int
1670 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1671 {
1672 	boolean_t held;
1673 
1674 	if (!dmu_tx_is_syncing(tx))
1675 		return (0);
1676 
1677 	if (owner != NULL) {
1678 		VERIFY3P(ds->ds_owner, ==, owner);
1679 		dsl_dataset_long_rele(ds, owner);
1680 	}
1681 
1682 	held = dsl_dataset_long_held(ds);
1683 
1684 	if (owner != NULL)
1685 		dsl_dataset_long_hold(ds, owner);
1686 
1687 	if (held)
1688 		return (SET_ERROR(EBUSY));
1689 
1690 	return (0);
1691 }
1692 
1693 typedef struct dsl_dataset_rollback_arg {
1694 	const char *ddra_fsname;
1695 	void *ddra_owner;
1696 	nvlist_t *ddra_result;
1697 } dsl_dataset_rollback_arg_t;
1698 
1699 static int
1700 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1701 {
1702 	dsl_dataset_rollback_arg_t *ddra = arg;
1703 	dsl_pool_t *dp = dmu_tx_pool(tx);
1704 	dsl_dataset_t *ds;
1705 	int64_t unused_refres_delta;
1706 	int error;
1707 
1708 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1709 	if (error != 0)
1710 		return (error);
1711 
1712 	/* must not be a snapshot */
1713 	if (dsl_dataset_is_snapshot(ds)) {
1714 		dsl_dataset_rele(ds, FTAG);
1715 		return (SET_ERROR(EINVAL));
1716 	}
1717 
1718 	/* must have a most recent snapshot */
1719 	if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1720 		dsl_dataset_rele(ds, FTAG);
1721 		return (SET_ERROR(EINVAL));
1722 	}
1723 
1724 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1725 	if (error != 0) {
1726 		dsl_dataset_rele(ds, FTAG);
1727 		return (error);
1728 	}
1729 
1730 	/*
1731 	 * Check if the snap we are rolling back to uses more than
1732 	 * the refquota.
1733 	 */
1734 	if (ds->ds_quota != 0 &&
1735 	    ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1736 		dsl_dataset_rele(ds, FTAG);
1737 		return (SET_ERROR(EDQUOT));
1738 	}
1739 
1740 	/*
1741 	 * When we do the clone swap, we will temporarily use more space
1742 	 * due to the refreservation (the head will no longer have any
1743 	 * unique space, so the entire amount of the refreservation will need
1744 	 * to be free).  We will immediately destroy the clone, freeing
1745 	 * this space, but the freeing happens over many txg's.
1746 	 */
1747 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1748 	    ds->ds_phys->ds_unique_bytes);
1749 
1750 	if (unused_refres_delta > 0 &&
1751 	    unused_refres_delta >
1752 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1753 		dsl_dataset_rele(ds, FTAG);
1754 		return (SET_ERROR(ENOSPC));
1755 	}
1756 
1757 	dsl_dataset_rele(ds, FTAG);
1758 	return (0);
1759 }
1760 
1761 static void
1762 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1763 {
1764 	dsl_dataset_rollback_arg_t *ddra = arg;
1765 	dsl_pool_t *dp = dmu_tx_pool(tx);
1766 	dsl_dataset_t *ds, *clone;
1767 	uint64_t cloneobj;
1768 	char namebuf[ZFS_MAXNAMELEN];
1769 
1770 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
1771 
1772 	dsl_dataset_name(ds->ds_prev, namebuf);
1773 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
1774 
1775 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1776 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1777 
1778 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1779 
1780 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1781 	dsl_dataset_zero_zil(ds, tx);
1782 
1783 	dsl_destroy_head_sync_impl(clone, tx);
1784 
1785 	dsl_dataset_rele(clone, FTAG);
1786 	dsl_dataset_rele(ds, FTAG);
1787 }
1788 
1789 /*
1790  * Rolls back the given filesystem or volume to the most recent snapshot.
1791  * The name of the most recent snapshot will be returned under key "target"
1792  * in the result nvlist.
1793  *
1794  * If owner != NULL:
1795  * - The existing dataset MUST be owned by the specified owner at entry
1796  * - Upon return, dataset will still be held by the same owner, whether we
1797  *   succeed or not.
1798  *
1799  * This mode is required any time the existing filesystem is mounted.  See
1800  * notes above zfs_suspend_fs() for further details.
1801  */
1802 int
1803 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
1804 {
1805 	dsl_dataset_rollback_arg_t ddra;
1806 
1807 	ddra.ddra_fsname = fsname;
1808 	ddra.ddra_owner = owner;
1809 	ddra.ddra_result = result;
1810 
1811 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1812 	    dsl_dataset_rollback_sync, &ddra, 1));
1813 }
1814 
1815 struct promotenode {
1816 	list_node_t link;
1817 	dsl_dataset_t *ds;
1818 };
1819 
1820 typedef struct dsl_dataset_promote_arg {
1821 	const char *ddpa_clonename;
1822 	dsl_dataset_t *ddpa_clone;
1823 	list_t shared_snaps, origin_snaps, clone_snaps;
1824 	dsl_dataset_t *origin_origin; /* origin of the origin */
1825 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
1826 	char *err_ds;
1827 } dsl_dataset_promote_arg_t;
1828 
1829 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
1830 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
1831     void *tag);
1832 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
1833 
1834 static int
1835 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
1836 {
1837 	dsl_dataset_promote_arg_t *ddpa = arg;
1838 	dsl_pool_t *dp = dmu_tx_pool(tx);
1839 	dsl_dataset_t *hds;
1840 	struct promotenode *snap;
1841 	dsl_dataset_t *origin_ds;
1842 	int err;
1843 	uint64_t unused;
1844 
1845 	err = promote_hold(ddpa, dp, FTAG);
1846 	if (err != 0)
1847 		return (err);
1848 
1849 	hds = ddpa->ddpa_clone;
1850 
1851 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1852 		promote_rele(ddpa, FTAG);
1853 		return (SET_ERROR(EXDEV));
1854 	}
1855 
1856 	/*
1857 	 * Compute and check the amount of space to transfer.  Since this is
1858 	 * so expensive, don't do the preliminary check.
1859 	 */
1860 	if (!dmu_tx_is_syncing(tx)) {
1861 		promote_rele(ddpa, FTAG);
1862 		return (0);
1863 	}
1864 
1865 	snap = list_head(&ddpa->shared_snaps);
1866 	origin_ds = snap->ds;
1867 
1868 	/* compute origin's new unique space */
1869 	snap = list_tail(&ddpa->clone_snaps);
1870 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1871 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
1872 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1873 	    &ddpa->unique, &unused, &unused);
1874 
1875 	/*
1876 	 * Walk the snapshots that we are moving
1877 	 *
1878 	 * Compute space to transfer.  Consider the incremental changes
1879 	 * to used by each snapshot:
1880 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1881 	 * So each snapshot gave birth to:
1882 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1883 	 * So a sequence would look like:
1884 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1885 	 * Which simplifies to:
1886 	 * uN + kN + kN-1 + ... + k1 + k0
1887 	 * Note however, if we stop before we reach the ORIGIN we get:
1888 	 * uN + kN + kN-1 + ... + kM - uM-1
1889 	 */
1890 	ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
1891 	ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
1892 	ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
1893 	for (snap = list_head(&ddpa->shared_snaps); snap;
1894 	    snap = list_next(&ddpa->shared_snaps, snap)) {
1895 		uint64_t val, dlused, dlcomp, dluncomp;
1896 		dsl_dataset_t *ds = snap->ds;
1897 
1898 		/*
1899 		 * If there are long holds, we won't be able to evict
1900 		 * the objset.
1901 		 */
1902 		if (dsl_dataset_long_held(ds)) {
1903 			err = SET_ERROR(EBUSY);
1904 			goto out;
1905 		}
1906 
1907 		/* Check that the snapshot name does not conflict */
1908 		VERIFY0(dsl_dataset_get_snapname(ds));
1909 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
1910 		if (err == 0) {
1911 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
1912 			err = SET_ERROR(EEXIST);
1913 			goto out;
1914 		}
1915 		if (err != ENOENT)
1916 			goto out;
1917 
1918 		/* The very first snapshot does not have a deadlist */
1919 		if (ds->ds_phys->ds_prev_snap_obj == 0)
1920 			continue;
1921 
1922 		dsl_deadlist_space(&ds->ds_deadlist,
1923 		    &dlused, &dlcomp, &dluncomp);
1924 		ddpa->used += dlused;
1925 		ddpa->comp += dlcomp;
1926 		ddpa->uncomp += dluncomp;
1927 	}
1928 
1929 	/*
1930 	 * If we are a clone of a clone then we never reached ORIGIN,
1931 	 * so we need to subtract out the clone origin's used space.
1932 	 */
1933 	if (ddpa->origin_origin) {
1934 		ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
1935 		ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
1936 		ddpa->uncomp -=
1937 		    ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
1938 	}
1939 
1940 	/* Check that there is enough space here */
1941 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
1942 	    ddpa->used);
1943 	if (err != 0)
1944 		goto out;
1945 
1946 	/*
1947 	 * Compute the amounts of space that will be used by snapshots
1948 	 * after the promotion (for both origin and clone).  For each,
1949 	 * it is the amount of space that will be on all of their
1950 	 * deadlists (that was not born before their new origin).
1951 	 */
1952 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1953 		uint64_t space;
1954 
1955 		/*
1956 		 * Note, typically this will not be a clone of a clone,
1957 		 * so dd_origin_txg will be < TXG_INITIAL, so
1958 		 * these snaplist_space() -> dsl_deadlist_space_range()
1959 		 * calls will be fast because they do not have to
1960 		 * iterate over all bps.
1961 		 */
1962 		snap = list_head(&ddpa->origin_snaps);
1963 		err = snaplist_space(&ddpa->shared_snaps,
1964 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
1965 		if (err != 0)
1966 			goto out;
1967 
1968 		err = snaplist_space(&ddpa->clone_snaps,
1969 		    snap->ds->ds_dir->dd_origin_txg, &space);
1970 		if (err != 0)
1971 			goto out;
1972 		ddpa->cloneusedsnap += space;
1973 	}
1974 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1975 		err = snaplist_space(&ddpa->origin_snaps,
1976 		    origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
1977 		if (err != 0)
1978 			goto out;
1979 	}
1980 
1981 out:
1982 	promote_rele(ddpa, FTAG);
1983 	return (err);
1984 }
1985 
1986 static void
1987 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
1988 {
1989 	dsl_dataset_promote_arg_t *ddpa = arg;
1990 	dsl_pool_t *dp = dmu_tx_pool(tx);
1991 	dsl_dataset_t *hds;
1992 	struct promotenode *snap;
1993 	dsl_dataset_t *origin_ds;
1994 	dsl_dataset_t *origin_head;
1995 	dsl_dir_t *dd;
1996 	dsl_dir_t *odd = NULL;
1997 	uint64_t oldnext_obj;
1998 	int64_t delta;
1999 
2000 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2001 	hds = ddpa->ddpa_clone;
2002 
2003 	ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
2004 
2005 	snap = list_head(&ddpa->shared_snaps);
2006 	origin_ds = snap->ds;
2007 	dd = hds->ds_dir;
2008 
2009 	snap = list_head(&ddpa->origin_snaps);
2010 	origin_head = snap->ds;
2011 
2012 	/*
2013 	 * We need to explicitly open odd, since origin_ds's dd will be
2014 	 * changing.
2015 	 */
2016 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2017 	    NULL, FTAG, &odd));
2018 
2019 	/* change origin's next snap */
2020 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2021 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2022 	snap = list_tail(&ddpa->clone_snaps);
2023 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2024 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2025 
2026 	/* change the origin's next clone */
2027 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2028 		dsl_dataset_remove_from_next_clones(origin_ds,
2029 		    snap->ds->ds_object, tx);
2030 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2031 		    origin_ds->ds_phys->ds_next_clones_obj,
2032 		    oldnext_obj, tx));
2033 	}
2034 
2035 	/* change origin */
2036 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2037 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2038 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2039 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2040 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2041 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2042 	origin_head->ds_dir->dd_origin_txg =
2043 	    origin_ds->ds_phys->ds_creation_txg;
2044 
2045 	/* change dd_clone entries */
2046 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2047 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2048 		    odd->dd_phys->dd_clones, hds->ds_object, tx));
2049 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2050 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2051 		    hds->ds_object, tx));
2052 
2053 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2054 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2055 		    origin_head->ds_object, tx));
2056 		if (dd->dd_phys->dd_clones == 0) {
2057 			dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2058 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2059 		}
2060 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2061 		    dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2062 	}
2063 
2064 	/* move snapshots to this dir */
2065 	for (snap = list_head(&ddpa->shared_snaps); snap;
2066 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2067 		dsl_dataset_t *ds = snap->ds;
2068 
2069 		/*
2070 		 * Property callbacks are registered to a particular
2071 		 * dsl_dir.  Since ours is changing, evict the objset
2072 		 * so that they will be unregistered from the old dsl_dir.
2073 		 */
2074 		if (ds->ds_objset) {
2075 			dmu_objset_evict(ds->ds_objset);
2076 			ds->ds_objset = NULL;
2077 		}
2078 
2079 		/* move snap name entry */
2080 		VERIFY0(dsl_dataset_get_snapname(ds));
2081 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2082 		    ds->ds_snapname, tx));
2083 		VERIFY0(zap_add(dp->dp_meta_objset,
2084 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2085 		    8, 1, &ds->ds_object, tx));
2086 
2087 		/* change containing dsl_dir */
2088 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2089 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2090 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2091 		ASSERT3P(ds->ds_dir, ==, odd);
2092 		dsl_dir_rele(ds->ds_dir, ds);
2093 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2094 		    NULL, ds, &ds->ds_dir));
2095 
2096 		/* move any clone references */
2097 		if (ds->ds_phys->ds_next_clones_obj &&
2098 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2099 			zap_cursor_t zc;
2100 			zap_attribute_t za;
2101 
2102 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2103 			    ds->ds_phys->ds_next_clones_obj);
2104 			    zap_cursor_retrieve(&zc, &za) == 0;
2105 			    zap_cursor_advance(&zc)) {
2106 				dsl_dataset_t *cnds;
2107 				uint64_t o;
2108 
2109 				if (za.za_first_integer == oldnext_obj) {
2110 					/*
2111 					 * We've already moved the
2112 					 * origin's reference.
2113 					 */
2114 					continue;
2115 				}
2116 
2117 				VERIFY0(dsl_dataset_hold_obj(dp,
2118 				    za.za_first_integer, FTAG, &cnds));
2119 				o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2120 
2121 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2122 				    odd->dd_phys->dd_clones, o, tx));
2123 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2124 				    dd->dd_phys->dd_clones, o, tx));
2125 				dsl_dataset_rele(cnds, FTAG);
2126 			}
2127 			zap_cursor_fini(&zc);
2128 		}
2129 
2130 		ASSERT(!dsl_prop_hascb(ds));
2131 	}
2132 
2133 	/*
2134 	 * Change space accounting.
2135 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2136 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2137 	 * is true for each of {clone,origin} independently.
2138 	 */
2139 
2140 	delta = ddpa->cloneusedsnap -
2141 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2142 	ASSERT3S(delta, >=, 0);
2143 	ASSERT3U(ddpa->used, >=, delta);
2144 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2145 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2146 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2147 
2148 	delta = ddpa->originusedsnap -
2149 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2150 	ASSERT3S(delta, <=, 0);
2151 	ASSERT3U(ddpa->used, >=, -delta);
2152 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2153 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2154 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2155 
2156 	origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2157 
2158 	/* log history record */
2159 	spa_history_log_internal_ds(hds, "promote", tx, "");
2160 
2161 	dsl_dir_rele(odd, FTAG);
2162 	promote_rele(ddpa, FTAG);
2163 }
2164 
2165 /*
2166  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2167  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2168  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2169  * snapshots back to this dataset's origin.
2170  */
2171 static int
2172 snaplist_make(dsl_pool_t *dp,
2173     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2174 {
2175 	uint64_t obj = last_obj;
2176 
2177 	list_create(l, sizeof (struct promotenode),
2178 	    offsetof(struct promotenode, link));
2179 
2180 	while (obj != first_obj) {
2181 		dsl_dataset_t *ds;
2182 		struct promotenode *snap;
2183 		int err;
2184 
2185 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2186 		ASSERT(err != ENOENT);
2187 		if (err != 0)
2188 			return (err);
2189 
2190 		if (first_obj == 0)
2191 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2192 
2193 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2194 		snap->ds = ds;
2195 		list_insert_tail(l, snap);
2196 		obj = ds->ds_phys->ds_prev_snap_obj;
2197 	}
2198 
2199 	return (0);
2200 }
2201 
2202 static int
2203 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2204 {
2205 	struct promotenode *snap;
2206 
2207 	*spacep = 0;
2208 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2209 		uint64_t used, comp, uncomp;
2210 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2211 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2212 		*spacep += used;
2213 	}
2214 	return (0);
2215 }
2216 
2217 static void
2218 snaplist_destroy(list_t *l, void *tag)
2219 {
2220 	struct promotenode *snap;
2221 
2222 	if (l == NULL || !list_link_active(&l->list_head))
2223 		return;
2224 
2225 	while ((snap = list_tail(l)) != NULL) {
2226 		list_remove(l, snap);
2227 		dsl_dataset_rele(snap->ds, tag);
2228 		kmem_free(snap, sizeof (*snap));
2229 	}
2230 	list_destroy(l);
2231 }
2232 
2233 static int
2234 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2235 {
2236 	int error;
2237 	dsl_dir_t *dd;
2238 	struct promotenode *snap;
2239 
2240 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2241 	    &ddpa->ddpa_clone);
2242 	if (error != 0)
2243 		return (error);
2244 	dd = ddpa->ddpa_clone->ds_dir;
2245 
2246 	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2247 	    !dsl_dir_is_clone(dd)) {
2248 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2249 		return (SET_ERROR(EINVAL));
2250 	}
2251 
2252 	error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2253 	    &ddpa->shared_snaps, tag);
2254 	if (error != 0)
2255 		goto out;
2256 
2257 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2258 	    &ddpa->clone_snaps, tag);
2259 	if (error != 0)
2260 		goto out;
2261 
2262 	snap = list_head(&ddpa->shared_snaps);
2263 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2264 	error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2265 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2266 	    &ddpa->origin_snaps, tag);
2267 	if (error != 0)
2268 		goto out;
2269 
2270 	if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2271 		error = dsl_dataset_hold_obj(dp,
2272 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2273 		    tag, &ddpa->origin_origin);
2274 		if (error != 0)
2275 			goto out;
2276 	}
2277 out:
2278 	if (error != 0)
2279 		promote_rele(ddpa, tag);
2280 	return (error);
2281 }
2282 
2283 static void
2284 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2285 {
2286 	snaplist_destroy(&ddpa->shared_snaps, tag);
2287 	snaplist_destroy(&ddpa->clone_snaps, tag);
2288 	snaplist_destroy(&ddpa->origin_snaps, tag);
2289 	if (ddpa->origin_origin != NULL)
2290 		dsl_dataset_rele(ddpa->origin_origin, tag);
2291 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2292 }
2293 
2294 /*
2295  * Promote a clone.
2296  *
2297  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2298  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2299  */
2300 int
2301 dsl_dataset_promote(const char *name, char *conflsnap)
2302 {
2303 	dsl_dataset_promote_arg_t ddpa = { 0 };
2304 	uint64_t numsnaps;
2305 	int error;
2306 	objset_t *os;
2307 
2308 	/*
2309 	 * We will modify space proportional to the number of
2310 	 * snapshots.  Compute numsnaps.
2311 	 */
2312 	error = dmu_objset_hold(name, FTAG, &os);
2313 	if (error != 0)
2314 		return (error);
2315 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2316 	    dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2317 	dmu_objset_rele(os, FTAG);
2318 	if (error != 0)
2319 		return (error);
2320 
2321 	ddpa.ddpa_clonename = name;
2322 	ddpa.err_ds = conflsnap;
2323 
2324 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2325 	    dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2326 }
2327 
2328 int
2329 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2330     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2331 {
2332 	int64_t unused_refres_delta;
2333 
2334 	/* they should both be heads */
2335 	if (dsl_dataset_is_snapshot(clone) ||
2336 	    dsl_dataset_is_snapshot(origin_head))
2337 		return (SET_ERROR(EINVAL));
2338 
2339 	/* if we are not forcing, the branch point should be just before them */
2340 	if (!force && clone->ds_prev != origin_head->ds_prev)
2341 		return (SET_ERROR(EINVAL));
2342 
2343 	/* clone should be the clone (unless they are unrelated) */
2344 	if (clone->ds_prev != NULL &&
2345 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2346 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2347 		return (SET_ERROR(EINVAL));
2348 
2349 	/* the clone should be a child of the origin */
2350 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2351 		return (SET_ERROR(EINVAL));
2352 
2353 	/* origin_head shouldn't be modified unless 'force' */
2354 	if (!force &&
2355 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2356 		return (SET_ERROR(ETXTBSY));
2357 
2358 	/* origin_head should have no long holds (e.g. is not mounted) */
2359 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2360 		return (SET_ERROR(EBUSY));
2361 
2362 	/* check amount of any unconsumed refreservation */
2363 	unused_refres_delta =
2364 	    (int64_t)MIN(origin_head->ds_reserved,
2365 	    origin_head->ds_phys->ds_unique_bytes) -
2366 	    (int64_t)MIN(origin_head->ds_reserved,
2367 	    clone->ds_phys->ds_unique_bytes);
2368 
2369 	if (unused_refres_delta > 0 &&
2370 	    unused_refres_delta >
2371 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2372 		return (SET_ERROR(ENOSPC));
2373 
2374 	/* clone can't be over the head's refquota */
2375 	if (origin_head->ds_quota != 0 &&
2376 	    clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2377 		return (SET_ERROR(EDQUOT));
2378 
2379 	return (0);
2380 }
2381 
2382 void
2383 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2384     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2385 {
2386 	dsl_pool_t *dp = dmu_tx_pool(tx);
2387 	int64_t unused_refres_delta;
2388 
2389 	ASSERT(clone->ds_reserved == 0);
2390 	ASSERT(origin_head->ds_quota == 0 ||
2391 	    clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2392 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2393 
2394 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2395 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2396 
2397 	if (clone->ds_objset != NULL) {
2398 		dmu_objset_evict(clone->ds_objset);
2399 		clone->ds_objset = NULL;
2400 	}
2401 
2402 	if (origin_head->ds_objset != NULL) {
2403 		dmu_objset_evict(origin_head->ds_objset);
2404 		origin_head->ds_objset = NULL;
2405 	}
2406 
2407 	unused_refres_delta =
2408 	    (int64_t)MIN(origin_head->ds_reserved,
2409 	    origin_head->ds_phys->ds_unique_bytes) -
2410 	    (int64_t)MIN(origin_head->ds_reserved,
2411 	    clone->ds_phys->ds_unique_bytes);
2412 
2413 	/*
2414 	 * Reset origin's unique bytes, if it exists.
2415 	 */
2416 	if (clone->ds_prev) {
2417 		dsl_dataset_t *origin = clone->ds_prev;
2418 		uint64_t comp, uncomp;
2419 
2420 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2421 		dsl_deadlist_space_range(&clone->ds_deadlist,
2422 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2423 		    &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2424 	}
2425 
2426 	/* swap blkptrs */
2427 	{
2428 		blkptr_t tmp;
2429 		tmp = origin_head->ds_phys->ds_bp;
2430 		origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2431 		clone->ds_phys->ds_bp = tmp;
2432 	}
2433 
2434 	/* set dd_*_bytes */
2435 	{
2436 		int64_t dused, dcomp, duncomp;
2437 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2438 		uint64_t odl_used, odl_comp, odl_uncomp;
2439 
2440 		ASSERT3U(clone->ds_dir->dd_phys->
2441 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2442 
2443 		dsl_deadlist_space(&clone->ds_deadlist,
2444 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2445 		dsl_deadlist_space(&origin_head->ds_deadlist,
2446 		    &odl_used, &odl_comp, &odl_uncomp);
2447 
2448 		dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2449 		    (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2450 		dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2451 		    (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2452 		duncomp = clone->ds_phys->ds_uncompressed_bytes +
2453 		    cdl_uncomp -
2454 		    (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2455 
2456 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2457 		    dused, dcomp, duncomp, tx);
2458 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2459 		    -dused, -dcomp, -duncomp, tx);
2460 
2461 		/*
2462 		 * The difference in the space used by snapshots is the
2463 		 * difference in snapshot space due to the head's
2464 		 * deadlist (since that's the only thing that's
2465 		 * changing that affects the snapused).
2466 		 */
2467 		dsl_deadlist_space_range(&clone->ds_deadlist,
2468 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2469 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2470 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2471 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2472 		    &odl_used, &odl_comp, &odl_uncomp);
2473 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2474 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2475 	}
2476 
2477 	/* swap ds_*_bytes */
2478 	SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2479 	    clone->ds_phys->ds_referenced_bytes);
2480 	SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2481 	    clone->ds_phys->ds_compressed_bytes);
2482 	SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2483 	    clone->ds_phys->ds_uncompressed_bytes);
2484 	SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2485 	    clone->ds_phys->ds_unique_bytes);
2486 
2487 	/* apply any parent delta for change in unconsumed refreservation */
2488 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2489 	    unused_refres_delta, 0, 0, tx);
2490 
2491 	/*
2492 	 * Swap deadlists.
2493 	 */
2494 	dsl_deadlist_close(&clone->ds_deadlist);
2495 	dsl_deadlist_close(&origin_head->ds_deadlist);
2496 	SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2497 	    clone->ds_phys->ds_deadlist_obj);
2498 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2499 	    clone->ds_phys->ds_deadlist_obj);
2500 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2501 	    origin_head->ds_phys->ds_deadlist_obj);
2502 
2503 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2504 
2505 	spa_history_log_internal_ds(clone, "clone swap", tx,
2506 	    "parent=%s", origin_head->ds_dir->dd_myname);
2507 }
2508 
2509 /*
2510  * Given a pool name and a dataset object number in that pool,
2511  * return the name of that dataset.
2512  */
2513 int
2514 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2515 {
2516 	dsl_pool_t *dp;
2517 	dsl_dataset_t *ds;
2518 	int error;
2519 
2520 	error = dsl_pool_hold(pname, FTAG, &dp);
2521 	if (error != 0)
2522 		return (error);
2523 
2524 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2525 	if (error == 0) {
2526 		dsl_dataset_name(ds, buf);
2527 		dsl_dataset_rele(ds, FTAG);
2528 	}
2529 	dsl_pool_rele(dp, FTAG);
2530 
2531 	return (error);
2532 }
2533 
2534 int
2535 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2536     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2537 {
2538 	int error = 0;
2539 
2540 	ASSERT3S(asize, >, 0);
2541 
2542 	/*
2543 	 * *ref_rsrv is the portion of asize that will come from any
2544 	 * unconsumed refreservation space.
2545 	 */
2546 	*ref_rsrv = 0;
2547 
2548 	mutex_enter(&ds->ds_lock);
2549 	/*
2550 	 * Make a space adjustment for reserved bytes.
2551 	 */
2552 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2553 		ASSERT3U(*used, >=,
2554 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2555 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2556 		*ref_rsrv =
2557 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2558 	}
2559 
2560 	if (!check_quota || ds->ds_quota == 0) {
2561 		mutex_exit(&ds->ds_lock);
2562 		return (0);
2563 	}
2564 	/*
2565 	 * If they are requesting more space, and our current estimate
2566 	 * is over quota, they get to try again unless the actual
2567 	 * on-disk is over quota and there are no pending changes (which
2568 	 * may free up space for us).
2569 	 */
2570 	if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2571 		if (inflight > 0 ||
2572 		    ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2573 			error = SET_ERROR(ERESTART);
2574 		else
2575 			error = SET_ERROR(EDQUOT);
2576 	}
2577 	mutex_exit(&ds->ds_lock);
2578 
2579 	return (error);
2580 }
2581 
2582 typedef struct dsl_dataset_set_qr_arg {
2583 	const char *ddsqra_name;
2584 	zprop_source_t ddsqra_source;
2585 	uint64_t ddsqra_value;
2586 } dsl_dataset_set_qr_arg_t;
2587 
2588 
2589 /* ARGSUSED */
2590 static int
2591 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2592 {
2593 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2594 	dsl_pool_t *dp = dmu_tx_pool(tx);
2595 	dsl_dataset_t *ds;
2596 	int error;
2597 	uint64_t newval;
2598 
2599 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2600 		return (SET_ERROR(ENOTSUP));
2601 
2602 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2603 	if (error != 0)
2604 		return (error);
2605 
2606 	if (dsl_dataset_is_snapshot(ds)) {
2607 		dsl_dataset_rele(ds, FTAG);
2608 		return (SET_ERROR(EINVAL));
2609 	}
2610 
2611 	error = dsl_prop_predict(ds->ds_dir,
2612 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2613 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2614 	if (error != 0) {
2615 		dsl_dataset_rele(ds, FTAG);
2616 		return (error);
2617 	}
2618 
2619 	if (newval == 0) {
2620 		dsl_dataset_rele(ds, FTAG);
2621 		return (0);
2622 	}
2623 
2624 	if (newval < ds->ds_phys->ds_referenced_bytes ||
2625 	    newval < ds->ds_reserved) {
2626 		dsl_dataset_rele(ds, FTAG);
2627 		return (SET_ERROR(ENOSPC));
2628 	}
2629 
2630 	dsl_dataset_rele(ds, FTAG);
2631 	return (0);
2632 }
2633 
2634 static void
2635 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2636 {
2637 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2638 	dsl_pool_t *dp = dmu_tx_pool(tx);
2639 	dsl_dataset_t *ds;
2640 	uint64_t newval;
2641 
2642 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2643 
2644 	dsl_prop_set_sync_impl(ds,
2645 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2646 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2647 	    &ddsqra->ddsqra_value, tx);
2648 
2649 	VERIFY0(dsl_prop_get_int_ds(ds,
2650 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2651 
2652 	if (ds->ds_quota != newval) {
2653 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2654 		ds->ds_quota = newval;
2655 	}
2656 	dsl_dataset_rele(ds, FTAG);
2657 }
2658 
2659 int
2660 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2661     uint64_t refquota)
2662 {
2663 	dsl_dataset_set_qr_arg_t ddsqra;
2664 
2665 	ddsqra.ddsqra_name = dsname;
2666 	ddsqra.ddsqra_source = source;
2667 	ddsqra.ddsqra_value = refquota;
2668 
2669 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2670 	    dsl_dataset_set_refquota_sync, &ddsqra, 0));
2671 }
2672 
2673 static int
2674 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2675 {
2676 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2677 	dsl_pool_t *dp = dmu_tx_pool(tx);
2678 	dsl_dataset_t *ds;
2679 	int error;
2680 	uint64_t newval, unique;
2681 
2682 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2683 		return (SET_ERROR(ENOTSUP));
2684 
2685 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2686 	if (error != 0)
2687 		return (error);
2688 
2689 	if (dsl_dataset_is_snapshot(ds)) {
2690 		dsl_dataset_rele(ds, FTAG);
2691 		return (SET_ERROR(EINVAL));
2692 	}
2693 
2694 	error = dsl_prop_predict(ds->ds_dir,
2695 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2696 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2697 	if (error != 0) {
2698 		dsl_dataset_rele(ds, FTAG);
2699 		return (error);
2700 	}
2701 
2702 	/*
2703 	 * If we are doing the preliminary check in open context, the
2704 	 * space estimates may be inaccurate.
2705 	 */
2706 	if (!dmu_tx_is_syncing(tx)) {
2707 		dsl_dataset_rele(ds, FTAG);
2708 		return (0);
2709 	}
2710 
2711 	mutex_enter(&ds->ds_lock);
2712 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2713 		dsl_dataset_recalc_head_uniq(ds);
2714 	unique = ds->ds_phys->ds_unique_bytes;
2715 	mutex_exit(&ds->ds_lock);
2716 
2717 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2718 		uint64_t delta = MAX(unique, newval) -
2719 		    MAX(unique, ds->ds_reserved);
2720 
2721 		if (delta >
2722 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2723 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2724 			dsl_dataset_rele(ds, FTAG);
2725 			return (SET_ERROR(ENOSPC));
2726 		}
2727 	}
2728 
2729 	dsl_dataset_rele(ds, FTAG);
2730 	return (0);
2731 }
2732 
2733 void
2734 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2735     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2736 {
2737 	uint64_t newval;
2738 	uint64_t unique;
2739 	int64_t delta;
2740 
2741 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2742 	    source, sizeof (value), 1, &value, tx);
2743 
2744 	VERIFY0(dsl_prop_get_int_ds(ds,
2745 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2746 
2747 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2748 	mutex_enter(&ds->ds_dir->dd_lock);
2749 	mutex_enter(&ds->ds_lock);
2750 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2751 	unique = ds->ds_phys->ds_unique_bytes;
2752 	delta = MAX(0, (int64_t)(newval - unique)) -
2753 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2754 	ds->ds_reserved = newval;
2755 	mutex_exit(&ds->ds_lock);
2756 
2757 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2758 	mutex_exit(&ds->ds_dir->dd_lock);
2759 }
2760 
2761 static void
2762 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2763 {
2764 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2765 	dsl_pool_t *dp = dmu_tx_pool(tx);
2766 	dsl_dataset_t *ds;
2767 
2768 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2769 	dsl_dataset_set_refreservation_sync_impl(ds,
2770 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2771 	dsl_dataset_rele(ds, FTAG);
2772 }
2773 
2774 int
2775 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2776     uint64_t refreservation)
2777 {
2778 	dsl_dataset_set_qr_arg_t ddsqra;
2779 
2780 	ddsqra.ddsqra_name = dsname;
2781 	ddsqra.ddsqra_source = source;
2782 	ddsqra.ddsqra_value = refreservation;
2783 
2784 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2785 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0));
2786 }
2787 
2788 /*
2789  * Return (in *usedp) the amount of space written in new that is not
2790  * present in oldsnap.  New may be a snapshot or the head.  Old must be
2791  * a snapshot before new, in new's filesystem (or its origin).  If not then
2792  * fail and return EINVAL.
2793  *
2794  * The written space is calculated by considering two components:  First, we
2795  * ignore any freed space, and calculate the written as new's used space
2796  * minus old's used space.  Next, we add in the amount of space that was freed
2797  * between the two snapshots, thus reducing new's used space relative to old's.
2798  * Specifically, this is the space that was born before old->ds_creation_txg,
2799  * and freed before new (ie. on new's deadlist or a previous deadlist).
2800  *
2801  * space freed                         [---------------------]
2802  * snapshots                       ---O-------O--------O-------O------
2803  *                                         oldsnap            new
2804  */
2805 int
2806 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2807     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2808 {
2809 	int err = 0;
2810 	uint64_t snapobj;
2811 	dsl_pool_t *dp = new->ds_dir->dd_pool;
2812 
2813 	ASSERT(dsl_pool_config_held(dp));
2814 
2815 	*usedp = 0;
2816 	*usedp += new->ds_phys->ds_referenced_bytes;
2817 	*usedp -= oldsnap->ds_phys->ds_referenced_bytes;
2818 
2819 	*compp = 0;
2820 	*compp += new->ds_phys->ds_compressed_bytes;
2821 	*compp -= oldsnap->ds_phys->ds_compressed_bytes;
2822 
2823 	*uncompp = 0;
2824 	*uncompp += new->ds_phys->ds_uncompressed_bytes;
2825 	*uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
2826 
2827 	snapobj = new->ds_object;
2828 	while (snapobj != oldsnap->ds_object) {
2829 		dsl_dataset_t *snap;
2830 		uint64_t used, comp, uncomp;
2831 
2832 		if (snapobj == new->ds_object) {
2833 			snap = new;
2834 		} else {
2835 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
2836 			if (err != 0)
2837 				break;
2838 		}
2839 
2840 		if (snap->ds_phys->ds_prev_snap_txg ==
2841 		    oldsnap->ds_phys->ds_creation_txg) {
2842 			/*
2843 			 * The blocks in the deadlist can not be born after
2844 			 * ds_prev_snap_txg, so get the whole deadlist space,
2845 			 * which is more efficient (especially for old-format
2846 			 * deadlists).  Unfortunately the deadlist code
2847 			 * doesn't have enough information to make this
2848 			 * optimization itself.
2849 			 */
2850 			dsl_deadlist_space(&snap->ds_deadlist,
2851 			    &used, &comp, &uncomp);
2852 		} else {
2853 			dsl_deadlist_space_range(&snap->ds_deadlist,
2854 			    0, oldsnap->ds_phys->ds_creation_txg,
2855 			    &used, &comp, &uncomp);
2856 		}
2857 		*usedp += used;
2858 		*compp += comp;
2859 		*uncompp += uncomp;
2860 
2861 		/*
2862 		 * If we get to the beginning of the chain of snapshots
2863 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2864 		 * was not a snapshot of/before new.
2865 		 */
2866 		snapobj = snap->ds_phys->ds_prev_snap_obj;
2867 		if (snap != new)
2868 			dsl_dataset_rele(snap, FTAG);
2869 		if (snapobj == 0) {
2870 			err = SET_ERROR(EINVAL);
2871 			break;
2872 		}
2873 
2874 	}
2875 	return (err);
2876 }
2877 
2878 /*
2879  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2880  * lastsnap, and all snapshots in between are deleted.
2881  *
2882  * blocks that would be freed            [---------------------------]
2883  * snapshots                       ---O-------O--------O-------O--------O
2884  *                                        firstsnap        lastsnap
2885  *
2886  * This is the set of blocks that were born after the snap before firstsnap,
2887  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2888  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2889  * We calculate this by iterating over the relevant deadlists (from the snap
2890  * after lastsnap, backward to the snap after firstsnap), summing up the
2891  * space on the deadlist that was born after the snap before firstsnap.
2892  */
2893 int
2894 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
2895     dsl_dataset_t *lastsnap,
2896     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2897 {
2898 	int err = 0;
2899 	uint64_t snapobj;
2900 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
2901 
2902 	ASSERT(dsl_dataset_is_snapshot(firstsnap));
2903 	ASSERT(dsl_dataset_is_snapshot(lastsnap));
2904 
2905 	/*
2906 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
2907 	 * is before lastsnap.
2908 	 */
2909 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
2910 	    firstsnap->ds_phys->ds_creation_txg >
2911 	    lastsnap->ds_phys->ds_creation_txg)
2912 		return (SET_ERROR(EINVAL));
2913 
2914 	*usedp = *compp = *uncompp = 0;
2915 
2916 	snapobj = lastsnap->ds_phys->ds_next_snap_obj;
2917 	while (snapobj != firstsnap->ds_object) {
2918 		dsl_dataset_t *ds;
2919 		uint64_t used, comp, uncomp;
2920 
2921 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
2922 		if (err != 0)
2923 			break;
2924 
2925 		dsl_deadlist_space_range(&ds->ds_deadlist,
2926 		    firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2927 		    &used, &comp, &uncomp);
2928 		*usedp += used;
2929 		*compp += comp;
2930 		*uncompp += uncomp;
2931 
2932 		snapobj = ds->ds_phys->ds_prev_snap_obj;
2933 		ASSERT3U(snapobj, !=, 0);
2934 		dsl_dataset_rele(ds, FTAG);
2935 	}
2936 	return (err);
2937 }
2938 
2939 /*
2940  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2941  * For example, they could both be snapshots of the same filesystem, and
2942  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
2943  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
2944  * filesystem.  Or 'earlier' could be the origin's origin.
2945  */
2946 boolean_t
2947 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
2948 {
2949 	dsl_pool_t *dp = later->ds_dir->dd_pool;
2950 	int error;
2951 	boolean_t ret;
2952 
2953 	ASSERT(dsl_pool_config_held(dp));
2954 
2955 	if (earlier->ds_phys->ds_creation_txg >=
2956 	    later->ds_phys->ds_creation_txg)
2957 		return (B_FALSE);
2958 
2959 	if (later->ds_dir == earlier->ds_dir)
2960 		return (B_TRUE);
2961 	if (!dsl_dir_is_clone(later->ds_dir))
2962 		return (B_FALSE);
2963 
2964 	if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
2965 		return (B_TRUE);
2966 	dsl_dataset_t *origin;
2967 	error = dsl_dataset_hold_obj(dp,
2968 	    later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
2969 	if (error != 0)
2970 		return (B_FALSE);
2971 	ret = dsl_dataset_is_before(origin, earlier);
2972 	dsl_dataset_rele(origin, FTAG);
2973 	return (ret);
2974 }
2975 
2976 
2977 void
2978 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
2979 {
2980 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2981 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
2982 }
2983