xref: /titanic_50/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision ef16f6b5e980be84e2248939502c5aa269ef026d)
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 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
104 
105 	mutex_enter(&ds->ds_dir->dd_lock);
106 	mutex_enter(&ds->ds_lock);
107 	delta = parent_delta(ds, used);
108 	ds->ds_phys->ds_referenced_bytes += used;
109 	ds->ds_phys->ds_compressed_bytes += compressed;
110 	ds->ds_phys->ds_uncompressed_bytes += uncompressed;
111 	ds->ds_phys->ds_unique_bytes += used;
112 	mutex_exit(&ds->ds_lock);
113 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
114 	    compressed, uncompressed, tx);
115 	dsl_dir_transfer_space(ds->ds_dir, used - delta,
116 	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
117 	mutex_exit(&ds->ds_dir->dd_lock);
118 }
119 
120 int
121 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
122     boolean_t async)
123 {
124 	if (BP_IS_HOLE(bp))
125 		return (0);
126 
127 	ASSERT(dmu_tx_is_syncing(tx));
128 	ASSERT(bp->blk_birth <= tx->tx_txg);
129 
130 	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
131 	int compressed = BP_GET_PSIZE(bp);
132 	int uncompressed = BP_GET_UCSIZE(bp);
133 
134 	ASSERT(used > 0);
135 	if (ds == NULL) {
136 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
137 		dsl_pool_mos_diduse_space(tx->tx_pool,
138 		    -used, -compressed, -uncompressed);
139 		return (used);
140 	}
141 	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
142 
143 	ASSERT(!dsl_dataset_is_snapshot(ds));
144 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
145 
146 	if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
147 		int64_t delta;
148 
149 		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
150 		dsl_free(tx->tx_pool, tx->tx_txg, bp);
151 
152 		mutex_enter(&ds->ds_dir->dd_lock);
153 		mutex_enter(&ds->ds_lock);
154 		ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
155 		    !DS_UNIQUE_IS_ACCURATE(ds));
156 		delta = parent_delta(ds, -used);
157 		ds->ds_phys->ds_unique_bytes -= used;
158 		mutex_exit(&ds->ds_lock);
159 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
160 		    delta, -compressed, -uncompressed, tx);
161 		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
162 		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
163 		mutex_exit(&ds->ds_dir->dd_lock);
164 	} else {
165 		dprintf_bp(bp, "putting on dead list: %s", "");
166 		if (async) {
167 			/*
168 			 * We are here as part of zio's write done callback,
169 			 * which means we're a zio interrupt thread.  We can't
170 			 * call dsl_deadlist_insert() now because it may block
171 			 * waiting for I/O.  Instead, put bp on the deferred
172 			 * queue and let dsl_pool_sync() finish the job.
173 			 */
174 			bplist_append(&ds->ds_pending_deadlist, bp);
175 		} else {
176 			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
177 		}
178 		ASSERT3U(ds->ds_prev->ds_object, ==,
179 		    ds->ds_phys->ds_prev_snap_obj);
180 		ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
181 		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
182 		if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
183 		    ds->ds_object && bp->blk_birth >
184 		    ds->ds_prev->ds_phys->ds_prev_snap_txg) {
185 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
186 			mutex_enter(&ds->ds_prev->ds_lock);
187 			ds->ds_prev->ds_phys->ds_unique_bytes += used;
188 			mutex_exit(&ds->ds_prev->ds_lock);
189 		}
190 		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
191 			dsl_dir_transfer_space(ds->ds_dir, used,
192 			    DD_USED_HEAD, DD_USED_SNAP, tx);
193 		}
194 	}
195 	mutex_enter(&ds->ds_lock);
196 	ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
197 	ds->ds_phys->ds_referenced_bytes -= used;
198 	ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
199 	ds->ds_phys->ds_compressed_bytes -= compressed;
200 	ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
201 	ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
202 	mutex_exit(&ds->ds_lock);
203 
204 	return (used);
205 }
206 
207 uint64_t
208 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
209 {
210 	uint64_t trysnap = 0;
211 
212 	if (ds == NULL)
213 		return (0);
214 	/*
215 	 * The snapshot creation could fail, but that would cause an
216 	 * incorrect FALSE return, which would only result in an
217 	 * overestimation of the amount of space that an operation would
218 	 * consume, which is OK.
219 	 *
220 	 * There's also a small window where we could miss a pending
221 	 * snapshot, because we could set the sync task in the quiescing
222 	 * phase.  So this should only be used as a guess.
223 	 */
224 	if (ds->ds_trysnap_txg >
225 	    spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
226 		trysnap = ds->ds_trysnap_txg;
227 	return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
228 }
229 
230 boolean_t
231 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
232     uint64_t blk_birth)
233 {
234 	if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
235 		return (B_FALSE);
236 
237 	ddt_prefetch(dsl_dataset_get_spa(ds), bp);
238 
239 	return (B_TRUE);
240 }
241 
242 /* ARGSUSED */
243 static void
244 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
245 {
246 	dsl_dataset_t *ds = dsv;
247 
248 	ASSERT(ds->ds_owner == NULL);
249 
250 	unique_remove(ds->ds_fsid_guid);
251 
252 	if (ds->ds_objset != NULL)
253 		dmu_objset_evict(ds->ds_objset);
254 
255 	if (ds->ds_prev) {
256 		dsl_dataset_rele(ds->ds_prev, ds);
257 		ds->ds_prev = NULL;
258 	}
259 
260 	bplist_destroy(&ds->ds_pending_deadlist);
261 	if (ds->ds_phys->ds_deadlist_obj != 0)
262 		dsl_deadlist_close(&ds->ds_deadlist);
263 	if (ds->ds_dir)
264 		dsl_dir_rele(ds->ds_dir, ds);
265 
266 	ASSERT(!list_link_active(&ds->ds_synced_link));
267 
268 	mutex_destroy(&ds->ds_lock);
269 	mutex_destroy(&ds->ds_opening_lock);
270 	refcount_destroy(&ds->ds_longholds);
271 
272 	kmem_free(ds, sizeof (dsl_dataset_t));
273 }
274 
275 int
276 dsl_dataset_get_snapname(dsl_dataset_t *ds)
277 {
278 	dsl_dataset_phys_t *headphys;
279 	int err;
280 	dmu_buf_t *headdbuf;
281 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
282 	objset_t *mos = dp->dp_meta_objset;
283 
284 	if (ds->ds_snapname[0])
285 		return (0);
286 	if (ds->ds_phys->ds_next_snap_obj == 0)
287 		return (0);
288 
289 	err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
290 	    FTAG, &headdbuf);
291 	if (err != 0)
292 		return (err);
293 	headphys = headdbuf->db_data;
294 	err = zap_value_search(dp->dp_meta_objset,
295 	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
296 	dmu_buf_rele(headdbuf, FTAG);
297 	return (err);
298 }
299 
300 int
301 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
302 {
303 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
304 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
305 	matchtype_t mt;
306 	int err;
307 
308 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
309 		mt = MT_FIRST;
310 	else
311 		mt = MT_EXACT;
312 
313 	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
314 	    value, mt, NULL, 0, NULL);
315 	if (err == ENOTSUP && mt == MT_FIRST)
316 		err = zap_lookup(mos, snapobj, name, 8, 1, value);
317 	return (err);
318 }
319 
320 int
321 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx)
322 {
323 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
324 	uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
325 	matchtype_t mt;
326 	int err;
327 
328 	dsl_dir_snap_cmtime_update(ds->ds_dir);
329 
330 	if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
331 		mt = MT_FIRST;
332 	else
333 		mt = MT_EXACT;
334 
335 	err = zap_remove_norm(mos, snapobj, name, mt, tx);
336 	if (err == ENOTSUP && mt == MT_FIRST)
337 		err = zap_remove(mos, snapobj, name, tx);
338 	return (err);
339 }
340 
341 int
342 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
343     dsl_dataset_t **dsp)
344 {
345 	objset_t *mos = dp->dp_meta_objset;
346 	dmu_buf_t *dbuf;
347 	dsl_dataset_t *ds;
348 	int err;
349 	dmu_object_info_t doi;
350 
351 	ASSERT(dsl_pool_config_held(dp));
352 
353 	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
354 	if (err != 0)
355 		return (err);
356 
357 	/* Make sure dsobj has the correct object type. */
358 	dmu_object_info_from_db(dbuf, &doi);
359 	if (doi.doi_type != DMU_OT_DSL_DATASET)
360 		return (SET_ERROR(EINVAL));
361 
362 	ds = dmu_buf_get_user(dbuf);
363 	if (ds == NULL) {
364 		dsl_dataset_t *winner = NULL;
365 
366 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
367 		ds->ds_dbuf = dbuf;
368 		ds->ds_object = dsobj;
369 		ds->ds_phys = dbuf->db_data;
370 
371 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
372 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
373 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
374 		refcount_create(&ds->ds_longholds);
375 
376 		bplist_create(&ds->ds_pending_deadlist);
377 		dsl_deadlist_open(&ds->ds_deadlist,
378 		    mos, ds->ds_phys->ds_deadlist_obj);
379 
380 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
381 		    offsetof(dmu_sendarg_t, dsa_link));
382 
383 		if (err == 0) {
384 			err = dsl_dir_hold_obj(dp,
385 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
386 		}
387 		if (err != 0) {
388 			mutex_destroy(&ds->ds_lock);
389 			mutex_destroy(&ds->ds_opening_lock);
390 			refcount_destroy(&ds->ds_longholds);
391 			bplist_destroy(&ds->ds_pending_deadlist);
392 			dsl_deadlist_close(&ds->ds_deadlist);
393 			kmem_free(ds, sizeof (dsl_dataset_t));
394 			dmu_buf_rele(dbuf, tag);
395 			return (err);
396 		}
397 
398 		if (!dsl_dataset_is_snapshot(ds)) {
399 			ds->ds_snapname[0] = '\0';
400 			if (ds->ds_phys->ds_prev_snap_obj != 0) {
401 				err = dsl_dataset_hold_obj(dp,
402 				    ds->ds_phys->ds_prev_snap_obj,
403 				    ds, &ds->ds_prev);
404 			}
405 		} else {
406 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
407 				err = dsl_dataset_get_snapname(ds);
408 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
409 				err = zap_count(
410 				    ds->ds_dir->dd_pool->dp_meta_objset,
411 				    ds->ds_phys->ds_userrefs_obj,
412 				    &ds->ds_userrefs);
413 			}
414 		}
415 
416 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
417 			err = dsl_prop_get_int_ds(ds,
418 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
419 			    &ds->ds_reserved);
420 			if (err == 0) {
421 				err = dsl_prop_get_int_ds(ds,
422 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
423 				    &ds->ds_quota);
424 			}
425 		} else {
426 			ds->ds_reserved = ds->ds_quota = 0;
427 		}
428 
429 		if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
430 		    &ds->ds_phys, dsl_dataset_evict)) != NULL) {
431 			bplist_destroy(&ds->ds_pending_deadlist);
432 			dsl_deadlist_close(&ds->ds_deadlist);
433 			if (ds->ds_prev)
434 				dsl_dataset_rele(ds->ds_prev, ds);
435 			dsl_dir_rele(ds->ds_dir, ds);
436 			mutex_destroy(&ds->ds_lock);
437 			mutex_destroy(&ds->ds_opening_lock);
438 			refcount_destroy(&ds->ds_longholds);
439 			kmem_free(ds, sizeof (dsl_dataset_t));
440 			if (err != 0) {
441 				dmu_buf_rele(dbuf, tag);
442 				return (err);
443 			}
444 			ds = winner;
445 		} else {
446 			ds->ds_fsid_guid =
447 			    unique_insert(ds->ds_phys->ds_fsid_guid);
448 		}
449 	}
450 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
451 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
452 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
453 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
454 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
455 	*dsp = ds;
456 	return (0);
457 }
458 
459 int
460 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
461     void *tag, dsl_dataset_t **dsp)
462 {
463 	dsl_dir_t *dd;
464 	const char *snapname;
465 	uint64_t obj;
466 	int err = 0;
467 
468 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
469 	if (err != 0)
470 		return (err);
471 
472 	ASSERT(dsl_pool_config_held(dp));
473 	obj = dd->dd_phys->dd_head_dataset_obj;
474 	if (obj != 0)
475 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
476 	else
477 		err = SET_ERROR(ENOENT);
478 
479 	/* we may be looking for a snapshot */
480 	if (err == 0 && snapname != NULL) {
481 		dsl_dataset_t *ds;
482 
483 		if (*snapname++ != '@') {
484 			dsl_dataset_rele(*dsp, tag);
485 			dsl_dir_rele(dd, FTAG);
486 			return (SET_ERROR(ENOENT));
487 		}
488 
489 		dprintf("looking for snapshot '%s'\n", snapname);
490 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
491 		if (err == 0)
492 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
493 		dsl_dataset_rele(*dsp, tag);
494 
495 		if (err == 0) {
496 			mutex_enter(&ds->ds_lock);
497 			if (ds->ds_snapname[0] == 0)
498 				(void) strlcpy(ds->ds_snapname, snapname,
499 				    sizeof (ds->ds_snapname));
500 			mutex_exit(&ds->ds_lock);
501 			*dsp = ds;
502 		}
503 	}
504 
505 	dsl_dir_rele(dd, FTAG);
506 	return (err);
507 }
508 
509 int
510 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
511     void *tag, dsl_dataset_t **dsp)
512 {
513 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
514 	if (err != 0)
515 		return (err);
516 	if (!dsl_dataset_tryown(*dsp, tag)) {
517 		dsl_dataset_rele(*dsp, tag);
518 		*dsp = NULL;
519 		return (SET_ERROR(EBUSY));
520 	}
521 	return (0);
522 }
523 
524 int
525 dsl_dataset_own(dsl_pool_t *dp, const char *name,
526     void *tag, dsl_dataset_t **dsp)
527 {
528 	int err = dsl_dataset_hold(dp, name, tag, dsp);
529 	if (err != 0)
530 		return (err);
531 	if (!dsl_dataset_tryown(*dsp, tag)) {
532 		dsl_dataset_rele(*dsp, tag);
533 		return (SET_ERROR(EBUSY));
534 	}
535 	return (0);
536 }
537 
538 /*
539  * See the comment above dsl_pool_hold() for details.  In summary, a long
540  * hold is used to prevent destruction of a dataset while the pool hold
541  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
542  *
543  * The dataset and pool must be held when this function is called.  After it
544  * is called, the pool hold may be released while the dataset is still held
545  * and accessed.
546  */
547 void
548 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
549 {
550 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
551 	(void) refcount_add(&ds->ds_longholds, tag);
552 }
553 
554 void
555 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
556 {
557 	(void) refcount_remove(&ds->ds_longholds, tag);
558 }
559 
560 /* Return B_TRUE if there are any long holds on this dataset. */
561 boolean_t
562 dsl_dataset_long_held(dsl_dataset_t *ds)
563 {
564 	return (!refcount_is_zero(&ds->ds_longholds));
565 }
566 
567 void
568 dsl_dataset_name(dsl_dataset_t *ds, char *name)
569 {
570 	if (ds == NULL) {
571 		(void) strcpy(name, "mos");
572 	} else {
573 		dsl_dir_name(ds->ds_dir, name);
574 		VERIFY0(dsl_dataset_get_snapname(ds));
575 		if (ds->ds_snapname[0]) {
576 			(void) strcat(name, "@");
577 			/*
578 			 * We use a "recursive" mutex so that we
579 			 * can call dprintf_ds() with ds_lock held.
580 			 */
581 			if (!MUTEX_HELD(&ds->ds_lock)) {
582 				mutex_enter(&ds->ds_lock);
583 				(void) strcat(name, ds->ds_snapname);
584 				mutex_exit(&ds->ds_lock);
585 			} else {
586 				(void) strcat(name, ds->ds_snapname);
587 			}
588 		}
589 	}
590 }
591 
592 static int
593 dsl_dataset_namelen(dsl_dataset_t *ds)
594 {
595 	int result;
596 
597 	if (ds == NULL) {
598 		result = 3;	/* "mos" */
599 	} else {
600 		result = dsl_dir_namelen(ds->ds_dir);
601 		VERIFY0(dsl_dataset_get_snapname(ds));
602 		if (ds->ds_snapname[0]) {
603 			++result;	/* adding one for the @-sign */
604 			if (!MUTEX_HELD(&ds->ds_lock)) {
605 				mutex_enter(&ds->ds_lock);
606 				result += strlen(ds->ds_snapname);
607 				mutex_exit(&ds->ds_lock);
608 			} else {
609 				result += strlen(ds->ds_snapname);
610 			}
611 		}
612 	}
613 
614 	return (result);
615 }
616 
617 void
618 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
619 {
620 	dmu_buf_rele(ds->ds_dbuf, tag);
621 }
622 
623 void
624 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
625 {
626 	ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
627 
628 	mutex_enter(&ds->ds_lock);
629 	ds->ds_owner = NULL;
630 	mutex_exit(&ds->ds_lock);
631 	dsl_dataset_long_rele(ds, tag);
632 	if (ds->ds_dbuf != NULL)
633 		dsl_dataset_rele(ds, tag);
634 	else
635 		dsl_dataset_evict(NULL, ds);
636 }
637 
638 boolean_t
639 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
640 {
641 	boolean_t gotit = FALSE;
642 
643 	mutex_enter(&ds->ds_lock);
644 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
645 		ds->ds_owner = tag;
646 		dsl_dataset_long_hold(ds, tag);
647 		gotit = TRUE;
648 	}
649 	mutex_exit(&ds->ds_lock);
650 	return (gotit);
651 }
652 
653 uint64_t
654 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
655     uint64_t flags, dmu_tx_t *tx)
656 {
657 	dsl_pool_t *dp = dd->dd_pool;
658 	dmu_buf_t *dbuf;
659 	dsl_dataset_phys_t *dsphys;
660 	uint64_t dsobj;
661 	objset_t *mos = dp->dp_meta_objset;
662 
663 	if (origin == NULL)
664 		origin = dp->dp_origin_snap;
665 
666 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
667 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
668 	ASSERT(dmu_tx_is_syncing(tx));
669 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
670 
671 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
672 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
673 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
674 	dmu_buf_will_dirty(dbuf, tx);
675 	dsphys = dbuf->db_data;
676 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
677 	dsphys->ds_dir_obj = dd->dd_object;
678 	dsphys->ds_flags = flags;
679 	dsphys->ds_fsid_guid = unique_create();
680 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
681 	    sizeof (dsphys->ds_guid));
682 	dsphys->ds_snapnames_zapobj =
683 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
684 	    DMU_OT_NONE, 0, tx);
685 	dsphys->ds_creation_time = gethrestime_sec();
686 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
687 
688 	if (origin == NULL) {
689 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
690 	} else {
691 		dsl_dataset_t *ohds; /* head of the origin snapshot */
692 
693 		dsphys->ds_prev_snap_obj = origin->ds_object;
694 		dsphys->ds_prev_snap_txg =
695 		    origin->ds_phys->ds_creation_txg;
696 		dsphys->ds_referenced_bytes =
697 		    origin->ds_phys->ds_referenced_bytes;
698 		dsphys->ds_compressed_bytes =
699 		    origin->ds_phys->ds_compressed_bytes;
700 		dsphys->ds_uncompressed_bytes =
701 		    origin->ds_phys->ds_uncompressed_bytes;
702 		dsphys->ds_bp = origin->ds_phys->ds_bp;
703 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
704 
705 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
706 		origin->ds_phys->ds_num_children++;
707 
708 		VERIFY0(dsl_dataset_hold_obj(dp,
709 		    origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
710 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
711 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
712 		dsl_dataset_rele(ohds, FTAG);
713 
714 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
715 			if (origin->ds_phys->ds_next_clones_obj == 0) {
716 				origin->ds_phys->ds_next_clones_obj =
717 				    zap_create(mos,
718 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
719 			}
720 			VERIFY0(zap_add_int(mos,
721 			    origin->ds_phys->ds_next_clones_obj, dsobj, tx));
722 		}
723 
724 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
725 		dd->dd_phys->dd_origin_obj = origin->ds_object;
726 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
727 			if (origin->ds_dir->dd_phys->dd_clones == 0) {
728 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
729 				origin->ds_dir->dd_phys->dd_clones =
730 				    zap_create(mos,
731 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
732 			}
733 			VERIFY0(zap_add_int(mos,
734 			    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
735 		}
736 	}
737 
738 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
739 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
740 
741 	dmu_buf_rele(dbuf, FTAG);
742 
743 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
744 	dd->dd_phys->dd_head_dataset_obj = dsobj;
745 
746 	return (dsobj);
747 }
748 
749 static void
750 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
751 {
752 	objset_t *os;
753 
754 	VERIFY0(dmu_objset_from_ds(ds, &os));
755 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
756 	dsl_dataset_dirty(ds, tx);
757 }
758 
759 uint64_t
760 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
761     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
762 {
763 	dsl_pool_t *dp = pdd->dd_pool;
764 	uint64_t dsobj, ddobj;
765 	dsl_dir_t *dd;
766 
767 	ASSERT(dmu_tx_is_syncing(tx));
768 	ASSERT(lastname[0] != '@');
769 
770 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
771 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
772 
773 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
774 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
775 
776 	dsl_deleg_set_create_perms(dd, tx, cr);
777 
778 	dsl_dir_rele(dd, FTAG);
779 
780 	/*
781 	 * If we are creating a clone, make sure we zero out any stale
782 	 * data from the origin snapshots zil header.
783 	 */
784 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
785 		dsl_dataset_t *ds;
786 
787 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
788 		dsl_dataset_zero_zil(ds, tx);
789 		dsl_dataset_rele(ds, FTAG);
790 	}
791 
792 	return (dsobj);
793 }
794 
795 /*
796  * The unique space in the head dataset can be calculated by subtracting
797  * the space used in the most recent snapshot, that is still being used
798  * in this file system, from the space currently in use.  To figure out
799  * the space in the most recent snapshot still in use, we need to take
800  * the total space used in the snapshot and subtract out the space that
801  * has been freed up since the snapshot was taken.
802  */
803 void
804 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
805 {
806 	uint64_t mrs_used;
807 	uint64_t dlused, dlcomp, dluncomp;
808 
809 	ASSERT(!dsl_dataset_is_snapshot(ds));
810 
811 	if (ds->ds_phys->ds_prev_snap_obj != 0)
812 		mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
813 	else
814 		mrs_used = 0;
815 
816 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
817 
818 	ASSERT3U(dlused, <=, mrs_used);
819 	ds->ds_phys->ds_unique_bytes =
820 	    ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
821 
822 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
823 	    SPA_VERSION_UNIQUE_ACCURATE)
824 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
825 }
826 
827 void
828 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
829     dmu_tx_t *tx)
830 {
831 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
832 	uint64_t count;
833 	int err;
834 
835 	ASSERT(ds->ds_phys->ds_num_children >= 2);
836 	err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
837 	/*
838 	 * The err should not be ENOENT, but a bug in a previous version
839 	 * of the code could cause upgrade_clones_cb() to not set
840 	 * ds_next_snap_obj when it should, leading to a missing entry.
841 	 * If we knew that the pool was created after
842 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
843 	 * ENOENT.  However, at least we can check that we don't have
844 	 * too many entries in the next_clones_obj even after failing to
845 	 * remove this one.
846 	 */
847 	if (err != ENOENT)
848 		VERIFY0(err);
849 	ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
850 	    &count));
851 	ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
852 }
853 
854 
855 blkptr_t *
856 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
857 {
858 	return (&ds->ds_phys->ds_bp);
859 }
860 
861 void
862 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
863 {
864 	ASSERT(dmu_tx_is_syncing(tx));
865 	/* If it's the meta-objset, set dp_meta_rootbp */
866 	if (ds == NULL) {
867 		tx->tx_pool->dp_meta_rootbp = *bp;
868 	} else {
869 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
870 		ds->ds_phys->ds_bp = *bp;
871 	}
872 }
873 
874 spa_t *
875 dsl_dataset_get_spa(dsl_dataset_t *ds)
876 {
877 	return (ds->ds_dir->dd_pool->dp_spa);
878 }
879 
880 void
881 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
882 {
883 	dsl_pool_t *dp;
884 
885 	if (ds == NULL) /* this is the meta-objset */
886 		return;
887 
888 	ASSERT(ds->ds_objset != NULL);
889 
890 	if (ds->ds_phys->ds_next_snap_obj != 0)
891 		panic("dirtying snapshot!");
892 
893 	dp = ds->ds_dir->dd_pool;
894 
895 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
896 		/* up the hold count until we can be written out */
897 		dmu_buf_add_ref(ds->ds_dbuf, ds);
898 	}
899 }
900 
901 boolean_t
902 dsl_dataset_is_dirty(dsl_dataset_t *ds)
903 {
904 	for (int t = 0; t < TXG_SIZE; t++) {
905 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
906 		    ds, t))
907 			return (B_TRUE);
908 	}
909 	return (B_FALSE);
910 }
911 
912 static int
913 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
914 {
915 	uint64_t asize;
916 
917 	if (!dmu_tx_is_syncing(tx))
918 		return (0);
919 
920 	/*
921 	 * If there's an fs-only reservation, any blocks that might become
922 	 * owned by the snapshot dataset must be accommodated by space
923 	 * outside of the reservation.
924 	 */
925 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
926 	asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
927 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
928 		return (SET_ERROR(ENOSPC));
929 
930 	/*
931 	 * Propagate any reserved space for this snapshot to other
932 	 * snapshot checks in this sync group.
933 	 */
934 	if (asize > 0)
935 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
936 
937 	return (0);
938 }
939 
940 typedef struct dsl_dataset_snapshot_arg {
941 	nvlist_t *ddsa_snaps;
942 	nvlist_t *ddsa_props;
943 	nvlist_t *ddsa_errors;
944 } dsl_dataset_snapshot_arg_t;
945 
946 int
947 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
948     dmu_tx_t *tx)
949 {
950 	int error;
951 	uint64_t value;
952 
953 	ds->ds_trysnap_txg = tx->tx_txg;
954 
955 	if (!dmu_tx_is_syncing(tx))
956 		return (0);
957 
958 	/*
959 	 * We don't allow multiple snapshots of the same txg.  If there
960 	 * is already one, try again.
961 	 */
962 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
963 		return (SET_ERROR(EAGAIN));
964 
965 	/*
966 	 * Check for conflicting snapshot name.
967 	 */
968 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
969 	if (error == 0)
970 		return (SET_ERROR(EEXIST));
971 	if (error != ENOENT)
972 		return (error);
973 
974 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
975 	if (error != 0)
976 		return (error);
977 
978 	return (0);
979 }
980 
981 static int
982 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
983 {
984 	dsl_dataset_snapshot_arg_t *ddsa = arg;
985 	dsl_pool_t *dp = dmu_tx_pool(tx);
986 	nvpair_t *pair;
987 	int rv = 0;
988 
989 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
990 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
991 		int error = 0;
992 		dsl_dataset_t *ds;
993 		char *name, *atp;
994 		char dsname[MAXNAMELEN];
995 
996 		name = nvpair_name(pair);
997 		if (strlen(name) >= MAXNAMELEN)
998 			error = SET_ERROR(ENAMETOOLONG);
999 		if (error == 0) {
1000 			atp = strchr(name, '@');
1001 			if (atp == NULL)
1002 				error = SET_ERROR(EINVAL);
1003 			if (error == 0)
1004 				(void) strlcpy(dsname, name, atp - name + 1);
1005 		}
1006 		if (error == 0)
1007 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1008 		if (error == 0) {
1009 			error = dsl_dataset_snapshot_check_impl(ds,
1010 			    atp + 1, tx);
1011 			dsl_dataset_rele(ds, FTAG);
1012 		}
1013 
1014 		if (error != 0) {
1015 			if (ddsa->ddsa_errors != NULL) {
1016 				fnvlist_add_int32(ddsa->ddsa_errors,
1017 				    name, error);
1018 			}
1019 			rv = error;
1020 		}
1021 	}
1022 	return (rv);
1023 }
1024 
1025 void
1026 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1027     dmu_tx_t *tx)
1028 {
1029 	static zil_header_t zero_zil;
1030 
1031 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1032 	dmu_buf_t *dbuf;
1033 	dsl_dataset_phys_t *dsphys;
1034 	uint64_t dsobj, crtxg;
1035 	objset_t *mos = dp->dp_meta_objset;
1036 	objset_t *os;
1037 
1038 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1039 
1040 	/*
1041 	 * If we are on an old pool, the zil must not be active, in which
1042 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1043 	 */
1044 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1045 	    dmu_objset_from_ds(ds, &os) != 0 ||
1046 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1047 	    sizeof (zero_zil)) == 0);
1048 
1049 
1050 	/*
1051 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1052 	 */
1053 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1054 		crtxg = 1;
1055 	else
1056 		crtxg = tx->tx_txg;
1057 
1058 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1059 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1060 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1061 	dmu_buf_will_dirty(dbuf, tx);
1062 	dsphys = dbuf->db_data;
1063 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1064 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1065 	dsphys->ds_fsid_guid = unique_create();
1066 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1067 	    sizeof (dsphys->ds_guid));
1068 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1069 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1070 	dsphys->ds_next_snap_obj = ds->ds_object;
1071 	dsphys->ds_num_children = 1;
1072 	dsphys->ds_creation_time = gethrestime_sec();
1073 	dsphys->ds_creation_txg = crtxg;
1074 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1075 	dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1076 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1077 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1078 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1079 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1080 	dmu_buf_rele(dbuf, FTAG);
1081 
1082 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1083 	if (ds->ds_prev) {
1084 		uint64_t next_clones_obj =
1085 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1086 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1087 		    ds->ds_object ||
1088 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1089 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1090 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1091 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1092 			    ds->ds_prev->ds_phys->ds_creation_txg);
1093 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1094 		} else if (next_clones_obj != 0) {
1095 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1096 			    dsphys->ds_next_snap_obj, tx);
1097 			VERIFY0(zap_add_int(mos,
1098 			    next_clones_obj, dsobj, tx));
1099 		}
1100 	}
1101 
1102 	/*
1103 	 * If we have a reference-reservation on this dataset, we will
1104 	 * need to increase the amount of refreservation being charged
1105 	 * since our unique space is going to zero.
1106 	 */
1107 	if (ds->ds_reserved) {
1108 		int64_t delta;
1109 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1110 		delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1111 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1112 		    delta, 0, 0, tx);
1113 	}
1114 
1115 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1116 	ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1117 	    UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1118 	dsl_deadlist_close(&ds->ds_deadlist);
1119 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1120 	dsl_deadlist_add_key(&ds->ds_deadlist,
1121 	    ds->ds_phys->ds_prev_snap_txg, tx);
1122 
1123 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1124 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1125 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1126 	ds->ds_phys->ds_unique_bytes = 0;
1127 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1128 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1129 
1130 	VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1131 	    snapname, 8, 1, &dsobj, tx));
1132 
1133 	if (ds->ds_prev)
1134 		dsl_dataset_rele(ds->ds_prev, ds);
1135 	VERIFY0(dsl_dataset_hold_obj(dp,
1136 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1137 
1138 	dsl_scan_ds_snapshotted(ds, tx);
1139 
1140 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1141 
1142 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1143 }
1144 
1145 static void
1146 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1147 {
1148 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1149 	dsl_pool_t *dp = dmu_tx_pool(tx);
1150 	nvpair_t *pair;
1151 
1152 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1153 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1154 		dsl_dataset_t *ds;
1155 		char *name, *atp;
1156 		char dsname[MAXNAMELEN];
1157 
1158 		name = nvpair_name(pair);
1159 		atp = strchr(name, '@');
1160 		(void) strlcpy(dsname, name, atp - name + 1);
1161 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1162 
1163 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1164 		if (ddsa->ddsa_props != NULL) {
1165 			dsl_props_set_sync_impl(ds->ds_prev,
1166 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1167 		}
1168 		dsl_dataset_rele(ds, FTAG);
1169 	}
1170 }
1171 
1172 /*
1173  * The snapshots must all be in the same pool.
1174  * All-or-nothing: if there are any failures, nothing will be modified.
1175  */
1176 int
1177 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1178 {
1179 	dsl_dataset_snapshot_arg_t ddsa;
1180 	nvpair_t *pair;
1181 	boolean_t needsuspend;
1182 	int error;
1183 	spa_t *spa;
1184 	char *firstname;
1185 	nvlist_t *suspended = NULL;
1186 
1187 	pair = nvlist_next_nvpair(snaps, NULL);
1188 	if (pair == NULL)
1189 		return (0);
1190 	firstname = nvpair_name(pair);
1191 
1192 	error = spa_open(firstname, &spa, FTAG);
1193 	if (error != 0)
1194 		return (error);
1195 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1196 	spa_close(spa, FTAG);
1197 
1198 	if (needsuspend) {
1199 		suspended = fnvlist_alloc();
1200 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1201 		    pair = nvlist_next_nvpair(snaps, pair)) {
1202 			char fsname[MAXNAMELEN];
1203 			char *snapname = nvpair_name(pair);
1204 			char *atp;
1205 			void *cookie;
1206 
1207 			atp = strchr(snapname, '@');
1208 			if (atp == NULL) {
1209 				error = SET_ERROR(EINVAL);
1210 				break;
1211 			}
1212 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1213 
1214 			error = zil_suspend(fsname, &cookie);
1215 			if (error != 0)
1216 				break;
1217 			fnvlist_add_uint64(suspended, fsname,
1218 			    (uintptr_t)cookie);
1219 		}
1220 	}
1221 
1222 	ddsa.ddsa_snaps = snaps;
1223 	ddsa.ddsa_props = props;
1224 	ddsa.ddsa_errors = errors;
1225 
1226 	if (error == 0) {
1227 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1228 		    dsl_dataset_snapshot_sync, &ddsa,
1229 		    fnvlist_num_pairs(snaps) * 3);
1230 	}
1231 
1232 	if (suspended != NULL) {
1233 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1234 		    pair = nvlist_next_nvpair(suspended, pair)) {
1235 			zil_resume((void *)(uintptr_t)
1236 			    fnvpair_value_uint64(pair));
1237 		}
1238 		fnvlist_free(suspended);
1239 	}
1240 
1241 	return (error);
1242 }
1243 
1244 typedef struct dsl_dataset_snapshot_tmp_arg {
1245 	const char *ddsta_fsname;
1246 	const char *ddsta_snapname;
1247 	minor_t ddsta_cleanup_minor;
1248 	const char *ddsta_htag;
1249 } dsl_dataset_snapshot_tmp_arg_t;
1250 
1251 static int
1252 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1253 {
1254 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1255 	dsl_pool_t *dp = dmu_tx_pool(tx);
1256 	dsl_dataset_t *ds;
1257 	int error;
1258 
1259 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1260 	if (error != 0)
1261 		return (error);
1262 
1263 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, tx);
1264 	if (error != 0) {
1265 		dsl_dataset_rele(ds, FTAG);
1266 		return (error);
1267 	}
1268 
1269 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1270 		dsl_dataset_rele(ds, FTAG);
1271 		return (SET_ERROR(ENOTSUP));
1272 	}
1273 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1274 	    B_TRUE, tx);
1275 	if (error != 0) {
1276 		dsl_dataset_rele(ds, FTAG);
1277 		return (error);
1278 	}
1279 
1280 	dsl_dataset_rele(ds, FTAG);
1281 	return (0);
1282 }
1283 
1284 static void
1285 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1286 {
1287 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1288 	dsl_pool_t *dp = dmu_tx_pool(tx);
1289 	dsl_dataset_t *ds;
1290 
1291 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1292 
1293 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1294 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1295 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1296 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1297 
1298 	dsl_dataset_rele(ds, FTAG);
1299 }
1300 
1301 int
1302 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1303     minor_t cleanup_minor, const char *htag)
1304 {
1305 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1306 	int error;
1307 	spa_t *spa;
1308 	boolean_t needsuspend;
1309 	void *cookie;
1310 
1311 	ddsta.ddsta_fsname = fsname;
1312 	ddsta.ddsta_snapname = snapname;
1313 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1314 	ddsta.ddsta_htag = htag;
1315 
1316 	error = spa_open(fsname, &spa, FTAG);
1317 	if (error != 0)
1318 		return (error);
1319 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1320 	spa_close(spa, FTAG);
1321 
1322 	if (needsuspend) {
1323 		error = zil_suspend(fsname, &cookie);
1324 		if (error != 0)
1325 			return (error);
1326 	}
1327 
1328 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1329 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1330 
1331 	if (needsuspend)
1332 		zil_resume(cookie);
1333 	return (error);
1334 }
1335 
1336 
1337 void
1338 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1339 {
1340 	ASSERT(dmu_tx_is_syncing(tx));
1341 	ASSERT(ds->ds_objset != NULL);
1342 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1343 
1344 	/*
1345 	 * in case we had to change ds_fsid_guid when we opened it,
1346 	 * sync it out now.
1347 	 */
1348 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1349 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1350 
1351 	dmu_objset_sync(ds->ds_objset, zio, tx);
1352 }
1353 
1354 static void
1355 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1356 {
1357 	uint64_t count = 0;
1358 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1359 	zap_cursor_t zc;
1360 	zap_attribute_t za;
1361 	nvlist_t *propval = fnvlist_alloc();
1362 	nvlist_t *val = fnvlist_alloc();
1363 
1364 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1365 
1366 	/*
1367 	 * There may be missing entries in ds_next_clones_obj
1368 	 * due to a bug in a previous version of the code.
1369 	 * Only trust it if it has the right number of entries.
1370 	 */
1371 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1372 		ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1373 		    &count));
1374 	}
1375 	if (count != ds->ds_phys->ds_num_children - 1)
1376 		goto fail;
1377 	for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1378 	    zap_cursor_retrieve(&zc, &za) == 0;
1379 	    zap_cursor_advance(&zc)) {
1380 		dsl_dataset_t *clone;
1381 		char buf[ZFS_MAXNAMELEN];
1382 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1383 		    za.za_first_integer, FTAG, &clone));
1384 		dsl_dir_name(clone->ds_dir, buf);
1385 		fnvlist_add_boolean(val, buf);
1386 		dsl_dataset_rele(clone, FTAG);
1387 	}
1388 	zap_cursor_fini(&zc);
1389 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1390 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1391 fail:
1392 	nvlist_free(val);
1393 	nvlist_free(propval);
1394 }
1395 
1396 void
1397 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1398 {
1399 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1400 	uint64_t refd, avail, uobjs, aobjs, ratio;
1401 
1402 	ASSERT(dsl_pool_config_held(dp));
1403 
1404 	ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1405 	    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1406 	    ds->ds_phys->ds_compressed_bytes);
1407 
1408 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1409 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1410 	    ds->ds_phys->ds_uncompressed_bytes);
1411 
1412 	if (dsl_dataset_is_snapshot(ds)) {
1413 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1414 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1415 		    ds->ds_phys->ds_unique_bytes);
1416 		get_clones_stat(ds, nv);
1417 	} else {
1418 		dsl_dir_stats(ds->ds_dir, nv);
1419 	}
1420 
1421 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1422 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1423 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1424 
1425 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1426 	    ds->ds_phys->ds_creation_time);
1427 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1428 	    ds->ds_phys->ds_creation_txg);
1429 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1430 	    ds->ds_quota);
1431 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1432 	    ds->ds_reserved);
1433 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1434 	    ds->ds_phys->ds_guid);
1435 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1436 	    ds->ds_phys->ds_unique_bytes);
1437 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1438 	    ds->ds_object);
1439 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1440 	    ds->ds_userrefs);
1441 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1442 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1443 
1444 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1445 		uint64_t written, comp, uncomp;
1446 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1447 		dsl_dataset_t *prev;
1448 
1449 		int err = dsl_dataset_hold_obj(dp,
1450 		    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1451 		if (err == 0) {
1452 			err = dsl_dataset_space_written(prev, ds, &written,
1453 			    &comp, &uncomp);
1454 			dsl_dataset_rele(prev, FTAG);
1455 			if (err == 0) {
1456 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1457 				    written);
1458 			}
1459 		}
1460 	}
1461 }
1462 
1463 void
1464 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1465 {
1466 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1467 	ASSERT(dsl_pool_config_held(dp));
1468 
1469 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1470 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1471 	stat->dds_guid = ds->ds_phys->ds_guid;
1472 	stat->dds_origin[0] = '\0';
1473 	if (dsl_dataset_is_snapshot(ds)) {
1474 		stat->dds_is_snapshot = B_TRUE;
1475 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1476 	} else {
1477 		stat->dds_is_snapshot = B_FALSE;
1478 		stat->dds_num_clones = 0;
1479 
1480 		if (dsl_dir_is_clone(ds->ds_dir)) {
1481 			dsl_dataset_t *ods;
1482 
1483 			VERIFY0(dsl_dataset_hold_obj(dp,
1484 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1485 			dsl_dataset_name(ods, stat->dds_origin);
1486 			dsl_dataset_rele(ods, FTAG);
1487 		}
1488 	}
1489 }
1490 
1491 uint64_t
1492 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1493 {
1494 	return (ds->ds_fsid_guid);
1495 }
1496 
1497 void
1498 dsl_dataset_space(dsl_dataset_t *ds,
1499     uint64_t *refdbytesp, uint64_t *availbytesp,
1500     uint64_t *usedobjsp, uint64_t *availobjsp)
1501 {
1502 	*refdbytesp = ds->ds_phys->ds_referenced_bytes;
1503 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1504 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1505 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1506 	if (ds->ds_quota != 0) {
1507 		/*
1508 		 * Adjust available bytes according to refquota
1509 		 */
1510 		if (*refdbytesp < ds->ds_quota)
1511 			*availbytesp = MIN(*availbytesp,
1512 			    ds->ds_quota - *refdbytesp);
1513 		else
1514 			*availbytesp = 0;
1515 	}
1516 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1517 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1518 }
1519 
1520 boolean_t
1521 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1522 {
1523 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1524 
1525 	ASSERT(dsl_pool_config_held(dp));
1526 	if (ds->ds_prev == NULL)
1527 		return (B_FALSE);
1528 	if (ds->ds_phys->ds_bp.blk_birth >
1529 	    ds->ds_prev->ds_phys->ds_creation_txg) {
1530 		objset_t *os, *os_prev;
1531 		/*
1532 		 * It may be that only the ZIL differs, because it was
1533 		 * reset in the head.  Don't count that as being
1534 		 * modified.
1535 		 */
1536 		if (dmu_objset_from_ds(ds, &os) != 0)
1537 			return (B_TRUE);
1538 		if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
1539 			return (B_TRUE);
1540 		return (bcmp(&os->os_phys->os_meta_dnode,
1541 		    &os_prev->os_phys->os_meta_dnode,
1542 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1543 	}
1544 	return (B_FALSE);
1545 }
1546 
1547 typedef struct dsl_dataset_rename_snapshot_arg {
1548 	const char *ddrsa_fsname;
1549 	const char *ddrsa_oldsnapname;
1550 	const char *ddrsa_newsnapname;
1551 	boolean_t ddrsa_recursive;
1552 	dmu_tx_t *ddrsa_tx;
1553 } dsl_dataset_rename_snapshot_arg_t;
1554 
1555 /* ARGSUSED */
1556 static int
1557 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1558     dsl_dataset_t *hds, void *arg)
1559 {
1560 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1561 	int error;
1562 	uint64_t val;
1563 
1564 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1565 	if (error != 0) {
1566 		/* ignore nonexistent snapshots */
1567 		return (error == ENOENT ? 0 : error);
1568 	}
1569 
1570 	/* new name should not exist */
1571 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1572 	if (error == 0)
1573 		error = SET_ERROR(EEXIST);
1574 	else if (error == ENOENT)
1575 		error = 0;
1576 
1577 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1578 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1579 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1580 		error = SET_ERROR(ENAMETOOLONG);
1581 
1582 	return (error);
1583 }
1584 
1585 static int
1586 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1587 {
1588 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1589 	dsl_pool_t *dp = dmu_tx_pool(tx);
1590 	dsl_dataset_t *hds;
1591 	int error;
1592 
1593 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1594 	if (error != 0)
1595 		return (error);
1596 
1597 	if (ddrsa->ddrsa_recursive) {
1598 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1599 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1600 		    DS_FIND_CHILDREN);
1601 	} else {
1602 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1603 	}
1604 	dsl_dataset_rele(hds, FTAG);
1605 	return (error);
1606 }
1607 
1608 static int
1609 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1610     dsl_dataset_t *hds, void *arg)
1611 {
1612 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1613 	dsl_dataset_t *ds;
1614 	uint64_t val;
1615 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1616 	int error;
1617 
1618 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1619 	ASSERT(error == 0 || error == ENOENT);
1620 	if (error == ENOENT) {
1621 		/* ignore nonexistent snapshots */
1622 		return (0);
1623 	}
1624 
1625 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1626 
1627 	/* log before we change the name */
1628 	spa_history_log_internal_ds(ds, "rename", tx,
1629 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1630 
1631 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx));
1632 	mutex_enter(&ds->ds_lock);
1633 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1634 	mutex_exit(&ds->ds_lock);
1635 	VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1636 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1637 
1638 	dsl_dataset_rele(ds, FTAG);
1639 	return (0);
1640 }
1641 
1642 static void
1643 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1644 {
1645 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1646 	dsl_pool_t *dp = dmu_tx_pool(tx);
1647 	dsl_dataset_t *hds;
1648 
1649 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1650 	ddrsa->ddrsa_tx = tx;
1651 	if (ddrsa->ddrsa_recursive) {
1652 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1653 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1654 		    DS_FIND_CHILDREN));
1655 	} else {
1656 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1657 	}
1658 	dsl_dataset_rele(hds, FTAG);
1659 }
1660 
1661 int
1662 dsl_dataset_rename_snapshot(const char *fsname,
1663     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1664 {
1665 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1666 
1667 	ddrsa.ddrsa_fsname = fsname;
1668 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1669 	ddrsa.ddrsa_newsnapname = newsnapname;
1670 	ddrsa.ddrsa_recursive = recursive;
1671 
1672 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1673 	    dsl_dataset_rename_snapshot_sync, &ddrsa, 1));
1674 }
1675 
1676 static int
1677 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1678 {
1679 	const char *fsname = arg;
1680 	dsl_pool_t *dp = dmu_tx_pool(tx);
1681 	dsl_dataset_t *ds;
1682 	int64_t unused_refres_delta;
1683 	int error;
1684 
1685 	error = dsl_dataset_hold(dp, fsname, FTAG, &ds);
1686 	if (error != 0)
1687 		return (error);
1688 
1689 	/* must not be a snapshot */
1690 	if (dsl_dataset_is_snapshot(ds)) {
1691 		dsl_dataset_rele(ds, FTAG);
1692 		return (SET_ERROR(EINVAL));
1693 	}
1694 
1695 	/* must have a most recent snapshot */
1696 	if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1697 		dsl_dataset_rele(ds, FTAG);
1698 		return (SET_ERROR(EINVAL));
1699 	}
1700 
1701 	if (dsl_dataset_long_held(ds)) {
1702 		dsl_dataset_rele(ds, FTAG);
1703 		return (SET_ERROR(EBUSY));
1704 	}
1705 
1706 	/*
1707 	 * Check if the snap we are rolling back to uses more than
1708 	 * the refquota.
1709 	 */
1710 	if (ds->ds_quota != 0 &&
1711 	    ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1712 		dsl_dataset_rele(ds, FTAG);
1713 		return (SET_ERROR(EDQUOT));
1714 	}
1715 
1716 	/*
1717 	 * When we do the clone swap, we will temporarily use more space
1718 	 * due to the refreservation (the head will no longer have any
1719 	 * unique space, so the entire amount of the refreservation will need
1720 	 * to be free).  We will immediately destroy the clone, freeing
1721 	 * this space, but the freeing happens over many txg's.
1722 	 */
1723 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1724 	    ds->ds_phys->ds_unique_bytes);
1725 
1726 	if (unused_refres_delta > 0 &&
1727 	    unused_refres_delta >
1728 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1729 		dsl_dataset_rele(ds, FTAG);
1730 		return (SET_ERROR(ENOSPC));
1731 	}
1732 
1733 	dsl_dataset_rele(ds, FTAG);
1734 	return (0);
1735 }
1736 
1737 static void
1738 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1739 {
1740 	const char *fsname = arg;
1741 	dsl_pool_t *dp = dmu_tx_pool(tx);
1742 	dsl_dataset_t *ds, *clone;
1743 	uint64_t cloneobj;
1744 
1745 	VERIFY0(dsl_dataset_hold(dp, fsname, FTAG, &ds));
1746 
1747 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1748 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1749 
1750 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1751 
1752 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1753 	dsl_dataset_zero_zil(ds, tx);
1754 
1755 	dsl_destroy_head_sync_impl(clone, tx);
1756 
1757 	dsl_dataset_rele(clone, FTAG);
1758 	dsl_dataset_rele(ds, FTAG);
1759 }
1760 
1761 int
1762 dsl_dataset_rollback(const char *fsname)
1763 {
1764 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1765 	    dsl_dataset_rollback_sync, (void *)fsname, 1));
1766 }
1767 
1768 struct promotenode {
1769 	list_node_t link;
1770 	dsl_dataset_t *ds;
1771 };
1772 
1773 typedef struct dsl_dataset_promote_arg {
1774 	const char *ddpa_clonename;
1775 	dsl_dataset_t *ddpa_clone;
1776 	list_t shared_snaps, origin_snaps, clone_snaps;
1777 	dsl_dataset_t *origin_origin; /* origin of the origin */
1778 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
1779 	char *err_ds;
1780 } dsl_dataset_promote_arg_t;
1781 
1782 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
1783 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
1784     void *tag);
1785 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
1786 
1787 static int
1788 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
1789 {
1790 	dsl_dataset_promote_arg_t *ddpa = arg;
1791 	dsl_pool_t *dp = dmu_tx_pool(tx);
1792 	dsl_dataset_t *hds;
1793 	struct promotenode *snap;
1794 	dsl_dataset_t *origin_ds;
1795 	int err;
1796 	uint64_t unused;
1797 
1798 	err = promote_hold(ddpa, dp, FTAG);
1799 	if (err != 0)
1800 		return (err);
1801 
1802 	hds = ddpa->ddpa_clone;
1803 
1804 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1805 		promote_rele(ddpa, FTAG);
1806 		return (SET_ERROR(EXDEV));
1807 	}
1808 
1809 	/*
1810 	 * Compute and check the amount of space to transfer.  Since this is
1811 	 * so expensive, don't do the preliminary check.
1812 	 */
1813 	if (!dmu_tx_is_syncing(tx)) {
1814 		promote_rele(ddpa, FTAG);
1815 		return (0);
1816 	}
1817 
1818 	snap = list_head(&ddpa->shared_snaps);
1819 	origin_ds = snap->ds;
1820 
1821 	/* compute origin's new unique space */
1822 	snap = list_tail(&ddpa->clone_snaps);
1823 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1824 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
1825 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1826 	    &ddpa->unique, &unused, &unused);
1827 
1828 	/*
1829 	 * Walk the snapshots that we are moving
1830 	 *
1831 	 * Compute space to transfer.  Consider the incremental changes
1832 	 * to used by each snapshot:
1833 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1834 	 * So each snapshot gave birth to:
1835 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1836 	 * So a sequence would look like:
1837 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1838 	 * Which simplifies to:
1839 	 * uN + kN + kN-1 + ... + k1 + k0
1840 	 * Note however, if we stop before we reach the ORIGIN we get:
1841 	 * uN + kN + kN-1 + ... + kM - uM-1
1842 	 */
1843 	ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
1844 	ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
1845 	ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
1846 	for (snap = list_head(&ddpa->shared_snaps); snap;
1847 	    snap = list_next(&ddpa->shared_snaps, snap)) {
1848 		uint64_t val, dlused, dlcomp, dluncomp;
1849 		dsl_dataset_t *ds = snap->ds;
1850 
1851 		/*
1852 		 * If there are long holds, we won't be able to evict
1853 		 * the objset.
1854 		 */
1855 		if (dsl_dataset_long_held(ds)) {
1856 			err = SET_ERROR(EBUSY);
1857 			goto out;
1858 		}
1859 
1860 		/* Check that the snapshot name does not conflict */
1861 		VERIFY0(dsl_dataset_get_snapname(ds));
1862 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
1863 		if (err == 0) {
1864 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
1865 			err = SET_ERROR(EEXIST);
1866 			goto out;
1867 		}
1868 		if (err != ENOENT)
1869 			goto out;
1870 
1871 		/* The very first snapshot does not have a deadlist */
1872 		if (ds->ds_phys->ds_prev_snap_obj == 0)
1873 			continue;
1874 
1875 		dsl_deadlist_space(&ds->ds_deadlist,
1876 		    &dlused, &dlcomp, &dluncomp);
1877 		ddpa->used += dlused;
1878 		ddpa->comp += dlcomp;
1879 		ddpa->uncomp += dluncomp;
1880 	}
1881 
1882 	/*
1883 	 * If we are a clone of a clone then we never reached ORIGIN,
1884 	 * so we need to subtract out the clone origin's used space.
1885 	 */
1886 	if (ddpa->origin_origin) {
1887 		ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
1888 		ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
1889 		ddpa->uncomp -=
1890 		    ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
1891 	}
1892 
1893 	/* Check that there is enough space here */
1894 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
1895 	    ddpa->used);
1896 	if (err != 0)
1897 		goto out;
1898 
1899 	/*
1900 	 * Compute the amounts of space that will be used by snapshots
1901 	 * after the promotion (for both origin and clone).  For each,
1902 	 * it is the amount of space that will be on all of their
1903 	 * deadlists (that was not born before their new origin).
1904 	 */
1905 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1906 		uint64_t space;
1907 
1908 		/*
1909 		 * Note, typically this will not be a clone of a clone,
1910 		 * so dd_origin_txg will be < TXG_INITIAL, so
1911 		 * these snaplist_space() -> dsl_deadlist_space_range()
1912 		 * calls will be fast because they do not have to
1913 		 * iterate over all bps.
1914 		 */
1915 		snap = list_head(&ddpa->origin_snaps);
1916 		err = snaplist_space(&ddpa->shared_snaps,
1917 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
1918 		if (err != 0)
1919 			goto out;
1920 
1921 		err = snaplist_space(&ddpa->clone_snaps,
1922 		    snap->ds->ds_dir->dd_origin_txg, &space);
1923 		if (err != 0)
1924 			goto out;
1925 		ddpa->cloneusedsnap += space;
1926 	}
1927 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1928 		err = snaplist_space(&ddpa->origin_snaps,
1929 		    origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
1930 		if (err != 0)
1931 			goto out;
1932 	}
1933 
1934 out:
1935 	promote_rele(ddpa, FTAG);
1936 	return (err);
1937 }
1938 
1939 static void
1940 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
1941 {
1942 	dsl_dataset_promote_arg_t *ddpa = arg;
1943 	dsl_pool_t *dp = dmu_tx_pool(tx);
1944 	dsl_dataset_t *hds;
1945 	struct promotenode *snap;
1946 	dsl_dataset_t *origin_ds;
1947 	dsl_dataset_t *origin_head;
1948 	dsl_dir_t *dd;
1949 	dsl_dir_t *odd = NULL;
1950 	uint64_t oldnext_obj;
1951 	int64_t delta;
1952 
1953 	VERIFY0(promote_hold(ddpa, dp, FTAG));
1954 	hds = ddpa->ddpa_clone;
1955 
1956 	ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
1957 
1958 	snap = list_head(&ddpa->shared_snaps);
1959 	origin_ds = snap->ds;
1960 	dd = hds->ds_dir;
1961 
1962 	snap = list_head(&ddpa->origin_snaps);
1963 	origin_head = snap->ds;
1964 
1965 	/*
1966 	 * We need to explicitly open odd, since origin_ds's dd will be
1967 	 * changing.
1968 	 */
1969 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
1970 	    NULL, FTAG, &odd));
1971 
1972 	/* change origin's next snap */
1973 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
1974 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
1975 	snap = list_tail(&ddpa->clone_snaps);
1976 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1977 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
1978 
1979 	/* change the origin's next clone */
1980 	if (origin_ds->ds_phys->ds_next_clones_obj) {
1981 		dsl_dataset_remove_from_next_clones(origin_ds,
1982 		    snap->ds->ds_object, tx);
1983 		VERIFY0(zap_add_int(dp->dp_meta_objset,
1984 		    origin_ds->ds_phys->ds_next_clones_obj,
1985 		    oldnext_obj, tx));
1986 	}
1987 
1988 	/* change origin */
1989 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1990 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
1991 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
1992 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
1993 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
1994 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
1995 	origin_head->ds_dir->dd_origin_txg =
1996 	    origin_ds->ds_phys->ds_creation_txg;
1997 
1998 	/* change dd_clone entries */
1999 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2000 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2001 		    odd->dd_phys->dd_clones, hds->ds_object, tx));
2002 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2003 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2004 		    hds->ds_object, tx));
2005 
2006 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2007 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2008 		    origin_head->ds_object, tx));
2009 		if (dd->dd_phys->dd_clones == 0) {
2010 			dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2011 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2012 		}
2013 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2014 		    dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2015 	}
2016 
2017 	/* move snapshots to this dir */
2018 	for (snap = list_head(&ddpa->shared_snaps); snap;
2019 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2020 		dsl_dataset_t *ds = snap->ds;
2021 
2022 		/*
2023 		 * Property callbacks are registered to a particular
2024 		 * dsl_dir.  Since ours is changing, evict the objset
2025 		 * so that they will be unregistered from the old dsl_dir.
2026 		 */
2027 		if (ds->ds_objset) {
2028 			dmu_objset_evict(ds->ds_objset);
2029 			ds->ds_objset = NULL;
2030 		}
2031 
2032 		/* move snap name entry */
2033 		VERIFY0(dsl_dataset_get_snapname(ds));
2034 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2035 		    ds->ds_snapname, tx));
2036 		VERIFY0(zap_add(dp->dp_meta_objset,
2037 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2038 		    8, 1, &ds->ds_object, tx));
2039 
2040 		/* change containing dsl_dir */
2041 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2042 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2043 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2044 		ASSERT3P(ds->ds_dir, ==, odd);
2045 		dsl_dir_rele(ds->ds_dir, ds);
2046 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2047 		    NULL, ds, &ds->ds_dir));
2048 
2049 		/* move any clone references */
2050 		if (ds->ds_phys->ds_next_clones_obj &&
2051 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2052 			zap_cursor_t zc;
2053 			zap_attribute_t za;
2054 
2055 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2056 			    ds->ds_phys->ds_next_clones_obj);
2057 			    zap_cursor_retrieve(&zc, &za) == 0;
2058 			    zap_cursor_advance(&zc)) {
2059 				dsl_dataset_t *cnds;
2060 				uint64_t o;
2061 
2062 				if (za.za_first_integer == oldnext_obj) {
2063 					/*
2064 					 * We've already moved the
2065 					 * origin's reference.
2066 					 */
2067 					continue;
2068 				}
2069 
2070 				VERIFY0(dsl_dataset_hold_obj(dp,
2071 				    za.za_first_integer, FTAG, &cnds));
2072 				o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2073 
2074 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2075 				    odd->dd_phys->dd_clones, o, tx));
2076 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2077 				    dd->dd_phys->dd_clones, o, tx));
2078 				dsl_dataset_rele(cnds, FTAG);
2079 			}
2080 			zap_cursor_fini(&zc);
2081 		}
2082 
2083 		ASSERT(!dsl_prop_hascb(ds));
2084 	}
2085 
2086 	/*
2087 	 * Change space accounting.
2088 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2089 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2090 	 * is true for each of {clone,origin} independently.
2091 	 */
2092 
2093 	delta = ddpa->cloneusedsnap -
2094 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2095 	ASSERT3S(delta, >=, 0);
2096 	ASSERT3U(ddpa->used, >=, delta);
2097 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2098 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2099 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2100 
2101 	delta = ddpa->originusedsnap -
2102 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2103 	ASSERT3S(delta, <=, 0);
2104 	ASSERT3U(ddpa->used, >=, -delta);
2105 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2106 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2107 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2108 
2109 	origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2110 
2111 	/* log history record */
2112 	spa_history_log_internal_ds(hds, "promote", tx, "");
2113 
2114 	dsl_dir_rele(odd, FTAG);
2115 	promote_rele(ddpa, FTAG);
2116 }
2117 
2118 /*
2119  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2120  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2121  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2122  * snapshots back to this dataset's origin.
2123  */
2124 static int
2125 snaplist_make(dsl_pool_t *dp,
2126     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2127 {
2128 	uint64_t obj = last_obj;
2129 
2130 	list_create(l, sizeof (struct promotenode),
2131 	    offsetof(struct promotenode, link));
2132 
2133 	while (obj != first_obj) {
2134 		dsl_dataset_t *ds;
2135 		struct promotenode *snap;
2136 		int err;
2137 
2138 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2139 		ASSERT(err != ENOENT);
2140 		if (err != 0)
2141 			return (err);
2142 
2143 		if (first_obj == 0)
2144 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2145 
2146 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2147 		snap->ds = ds;
2148 		list_insert_tail(l, snap);
2149 		obj = ds->ds_phys->ds_prev_snap_obj;
2150 	}
2151 
2152 	return (0);
2153 }
2154 
2155 static int
2156 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2157 {
2158 	struct promotenode *snap;
2159 
2160 	*spacep = 0;
2161 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2162 		uint64_t used, comp, uncomp;
2163 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2164 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2165 		*spacep += used;
2166 	}
2167 	return (0);
2168 }
2169 
2170 static void
2171 snaplist_destroy(list_t *l, void *tag)
2172 {
2173 	struct promotenode *snap;
2174 
2175 	if (l == NULL || !list_link_active(&l->list_head))
2176 		return;
2177 
2178 	while ((snap = list_tail(l)) != NULL) {
2179 		list_remove(l, snap);
2180 		dsl_dataset_rele(snap->ds, tag);
2181 		kmem_free(snap, sizeof (*snap));
2182 	}
2183 	list_destroy(l);
2184 }
2185 
2186 static int
2187 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2188 {
2189 	int error;
2190 	dsl_dir_t *dd;
2191 	struct promotenode *snap;
2192 
2193 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2194 	    &ddpa->ddpa_clone);
2195 	if (error != 0)
2196 		return (error);
2197 	dd = ddpa->ddpa_clone->ds_dir;
2198 
2199 	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2200 	    !dsl_dir_is_clone(dd)) {
2201 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2202 		return (SET_ERROR(EINVAL));
2203 	}
2204 
2205 	error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2206 	    &ddpa->shared_snaps, tag);
2207 	if (error != 0)
2208 		goto out;
2209 
2210 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2211 	    &ddpa->clone_snaps, tag);
2212 	if (error != 0)
2213 		goto out;
2214 
2215 	snap = list_head(&ddpa->shared_snaps);
2216 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2217 	error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2218 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2219 	    &ddpa->origin_snaps, tag);
2220 	if (error != 0)
2221 		goto out;
2222 
2223 	if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2224 		error = dsl_dataset_hold_obj(dp,
2225 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2226 		    tag, &ddpa->origin_origin);
2227 		if (error != 0)
2228 			goto out;
2229 	}
2230 out:
2231 	if (error != 0)
2232 		promote_rele(ddpa, tag);
2233 	return (error);
2234 }
2235 
2236 static void
2237 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2238 {
2239 	snaplist_destroy(&ddpa->shared_snaps, tag);
2240 	snaplist_destroy(&ddpa->clone_snaps, tag);
2241 	snaplist_destroy(&ddpa->origin_snaps, tag);
2242 	if (ddpa->origin_origin != NULL)
2243 		dsl_dataset_rele(ddpa->origin_origin, tag);
2244 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2245 }
2246 
2247 /*
2248  * Promote a clone.
2249  *
2250  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2251  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2252  */
2253 int
2254 dsl_dataset_promote(const char *name, char *conflsnap)
2255 {
2256 	dsl_dataset_promote_arg_t ddpa = { 0 };
2257 	uint64_t numsnaps;
2258 	int error;
2259 	objset_t *os;
2260 
2261 	/*
2262 	 * We will modify space proportional to the number of
2263 	 * snapshots.  Compute numsnaps.
2264 	 */
2265 	error = dmu_objset_hold(name, FTAG, &os);
2266 	if (error != 0)
2267 		return (error);
2268 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2269 	    dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2270 	dmu_objset_rele(os, FTAG);
2271 	if (error != 0)
2272 		return (error);
2273 
2274 	ddpa.ddpa_clonename = name;
2275 	ddpa.err_ds = conflsnap;
2276 
2277 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2278 	    dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2279 }
2280 
2281 int
2282 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2283     dsl_dataset_t *origin_head, boolean_t force)
2284 {
2285 	int64_t unused_refres_delta;
2286 
2287 	/* they should both be heads */
2288 	if (dsl_dataset_is_snapshot(clone) ||
2289 	    dsl_dataset_is_snapshot(origin_head))
2290 		return (SET_ERROR(EINVAL));
2291 
2292 	/* the branch point should be just before them */
2293 	if (clone->ds_prev != origin_head->ds_prev)
2294 		return (SET_ERROR(EINVAL));
2295 
2296 	/* clone should be the clone (unless they are unrelated) */
2297 	if (clone->ds_prev != NULL &&
2298 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2299 	    origin_head->ds_object !=
2300 	    clone->ds_prev->ds_phys->ds_next_snap_obj)
2301 		return (SET_ERROR(EINVAL));
2302 
2303 	/* the clone should be a child of the origin */
2304 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2305 		return (SET_ERROR(EINVAL));
2306 
2307 	/* origin_head shouldn't be modified unless 'force' */
2308 	if (!force && dsl_dataset_modified_since_lastsnap(origin_head))
2309 		return (SET_ERROR(ETXTBSY));
2310 
2311 	/* origin_head should have no long holds (e.g. is not mounted) */
2312 	if (dsl_dataset_long_held(origin_head))
2313 		return (SET_ERROR(EBUSY));
2314 
2315 	/* check amount of any unconsumed refreservation */
2316 	unused_refres_delta =
2317 	    (int64_t)MIN(origin_head->ds_reserved,
2318 	    origin_head->ds_phys->ds_unique_bytes) -
2319 	    (int64_t)MIN(origin_head->ds_reserved,
2320 	    clone->ds_phys->ds_unique_bytes);
2321 
2322 	if (unused_refres_delta > 0 &&
2323 	    unused_refres_delta >
2324 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2325 		return (SET_ERROR(ENOSPC));
2326 
2327 	/* clone can't be over the head's refquota */
2328 	if (origin_head->ds_quota != 0 &&
2329 	    clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2330 		return (SET_ERROR(EDQUOT));
2331 
2332 	return (0);
2333 }
2334 
2335 void
2336 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2337     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2338 {
2339 	dsl_pool_t *dp = dmu_tx_pool(tx);
2340 	int64_t unused_refres_delta;
2341 
2342 	ASSERT(clone->ds_reserved == 0);
2343 	ASSERT(origin_head->ds_quota == 0 ||
2344 	    clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2345 
2346 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2347 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2348 
2349 	if (clone->ds_objset != NULL) {
2350 		dmu_objset_evict(clone->ds_objset);
2351 		clone->ds_objset = NULL;
2352 	}
2353 
2354 	if (origin_head->ds_objset != NULL) {
2355 		dmu_objset_evict(origin_head->ds_objset);
2356 		origin_head->ds_objset = NULL;
2357 	}
2358 
2359 	unused_refres_delta =
2360 	    (int64_t)MIN(origin_head->ds_reserved,
2361 	    origin_head->ds_phys->ds_unique_bytes) -
2362 	    (int64_t)MIN(origin_head->ds_reserved,
2363 	    clone->ds_phys->ds_unique_bytes);
2364 
2365 	/*
2366 	 * Reset origin's unique bytes, if it exists.
2367 	 */
2368 	if (clone->ds_prev) {
2369 		dsl_dataset_t *origin = clone->ds_prev;
2370 		uint64_t comp, uncomp;
2371 
2372 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2373 		dsl_deadlist_space_range(&clone->ds_deadlist,
2374 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2375 		    &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2376 	}
2377 
2378 	/* swap blkptrs */
2379 	{
2380 		blkptr_t tmp;
2381 		tmp = origin_head->ds_phys->ds_bp;
2382 		origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2383 		clone->ds_phys->ds_bp = tmp;
2384 	}
2385 
2386 	/* set dd_*_bytes */
2387 	{
2388 		int64_t dused, dcomp, duncomp;
2389 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2390 		uint64_t odl_used, odl_comp, odl_uncomp;
2391 
2392 		ASSERT3U(clone->ds_dir->dd_phys->
2393 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2394 
2395 		dsl_deadlist_space(&clone->ds_deadlist,
2396 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2397 		dsl_deadlist_space(&origin_head->ds_deadlist,
2398 		    &odl_used, &odl_comp, &odl_uncomp);
2399 
2400 		dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2401 		    (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2402 		dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2403 		    (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2404 		duncomp = clone->ds_phys->ds_uncompressed_bytes +
2405 		    cdl_uncomp -
2406 		    (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2407 
2408 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2409 		    dused, dcomp, duncomp, tx);
2410 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2411 		    -dused, -dcomp, -duncomp, tx);
2412 
2413 		/*
2414 		 * The difference in the space used by snapshots is the
2415 		 * difference in snapshot space due to the head's
2416 		 * deadlist (since that's the only thing that's
2417 		 * changing that affects the snapused).
2418 		 */
2419 		dsl_deadlist_space_range(&clone->ds_deadlist,
2420 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2421 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2422 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2423 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2424 		    &odl_used, &odl_comp, &odl_uncomp);
2425 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2426 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2427 	}
2428 
2429 	/* swap ds_*_bytes */
2430 	SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2431 	    clone->ds_phys->ds_referenced_bytes);
2432 	SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2433 	    clone->ds_phys->ds_compressed_bytes);
2434 	SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2435 	    clone->ds_phys->ds_uncompressed_bytes);
2436 	SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2437 	    clone->ds_phys->ds_unique_bytes);
2438 
2439 	/* apply any parent delta for change in unconsumed refreservation */
2440 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2441 	    unused_refres_delta, 0, 0, tx);
2442 
2443 	/*
2444 	 * Swap deadlists.
2445 	 */
2446 	dsl_deadlist_close(&clone->ds_deadlist);
2447 	dsl_deadlist_close(&origin_head->ds_deadlist);
2448 	SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2449 	    clone->ds_phys->ds_deadlist_obj);
2450 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2451 	    clone->ds_phys->ds_deadlist_obj);
2452 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2453 	    origin_head->ds_phys->ds_deadlist_obj);
2454 
2455 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2456 
2457 	spa_history_log_internal_ds(clone, "clone swap", tx,
2458 	    "parent=%s", origin_head->ds_dir->dd_myname);
2459 }
2460 
2461 /*
2462  * Given a pool name and a dataset object number in that pool,
2463  * return the name of that dataset.
2464  */
2465 int
2466 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2467 {
2468 	dsl_pool_t *dp;
2469 	dsl_dataset_t *ds;
2470 	int error;
2471 
2472 	error = dsl_pool_hold(pname, FTAG, &dp);
2473 	if (error != 0)
2474 		return (error);
2475 
2476 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2477 	if (error == 0) {
2478 		dsl_dataset_name(ds, buf);
2479 		dsl_dataset_rele(ds, FTAG);
2480 	}
2481 	dsl_pool_rele(dp, FTAG);
2482 
2483 	return (error);
2484 }
2485 
2486 int
2487 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2488     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2489 {
2490 	int error = 0;
2491 
2492 	ASSERT3S(asize, >, 0);
2493 
2494 	/*
2495 	 * *ref_rsrv is the portion of asize that will come from any
2496 	 * unconsumed refreservation space.
2497 	 */
2498 	*ref_rsrv = 0;
2499 
2500 	mutex_enter(&ds->ds_lock);
2501 	/*
2502 	 * Make a space adjustment for reserved bytes.
2503 	 */
2504 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2505 		ASSERT3U(*used, >=,
2506 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2507 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2508 		*ref_rsrv =
2509 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2510 	}
2511 
2512 	if (!check_quota || ds->ds_quota == 0) {
2513 		mutex_exit(&ds->ds_lock);
2514 		return (0);
2515 	}
2516 	/*
2517 	 * If they are requesting more space, and our current estimate
2518 	 * is over quota, they get to try again unless the actual
2519 	 * on-disk is over quota and there are no pending changes (which
2520 	 * may free up space for us).
2521 	 */
2522 	if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2523 		if (inflight > 0 ||
2524 		    ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2525 			error = SET_ERROR(ERESTART);
2526 		else
2527 			error = SET_ERROR(EDQUOT);
2528 	}
2529 	mutex_exit(&ds->ds_lock);
2530 
2531 	return (error);
2532 }
2533 
2534 typedef struct dsl_dataset_set_qr_arg {
2535 	const char *ddsqra_name;
2536 	zprop_source_t ddsqra_source;
2537 	uint64_t ddsqra_value;
2538 } dsl_dataset_set_qr_arg_t;
2539 
2540 
2541 /* ARGSUSED */
2542 static int
2543 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2544 {
2545 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2546 	dsl_pool_t *dp = dmu_tx_pool(tx);
2547 	dsl_dataset_t *ds;
2548 	int error;
2549 	uint64_t newval;
2550 
2551 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2552 		return (SET_ERROR(ENOTSUP));
2553 
2554 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2555 	if (error != 0)
2556 		return (error);
2557 
2558 	if (dsl_dataset_is_snapshot(ds)) {
2559 		dsl_dataset_rele(ds, FTAG);
2560 		return (SET_ERROR(EINVAL));
2561 	}
2562 
2563 	error = dsl_prop_predict(ds->ds_dir,
2564 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2565 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2566 	if (error != 0) {
2567 		dsl_dataset_rele(ds, FTAG);
2568 		return (error);
2569 	}
2570 
2571 	if (newval == 0) {
2572 		dsl_dataset_rele(ds, FTAG);
2573 		return (0);
2574 	}
2575 
2576 	if (newval < ds->ds_phys->ds_referenced_bytes ||
2577 	    newval < ds->ds_reserved) {
2578 		dsl_dataset_rele(ds, FTAG);
2579 		return (SET_ERROR(ENOSPC));
2580 	}
2581 
2582 	dsl_dataset_rele(ds, FTAG);
2583 	return (0);
2584 }
2585 
2586 static void
2587 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2588 {
2589 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2590 	dsl_pool_t *dp = dmu_tx_pool(tx);
2591 	dsl_dataset_t *ds;
2592 	uint64_t newval;
2593 
2594 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2595 
2596 	dsl_prop_set_sync_impl(ds,
2597 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2598 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2599 	    &ddsqra->ddsqra_value, tx);
2600 
2601 	VERIFY0(dsl_prop_get_int_ds(ds,
2602 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2603 
2604 	if (ds->ds_quota != newval) {
2605 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2606 		ds->ds_quota = newval;
2607 	}
2608 	dsl_dataset_rele(ds, FTAG);
2609 }
2610 
2611 int
2612 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2613     uint64_t refquota)
2614 {
2615 	dsl_dataset_set_qr_arg_t ddsqra;
2616 
2617 	ddsqra.ddsqra_name = dsname;
2618 	ddsqra.ddsqra_source = source;
2619 	ddsqra.ddsqra_value = refquota;
2620 
2621 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2622 	    dsl_dataset_set_refquota_sync, &ddsqra, 0));
2623 }
2624 
2625 static int
2626 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2627 {
2628 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2629 	dsl_pool_t *dp = dmu_tx_pool(tx);
2630 	dsl_dataset_t *ds;
2631 	int error;
2632 	uint64_t newval, unique;
2633 
2634 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2635 		return (SET_ERROR(ENOTSUP));
2636 
2637 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2638 	if (error != 0)
2639 		return (error);
2640 
2641 	if (dsl_dataset_is_snapshot(ds)) {
2642 		dsl_dataset_rele(ds, FTAG);
2643 		return (SET_ERROR(EINVAL));
2644 	}
2645 
2646 	error = dsl_prop_predict(ds->ds_dir,
2647 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2648 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2649 	if (error != 0) {
2650 		dsl_dataset_rele(ds, FTAG);
2651 		return (error);
2652 	}
2653 
2654 	/*
2655 	 * If we are doing the preliminary check in open context, the
2656 	 * space estimates may be inaccurate.
2657 	 */
2658 	if (!dmu_tx_is_syncing(tx)) {
2659 		dsl_dataset_rele(ds, FTAG);
2660 		return (0);
2661 	}
2662 
2663 	mutex_enter(&ds->ds_lock);
2664 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2665 		dsl_dataset_recalc_head_uniq(ds);
2666 	unique = ds->ds_phys->ds_unique_bytes;
2667 	mutex_exit(&ds->ds_lock);
2668 
2669 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2670 		uint64_t delta = MAX(unique, newval) -
2671 		    MAX(unique, ds->ds_reserved);
2672 
2673 		if (delta >
2674 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2675 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2676 			dsl_dataset_rele(ds, FTAG);
2677 			return (SET_ERROR(ENOSPC));
2678 		}
2679 	}
2680 
2681 	dsl_dataset_rele(ds, FTAG);
2682 	return (0);
2683 }
2684 
2685 void
2686 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2687     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2688 {
2689 	uint64_t newval;
2690 	uint64_t unique;
2691 	int64_t delta;
2692 
2693 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2694 	    source, sizeof (value), 1, &value, tx);
2695 
2696 	VERIFY0(dsl_prop_get_int_ds(ds,
2697 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2698 
2699 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2700 	mutex_enter(&ds->ds_dir->dd_lock);
2701 	mutex_enter(&ds->ds_lock);
2702 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2703 	unique = ds->ds_phys->ds_unique_bytes;
2704 	delta = MAX(0, (int64_t)(newval - unique)) -
2705 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2706 	ds->ds_reserved = newval;
2707 	mutex_exit(&ds->ds_lock);
2708 
2709 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2710 	mutex_exit(&ds->ds_dir->dd_lock);
2711 }
2712 
2713 static void
2714 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2715 {
2716 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2717 	dsl_pool_t *dp = dmu_tx_pool(tx);
2718 	dsl_dataset_t *ds;
2719 
2720 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2721 	dsl_dataset_set_refreservation_sync_impl(ds,
2722 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2723 	dsl_dataset_rele(ds, FTAG);
2724 }
2725 
2726 int
2727 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2728     uint64_t refreservation)
2729 {
2730 	dsl_dataset_set_qr_arg_t ddsqra;
2731 
2732 	ddsqra.ddsqra_name = dsname;
2733 	ddsqra.ddsqra_source = source;
2734 	ddsqra.ddsqra_value = refreservation;
2735 
2736 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2737 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0));
2738 }
2739 
2740 /*
2741  * Return (in *usedp) the amount of space written in new that is not
2742  * present in oldsnap.  New may be a snapshot or the head.  Old must be
2743  * a snapshot before new, in new's filesystem (or its origin).  If not then
2744  * fail and return EINVAL.
2745  *
2746  * The written space is calculated by considering two components:  First, we
2747  * ignore any freed space, and calculate the written as new's used space
2748  * minus old's used space.  Next, we add in the amount of space that was freed
2749  * between the two snapshots, thus reducing new's used space relative to old's.
2750  * Specifically, this is the space that was born before old->ds_creation_txg,
2751  * and freed before new (ie. on new's deadlist or a previous deadlist).
2752  *
2753  * space freed                         [---------------------]
2754  * snapshots                       ---O-------O--------O-------O------
2755  *                                         oldsnap            new
2756  */
2757 int
2758 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2759     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2760 {
2761 	int err = 0;
2762 	uint64_t snapobj;
2763 	dsl_pool_t *dp = new->ds_dir->dd_pool;
2764 
2765 	ASSERT(dsl_pool_config_held(dp));
2766 
2767 	*usedp = 0;
2768 	*usedp += new->ds_phys->ds_referenced_bytes;
2769 	*usedp -= oldsnap->ds_phys->ds_referenced_bytes;
2770 
2771 	*compp = 0;
2772 	*compp += new->ds_phys->ds_compressed_bytes;
2773 	*compp -= oldsnap->ds_phys->ds_compressed_bytes;
2774 
2775 	*uncompp = 0;
2776 	*uncompp += new->ds_phys->ds_uncompressed_bytes;
2777 	*uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
2778 
2779 	snapobj = new->ds_object;
2780 	while (snapobj != oldsnap->ds_object) {
2781 		dsl_dataset_t *snap;
2782 		uint64_t used, comp, uncomp;
2783 
2784 		if (snapobj == new->ds_object) {
2785 			snap = new;
2786 		} else {
2787 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
2788 			if (err != 0)
2789 				break;
2790 		}
2791 
2792 		if (snap->ds_phys->ds_prev_snap_txg ==
2793 		    oldsnap->ds_phys->ds_creation_txg) {
2794 			/*
2795 			 * The blocks in the deadlist can not be born after
2796 			 * ds_prev_snap_txg, so get the whole deadlist space,
2797 			 * which is more efficient (especially for old-format
2798 			 * deadlists).  Unfortunately the deadlist code
2799 			 * doesn't have enough information to make this
2800 			 * optimization itself.
2801 			 */
2802 			dsl_deadlist_space(&snap->ds_deadlist,
2803 			    &used, &comp, &uncomp);
2804 		} else {
2805 			dsl_deadlist_space_range(&snap->ds_deadlist,
2806 			    0, oldsnap->ds_phys->ds_creation_txg,
2807 			    &used, &comp, &uncomp);
2808 		}
2809 		*usedp += used;
2810 		*compp += comp;
2811 		*uncompp += uncomp;
2812 
2813 		/*
2814 		 * If we get to the beginning of the chain of snapshots
2815 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2816 		 * was not a snapshot of/before new.
2817 		 */
2818 		snapobj = snap->ds_phys->ds_prev_snap_obj;
2819 		if (snap != new)
2820 			dsl_dataset_rele(snap, FTAG);
2821 		if (snapobj == 0) {
2822 			err = SET_ERROR(EINVAL);
2823 			break;
2824 		}
2825 
2826 	}
2827 	return (err);
2828 }
2829 
2830 /*
2831  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2832  * lastsnap, and all snapshots in between are deleted.
2833  *
2834  * blocks that would be freed            [---------------------------]
2835  * snapshots                       ---O-------O--------O-------O--------O
2836  *                                        firstsnap        lastsnap
2837  *
2838  * This is the set of blocks that were born after the snap before firstsnap,
2839  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2840  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2841  * We calculate this by iterating over the relevant deadlists (from the snap
2842  * after lastsnap, backward to the snap after firstsnap), summing up the
2843  * space on the deadlist that was born after the snap before firstsnap.
2844  */
2845 int
2846 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
2847     dsl_dataset_t *lastsnap,
2848     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2849 {
2850 	int err = 0;
2851 	uint64_t snapobj;
2852 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
2853 
2854 	ASSERT(dsl_dataset_is_snapshot(firstsnap));
2855 	ASSERT(dsl_dataset_is_snapshot(lastsnap));
2856 
2857 	/*
2858 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
2859 	 * is before lastsnap.
2860 	 */
2861 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
2862 	    firstsnap->ds_phys->ds_creation_txg >
2863 	    lastsnap->ds_phys->ds_creation_txg)
2864 		return (SET_ERROR(EINVAL));
2865 
2866 	*usedp = *compp = *uncompp = 0;
2867 
2868 	snapobj = lastsnap->ds_phys->ds_next_snap_obj;
2869 	while (snapobj != firstsnap->ds_object) {
2870 		dsl_dataset_t *ds;
2871 		uint64_t used, comp, uncomp;
2872 
2873 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
2874 		if (err != 0)
2875 			break;
2876 
2877 		dsl_deadlist_space_range(&ds->ds_deadlist,
2878 		    firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2879 		    &used, &comp, &uncomp);
2880 		*usedp += used;
2881 		*compp += comp;
2882 		*uncompp += uncomp;
2883 
2884 		snapobj = ds->ds_phys->ds_prev_snap_obj;
2885 		ASSERT3U(snapobj, !=, 0);
2886 		dsl_dataset_rele(ds, FTAG);
2887 	}
2888 	return (err);
2889 }
2890 
2891 /*
2892  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2893  * For example, they could both be snapshots of the same filesystem, and
2894  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
2895  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
2896  * filesystem.  Or 'earlier' could be the origin's origin.
2897  */
2898 boolean_t
2899 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
2900 {
2901 	dsl_pool_t *dp = later->ds_dir->dd_pool;
2902 	int error;
2903 	boolean_t ret;
2904 
2905 	ASSERT(dsl_pool_config_held(dp));
2906 
2907 	if (earlier->ds_phys->ds_creation_txg >=
2908 	    later->ds_phys->ds_creation_txg)
2909 		return (B_FALSE);
2910 
2911 	if (later->ds_dir == earlier->ds_dir)
2912 		return (B_TRUE);
2913 	if (!dsl_dir_is_clone(later->ds_dir))
2914 		return (B_FALSE);
2915 
2916 	if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
2917 		return (B_TRUE);
2918 	dsl_dataset_t *origin;
2919 	error = dsl_dataset_hold_obj(dp,
2920 	    later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
2921 	if (error != 0)
2922 		return (B_FALSE);
2923 	ret = dsl_dataset_is_before(origin, earlier);
2924 	dsl_dataset_rele(origin, FTAG);
2925 	return (ret);
2926 }
2927