xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 0a47c91c895e274dd0990009919e30e984364a8b)
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 		dmu_buf_rele(dbuf, tag);
361 		return (SET_ERROR(EINVAL));
362 	}
363 
364 	ds = dmu_buf_get_user(dbuf);
365 	if (ds == NULL) {
366 		dsl_dataset_t *winner = NULL;
367 
368 		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
369 		ds->ds_dbuf = dbuf;
370 		ds->ds_object = dsobj;
371 		ds->ds_phys = dbuf->db_data;
372 
373 		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
374 		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
375 		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
376 		refcount_create(&ds->ds_longholds);
377 
378 		bplist_create(&ds->ds_pending_deadlist);
379 		dsl_deadlist_open(&ds->ds_deadlist,
380 		    mos, ds->ds_phys->ds_deadlist_obj);
381 
382 		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
383 		    offsetof(dmu_sendarg_t, dsa_link));
384 
385 		if (err == 0) {
386 			err = dsl_dir_hold_obj(dp,
387 			    ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
388 		}
389 		if (err != 0) {
390 			mutex_destroy(&ds->ds_lock);
391 			mutex_destroy(&ds->ds_opening_lock);
392 			refcount_destroy(&ds->ds_longholds);
393 			bplist_destroy(&ds->ds_pending_deadlist);
394 			dsl_deadlist_close(&ds->ds_deadlist);
395 			kmem_free(ds, sizeof (dsl_dataset_t));
396 			dmu_buf_rele(dbuf, tag);
397 			return (err);
398 		}
399 
400 		if (!dsl_dataset_is_snapshot(ds)) {
401 			ds->ds_snapname[0] = '\0';
402 			if (ds->ds_phys->ds_prev_snap_obj != 0) {
403 				err = dsl_dataset_hold_obj(dp,
404 				    ds->ds_phys->ds_prev_snap_obj,
405 				    ds, &ds->ds_prev);
406 			}
407 		} else {
408 			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
409 				err = dsl_dataset_get_snapname(ds);
410 			if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
411 				err = zap_count(
412 				    ds->ds_dir->dd_pool->dp_meta_objset,
413 				    ds->ds_phys->ds_userrefs_obj,
414 				    &ds->ds_userrefs);
415 			}
416 		}
417 
418 		if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
419 			err = dsl_prop_get_int_ds(ds,
420 			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
421 			    &ds->ds_reserved);
422 			if (err == 0) {
423 				err = dsl_prop_get_int_ds(ds,
424 				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
425 				    &ds->ds_quota);
426 			}
427 		} else {
428 			ds->ds_reserved = ds->ds_quota = 0;
429 		}
430 
431 		if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
432 		    &ds->ds_phys, dsl_dataset_evict)) != NULL) {
433 			bplist_destroy(&ds->ds_pending_deadlist);
434 			dsl_deadlist_close(&ds->ds_deadlist);
435 			if (ds->ds_prev)
436 				dsl_dataset_rele(ds->ds_prev, ds);
437 			dsl_dir_rele(ds->ds_dir, ds);
438 			mutex_destroy(&ds->ds_lock);
439 			mutex_destroy(&ds->ds_opening_lock);
440 			refcount_destroy(&ds->ds_longholds);
441 			kmem_free(ds, sizeof (dsl_dataset_t));
442 			if (err != 0) {
443 				dmu_buf_rele(dbuf, tag);
444 				return (err);
445 			}
446 			ds = winner;
447 		} else {
448 			ds->ds_fsid_guid =
449 			    unique_insert(ds->ds_phys->ds_fsid_guid);
450 		}
451 	}
452 	ASSERT3P(ds->ds_dbuf, ==, dbuf);
453 	ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
454 	ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
455 	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
456 	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
457 	*dsp = ds;
458 	return (0);
459 }
460 
461 int
462 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
463     void *tag, dsl_dataset_t **dsp)
464 {
465 	dsl_dir_t *dd;
466 	const char *snapname;
467 	uint64_t obj;
468 	int err = 0;
469 
470 	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
471 	if (err != 0)
472 		return (err);
473 
474 	ASSERT(dsl_pool_config_held(dp));
475 	obj = dd->dd_phys->dd_head_dataset_obj;
476 	if (obj != 0)
477 		err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
478 	else
479 		err = SET_ERROR(ENOENT);
480 
481 	/* we may be looking for a snapshot */
482 	if (err == 0 && snapname != NULL) {
483 		dsl_dataset_t *ds;
484 
485 		if (*snapname++ != '@') {
486 			dsl_dataset_rele(*dsp, tag);
487 			dsl_dir_rele(dd, FTAG);
488 			return (SET_ERROR(ENOENT));
489 		}
490 
491 		dprintf("looking for snapshot '%s'\n", snapname);
492 		err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
493 		if (err == 0)
494 			err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
495 		dsl_dataset_rele(*dsp, tag);
496 
497 		if (err == 0) {
498 			mutex_enter(&ds->ds_lock);
499 			if (ds->ds_snapname[0] == 0)
500 				(void) strlcpy(ds->ds_snapname, snapname,
501 				    sizeof (ds->ds_snapname));
502 			mutex_exit(&ds->ds_lock);
503 			*dsp = ds;
504 		}
505 	}
506 
507 	dsl_dir_rele(dd, FTAG);
508 	return (err);
509 }
510 
511 int
512 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
513     void *tag, dsl_dataset_t **dsp)
514 {
515 	int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
516 	if (err != 0)
517 		return (err);
518 	if (!dsl_dataset_tryown(*dsp, tag)) {
519 		dsl_dataset_rele(*dsp, tag);
520 		*dsp = NULL;
521 		return (SET_ERROR(EBUSY));
522 	}
523 	return (0);
524 }
525 
526 int
527 dsl_dataset_own(dsl_pool_t *dp, const char *name,
528     void *tag, dsl_dataset_t **dsp)
529 {
530 	int err = dsl_dataset_hold(dp, name, tag, dsp);
531 	if (err != 0)
532 		return (err);
533 	if (!dsl_dataset_tryown(*dsp, tag)) {
534 		dsl_dataset_rele(*dsp, tag);
535 		return (SET_ERROR(EBUSY));
536 	}
537 	return (0);
538 }
539 
540 /*
541  * See the comment above dsl_pool_hold() for details.  In summary, a long
542  * hold is used to prevent destruction of a dataset while the pool hold
543  * is dropped, allowing other concurrent operations (e.g. spa_sync()).
544  *
545  * The dataset and pool must be held when this function is called.  After it
546  * is called, the pool hold may be released while the dataset is still held
547  * and accessed.
548  */
549 void
550 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
551 {
552 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
553 	(void) refcount_add(&ds->ds_longholds, tag);
554 }
555 
556 void
557 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
558 {
559 	(void) refcount_remove(&ds->ds_longholds, tag);
560 }
561 
562 /* Return B_TRUE if there are any long holds on this dataset. */
563 boolean_t
564 dsl_dataset_long_held(dsl_dataset_t *ds)
565 {
566 	return (!refcount_is_zero(&ds->ds_longholds));
567 }
568 
569 void
570 dsl_dataset_name(dsl_dataset_t *ds, char *name)
571 {
572 	if (ds == NULL) {
573 		(void) strcpy(name, "mos");
574 	} else {
575 		dsl_dir_name(ds->ds_dir, name);
576 		VERIFY0(dsl_dataset_get_snapname(ds));
577 		if (ds->ds_snapname[0]) {
578 			(void) strcat(name, "@");
579 			/*
580 			 * We use a "recursive" mutex so that we
581 			 * can call dprintf_ds() with ds_lock held.
582 			 */
583 			if (!MUTEX_HELD(&ds->ds_lock)) {
584 				mutex_enter(&ds->ds_lock);
585 				(void) strcat(name, ds->ds_snapname);
586 				mutex_exit(&ds->ds_lock);
587 			} else {
588 				(void) strcat(name, ds->ds_snapname);
589 			}
590 		}
591 	}
592 }
593 
594 static int
595 dsl_dataset_namelen(dsl_dataset_t *ds)
596 {
597 	int result;
598 
599 	if (ds == NULL) {
600 		result = 3;	/* "mos" */
601 	} else {
602 		result = dsl_dir_namelen(ds->ds_dir);
603 		VERIFY0(dsl_dataset_get_snapname(ds));
604 		if (ds->ds_snapname[0]) {
605 			++result;	/* adding one for the @-sign */
606 			if (!MUTEX_HELD(&ds->ds_lock)) {
607 				mutex_enter(&ds->ds_lock);
608 				result += strlen(ds->ds_snapname);
609 				mutex_exit(&ds->ds_lock);
610 			} else {
611 				result += strlen(ds->ds_snapname);
612 			}
613 		}
614 	}
615 
616 	return (result);
617 }
618 
619 void
620 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
621 {
622 	dmu_buf_rele(ds->ds_dbuf, tag);
623 }
624 
625 void
626 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
627 {
628 	ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
629 
630 	mutex_enter(&ds->ds_lock);
631 	ds->ds_owner = NULL;
632 	mutex_exit(&ds->ds_lock);
633 	dsl_dataset_long_rele(ds, tag);
634 	if (ds->ds_dbuf != NULL)
635 		dsl_dataset_rele(ds, tag);
636 	else
637 		dsl_dataset_evict(NULL, ds);
638 }
639 
640 boolean_t
641 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
642 {
643 	boolean_t gotit = FALSE;
644 
645 	mutex_enter(&ds->ds_lock);
646 	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
647 		ds->ds_owner = tag;
648 		dsl_dataset_long_hold(ds, tag);
649 		gotit = TRUE;
650 	}
651 	mutex_exit(&ds->ds_lock);
652 	return (gotit);
653 }
654 
655 uint64_t
656 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
657     uint64_t flags, dmu_tx_t *tx)
658 {
659 	dsl_pool_t *dp = dd->dd_pool;
660 	dmu_buf_t *dbuf;
661 	dsl_dataset_phys_t *dsphys;
662 	uint64_t dsobj;
663 	objset_t *mos = dp->dp_meta_objset;
664 
665 	if (origin == NULL)
666 		origin = dp->dp_origin_snap;
667 
668 	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
669 	ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
670 	ASSERT(dmu_tx_is_syncing(tx));
671 	ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
672 
673 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
674 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
675 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
676 	dmu_buf_will_dirty(dbuf, tx);
677 	dsphys = dbuf->db_data;
678 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
679 	dsphys->ds_dir_obj = dd->dd_object;
680 	dsphys->ds_flags = flags;
681 	dsphys->ds_fsid_guid = unique_create();
682 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
683 	    sizeof (dsphys->ds_guid));
684 	dsphys->ds_snapnames_zapobj =
685 	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
686 	    DMU_OT_NONE, 0, tx);
687 	dsphys->ds_creation_time = gethrestime_sec();
688 	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
689 
690 	if (origin == NULL) {
691 		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
692 	} else {
693 		dsl_dataset_t *ohds; /* head of the origin snapshot */
694 
695 		dsphys->ds_prev_snap_obj = origin->ds_object;
696 		dsphys->ds_prev_snap_txg =
697 		    origin->ds_phys->ds_creation_txg;
698 		dsphys->ds_referenced_bytes =
699 		    origin->ds_phys->ds_referenced_bytes;
700 		dsphys->ds_compressed_bytes =
701 		    origin->ds_phys->ds_compressed_bytes;
702 		dsphys->ds_uncompressed_bytes =
703 		    origin->ds_phys->ds_uncompressed_bytes;
704 		dsphys->ds_bp = origin->ds_phys->ds_bp;
705 		dsphys->ds_flags |= origin->ds_phys->ds_flags;
706 
707 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
708 		origin->ds_phys->ds_num_children++;
709 
710 		VERIFY0(dsl_dataset_hold_obj(dp,
711 		    origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
712 		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
713 		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
714 		dsl_dataset_rele(ohds, FTAG);
715 
716 		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
717 			if (origin->ds_phys->ds_next_clones_obj == 0) {
718 				origin->ds_phys->ds_next_clones_obj =
719 				    zap_create(mos,
720 				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
721 			}
722 			VERIFY0(zap_add_int(mos,
723 			    origin->ds_phys->ds_next_clones_obj, dsobj, tx));
724 		}
725 
726 		dmu_buf_will_dirty(dd->dd_dbuf, tx);
727 		dd->dd_phys->dd_origin_obj = origin->ds_object;
728 		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
729 			if (origin->ds_dir->dd_phys->dd_clones == 0) {
730 				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
731 				origin->ds_dir->dd_phys->dd_clones =
732 				    zap_create(mos,
733 				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
734 			}
735 			VERIFY0(zap_add_int(mos,
736 			    origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
737 		}
738 	}
739 
740 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
741 		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
742 
743 	dmu_buf_rele(dbuf, FTAG);
744 
745 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
746 	dd->dd_phys->dd_head_dataset_obj = dsobj;
747 
748 	return (dsobj);
749 }
750 
751 static void
752 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
753 {
754 	objset_t *os;
755 
756 	VERIFY0(dmu_objset_from_ds(ds, &os));
757 	bzero(&os->os_zil_header, sizeof (os->os_zil_header));
758 	dsl_dataset_dirty(ds, tx);
759 }
760 
761 uint64_t
762 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
763     dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
764 {
765 	dsl_pool_t *dp = pdd->dd_pool;
766 	uint64_t dsobj, ddobj;
767 	dsl_dir_t *dd;
768 
769 	ASSERT(dmu_tx_is_syncing(tx));
770 	ASSERT(lastname[0] != '@');
771 
772 	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
773 	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
774 
775 	dsobj = dsl_dataset_create_sync_dd(dd, origin,
776 	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
777 
778 	dsl_deleg_set_create_perms(dd, tx, cr);
779 
780 	dsl_dir_rele(dd, FTAG);
781 
782 	/*
783 	 * If we are creating a clone, make sure we zero out any stale
784 	 * data from the origin snapshots zil header.
785 	 */
786 	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
787 		dsl_dataset_t *ds;
788 
789 		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
790 		dsl_dataset_zero_zil(ds, tx);
791 		dsl_dataset_rele(ds, FTAG);
792 	}
793 
794 	return (dsobj);
795 }
796 
797 /*
798  * The unique space in the head dataset can be calculated by subtracting
799  * the space used in the most recent snapshot, that is still being used
800  * in this file system, from the space currently in use.  To figure out
801  * the space in the most recent snapshot still in use, we need to take
802  * the total space used in the snapshot and subtract out the space that
803  * has been freed up since the snapshot was taken.
804  */
805 void
806 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
807 {
808 	uint64_t mrs_used;
809 	uint64_t dlused, dlcomp, dluncomp;
810 
811 	ASSERT(!dsl_dataset_is_snapshot(ds));
812 
813 	if (ds->ds_phys->ds_prev_snap_obj != 0)
814 		mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
815 	else
816 		mrs_used = 0;
817 
818 	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
819 
820 	ASSERT3U(dlused, <=, mrs_used);
821 	ds->ds_phys->ds_unique_bytes =
822 	    ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
823 
824 	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
825 	    SPA_VERSION_UNIQUE_ACCURATE)
826 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
827 }
828 
829 void
830 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
831     dmu_tx_t *tx)
832 {
833 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
834 	uint64_t count;
835 	int err;
836 
837 	ASSERT(ds->ds_phys->ds_num_children >= 2);
838 	err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
839 	/*
840 	 * The err should not be ENOENT, but a bug in a previous version
841 	 * of the code could cause upgrade_clones_cb() to not set
842 	 * ds_next_snap_obj when it should, leading to a missing entry.
843 	 * If we knew that the pool was created after
844 	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
845 	 * ENOENT.  However, at least we can check that we don't have
846 	 * too many entries in the next_clones_obj even after failing to
847 	 * remove this one.
848 	 */
849 	if (err != ENOENT)
850 		VERIFY0(err);
851 	ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
852 	    &count));
853 	ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
854 }
855 
856 
857 blkptr_t *
858 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
859 {
860 	return (&ds->ds_phys->ds_bp);
861 }
862 
863 void
864 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
865 {
866 	ASSERT(dmu_tx_is_syncing(tx));
867 	/* If it's the meta-objset, set dp_meta_rootbp */
868 	if (ds == NULL) {
869 		tx->tx_pool->dp_meta_rootbp = *bp;
870 	} else {
871 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
872 		ds->ds_phys->ds_bp = *bp;
873 	}
874 }
875 
876 spa_t *
877 dsl_dataset_get_spa(dsl_dataset_t *ds)
878 {
879 	return (ds->ds_dir->dd_pool->dp_spa);
880 }
881 
882 void
883 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
884 {
885 	dsl_pool_t *dp;
886 
887 	if (ds == NULL) /* this is the meta-objset */
888 		return;
889 
890 	ASSERT(ds->ds_objset != NULL);
891 
892 	if (ds->ds_phys->ds_next_snap_obj != 0)
893 		panic("dirtying snapshot!");
894 
895 	dp = ds->ds_dir->dd_pool;
896 
897 	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
898 		/* up the hold count until we can be written out */
899 		dmu_buf_add_ref(ds->ds_dbuf, ds);
900 	}
901 }
902 
903 boolean_t
904 dsl_dataset_is_dirty(dsl_dataset_t *ds)
905 {
906 	for (int t = 0; t < TXG_SIZE; t++) {
907 		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
908 		    ds, t))
909 			return (B_TRUE);
910 	}
911 	return (B_FALSE);
912 }
913 
914 static int
915 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
916 {
917 	uint64_t asize;
918 
919 	if (!dmu_tx_is_syncing(tx))
920 		return (0);
921 
922 	/*
923 	 * If there's an fs-only reservation, any blocks that might become
924 	 * owned by the snapshot dataset must be accommodated by space
925 	 * outside of the reservation.
926 	 */
927 	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
928 	asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
929 	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
930 		return (SET_ERROR(ENOSPC));
931 
932 	/*
933 	 * Propagate any reserved space for this snapshot to other
934 	 * snapshot checks in this sync group.
935 	 */
936 	if (asize > 0)
937 		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
938 
939 	return (0);
940 }
941 
942 typedef struct dsl_dataset_snapshot_arg {
943 	nvlist_t *ddsa_snaps;
944 	nvlist_t *ddsa_props;
945 	nvlist_t *ddsa_errors;
946 } dsl_dataset_snapshot_arg_t;
947 
948 int
949 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
950     dmu_tx_t *tx, boolean_t recv)
951 {
952 	int error;
953 	uint64_t value;
954 
955 	ds->ds_trysnap_txg = tx->tx_txg;
956 
957 	if (!dmu_tx_is_syncing(tx))
958 		return (0);
959 
960 	/*
961 	 * We don't allow multiple snapshots of the same txg.  If there
962 	 * is already one, try again.
963 	 */
964 	if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
965 		return (SET_ERROR(EAGAIN));
966 
967 	/*
968 	 * Check for conflicting snapshot name.
969 	 */
970 	error = dsl_dataset_snap_lookup(ds, snapname, &value);
971 	if (error == 0)
972 		return (SET_ERROR(EEXIST));
973 	if (error != ENOENT)
974 		return (error);
975 
976 	/*
977 	 * We don't allow taking snapshots of inconsistent datasets, such as
978 	 * those into which we are currently receiving.  However, if we are
979 	 * creating this snapshot as part of a receive, this check will be
980 	 * executed atomically with respect to the completion of the receive
981 	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
982 	 * case we ignore this, knowing it will be fixed up for us shortly in
983 	 * dmu_recv_end_sync().
984 	 */
985 	if (!recv && DS_IS_INCONSISTENT(ds))
986 		return (SET_ERROR(EBUSY));
987 
988 	error = dsl_dataset_snapshot_reserve_space(ds, tx);
989 	if (error != 0)
990 		return (error);
991 
992 	return (0);
993 }
994 
995 static int
996 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
997 {
998 	dsl_dataset_snapshot_arg_t *ddsa = arg;
999 	dsl_pool_t *dp = dmu_tx_pool(tx);
1000 	nvpair_t *pair;
1001 	int rv = 0;
1002 
1003 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1004 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1005 		int error = 0;
1006 		dsl_dataset_t *ds;
1007 		char *name, *atp;
1008 		char dsname[MAXNAMELEN];
1009 
1010 		name = nvpair_name(pair);
1011 		if (strlen(name) >= MAXNAMELEN)
1012 			error = SET_ERROR(ENAMETOOLONG);
1013 		if (error == 0) {
1014 			atp = strchr(name, '@');
1015 			if (atp == NULL)
1016 				error = SET_ERROR(EINVAL);
1017 			if (error == 0)
1018 				(void) strlcpy(dsname, name, atp - name + 1);
1019 		}
1020 		if (error == 0)
1021 			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1022 		if (error == 0) {
1023 			error = dsl_dataset_snapshot_check_impl(ds,
1024 			    atp + 1, tx, B_FALSE);
1025 			dsl_dataset_rele(ds, FTAG);
1026 		}
1027 
1028 		if (error != 0) {
1029 			if (ddsa->ddsa_errors != NULL) {
1030 				fnvlist_add_int32(ddsa->ddsa_errors,
1031 				    name, error);
1032 			}
1033 			rv = error;
1034 		}
1035 	}
1036 	return (rv);
1037 }
1038 
1039 void
1040 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1041     dmu_tx_t *tx)
1042 {
1043 	static zil_header_t zero_zil;
1044 
1045 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1046 	dmu_buf_t *dbuf;
1047 	dsl_dataset_phys_t *dsphys;
1048 	uint64_t dsobj, crtxg;
1049 	objset_t *mos = dp->dp_meta_objset;
1050 	objset_t *os;
1051 
1052 	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1053 
1054 	/*
1055 	 * If we are on an old pool, the zil must not be active, in which
1056 	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1057 	 */
1058 	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1059 	    dmu_objset_from_ds(ds, &os) != 0 ||
1060 	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1061 	    sizeof (zero_zil)) == 0);
1062 
1063 
1064 	/*
1065 	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1066 	 */
1067 	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1068 		crtxg = 1;
1069 	else
1070 		crtxg = tx->tx_txg;
1071 
1072 	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1073 	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1074 	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1075 	dmu_buf_will_dirty(dbuf, tx);
1076 	dsphys = dbuf->db_data;
1077 	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1078 	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1079 	dsphys->ds_fsid_guid = unique_create();
1080 	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1081 	    sizeof (dsphys->ds_guid));
1082 	dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1083 	dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1084 	dsphys->ds_next_snap_obj = ds->ds_object;
1085 	dsphys->ds_num_children = 1;
1086 	dsphys->ds_creation_time = gethrestime_sec();
1087 	dsphys->ds_creation_txg = crtxg;
1088 	dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1089 	dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1090 	dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1091 	dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1092 	dsphys->ds_flags = ds->ds_phys->ds_flags;
1093 	dsphys->ds_bp = ds->ds_phys->ds_bp;
1094 	dmu_buf_rele(dbuf, FTAG);
1095 
1096 	ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1097 	if (ds->ds_prev) {
1098 		uint64_t next_clones_obj =
1099 		    ds->ds_prev->ds_phys->ds_next_clones_obj;
1100 		ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1101 		    ds->ds_object ||
1102 		    ds->ds_prev->ds_phys->ds_num_children > 1);
1103 		if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1104 			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1105 			ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1106 			    ds->ds_prev->ds_phys->ds_creation_txg);
1107 			ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1108 		} else if (next_clones_obj != 0) {
1109 			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1110 			    dsphys->ds_next_snap_obj, tx);
1111 			VERIFY0(zap_add_int(mos,
1112 			    next_clones_obj, dsobj, tx));
1113 		}
1114 	}
1115 
1116 	/*
1117 	 * If we have a reference-reservation on this dataset, we will
1118 	 * need to increase the amount of refreservation being charged
1119 	 * since our unique space is going to zero.
1120 	 */
1121 	if (ds->ds_reserved) {
1122 		int64_t delta;
1123 		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1124 		delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1125 		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1126 		    delta, 0, 0, tx);
1127 	}
1128 
1129 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1130 	ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1131 	    UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1132 	dsl_deadlist_close(&ds->ds_deadlist);
1133 	dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1134 	dsl_deadlist_add_key(&ds->ds_deadlist,
1135 	    ds->ds_phys->ds_prev_snap_txg, tx);
1136 
1137 	ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1138 	ds->ds_phys->ds_prev_snap_obj = dsobj;
1139 	ds->ds_phys->ds_prev_snap_txg = crtxg;
1140 	ds->ds_phys->ds_unique_bytes = 0;
1141 	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1142 		ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1143 
1144 	VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1145 	    snapname, 8, 1, &dsobj, tx));
1146 
1147 	if (ds->ds_prev)
1148 		dsl_dataset_rele(ds->ds_prev, ds);
1149 	VERIFY0(dsl_dataset_hold_obj(dp,
1150 	    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1151 
1152 	dsl_scan_ds_snapshotted(ds, tx);
1153 
1154 	dsl_dir_snap_cmtime_update(ds->ds_dir);
1155 
1156 	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1157 }
1158 
1159 static void
1160 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1161 {
1162 	dsl_dataset_snapshot_arg_t *ddsa = arg;
1163 	dsl_pool_t *dp = dmu_tx_pool(tx);
1164 	nvpair_t *pair;
1165 
1166 	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1167 	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1168 		dsl_dataset_t *ds;
1169 		char *name, *atp;
1170 		char dsname[MAXNAMELEN];
1171 
1172 		name = nvpair_name(pair);
1173 		atp = strchr(name, '@');
1174 		(void) strlcpy(dsname, name, atp - name + 1);
1175 		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1176 
1177 		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1178 		if (ddsa->ddsa_props != NULL) {
1179 			dsl_props_set_sync_impl(ds->ds_prev,
1180 			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1181 		}
1182 		dsl_dataset_rele(ds, FTAG);
1183 	}
1184 }
1185 
1186 /*
1187  * The snapshots must all be in the same pool.
1188  * All-or-nothing: if there are any failures, nothing will be modified.
1189  */
1190 int
1191 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1192 {
1193 	dsl_dataset_snapshot_arg_t ddsa;
1194 	nvpair_t *pair;
1195 	boolean_t needsuspend;
1196 	int error;
1197 	spa_t *spa;
1198 	char *firstname;
1199 	nvlist_t *suspended = NULL;
1200 
1201 	pair = nvlist_next_nvpair(snaps, NULL);
1202 	if (pair == NULL)
1203 		return (0);
1204 	firstname = nvpair_name(pair);
1205 
1206 	error = spa_open(firstname, &spa, FTAG);
1207 	if (error != 0)
1208 		return (error);
1209 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1210 	spa_close(spa, FTAG);
1211 
1212 	if (needsuspend) {
1213 		suspended = fnvlist_alloc();
1214 		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1215 		    pair = nvlist_next_nvpair(snaps, pair)) {
1216 			char fsname[MAXNAMELEN];
1217 			char *snapname = nvpair_name(pair);
1218 			char *atp;
1219 			void *cookie;
1220 
1221 			atp = strchr(snapname, '@');
1222 			if (atp == NULL) {
1223 				error = SET_ERROR(EINVAL);
1224 				break;
1225 			}
1226 			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1227 
1228 			error = zil_suspend(fsname, &cookie);
1229 			if (error != 0)
1230 				break;
1231 			fnvlist_add_uint64(suspended, fsname,
1232 			    (uintptr_t)cookie);
1233 		}
1234 	}
1235 
1236 	ddsa.ddsa_snaps = snaps;
1237 	ddsa.ddsa_props = props;
1238 	ddsa.ddsa_errors = errors;
1239 
1240 	if (error == 0) {
1241 		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1242 		    dsl_dataset_snapshot_sync, &ddsa,
1243 		    fnvlist_num_pairs(snaps) * 3);
1244 	}
1245 
1246 	if (suspended != NULL) {
1247 		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1248 		    pair = nvlist_next_nvpair(suspended, pair)) {
1249 			zil_resume((void *)(uintptr_t)
1250 			    fnvpair_value_uint64(pair));
1251 		}
1252 		fnvlist_free(suspended);
1253 	}
1254 
1255 	return (error);
1256 }
1257 
1258 typedef struct dsl_dataset_snapshot_tmp_arg {
1259 	const char *ddsta_fsname;
1260 	const char *ddsta_snapname;
1261 	minor_t ddsta_cleanup_minor;
1262 	const char *ddsta_htag;
1263 } dsl_dataset_snapshot_tmp_arg_t;
1264 
1265 static int
1266 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1267 {
1268 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1269 	dsl_pool_t *dp = dmu_tx_pool(tx);
1270 	dsl_dataset_t *ds;
1271 	int error;
1272 
1273 	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1274 	if (error != 0)
1275 		return (error);
1276 
1277 	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1278 	    tx, B_FALSE);
1279 	if (error != 0) {
1280 		dsl_dataset_rele(ds, FTAG);
1281 		return (error);
1282 	}
1283 
1284 	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1285 		dsl_dataset_rele(ds, FTAG);
1286 		return (SET_ERROR(ENOTSUP));
1287 	}
1288 	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1289 	    B_TRUE, tx);
1290 	if (error != 0) {
1291 		dsl_dataset_rele(ds, FTAG);
1292 		return (error);
1293 	}
1294 
1295 	dsl_dataset_rele(ds, FTAG);
1296 	return (0);
1297 }
1298 
1299 static void
1300 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1301 {
1302 	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1303 	dsl_pool_t *dp = dmu_tx_pool(tx);
1304 	dsl_dataset_t *ds;
1305 
1306 	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1307 
1308 	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1309 	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1310 	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1311 	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1312 
1313 	dsl_dataset_rele(ds, FTAG);
1314 }
1315 
1316 int
1317 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1318     minor_t cleanup_minor, const char *htag)
1319 {
1320 	dsl_dataset_snapshot_tmp_arg_t ddsta;
1321 	int error;
1322 	spa_t *spa;
1323 	boolean_t needsuspend;
1324 	void *cookie;
1325 
1326 	ddsta.ddsta_fsname = fsname;
1327 	ddsta.ddsta_snapname = snapname;
1328 	ddsta.ddsta_cleanup_minor = cleanup_minor;
1329 	ddsta.ddsta_htag = htag;
1330 
1331 	error = spa_open(fsname, &spa, FTAG);
1332 	if (error != 0)
1333 		return (error);
1334 	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1335 	spa_close(spa, FTAG);
1336 
1337 	if (needsuspend) {
1338 		error = zil_suspend(fsname, &cookie);
1339 		if (error != 0)
1340 			return (error);
1341 	}
1342 
1343 	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1344 	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1345 
1346 	if (needsuspend)
1347 		zil_resume(cookie);
1348 	return (error);
1349 }
1350 
1351 
1352 void
1353 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1354 {
1355 	ASSERT(dmu_tx_is_syncing(tx));
1356 	ASSERT(ds->ds_objset != NULL);
1357 	ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1358 
1359 	/*
1360 	 * in case we had to change ds_fsid_guid when we opened it,
1361 	 * sync it out now.
1362 	 */
1363 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1364 	ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1365 
1366 	dmu_objset_sync(ds->ds_objset, zio, tx);
1367 }
1368 
1369 static void
1370 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1371 {
1372 	uint64_t count = 0;
1373 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1374 	zap_cursor_t zc;
1375 	zap_attribute_t za;
1376 	nvlist_t *propval = fnvlist_alloc();
1377 	nvlist_t *val = fnvlist_alloc();
1378 
1379 	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1380 
1381 	/*
1382 	 * There may be missing entries in ds_next_clones_obj
1383 	 * due to a bug in a previous version of the code.
1384 	 * Only trust it if it has the right number of entries.
1385 	 */
1386 	if (ds->ds_phys->ds_next_clones_obj != 0) {
1387 		ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1388 		    &count));
1389 	}
1390 	if (count != ds->ds_phys->ds_num_children - 1)
1391 		goto fail;
1392 	for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1393 	    zap_cursor_retrieve(&zc, &za) == 0;
1394 	    zap_cursor_advance(&zc)) {
1395 		dsl_dataset_t *clone;
1396 		char buf[ZFS_MAXNAMELEN];
1397 		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1398 		    za.za_first_integer, FTAG, &clone));
1399 		dsl_dir_name(clone->ds_dir, buf);
1400 		fnvlist_add_boolean(val, buf);
1401 		dsl_dataset_rele(clone, FTAG);
1402 	}
1403 	zap_cursor_fini(&zc);
1404 	fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1405 	fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1406 fail:
1407 	nvlist_free(val);
1408 	nvlist_free(propval);
1409 }
1410 
1411 void
1412 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1413 {
1414 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1415 	uint64_t refd, avail, uobjs, aobjs, ratio;
1416 
1417 	ASSERT(dsl_pool_config_held(dp));
1418 
1419 	ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1420 	    (ds->ds_phys->ds_uncompressed_bytes * 100 /
1421 	    ds->ds_phys->ds_compressed_bytes);
1422 
1423 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1424 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1425 	    ds->ds_phys->ds_uncompressed_bytes);
1426 
1427 	if (dsl_dataset_is_snapshot(ds)) {
1428 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1429 		dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1430 		    ds->ds_phys->ds_unique_bytes);
1431 		get_clones_stat(ds, nv);
1432 	} else {
1433 		dsl_dir_stats(ds->ds_dir, nv);
1434 	}
1435 
1436 	dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1437 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1438 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1439 
1440 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1441 	    ds->ds_phys->ds_creation_time);
1442 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1443 	    ds->ds_phys->ds_creation_txg);
1444 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1445 	    ds->ds_quota);
1446 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1447 	    ds->ds_reserved);
1448 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1449 	    ds->ds_phys->ds_guid);
1450 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1451 	    ds->ds_phys->ds_unique_bytes);
1452 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1453 	    ds->ds_object);
1454 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1455 	    ds->ds_userrefs);
1456 	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1457 	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1458 
1459 	if (ds->ds_phys->ds_prev_snap_obj != 0) {
1460 		uint64_t written, comp, uncomp;
1461 		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1462 		dsl_dataset_t *prev;
1463 
1464 		int err = dsl_dataset_hold_obj(dp,
1465 		    ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1466 		if (err == 0) {
1467 			err = dsl_dataset_space_written(prev, ds, &written,
1468 			    &comp, &uncomp);
1469 			dsl_dataset_rele(prev, FTAG);
1470 			if (err == 0) {
1471 				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1472 				    written);
1473 			}
1474 		}
1475 	}
1476 }
1477 
1478 void
1479 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1480 {
1481 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1482 	ASSERT(dsl_pool_config_held(dp));
1483 
1484 	stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1485 	stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1486 	stat->dds_guid = ds->ds_phys->ds_guid;
1487 	stat->dds_origin[0] = '\0';
1488 	if (dsl_dataset_is_snapshot(ds)) {
1489 		stat->dds_is_snapshot = B_TRUE;
1490 		stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1491 	} else {
1492 		stat->dds_is_snapshot = B_FALSE;
1493 		stat->dds_num_clones = 0;
1494 
1495 		if (dsl_dir_is_clone(ds->ds_dir)) {
1496 			dsl_dataset_t *ods;
1497 
1498 			VERIFY0(dsl_dataset_hold_obj(dp,
1499 			    ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1500 			dsl_dataset_name(ods, stat->dds_origin);
1501 			dsl_dataset_rele(ods, FTAG);
1502 		}
1503 	}
1504 }
1505 
1506 uint64_t
1507 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1508 {
1509 	return (ds->ds_fsid_guid);
1510 }
1511 
1512 void
1513 dsl_dataset_space(dsl_dataset_t *ds,
1514     uint64_t *refdbytesp, uint64_t *availbytesp,
1515     uint64_t *usedobjsp, uint64_t *availobjsp)
1516 {
1517 	*refdbytesp = ds->ds_phys->ds_referenced_bytes;
1518 	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1519 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1520 		*availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1521 	if (ds->ds_quota != 0) {
1522 		/*
1523 		 * Adjust available bytes according to refquota
1524 		 */
1525 		if (*refdbytesp < ds->ds_quota)
1526 			*availbytesp = MIN(*availbytesp,
1527 			    ds->ds_quota - *refdbytesp);
1528 		else
1529 			*availbytesp = 0;
1530 	}
1531 	*usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1532 	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
1533 }
1534 
1535 boolean_t
1536 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1537 {
1538 	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1539 
1540 	ASSERT(dsl_pool_config_held(dp));
1541 	if (snap == NULL)
1542 		return (B_FALSE);
1543 	if (ds->ds_phys->ds_bp.blk_birth >
1544 	    snap->ds_phys->ds_creation_txg) {
1545 		objset_t *os, *os_snap;
1546 		/*
1547 		 * It may be that only the ZIL differs, because it was
1548 		 * reset in the head.  Don't count that as being
1549 		 * modified.
1550 		 */
1551 		if (dmu_objset_from_ds(ds, &os) != 0)
1552 			return (B_TRUE);
1553 		if (dmu_objset_from_ds(snap, &os_snap) != 0)
1554 			return (B_TRUE);
1555 		return (bcmp(&os->os_phys->os_meta_dnode,
1556 		    &os_snap->os_phys->os_meta_dnode,
1557 		    sizeof (os->os_phys->os_meta_dnode)) != 0);
1558 	}
1559 	return (B_FALSE);
1560 }
1561 
1562 typedef struct dsl_dataset_rename_snapshot_arg {
1563 	const char *ddrsa_fsname;
1564 	const char *ddrsa_oldsnapname;
1565 	const char *ddrsa_newsnapname;
1566 	boolean_t ddrsa_recursive;
1567 	dmu_tx_t *ddrsa_tx;
1568 } dsl_dataset_rename_snapshot_arg_t;
1569 
1570 /* ARGSUSED */
1571 static int
1572 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1573     dsl_dataset_t *hds, void *arg)
1574 {
1575 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1576 	int error;
1577 	uint64_t val;
1578 
1579 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1580 	if (error != 0) {
1581 		/* ignore nonexistent snapshots */
1582 		return (error == ENOENT ? 0 : error);
1583 	}
1584 
1585 	/* new name should not exist */
1586 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1587 	if (error == 0)
1588 		error = SET_ERROR(EEXIST);
1589 	else if (error == ENOENT)
1590 		error = 0;
1591 
1592 	/* dataset name + 1 for the "@" + the new snapshot name must fit */
1593 	if (dsl_dir_namelen(hds->ds_dir) + 1 +
1594 	    strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1595 		error = SET_ERROR(ENAMETOOLONG);
1596 
1597 	return (error);
1598 }
1599 
1600 static int
1601 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1602 {
1603 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1604 	dsl_pool_t *dp = dmu_tx_pool(tx);
1605 	dsl_dataset_t *hds;
1606 	int error;
1607 
1608 	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1609 	if (error != 0)
1610 		return (error);
1611 
1612 	if (ddrsa->ddrsa_recursive) {
1613 		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1614 		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
1615 		    DS_FIND_CHILDREN);
1616 	} else {
1617 		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1618 	}
1619 	dsl_dataset_rele(hds, FTAG);
1620 	return (error);
1621 }
1622 
1623 static int
1624 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1625     dsl_dataset_t *hds, void *arg)
1626 {
1627 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1628 	dsl_dataset_t *ds;
1629 	uint64_t val;
1630 	dmu_tx_t *tx = ddrsa->ddrsa_tx;
1631 	int error;
1632 
1633 	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1634 	ASSERT(error == 0 || error == ENOENT);
1635 	if (error == ENOENT) {
1636 		/* ignore nonexistent snapshots */
1637 		return (0);
1638 	}
1639 
1640 	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1641 
1642 	/* log before we change the name */
1643 	spa_history_log_internal_ds(ds, "rename", tx,
1644 	    "-> @%s", ddrsa->ddrsa_newsnapname);
1645 
1646 	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx));
1647 	mutex_enter(&ds->ds_lock);
1648 	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1649 	mutex_exit(&ds->ds_lock);
1650 	VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1651 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1652 
1653 	dsl_dataset_rele(ds, FTAG);
1654 	return (0);
1655 }
1656 
1657 static void
1658 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1659 {
1660 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1661 	dsl_pool_t *dp = dmu_tx_pool(tx);
1662 	dsl_dataset_t *hds;
1663 
1664 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1665 	ddrsa->ddrsa_tx = tx;
1666 	if (ddrsa->ddrsa_recursive) {
1667 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1668 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1669 		    DS_FIND_CHILDREN));
1670 	} else {
1671 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1672 	}
1673 	dsl_dataset_rele(hds, FTAG);
1674 }
1675 
1676 int
1677 dsl_dataset_rename_snapshot(const char *fsname,
1678     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1679 {
1680 	dsl_dataset_rename_snapshot_arg_t ddrsa;
1681 
1682 	ddrsa.ddrsa_fsname = fsname;
1683 	ddrsa.ddrsa_oldsnapname = oldsnapname;
1684 	ddrsa.ddrsa_newsnapname = newsnapname;
1685 	ddrsa.ddrsa_recursive = recursive;
1686 
1687 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1688 	    dsl_dataset_rename_snapshot_sync, &ddrsa, 1));
1689 }
1690 
1691 /*
1692  * If we're doing an ownership handoff, we need to make sure that there is
1693  * only one long hold on the dataset.  We're not allowed to change anything here
1694  * so we don't permanently release the long hold or regular hold here.  We want
1695  * to do this only when syncing to avoid the dataset unexpectedly going away
1696  * when we release the long hold.
1697  */
1698 static int
1699 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1700 {
1701 	boolean_t held;
1702 
1703 	if (!dmu_tx_is_syncing(tx))
1704 		return (0);
1705 
1706 	if (owner != NULL) {
1707 		VERIFY3P(ds->ds_owner, ==, owner);
1708 		dsl_dataset_long_rele(ds, owner);
1709 	}
1710 
1711 	held = dsl_dataset_long_held(ds);
1712 
1713 	if (owner != NULL)
1714 		dsl_dataset_long_hold(ds, owner);
1715 
1716 	if (held)
1717 		return (SET_ERROR(EBUSY));
1718 
1719 	return (0);
1720 }
1721 
1722 typedef struct dsl_dataset_rollback_arg {
1723 	const char *ddra_fsname;
1724 	void *ddra_owner;
1725 	nvlist_t *ddra_result;
1726 } dsl_dataset_rollback_arg_t;
1727 
1728 static int
1729 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1730 {
1731 	dsl_dataset_rollback_arg_t *ddra = arg;
1732 	dsl_pool_t *dp = dmu_tx_pool(tx);
1733 	dsl_dataset_t *ds;
1734 	int64_t unused_refres_delta;
1735 	int error;
1736 
1737 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1738 	if (error != 0)
1739 		return (error);
1740 
1741 	/* must not be a snapshot */
1742 	if (dsl_dataset_is_snapshot(ds)) {
1743 		dsl_dataset_rele(ds, FTAG);
1744 		return (SET_ERROR(EINVAL));
1745 	}
1746 
1747 	/* must have a most recent snapshot */
1748 	if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1749 		dsl_dataset_rele(ds, FTAG);
1750 		return (SET_ERROR(EINVAL));
1751 	}
1752 
1753 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1754 	if (error != 0) {
1755 		dsl_dataset_rele(ds, FTAG);
1756 		return (error);
1757 	}
1758 
1759 	/*
1760 	 * Check if the snap we are rolling back to uses more than
1761 	 * the refquota.
1762 	 */
1763 	if (ds->ds_quota != 0 &&
1764 	    ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1765 		dsl_dataset_rele(ds, FTAG);
1766 		return (SET_ERROR(EDQUOT));
1767 	}
1768 
1769 	/*
1770 	 * When we do the clone swap, we will temporarily use more space
1771 	 * due to the refreservation (the head will no longer have any
1772 	 * unique space, so the entire amount of the refreservation will need
1773 	 * to be free).  We will immediately destroy the clone, freeing
1774 	 * this space, but the freeing happens over many txg's.
1775 	 */
1776 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1777 	    ds->ds_phys->ds_unique_bytes);
1778 
1779 	if (unused_refres_delta > 0 &&
1780 	    unused_refres_delta >
1781 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1782 		dsl_dataset_rele(ds, FTAG);
1783 		return (SET_ERROR(ENOSPC));
1784 	}
1785 
1786 	dsl_dataset_rele(ds, FTAG);
1787 	return (0);
1788 }
1789 
1790 static void
1791 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1792 {
1793 	dsl_dataset_rollback_arg_t *ddra = arg;
1794 	dsl_pool_t *dp = dmu_tx_pool(tx);
1795 	dsl_dataset_t *ds, *clone;
1796 	uint64_t cloneobj;
1797 	char namebuf[ZFS_MAXNAMELEN];
1798 
1799 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
1800 
1801 	dsl_dataset_name(ds->ds_prev, namebuf);
1802 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
1803 
1804 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1805 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1806 
1807 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1808 
1809 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1810 	dsl_dataset_zero_zil(ds, tx);
1811 
1812 	dsl_destroy_head_sync_impl(clone, tx);
1813 
1814 	dsl_dataset_rele(clone, FTAG);
1815 	dsl_dataset_rele(ds, FTAG);
1816 }
1817 
1818 /*
1819  * Rolls back the given filesystem or volume to the most recent snapshot.
1820  * The name of the most recent snapshot will be returned under key "target"
1821  * in the result nvlist.
1822  *
1823  * If owner != NULL:
1824  * - The existing dataset MUST be owned by the specified owner at entry
1825  * - Upon return, dataset will still be held by the same owner, whether we
1826  *   succeed or not.
1827  *
1828  * This mode is required any time the existing filesystem is mounted.  See
1829  * notes above zfs_suspend_fs() for further details.
1830  */
1831 int
1832 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
1833 {
1834 	dsl_dataset_rollback_arg_t ddra;
1835 
1836 	ddra.ddra_fsname = fsname;
1837 	ddra.ddra_owner = owner;
1838 	ddra.ddra_result = result;
1839 
1840 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1841 	    dsl_dataset_rollback_sync, &ddra, 1));
1842 }
1843 
1844 struct promotenode {
1845 	list_node_t link;
1846 	dsl_dataset_t *ds;
1847 };
1848 
1849 typedef struct dsl_dataset_promote_arg {
1850 	const char *ddpa_clonename;
1851 	dsl_dataset_t *ddpa_clone;
1852 	list_t shared_snaps, origin_snaps, clone_snaps;
1853 	dsl_dataset_t *origin_origin; /* origin of the origin */
1854 	uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
1855 	char *err_ds;
1856 } dsl_dataset_promote_arg_t;
1857 
1858 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
1859 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
1860     void *tag);
1861 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
1862 
1863 static int
1864 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
1865 {
1866 	dsl_dataset_promote_arg_t *ddpa = arg;
1867 	dsl_pool_t *dp = dmu_tx_pool(tx);
1868 	dsl_dataset_t *hds;
1869 	struct promotenode *snap;
1870 	dsl_dataset_t *origin_ds;
1871 	int err;
1872 	uint64_t unused;
1873 
1874 	err = promote_hold(ddpa, dp, FTAG);
1875 	if (err != 0)
1876 		return (err);
1877 
1878 	hds = ddpa->ddpa_clone;
1879 
1880 	if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1881 		promote_rele(ddpa, FTAG);
1882 		return (SET_ERROR(EXDEV));
1883 	}
1884 
1885 	/*
1886 	 * Compute and check the amount of space to transfer.  Since this is
1887 	 * so expensive, don't do the preliminary check.
1888 	 */
1889 	if (!dmu_tx_is_syncing(tx)) {
1890 		promote_rele(ddpa, FTAG);
1891 		return (0);
1892 	}
1893 
1894 	snap = list_head(&ddpa->shared_snaps);
1895 	origin_ds = snap->ds;
1896 
1897 	/* compute origin's new unique space */
1898 	snap = list_tail(&ddpa->clone_snaps);
1899 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1900 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
1901 	    origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1902 	    &ddpa->unique, &unused, &unused);
1903 
1904 	/*
1905 	 * Walk the snapshots that we are moving
1906 	 *
1907 	 * Compute space to transfer.  Consider the incremental changes
1908 	 * to used by each snapshot:
1909 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1910 	 * So each snapshot gave birth to:
1911 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1912 	 * So a sequence would look like:
1913 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1914 	 * Which simplifies to:
1915 	 * uN + kN + kN-1 + ... + k1 + k0
1916 	 * Note however, if we stop before we reach the ORIGIN we get:
1917 	 * uN + kN + kN-1 + ... + kM - uM-1
1918 	 */
1919 	ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
1920 	ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
1921 	ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
1922 	for (snap = list_head(&ddpa->shared_snaps); snap;
1923 	    snap = list_next(&ddpa->shared_snaps, snap)) {
1924 		uint64_t val, dlused, dlcomp, dluncomp;
1925 		dsl_dataset_t *ds = snap->ds;
1926 
1927 		/*
1928 		 * If there are long holds, we won't be able to evict
1929 		 * the objset.
1930 		 */
1931 		if (dsl_dataset_long_held(ds)) {
1932 			err = SET_ERROR(EBUSY);
1933 			goto out;
1934 		}
1935 
1936 		/* Check that the snapshot name does not conflict */
1937 		VERIFY0(dsl_dataset_get_snapname(ds));
1938 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
1939 		if (err == 0) {
1940 			(void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
1941 			err = SET_ERROR(EEXIST);
1942 			goto out;
1943 		}
1944 		if (err != ENOENT)
1945 			goto out;
1946 
1947 		/* The very first snapshot does not have a deadlist */
1948 		if (ds->ds_phys->ds_prev_snap_obj == 0)
1949 			continue;
1950 
1951 		dsl_deadlist_space(&ds->ds_deadlist,
1952 		    &dlused, &dlcomp, &dluncomp);
1953 		ddpa->used += dlused;
1954 		ddpa->comp += dlcomp;
1955 		ddpa->uncomp += dluncomp;
1956 	}
1957 
1958 	/*
1959 	 * If we are a clone of a clone then we never reached ORIGIN,
1960 	 * so we need to subtract out the clone origin's used space.
1961 	 */
1962 	if (ddpa->origin_origin) {
1963 		ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
1964 		ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
1965 		ddpa->uncomp -=
1966 		    ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
1967 	}
1968 
1969 	/* Check that there is enough space here */
1970 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
1971 	    ddpa->used);
1972 	if (err != 0)
1973 		goto out;
1974 
1975 	/*
1976 	 * Compute the amounts of space that will be used by snapshots
1977 	 * after the promotion (for both origin and clone).  For each,
1978 	 * it is the amount of space that will be on all of their
1979 	 * deadlists (that was not born before their new origin).
1980 	 */
1981 	if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1982 		uint64_t space;
1983 
1984 		/*
1985 		 * Note, typically this will not be a clone of a clone,
1986 		 * so dd_origin_txg will be < TXG_INITIAL, so
1987 		 * these snaplist_space() -> dsl_deadlist_space_range()
1988 		 * calls will be fast because they do not have to
1989 		 * iterate over all bps.
1990 		 */
1991 		snap = list_head(&ddpa->origin_snaps);
1992 		err = snaplist_space(&ddpa->shared_snaps,
1993 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
1994 		if (err != 0)
1995 			goto out;
1996 
1997 		err = snaplist_space(&ddpa->clone_snaps,
1998 		    snap->ds->ds_dir->dd_origin_txg, &space);
1999 		if (err != 0)
2000 			goto out;
2001 		ddpa->cloneusedsnap += space;
2002 	}
2003 	if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2004 		err = snaplist_space(&ddpa->origin_snaps,
2005 		    origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
2006 		if (err != 0)
2007 			goto out;
2008 	}
2009 
2010 out:
2011 	promote_rele(ddpa, FTAG);
2012 	return (err);
2013 }
2014 
2015 static void
2016 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2017 {
2018 	dsl_dataset_promote_arg_t *ddpa = arg;
2019 	dsl_pool_t *dp = dmu_tx_pool(tx);
2020 	dsl_dataset_t *hds;
2021 	struct promotenode *snap;
2022 	dsl_dataset_t *origin_ds;
2023 	dsl_dataset_t *origin_head;
2024 	dsl_dir_t *dd;
2025 	dsl_dir_t *odd = NULL;
2026 	uint64_t oldnext_obj;
2027 	int64_t delta;
2028 
2029 	VERIFY0(promote_hold(ddpa, dp, FTAG));
2030 	hds = ddpa->ddpa_clone;
2031 
2032 	ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
2033 
2034 	snap = list_head(&ddpa->shared_snaps);
2035 	origin_ds = snap->ds;
2036 	dd = hds->ds_dir;
2037 
2038 	snap = list_head(&ddpa->origin_snaps);
2039 	origin_head = snap->ds;
2040 
2041 	/*
2042 	 * We need to explicitly open odd, since origin_ds's dd will be
2043 	 * changing.
2044 	 */
2045 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2046 	    NULL, FTAG, &odd));
2047 
2048 	/* change origin's next snap */
2049 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2050 	oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2051 	snap = list_tail(&ddpa->clone_snaps);
2052 	ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2053 	origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2054 
2055 	/* change the origin's next clone */
2056 	if (origin_ds->ds_phys->ds_next_clones_obj) {
2057 		dsl_dataset_remove_from_next_clones(origin_ds,
2058 		    snap->ds->ds_object, tx);
2059 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2060 		    origin_ds->ds_phys->ds_next_clones_obj,
2061 		    oldnext_obj, tx));
2062 	}
2063 
2064 	/* change origin */
2065 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
2066 	ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2067 	dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2068 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2069 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
2070 	odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2071 	origin_head->ds_dir->dd_origin_txg =
2072 	    origin_ds->ds_phys->ds_creation_txg;
2073 
2074 	/* change dd_clone entries */
2075 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2076 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2077 		    odd->dd_phys->dd_clones, hds->ds_object, tx));
2078 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2079 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2080 		    hds->ds_object, tx));
2081 
2082 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
2083 		    ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2084 		    origin_head->ds_object, tx));
2085 		if (dd->dd_phys->dd_clones == 0) {
2086 			dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2087 			    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2088 		}
2089 		VERIFY0(zap_add_int(dp->dp_meta_objset,
2090 		    dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2091 	}
2092 
2093 	/* move snapshots to this dir */
2094 	for (snap = list_head(&ddpa->shared_snaps); snap;
2095 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2096 		dsl_dataset_t *ds = snap->ds;
2097 
2098 		/*
2099 		 * Property callbacks are registered to a particular
2100 		 * dsl_dir.  Since ours is changing, evict the objset
2101 		 * so that they will be unregistered from the old dsl_dir.
2102 		 */
2103 		if (ds->ds_objset) {
2104 			dmu_objset_evict(ds->ds_objset);
2105 			ds->ds_objset = NULL;
2106 		}
2107 
2108 		/* move snap name entry */
2109 		VERIFY0(dsl_dataset_get_snapname(ds));
2110 		VERIFY0(dsl_dataset_snap_remove(origin_head,
2111 		    ds->ds_snapname, tx));
2112 		VERIFY0(zap_add(dp->dp_meta_objset,
2113 		    hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2114 		    8, 1, &ds->ds_object, tx));
2115 
2116 		/* change containing dsl_dir */
2117 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2118 		ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2119 		ds->ds_phys->ds_dir_obj = dd->dd_object;
2120 		ASSERT3P(ds->ds_dir, ==, odd);
2121 		dsl_dir_rele(ds->ds_dir, ds);
2122 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2123 		    NULL, ds, &ds->ds_dir));
2124 
2125 		/* move any clone references */
2126 		if (ds->ds_phys->ds_next_clones_obj &&
2127 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2128 			zap_cursor_t zc;
2129 			zap_attribute_t za;
2130 
2131 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
2132 			    ds->ds_phys->ds_next_clones_obj);
2133 			    zap_cursor_retrieve(&zc, &za) == 0;
2134 			    zap_cursor_advance(&zc)) {
2135 				dsl_dataset_t *cnds;
2136 				uint64_t o;
2137 
2138 				if (za.za_first_integer == oldnext_obj) {
2139 					/*
2140 					 * We've already moved the
2141 					 * origin's reference.
2142 					 */
2143 					continue;
2144 				}
2145 
2146 				VERIFY0(dsl_dataset_hold_obj(dp,
2147 				    za.za_first_integer, FTAG, &cnds));
2148 				o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2149 
2150 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
2151 				    odd->dd_phys->dd_clones, o, tx));
2152 				VERIFY0(zap_add_int(dp->dp_meta_objset,
2153 				    dd->dd_phys->dd_clones, o, tx));
2154 				dsl_dataset_rele(cnds, FTAG);
2155 			}
2156 			zap_cursor_fini(&zc);
2157 		}
2158 
2159 		ASSERT(!dsl_prop_hascb(ds));
2160 	}
2161 
2162 	/*
2163 	 * Change space accounting.
2164 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2165 	 * both be valid, or both be 0 (resulting in delta == 0).  This
2166 	 * is true for each of {clone,origin} independently.
2167 	 */
2168 
2169 	delta = ddpa->cloneusedsnap -
2170 	    dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2171 	ASSERT3S(delta, >=, 0);
2172 	ASSERT3U(ddpa->used, >=, delta);
2173 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2174 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
2175 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2176 
2177 	delta = ddpa->originusedsnap -
2178 	    odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2179 	ASSERT3S(delta, <=, 0);
2180 	ASSERT3U(ddpa->used, >=, -delta);
2181 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2182 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
2183 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2184 
2185 	origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2186 
2187 	/* log history record */
2188 	spa_history_log_internal_ds(hds, "promote", tx, "");
2189 
2190 	dsl_dir_rele(odd, FTAG);
2191 	promote_rele(ddpa, FTAG);
2192 }
2193 
2194 /*
2195  * Make a list of dsl_dataset_t's for the snapshots between first_obj
2196  * (exclusive) and last_obj (inclusive).  The list will be in reverse
2197  * order (last_obj will be the list_head()).  If first_obj == 0, do all
2198  * snapshots back to this dataset's origin.
2199  */
2200 static int
2201 snaplist_make(dsl_pool_t *dp,
2202     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2203 {
2204 	uint64_t obj = last_obj;
2205 
2206 	list_create(l, sizeof (struct promotenode),
2207 	    offsetof(struct promotenode, link));
2208 
2209 	while (obj != first_obj) {
2210 		dsl_dataset_t *ds;
2211 		struct promotenode *snap;
2212 		int err;
2213 
2214 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2215 		ASSERT(err != ENOENT);
2216 		if (err != 0)
2217 			return (err);
2218 
2219 		if (first_obj == 0)
2220 			first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2221 
2222 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2223 		snap->ds = ds;
2224 		list_insert_tail(l, snap);
2225 		obj = ds->ds_phys->ds_prev_snap_obj;
2226 	}
2227 
2228 	return (0);
2229 }
2230 
2231 static int
2232 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2233 {
2234 	struct promotenode *snap;
2235 
2236 	*spacep = 0;
2237 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2238 		uint64_t used, comp, uncomp;
2239 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2240 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
2241 		*spacep += used;
2242 	}
2243 	return (0);
2244 }
2245 
2246 static void
2247 snaplist_destroy(list_t *l, void *tag)
2248 {
2249 	struct promotenode *snap;
2250 
2251 	if (l == NULL || !list_link_active(&l->list_head))
2252 		return;
2253 
2254 	while ((snap = list_tail(l)) != NULL) {
2255 		list_remove(l, snap);
2256 		dsl_dataset_rele(snap->ds, tag);
2257 		kmem_free(snap, sizeof (*snap));
2258 	}
2259 	list_destroy(l);
2260 }
2261 
2262 static int
2263 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2264 {
2265 	int error;
2266 	dsl_dir_t *dd;
2267 	struct promotenode *snap;
2268 
2269 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2270 	    &ddpa->ddpa_clone);
2271 	if (error != 0)
2272 		return (error);
2273 	dd = ddpa->ddpa_clone->ds_dir;
2274 
2275 	if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2276 	    !dsl_dir_is_clone(dd)) {
2277 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
2278 		return (SET_ERROR(EINVAL));
2279 	}
2280 
2281 	error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2282 	    &ddpa->shared_snaps, tag);
2283 	if (error != 0)
2284 		goto out;
2285 
2286 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2287 	    &ddpa->clone_snaps, tag);
2288 	if (error != 0)
2289 		goto out;
2290 
2291 	snap = list_head(&ddpa->shared_snaps);
2292 	ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2293 	error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2294 	    snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2295 	    &ddpa->origin_snaps, tag);
2296 	if (error != 0)
2297 		goto out;
2298 
2299 	if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2300 		error = dsl_dataset_hold_obj(dp,
2301 		    snap->ds->ds_dir->dd_phys->dd_origin_obj,
2302 		    tag, &ddpa->origin_origin);
2303 		if (error != 0)
2304 			goto out;
2305 	}
2306 out:
2307 	if (error != 0)
2308 		promote_rele(ddpa, tag);
2309 	return (error);
2310 }
2311 
2312 static void
2313 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2314 {
2315 	snaplist_destroy(&ddpa->shared_snaps, tag);
2316 	snaplist_destroy(&ddpa->clone_snaps, tag);
2317 	snaplist_destroy(&ddpa->origin_snaps, tag);
2318 	if (ddpa->origin_origin != NULL)
2319 		dsl_dataset_rele(ddpa->origin_origin, tag);
2320 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
2321 }
2322 
2323 /*
2324  * Promote a clone.
2325  *
2326  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2327  * in with the name.  (It must be at least MAXNAMELEN bytes long.)
2328  */
2329 int
2330 dsl_dataset_promote(const char *name, char *conflsnap)
2331 {
2332 	dsl_dataset_promote_arg_t ddpa = { 0 };
2333 	uint64_t numsnaps;
2334 	int error;
2335 	objset_t *os;
2336 
2337 	/*
2338 	 * We will modify space proportional to the number of
2339 	 * snapshots.  Compute numsnaps.
2340 	 */
2341 	error = dmu_objset_hold(name, FTAG, &os);
2342 	if (error != 0)
2343 		return (error);
2344 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2345 	    dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2346 	dmu_objset_rele(os, FTAG);
2347 	if (error != 0)
2348 		return (error);
2349 
2350 	ddpa.ddpa_clonename = name;
2351 	ddpa.err_ds = conflsnap;
2352 
2353 	return (dsl_sync_task(name, dsl_dataset_promote_check,
2354 	    dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2355 }
2356 
2357 int
2358 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2359     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2360 {
2361 	int64_t unused_refres_delta;
2362 
2363 	/* they should both be heads */
2364 	if (dsl_dataset_is_snapshot(clone) ||
2365 	    dsl_dataset_is_snapshot(origin_head))
2366 		return (SET_ERROR(EINVAL));
2367 
2368 	/* if we are not forcing, the branch point should be just before them */
2369 	if (!force && clone->ds_prev != origin_head->ds_prev)
2370 		return (SET_ERROR(EINVAL));
2371 
2372 	/* clone should be the clone (unless they are unrelated) */
2373 	if (clone->ds_prev != NULL &&
2374 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2375 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
2376 		return (SET_ERROR(EINVAL));
2377 
2378 	/* the clone should be a child of the origin */
2379 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2380 		return (SET_ERROR(EINVAL));
2381 
2382 	/* origin_head shouldn't be modified unless 'force' */
2383 	if (!force &&
2384 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2385 		return (SET_ERROR(ETXTBSY));
2386 
2387 	/* origin_head should have no long holds (e.g. is not mounted) */
2388 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
2389 		return (SET_ERROR(EBUSY));
2390 
2391 	/* check amount of any unconsumed refreservation */
2392 	unused_refres_delta =
2393 	    (int64_t)MIN(origin_head->ds_reserved,
2394 	    origin_head->ds_phys->ds_unique_bytes) -
2395 	    (int64_t)MIN(origin_head->ds_reserved,
2396 	    clone->ds_phys->ds_unique_bytes);
2397 
2398 	if (unused_refres_delta > 0 &&
2399 	    unused_refres_delta >
2400 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2401 		return (SET_ERROR(ENOSPC));
2402 
2403 	/* clone can't be over the head's refquota */
2404 	if (origin_head->ds_quota != 0 &&
2405 	    clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2406 		return (SET_ERROR(EDQUOT));
2407 
2408 	return (0);
2409 }
2410 
2411 void
2412 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2413     dsl_dataset_t *origin_head, dmu_tx_t *tx)
2414 {
2415 	dsl_pool_t *dp = dmu_tx_pool(tx);
2416 	int64_t unused_refres_delta;
2417 
2418 	ASSERT(clone->ds_reserved == 0);
2419 	ASSERT(origin_head->ds_quota == 0 ||
2420 	    clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2421 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2422 
2423 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
2424 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2425 
2426 	if (clone->ds_objset != NULL) {
2427 		dmu_objset_evict(clone->ds_objset);
2428 		clone->ds_objset = NULL;
2429 	}
2430 
2431 	if (origin_head->ds_objset != NULL) {
2432 		dmu_objset_evict(origin_head->ds_objset);
2433 		origin_head->ds_objset = NULL;
2434 	}
2435 
2436 	unused_refres_delta =
2437 	    (int64_t)MIN(origin_head->ds_reserved,
2438 	    origin_head->ds_phys->ds_unique_bytes) -
2439 	    (int64_t)MIN(origin_head->ds_reserved,
2440 	    clone->ds_phys->ds_unique_bytes);
2441 
2442 	/*
2443 	 * Reset origin's unique bytes, if it exists.
2444 	 */
2445 	if (clone->ds_prev) {
2446 		dsl_dataset_t *origin = clone->ds_prev;
2447 		uint64_t comp, uncomp;
2448 
2449 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
2450 		dsl_deadlist_space_range(&clone->ds_deadlist,
2451 		    origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2452 		    &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2453 	}
2454 
2455 	/* swap blkptrs */
2456 	{
2457 		blkptr_t tmp;
2458 		tmp = origin_head->ds_phys->ds_bp;
2459 		origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2460 		clone->ds_phys->ds_bp = tmp;
2461 	}
2462 
2463 	/* set dd_*_bytes */
2464 	{
2465 		int64_t dused, dcomp, duncomp;
2466 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
2467 		uint64_t odl_used, odl_comp, odl_uncomp;
2468 
2469 		ASSERT3U(clone->ds_dir->dd_phys->
2470 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
2471 
2472 		dsl_deadlist_space(&clone->ds_deadlist,
2473 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2474 		dsl_deadlist_space(&origin_head->ds_deadlist,
2475 		    &odl_used, &odl_comp, &odl_uncomp);
2476 
2477 		dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2478 		    (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2479 		dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2480 		    (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2481 		duncomp = clone->ds_phys->ds_uncompressed_bytes +
2482 		    cdl_uncomp -
2483 		    (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2484 
2485 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2486 		    dused, dcomp, duncomp, tx);
2487 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2488 		    -dused, -dcomp, -duncomp, tx);
2489 
2490 		/*
2491 		 * The difference in the space used by snapshots is the
2492 		 * difference in snapshot space due to the head's
2493 		 * deadlist (since that's the only thing that's
2494 		 * changing that affects the snapused).
2495 		 */
2496 		dsl_deadlist_space_range(&clone->ds_deadlist,
2497 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2498 		    &cdl_used, &cdl_comp, &cdl_uncomp);
2499 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
2500 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2501 		    &odl_used, &odl_comp, &odl_uncomp);
2502 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2503 		    DD_USED_HEAD, DD_USED_SNAP, tx);
2504 	}
2505 
2506 	/* swap ds_*_bytes */
2507 	SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2508 	    clone->ds_phys->ds_referenced_bytes);
2509 	SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2510 	    clone->ds_phys->ds_compressed_bytes);
2511 	SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2512 	    clone->ds_phys->ds_uncompressed_bytes);
2513 	SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2514 	    clone->ds_phys->ds_unique_bytes);
2515 
2516 	/* apply any parent delta for change in unconsumed refreservation */
2517 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2518 	    unused_refres_delta, 0, 0, tx);
2519 
2520 	/*
2521 	 * Swap deadlists.
2522 	 */
2523 	dsl_deadlist_close(&clone->ds_deadlist);
2524 	dsl_deadlist_close(&origin_head->ds_deadlist);
2525 	SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2526 	    clone->ds_phys->ds_deadlist_obj);
2527 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2528 	    clone->ds_phys->ds_deadlist_obj);
2529 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2530 	    origin_head->ds_phys->ds_deadlist_obj);
2531 
2532 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2533 
2534 	spa_history_log_internal_ds(clone, "clone swap", tx,
2535 	    "parent=%s", origin_head->ds_dir->dd_myname);
2536 }
2537 
2538 /*
2539  * Given a pool name and a dataset object number in that pool,
2540  * return the name of that dataset.
2541  */
2542 int
2543 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2544 {
2545 	dsl_pool_t *dp;
2546 	dsl_dataset_t *ds;
2547 	int error;
2548 
2549 	error = dsl_pool_hold(pname, FTAG, &dp);
2550 	if (error != 0)
2551 		return (error);
2552 
2553 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2554 	if (error == 0) {
2555 		dsl_dataset_name(ds, buf);
2556 		dsl_dataset_rele(ds, FTAG);
2557 	}
2558 	dsl_pool_rele(dp, FTAG);
2559 
2560 	return (error);
2561 }
2562 
2563 int
2564 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2565     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2566 {
2567 	int error = 0;
2568 
2569 	ASSERT3S(asize, >, 0);
2570 
2571 	/*
2572 	 * *ref_rsrv is the portion of asize that will come from any
2573 	 * unconsumed refreservation space.
2574 	 */
2575 	*ref_rsrv = 0;
2576 
2577 	mutex_enter(&ds->ds_lock);
2578 	/*
2579 	 * Make a space adjustment for reserved bytes.
2580 	 */
2581 	if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2582 		ASSERT3U(*used, >=,
2583 		    ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2584 		*used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2585 		*ref_rsrv =
2586 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
2587 	}
2588 
2589 	if (!check_quota || ds->ds_quota == 0) {
2590 		mutex_exit(&ds->ds_lock);
2591 		return (0);
2592 	}
2593 	/*
2594 	 * If they are requesting more space, and our current estimate
2595 	 * is over quota, they get to try again unless the actual
2596 	 * on-disk is over quota and there are no pending changes (which
2597 	 * may free up space for us).
2598 	 */
2599 	if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2600 		if (inflight > 0 ||
2601 		    ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2602 			error = SET_ERROR(ERESTART);
2603 		else
2604 			error = SET_ERROR(EDQUOT);
2605 	}
2606 	mutex_exit(&ds->ds_lock);
2607 
2608 	return (error);
2609 }
2610 
2611 typedef struct dsl_dataset_set_qr_arg {
2612 	const char *ddsqra_name;
2613 	zprop_source_t ddsqra_source;
2614 	uint64_t ddsqra_value;
2615 } dsl_dataset_set_qr_arg_t;
2616 
2617 
2618 /* ARGSUSED */
2619 static int
2620 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2621 {
2622 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2623 	dsl_pool_t *dp = dmu_tx_pool(tx);
2624 	dsl_dataset_t *ds;
2625 	int error;
2626 	uint64_t newval;
2627 
2628 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2629 		return (SET_ERROR(ENOTSUP));
2630 
2631 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2632 	if (error != 0)
2633 		return (error);
2634 
2635 	if (dsl_dataset_is_snapshot(ds)) {
2636 		dsl_dataset_rele(ds, FTAG);
2637 		return (SET_ERROR(EINVAL));
2638 	}
2639 
2640 	error = dsl_prop_predict(ds->ds_dir,
2641 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2642 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2643 	if (error != 0) {
2644 		dsl_dataset_rele(ds, FTAG);
2645 		return (error);
2646 	}
2647 
2648 	if (newval == 0) {
2649 		dsl_dataset_rele(ds, FTAG);
2650 		return (0);
2651 	}
2652 
2653 	if (newval < ds->ds_phys->ds_referenced_bytes ||
2654 	    newval < ds->ds_reserved) {
2655 		dsl_dataset_rele(ds, FTAG);
2656 		return (SET_ERROR(ENOSPC));
2657 	}
2658 
2659 	dsl_dataset_rele(ds, FTAG);
2660 	return (0);
2661 }
2662 
2663 static void
2664 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2665 {
2666 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2667 	dsl_pool_t *dp = dmu_tx_pool(tx);
2668 	dsl_dataset_t *ds;
2669 	uint64_t newval;
2670 
2671 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2672 
2673 	dsl_prop_set_sync_impl(ds,
2674 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2675 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2676 	    &ddsqra->ddsqra_value, tx);
2677 
2678 	VERIFY0(dsl_prop_get_int_ds(ds,
2679 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2680 
2681 	if (ds->ds_quota != newval) {
2682 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
2683 		ds->ds_quota = newval;
2684 	}
2685 	dsl_dataset_rele(ds, FTAG);
2686 }
2687 
2688 int
2689 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2690     uint64_t refquota)
2691 {
2692 	dsl_dataset_set_qr_arg_t ddsqra;
2693 
2694 	ddsqra.ddsqra_name = dsname;
2695 	ddsqra.ddsqra_source = source;
2696 	ddsqra.ddsqra_value = refquota;
2697 
2698 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2699 	    dsl_dataset_set_refquota_sync, &ddsqra, 0));
2700 }
2701 
2702 static int
2703 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2704 {
2705 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2706 	dsl_pool_t *dp = dmu_tx_pool(tx);
2707 	dsl_dataset_t *ds;
2708 	int error;
2709 	uint64_t newval, unique;
2710 
2711 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2712 		return (SET_ERROR(ENOTSUP));
2713 
2714 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2715 	if (error != 0)
2716 		return (error);
2717 
2718 	if (dsl_dataset_is_snapshot(ds)) {
2719 		dsl_dataset_rele(ds, FTAG);
2720 		return (SET_ERROR(EINVAL));
2721 	}
2722 
2723 	error = dsl_prop_predict(ds->ds_dir,
2724 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2725 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2726 	if (error != 0) {
2727 		dsl_dataset_rele(ds, FTAG);
2728 		return (error);
2729 	}
2730 
2731 	/*
2732 	 * If we are doing the preliminary check in open context, the
2733 	 * space estimates may be inaccurate.
2734 	 */
2735 	if (!dmu_tx_is_syncing(tx)) {
2736 		dsl_dataset_rele(ds, FTAG);
2737 		return (0);
2738 	}
2739 
2740 	mutex_enter(&ds->ds_lock);
2741 	if (!DS_UNIQUE_IS_ACCURATE(ds))
2742 		dsl_dataset_recalc_head_uniq(ds);
2743 	unique = ds->ds_phys->ds_unique_bytes;
2744 	mutex_exit(&ds->ds_lock);
2745 
2746 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2747 		uint64_t delta = MAX(unique, newval) -
2748 		    MAX(unique, ds->ds_reserved);
2749 
2750 		if (delta >
2751 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2752 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2753 			dsl_dataset_rele(ds, FTAG);
2754 			return (SET_ERROR(ENOSPC));
2755 		}
2756 	}
2757 
2758 	dsl_dataset_rele(ds, FTAG);
2759 	return (0);
2760 }
2761 
2762 void
2763 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2764     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2765 {
2766 	uint64_t newval;
2767 	uint64_t unique;
2768 	int64_t delta;
2769 
2770 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2771 	    source, sizeof (value), 1, &value, tx);
2772 
2773 	VERIFY0(dsl_prop_get_int_ds(ds,
2774 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2775 
2776 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
2777 	mutex_enter(&ds->ds_dir->dd_lock);
2778 	mutex_enter(&ds->ds_lock);
2779 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2780 	unique = ds->ds_phys->ds_unique_bytes;
2781 	delta = MAX(0, (int64_t)(newval - unique)) -
2782 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
2783 	ds->ds_reserved = newval;
2784 	mutex_exit(&ds->ds_lock);
2785 
2786 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2787 	mutex_exit(&ds->ds_dir->dd_lock);
2788 }
2789 
2790 static void
2791 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2792 {
2793 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
2794 	dsl_pool_t *dp = dmu_tx_pool(tx);
2795 	dsl_dataset_t *ds;
2796 
2797 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2798 	dsl_dataset_set_refreservation_sync_impl(ds,
2799 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2800 	dsl_dataset_rele(ds, FTAG);
2801 }
2802 
2803 int
2804 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2805     uint64_t refreservation)
2806 {
2807 	dsl_dataset_set_qr_arg_t ddsqra;
2808 
2809 	ddsqra.ddsqra_name = dsname;
2810 	ddsqra.ddsqra_source = source;
2811 	ddsqra.ddsqra_value = refreservation;
2812 
2813 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2814 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0));
2815 }
2816 
2817 /*
2818  * Return (in *usedp) the amount of space written in new that is not
2819  * present in oldsnap.  New may be a snapshot or the head.  Old must be
2820  * a snapshot before new, in new's filesystem (or its origin).  If not then
2821  * fail and return EINVAL.
2822  *
2823  * The written space is calculated by considering two components:  First, we
2824  * ignore any freed space, and calculate the written as new's used space
2825  * minus old's used space.  Next, we add in the amount of space that was freed
2826  * between the two snapshots, thus reducing new's used space relative to old's.
2827  * Specifically, this is the space that was born before old->ds_creation_txg,
2828  * and freed before new (ie. on new's deadlist or a previous deadlist).
2829  *
2830  * space freed                         [---------------------]
2831  * snapshots                       ---O-------O--------O-------O------
2832  *                                         oldsnap            new
2833  */
2834 int
2835 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2836     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2837 {
2838 	int err = 0;
2839 	uint64_t snapobj;
2840 	dsl_pool_t *dp = new->ds_dir->dd_pool;
2841 
2842 	ASSERT(dsl_pool_config_held(dp));
2843 
2844 	*usedp = 0;
2845 	*usedp += new->ds_phys->ds_referenced_bytes;
2846 	*usedp -= oldsnap->ds_phys->ds_referenced_bytes;
2847 
2848 	*compp = 0;
2849 	*compp += new->ds_phys->ds_compressed_bytes;
2850 	*compp -= oldsnap->ds_phys->ds_compressed_bytes;
2851 
2852 	*uncompp = 0;
2853 	*uncompp += new->ds_phys->ds_uncompressed_bytes;
2854 	*uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
2855 
2856 	snapobj = new->ds_object;
2857 	while (snapobj != oldsnap->ds_object) {
2858 		dsl_dataset_t *snap;
2859 		uint64_t used, comp, uncomp;
2860 
2861 		if (snapobj == new->ds_object) {
2862 			snap = new;
2863 		} else {
2864 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
2865 			if (err != 0)
2866 				break;
2867 		}
2868 
2869 		if (snap->ds_phys->ds_prev_snap_txg ==
2870 		    oldsnap->ds_phys->ds_creation_txg) {
2871 			/*
2872 			 * The blocks in the deadlist can not be born after
2873 			 * ds_prev_snap_txg, so get the whole deadlist space,
2874 			 * which is more efficient (especially for old-format
2875 			 * deadlists).  Unfortunately the deadlist code
2876 			 * doesn't have enough information to make this
2877 			 * optimization itself.
2878 			 */
2879 			dsl_deadlist_space(&snap->ds_deadlist,
2880 			    &used, &comp, &uncomp);
2881 		} else {
2882 			dsl_deadlist_space_range(&snap->ds_deadlist,
2883 			    0, oldsnap->ds_phys->ds_creation_txg,
2884 			    &used, &comp, &uncomp);
2885 		}
2886 		*usedp += used;
2887 		*compp += comp;
2888 		*uncompp += uncomp;
2889 
2890 		/*
2891 		 * If we get to the beginning of the chain of snapshots
2892 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2893 		 * was not a snapshot of/before new.
2894 		 */
2895 		snapobj = snap->ds_phys->ds_prev_snap_obj;
2896 		if (snap != new)
2897 			dsl_dataset_rele(snap, FTAG);
2898 		if (snapobj == 0) {
2899 			err = SET_ERROR(EINVAL);
2900 			break;
2901 		}
2902 
2903 	}
2904 	return (err);
2905 }
2906 
2907 /*
2908  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2909  * lastsnap, and all snapshots in between are deleted.
2910  *
2911  * blocks that would be freed            [---------------------------]
2912  * snapshots                       ---O-------O--------O-------O--------O
2913  *                                        firstsnap        lastsnap
2914  *
2915  * This is the set of blocks that were born after the snap before firstsnap,
2916  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2917  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2918  * We calculate this by iterating over the relevant deadlists (from the snap
2919  * after lastsnap, backward to the snap after firstsnap), summing up the
2920  * space on the deadlist that was born after the snap before firstsnap.
2921  */
2922 int
2923 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
2924     dsl_dataset_t *lastsnap,
2925     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2926 {
2927 	int err = 0;
2928 	uint64_t snapobj;
2929 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
2930 
2931 	ASSERT(dsl_dataset_is_snapshot(firstsnap));
2932 	ASSERT(dsl_dataset_is_snapshot(lastsnap));
2933 
2934 	/*
2935 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
2936 	 * is before lastsnap.
2937 	 */
2938 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
2939 	    firstsnap->ds_phys->ds_creation_txg >
2940 	    lastsnap->ds_phys->ds_creation_txg)
2941 		return (SET_ERROR(EINVAL));
2942 
2943 	*usedp = *compp = *uncompp = 0;
2944 
2945 	snapobj = lastsnap->ds_phys->ds_next_snap_obj;
2946 	while (snapobj != firstsnap->ds_object) {
2947 		dsl_dataset_t *ds;
2948 		uint64_t used, comp, uncomp;
2949 
2950 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
2951 		if (err != 0)
2952 			break;
2953 
2954 		dsl_deadlist_space_range(&ds->ds_deadlist,
2955 		    firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2956 		    &used, &comp, &uncomp);
2957 		*usedp += used;
2958 		*compp += comp;
2959 		*uncompp += uncomp;
2960 
2961 		snapobj = ds->ds_phys->ds_prev_snap_obj;
2962 		ASSERT3U(snapobj, !=, 0);
2963 		dsl_dataset_rele(ds, FTAG);
2964 	}
2965 	return (err);
2966 }
2967 
2968 /*
2969  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2970  * For example, they could both be snapshots of the same filesystem, and
2971  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
2972  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
2973  * filesystem.  Or 'earlier' could be the origin's origin.
2974  */
2975 boolean_t
2976 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
2977 {
2978 	dsl_pool_t *dp = later->ds_dir->dd_pool;
2979 	int error;
2980 	boolean_t ret;
2981 
2982 	ASSERT(dsl_pool_config_held(dp));
2983 
2984 	if (earlier->ds_phys->ds_creation_txg >=
2985 	    later->ds_phys->ds_creation_txg)
2986 		return (B_FALSE);
2987 
2988 	if (later->ds_dir == earlier->ds_dir)
2989 		return (B_TRUE);
2990 	if (!dsl_dir_is_clone(later->ds_dir))
2991 		return (B_FALSE);
2992 
2993 	if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
2994 		return (B_TRUE);
2995 	dsl_dataset_t *origin;
2996 	error = dsl_dataset_hold_obj(dp,
2997 	    later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
2998 	if (error != 0)
2999 		return (B_FALSE);
3000 	ret = dsl_dataset_is_before(origin, earlier);
3001 	dsl_dataset_rele(origin, FTAG);
3002 	return (ret);
3003 }
3004